2023-03-07 10:14:36,176 INFO [train2.py:879] (3/4) Training started 2023-03-07 10:14:36,176 INFO [train2.py:880] (3/4) {'frame_shift_ms': 10.0, 'allowed_excess_duration_ratio': 0.1, 'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23.3', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': '3b81ac9686aee539d447bb2085b2cdfc131c7c91', 'k2-git-date': 'Thu Jan 26 20:40:25 2023', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'surt', 'icefall-git-sha1': 'e9931b7-dirty', 'icefall-git-date': 'Fri Mar 3 16:27:17 2023', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r8n04', 'IP address': '10.1.8.4'}, 'beam_size': 10, 'reduction': 'sum', 'use_double_scores': True, 'world_size': 4, 'master_port': 12368, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('zipformer_ctc_att/exp/v0'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'att_rate': 0.8, 'num_decoder_layers': 6, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 10, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'full_libri': True, 'manifest_dir': PosixPath('data/manifests'), 'max_duration': 1000, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures'} 2023-03-07 10:14:36,469 INFO [lexicon.py:168] (3/4) Loading pre-compiled data/lang_bpe_500/Linv.pt 2023-03-07 10:14:37,048 INFO [train2.py:902] (3/4) About to create model 2023-03-07 10:14:37,545 INFO [zipformer.py:178] (3/4) At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8. 2023-03-07 10:14:37,606 INFO [train2.py:906] (3/4) Number of model parameters: 86083707 2023-03-07 10:14:42,025 INFO [train2.py:921] (3/4) Using DDP 2023-03-07 10:14:42,773 INFO [asr_datamodule.py:420] (3/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts 2023-03-07 10:14:42,878 INFO [asr_datamodule.py:224] (3/4) Enable MUSAN 2023-03-07 10:14:42,878 INFO [asr_datamodule.py:225] (3/4) About to get Musan cuts 2023-03-07 10:14:44,217 INFO [asr_datamodule.py:249] (3/4) Enable SpecAugment 2023-03-07 10:14:44,218 INFO [asr_datamodule.py:250] (3/4) Time warp factor: 80 2023-03-07 10:14:44,218 INFO [asr_datamodule.py:260] (3/4) Num frame mask: 10 2023-03-07 10:14:44,218 INFO [asr_datamodule.py:273] (3/4) About to create train dataset 2023-03-07 10:14:44,218 INFO [asr_datamodule.py:300] (3/4) Using DynamicBucketingSampler. 2023-03-07 10:14:46,633 INFO [asr_datamodule.py:316] (3/4) About to create train dataloader 2023-03-07 10:14:46,634 INFO [asr_datamodule.py:440] (3/4) About to get dev-clean cuts 2023-03-07 10:14:46,635 INFO [asr_datamodule.py:447] (3/4) About to get dev-other cuts 2023-03-07 10:14:46,637 INFO [asr_datamodule.py:347] (3/4) About to create dev dataset 2023-03-07 10:14:46,930 INFO [asr_datamodule.py:364] (3/4) About to create dev dataloader 2023-03-07 10:15:00,125 INFO [train2.py:809] (3/4) Epoch 1, batch 0, loss[ctc_loss=5.375, att_loss=1.186, loss=2.024, over 15375.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01096, over 35.00 utterances.], tot_loss[ctc_loss=5.375, att_loss=1.186, loss=2.024, over 15375.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01096, over 35.00 utterances.], batch size: 35, lr: 2.50e-02, grad_scale: 2.0 2023-03-07 10:15:00,125 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 10:15:12,517 INFO [train2.py:843] (3/4) Epoch 1, validation: ctc_loss=5.278, att_loss=1.419, loss=2.191, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 10:15:12,518 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 12486MB 2023-03-07 10:15:18,052 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 10:15:30,417 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.5946, 6.5952, 6.5951, 6.5945, 6.5952, 6.5949, 6.5952, 6.5941], device='cuda:3'), covar=tensor([0.0008, 0.0009, 0.0014, 0.0007, 0.0010, 0.0011, 0.0010, 0.0011], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([8.9534e-06, 8.9703e-06, 8.8950e-06, 8.7179e-06, 8.8979e-06, 8.8188e-06, 8.7280e-06, 8.8258e-06], device='cuda:3') 2023-03-07 10:15:43,096 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:16:22,273 INFO [train2.py:809] (3/4) Epoch 1, batch 50, loss[ctc_loss=1.308, att_loss=1.101, loss=1.142, over 17471.00 frames. utt_duration=1014 frames, utt_pad_proportion=0.04405, over 69.00 utterances.], tot_loss[ctc_loss=2.286, att_loss=1.137, loss=1.367, over 745186.62 frames. utt_duration=1297 frames, utt_pad_proportion=0.03435, over 2300.87 utterances.], batch size: 69, lr: 2.75e-02, grad_scale: 2.0 2023-03-07 10:16:36,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=5.52 vs. limit=2.0 2023-03-07 10:16:37,170 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=5.83 vs. limit=2.0 2023-03-07 10:16:47,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.76 vs. limit=2.0 2023-03-07 10:16:55,519 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=10.28 vs. limit=2.0 2023-03-07 10:17:06,745 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:17:22,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=4.04 vs. limit=2.0 2023-03-07 10:17:30,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=4.18 vs. limit=2.0 2023-03-07 10:17:31,217 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.328e+01 1.076e+02 2.257e+02 4.349e+02 2.620e+03, threshold=4.514e+02, percent-clipped=0.0 2023-03-07 10:17:31,262 INFO [train2.py:809] (3/4) Epoch 1, batch 100, loss[ctc_loss=1.26, att_loss=1.051, loss=1.093, over 17359.00 frames. utt_duration=880.5 frames, utt_pad_proportion=0.07902, over 79.00 utterances.], tot_loss[ctc_loss=1.672, att_loss=1.057, loss=1.18, over 1309765.05 frames. utt_duration=1285 frames, utt_pad_proportion=0.03471, over 4081.74 utterances.], batch size: 79, lr: 3.00e-02, grad_scale: 2.0 2023-03-07 10:18:04,391 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=31.25 vs. limit=5.0 2023-03-07 10:18:33,467 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:18:42,514 INFO [train2.py:809] (3/4) Epoch 1, batch 150, loss[ctc_loss=1.103, att_loss=0.9177, loss=0.9548, over 15876.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009767, over 39.00 utterances.], tot_loss[ctc_loss=1.466, att_loss=1.023, loss=1.111, over 1743511.36 frames. utt_duration=1231 frames, utt_pad_proportion=0.05379, over 5672.71 utterances.], batch size: 39, lr: 3.25e-02, grad_scale: 2.0 2023-03-07 10:19:48,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.064e+01 7.732e+01 1.287e+02 1.925e+02 9.711e+02, threshold=2.575e+02, percent-clipped=2.0 2023-03-07 10:19:48,667 INFO [train2.py:809] (3/4) Epoch 1, batch 200, loss[ctc_loss=1.175, att_loss=0.9671, loss=1.009, over 16965.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007597, over 50.00 utterances.], tot_loss[ctc_loss=1.362, att_loss=1.001, loss=1.073, over 2081305.86 frames. utt_duration=1250 frames, utt_pad_proportion=0.05047, over 6666.76 utterances.], batch size: 50, lr: 3.50e-02, grad_scale: 2.0 2023-03-07 10:20:40,904 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=13.28 vs. limit=5.0 2023-03-07 10:20:54,633 INFO [train2.py:809] (3/4) Epoch 1, batch 250, loss[ctc_loss=1.186, att_loss=0.9756, loss=1.018, over 16113.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.007095, over 42.00 utterances.], tot_loss[ctc_loss=1.307, att_loss=0.9913, loss=1.054, over 2349020.82 frames. utt_duration=1191 frames, utt_pad_proportion=0.06551, over 7901.43 utterances.], batch size: 42, lr: 3.75e-02, grad_scale: 2.0 2023-03-07 10:21:54,679 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:21:59,470 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 10:22:00,457 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.252e+01 6.827e+01 1.043e+02 1.996e+02 1.331e+03, threshold=2.085e+02, percent-clipped=17.0 2023-03-07 10:22:00,501 INFO [train2.py:809] (3/4) Epoch 1, batch 300, loss[ctc_loss=1.201, att_loss=0.972, loss=1.018, over 16486.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005439, over 46.00 utterances.], tot_loss[ctc_loss=1.263, att_loss=0.9755, loss=1.033, over 2553451.95 frames. utt_duration=1218 frames, utt_pad_proportion=0.06075, over 8393.71 utterances.], batch size: 46, lr: 4.00e-02, grad_scale: 2.0 2023-03-07 10:22:19,373 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.26 vs. limit=5.0 2023-03-07 10:22:50,011 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=14.09 vs. limit=5.0 2023-03-07 10:23:06,530 INFO [train2.py:809] (3/4) Epoch 1, batch 350, loss[ctc_loss=1.228, att_loss=0.9507, loss=1.006, over 17465.00 frames. utt_duration=1014 frames, utt_pad_proportion=0.04346, over 69.00 utterances.], tot_loss[ctc_loss=1.24, att_loss=0.9662, loss=1.021, over 2720792.19 frames. utt_duration=1233 frames, utt_pad_proportion=0.05541, over 8837.65 utterances.], batch size: 69, lr: 4.25e-02, grad_scale: 2.0 2023-03-07 10:23:14,299 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 10:23:54,078 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:24:12,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.983e+01 8.126e+01 1.923e+02 2.981e+02 6.249e+02, threshold=3.846e+02, percent-clipped=46.0 2023-03-07 10:24:12,820 INFO [train2.py:809] (3/4) Epoch 1, batch 400, loss[ctc_loss=1.186, att_loss=0.9648, loss=1.009, over 17082.00 frames. utt_duration=698.7 frames, utt_pad_proportion=0.1244, over 98.00 utterances.], tot_loss[ctc_loss=1.217, att_loss=0.9555, loss=1.008, over 2836613.05 frames. utt_duration=1207 frames, utt_pad_proportion=0.06374, over 9415.81 utterances.], batch size: 98, lr: 4.50e-02, grad_scale: 4.0 2023-03-07 10:24:18,985 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.6657, 6.7674, 6.7706, 6.7485, 6.6849, 6.7677, 6.7285, 6.7690], device='cuda:3'), covar=tensor([0.0548, 0.0050, 0.0041, 0.0145, 0.0285, 0.0067, 0.0153, 0.0048], device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0009, 0.0008, 0.0008, 0.0008, 0.0008, 0.0008, 0.0008], device='cuda:3'), out_proj_covar=tensor([8.7286e-06, 8.4095e-06, 8.2633e-06, 8.1789e-06, 8.3487e-06, 8.4705e-06, 8.0797e-06, 8.3627e-06], device='cuda:3') 2023-03-07 10:25:02,043 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-07 10:25:13,974 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:25:17,465 INFO [train2.py:809] (3/4) Epoch 1, batch 450, loss[ctc_loss=1.135, att_loss=0.9248, loss=0.9668, over 16761.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006189, over 48.00 utterances.], tot_loss[ctc_loss=1.194, att_loss=0.9426, loss=0.993, over 2935234.20 frames. utt_duration=1219 frames, utt_pad_proportion=0.06102, over 9640.70 utterances.], batch size: 48, lr: 4.75e-02, grad_scale: 4.0 2023-03-07 10:26:22,551 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.753e+01 1.641e+02 2.817e+02 4.953e+02 1.195e+03, threshold=5.634e+02, percent-clipped=34.0 2023-03-07 10:26:22,598 INFO [train2.py:809] (3/4) Epoch 1, batch 500, loss[ctc_loss=1.043, att_loss=0.798, loss=0.847, over 16181.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006213, over 41.00 utterances.], tot_loss[ctc_loss=1.171, att_loss=0.926, loss=0.975, over 3008936.93 frames. utt_duration=1249 frames, utt_pad_proportion=0.0552, over 9648.57 utterances.], batch size: 41, lr: 4.99e-02, grad_scale: 4.0 2023-03-07 10:26:28,410 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.32 vs. limit=5.0 2023-03-07 10:26:37,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=12.90 vs. limit=5.0 2023-03-07 10:27:06,309 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5432, 2.9391, 4.6262, 3.6366, 3.0001, 3.9508, 4.7172, 3.7899], device='cuda:3'), covar=tensor([0.0475, 0.2579, 0.0209, 0.0835, 0.2547, 0.0695, 0.0362, 0.0438], device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0010, 0.0010, 0.0010, 0.0010, 0.0010], device='cuda:3'), out_proj_covar=tensor([9.2977e-06, 9.3605e-06, 9.0215e-06, 9.5736e-06, 1.0530e-05, 9.0789e-06, 9.0111e-06, 8.8934e-06], device='cuda:3') 2023-03-07 10:27:27,832 INFO [train2.py:809] (3/4) Epoch 1, batch 550, loss[ctc_loss=1.074, att_loss=0.8808, loss=0.9196, over 17343.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02118, over 59.00 utterances.], tot_loss[ctc_loss=1.15, att_loss=0.9133, loss=0.9606, over 3068614.50 frames. utt_duration=1217 frames, utt_pad_proportion=0.06307, over 10100.89 utterances.], batch size: 59, lr: 4.98e-02, grad_scale: 4.0 2023-03-07 10:27:41,738 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:27:49,946 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:28:06,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.19 vs. limit=5.0 2023-03-07 10:28:19,163 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:28:32,099 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:28:33,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.073e+01 2.172e+02 3.350e+02 5.891e+02 2.324e+03, threshold=6.699e+02, percent-clipped=26.0 2023-03-07 10:28:33,109 INFO [train2.py:809] (3/4) Epoch 1, batch 600, loss[ctc_loss=1.028, att_loss=0.8641, loss=0.8968, over 17405.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.04587, over 69.00 utterances.], tot_loss[ctc_loss=1.126, att_loss=0.9011, loss=0.9461, over 3123683.29 frames. utt_duration=1231 frames, utt_pad_proportion=0.057, over 10158.96 utterances.], batch size: 69, lr: 4.98e-02, grad_scale: 4.0 2023-03-07 10:29:01,862 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 10:29:09,931 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:29:33,576 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:29:37,167 INFO [train2.py:809] (3/4) Epoch 1, batch 650, loss[ctc_loss=0.9522, att_loss=0.8631, loss=0.8809, over 16532.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006632, over 45.00 utterances.], tot_loss[ctc_loss=1.093, att_loss=0.8899, loss=0.9305, over 3157096.29 frames. utt_duration=1223 frames, utt_pad_proportion=0.05968, over 10338.54 utterances.], batch size: 45, lr: 4.98e-02, grad_scale: 4.0 2023-03-07 10:29:37,424 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 10:29:38,582 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:30:42,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 2.720e+02 3.813e+02 5.930e+02 1.074e+03, threshold=7.626e+02, percent-clipped=15.0 2023-03-07 10:30:42,322 INFO [train2.py:809] (3/4) Epoch 1, batch 700, loss[ctc_loss=0.849, att_loss=0.8357, loss=0.8384, over 17259.00 frames. utt_duration=1257 frames, utt_pad_proportion=0.01283, over 55.00 utterances.], tot_loss[ctc_loss=1.051, att_loss=0.8796, loss=0.9138, over 3187358.59 frames. utt_duration=1213 frames, utt_pad_proportion=0.06194, over 10526.19 utterances.], batch size: 55, lr: 4.98e-02, grad_scale: 4.0 2023-03-07 10:31:13,128 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-03-07 10:31:21,616 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 2023-03-07 10:31:32,120 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 10:31:37,693 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:31:47,681 INFO [train2.py:809] (3/4) Epoch 1, batch 750, loss[ctc_loss=0.856, att_loss=0.8707, loss=0.8677, over 17071.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007794, over 52.00 utterances.], tot_loss[ctc_loss=1.002, att_loss=0.8686, loss=0.8952, over 3210446.89 frames. utt_duration=1220 frames, utt_pad_proportion=0.05904, over 10534.98 utterances.], batch size: 52, lr: 4.97e-02, grad_scale: 4.0 2023-03-07 10:31:56,556 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3565, 3.6547, 3.8341, 4.0807, 3.7876, 3.5601, 3.7073, 3.7723], device='cuda:3'), covar=tensor([0.4765, 0.3589, 0.3014, 0.2008, 0.3304, 0.4140, 0.3488, 0.3308], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0028, 0.0027, 0.0025, 0.0028, 0.0032, 0.0028, 0.0028], device='cuda:3'), out_proj_covar=tensor([3.1635e-05, 2.7351e-05, 2.7502e-05, 2.3528e-05, 2.5926e-05, 3.0017e-05, 2.7623e-05, 2.6838e-05], device='cuda:3') 2023-03-07 10:32:34,594 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:32:52,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-07 10:32:52,870 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.221e+02 2.689e+02 3.178e+02 4.558e+02 8.298e+02, threshold=6.355e+02, percent-clipped=3.0 2023-03-07 10:32:52,913 INFO [train2.py:809] (3/4) Epoch 1, batch 800, loss[ctc_loss=0.7601, att_loss=0.8355, loss=0.8204, over 16544.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006002, over 45.00 utterances.], tot_loss[ctc_loss=0.9567, att_loss=0.8597, loss=0.8791, over 3230618.45 frames. utt_duration=1218 frames, utt_pad_proportion=0.05855, over 10619.49 utterances.], batch size: 45, lr: 4.97e-02, grad_scale: 8.0 2023-03-07 10:32:59,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 10:33:52,605 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:33:57,214 INFO [train2.py:809] (3/4) Epoch 1, batch 850, loss[ctc_loss=0.7201, att_loss=0.7577, loss=0.7502, over 15936.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007326, over 41.00 utterances.], tot_loss[ctc_loss=0.9131, att_loss=0.8459, loss=0.8594, over 3240746.89 frames. utt_duration=1219 frames, utt_pad_proportion=0.05919, over 10646.47 utterances.], batch size: 41, lr: 4.96e-02, grad_scale: 8.0 2023-03-07 10:34:39,540 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.33 vs. limit=2.0 2023-03-07 10:34:53,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.10 vs. limit=2.0 2023-03-07 10:35:01,442 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.412e+02 3.625e+02 4.631e+02 5.738e+02 1.582e+03, threshold=9.262e+02, percent-clipped=18.0 2023-03-07 10:35:01,486 INFO [train2.py:809] (3/4) Epoch 1, batch 900, loss[ctc_loss=0.7673, att_loss=0.7549, loss=0.7573, over 17321.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01029, over 55.00 utterances.], tot_loss[ctc_loss=0.8718, att_loss=0.8231, loss=0.8328, over 3238049.07 frames. utt_duration=1231 frames, utt_pad_proportion=0.06079, over 10532.63 utterances.], batch size: 55, lr: 4.96e-02, grad_scale: 8.0 2023-03-07 10:35:10,296 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 10:35:13,295 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2023-03-07 10:35:23,187 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 10:35:31,242 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 10:35:42,317 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-03-07 10:35:59,563 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 10:36:05,678 INFO [train2.py:809] (3/4) Epoch 1, batch 950, loss[ctc_loss=0.7491, att_loss=0.711, loss=0.7186, over 16953.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007646, over 50.00 utterances.], tot_loss[ctc_loss=0.8374, att_loss=0.7943, loss=0.8029, over 3238414.18 frames. utt_duration=1210 frames, utt_pad_proportion=0.06804, over 10717.55 utterances.], batch size: 50, lr: 4.96e-02, grad_scale: 8.0 2023-03-07 10:36:07,097 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:37:08,810 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:37:09,962 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.067e+02 5.280e+02 6.177e+02 7.418e+02 1.446e+03, threshold=1.235e+03, percent-clipped=9.0 2023-03-07 10:37:10,007 INFO [train2.py:809] (3/4) Epoch 1, batch 1000, loss[ctc_loss=0.7709, att_loss=0.6966, loss=0.7115, over 17056.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009456, over 53.00 utterances.], tot_loss[ctc_loss=0.8078, att_loss=0.7629, loss=0.7719, over 3245371.40 frames. utt_duration=1231 frames, utt_pad_proportion=0.06134, over 10557.94 utterances.], batch size: 53, lr: 4.95e-02, grad_scale: 8.0 2023-03-07 10:37:31,628 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8940, 3.8773, 3.9963, 3.9726, 4.1440, 4.0941, 4.1532, 4.0405], device='cuda:3'), covar=tensor([0.1538, 0.1325, 0.1313, 0.1047, 0.1243, 0.0924, 0.1078, 0.1267], device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0038, 0.0038, 0.0039, 0.0036, 0.0042, 0.0030, 0.0033], device='cuda:3'), out_proj_covar=tensor([3.6712e-05, 3.4556e-05, 3.2668e-05, 3.4212e-05, 3.1727e-05, 3.6342e-05, 2.6171e-05, 2.9764e-05], device='cuda:3') 2023-03-07 10:37:38,106 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3380, 4.6421, 4.4263, 4.7065, 4.1219, 4.5486, 4.6227, 4.6345], device='cuda:3'), covar=tensor([0.1028, 0.0678, 0.0895, 0.0603, 0.1066, 0.0748, 0.0915, 0.0677], device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0038, 0.0040, 0.0032, 0.0041, 0.0034, 0.0039, 0.0034], device='cuda:3'), out_proj_covar=tensor([3.8359e-05, 3.2854e-05, 3.5520e-05, 2.7236e-05, 3.6197e-05, 3.0069e-05, 3.4801e-05, 2.7717e-05], device='cuda:3') 2023-03-07 10:38:04,776 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:38:15,417 INFO [train2.py:809] (3/4) Epoch 1, batch 1050, loss[ctc_loss=0.7108, att_loss=0.6286, loss=0.645, over 17285.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01272, over 55.00 utterances.], tot_loss[ctc_loss=0.7791, att_loss=0.7291, loss=0.7391, over 3261377.83 frames. utt_duration=1225 frames, utt_pad_proportion=0.05834, over 10665.87 utterances.], batch size: 55, lr: 4.95e-02, grad_scale: 8.0 2023-03-07 10:38:57,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.57 vs. limit=5.0 2023-03-07 10:39:07,966 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:39:21,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.042e+02 5.471e+02 6.537e+02 8.714e+02 1.508e+03, threshold=1.307e+03, percent-clipped=3.0 2023-03-07 10:39:21,111 INFO [train2.py:809] (3/4) Epoch 1, batch 1100, loss[ctc_loss=0.5443, att_loss=0.4927, loss=0.5031, over 15951.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.006766, over 41.00 utterances.], tot_loss[ctc_loss=0.7482, att_loss=0.6925, loss=0.7036, over 3266084.97 frames. utt_duration=1225 frames, utt_pad_proportion=0.05832, over 10678.68 utterances.], batch size: 41, lr: 4.94e-02, grad_scale: 8.0 2023-03-07 10:40:27,182 INFO [train2.py:809] (3/4) Epoch 1, batch 1150, loss[ctc_loss=0.6508, att_loss=0.5574, loss=0.5761, over 17369.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03455, over 63.00 utterances.], tot_loss[ctc_loss=0.7219, att_loss=0.6605, loss=0.6728, over 3266977.02 frames. utt_duration=1238 frames, utt_pad_proportion=0.05653, over 10570.19 utterances.], batch size: 63, lr: 4.94e-02, grad_scale: 8.0 2023-03-07 10:40:30,117 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:41:32,903 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 5.540e+02 7.016e+02 8.798e+02 2.623e+03, threshold=1.403e+03, percent-clipped=4.0 2023-03-07 10:41:32,954 INFO [train2.py:809] (3/4) Epoch 1, batch 1200, loss[ctc_loss=0.6443, att_loss=0.5437, loss=0.5638, over 17346.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03594, over 63.00 utterances.], tot_loss[ctc_loss=0.6989, att_loss=0.6326, loss=0.6459, over 3262917.72 frames. utt_duration=1226 frames, utt_pad_proportion=0.05856, over 10660.97 utterances.], batch size: 63, lr: 4.93e-02, grad_scale: 8.0 2023-03-07 10:41:35,584 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 10:41:43,935 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 2023-03-07 10:41:49,229 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 10:41:54,869 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:42:03,506 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:42:24,788 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6201, 3.2082, 3.4928, 4.3119, 3.2453, 3.6700, 3.4792, 3.6537], device='cuda:3'), covar=tensor([0.2140, 0.3218, 2.9417, 0.1357, 0.1952, 0.2431, 0.6611, 0.2041], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0020, 0.0029, 0.0025, 0.0017, 0.0024, 0.0022, 0.0024], device='cuda:3'), out_proj_covar=tensor([1.4644e-05, 1.2273e-05, 2.0249e-05, 1.4914e-05, 1.1252e-05, 1.5665e-05, 1.4383e-05, 1.4857e-05], device='cuda:3') 2023-03-07 10:42:29,481 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-07 10:42:31,574 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:42:37,645 INFO [train2.py:809] (3/4) Epoch 1, batch 1250, loss[ctc_loss=0.5273, att_loss=0.4709, loss=0.4822, over 15398.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.009545, over 35.00 utterances.], tot_loss[ctc_loss=0.6766, att_loss=0.6054, loss=0.6197, over 3267513.51 frames. utt_duration=1236 frames, utt_pad_proportion=0.05558, over 10583.61 utterances.], batch size: 35, lr: 4.92e-02, grad_scale: 8.0 2023-03-07 10:42:47,361 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5981, 3.9431, 3.9936, 4.0026, 4.0502, 3.8177, 3.5588, 3.8323], device='cuda:3'), covar=tensor([0.1984, 0.1178, 0.0955, 0.1237, 0.1153, 0.1643, 0.1614, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0047, 0.0050, 0.0051, 0.0050, 0.0053, 0.0055, 0.0056], device='cuda:3'), out_proj_covar=tensor([4.5942e-05, 3.6748e-05, 3.9757e-05, 4.2340e-05, 4.1252e-05, 4.2088e-05, 4.6926e-05, 4.6395e-05], device='cuda:3') 2023-03-07 10:42:56,414 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:43:05,043 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 10:43:34,643 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 10:43:43,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.330e+02 5.046e+02 6.291e+02 8.119e+02 1.875e+03, threshold=1.258e+03, percent-clipped=2.0 2023-03-07 10:43:43,414 INFO [train2.py:809] (3/4) Epoch 1, batch 1300, loss[ctc_loss=0.531, att_loss=0.4641, loss=0.4775, over 16108.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.007341, over 42.00 utterances.], tot_loss[ctc_loss=0.6583, att_loss=0.5829, loss=0.5979, over 3263970.39 frames. utt_duration=1221 frames, utt_pad_proportion=0.06189, over 10708.54 utterances.], batch size: 42, lr: 4.92e-02, grad_scale: 8.0 2023-03-07 10:44:50,650 INFO [train2.py:809] (3/4) Epoch 1, batch 1350, loss[ctc_loss=0.5518, att_loss=0.4789, loss=0.4935, over 17068.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008055, over 52.00 utterances.], tot_loss[ctc_loss=0.6377, att_loss=0.5609, loss=0.5763, over 3274460.67 frames. utt_duration=1247 frames, utt_pad_proportion=0.05338, over 10519.33 utterances.], batch size: 52, lr: 4.91e-02, grad_scale: 8.0 2023-03-07 10:44:51,574 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-03-07 10:45:59,937 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.741e+02 4.980e+02 5.949e+02 7.400e+02 1.342e+03, threshold=1.190e+03, percent-clipped=3.0 2023-03-07 10:45:59,980 INFO [train2.py:809] (3/4) Epoch 1, batch 1400, loss[ctc_loss=0.4863, att_loss=0.4244, loss=0.4368, over 13636.00 frames. utt_duration=1820 frames, utt_pad_proportion=0.07024, over 30.00 utterances.], tot_loss[ctc_loss=0.6211, att_loss=0.5431, loss=0.5587, over 3272893.45 frames. utt_duration=1229 frames, utt_pad_proportion=0.05823, over 10669.17 utterances.], batch size: 30, lr: 4.91e-02, grad_scale: 8.0 2023-03-07 10:46:45,495 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:46:50,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0323, 5.7842, 5.4587, 5.8251, 5.6936, 5.7383, 5.6484, 5.7409], device='cuda:3'), covar=tensor([0.0520, 0.0351, 0.0631, 0.0528, 0.0531, 0.0760, 0.0927, 0.0572], device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0063, 0.0056, 0.0059, 0.0060, 0.0058, 0.0053, 0.0063], device='cuda:3'), out_proj_covar=tensor([4.2389e-05, 5.2748e-05, 4.6703e-05, 5.3563e-05, 5.3514e-05, 5.3620e-05, 5.1150e-05, 5.4603e-05], device='cuda:3') 2023-03-07 10:47:08,138 INFO [train2.py:809] (3/4) Epoch 1, batch 1450, loss[ctc_loss=0.6936, att_loss=0.5497, loss=0.5785, over 13138.00 frames. utt_duration=359.1 frames, utt_pad_proportion=0.371, over 147.00 utterances.], tot_loss[ctc_loss=0.6066, att_loss=0.5279, loss=0.5437, over 3263253.43 frames. utt_duration=1209 frames, utt_pad_proportion=0.06716, over 10813.11 utterances.], batch size: 147, lr: 4.90e-02, grad_scale: 8.0 2023-03-07 10:48:08,728 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 10:48:16,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.693e+02 5.276e+02 6.895e+02 9.021e+02 2.982e+03, threshold=1.379e+03, percent-clipped=6.0 2023-03-07 10:48:16,298 INFO [train2.py:809] (3/4) Epoch 1, batch 1500, loss[ctc_loss=0.5357, att_loss=0.4661, loss=0.48, over 17095.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01548, over 56.00 utterances.], tot_loss[ctc_loss=0.5905, att_loss=0.513, loss=0.5285, over 3259530.90 frames. utt_duration=1200 frames, utt_pad_proportion=0.06907, over 10874.25 utterances.], batch size: 56, lr: 4.89e-02, grad_scale: 8.0 2023-03-07 10:48:19,335 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 10:48:27,287 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-07 10:48:53,010 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7156, 2.5878, 3.3932, 3.5649, 3.0591, 3.2773, 3.0331, 3.3803], device='cuda:3'), covar=tensor([0.1036, 0.5586, 0.1528, 0.1092, 0.4031, 0.1690, 0.1646, 0.1472], device='cuda:3'), in_proj_covar=tensor([0.0045, 0.0035, 0.0040, 0.0040, 0.0042, 0.0042, 0.0046, 0.0037], device='cuda:3'), out_proj_covar=tensor([3.9849e-05, 3.5336e-05, 3.4913e-05, 3.4323e-05, 3.8590e-05, 3.6303e-05, 4.2616e-05, 3.2401e-05], device='cuda:3') 2023-03-07 10:49:26,099 INFO [train2.py:809] (3/4) Epoch 1, batch 1550, loss[ctc_loss=0.5088, att_loss=0.454, loss=0.465, over 17024.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.00848, over 51.00 utterances.], tot_loss[ctc_loss=0.5741, att_loss=0.4988, loss=0.5139, over 3264852.13 frames. utt_duration=1226 frames, utt_pad_proportion=0.06266, over 10667.02 utterances.], batch size: 51, lr: 4.89e-02, grad_scale: 8.0 2023-03-07 10:49:26,252 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 10:49:46,971 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:50:35,851 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.29 vs. limit=5.0 2023-03-07 10:50:36,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.585e+02 5.591e+02 7.426e+02 9.323e+02 2.170e+03, threshold=1.485e+03, percent-clipped=5.0 2023-03-07 10:50:36,355 INFO [train2.py:809] (3/4) Epoch 1, batch 1600, loss[ctc_loss=0.654, att_loss=0.5322, loss=0.5566, over 13757.00 frames. utt_duration=373.2 frames, utt_pad_proportion=0.3418, over 148.00 utterances.], tot_loss[ctc_loss=0.56, att_loss=0.4869, loss=0.5015, over 3258388.62 frames. utt_duration=1217 frames, utt_pad_proportion=0.06769, over 10724.44 utterances.], batch size: 148, lr: 4.88e-02, grad_scale: 8.0 2023-03-07 10:50:50,125 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:51:00,823 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.36 vs. limit=2.0 2023-03-07 10:51:13,091 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:51:44,856 INFO [train2.py:809] (3/4) Epoch 1, batch 1650, loss[ctc_loss=0.445, att_loss=0.395, loss=0.405, over 15375.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01096, over 35.00 utterances.], tot_loss[ctc_loss=0.5481, att_loss=0.4771, loss=0.4913, over 3262978.48 frames. utt_duration=1201 frames, utt_pad_proportion=0.06991, over 10876.85 utterances.], batch size: 35, lr: 4.87e-02, grad_scale: 8.0 2023-03-07 10:52:15,860 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:52:55,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.825e+02 4.712e+02 6.880e+02 8.240e+02 2.356e+03, threshold=1.376e+03, percent-clipped=2.0 2023-03-07 10:52:55,869 INFO [train2.py:809] (3/4) Epoch 1, batch 1700, loss[ctc_loss=0.4277, att_loss=0.3934, loss=0.4003, over 15385.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.01009, over 35.00 utterances.], tot_loss[ctc_loss=0.5342, att_loss=0.4666, loss=0.4801, over 3263966.91 frames. utt_duration=1193 frames, utt_pad_proportion=0.07219, over 10959.75 utterances.], batch size: 35, lr: 4.86e-02, grad_scale: 8.0 2023-03-07 10:54:07,065 INFO [train2.py:809] (3/4) Epoch 1, batch 1750, loss[ctc_loss=0.5, att_loss=0.4423, loss=0.4539, over 17048.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009144, over 52.00 utterances.], tot_loss[ctc_loss=0.5189, att_loss=0.4564, loss=0.4689, over 3274584.88 frames. utt_duration=1229 frames, utt_pad_proportion=0.06074, over 10669.37 utterances.], batch size: 52, lr: 4.86e-02, grad_scale: 8.0 2023-03-07 10:55:02,384 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 10:55:17,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.104e+02 4.727e+02 5.545e+02 7.358e+02 1.291e+03, threshold=1.109e+03, percent-clipped=0.0 2023-03-07 10:55:17,513 INFO [train2.py:809] (3/4) Epoch 1, batch 1800, loss[ctc_loss=0.5319, att_loss=0.4595, loss=0.474, over 17289.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02443, over 59.00 utterances.], tot_loss[ctc_loss=0.5073, att_loss=0.4482, loss=0.46, over 3274078.65 frames. utt_duration=1246 frames, utt_pad_proportion=0.05664, over 10524.82 utterances.], batch size: 59, lr: 4.85e-02, grad_scale: 8.0 2023-03-07 10:55:28,588 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:56:10,009 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:56:27,798 INFO [train2.py:809] (3/4) Epoch 1, batch 1850, loss[ctc_loss=0.4665, att_loss=0.4364, loss=0.4424, over 17050.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009934, over 53.00 utterances.], tot_loss[ctc_loss=0.498, att_loss=0.4429, loss=0.4539, over 3266201.41 frames. utt_duration=1221 frames, utt_pad_proportion=0.0644, over 10712.22 utterances.], batch size: 53, lr: 4.84e-02, grad_scale: 8.0 2023-03-07 10:56:36,578 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 10:56:36,813 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 10:56:56,783 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 10:57:37,122 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 10:57:39,638 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.994e+02 5.025e+02 6.383e+02 8.643e+02 1.619e+03, threshold=1.277e+03, percent-clipped=10.0 2023-03-07 10:57:39,683 INFO [train2.py:809] (3/4) Epoch 1, batch 1900, loss[ctc_loss=0.5281, att_loss=0.474, loss=0.4849, over 17048.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008984, over 52.00 utterances.], tot_loss[ctc_loss=0.4865, att_loss=0.436, loss=0.4461, over 3268717.19 frames. utt_duration=1237 frames, utt_pad_proportion=0.0589, over 10584.55 utterances.], batch size: 52, lr: 4.83e-02, grad_scale: 8.0 2023-03-07 10:58:05,221 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 10:58:10,268 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 10:58:22,755 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 10:58:50,735 INFO [train2.py:809] (3/4) Epoch 1, batch 1950, loss[ctc_loss=0.4142, att_loss=0.4121, loss=0.4125, over 16816.00 frames. utt_duration=687.8 frames, utt_pad_proportion=0.1381, over 98.00 utterances.], tot_loss[ctc_loss=0.4756, att_loss=0.4299, loss=0.439, over 3274502.76 frames. utt_duration=1229 frames, utt_pad_proportion=0.0578, over 10673.36 utterances.], batch size: 98, lr: 4.83e-02, grad_scale: 8.0 2023-03-07 10:59:09,118 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4716, 4.7946, 4.7236, 3.8273, 4.9085, 4.2465, 4.3519, 4.2329], device='cuda:3'), covar=tensor([0.2030, 0.0373, 0.1197, 0.2360, 0.0463, 0.1772, 0.2349, 0.2228], device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0030, 0.0027, 0.0043, 0.0036, 0.0031, 0.0030, 0.0027], device='cuda:3'), out_proj_covar=tensor([1.4375e-05, 1.4620e-05, 1.4566e-05, 2.5480e-05, 1.8370e-05, 1.8470e-05, 1.8679e-05, 1.7053e-05], device='cuda:3') 2023-03-07 10:59:14,973 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 10:59:54,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.20 vs. limit=2.0 2023-03-07 11:00:06,971 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.041e+02 5.008e+02 6.125e+02 7.778e+02 1.503e+03, threshold=1.225e+03, percent-clipped=3.0 2023-03-07 11:00:07,015 INFO [train2.py:809] (3/4) Epoch 1, batch 2000, loss[ctc_loss=0.4199, att_loss=0.379, loss=0.3872, over 15364.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01111, over 35.00 utterances.], tot_loss[ctc_loss=0.4683, att_loss=0.4258, loss=0.4343, over 3279052.80 frames. utt_duration=1236 frames, utt_pad_proportion=0.05426, over 10625.77 utterances.], batch size: 35, lr: 4.82e-02, grad_scale: 16.0 2023-03-07 11:00:51,035 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8387, 5.6728, 5.1223, 5.8048, 5.5205, 5.1954, 5.5362, 5.2111], device='cuda:3'), covar=tensor([0.0572, 0.0589, 0.0671, 0.0399, 0.0568, 0.0917, 0.0670, 0.1164], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0123, 0.0124, 0.0101, 0.0102, 0.0136, 0.0111, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-07 11:01:23,778 INFO [train2.py:809] (3/4) Epoch 1, batch 2050, loss[ctc_loss=0.3949, att_loss=0.4022, loss=0.4008, over 17314.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01114, over 55.00 utterances.], tot_loss[ctc_loss=0.4537, att_loss=0.4186, loss=0.4256, over 3282616.42 frames. utt_duration=1225 frames, utt_pad_proportion=0.05706, over 10735.99 utterances.], batch size: 55, lr: 4.81e-02, grad_scale: 16.0 2023-03-07 11:01:35,742 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-07 11:02:23,680 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 11:02:39,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.629e+02 3.921e+02 4.697e+02 6.442e+02 1.106e+03, threshold=9.394e+02, percent-clipped=0.0 2023-03-07 11:02:39,934 INFO [train2.py:809] (3/4) Epoch 1, batch 2100, loss[ctc_loss=0.3729, att_loss=0.366, loss=0.3674, over 14488.00 frames. utt_duration=1812 frames, utt_pad_proportion=0.0436, over 32.00 utterances.], tot_loss[ctc_loss=0.4403, att_loss=0.4127, loss=0.4182, over 3279212.40 frames. utt_duration=1237 frames, utt_pad_proportion=0.0541, over 10612.41 utterances.], batch size: 32, lr: 4.80e-02, grad_scale: 16.0 2023-03-07 11:03:29,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3273, 4.6984, 4.7233, 4.0852, 4.9135, 4.5529, 4.7151, 4.5427], device='cuda:3'), covar=tensor([0.0290, 0.0178, 0.0187, 0.0452, 0.0171, 0.0223, 0.0184, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0060, 0.0065, 0.0049, 0.0067, 0.0082, 0.0058, 0.0072], device='cuda:3'), out_proj_covar=tensor([7.1018e-05, 5.3320e-05, 5.7812e-05, 4.5227e-05, 6.5977e-05, 8.2636e-05, 5.3046e-05, 7.1091e-05], device='cuda:3') 2023-03-07 11:03:35,254 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:03:55,160 INFO [train2.py:809] (3/4) Epoch 1, batch 2150, loss[ctc_loss=0.3992, att_loss=0.4155, loss=0.4122, over 16501.00 frames. utt_duration=1436 frames, utt_pad_proportion=0.005284, over 46.00 utterances.], tot_loss[ctc_loss=0.4295, att_loss=0.4079, loss=0.4122, over 3278147.79 frames. utt_duration=1229 frames, utt_pad_proportion=0.0565, over 10685.69 utterances.], batch size: 46, lr: 4.79e-02, grad_scale: 16.0 2023-03-07 11:04:17,923 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-03-07 11:05:00,220 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:05:10,569 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.636e+02 4.045e+02 5.653e+02 7.128e+02 1.127e+03, threshold=1.131e+03, percent-clipped=5.0 2023-03-07 11:05:10,615 INFO [train2.py:809] (3/4) Epoch 1, batch 2200, loss[ctc_loss=0.3806, att_loss=0.3943, loss=0.3916, over 16980.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.006545, over 50.00 utterances.], tot_loss[ctc_loss=0.4189, att_loss=0.4033, loss=0.4064, over 3273584.31 frames. utt_duration=1239 frames, utt_pad_proportion=0.05552, over 10584.87 utterances.], batch size: 50, lr: 4.78e-02, grad_scale: 16.0 2023-03-07 11:05:29,598 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-03-07 11:05:30,193 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:05:43,457 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-07 11:05:45,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.60 vs. limit=5.0 2023-03-07 11:05:49,151 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:06:26,413 INFO [train2.py:809] (3/4) Epoch 1, batch 2250, loss[ctc_loss=0.4273, att_loss=0.4035, loss=0.4083, over 17073.00 frames. utt_duration=691.5 frames, utt_pad_proportion=0.1335, over 99.00 utterances.], tot_loss[ctc_loss=0.407, att_loss=0.3975, loss=0.3994, over 3273092.07 frames. utt_duration=1242 frames, utt_pad_proportion=0.05396, over 10551.16 utterances.], batch size: 99, lr: 4.77e-02, grad_scale: 16.0 2023-03-07 11:06:33,476 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8401, 5.7488, 5.1089, 5.9604, 5.4797, 5.3873, 5.5055, 5.3230], device='cuda:3'), covar=tensor([0.0602, 0.0594, 0.0596, 0.0376, 0.0486, 0.0758, 0.0650, 0.1161], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0150, 0.0140, 0.0121, 0.0119, 0.0158, 0.0136, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-07 11:06:52,798 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 11:06:57,127 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:07:06,102 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5142, 3.5609, 3.7507, 3.6724, 4.0434, 3.7306, 3.2108, 3.5979], device='cuda:3'), covar=tensor([0.0275, 0.0245, 0.0167, 0.0192, 0.0171, 0.0193, 0.0468, 0.0259], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0024, 0.0024, 0.0025, 0.0025, 0.0024, 0.0031, 0.0027], device='cuda:3'), out_proj_covar=tensor([1.9133e-05, 1.6794e-05, 1.6614e-05, 1.7546e-05, 1.7986e-05, 1.6707e-05, 2.3579e-05, 1.8728e-05], device='cuda:3') 2023-03-07 11:07:43,375 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.835e+02 4.889e+02 6.269e+02 9.585e+02, threshold=9.779e+02, percent-clipped=0.0 2023-03-07 11:07:43,421 INFO [train2.py:809] (3/4) Epoch 1, batch 2300, loss[ctc_loss=0.3237, att_loss=0.3504, loss=0.345, over 16181.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006274, over 41.00 utterances.], tot_loss[ctc_loss=0.3978, att_loss=0.394, loss=0.3947, over 3282698.37 frames. utt_duration=1246 frames, utt_pad_proportion=0.05091, over 10549.51 utterances.], batch size: 41, lr: 4.77e-02, grad_scale: 16.0 2023-03-07 11:08:06,520 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:08:15,964 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 2023-03-07 11:08:43,536 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:08:59,804 INFO [train2.py:809] (3/4) Epoch 1, batch 2350, loss[ctc_loss=0.3627, att_loss=0.3767, loss=0.3739, over 17383.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03391, over 63.00 utterances.], tot_loss[ctc_loss=0.3883, att_loss=0.39, loss=0.3896, over 3282275.38 frames. utt_duration=1260 frames, utt_pad_proportion=0.04787, over 10433.73 utterances.], batch size: 63, lr: 4.76e-02, grad_scale: 16.0 2023-03-07 11:09:13,345 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6022, 4.7563, 4.4518, 4.7025, 3.2842, 4.4257, 5.2081, 5.0493], device='cuda:3'), covar=tensor([0.0807, 0.0474, 0.0206, 0.0231, 0.2151, 0.0351, 0.0092, 0.0096], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0038, 0.0040, 0.0052, 0.0080, 0.0034, 0.0047, 0.0047], device='cuda:3'), out_proj_covar=tensor([2.9917e-05, 2.6247e-05, 2.2382e-05, 3.0548e-05, 5.6133e-05, 2.2403e-05, 2.5495e-05, 2.4088e-05], device='cuda:3') 2023-03-07 11:09:52,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2023-03-07 11:10:02,030 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2023-03-07 11:10:15,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.741e+02 4.246e+02 5.457e+02 7.092e+02 1.687e+03, threshold=1.091e+03, percent-clipped=6.0 2023-03-07 11:10:15,599 INFO [train2.py:809] (3/4) Epoch 1, batch 2400, loss[ctc_loss=0.35, att_loss=0.3846, loss=0.3777, over 16996.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009127, over 51.00 utterances.], tot_loss[ctc_loss=0.3804, att_loss=0.3866, loss=0.3854, over 3285676.25 frames. utt_duration=1272 frames, utt_pad_proportion=0.04477, over 10341.42 utterances.], batch size: 51, lr: 4.75e-02, grad_scale: 16.0 2023-03-07 11:10:15,974 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-07 11:11:30,934 INFO [train2.py:809] (3/4) Epoch 1, batch 2450, loss[ctc_loss=0.3514, att_loss=0.3771, loss=0.372, over 17055.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009311, over 53.00 utterances.], tot_loss[ctc_loss=0.374, att_loss=0.3841, loss=0.3821, over 3288251.63 frames. utt_duration=1265 frames, utt_pad_proportion=0.04484, over 10407.07 utterances.], batch size: 53, lr: 4.74e-02, grad_scale: 16.0 2023-03-07 11:12:36,945 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 11:12:47,386 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.845e+02 4.360e+02 5.388e+02 6.966e+02 1.787e+03, threshold=1.078e+03, percent-clipped=1.0 2023-03-07 11:12:47,431 INFO [train2.py:809] (3/4) Epoch 1, batch 2500, loss[ctc_loss=0.3689, att_loss=0.3938, loss=0.3888, over 17398.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03155, over 63.00 utterances.], tot_loss[ctc_loss=0.3677, att_loss=0.3819, loss=0.3791, over 3282705.69 frames. utt_duration=1254 frames, utt_pad_proportion=0.05051, over 10487.16 utterances.], batch size: 63, lr: 4.73e-02, grad_scale: 16.0 2023-03-07 11:13:06,123 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:13:25,391 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:13:28,345 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0893, 3.8824, 3.9953, 4.3909, 4.4552, 3.9888, 4.4368, 4.5273], device='cuda:3'), covar=tensor([0.0215, 0.0318, 0.0197, 0.0119, 0.0140, 0.0212, 0.0158, 0.0094], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0032, 0.0031, 0.0029, 0.0030, 0.0030, 0.0026, 0.0025], device='cuda:3'), out_proj_covar=tensor([3.0239e-05, 3.0718e-05, 2.8617e-05, 2.4921e-05, 2.4646e-05, 2.8425e-05, 2.1496e-05, 2.1398e-05], device='cuda:3') 2023-03-07 11:13:49,504 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:14:02,698 INFO [train2.py:809] (3/4) Epoch 1, batch 2550, loss[ctc_loss=0.3345, att_loss=0.377, loss=0.3685, over 16108.00 frames. utt_duration=1535 frames, utt_pad_proportion=0.0069, over 42.00 utterances.], tot_loss[ctc_loss=0.3626, att_loss=0.3803, loss=0.3768, over 3288578.16 frames. utt_duration=1260 frames, utt_pad_proportion=0.04617, over 10450.73 utterances.], batch size: 42, lr: 4.72e-02, grad_scale: 16.0 2023-03-07 11:14:19,980 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:14:39,284 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:15:19,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.773e+02 4.230e+02 5.272e+02 6.534e+02 1.107e+03, threshold=1.054e+03, percent-clipped=1.0 2023-03-07 11:15:20,037 INFO [train2.py:809] (3/4) Epoch 1, batch 2600, loss[ctc_loss=0.3282, att_loss=0.3805, loss=0.37, over 17019.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008713, over 51.00 utterances.], tot_loss[ctc_loss=0.3548, att_loss=0.3761, loss=0.3718, over 3288648.93 frames. utt_duration=1266 frames, utt_pad_proportion=0.04494, over 10401.61 utterances.], batch size: 51, lr: 4.71e-02, grad_scale: 16.0 2023-03-07 11:16:12,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.12 vs. limit=2.0 2023-03-07 11:16:37,081 INFO [train2.py:809] (3/4) Epoch 1, batch 2650, loss[ctc_loss=0.3435, att_loss=0.3791, loss=0.372, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.00688, over 46.00 utterances.], tot_loss[ctc_loss=0.3496, att_loss=0.3742, loss=0.3693, over 3289794.72 frames. utt_duration=1255 frames, utt_pad_proportion=0.04742, over 10500.39 utterances.], batch size: 46, lr: 4.70e-02, grad_scale: 16.0 2023-03-07 11:17:29,904 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:17:46,027 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 11:17:53,972 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 4.149e+02 5.114e+02 6.615e+02 1.308e+03, threshold=1.023e+03, percent-clipped=1.0 2023-03-07 11:17:54,017 INFO [train2.py:809] (3/4) Epoch 1, batch 2700, loss[ctc_loss=0.3713, att_loss=0.3702, loss=0.3704, over 15875.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009586, over 39.00 utterances.], tot_loss[ctc_loss=0.3446, att_loss=0.3717, loss=0.3663, over 3287429.65 frames. utt_duration=1281 frames, utt_pad_proportion=0.04341, over 10278.20 utterances.], batch size: 39, lr: 4.69e-02, grad_scale: 16.0 2023-03-07 11:19:03,801 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-07 11:19:03,855 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2218, 4.8934, 5.0842, 4.2194, 4.4141, 4.1642, 4.6663, 4.3827], device='cuda:3'), covar=tensor([0.1082, 0.0359, 0.0549, 0.2695, 0.1800, 0.4033, 0.2143, 0.3353], device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0057, 0.0053, 0.0078, 0.0076, 0.0064, 0.0046, 0.0041], device='cuda:3'), out_proj_covar=tensor([1.9690e-05, 2.2986e-05, 2.2222e-05, 4.9003e-05, 3.9717e-05, 3.8303e-05, 2.8784e-05, 2.4895e-05], device='cuda:3') 2023-03-07 11:19:10,756 INFO [train2.py:809] (3/4) Epoch 1, batch 2750, loss[ctc_loss=0.2826, att_loss=0.3222, loss=0.3143, over 15979.00 frames. utt_duration=1599 frames, utt_pad_proportion=0.009015, over 40.00 utterances.], tot_loss[ctc_loss=0.3419, att_loss=0.3707, loss=0.3649, over 3280435.01 frames. utt_duration=1236 frames, utt_pad_proportion=0.0566, over 10627.16 utterances.], batch size: 40, lr: 4.68e-02, grad_scale: 16.0 2023-03-07 11:20:27,836 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 4.978e+02 5.756e+02 7.064e+02 1.353e+03, threshold=1.151e+03, percent-clipped=3.0 2023-03-07 11:20:27,879 INFO [train2.py:809] (3/4) Epoch 1, batch 2800, loss[ctc_loss=0.3296, att_loss=0.3647, loss=0.3577, over 16468.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006553, over 46.00 utterances.], tot_loss[ctc_loss=0.3384, att_loss=0.3692, loss=0.3631, over 3278168.21 frames. utt_duration=1263 frames, utt_pad_proportion=0.04981, over 10397.66 utterances.], batch size: 46, lr: 4.67e-02, grad_scale: 16.0 2023-03-07 11:21:04,290 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 11:21:07,536 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2023-03-07 11:21:44,977 INFO [train2.py:809] (3/4) Epoch 1, batch 2850, loss[ctc_loss=0.3866, att_loss=0.4038, loss=0.4004, over 17421.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.0292, over 63.00 utterances.], tot_loss[ctc_loss=0.336, att_loss=0.3688, loss=0.3623, over 3284064.11 frames. utt_duration=1262 frames, utt_pad_proportion=0.04801, over 10421.25 utterances.], batch size: 63, lr: 4.66e-02, grad_scale: 16.0 2023-03-07 11:22:36,889 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 11:23:01,458 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.575e+02 4.414e+02 5.747e+02 7.042e+02 1.597e+03, threshold=1.149e+03, percent-clipped=5.0 2023-03-07 11:23:01,502 INFO [train2.py:809] (3/4) Epoch 1, batch 2900, loss[ctc_loss=0.3335, att_loss=0.3795, loss=0.3703, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01097, over 55.00 utterances.], tot_loss[ctc_loss=0.3342, att_loss=0.3682, loss=0.3614, over 3281247.49 frames. utt_duration=1231 frames, utt_pad_proportion=0.05571, over 10679.16 utterances.], batch size: 55, lr: 4.65e-02, grad_scale: 16.0 2023-03-07 11:23:19,881 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8849, 2.6550, 2.7144, 3.6626, 2.5091, 3.2806, 3.0895, 2.8085], device='cuda:3'), covar=tensor([0.0209, 0.0549, 0.1042, 0.0340, 0.0718, 0.0371, 0.0513, 0.0352], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0020, 0.0034, 0.0027, 0.0025, 0.0025, 0.0027, 0.0024], device='cuda:3'), out_proj_covar=tensor([1.9229e-05, 2.0281e-05, 3.2301e-05, 2.1447e-05, 2.1480e-05, 1.8861e-05, 2.1689e-05, 1.9732e-05], device='cuda:3') 2023-03-07 11:23:57,980 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:24:18,943 INFO [train2.py:809] (3/4) Epoch 1, batch 2950, loss[ctc_loss=0.3784, att_loss=0.3951, loss=0.3918, over 14461.00 frames. utt_duration=397.7 frames, utt_pad_proportion=0.3071, over 146.00 utterances.], tot_loss[ctc_loss=0.328, att_loss=0.3647, loss=0.3574, over 3275272.76 frames. utt_duration=1206 frames, utt_pad_proportion=0.06273, over 10878.61 utterances.], batch size: 146, lr: 4.64e-02, grad_scale: 16.0 2023-03-07 11:25:29,093 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 11:25:32,190 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-07 11:25:36,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.432e+02 4.332e+02 5.294e+02 6.689e+02 1.698e+03, threshold=1.059e+03, percent-clipped=3.0 2023-03-07 11:25:36,516 INFO [train2.py:809] (3/4) Epoch 1, batch 3000, loss[ctc_loss=0.346, att_loss=0.4059, loss=0.3939, over 17105.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01602, over 56.00 utterances.], tot_loss[ctc_loss=0.3255, att_loss=0.3639, loss=0.3563, over 3282550.74 frames. utt_duration=1204 frames, utt_pad_proportion=0.06071, over 10917.74 utterances.], batch size: 56, lr: 4.63e-02, grad_scale: 16.0 2023-03-07 11:25:36,516 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 11:25:44,557 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8560, 2.6535, 2.8564, 2.9569, 3.2382, 2.7635, 2.3858, 2.5832], device='cuda:3'), covar=tensor([0.0426, 0.0533, 0.0448, 0.0490, 0.0282, 0.0478, 0.1542, 0.0830], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0031, 0.0035, 0.0031, 0.0029, 0.0031, 0.0049, 0.0039], device='cuda:3'), out_proj_covar=tensor([2.2131e-05, 2.4384e-05, 2.4697e-05, 2.2790e-05, 2.2585e-05, 2.2213e-05, 4.0041e-05, 2.8972e-05], device='cuda:3') 2023-03-07 11:25:51,182 INFO [train2.py:843] (3/4) Epoch 1, validation: ctc_loss=0.2388, att_loss=0.3154, loss=0.3001, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 11:25:51,182 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 14441MB 2023-03-07 11:26:17,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-07 11:26:18,685 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:26:36,793 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.09 vs. limit=2.0 2023-03-07 11:26:51,053 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:26:55,337 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 11:27:07,505 INFO [train2.py:809] (3/4) Epoch 1, batch 3050, loss[ctc_loss=0.2573, att_loss=0.3196, loss=0.3071, over 15632.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.0097, over 37.00 utterances.], tot_loss[ctc_loss=0.3227, att_loss=0.363, loss=0.3549, over 3279684.19 frames. utt_duration=1190 frames, utt_pad_proportion=0.06582, over 11036.27 utterances.], batch size: 37, lr: 4.62e-02, grad_scale: 16.0 2023-03-07 11:27:45,135 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8997, 4.0256, 4.0671, 4.2825, 4.3535, 4.1189, 4.1602, 4.1280], device='cuda:3'), covar=tensor([0.0270, 0.0339, 0.0144, 0.0134, 0.0122, 0.0155, 0.0215, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0032, 0.0032, 0.0027, 0.0028, 0.0030, 0.0028, 0.0027], device='cuda:3'), out_proj_covar=tensor([3.5354e-05, 3.3276e-05, 3.3827e-05, 2.6504e-05, 2.5850e-05, 3.1109e-05, 2.6479e-05, 2.5949e-05], device='cuda:3') 2023-03-07 11:27:52,685 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 11:27:58,567 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3813, 4.6339, 4.7018, 4.8670, 4.9108, 4.8365, 4.6197, 4.7632], device='cuda:3'), covar=tensor([0.0128, 0.0170, 0.0077, 0.0068, 0.0069, 0.0089, 0.0139, 0.0090], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0031, 0.0031, 0.0027, 0.0028, 0.0029, 0.0028, 0.0027], device='cuda:3'), out_proj_covar=tensor([3.5018e-05, 3.3089e-05, 3.3380e-05, 2.6144e-05, 2.5620e-05, 3.0944e-05, 2.6533e-05, 2.5870e-05], device='cuda:3') 2023-03-07 11:28:24,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.428e+02 4.335e+02 5.357e+02 6.400e+02 1.084e+03, threshold=1.071e+03, percent-clipped=0.0 2023-03-07 11:28:24,899 INFO [train2.py:809] (3/4) Epoch 1, batch 3100, loss[ctc_loss=0.3391, att_loss=0.3596, loss=0.3555, over 15785.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007662, over 38.00 utterances.], tot_loss[ctc_loss=0.3204, att_loss=0.3622, loss=0.3538, over 3275714.72 frames. utt_duration=1194 frames, utt_pad_proportion=0.06519, over 10987.39 utterances.], batch size: 38, lr: 4.61e-02, grad_scale: 16.0 2023-03-07 11:29:34,084 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7147, 4.4295, 4.4477, 4.5565, 4.0516, 4.3641, 4.4374, 4.5345], device='cuda:3'), covar=tensor([0.0299, 0.0133, 0.0140, 0.0103, 0.0224, 0.0094, 0.0230, 0.0091], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0056, 0.0062, 0.0051, 0.0068, 0.0058, 0.0068, 0.0051], device='cuda:3'), out_proj_covar=tensor([8.9012e-05, 5.8852e-05, 6.3250e-05, 5.1891e-05, 7.5190e-05, 6.8496e-05, 7.6711e-05, 4.8455e-05], device='cuda:3') 2023-03-07 11:29:41,933 INFO [train2.py:809] (3/4) Epoch 1, batch 3150, loss[ctc_loss=0.3047, att_loss=0.3722, loss=0.3587, over 16963.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007772, over 50.00 utterances.], tot_loss[ctc_loss=0.3162, att_loss=0.3598, loss=0.3511, over 3265536.97 frames. utt_duration=1209 frames, utt_pad_proportion=0.06435, over 10820.69 utterances.], batch size: 50, lr: 4.60e-02, grad_scale: 16.0 2023-03-07 11:30:02,331 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8972, 3.7877, 3.7295, 3.7053, 4.1627, 3.8526, 2.8528, 3.4333], device='cuda:3'), covar=tensor([0.0165, 0.0233, 0.0236, 0.0268, 0.0118, 0.0160, 0.0921, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0032, 0.0039, 0.0030, 0.0031, 0.0032, 0.0053, 0.0043], device='cuda:3'), out_proj_covar=tensor([2.5175e-05, 2.6531e-05, 2.7628e-05, 2.3138e-05, 2.4305e-05, 2.3478e-05, 4.4592e-05, 3.2598e-05], device='cuda:3') 2023-03-07 11:30:25,840 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:30:58,390 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.582e+02 5.665e+02 7.020e+02 1.937e+03, threshold=1.133e+03, percent-clipped=6.0 2023-03-07 11:30:58,433 INFO [train2.py:809] (3/4) Epoch 1, batch 3200, loss[ctc_loss=0.3933, att_loss=0.4081, loss=0.4051, over 14695.00 frames. utt_duration=406.7 frames, utt_pad_proportion=0.2938, over 145.00 utterances.], tot_loss[ctc_loss=0.3173, att_loss=0.3616, loss=0.3527, over 3274229.46 frames. utt_duration=1189 frames, utt_pad_proportion=0.06781, over 11031.81 utterances.], batch size: 145, lr: 4.59e-02, grad_scale: 16.0 2023-03-07 11:31:04,450 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5568, 2.3173, 2.6920, 3.3149, 2.9927, 3.1698, 2.6041, 2.9505], device='cuda:3'), covar=tensor([0.0155, 0.0607, 0.0717, 0.0342, 0.0367, 0.0291, 0.0507, 0.0299], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0022, 0.0036, 0.0028, 0.0025, 0.0025, 0.0031, 0.0026], device='cuda:3'), out_proj_covar=tensor([2.0521e-05, 2.2511e-05, 3.6158e-05, 2.2757e-05, 2.1652e-05, 2.0921e-05, 2.6215e-05, 2.2641e-05], device='cuda:3') 2023-03-07 11:32:13,829 INFO [train2.py:809] (3/4) Epoch 1, batch 3250, loss[ctc_loss=0.2139, att_loss=0.2906, loss=0.2752, over 15390.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.01004, over 35.00 utterances.], tot_loss[ctc_loss=0.3145, att_loss=0.36, loss=0.3509, over 3281512.46 frames. utt_duration=1217 frames, utt_pad_proportion=0.0586, over 10795.23 utterances.], batch size: 35, lr: 4.58e-02, grad_scale: 16.0 2023-03-07 11:33:09,819 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:33:18,489 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:33:29,895 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.939e+02 4.361e+02 5.805e+02 6.999e+02 1.268e+03, threshold=1.161e+03, percent-clipped=3.0 2023-03-07 11:33:29,939 INFO [train2.py:809] (3/4) Epoch 1, batch 3300, loss[ctc_loss=0.2905, att_loss=0.3593, loss=0.3455, over 17389.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03176, over 63.00 utterances.], tot_loss[ctc_loss=0.3096, att_loss=0.357, loss=0.3475, over 3284182.78 frames. utt_duration=1235 frames, utt_pad_proportion=0.05267, over 10645.86 utterances.], batch size: 63, lr: 4.57e-02, grad_scale: 16.0 2023-03-07 11:33:30,203 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6840, 3.9156, 3.8521, 4.0283, 4.2518, 3.9944, 4.0840, 4.0318], device='cuda:3'), covar=tensor([0.0216, 0.0192, 0.0146, 0.0121, 0.0111, 0.0120, 0.0133, 0.0121], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0033, 0.0033, 0.0029, 0.0029, 0.0031, 0.0030, 0.0029], device='cuda:3'), out_proj_covar=tensor([3.8908e-05, 3.6290e-05, 3.6867e-05, 3.0199e-05, 2.8642e-05, 3.4324e-05, 3.0055e-05, 2.9188e-05], device='cuda:3') 2023-03-07 11:33:48,654 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.1486, 1.9965, 2.5038, 2.4964, 2.4263, 2.3739, 2.3922, 2.0249], device='cuda:3'), covar=tensor([0.0533, 0.0519, 0.0339, 0.0392, 0.0423, 0.0758, 0.0403, 0.1443], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0068, 0.0066, 0.0092, 0.0082, 0.0090, 0.0057, 0.0107], device='cuda:3'), out_proj_covar=tensor([5.0496e-05, 5.0881e-05, 4.4679e-05, 5.9611e-05, 4.9419e-05, 8.6543e-05, 4.2622e-05, 8.8419e-05], device='cuda:3') 2023-03-07 11:34:30,984 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:34:43,773 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-07 11:34:46,415 INFO [train2.py:809] (3/4) Epoch 1, batch 3350, loss[ctc_loss=0.2513, att_loss=0.2976, loss=0.2883, over 16176.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006427, over 41.00 utterances.], tot_loss[ctc_loss=0.3076, att_loss=0.3562, loss=0.3465, over 3287092.74 frames. utt_duration=1219 frames, utt_pad_proportion=0.05574, over 10801.01 utterances.], batch size: 41, lr: 4.56e-02, grad_scale: 16.0 2023-03-07 11:35:04,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 11:35:23,104 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:35:44,661 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:36:04,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.598e+02 4.455e+02 5.465e+02 6.622e+02 1.273e+03, threshold=1.093e+03, percent-clipped=2.0 2023-03-07 11:36:04,161 INFO [train2.py:809] (3/4) Epoch 1, batch 3400, loss[ctc_loss=0.2596, att_loss=0.3213, loss=0.3089, over 15873.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01008, over 39.00 utterances.], tot_loss[ctc_loss=0.3042, att_loss=0.3547, loss=0.3446, over 3275624.71 frames. utt_duration=1221 frames, utt_pad_proportion=0.05963, over 10740.95 utterances.], batch size: 39, lr: 4.55e-02, grad_scale: 16.0 2023-03-07 11:36:35,128 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 2023-03-07 11:37:19,748 INFO [train2.py:809] (3/4) Epoch 1, batch 3450, loss[ctc_loss=0.2959, att_loss=0.3465, loss=0.3364, over 16479.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006489, over 46.00 utterances.], tot_loss[ctc_loss=0.3046, att_loss=0.3549, loss=0.3448, over 3268744.12 frames. utt_duration=1190 frames, utt_pad_proportion=0.06932, over 11003.55 utterances.], batch size: 46, lr: 4.54e-02, grad_scale: 16.0 2023-03-07 11:38:03,912 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:38:35,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.427e+02 4.628e+02 6.013e+02 7.728e+02 1.949e+03, threshold=1.203e+03, percent-clipped=6.0 2023-03-07 11:38:35,941 INFO [train2.py:809] (3/4) Epoch 1, batch 3500, loss[ctc_loss=0.2998, att_loss=0.3553, loss=0.3442, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005244, over 47.00 utterances.], tot_loss[ctc_loss=0.3039, att_loss=0.3546, loss=0.3445, over 3267542.12 frames. utt_duration=1185 frames, utt_pad_proportion=0.0707, over 11043.31 utterances.], batch size: 47, lr: 4.53e-02, grad_scale: 16.0 2023-03-07 11:39:08,558 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8697, 1.8031, 2.8497, 3.2393, 3.2877, 2.5578, 2.7771, 1.6635], device='cuda:3'), covar=tensor([0.0807, 0.1421, 0.0410, 0.0390, 0.0466, 0.0901, 0.0933, 0.2576], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0066, 0.0067, 0.0093, 0.0088, 0.0084, 0.0062, 0.0108], device='cuda:3'), out_proj_covar=tensor([5.4802e-05, 4.9577e-05, 4.5471e-05, 6.0359e-05, 5.3512e-05, 7.8708e-05, 4.5525e-05, 8.9960e-05], device='cuda:3') 2023-03-07 11:39:16,982 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:39:45,683 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:39:52,691 INFO [train2.py:809] (3/4) Epoch 1, batch 3550, loss[ctc_loss=0.2805, att_loss=0.3496, loss=0.3357, over 16894.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006171, over 49.00 utterances.], tot_loss[ctc_loss=0.3049, att_loss=0.3545, loss=0.3446, over 3262831.77 frames. utt_duration=1205 frames, utt_pad_proportion=0.06654, over 10845.06 utterances.], batch size: 49, lr: 4.51e-02, grad_scale: 16.0 2023-03-07 11:40:59,022 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:41:10,752 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.449e+02 6.503e+02 7.868e+02 1.011e+03 2.300e+03, threshold=1.574e+03, percent-clipped=15.0 2023-03-07 11:41:10,796 INFO [train2.py:809] (3/4) Epoch 1, batch 3600, loss[ctc_loss=0.3235, att_loss=0.3795, loss=0.3683, over 17316.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01099, over 55.00 utterances.], tot_loss[ctc_loss=0.3034, att_loss=0.3541, loss=0.3439, over 3260748.32 frames. utt_duration=1214 frames, utt_pad_proportion=0.06489, over 10753.61 utterances.], batch size: 55, lr: 4.50e-02, grad_scale: 16.0 2023-03-07 11:41:20,366 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 11:41:40,437 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2262, 4.6666, 4.2822, 4.5075, 4.7426, 4.6149, 4.5486, 4.7723], device='cuda:3'), covar=tensor([0.0138, 0.0221, 0.0148, 0.0162, 0.0116, 0.0137, 0.0145, 0.0146], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0033, 0.0033, 0.0028, 0.0027, 0.0030, 0.0030, 0.0029], device='cuda:3'), out_proj_covar=tensor([3.9213e-05, 3.8252e-05, 3.8846e-05, 3.1870e-05, 2.9287e-05, 3.6232e-05, 3.2790e-05, 3.1589e-05], device='cuda:3') 2023-03-07 11:41:58,293 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5551, 4.1500, 4.1326, 3.7875, 2.0791, 3.0245, 3.6189, 3.3161], device='cuda:3'), covar=tensor([0.3461, 0.1167, 0.1612, 0.2189, 2.3761, 0.4191, 0.2793, 1.0424], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0056, 0.0074, 0.0074, 0.0194, 0.0102, 0.0081, 0.0074], device='cuda:3'), out_proj_covar=tensor([4.6179e-05, 2.8478e-05, 2.8547e-05, 2.9624e-05, 1.1489e-04, 4.9064e-05, 3.2228e-05, 4.7119e-05], device='cuda:3') 2023-03-07 11:42:12,939 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:42:17,625 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 11:42:25,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-03-07 11:42:27,538 INFO [train2.py:809] (3/4) Epoch 1, batch 3650, loss[ctc_loss=0.2415, att_loss=0.3096, loss=0.296, over 15783.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008102, over 38.00 utterances.], tot_loss[ctc_loss=0.3016, att_loss=0.3535, loss=0.3431, over 3269158.81 frames. utt_duration=1226 frames, utt_pad_proportion=0.06139, over 10682.47 utterances.], batch size: 38, lr: 4.49e-02, grad_scale: 16.0 2023-03-07 11:43:05,274 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:43:06,917 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:43:20,772 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1059, 4.6388, 4.2417, 4.1679, 4.8195, 4.5976, 4.4061, 4.3459], device='cuda:3'), covar=tensor([0.0525, 0.0239, 0.0402, 0.0598, 0.0240, 0.0239, 0.0293, 0.0316], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0105, 0.0085, 0.0084, 0.0114, 0.0131, 0.0096, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-07 11:43:47,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.430e+02 5.749e+02 7.107e+02 9.009e+02 1.799e+03, threshold=1.421e+03, percent-clipped=2.0 2023-03-07 11:43:47,136 INFO [train2.py:809] (3/4) Epoch 1, batch 3700, loss[ctc_loss=0.2892, att_loss=0.3488, loss=0.3369, over 16867.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006764, over 49.00 utterances.], tot_loss[ctc_loss=0.2996, att_loss=0.3522, loss=0.3417, over 3264698.02 frames. utt_duration=1219 frames, utt_pad_proportion=0.06401, over 10724.47 utterances.], batch size: 49, lr: 4.48e-02, grad_scale: 16.0 2023-03-07 11:44:03,487 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0011, 3.7309, 4.0391, 4.1107, 2.7049, 4.1369, 3.4934, 4.1158], device='cuda:3'), covar=tensor([0.0444, 0.0399, 0.0305, 0.0577, 0.4616, 0.0392, 0.0756, 0.0257], device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0078, 0.0095, 0.0145, 0.0220, 0.0072, 0.0114, 0.0108], device='cuda:3'), out_proj_covar=tensor([5.1907e-05, 5.3159e-05, 5.4478e-05, 8.9512e-05, 1.4782e-04, 4.3088e-05, 6.4013e-05, 5.3124e-05], device='cuda:3') 2023-03-07 11:44:21,464 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:44:45,178 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-07 11:45:06,156 INFO [train2.py:809] (3/4) Epoch 1, batch 3750, loss[ctc_loss=0.3775, att_loss=0.3964, loss=0.3927, over 17203.00 frames. utt_duration=696.5 frames, utt_pad_proportion=0.1272, over 99.00 utterances.], tot_loss[ctc_loss=0.2986, att_loss=0.3523, loss=0.3415, over 3276539.79 frames. utt_duration=1196 frames, utt_pad_proportion=0.0658, over 10969.25 utterances.], batch size: 99, lr: 4.47e-02, grad_scale: 16.0 2023-03-07 11:45:18,741 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-03-07 11:46:23,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.021e+02 4.717e+02 6.819e+02 8.989e+02 1.986e+03, threshold=1.364e+03, percent-clipped=6.0 2023-03-07 11:46:24,041 INFO [train2.py:809] (3/4) Epoch 1, batch 3800, loss[ctc_loss=0.2943, att_loss=0.3364, loss=0.328, over 15863.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01078, over 39.00 utterances.], tot_loss[ctc_loss=0.2974, att_loss=0.3514, loss=0.3406, over 3272726.16 frames. utt_duration=1219 frames, utt_pad_proportion=0.05936, over 10753.87 utterances.], batch size: 39, lr: 4.46e-02, grad_scale: 16.0 2023-03-07 11:47:11,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-07 11:47:43,585 INFO [train2.py:809] (3/4) Epoch 1, batch 3850, loss[ctc_loss=0.2847, att_loss=0.3435, loss=0.3318, over 16677.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007152, over 46.00 utterances.], tot_loss[ctc_loss=0.2939, att_loss=0.3498, loss=0.3386, over 3269608.85 frames. utt_duration=1251 frames, utt_pad_proportion=0.05198, over 10469.74 utterances.], batch size: 46, lr: 4.45e-02, grad_scale: 16.0 2023-03-07 11:48:57,276 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.62 vs. limit=5.0 2023-03-07 11:49:01,031 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.357e+02 5.503e+02 6.521e+02 8.081e+02 1.434e+03, threshold=1.304e+03, percent-clipped=2.0 2023-03-07 11:49:01,075 INFO [train2.py:809] (3/4) Epoch 1, batch 3900, loss[ctc_loss=0.2741, att_loss=0.355, loss=0.3388, over 17061.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.00865, over 52.00 utterances.], tot_loss[ctc_loss=0.2916, att_loss=0.3484, loss=0.3371, over 3266452.85 frames. utt_duration=1265 frames, utt_pad_proportion=0.04943, over 10343.56 utterances.], batch size: 52, lr: 4.44e-02, grad_scale: 16.0 2023-03-07 11:49:02,686 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 11:49:39,858 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:50:08,120 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:50:18,520 INFO [train2.py:809] (3/4) Epoch 1, batch 3950, loss[ctc_loss=0.2565, att_loss=0.3186, loss=0.3062, over 15883.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009331, over 39.00 utterances.], tot_loss[ctc_loss=0.2896, att_loss=0.3475, loss=0.3359, over 3263204.34 frames. utt_duration=1256 frames, utt_pad_proportion=0.05394, over 10404.02 utterances.], batch size: 39, lr: 4.43e-02, grad_scale: 16.0 2023-03-07 11:50:58,070 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-03-07 11:51:37,721 INFO [train2.py:809] (3/4) Epoch 2, batch 0, loss[ctc_loss=0.2284, att_loss=0.3009, loss=0.2864, over 15778.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008243, over 38.00 utterances.], tot_loss[ctc_loss=0.2284, att_loss=0.3009, loss=0.2864, over 15778.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008243, over 38.00 utterances.], batch size: 38, lr: 4.34e-02, grad_scale: 8.0 2023-03-07 11:51:37,722 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 11:51:49,514 INFO [train2.py:843] (3/4) Epoch 2, validation: ctc_loss=0.1604, att_loss=0.2954, loss=0.2684, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 11:51:49,516 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 15746MB 2023-03-07 11:51:53,001 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:52:01,276 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:52:20,461 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.599e+02 5.286e+02 6.545e+02 8.035e+02 1.268e+03, threshold=1.309e+03, percent-clipped=0.0 2023-03-07 11:53:10,081 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:53:14,642 INFO [train2.py:809] (3/4) Epoch 2, batch 50, loss[ctc_loss=0.2938, att_loss=0.3629, loss=0.3491, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008297, over 51.00 utterances.], tot_loss[ctc_loss=0.279, att_loss=0.3428, loss=0.33, over 734426.84 frames. utt_duration=1282 frames, utt_pad_proportion=0.05345, over 2294.04 utterances.], batch size: 51, lr: 4.33e-02, grad_scale: 8.0 2023-03-07 11:53:45,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-03-07 11:54:37,356 INFO [train2.py:809] (3/4) Epoch 2, batch 100, loss[ctc_loss=0.2936, att_loss=0.3544, loss=0.3422, over 17332.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02187, over 59.00 utterances.], tot_loss[ctc_loss=0.2848, att_loss=0.3472, loss=0.3347, over 1306124.05 frames. utt_duration=1251 frames, utt_pad_proportion=0.04902, over 4180.39 utterances.], batch size: 59, lr: 4.31e-02, grad_scale: 8.0 2023-03-07 11:54:52,795 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3679, 5.0874, 4.6731, 5.0843, 4.9592, 5.1427, 4.8289, 2.8120], device='cuda:3'), covar=tensor([0.1138, 0.0486, 0.0598, 0.0277, 0.0740, 0.0417, 0.0637, 0.7445], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0086, 0.0059, 0.0094, 0.0104, 0.0098, 0.0067, 0.0201], device='cuda:3'), out_proj_covar=tensor([8.5471e-05, 4.0829e-05, 4.0088e-05, 4.2485e-05, 6.7439e-05, 5.3082e-05, 4.3964e-05, 1.2397e-04], device='cuda:3') 2023-03-07 11:55:05,386 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.894e+02 5.347e+02 6.194e+02 8.279e+02 2.356e+03, threshold=1.239e+03, percent-clipped=4.0 2023-03-07 11:55:20,854 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1541, 3.0950, 2.9456, 3.6958, 4.5032, 4.0152, 2.7347, 2.4751], device='cuda:3'), covar=tensor([0.0235, 0.0419, 0.0581, 0.0444, 0.0101, 0.0235, 0.1392, 0.1242], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0046, 0.0059, 0.0042, 0.0036, 0.0051, 0.0079, 0.0073], device='cuda:3'), out_proj_covar=tensor([4.0903e-05, 4.2673e-05, 4.7929e-05, 4.1382e-05, 3.2701e-05, 3.7783e-05, 7.3989e-05, 6.0918e-05], device='cuda:3') 2023-03-07 11:56:00,911 INFO [train2.py:809] (3/4) Epoch 2, batch 150, loss[ctc_loss=0.2583, att_loss=0.3288, loss=0.3147, over 16392.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007606, over 44.00 utterances.], tot_loss[ctc_loss=0.2822, att_loss=0.3446, loss=0.3321, over 1739312.82 frames. utt_duration=1234 frames, utt_pad_proportion=0.05848, over 5646.21 utterances.], batch size: 44, lr: 4.30e-02, grad_scale: 8.0 2023-03-07 11:56:11,623 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:57:24,054 INFO [train2.py:809] (3/4) Epoch 2, batch 200, loss[ctc_loss=0.2991, att_loss=0.3633, loss=0.3504, over 17115.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01426, over 56.00 utterances.], tot_loss[ctc_loss=0.2766, att_loss=0.3401, loss=0.3274, over 2066606.38 frames. utt_duration=1304 frames, utt_pad_proportion=0.04649, over 6346.60 utterances.], batch size: 56, lr: 4.29e-02, grad_scale: 8.0 2023-03-07 11:57:52,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.168e+02 5.150e+02 6.618e+02 8.460e+02 1.655e+03, threshold=1.324e+03, percent-clipped=5.0 2023-03-07 11:57:52,604 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 11:57:52,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 11:58:46,861 INFO [train2.py:809] (3/4) Epoch 2, batch 250, loss[ctc_loss=0.3601, att_loss=0.3946, loss=0.3877, over 17381.00 frames. utt_duration=1009 frames, utt_pad_proportion=0.04934, over 69.00 utterances.], tot_loss[ctc_loss=0.278, att_loss=0.3415, loss=0.3288, over 2334665.99 frames. utt_duration=1253 frames, utt_pad_proportion=0.05801, over 7460.02 utterances.], batch size: 69, lr: 4.28e-02, grad_scale: 8.0 2023-03-07 11:59:11,790 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 11:59:16,707 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3341, 2.5212, 3.1727, 2.6353, 2.6331, 2.5183, 2.4657, 3.0854], device='cuda:3'), covar=tensor([0.0215, 0.0443, 0.0555, 0.0465, 0.0649, 0.0718, 0.0579, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0025, 0.0044, 0.0036, 0.0027, 0.0041, 0.0041, 0.0024], device='cuda:3'), out_proj_covar=tensor([2.9453e-05, 3.1124e-05, 4.9704e-05, 3.3979e-05, 2.9102e-05, 4.3086e-05, 4.0563e-05, 2.6897e-05], device='cuda:3') 2023-03-07 11:59:56,331 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0470, 5.3051, 4.6715, 5.3110, 4.8334, 4.7857, 4.6278, 4.7641], device='cuda:3'), covar=tensor([0.0966, 0.0727, 0.0735, 0.0483, 0.0656, 0.0994, 0.1988, 0.1385], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0226, 0.0190, 0.0167, 0.0156, 0.0231, 0.0248, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-07 12:00:04,040 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:00:08,480 INFO [train2.py:809] (3/4) Epoch 2, batch 300, loss[ctc_loss=0.2983, att_loss=0.3587, loss=0.3466, over 17468.00 frames. utt_duration=886.1 frames, utt_pad_proportion=0.07216, over 79.00 utterances.], tot_loss[ctc_loss=0.2765, att_loss=0.3414, loss=0.3284, over 2544107.39 frames. utt_duration=1281 frames, utt_pad_proportion=0.04929, over 7952.89 utterances.], batch size: 79, lr: 4.27e-02, grad_scale: 8.0 2023-03-07 12:00:14,412 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5294, 2.2589, 3.1637, 3.1517, 2.9102, 2.6797, 2.2299, 2.6885], device='cuda:3'), covar=tensor([0.0178, 0.0528, 0.0428, 0.0309, 0.0316, 0.0550, 0.0624, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0025, 0.0044, 0.0036, 0.0026, 0.0041, 0.0042, 0.0024], device='cuda:3'), out_proj_covar=tensor([2.9174e-05, 3.1178e-05, 5.0491e-05, 3.4911e-05, 2.8316e-05, 4.2579e-05, 4.1072e-05, 2.6937e-05], device='cuda:3') 2023-03-07 12:00:14,422 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:00:36,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.675e+02 5.336e+02 6.631e+02 8.677e+02 1.956e+03, threshold=1.326e+03, percent-clipped=4.0 2023-03-07 12:00:38,502 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7231, 3.5813, 3.6893, 4.6248, 4.7422, 4.5801, 3.2715, 2.4447], device='cuda:3'), covar=tensor([0.0098, 0.0271, 0.0302, 0.0193, 0.0088, 0.0083, 0.0864, 0.1062], device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0051, 0.0061, 0.0044, 0.0038, 0.0052, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([4.3115e-05, 4.7661e-05, 5.0190e-05, 4.3545e-05, 3.6582e-05, 3.9908e-05, 7.9377e-05, 6.7246e-05], device='cuda:3') 2023-03-07 12:01:25,972 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 12:01:30,248 INFO [train2.py:809] (3/4) Epoch 2, batch 350, loss[ctc_loss=0.2823, att_loss=0.3606, loss=0.3449, over 17034.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01072, over 53.00 utterances.], tot_loss[ctc_loss=0.2765, att_loss=0.3418, loss=0.3287, over 2705682.06 frames. utt_duration=1265 frames, utt_pad_proportion=0.05178, over 8565.69 utterances.], batch size: 53, lr: 4.26e-02, grad_scale: 8.0 2023-03-07 12:01:52,916 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:02:42,478 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 12:02:49,848 INFO [train2.py:809] (3/4) Epoch 2, batch 400, loss[ctc_loss=0.2815, att_loss=0.371, loss=0.3531, over 16130.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005467, over 42.00 utterances.], tot_loss[ctc_loss=0.276, att_loss=0.3419, loss=0.3287, over 2831838.74 frames. utt_duration=1270 frames, utt_pad_proportion=0.05039, over 8927.61 utterances.], batch size: 42, lr: 4.25e-02, grad_scale: 8.0 2023-03-07 12:02:57,007 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2505, 4.6617, 4.3796, 4.2487, 4.7896, 4.7307, 4.6164, 4.5967], device='cuda:3'), covar=tensor([0.0519, 0.0269, 0.0357, 0.0487, 0.0295, 0.0233, 0.0221, 0.0263], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0121, 0.0090, 0.0090, 0.0127, 0.0145, 0.0103, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-07 12:03:17,593 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.419e+02 5.648e+02 6.669e+02 8.746e+02 3.020e+03, threshold=1.334e+03, percent-clipped=4.0 2023-03-07 12:04:10,866 INFO [train2.py:809] (3/4) Epoch 2, batch 450, loss[ctc_loss=0.2975, att_loss=0.3594, loss=0.347, over 17116.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01444, over 56.00 utterances.], tot_loss[ctc_loss=0.2769, att_loss=0.3431, loss=0.3299, over 2927673.67 frames. utt_duration=1281 frames, utt_pad_proportion=0.04646, over 9150.78 utterances.], batch size: 56, lr: 4.24e-02, grad_scale: 8.0 2023-03-07 12:04:11,113 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4443, 4.8961, 4.8057, 4.5430, 5.1997, 5.0188, 4.7503, 4.7605], device='cuda:3'), covar=tensor([0.0637, 0.0302, 0.0235, 0.0560, 0.0246, 0.0227, 0.0259, 0.0298], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0123, 0.0090, 0.0093, 0.0127, 0.0143, 0.0105, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-07 12:05:19,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2023-03-07 12:05:25,063 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5843, 5.8979, 5.2622, 5.8585, 5.5182, 5.3422, 5.1509, 5.2163], device='cuda:3'), covar=tensor([0.1012, 0.0658, 0.0573, 0.0462, 0.0534, 0.0947, 0.2253, 0.1542], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0237, 0.0194, 0.0172, 0.0159, 0.0235, 0.0258, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-07 12:05:26,964 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7897, 3.9276, 4.7782, 3.5481, 3.7791, 4.1205, 4.1923, 4.1835], device='cuda:3'), covar=tensor([0.0837, 0.1941, 0.0352, 0.4072, 0.4312, 0.3099, 0.1339, 0.1518], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0126, 0.0085, 0.0134, 0.0158, 0.0096, 0.0062, 0.0063], device='cuda:3'), out_proj_covar=tensor([3.3066e-05, 6.1177e-05, 3.7961e-05, 8.0927e-05, 9.5285e-05, 6.0685e-05, 3.4684e-05, 3.4164e-05], device='cuda:3') 2023-03-07 12:05:33,260 INFO [train2.py:809] (3/4) Epoch 2, batch 500, loss[ctc_loss=0.2579, att_loss=0.3249, loss=0.3115, over 16417.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.00519, over 44.00 utterances.], tot_loss[ctc_loss=0.2756, att_loss=0.3423, loss=0.329, over 2998338.09 frames. utt_duration=1255 frames, utt_pad_proportion=0.05426, over 9571.06 utterances.], batch size: 44, lr: 4.23e-02, grad_scale: 8.0 2023-03-07 12:05:52,586 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:06:00,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.132e+02 4.968e+02 5.977e+02 8.490e+02 1.955e+03, threshold=1.195e+03, percent-clipped=9.0 2023-03-07 12:06:18,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2023-03-07 12:06:53,455 INFO [train2.py:809] (3/4) Epoch 2, batch 550, loss[ctc_loss=0.2473, att_loss=0.3356, loss=0.3179, over 17419.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.03172, over 63.00 utterances.], tot_loss[ctc_loss=0.2743, att_loss=0.3411, loss=0.3278, over 3056117.88 frames. utt_duration=1271 frames, utt_pad_proportion=0.04988, over 9628.96 utterances.], batch size: 63, lr: 4.22e-02, grad_scale: 8.0 2023-03-07 12:07:48,982 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7604, 2.4847, 2.9098, 3.8582, 3.9783, 4.0339, 2.4094, 1.5807], device='cuda:3'), covar=tensor([0.0241, 0.0686, 0.0649, 0.0279, 0.0175, 0.0129, 0.1309, 0.1851], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0061, 0.0070, 0.0049, 0.0041, 0.0055, 0.0085, 0.0088], device='cuda:3'), out_proj_covar=tensor([4.6619e-05, 5.6371e-05, 5.8238e-05, 4.8364e-05, 3.9132e-05, 4.1735e-05, 8.1398e-05, 7.5128e-05], device='cuda:3') 2023-03-07 12:08:08,859 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:08:13,846 INFO [train2.py:809] (3/4) Epoch 2, batch 600, loss[ctc_loss=0.3726, att_loss=0.3842, loss=0.3819, over 13494.00 frames. utt_duration=373.8 frames, utt_pad_proportion=0.351, over 145.00 utterances.], tot_loss[ctc_loss=0.2754, att_loss=0.3413, loss=0.3281, over 3103675.08 frames. utt_duration=1278 frames, utt_pad_proportion=0.04806, over 9728.14 utterances.], batch size: 145, lr: 4.21e-02, grad_scale: 8.0 2023-03-07 12:08:21,755 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-03-07 12:08:40,957 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 4.778e+02 6.448e+02 8.266e+02 2.224e+03, threshold=1.290e+03, percent-clipped=6.0 2023-03-07 12:09:26,945 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:09:35,422 INFO [train2.py:809] (3/4) Epoch 2, batch 650, loss[ctc_loss=0.2447, att_loss=0.3185, loss=0.3038, over 15641.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008423, over 37.00 utterances.], tot_loss[ctc_loss=0.2745, att_loss=0.3404, loss=0.3272, over 3140666.25 frames. utt_duration=1274 frames, utt_pad_proportion=0.04804, over 9875.73 utterances.], batch size: 37, lr: 4.20e-02, grad_scale: 8.0 2023-03-07 12:09:50,188 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:10:56,698 INFO [train2.py:809] (3/4) Epoch 2, batch 700, loss[ctc_loss=0.2647, att_loss=0.3162, loss=0.3059, over 15652.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008181, over 37.00 utterances.], tot_loss[ctc_loss=0.2727, att_loss=0.3389, loss=0.3257, over 3167270.37 frames. utt_duration=1285 frames, utt_pad_proportion=0.04625, over 9872.16 utterances.], batch size: 37, lr: 4.19e-02, grad_scale: 8.0 2023-03-07 12:11:23,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 4.876e+02 6.372e+02 7.940e+02 1.497e+03, threshold=1.274e+03, percent-clipped=2.0 2023-03-07 12:12:04,020 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6578, 2.6260, 3.1460, 3.7498, 3.8352, 3.8049, 2.6052, 2.1867], device='cuda:3'), covar=tensor([0.0161, 0.0536, 0.0419, 0.0165, 0.0126, 0.0109, 0.0898, 0.1164], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0064, 0.0072, 0.0050, 0.0040, 0.0054, 0.0090, 0.0088], device='cuda:3'), out_proj_covar=tensor([4.7924e-05, 5.8600e-05, 6.1020e-05, 5.0900e-05, 3.9659e-05, 4.1850e-05, 8.5633e-05, 7.5697e-05], device='cuda:3') 2023-03-07 12:12:18,129 INFO [train2.py:809] (3/4) Epoch 2, batch 750, loss[ctc_loss=0.2301, att_loss=0.3151, loss=0.2981, over 16175.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006535, over 41.00 utterances.], tot_loss[ctc_loss=0.2705, att_loss=0.3383, loss=0.3247, over 3201601.05 frames. utt_duration=1271 frames, utt_pad_proportion=0.04516, over 10086.79 utterances.], batch size: 41, lr: 4.18e-02, grad_scale: 8.0 2023-03-07 12:13:38,641 INFO [train2.py:809] (3/4) Epoch 2, batch 800, loss[ctc_loss=0.2564, att_loss=0.3404, loss=0.3236, over 16455.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007231, over 46.00 utterances.], tot_loss[ctc_loss=0.2715, att_loss=0.3397, loss=0.3261, over 3227469.46 frames. utt_duration=1254 frames, utt_pad_proportion=0.04747, over 10308.51 utterances.], batch size: 46, lr: 4.17e-02, grad_scale: 8.0 2023-03-07 12:13:57,564 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:14:05,888 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.406e+02 5.666e+02 6.834e+02 8.456e+02 1.435e+03, threshold=1.367e+03, percent-clipped=1.0 2023-03-07 12:14:58,935 INFO [train2.py:809] (3/4) Epoch 2, batch 850, loss[ctc_loss=0.2604, att_loss=0.3392, loss=0.3235, over 16776.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005933, over 48.00 utterances.], tot_loss[ctc_loss=0.2699, att_loss=0.339, loss=0.3252, over 3239797.88 frames. utt_duration=1273 frames, utt_pad_proportion=0.04408, over 10194.62 utterances.], batch size: 48, lr: 4.16e-02, grad_scale: 8.0 2023-03-07 12:15:14,896 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:15:34,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-03-07 12:16:17,035 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-07 12:16:19,356 INFO [train2.py:809] (3/4) Epoch 2, batch 900, loss[ctc_loss=0.2396, att_loss=0.3124, loss=0.2978, over 15815.00 frames. utt_duration=1666 frames, utt_pad_proportion=0.006732, over 38.00 utterances.], tot_loss[ctc_loss=0.2667, att_loss=0.3366, loss=0.3226, over 3247797.29 frames. utt_duration=1304 frames, utt_pad_proportion=0.03797, over 9974.91 utterances.], batch size: 38, lr: 4.15e-02, grad_scale: 8.0 2023-03-07 12:16:46,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.888e+02 4.698e+02 5.859e+02 7.531e+02 1.383e+03, threshold=1.172e+03, percent-clipped=2.0 2023-03-07 12:17:40,696 INFO [train2.py:809] (3/4) Epoch 2, batch 950, loss[ctc_loss=0.3804, att_loss=0.3937, loss=0.391, over 14441.00 frames. utt_duration=397 frames, utt_pad_proportion=0.3071, over 146.00 utterances.], tot_loss[ctc_loss=0.2668, att_loss=0.3369, loss=0.3229, over 3243142.28 frames. utt_duration=1273 frames, utt_pad_proportion=0.04803, over 10201.67 utterances.], batch size: 146, lr: 4.14e-02, grad_scale: 8.0 2023-03-07 12:17:55,079 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:19:01,005 INFO [train2.py:809] (3/4) Epoch 2, batch 1000, loss[ctc_loss=0.2012, att_loss=0.2873, loss=0.2701, over 15368.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01084, over 35.00 utterances.], tot_loss[ctc_loss=0.2646, att_loss=0.3357, loss=0.3215, over 3244704.18 frames. utt_duration=1250 frames, utt_pad_proportion=0.05636, over 10399.06 utterances.], batch size: 35, lr: 4.13e-02, grad_scale: 8.0 2023-03-07 12:19:12,035 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:19:27,689 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.851e+02 4.989e+02 6.201e+02 7.841e+02 1.460e+03, threshold=1.240e+03, percent-clipped=6.0 2023-03-07 12:20:21,708 INFO [train2.py:809] (3/4) Epoch 2, batch 1050, loss[ctc_loss=0.2985, att_loss=0.3608, loss=0.3484, over 16847.00 frames. utt_duration=682.1 frames, utt_pad_proportion=0.1431, over 99.00 utterances.], tot_loss[ctc_loss=0.2646, att_loss=0.3357, loss=0.3215, over 3251048.16 frames. utt_duration=1250 frames, utt_pad_proportion=0.05506, over 10416.90 utterances.], batch size: 99, lr: 4.12e-02, grad_scale: 8.0 2023-03-07 12:20:58,835 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7095, 4.8844, 5.1952, 5.4178, 4.7199, 5.5615, 5.0439, 5.6000], device='cuda:3'), covar=tensor([0.0414, 0.0594, 0.0391, 0.0458, 0.1465, 0.0487, 0.0372, 0.0432], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0189, 0.0160, 0.0186, 0.0288, 0.0186, 0.0150, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 12:21:42,518 INFO [train2.py:809] (3/4) Epoch 2, batch 1100, loss[ctc_loss=0.2724, att_loss=0.3231, loss=0.3129, over 15779.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008118, over 38.00 utterances.], tot_loss[ctc_loss=0.2648, att_loss=0.3359, loss=0.3217, over 3257986.40 frames. utt_duration=1245 frames, utt_pad_proportion=0.05565, over 10483.64 utterances.], batch size: 38, lr: 4.11e-02, grad_scale: 8.0 2023-03-07 12:22:09,662 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.609e+02 5.009e+02 6.335e+02 7.458e+02 1.646e+03, threshold=1.267e+03, percent-clipped=3.0 2023-03-07 12:23:03,364 INFO [train2.py:809] (3/4) Epoch 2, batch 1150, loss[ctc_loss=0.2931, att_loss=0.3548, loss=0.3425, over 17295.00 frames. utt_duration=877.4 frames, utt_pad_proportion=0.08129, over 79.00 utterances.], tot_loss[ctc_loss=0.2645, att_loss=0.3355, loss=0.3213, over 3262691.35 frames. utt_duration=1241 frames, utt_pad_proportion=0.0556, over 10526.97 utterances.], batch size: 79, lr: 4.10e-02, grad_scale: 8.0 2023-03-07 12:24:24,942 INFO [train2.py:809] (3/4) Epoch 2, batch 1200, loss[ctc_loss=0.249, att_loss=0.3274, loss=0.3117, over 16127.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.006095, over 42.00 utterances.], tot_loss[ctc_loss=0.2652, att_loss=0.3363, loss=0.3221, over 3264488.31 frames. utt_duration=1221 frames, utt_pad_proportion=0.06131, over 10704.82 utterances.], batch size: 42, lr: 4.08e-02, grad_scale: 8.0 2023-03-07 12:24:28,251 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9752, 2.0869, 2.1861, 2.3595, 1.9619, 2.5520, 2.6742, 1.4999], device='cuda:3'), covar=tensor([0.0474, 0.1352, 0.0690, 0.0679, 0.1174, 0.0508, 0.0532, 0.2222], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0071, 0.0064, 0.0088, 0.0086, 0.0061, 0.0064, 0.0103], device='cuda:3'), out_proj_covar=tensor([5.1959e-05, 5.4300e-05, 5.0453e-05, 5.5480e-05, 5.1427e-05, 5.0466e-05, 4.6578e-05, 8.4525e-05], device='cuda:3') 2023-03-07 12:24:48,576 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.60 vs. limit=5.0 2023-03-07 12:24:53,171 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.294e+02 5.285e+02 6.236e+02 7.599e+02 1.317e+03, threshold=1.247e+03, percent-clipped=1.0 2023-03-07 12:25:32,867 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-03-07 12:25:45,336 INFO [train2.py:809] (3/4) Epoch 2, batch 1250, loss[ctc_loss=0.2106, att_loss=0.283, loss=0.2685, over 15786.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007851, over 38.00 utterances.], tot_loss[ctc_loss=0.2622, att_loss=0.3338, loss=0.3195, over 3254610.05 frames. utt_duration=1252 frames, utt_pad_proportion=0.05662, over 10413.45 utterances.], batch size: 38, lr: 4.07e-02, grad_scale: 8.0 2023-03-07 12:25:50,326 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9818, 4.2292, 3.9373, 4.3174, 4.3115, 4.1882, 3.9973, 4.2099], device='cuda:3'), covar=tensor([0.0117, 0.0214, 0.0168, 0.0135, 0.0125, 0.0126, 0.0223, 0.0204], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0040, 0.0040, 0.0032, 0.0027, 0.0032, 0.0042, 0.0038], device='cuda:3'), out_proj_covar=tensor([5.3986e-05, 6.0489e-05, 6.6474e-05, 4.8868e-05, 3.9752e-05, 5.2023e-05, 6.1959e-05, 5.6790e-05], device='cuda:3') 2023-03-07 12:26:09,748 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9485, 1.8521, 2.5905, 1.9386, 2.6259, 2.5187, 2.4458, 1.5225], device='cuda:3'), covar=tensor([0.0475, 0.1121, 0.0552, 0.0780, 0.0565, 0.0515, 0.0653, 0.1622], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0072, 0.0068, 0.0093, 0.0089, 0.0064, 0.0068, 0.0104], device='cuda:3'), out_proj_covar=tensor([5.3872e-05, 5.6579e-05, 5.3092e-05, 5.8008e-05, 5.2623e-05, 5.2177e-05, 4.8187e-05, 8.6526e-05], device='cuda:3') 2023-03-07 12:27:05,709 INFO [train2.py:809] (3/4) Epoch 2, batch 1300, loss[ctc_loss=0.2201, att_loss=0.289, loss=0.2752, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.007876, over 39.00 utterances.], tot_loss[ctc_loss=0.2618, att_loss=0.3334, loss=0.319, over 3254535.05 frames. utt_duration=1242 frames, utt_pad_proportion=0.0599, over 10498.05 utterances.], batch size: 39, lr: 4.06e-02, grad_scale: 8.0 2023-03-07 12:27:28,065 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 2023-03-07 12:27:33,503 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.402e+02 4.574e+02 5.700e+02 7.129e+02 3.450e+03, threshold=1.140e+03, percent-clipped=2.0 2023-03-07 12:27:36,751 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9037, 5.4500, 5.5585, 5.8747, 5.0840, 5.8995, 5.2168, 5.8745], device='cuda:3'), covar=tensor([0.0508, 0.0363, 0.0359, 0.0314, 0.1863, 0.0472, 0.0330, 0.0456], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0194, 0.0170, 0.0189, 0.0306, 0.0189, 0.0153, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 12:28:17,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2023-03-07 12:28:19,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2023-03-07 12:28:26,301 INFO [train2.py:809] (3/4) Epoch 2, batch 1350, loss[ctc_loss=0.2417, att_loss=0.3274, loss=0.3102, over 16881.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007023, over 49.00 utterances.], tot_loss[ctc_loss=0.2611, att_loss=0.3334, loss=0.3189, over 3261362.00 frames. utt_duration=1231 frames, utt_pad_proportion=0.05961, over 10613.90 utterances.], batch size: 49, lr: 4.05e-02, grad_scale: 8.0 2023-03-07 12:28:35,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.53 vs. limit=5.0 2023-03-07 12:29:47,500 INFO [train2.py:809] (3/4) Epoch 2, batch 1400, loss[ctc_loss=0.1819, att_loss=0.2749, loss=0.2563, over 15385.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.009824, over 35.00 utterances.], tot_loss[ctc_loss=0.2588, att_loss=0.3321, loss=0.3174, over 3262379.61 frames. utt_duration=1226 frames, utt_pad_proportion=0.06144, over 10661.26 utterances.], batch size: 35, lr: 4.04e-02, grad_scale: 8.0 2023-03-07 12:30:15,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.881e+02 4.556e+02 5.757e+02 7.116e+02 2.241e+03, threshold=1.151e+03, percent-clipped=2.0 2023-03-07 12:30:38,185 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2923, 4.8646, 4.9635, 4.9942, 4.4235, 4.5299, 5.1498, 4.9683], device='cuda:3'), covar=tensor([0.0294, 0.0222, 0.0195, 0.0187, 0.0278, 0.0170, 0.0282, 0.0130], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0083, 0.0090, 0.0069, 0.0096, 0.0080, 0.0091, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 12:30:38,741 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-07 12:30:53,265 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6057, 5.0335, 5.1568, 5.4557, 4.6605, 5.5314, 5.0528, 5.5515], device='cuda:3'), covar=tensor([0.0443, 0.0405, 0.0391, 0.0329, 0.1890, 0.0452, 0.0442, 0.0415], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0196, 0.0169, 0.0191, 0.0311, 0.0191, 0.0159, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 12:31:08,626 INFO [train2.py:809] (3/4) Epoch 2, batch 1450, loss[ctc_loss=0.4045, att_loss=0.4308, loss=0.4255, over 13970.00 frames. utt_duration=384.4 frames, utt_pad_proportion=0.3303, over 146.00 utterances.], tot_loss[ctc_loss=0.2583, att_loss=0.3324, loss=0.3176, over 3266151.62 frames. utt_duration=1233 frames, utt_pad_proportion=0.05839, over 10606.76 utterances.], batch size: 146, lr: 4.03e-02, grad_scale: 8.0 2023-03-07 12:31:53,221 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-03-07 12:32:29,330 INFO [train2.py:809] (3/4) Epoch 2, batch 1500, loss[ctc_loss=0.2064, att_loss=0.2838, loss=0.2683, over 15867.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01045, over 39.00 utterances.], tot_loss[ctc_loss=0.2572, att_loss=0.3325, loss=0.3174, over 3263126.97 frames. utt_duration=1225 frames, utt_pad_proportion=0.06169, over 10672.53 utterances.], batch size: 39, lr: 4.02e-02, grad_scale: 8.0 2023-03-07 12:32:57,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.303e+02 4.697e+02 5.525e+02 6.949e+02 3.924e+03, threshold=1.105e+03, percent-clipped=4.0 2023-03-07 12:33:01,954 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6596, 1.4485, 1.6789, 3.4820, 3.2703, 3.0254, 1.9524, 1.3427], device='cuda:3'), covar=tensor([0.0576, 0.1415, 0.0826, 0.0205, 0.0359, 0.0241, 0.1053, 0.1745], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0069, 0.0066, 0.0087, 0.0088, 0.0063, 0.0072, 0.0103], device='cuda:3'), out_proj_covar=tensor([5.3670e-05, 5.4786e-05, 5.2685e-05, 5.3755e-05, 5.2716e-05, 4.7175e-05, 5.1585e-05, 8.3577e-05], device='cuda:3') 2023-03-07 12:33:07,209 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4063, 4.9313, 5.1187, 5.1394, 4.4764, 4.7837, 5.1226, 5.0826], device='cuda:3'), covar=tensor([0.0297, 0.0218, 0.0153, 0.0125, 0.0352, 0.0138, 0.0328, 0.0119], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0086, 0.0092, 0.0069, 0.0098, 0.0081, 0.0094, 0.0074], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 12:33:07,436 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3806, 3.9331, 5.0804, 3.9641, 3.8709, 4.2236, 4.7238, 4.3370], device='cuda:3'), covar=tensor([0.0314, 0.1439, 0.0245, 0.2002, 0.2696, 0.1284, 0.0256, 0.0610], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0159, 0.0095, 0.0182, 0.0204, 0.0114, 0.0074, 0.0078], device='cuda:3'), out_proj_covar=tensor([4.5369e-05, 9.0054e-05, 5.2920e-05, 1.1985e-04, 1.2824e-04, 7.8716e-05, 4.8150e-05, 5.0836e-05], device='cuda:3') 2023-03-07 12:33:49,179 INFO [train2.py:809] (3/4) Epoch 2, batch 1550, loss[ctc_loss=0.2344, att_loss=0.2963, loss=0.2839, over 15479.00 frames. utt_duration=1721 frames, utt_pad_proportion=0.01014, over 36.00 utterances.], tot_loss[ctc_loss=0.2575, att_loss=0.3328, loss=0.3177, over 3267835.37 frames. utt_duration=1227 frames, utt_pad_proportion=0.06061, over 10666.72 utterances.], batch size: 36, lr: 4.01e-02, grad_scale: 8.0 2023-03-07 12:35:10,351 INFO [train2.py:809] (3/4) Epoch 2, batch 1600, loss[ctc_loss=0.2506, att_loss=0.3385, loss=0.3209, over 16477.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006026, over 46.00 utterances.], tot_loss[ctc_loss=0.2547, att_loss=0.3314, loss=0.316, over 3270014.39 frames. utt_duration=1246 frames, utt_pad_proportion=0.05567, over 10506.25 utterances.], batch size: 46, lr: 4.00e-02, grad_scale: 8.0 2023-03-07 12:35:33,690 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6542, 4.8811, 5.2317, 5.5643, 4.6566, 5.6378, 5.1337, 5.5760], device='cuda:3'), covar=tensor([0.0510, 0.0619, 0.0448, 0.0394, 0.2058, 0.0589, 0.0465, 0.0540], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0205, 0.0177, 0.0198, 0.0325, 0.0196, 0.0162, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 12:35:38,630 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.176e+02 4.466e+02 5.617e+02 6.515e+02 1.895e+03, threshold=1.123e+03, percent-clipped=5.0 2023-03-07 12:36:07,127 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.47 vs. limit=2.0 2023-03-07 12:36:31,276 INFO [train2.py:809] (3/4) Epoch 2, batch 1650, loss[ctc_loss=0.2444, att_loss=0.3232, loss=0.3074, over 16387.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007727, over 44.00 utterances.], tot_loss[ctc_loss=0.255, att_loss=0.3314, loss=0.3161, over 3260700.01 frames. utt_duration=1201 frames, utt_pad_proportion=0.06852, over 10869.34 utterances.], batch size: 44, lr: 3.99e-02, grad_scale: 8.0 2023-03-07 12:36:55,835 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 12:37:03,549 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1286, 4.6694, 4.7275, 4.6942, 4.2161, 4.3691, 4.7690, 4.6589], device='cuda:3'), covar=tensor([0.0293, 0.0153, 0.0150, 0.0136, 0.0298, 0.0151, 0.0304, 0.0114], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0084, 0.0090, 0.0069, 0.0096, 0.0081, 0.0094, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 12:37:47,017 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2733, 4.9225, 5.0474, 5.2232, 4.4948, 5.0023, 4.9931, 5.0234], device='cuda:3'), covar=tensor([0.0354, 0.0264, 0.0198, 0.0131, 0.0270, 0.0097, 0.0445, 0.0135], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0085, 0.0091, 0.0069, 0.0096, 0.0081, 0.0095, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 12:37:50,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.8632, 1.1614, 1.9821, 0.9390, 2.1482, 1.9561, 1.2883, 2.3923], device='cuda:3'), covar=tensor([0.0660, 0.1213, 0.0658, 0.1176, 0.0379, 0.0703, 0.1118, 0.0368], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0064, 0.0062, 0.0074, 0.0060, 0.0071, 0.0067, 0.0082], device='cuda:3'), out_proj_covar=tensor([5.1847e-05, 5.5595e-05, 5.5480e-05, 5.2839e-05, 4.5239e-05, 7.1076e-05, 6.4675e-05, 5.0285e-05], device='cuda:3') 2023-03-07 12:37:51,241 INFO [train2.py:809] (3/4) Epoch 2, batch 1700, loss[ctc_loss=0.2824, att_loss=0.36, loss=0.3445, over 17331.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02218, over 59.00 utterances.], tot_loss[ctc_loss=0.2549, att_loss=0.332, loss=0.3166, over 3266018.36 frames. utt_duration=1204 frames, utt_pad_proportion=0.06617, over 10867.58 utterances.], batch size: 59, lr: 3.98e-02, grad_scale: 8.0 2023-03-07 12:37:55,726 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.10 vs. limit=2.0 2023-03-07 12:38:18,537 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.296e+02 4.646e+02 5.757e+02 7.379e+02 2.321e+03, threshold=1.151e+03, percent-clipped=8.0 2023-03-07 12:38:34,124 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 12:38:34,744 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2023-03-07 12:39:10,812 INFO [train2.py:809] (3/4) Epoch 2, batch 1750, loss[ctc_loss=0.2241, att_loss=0.3235, loss=0.3036, over 17029.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.008073, over 51.00 utterances.], tot_loss[ctc_loss=0.2538, att_loss=0.331, loss=0.3156, over 3268138.64 frames. utt_duration=1223 frames, utt_pad_proportion=0.06131, over 10701.95 utterances.], batch size: 51, lr: 3.97e-02, grad_scale: 8.0 2023-03-07 12:39:44,217 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:40:31,797 INFO [train2.py:809] (3/4) Epoch 2, batch 1800, loss[ctc_loss=0.2424, att_loss=0.3301, loss=0.3126, over 16622.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005648, over 47.00 utterances.], tot_loss[ctc_loss=0.2522, att_loss=0.33, loss=0.3145, over 3278794.52 frames. utt_duration=1241 frames, utt_pad_proportion=0.05389, over 10584.69 utterances.], batch size: 47, lr: 3.96e-02, grad_scale: 8.0 2023-03-07 12:40:59,739 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.935e+02 4.735e+02 5.722e+02 7.362e+02 1.615e+03, threshold=1.144e+03, percent-clipped=5.0 2023-03-07 12:41:23,379 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:41:53,274 INFO [train2.py:809] (3/4) Epoch 2, batch 1850, loss[ctc_loss=0.3093, att_loss=0.3623, loss=0.3517, over 17310.00 frames. utt_duration=1100 frames, utt_pad_proportion=0.03653, over 63.00 utterances.], tot_loss[ctc_loss=0.252, att_loss=0.3297, loss=0.3142, over 3269375.15 frames. utt_duration=1244 frames, utt_pad_proportion=0.05511, over 10529.38 utterances.], batch size: 63, lr: 3.95e-02, grad_scale: 8.0 2023-03-07 12:41:55,929 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-07 12:41:56,819 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9263, 3.2412, 3.4034, 3.0614, 2.8139, 3.3774, 2.1789, 3.9734], device='cuda:3'), covar=tensor([0.0425, 0.0152, 0.0551, 0.0434, 0.0330, 0.0565, 0.0873, 0.0077], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0034, 0.0071, 0.0058, 0.0042, 0.0075, 0.0065, 0.0035], device='cuda:3'), out_proj_covar=tensor([6.1200e-05, 4.9874e-05, 9.8152e-05, 6.7860e-05, 5.6352e-05, 9.7753e-05, 7.5957e-05, 4.6052e-05], device='cuda:3') 2023-03-07 12:42:30,773 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:43:13,674 INFO [train2.py:809] (3/4) Epoch 2, batch 1900, loss[ctc_loss=0.2377, att_loss=0.3341, loss=0.3148, over 16960.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007743, over 50.00 utterances.], tot_loss[ctc_loss=0.2509, att_loss=0.3294, loss=0.3137, over 3271616.82 frames. utt_duration=1249 frames, utt_pad_proportion=0.05291, over 10487.95 utterances.], batch size: 50, lr: 3.95e-02, grad_scale: 8.0 2023-03-07 12:43:41,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.885e+02 5.042e+02 5.993e+02 7.245e+02 1.405e+03, threshold=1.199e+03, percent-clipped=1.0 2023-03-07 12:44:09,842 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:44:35,114 INFO [train2.py:809] (3/4) Epoch 2, batch 1950, loss[ctc_loss=0.28, att_loss=0.3566, loss=0.3413, over 17066.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.00876, over 53.00 utterances.], tot_loss[ctc_loss=0.2508, att_loss=0.3301, loss=0.3142, over 3277024.06 frames. utt_duration=1258 frames, utt_pad_proportion=0.04945, over 10429.43 utterances.], batch size: 53, lr: 3.94e-02, grad_scale: 8.0 2023-03-07 12:45:55,277 INFO [train2.py:809] (3/4) Epoch 2, batch 2000, loss[ctc_loss=0.1953, att_loss=0.2887, loss=0.27, over 15908.00 frames. utt_duration=1633 frames, utt_pad_proportion=0.007789, over 39.00 utterances.], tot_loss[ctc_loss=0.2514, att_loss=0.3304, loss=0.3146, over 3283418.16 frames. utt_duration=1255 frames, utt_pad_proportion=0.04875, over 10480.90 utterances.], batch size: 39, lr: 3.93e-02, grad_scale: 16.0 2023-03-07 12:46:04,109 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:46:05,958 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-07 12:46:26,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.443e+02 4.806e+02 6.010e+02 7.878e+02 2.247e+03, threshold=1.202e+03, percent-clipped=5.0 2023-03-07 12:46:33,531 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 12:47:19,310 INFO [train2.py:809] (3/4) Epoch 2, batch 2050, loss[ctc_loss=0.2432, att_loss=0.3216, loss=0.3059, over 15873.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01008, over 39.00 utterances.], tot_loss[ctc_loss=0.252, att_loss=0.3301, loss=0.3145, over 3278258.49 frames. utt_duration=1254 frames, utt_pad_proportion=0.05183, over 10472.14 utterances.], batch size: 39, lr: 3.92e-02, grad_scale: 8.0 2023-03-07 12:47:33,234 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:47:45,319 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:48:40,080 INFO [train2.py:809] (3/4) Epoch 2, batch 2100, loss[ctc_loss=0.2142, att_loss=0.3059, loss=0.2876, over 16397.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007334, over 44.00 utterances.], tot_loss[ctc_loss=0.251, att_loss=0.3295, loss=0.3138, over 3279933.38 frames. utt_duration=1257 frames, utt_pad_proportion=0.05231, over 10452.19 utterances.], batch size: 44, lr: 3.91e-02, grad_scale: 8.0 2023-03-07 12:48:54,894 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8831, 5.9783, 5.4395, 6.0262, 5.5173, 5.4349, 5.4823, 5.4961], device='cuda:3'), covar=tensor([0.0823, 0.0634, 0.0638, 0.0478, 0.0474, 0.1060, 0.1744, 0.1580], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0256, 0.0214, 0.0190, 0.0166, 0.0256, 0.0278, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 12:49:08,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.112e+02 4.379e+02 5.307e+02 7.391e+02 2.758e+03, threshold=1.061e+03, percent-clipped=4.0 2023-03-07 12:49:10,922 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:49:23,163 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:49:30,518 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-07 12:50:01,355 INFO [train2.py:809] (3/4) Epoch 2, batch 2150, loss[ctc_loss=0.1996, att_loss=0.2918, loss=0.2734, over 16284.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007158, over 43.00 utterances.], tot_loss[ctc_loss=0.2488, att_loss=0.3283, loss=0.3124, over 3266744.69 frames. utt_duration=1255 frames, utt_pad_proportion=0.05532, over 10426.74 utterances.], batch size: 43, lr: 3.90e-02, grad_scale: 8.0 2023-03-07 12:50:01,721 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:50:57,743 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:51:21,402 INFO [train2.py:809] (3/4) Epoch 2, batch 2200, loss[ctc_loss=0.2475, att_loss=0.3173, loss=0.3033, over 16017.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.005905, over 40.00 utterances.], tot_loss[ctc_loss=0.2495, att_loss=0.3287, loss=0.3128, over 3275735.00 frames. utt_duration=1271 frames, utt_pad_proportion=0.04874, over 10321.10 utterances.], batch size: 40, lr: 3.89e-02, grad_scale: 8.0 2023-03-07 12:51:39,659 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:51:41,688 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.51 vs. limit=5.0 2023-03-07 12:51:50,203 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.181e+02 4.728e+02 5.597e+02 7.112e+02 1.691e+03, threshold=1.119e+03, percent-clipped=4.0 2023-03-07 12:51:50,684 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:52:04,332 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:52:06,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-07 12:52:08,956 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:52:23,018 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4749, 3.6076, 3.5158, 2.2973, 3.5864, 4.3987, 4.0016, 3.0292], device='cuda:3'), covar=tensor([0.0213, 0.0500, 0.0682, 0.1532, 0.0601, 0.0105, 0.0444, 0.1450], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0094, 0.0090, 0.0119, 0.0118, 0.0060, 0.0071, 0.0124], device='cuda:3'), out_proj_covar=tensor([8.9761e-05, 8.8641e-05, 1.0188e-04, 1.1015e-04, 1.1325e-04, 6.1045e-05, 8.2892e-05, 1.1603e-04], device='cuda:3') 2023-03-07 12:52:36,125 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:52:42,166 INFO [train2.py:809] (3/4) Epoch 2, batch 2250, loss[ctc_loss=0.221, att_loss=0.3001, loss=0.2843, over 15946.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.006984, over 41.00 utterances.], tot_loss[ctc_loss=0.2473, att_loss=0.3273, loss=0.3113, over 3270541.02 frames. utt_duration=1281 frames, utt_pad_proportion=0.04711, over 10221.44 utterances.], batch size: 41, lr: 3.88e-02, grad_scale: 8.0 2023-03-07 12:53:29,771 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:53:32,827 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:53:42,135 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:53:49,777 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2598, 3.3123, 3.3972, 2.0574, 3.3882, 4.2165, 3.5108, 2.8798], device='cuda:3'), covar=tensor([0.0289, 0.0570, 0.0736, 0.1895, 0.1104, 0.0142, 0.0986, 0.1457], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0097, 0.0096, 0.0122, 0.0127, 0.0062, 0.0074, 0.0126], device='cuda:3'), out_proj_covar=tensor([9.2508e-05, 9.1189e-05, 1.0955e-04, 1.1398e-04, 1.2297e-04, 6.3798e-05, 8.6761e-05, 1.1897e-04], device='cuda:3') 2023-03-07 12:54:02,555 INFO [train2.py:809] (3/4) Epoch 2, batch 2300, loss[ctc_loss=0.2059, att_loss=0.2912, loss=0.2741, over 15406.00 frames. utt_duration=1762 frames, utt_pad_proportion=0.008967, over 35.00 utterances.], tot_loss[ctc_loss=0.2459, att_loss=0.326, loss=0.31, over 3272529.85 frames. utt_duration=1284 frames, utt_pad_proportion=0.04544, over 10203.07 utterances.], batch size: 35, lr: 3.87e-02, grad_scale: 8.0 2023-03-07 12:54:30,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.781e+02 4.301e+02 5.622e+02 7.152e+02 1.270e+03, threshold=1.124e+03, percent-clipped=3.0 2023-03-07 12:54:36,280 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 12:55:06,196 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0458, 3.4472, 4.0938, 4.0967, 1.9291, 4.1843, 2.9624, 4.1731], device='cuda:3'), covar=tensor([0.0175, 0.0328, 0.0452, 0.0272, 0.3920, 0.0170, 0.0885, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0069, 0.0134, 0.0118, 0.0204, 0.0074, 0.0131, 0.0113], device='cuda:3'), out_proj_covar=tensor([5.8735e-05, 5.8544e-05, 1.0098e-04, 8.5469e-05, 1.4069e-04, 6.0105e-05, 9.4574e-05, 7.9182e-05], device='cuda:3') 2023-03-07 12:55:09,242 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:55:21,933 INFO [train2.py:809] (3/4) Epoch 2, batch 2350, loss[ctc_loss=0.2193, att_loss=0.2829, loss=0.2702, over 15506.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.007943, over 36.00 utterances.], tot_loss[ctc_loss=0.2477, att_loss=0.3274, loss=0.3115, over 3275810.42 frames. utt_duration=1257 frames, utt_pad_proportion=0.05164, over 10435.27 utterances.], batch size: 36, lr: 3.86e-02, grad_scale: 8.0 2023-03-07 12:55:39,726 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:55:52,499 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 12:56:41,289 INFO [train2.py:809] (3/4) Epoch 2, batch 2400, loss[ctc_loss=0.2682, att_loss=0.3446, loss=0.3293, over 17369.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.04953, over 69.00 utterances.], tot_loss[ctc_loss=0.2483, att_loss=0.3273, loss=0.3115, over 3271952.63 frames. utt_duration=1256 frames, utt_pad_proportion=0.05288, over 10429.11 utterances.], batch size: 69, lr: 3.85e-02, grad_scale: 8.0 2023-03-07 12:57:03,929 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:57:10,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.496e+02 4.794e+02 5.680e+02 6.939e+02 1.517e+03, threshold=1.136e+03, percent-clipped=5.0 2023-03-07 12:57:24,369 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:57:39,447 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2023-03-07 12:58:03,300 INFO [train2.py:809] (3/4) Epoch 2, batch 2450, loss[ctc_loss=0.2605, att_loss=0.3246, loss=0.3118, over 16632.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004961, over 47.00 utterances.], tot_loss[ctc_loss=0.2479, att_loss=0.3276, loss=0.3117, over 3281782.60 frames. utt_duration=1261 frames, utt_pad_proportion=0.04926, over 10424.70 utterances.], batch size: 47, lr: 3.84e-02, grad_scale: 8.0 2023-03-07 12:58:42,228 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:58:47,198 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:58:54,683 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:59:23,399 INFO [train2.py:809] (3/4) Epoch 2, batch 2500, loss[ctc_loss=0.2204, att_loss=0.2888, loss=0.2751, over 15489.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009568, over 36.00 utterances.], tot_loss[ctc_loss=0.2476, att_loss=0.3269, loss=0.311, over 3275985.71 frames. utt_duration=1271 frames, utt_pad_proportion=0.04743, over 10320.29 utterances.], batch size: 36, lr: 3.83e-02, grad_scale: 8.0 2023-03-07 12:59:33,172 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 12:59:51,987 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 4.550e+02 5.536e+02 6.688e+02 2.770e+03, threshold=1.107e+03, percent-clipped=3.0 2023-03-07 13:00:09,857 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:00:18,572 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-07 13:00:24,400 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:00:29,502 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:00:32,863 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:00:44,097 INFO [train2.py:809] (3/4) Epoch 2, batch 2550, loss[ctc_loss=0.2318, att_loss=0.3246, loss=0.3061, over 16466.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006136, over 46.00 utterances.], tot_loss[ctc_loss=0.2491, att_loss=0.328, loss=0.3122, over 3275607.52 frames. utt_duration=1251 frames, utt_pad_proportion=0.05247, over 10488.06 utterances.], batch size: 46, lr: 3.82e-02, grad_scale: 8.0 2023-03-07 13:01:23,801 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:01:28,552 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:01:36,570 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:02:05,598 INFO [train2.py:809] (3/4) Epoch 2, batch 2600, loss[ctc_loss=0.2362, att_loss=0.3237, loss=0.3062, over 17058.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009282, over 53.00 utterances.], tot_loss[ctc_loss=0.2475, att_loss=0.3277, loss=0.3116, over 3282374.15 frames. utt_duration=1255 frames, utt_pad_proportion=0.04997, over 10477.49 utterances.], batch size: 53, lr: 3.81e-02, grad_scale: 8.0 2023-03-07 13:02:35,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.190e+02 4.654e+02 5.537e+02 6.841e+02 1.492e+03, threshold=1.107e+03, percent-clipped=3.0 2023-03-07 13:02:43,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8991, 4.2128, 4.0539, 4.0958, 4.7485, 3.9715, 3.9450, 2.0975], device='cuda:3'), covar=tensor([0.0589, 0.0558, 0.0413, 0.0315, 0.0926, 0.0444, 0.0949, 0.5189], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0092, 0.0093, 0.0102, 0.0169, 0.0118, 0.0086, 0.0234], device='cuda:3'), out_proj_covar=tensor([9.5090e-05, 6.2961e-05, 6.7324e-05, 6.9132e-05, 1.3859e-04, 7.9749e-05, 6.5210e-05, 1.6556e-04], device='cuda:3') 2023-03-07 13:03:05,307 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:03:16,853 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:03:26,461 INFO [train2.py:809] (3/4) Epoch 2, batch 2650, loss[ctc_loss=0.338, att_loss=0.3857, loss=0.3762, over 14086.00 frames. utt_duration=390 frames, utt_pad_proportion=0.3217, over 145.00 utterances.], tot_loss[ctc_loss=0.2434, att_loss=0.3251, loss=0.3088, over 3281605.37 frames. utt_duration=1267 frames, utt_pad_proportion=0.04785, over 10376.26 utterances.], batch size: 145, lr: 3.80e-02, grad_scale: 8.0 2023-03-07 13:03:44,199 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:04:22,350 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0072, 5.3367, 5.5481, 5.8496, 5.2815, 5.9423, 5.1678, 5.9495], device='cuda:3'), covar=tensor([0.0408, 0.0380, 0.0352, 0.0330, 0.1710, 0.0427, 0.0359, 0.0501], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0223, 0.0203, 0.0219, 0.0363, 0.0200, 0.0174, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 13:04:45,710 INFO [train2.py:809] (3/4) Epoch 2, batch 2700, loss[ctc_loss=0.2091, att_loss=0.3172, loss=0.2956, over 17301.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02298, over 59.00 utterances.], tot_loss[ctc_loss=0.2446, att_loss=0.326, loss=0.3097, over 3285937.43 frames. utt_duration=1245 frames, utt_pad_proportion=0.05202, over 10567.79 utterances.], batch size: 59, lr: 3.79e-02, grad_scale: 8.0 2023-03-07 13:04:54,088 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:05:00,091 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:05:09,218 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:05:15,303 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.955e+02 4.204e+02 5.612e+02 7.162e+02 1.326e+03, threshold=1.122e+03, percent-clipped=2.0 2023-03-07 13:06:06,239 INFO [train2.py:809] (3/4) Epoch 2, batch 2750, loss[ctc_loss=0.2794, att_loss=0.3196, loss=0.3115, over 15510.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008114, over 36.00 utterances.], tot_loss[ctc_loss=0.2442, att_loss=0.3259, loss=0.3095, over 3285196.46 frames. utt_duration=1234 frames, utt_pad_proportion=0.05492, over 10661.98 utterances.], batch size: 36, lr: 3.79e-02, grad_scale: 8.0 2023-03-07 13:06:26,270 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:06:54,601 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3123, 4.9524, 4.3100, 4.8110, 4.7024, 4.5041, 4.4853, 4.7571], device='cuda:3'), covar=tensor([0.0118, 0.0200, 0.0144, 0.0135, 0.0140, 0.0126, 0.0242, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0040, 0.0041, 0.0029, 0.0029, 0.0034, 0.0048, 0.0041], device='cuda:3'), out_proj_covar=tensor([7.0363e-05, 7.4473e-05, 8.7046e-05, 5.7391e-05, 5.3450e-05, 6.8140e-05, 9.1110e-05, 8.1280e-05], device='cuda:3') 2023-03-07 13:07:25,950 INFO [train2.py:809] (3/4) Epoch 2, batch 2800, loss[ctc_loss=0.2149, att_loss=0.3001, loss=0.283, over 15503.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008769, over 36.00 utterances.], tot_loss[ctc_loss=0.2428, att_loss=0.3249, loss=0.3085, over 3277975.81 frames. utt_duration=1239 frames, utt_pad_proportion=0.05408, over 10595.51 utterances.], batch size: 36, lr: 3.78e-02, grad_scale: 8.0 2023-03-07 13:07:36,251 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:07:55,848 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.567e+02 4.477e+02 5.364e+02 6.587e+02 1.505e+03, threshold=1.073e+03, percent-clipped=3.0 2023-03-07 13:08:20,306 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:08:28,516 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:08:33,289 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:08:38,829 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0 2023-03-07 13:08:46,712 INFO [train2.py:809] (3/4) Epoch 2, batch 2850, loss[ctc_loss=0.2389, att_loss=0.3381, loss=0.3182, over 16880.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007023, over 49.00 utterances.], tot_loss[ctc_loss=0.2414, att_loss=0.3239, loss=0.3074, over 3268264.38 frames. utt_duration=1235 frames, utt_pad_proportion=0.05796, over 10596.05 utterances.], batch size: 49, lr: 3.77e-02, grad_scale: 8.0 2023-03-07 13:08:53,172 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:09:13,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.17 vs. limit=2.0 2023-03-07 13:09:25,269 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:09:30,323 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9973, 6.0797, 5.6072, 6.1083, 5.8215, 5.5992, 5.3825, 5.4272], device='cuda:3'), covar=tensor([0.0817, 0.0669, 0.0527, 0.0460, 0.0406, 0.0934, 0.2251, 0.1536], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0270, 0.0225, 0.0201, 0.0171, 0.0274, 0.0285, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 13:09:38,426 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:09:49,553 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:09:56,256 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 13:10:06,818 INFO [train2.py:809] (3/4) Epoch 2, batch 2900, loss[ctc_loss=0.207, att_loss=0.3151, loss=0.2935, over 17321.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02244, over 59.00 utterances.], tot_loss[ctc_loss=0.2416, att_loss=0.324, loss=0.3075, over 3265159.61 frames. utt_duration=1208 frames, utt_pad_proportion=0.06584, over 10824.70 utterances.], batch size: 59, lr: 3.76e-02, grad_scale: 8.0 2023-03-07 13:10:34,466 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 13:10:36,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.536e+02 4.377e+02 5.489e+02 7.218e+02 1.392e+03, threshold=1.098e+03, percent-clipped=5.0 2023-03-07 13:10:42,897 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:10:46,793 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:10:55,827 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:11:07,557 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:11:27,222 INFO [train2.py:809] (3/4) Epoch 2, batch 2950, loss[ctc_loss=0.2235, att_loss=0.2951, loss=0.2808, over 15771.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008636, over 38.00 utterances.], tot_loss[ctc_loss=0.24, att_loss=0.3227, loss=0.3061, over 3260246.45 frames. utt_duration=1216 frames, utt_pad_proportion=0.06581, over 10742.03 utterances.], batch size: 38, lr: 3.75e-02, grad_scale: 8.0 2023-03-07 13:11:35,244 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 13:12:13,077 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4313, 4.7615, 4.5094, 4.8222, 4.9392, 4.4347, 4.5366, 4.8911], device='cuda:3'), covar=tensor([0.0117, 0.0186, 0.0131, 0.0151, 0.0085, 0.0133, 0.0234, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0040, 0.0041, 0.0029, 0.0029, 0.0034, 0.0048, 0.0041], device='cuda:3'), out_proj_covar=tensor([7.1601e-05, 7.6477e-05, 8.9182e-05, 5.9181e-05, 5.3495e-05, 7.0190e-05, 9.2273e-05, 8.1942e-05], device='cuda:3') 2023-03-07 13:12:24,514 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:12:24,795 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:12:47,619 INFO [train2.py:809] (3/4) Epoch 2, batch 3000, loss[ctc_loss=0.2096, att_loss=0.3005, loss=0.2823, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.005931, over 45.00 utterances.], tot_loss[ctc_loss=0.2388, att_loss=0.3218, loss=0.3052, over 3261484.37 frames. utt_duration=1244 frames, utt_pad_proportion=0.05934, over 10499.80 utterances.], batch size: 45, lr: 3.74e-02, grad_scale: 8.0 2023-03-07 13:12:47,619 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 13:13:01,224 INFO [train2.py:843] (3/4) Epoch 2, validation: ctc_loss=0.1245, att_loss=0.2759, loss=0.2456, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 13:13:01,225 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 13:13:01,426 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:13:30,571 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.232e+02 4.393e+02 5.286e+02 7.029e+02 1.971e+03, threshold=1.057e+03, percent-clipped=9.0 2023-03-07 13:14:10,579 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7488, 4.5009, 4.5145, 4.2120, 4.7286, 3.7725, 4.3290, 2.1890], device='cuda:3'), covar=tensor([0.0131, 0.0083, 0.0228, 0.0198, 0.0099, 0.0143, 0.0167, 0.1239], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0028, 0.0028, 0.0040, 0.0031, 0.0033, 0.0039, 0.0071], device='cuda:3'), out_proj_covar=tensor([6.2075e-05, 6.6366e-05, 7.5874e-05, 7.8700e-05, 6.4809e-05, 8.2894e-05, 7.7557e-05, 1.3663e-04], device='cuda:3') 2023-03-07 13:14:20,771 INFO [train2.py:809] (3/4) Epoch 2, batch 3050, loss[ctc_loss=0.3055, att_loss=0.3399, loss=0.333, over 14521.00 frames. utt_duration=1817 frames, utt_pad_proportion=0.03581, over 32.00 utterances.], tot_loss[ctc_loss=0.2401, att_loss=0.3232, loss=0.3066, over 3261922.46 frames. utt_duration=1237 frames, utt_pad_proportion=0.06136, over 10564.99 utterances.], batch size: 32, lr: 3.73e-02, grad_scale: 8.0 2023-03-07 13:14:37,321 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.26 vs. limit=2.0 2023-03-07 13:14:42,948 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 13:15:27,281 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7133, 3.9800, 4.0673, 4.0409, 1.9393, 3.9812, 2.5662, 3.8969], device='cuda:3'), covar=tensor([0.0255, 0.0301, 0.0537, 0.0293, 0.4482, 0.0219, 0.1352, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0074, 0.0146, 0.0114, 0.0213, 0.0080, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([6.6953e-05, 6.4091e-05, 1.1380e-04, 8.5119e-05, 1.4954e-04, 6.6619e-05, 1.0968e-04, 9.1646e-05], device='cuda:3') 2023-03-07 13:15:30,148 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 13:15:38,733 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8364, 4.3182, 4.4404, 4.4526, 3.9436, 4.1520, 4.5973, 4.3789], device='cuda:3'), covar=tensor([0.0312, 0.0204, 0.0213, 0.0139, 0.0322, 0.0185, 0.0264, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0091, 0.0104, 0.0072, 0.0105, 0.0082, 0.0098, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 13:15:41,472 INFO [train2.py:809] (3/4) Epoch 2, batch 3100, loss[ctc_loss=0.2406, att_loss=0.3244, loss=0.3076, over 16970.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007042, over 50.00 utterances.], tot_loss[ctc_loss=0.2384, att_loss=0.3224, loss=0.3056, over 3267992.36 frames. utt_duration=1252 frames, utt_pad_proportion=0.05516, over 10449.62 utterances.], batch size: 50, lr: 3.72e-02, grad_scale: 8.0 2023-03-07 13:16:10,552 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.443e+02 4.796e+02 5.430e+02 6.688e+02 1.495e+03, threshold=1.086e+03, percent-clipped=6.0 2023-03-07 13:16:20,402 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:16:35,033 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:16:42,989 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:17:01,199 INFO [train2.py:809] (3/4) Epoch 2, batch 3150, loss[ctc_loss=0.2632, att_loss=0.3485, loss=0.3314, over 17372.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.0205, over 59.00 utterances.], tot_loss[ctc_loss=0.24, att_loss=0.3232, loss=0.3066, over 3265731.13 frames. utt_duration=1223 frames, utt_pad_proportion=0.06444, over 10696.67 utterances.], batch size: 59, lr: 3.71e-02, grad_scale: 8.0 2023-03-07 13:17:09,094 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0030, 4.4366, 4.5466, 4.5239, 2.0361, 4.2790, 2.6898, 4.7925], device='cuda:3'), covar=tensor([0.0231, 0.0203, 0.0467, 0.0254, 0.4080, 0.0176, 0.1245, 0.0228], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0072, 0.0146, 0.0115, 0.0211, 0.0079, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([6.6942e-05, 6.4157e-05, 1.1413e-04, 8.6409e-05, 1.4921e-04, 6.6104e-05, 1.1063e-04, 9.2332e-05], device='cuda:3') 2023-03-07 13:17:47,338 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:17:51,642 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:17:53,147 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9999, 5.9794, 5.6602, 6.1520, 5.7765, 5.7463, 5.6041, 5.4914], device='cuda:3'), covar=tensor([0.0719, 0.0660, 0.0509, 0.0457, 0.0414, 0.0801, 0.1957, 0.1673], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0272, 0.0222, 0.0198, 0.0176, 0.0269, 0.0286, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 13:17:59,116 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:18:00,752 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9855, 5.9739, 5.6286, 6.0931, 5.8140, 5.7225, 5.5398, 5.5181], device='cuda:3'), covar=tensor([0.0963, 0.0859, 0.0646, 0.0672, 0.0413, 0.0964, 0.2488, 0.2030], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0273, 0.0224, 0.0200, 0.0177, 0.0270, 0.0287, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 13:18:15,453 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1397, 4.7866, 4.9298, 4.7126, 1.9605, 2.8591, 4.8648, 3.7601], device='cuda:3'), covar=tensor([0.0974, 0.0258, 0.0149, 0.0431, 1.1908, 0.2553, 0.0230, 0.3198], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0115, 0.0133, 0.0153, 0.0375, 0.0251, 0.0131, 0.0189], device='cuda:3'), out_proj_covar=tensor([1.0803e-04, 6.0262e-05, 6.4407e-05, 7.4539e-05, 1.8132e-04, 1.2350e-04, 6.4887e-05, 1.0608e-04], device='cuda:3') 2023-03-07 13:18:22,231 INFO [train2.py:809] (3/4) Epoch 2, batch 3200, loss[ctc_loss=0.2261, att_loss=0.3, loss=0.2853, over 16020.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007183, over 40.00 utterances.], tot_loss[ctc_loss=0.2378, att_loss=0.3216, loss=0.3048, over 3269711.78 frames. utt_duration=1262 frames, utt_pad_proportion=0.05475, over 10372.22 utterances.], batch size: 40, lr: 3.71e-02, grad_scale: 8.0 2023-03-07 13:18:50,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 4.512e+02 5.517e+02 7.088e+02 1.562e+03, threshold=1.103e+03, percent-clipped=2.0 2023-03-07 13:19:25,244 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.7696, 3.0353, 3.3895, 3.1042, 2.9945, 3.0448, 1.9031, 3.3624], device='cuda:3'), covar=tensor([0.1230, 0.0412, 0.0619, 0.0534, 0.0437, 0.0825, 0.1048, 0.0126], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0054, 0.0096, 0.0081, 0.0059, 0.0102, 0.0088, 0.0045], device='cuda:3'), out_proj_covar=tensor([1.0919e-04, 8.5497e-05, 1.3892e-04, 1.0650e-04, 9.0930e-05, 1.4576e-04, 1.1265e-04, 6.8639e-05], device='cuda:3') 2023-03-07 13:19:25,282 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 13:19:26,576 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5667, 5.6532, 5.2500, 5.7954, 5.4635, 5.3308, 5.1388, 5.1961], device='cuda:3'), covar=tensor([0.0853, 0.0814, 0.0736, 0.0523, 0.0487, 0.0878, 0.2020, 0.1638], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0273, 0.0229, 0.0200, 0.0183, 0.0283, 0.0289, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 13:19:41,629 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:19:42,925 INFO [train2.py:809] (3/4) Epoch 2, batch 3250, loss[ctc_loss=0.2343, att_loss=0.3175, loss=0.3008, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006231, over 45.00 utterances.], tot_loss[ctc_loss=0.238, att_loss=0.3214, loss=0.3047, over 3267149.54 frames. utt_duration=1252 frames, utt_pad_proportion=0.05817, over 10452.18 utterances.], batch size: 45, lr: 3.70e-02, grad_scale: 8.0 2023-03-07 13:20:29,962 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:20:46,003 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:21:02,605 INFO [train2.py:809] (3/4) Epoch 2, batch 3300, loss[ctc_loss=0.2042, att_loss=0.2846, loss=0.2685, over 15361.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01188, over 35.00 utterances.], tot_loss[ctc_loss=0.2373, att_loss=0.3209, loss=0.3042, over 3263613.76 frames. utt_duration=1234 frames, utt_pad_proportion=0.06266, over 10591.14 utterances.], batch size: 35, lr: 3.69e-02, grad_scale: 8.0 2023-03-07 13:21:02,939 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:21:26,154 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3842, 4.9812, 5.1826, 5.5465, 4.2552, 5.3437, 4.7532, 5.2775], device='cuda:3'), covar=tensor([0.1133, 0.0913, 0.0773, 0.0673, 0.3851, 0.1188, 0.0959, 0.1242], device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0229, 0.0212, 0.0225, 0.0383, 0.0215, 0.0184, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 13:21:30,732 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.392e+02 4.714e+02 6.021e+02 7.352e+02 2.285e+03, threshold=1.204e+03, percent-clipped=4.0 2023-03-07 13:22:20,349 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:22:23,592 INFO [train2.py:809] (3/4) Epoch 2, batch 3350, loss[ctc_loss=0.2406, att_loss=0.342, loss=0.3217, over 17112.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01544, over 56.00 utterances.], tot_loss[ctc_loss=0.2357, att_loss=0.3203, loss=0.3034, over 3260060.52 frames. utt_duration=1224 frames, utt_pad_proportion=0.06507, over 10668.55 utterances.], batch size: 56, lr: 3.68e-02, grad_scale: 8.0 2023-03-07 13:22:25,556 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:23:10,079 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-07 13:23:43,567 INFO [train2.py:809] (3/4) Epoch 2, batch 3400, loss[ctc_loss=0.2319, att_loss=0.3143, loss=0.2979, over 16277.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007356, over 43.00 utterances.], tot_loss[ctc_loss=0.235, att_loss=0.3204, loss=0.3033, over 3266709.65 frames. utt_duration=1235 frames, utt_pad_proportion=0.05961, over 10589.68 utterances.], batch size: 43, lr: 3.67e-02, grad_scale: 8.0 2023-03-07 13:24:12,389 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.987e+02 4.124e+02 5.654e+02 7.446e+02 2.114e+03, threshold=1.131e+03, percent-clipped=4.0 2023-03-07 13:24:14,039 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:24:18,706 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:24:39,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-03-07 13:24:50,020 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-03-07 13:25:03,870 INFO [train2.py:809] (3/4) Epoch 2, batch 3450, loss[ctc_loss=0.2074, att_loss=0.2892, loss=0.2728, over 16402.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007001, over 44.00 utterances.], tot_loss[ctc_loss=0.2343, att_loss=0.3201, loss=0.3029, over 3273317.72 frames. utt_duration=1227 frames, utt_pad_proportion=0.05853, over 10685.98 utterances.], batch size: 44, lr: 3.66e-02, grad_scale: 8.0 2023-03-07 13:25:56,122 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:26:09,788 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.3635, 0.5262, 2.2220, 2.3291, 2.7588, 1.4915, 1.8466, 1.0807], device='cuda:3'), covar=tensor([0.1205, 0.1842, 0.0567, 0.0765, 0.0516, 0.0829, 0.1723, 0.1784], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0059, 0.0052, 0.0077, 0.0062, 0.0053, 0.0070, 0.0081], device='cuda:3'), out_proj_covar=tensor([4.1999e-05, 3.8167e-05, 3.5017e-05, 4.2886e-05, 3.6846e-05, 3.7254e-05, 4.1011e-05, 5.4602e-05], device='cuda:3') 2023-03-07 13:26:22,845 INFO [train2.py:809] (3/4) Epoch 2, batch 3500, loss[ctc_loss=0.212, att_loss=0.3217, loss=0.2998, over 16755.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006471, over 48.00 utterances.], tot_loss[ctc_loss=0.2345, att_loss=0.3201, loss=0.3029, over 3271587.26 frames. utt_duration=1219 frames, utt_pad_proportion=0.06133, over 10750.33 utterances.], batch size: 48, lr: 3.65e-02, grad_scale: 8.0 2023-03-07 13:26:49,881 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.43 vs. limit=5.0 2023-03-07 13:26:50,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 4.119e+02 4.861e+02 5.923e+02 1.212e+03, threshold=9.723e+02, percent-clipped=1.0 2023-03-07 13:27:03,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-03-07 13:27:16,187 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 13:27:39,881 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 13:27:41,016 INFO [train2.py:809] (3/4) Epoch 2, batch 3550, loss[ctc_loss=0.1935, att_loss=0.3001, loss=0.2788, over 15948.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007555, over 41.00 utterances.], tot_loss[ctc_loss=0.235, att_loss=0.3205, loss=0.3034, over 3277258.07 frames. utt_duration=1245 frames, utt_pad_proportion=0.05392, over 10546.19 utterances.], batch size: 41, lr: 3.65e-02, grad_scale: 8.0 2023-03-07 13:28:21,732 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3297, 4.7770, 4.7279, 5.0110, 4.3079, 4.6746, 4.9818, 4.7274], device='cuda:3'), covar=tensor([0.0299, 0.0209, 0.0301, 0.0110, 0.0332, 0.0163, 0.0323, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0094, 0.0108, 0.0073, 0.0112, 0.0086, 0.0101, 0.0082], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 13:28:27,996 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:28:55,956 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 13:29:00,314 INFO [train2.py:809] (3/4) Epoch 2, batch 3600, loss[ctc_loss=0.3098, att_loss=0.3746, loss=0.3616, over 17057.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009354, over 53.00 utterances.], tot_loss[ctc_loss=0.2361, att_loss=0.3213, loss=0.3043, over 3282088.76 frames. utt_duration=1257 frames, utt_pad_proportion=0.05002, over 10454.55 utterances.], batch size: 53, lr: 3.64e-02, grad_scale: 8.0 2023-03-07 13:29:28,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.614e+02 5.143e+02 6.522e+02 9.491e+02 2.921e+03, threshold=1.304e+03, percent-clipped=20.0 2023-03-07 13:29:44,798 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:30:03,095 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:30:14,832 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:30:18,884 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2023-03-07 13:30:20,988 INFO [train2.py:809] (3/4) Epoch 2, batch 3650, loss[ctc_loss=0.1745, att_loss=0.2949, loss=0.2708, over 16350.00 frames. utt_duration=1455 frames, utt_pad_proportion=0.004834, over 45.00 utterances.], tot_loss[ctc_loss=0.2341, att_loss=0.3204, loss=0.3031, over 3284883.71 frames. utt_duration=1261 frames, utt_pad_proportion=0.04827, over 10432.32 utterances.], batch size: 45, lr: 3.63e-02, grad_scale: 8.0 2023-03-07 13:30:44,574 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5323, 4.8978, 4.9548, 4.9442, 4.6222, 4.7690, 5.2308, 4.8746], device='cuda:3'), covar=tensor([0.0272, 0.0219, 0.0202, 0.0164, 0.0287, 0.0150, 0.0226, 0.0133], device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0100, 0.0113, 0.0078, 0.0116, 0.0089, 0.0105, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 13:31:41,708 INFO [train2.py:809] (3/4) Epoch 2, batch 3700, loss[ctc_loss=0.2602, att_loss=0.3435, loss=0.3269, over 17297.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01216, over 55.00 utterances.], tot_loss[ctc_loss=0.2331, att_loss=0.3196, loss=0.3023, over 3279992.81 frames. utt_duration=1237 frames, utt_pad_proportion=0.05693, over 10618.36 utterances.], batch size: 55, lr: 3.62e-02, grad_scale: 8.0 2023-03-07 13:31:42,161 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:31:48,461 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([0.8025, 1.4069, 1.5731, 1.2875, 1.6610, 2.0782, 2.2467, 2.2629], device='cuda:3'), covar=tensor([0.0911, 0.1185, 0.0725, 0.0680, 0.0782, 0.0619, 0.0670, 0.0462], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0074, 0.0064, 0.0061, 0.0077, 0.0075, 0.0069, 0.0083], device='cuda:3'), out_proj_covar=tensor([4.7034e-05, 5.6473e-05, 4.7192e-05, 4.2300e-05, 4.7078e-05, 5.4596e-05, 4.6594e-05, 4.2763e-05], device='cuda:3') 2023-03-07 13:32:06,146 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9784, 2.0860, 2.8297, 3.9911, 4.3492, 4.0264, 2.7918, 1.5974], device='cuda:3'), covar=tensor([0.0280, 0.2318, 0.1073, 0.0403, 0.0084, 0.0247, 0.1610, 0.2718], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0147, 0.0134, 0.0092, 0.0071, 0.0080, 0.0146, 0.0140], device='cuda:3'), out_proj_covar=tensor([8.8323e-05, 1.3483e-04, 1.2659e-04, 1.0235e-04, 7.2605e-05, 7.3620e-05, 1.3945e-04, 1.2897e-04], device='cuda:3') 2023-03-07 13:32:08,319 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.39 vs. limit=2.0 2023-03-07 13:32:10,450 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 4.093e+02 5.003e+02 6.020e+02 1.181e+03, threshold=1.001e+03, percent-clipped=0.0 2023-03-07 13:32:12,290 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 13:32:37,689 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3019, 5.4782, 4.8588, 5.4995, 5.1123, 4.9183, 4.8569, 4.8205], device='cuda:3'), covar=tensor([0.0936, 0.0736, 0.0731, 0.0578, 0.0565, 0.1195, 0.1818, 0.1675], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0265, 0.0226, 0.0207, 0.0178, 0.0279, 0.0283, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 13:33:01,946 INFO [train2.py:809] (3/4) Epoch 2, batch 3750, loss[ctc_loss=0.2283, att_loss=0.3154, loss=0.298, over 15385.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.01022, over 35.00 utterances.], tot_loss[ctc_loss=0.2334, att_loss=0.3197, loss=0.3024, over 3278611.19 frames. utt_duration=1245 frames, utt_pad_proportion=0.05433, over 10546.00 utterances.], batch size: 35, lr: 3.61e-02, grad_scale: 8.0 2023-03-07 13:33:14,935 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:33:17,356 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-07 13:33:29,476 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 13:33:46,967 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:33:53,784 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2023-03-07 13:34:18,217 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.60 vs. limit=2.0 2023-03-07 13:34:21,943 INFO [train2.py:809] (3/4) Epoch 2, batch 3800, loss[ctc_loss=0.2381, att_loss=0.3539, loss=0.3307, over 17020.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007963, over 51.00 utterances.], tot_loss[ctc_loss=0.2321, att_loss=0.3186, loss=0.3013, over 3261981.00 frames. utt_duration=1262 frames, utt_pad_proportion=0.05398, over 10351.17 utterances.], batch size: 51, lr: 3.60e-02, grad_scale: 8.0 2023-03-07 13:34:49,624 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:34:50,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.304e+02 4.201e+02 5.009e+02 6.863e+02 2.448e+03, threshold=1.002e+03, percent-clipped=6.0 2023-03-07 13:34:52,731 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:35:16,899 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 13:35:42,000 INFO [train2.py:809] (3/4) Epoch 2, batch 3850, loss[ctc_loss=0.2838, att_loss=0.3588, loss=0.3438, over 17409.00 frames. utt_duration=1182 frames, utt_pad_proportion=0.01861, over 59.00 utterances.], tot_loss[ctc_loss=0.233, att_loss=0.3194, loss=0.3021, over 3265454.62 frames. utt_duration=1243 frames, utt_pad_proportion=0.05776, over 10518.24 utterances.], batch size: 59, lr: 3.60e-02, grad_scale: 8.0 2023-03-07 13:35:46,692 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9638, 5.1734, 5.5187, 5.7333, 5.1924, 5.8820, 5.2427, 5.9363], device='cuda:3'), covar=tensor([0.0426, 0.0536, 0.0398, 0.0382, 0.1722, 0.0633, 0.0344, 0.0444], device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0229, 0.0212, 0.0238, 0.0392, 0.0215, 0.0181, 0.0256], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 13:36:25,211 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:36:31,180 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:36:58,471 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4789, 4.9643, 4.4029, 4.9392, 5.0250, 4.6269, 4.5603, 4.9024], device='cuda:3'), covar=tensor([0.0088, 0.0127, 0.0130, 0.0080, 0.0075, 0.0101, 0.0213, 0.0132], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0040, 0.0042, 0.0030, 0.0028, 0.0036, 0.0051, 0.0045], device='cuda:3'), out_proj_covar=tensor([8.3470e-05, 8.4692e-05, 1.0221e-04, 6.8400e-05, 5.8292e-05, 8.2158e-05, 1.0936e-04, 1.0026e-04], device='cuda:3') 2023-03-07 13:36:59,756 INFO [train2.py:809] (3/4) Epoch 2, batch 3900, loss[ctc_loss=0.218, att_loss=0.2953, loss=0.2798, over 15884.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009316, over 39.00 utterances.], tot_loss[ctc_loss=0.2325, att_loss=0.3192, loss=0.3019, over 3268800.97 frames. utt_duration=1221 frames, utt_pad_proportion=0.06218, over 10719.72 utterances.], batch size: 39, lr: 3.59e-02, grad_scale: 8.0 2023-03-07 13:37:02,009 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9870, 4.4399, 4.5108, 4.4281, 2.1190, 2.7965, 4.4855, 3.5426], device='cuda:3'), covar=tensor([0.0819, 0.0260, 0.0176, 0.0342, 1.0667, 0.2444, 0.0189, 0.3064], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0137, 0.0152, 0.0172, 0.0412, 0.0279, 0.0144, 0.0222], device='cuda:3'), out_proj_covar=tensor([1.2260e-04, 7.1757e-05, 7.6160e-05, 8.4489e-05, 2.0084e-04, 1.3797e-04, 7.2973e-05, 1.2422e-04], device='cuda:3') 2023-03-07 13:37:27,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 4.818e+02 5.738e+02 6.928e+02 1.942e+03, threshold=1.148e+03, percent-clipped=11.0 2023-03-07 13:37:33,761 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-07 13:38:11,522 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:38:17,769 INFO [train2.py:809] (3/4) Epoch 2, batch 3950, loss[ctc_loss=0.2466, att_loss=0.3326, loss=0.3154, over 16885.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006377, over 49.00 utterances.], tot_loss[ctc_loss=0.2316, att_loss=0.3186, loss=0.3012, over 3277090.64 frames. utt_duration=1236 frames, utt_pad_proportion=0.05592, over 10616.72 utterances.], batch size: 49, lr: 3.58e-02, grad_scale: 8.0 2023-03-07 13:38:54,935 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5647, 3.8364, 3.9356, 4.0193, 3.6443, 3.7309, 4.0727, 3.9599], device='cuda:3'), covar=tensor([0.0294, 0.0233, 0.0222, 0.0146, 0.0309, 0.0237, 0.0261, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0102, 0.0113, 0.0078, 0.0116, 0.0088, 0.0104, 0.0086], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 13:39:31,705 INFO [train2.py:809] (3/4) Epoch 3, batch 0, loss[ctc_loss=0.2174, att_loss=0.3213, loss=0.3005, over 17419.00 frames. utt_duration=883.6 frames, utt_pad_proportion=0.07284, over 79.00 utterances.], tot_loss[ctc_loss=0.2174, att_loss=0.3213, loss=0.3005, over 17419.00 frames. utt_duration=883.6 frames, utt_pad_proportion=0.07284, over 79.00 utterances.], batch size: 79, lr: 3.40e-02, grad_scale: 8.0 2023-03-07 13:39:31,705 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 13:39:44,016 INFO [train2.py:843] (3/4) Epoch 3, validation: ctc_loss=0.122, att_loss=0.2748, loss=0.2442, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 13:39:44,017 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 13:40:00,311 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:40:01,917 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:40:42,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.716e+02 4.111e+02 5.021e+02 6.373e+02 1.388e+03, threshold=1.004e+03, percent-clipped=3.0 2023-03-07 13:41:08,149 INFO [train2.py:809] (3/4) Epoch 3, batch 50, loss[ctc_loss=0.2225, att_loss=0.316, loss=0.2973, over 17403.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03295, over 63.00 utterances.], tot_loss[ctc_loss=0.2223, att_loss=0.3141, loss=0.2957, over 734163.62 frames. utt_duration=1290 frames, utt_pad_proportion=0.04634, over 2278.57 utterances.], batch size: 63, lr: 3.39e-02, grad_scale: 16.0 2023-03-07 13:42:17,934 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:42:27,412 INFO [train2.py:809] (3/4) Epoch 3, batch 100, loss[ctc_loss=0.2074, att_loss=0.3061, loss=0.2864, over 16183.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006699, over 41.00 utterances.], tot_loss[ctc_loss=0.2286, att_loss=0.3154, loss=0.2981, over 1281361.15 frames. utt_duration=1140 frames, utt_pad_proportion=0.09481, over 4501.17 utterances.], batch size: 41, lr: 3.38e-02, grad_scale: 16.0 2023-03-07 13:42:31,288 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-03-07 13:43:04,175 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4896, 4.9166, 5.1152, 5.3113, 4.7594, 5.5164, 5.0553, 5.5003], device='cuda:3'), covar=tensor([0.0623, 0.0592, 0.0480, 0.0426, 0.2088, 0.0568, 0.0501, 0.0624], device='cuda:3'), in_proj_covar=tensor([0.0323, 0.0240, 0.0211, 0.0245, 0.0406, 0.0217, 0.0191, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 13:43:15,553 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:43:21,306 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.756e+02 4.455e+02 5.328e+02 6.355e+02 1.061e+03, threshold=1.066e+03, percent-clipped=2.0 2023-03-07 13:43:33,803 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:43:47,467 INFO [train2.py:809] (3/4) Epoch 3, batch 150, loss[ctc_loss=0.2183, att_loss=0.3215, loss=0.3008, over 16476.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.00682, over 46.00 utterances.], tot_loss[ctc_loss=0.2257, att_loss=0.3146, loss=0.2968, over 1717251.81 frames. utt_duration=1195 frames, utt_pad_proportion=0.07878, over 5757.04 utterances.], batch size: 46, lr: 3.37e-02, grad_scale: 16.0 2023-03-07 13:44:50,078 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:44:51,775 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7028, 3.7217, 3.6319, 3.0607, 3.6231, 3.3944, 2.2074, 4.5780], device='cuda:3'), covar=tensor([0.0918, 0.0234, 0.0769, 0.0641, 0.0388, 0.0693, 0.0957, 0.0063], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0062, 0.0108, 0.0089, 0.0070, 0.0107, 0.0094, 0.0048], device='cuda:3'), out_proj_covar=tensor([1.3191e-04, 1.0276e-04, 1.6127e-04, 1.2147e-04, 1.1604e-04, 1.5805e-04, 1.2612e-04, 7.4906e-05], device='cuda:3') 2023-03-07 13:45:07,280 INFO [train2.py:809] (3/4) Epoch 3, batch 200, loss[ctc_loss=0.2036, att_loss=0.3174, loss=0.2946, over 17022.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008611, over 51.00 utterances.], tot_loss[ctc_loss=0.222, att_loss=0.3134, loss=0.2951, over 2064721.88 frames. utt_duration=1235 frames, utt_pad_proportion=0.06429, over 6697.33 utterances.], batch size: 51, lr: 3.37e-02, grad_scale: 16.0 2023-03-07 13:46:02,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.226e+02 3.842e+02 5.004e+02 6.502e+02 2.003e+03, threshold=1.001e+03, percent-clipped=2.0 2023-03-07 13:46:20,385 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6344, 5.0623, 5.0659, 5.1295, 4.5735, 5.0252, 5.2957, 5.1204], device='cuda:3'), covar=tensor([0.0284, 0.0169, 0.0294, 0.0122, 0.0325, 0.0118, 0.0224, 0.0112], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0100, 0.0116, 0.0078, 0.0116, 0.0088, 0.0108, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 13:46:28,356 INFO [train2.py:809] (3/4) Epoch 3, batch 250, loss[ctc_loss=0.2274, att_loss=0.3183, loss=0.3001, over 16883.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00692, over 49.00 utterances.], tot_loss[ctc_loss=0.2237, att_loss=0.3146, loss=0.2964, over 2331845.18 frames. utt_duration=1203 frames, utt_pad_proportion=0.0707, over 7760.28 utterances.], batch size: 49, lr: 3.36e-02, grad_scale: 16.0 2023-03-07 13:46:30,105 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9623, 1.9503, 3.1039, 4.3225, 4.3418, 4.1859, 2.6578, 1.4402], device='cuda:3'), covar=tensor([0.0332, 0.3119, 0.1306, 0.0445, 0.0228, 0.0188, 0.2054, 0.3115], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0153, 0.0140, 0.0094, 0.0072, 0.0081, 0.0151, 0.0142], device='cuda:3'), out_proj_covar=tensor([8.8923e-05, 1.4076e-04, 1.3286e-04, 1.0565e-04, 7.4442e-05, 7.5041e-05, 1.4567e-04, 1.3022e-04], device='cuda:3') 2023-03-07 13:47:40,621 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:47:46,793 INFO [train2.py:809] (3/4) Epoch 3, batch 300, loss[ctc_loss=0.3366, att_loss=0.3775, loss=0.3693, over 13843.00 frames. utt_duration=378.1 frames, utt_pad_proportion=0.3343, over 147.00 utterances.], tot_loss[ctc_loss=0.2233, att_loss=0.314, loss=0.2958, over 2533612.69 frames. utt_duration=1195 frames, utt_pad_proportion=0.07382, over 8494.21 utterances.], batch size: 147, lr: 3.35e-02, grad_scale: 16.0 2023-03-07 13:48:03,751 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:48:41,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.817e+02 4.123e+02 5.343e+02 7.099e+02 1.515e+03, threshold=1.069e+03, percent-clipped=5.0 2023-03-07 13:49:05,808 INFO [train2.py:809] (3/4) Epoch 3, batch 350, loss[ctc_loss=0.1894, att_loss=0.2982, loss=0.2765, over 17441.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04584, over 69.00 utterances.], tot_loss[ctc_loss=0.2222, att_loss=0.314, loss=0.2956, over 2705555.08 frames. utt_duration=1208 frames, utt_pad_proportion=0.06629, over 8972.33 utterances.], batch size: 69, lr: 3.34e-02, grad_scale: 8.0 2023-03-07 13:49:16,808 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:49:19,541 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:50:22,954 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-03-07 13:50:25,021 INFO [train2.py:809] (3/4) Epoch 3, batch 400, loss[ctc_loss=0.2266, att_loss=0.2962, loss=0.2823, over 15751.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009034, over 38.00 utterances.], tot_loss[ctc_loss=0.2198, att_loss=0.3119, loss=0.2935, over 2826834.85 frames. utt_duration=1233 frames, utt_pad_proportion=0.06221, over 9185.27 utterances.], batch size: 38, lr: 3.34e-02, grad_scale: 8.0 2023-03-07 13:51:12,613 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:51:20,127 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.563e+02 4.039e+02 5.098e+02 6.528e+02 1.596e+03, threshold=1.020e+03, percent-clipped=2.0 2023-03-07 13:51:28,571 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0314, 4.5346, 4.6994, 4.7383, 4.1049, 4.6327, 4.9241, 4.6657], device='cuda:3'), covar=tensor([0.0421, 0.0273, 0.0217, 0.0199, 0.0355, 0.0141, 0.0256, 0.0156], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0102, 0.0118, 0.0079, 0.0118, 0.0089, 0.0108, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 13:51:44,896 INFO [train2.py:809] (3/4) Epoch 3, batch 450, loss[ctc_loss=0.2883, att_loss=0.3607, loss=0.3462, over 16863.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.007684, over 49.00 utterances.], tot_loss[ctc_loss=0.2227, att_loss=0.3139, loss=0.2956, over 2920705.24 frames. utt_duration=1180 frames, utt_pad_proportion=0.07563, over 9910.48 utterances.], batch size: 49, lr: 3.33e-02, grad_scale: 8.0 2023-03-07 13:52:30,027 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:52:47,973 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:53:05,348 INFO [train2.py:809] (3/4) Epoch 3, batch 500, loss[ctc_loss=0.2401, att_loss=0.3296, loss=0.3117, over 17028.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01104, over 53.00 utterances.], tot_loss[ctc_loss=0.2224, att_loss=0.314, loss=0.2957, over 2998265.56 frames. utt_duration=1196 frames, utt_pad_proportion=0.07138, over 10043.00 utterances.], batch size: 53, lr: 3.32e-02, grad_scale: 8.0 2023-03-07 13:53:36,578 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2023-03-07 13:53:46,012 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-03-07 13:54:00,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 4.509e+02 5.776e+02 7.013e+02 1.879e+03, threshold=1.155e+03, percent-clipped=5.0 2023-03-07 13:54:04,275 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:54:24,639 INFO [train2.py:809] (3/4) Epoch 3, batch 550, loss[ctc_loss=0.2243, att_loss=0.3075, loss=0.2909, over 16268.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007361, over 43.00 utterances.], tot_loss[ctc_loss=0.222, att_loss=0.3136, loss=0.2953, over 3057319.79 frames. utt_duration=1196 frames, utt_pad_proportion=0.07189, over 10242.15 utterances.], batch size: 43, lr: 3.31e-02, grad_scale: 8.0 2023-03-07 13:55:45,136 INFO [train2.py:809] (3/4) Epoch 3, batch 600, loss[ctc_loss=0.2358, att_loss=0.3374, loss=0.3171, over 17104.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01515, over 56.00 utterances.], tot_loss[ctc_loss=0.2227, att_loss=0.3138, loss=0.2956, over 3098632.66 frames. utt_duration=1183 frames, utt_pad_proportion=0.07535, over 10488.19 utterances.], batch size: 56, lr: 3.31e-02, grad_scale: 8.0 2023-03-07 13:56:37,860 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-07 13:56:42,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 4.122e+02 4.877e+02 6.146e+02 9.874e+02, threshold=9.754e+02, percent-clipped=0.0 2023-03-07 13:57:05,728 INFO [train2.py:809] (3/4) Epoch 3, batch 650, loss[ctc_loss=0.211, att_loss=0.2934, loss=0.2769, over 14535.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.04416, over 32.00 utterances.], tot_loss[ctc_loss=0.2193, att_loss=0.3117, loss=0.2932, over 3139606.87 frames. utt_duration=1212 frames, utt_pad_proportion=0.06709, over 10371.42 utterances.], batch size: 32, lr: 3.30e-02, grad_scale: 8.0 2023-03-07 13:57:09,595 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 13:58:25,329 INFO [train2.py:809] (3/4) Epoch 3, batch 700, loss[ctc_loss=0.2151, att_loss=0.2888, loss=0.2741, over 15759.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009484, over 38.00 utterances.], tot_loss[ctc_loss=0.2203, att_loss=0.3128, loss=0.2943, over 3171075.02 frames. utt_duration=1217 frames, utt_pad_proportion=0.06632, over 10436.76 utterances.], batch size: 38, lr: 3.29e-02, grad_scale: 8.0 2023-03-07 13:58:44,685 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6390, 2.3660, 3.1387, 4.2243, 4.2820, 4.3260, 2.4836, 1.4802], device='cuda:3'), covar=tensor([0.0514, 0.2340, 0.1174, 0.0470, 0.0163, 0.0163, 0.2111, 0.3156], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0151, 0.0145, 0.0094, 0.0073, 0.0089, 0.0154, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.5829e-05, 1.3962e-04, 1.3854e-04, 1.0617e-04, 7.5713e-05, 8.2571e-05, 1.4946e-04, 1.3320e-04], device='cuda:3') 2023-03-07 13:58:52,560 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1710, 4.9009, 4.8548, 4.8480, 4.9320, 4.9888, 4.9352, 4.7010], device='cuda:3'), covar=tensor([0.1499, 0.0647, 0.0295, 0.0451, 0.0564, 0.0352, 0.0287, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0167, 0.0113, 0.0134, 0.0180, 0.0191, 0.0151, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 13:58:55,780 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8836, 2.6811, 4.9055, 3.4335, 3.3888, 4.2183, 4.4623, 4.4722], device='cuda:3'), covar=tensor([0.0077, 0.1448, 0.0110, 0.1466, 0.2487, 0.0493, 0.0166, 0.0242], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0193, 0.0111, 0.0240, 0.0289, 0.0142, 0.0089, 0.0104], device='cuda:3'), out_proj_covar=tensor([7.6620e-05, 1.3794e-04, 8.0385e-05, 1.8445e-04, 2.0525e-04, 1.1525e-04, 7.0997e-05, 8.3535e-05], device='cuda:3') 2023-03-07 13:59:21,219 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.427e+02 4.229e+02 5.521e+02 7.710e+02 1.687e+03, threshold=1.104e+03, percent-clipped=9.0 2023-03-07 13:59:44,323 INFO [train2.py:809] (3/4) Epoch 3, batch 750, loss[ctc_loss=0.185, att_loss=0.2751, loss=0.2571, over 12719.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.1228, over 28.00 utterances.], tot_loss[ctc_loss=0.2217, att_loss=0.3135, loss=0.2951, over 3189355.29 frames. utt_duration=1200 frames, utt_pad_proportion=0.07077, over 10645.41 utterances.], batch size: 28, lr: 3.29e-02, grad_scale: 8.0 2023-03-07 14:00:51,002 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:01:04,405 INFO [train2.py:809] (3/4) Epoch 3, batch 800, loss[ctc_loss=0.1784, att_loss=0.2813, loss=0.2607, over 15491.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009408, over 36.00 utterances.], tot_loss[ctc_loss=0.2197, att_loss=0.3126, loss=0.2941, over 3215688.32 frames. utt_duration=1229 frames, utt_pad_proportion=0.06063, over 10476.62 utterances.], batch size: 36, lr: 3.28e-02, grad_scale: 8.0 2023-03-07 14:02:01,731 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.534e+02 3.936e+02 4.764e+02 5.876e+02 1.316e+03, threshold=9.528e+02, percent-clipped=1.0 2023-03-07 14:02:24,973 INFO [train2.py:809] (3/4) Epoch 3, batch 850, loss[ctc_loss=0.2105, att_loss=0.3259, loss=0.3028, over 17300.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01126, over 55.00 utterances.], tot_loss[ctc_loss=0.2198, att_loss=0.3127, loss=0.2941, over 3224324.03 frames. utt_duration=1240 frames, utt_pad_proportion=0.05863, over 10409.68 utterances.], batch size: 55, lr: 3.27e-02, grad_scale: 8.0 2023-03-07 14:02:29,346 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:03:19,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4120, 4.9243, 4.8684, 5.0704, 4.4757, 4.7466, 5.1605, 4.8313], device='cuda:3'), covar=tensor([0.0305, 0.0173, 0.0223, 0.0103, 0.0294, 0.0137, 0.0234, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0121, 0.0080, 0.0118, 0.0091, 0.0108, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 14:03:43,843 INFO [train2.py:809] (3/4) Epoch 3, batch 900, loss[ctc_loss=0.3037, att_loss=0.3607, loss=0.3493, over 14473.00 frames. utt_duration=398 frames, utt_pad_proportion=0.3066, over 146.00 utterances.], tot_loss[ctc_loss=0.2205, att_loss=0.313, loss=0.2945, over 3228862.73 frames. utt_duration=1231 frames, utt_pad_proportion=0.06266, over 10502.33 utterances.], batch size: 146, lr: 3.26e-02, grad_scale: 8.0 2023-03-07 14:04:10,255 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-07 14:04:39,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.583e+02 4.192e+02 5.104e+02 6.098e+02 1.636e+03, threshold=1.021e+03, percent-clipped=5.0 2023-03-07 14:04:55,720 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3251, 4.4796, 4.6236, 4.6527, 4.9426, 4.4091, 4.4020, 2.0540], device='cuda:3'), covar=tensor([0.0322, 0.0467, 0.0261, 0.0251, 0.1160, 0.0296, 0.0418, 0.4162], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0108, 0.0101, 0.0108, 0.0192, 0.0125, 0.0092, 0.0239], device='cuda:3'), out_proj_covar=tensor([9.9874e-05, 8.0431e-05, 8.0846e-05, 8.4997e-05, 1.7001e-04, 9.5328e-05, 7.8294e-05, 1.8758e-04], device='cuda:3') 2023-03-07 14:05:03,061 INFO [train2.py:809] (3/4) Epoch 3, batch 950, loss[ctc_loss=0.1889, att_loss=0.2942, loss=0.2731, over 16403.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.00688, over 44.00 utterances.], tot_loss[ctc_loss=0.2172, att_loss=0.3107, loss=0.292, over 3236624.96 frames. utt_duration=1273 frames, utt_pad_proportion=0.05294, over 10183.53 utterances.], batch size: 44, lr: 3.26e-02, grad_scale: 8.0 2023-03-07 14:05:07,407 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:05:16,823 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:06:23,049 INFO [train2.py:809] (3/4) Epoch 3, batch 1000, loss[ctc_loss=0.1978, att_loss=0.3134, loss=0.2903, over 16877.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007752, over 49.00 utterances.], tot_loss[ctc_loss=0.2159, att_loss=0.3101, loss=0.2913, over 3236722.48 frames. utt_duration=1258 frames, utt_pad_proportion=0.05627, over 10307.93 utterances.], batch size: 49, lr: 3.25e-02, grad_scale: 8.0 2023-03-07 14:06:23,862 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:06:50,845 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:06:54,005 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:07:19,491 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.658e+02 4.100e+02 4.728e+02 5.740e+02 1.456e+03, threshold=9.457e+02, percent-clipped=2.0 2023-03-07 14:07:34,349 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:07:43,211 INFO [train2.py:809] (3/4) Epoch 3, batch 1050, loss[ctc_loss=0.1608, att_loss=0.2746, loss=0.2519, over 15989.00 frames. utt_duration=1600 frames, utt_pad_proportion=0.008519, over 40.00 utterances.], tot_loss[ctc_loss=0.2144, att_loss=0.3088, loss=0.2899, over 3235088.80 frames. utt_duration=1264 frames, utt_pad_proportion=0.05649, over 10246.38 utterances.], batch size: 40, lr: 3.24e-02, grad_scale: 8.0 2023-03-07 14:07:46,081 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-07 14:08:27,891 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:09:02,452 INFO [train2.py:809] (3/4) Epoch 3, batch 1100, loss[ctc_loss=0.2494, att_loss=0.3473, loss=0.3277, over 17367.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03308, over 63.00 utterances.], tot_loss[ctc_loss=0.2141, att_loss=0.3093, loss=0.2902, over 3255752.25 frames. utt_duration=1284 frames, utt_pad_proportion=0.04782, over 10155.16 utterances.], batch size: 63, lr: 3.24e-02, grad_scale: 8.0 2023-03-07 14:09:11,154 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:09:24,846 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-07 14:09:58,861 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.446e+02 3.857e+02 5.076e+02 6.311e+02 2.003e+03, threshold=1.015e+03, percent-clipped=7.0 2023-03-07 14:10:13,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-03-07 14:10:18,401 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:10:22,915 INFO [train2.py:809] (3/4) Epoch 3, batch 1150, loss[ctc_loss=0.1997, att_loss=0.2938, loss=0.275, over 16135.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005695, over 42.00 utterances.], tot_loss[ctc_loss=0.2159, att_loss=0.3105, loss=0.2916, over 3257642.57 frames. utt_duration=1236 frames, utt_pad_proportion=0.05979, over 10554.54 utterances.], batch size: 42, lr: 3.23e-02, grad_scale: 8.0 2023-03-07 14:10:58,380 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([0.7288, 1.0355, 1.8161, 2.2032, 0.8315, 1.7442, 2.2944, 2.1846], device='cuda:3'), covar=tensor([0.0821, 0.1028, 0.0646, 0.0465, 0.0927, 0.0734, 0.0595, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0072, 0.0069, 0.0068, 0.0071, 0.0072, 0.0077, 0.0088], device='cuda:3'), out_proj_covar=tensor([3.6612e-05, 5.1207e-05, 4.5292e-05, 4.1742e-05, 4.1607e-05, 4.4109e-05, 4.4692e-05, 4.5103e-05], device='cuda:3') 2023-03-07 14:11:39,915 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3629, 2.6544, 4.8846, 3.7426, 3.2370, 4.2721, 4.7122, 4.4882], device='cuda:3'), covar=tensor([0.0181, 0.1549, 0.0203, 0.1391, 0.2904, 0.0487, 0.0191, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0200, 0.0113, 0.0245, 0.0301, 0.0150, 0.0092, 0.0109], device='cuda:3'), out_proj_covar=tensor([8.1713e-05, 1.4423e-04, 8.2685e-05, 1.9244e-04, 2.1582e-04, 1.2137e-04, 7.5074e-05, 8.8723e-05], device='cuda:3') 2023-03-07 14:11:42,363 INFO [train2.py:809] (3/4) Epoch 3, batch 1200, loss[ctc_loss=0.3116, att_loss=0.3463, loss=0.3394, over 16532.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006051, over 45.00 utterances.], tot_loss[ctc_loss=0.2163, att_loss=0.3103, loss=0.2915, over 3262010.63 frames. utt_duration=1235 frames, utt_pad_proportion=0.05883, over 10579.02 utterances.], batch size: 45, lr: 3.22e-02, grad_scale: 8.0 2023-03-07 14:12:01,840 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8365, 4.7968, 4.5186, 3.9076, 4.6780, 4.1206, 4.5494, 2.5261], device='cuda:3'), covar=tensor([0.0118, 0.0105, 0.0305, 0.0313, 0.0135, 0.0165, 0.0139, 0.1288], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0038, 0.0033, 0.0053, 0.0037, 0.0042, 0.0048, 0.0085], device='cuda:3'), out_proj_covar=tensor([8.8251e-05, 1.0747e-04, 1.0058e-04, 1.3069e-04, 9.9168e-05, 1.2750e-04, 1.1788e-04, 1.9573e-04], device='cuda:3') 2023-03-07 14:12:18,400 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2652, 5.1683, 5.1577, 4.8103, 2.0814, 2.9908, 5.1575, 3.9415], device='cuda:3'), covar=tensor([0.0644, 0.0166, 0.0114, 0.0416, 0.8318, 0.1902, 0.0166, 0.2084], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0133, 0.0147, 0.0171, 0.0398, 0.0282, 0.0145, 0.0224], device='cuda:3'), out_proj_covar=tensor([1.2406e-04, 6.9596e-05, 7.7135e-05, 8.5091e-05, 1.9650e-04, 1.4133e-04, 7.3087e-05, 1.2702e-04], device='cuda:3') 2023-03-07 14:12:39,891 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4553, 4.7260, 4.2929, 4.6400, 4.6553, 4.5102, 4.4774, 4.5197], device='cuda:3'), covar=tensor([0.0104, 0.0177, 0.0138, 0.0139, 0.0105, 0.0115, 0.0267, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0043, 0.0044, 0.0032, 0.0032, 0.0039, 0.0058, 0.0050], device='cuda:3'), out_proj_covar=tensor([1.0240e-04, 1.0407e-04, 1.1874e-04, 8.2632e-05, 7.5111e-05, 1.0212e-04, 1.3925e-04, 1.2515e-04], device='cuda:3') 2023-03-07 14:12:41,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.934e+02 4.184e+02 5.248e+02 6.761e+02 1.156e+03, threshold=1.050e+03, percent-clipped=4.0 2023-03-07 14:13:06,018 INFO [train2.py:809] (3/4) Epoch 3, batch 1250, loss[ctc_loss=0.188, att_loss=0.2858, loss=0.2663, over 15363.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01188, over 35.00 utterances.], tot_loss[ctc_loss=0.2169, att_loss=0.3108, loss=0.2921, over 3263762.73 frames. utt_duration=1207 frames, utt_pad_proportion=0.06528, over 10829.91 utterances.], batch size: 35, lr: 3.22e-02, grad_scale: 8.0 2023-03-07 14:13:54,320 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8738, 5.9292, 5.4380, 6.0112, 5.5997, 5.3546, 5.5128, 5.4132], device='cuda:3'), covar=tensor([0.0863, 0.0821, 0.0635, 0.0592, 0.0524, 0.1204, 0.1848, 0.1699], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0299, 0.0245, 0.0225, 0.0206, 0.0302, 0.0314, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 14:14:30,175 INFO [train2.py:809] (3/4) Epoch 3, batch 1300, loss[ctc_loss=0.1814, att_loss=0.2747, loss=0.256, over 15862.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01076, over 39.00 utterances.], tot_loss[ctc_loss=0.2154, att_loss=0.3096, loss=0.2908, over 3261878.63 frames. utt_duration=1233 frames, utt_pad_proportion=0.05941, over 10593.23 utterances.], batch size: 39, lr: 3.21e-02, grad_scale: 8.0 2023-03-07 14:14:52,965 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:15:29,122 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.817e+02 3.943e+02 4.874e+02 6.147e+02 1.582e+03, threshold=9.748e+02, percent-clipped=2.0 2023-03-07 14:15:49,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7070, 5.2296, 5.0834, 5.1864, 4.7304, 4.9260, 5.5517, 5.2639], device='cuda:3'), covar=tensor([0.0292, 0.0176, 0.0374, 0.0165, 0.0294, 0.0149, 0.0174, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0109, 0.0127, 0.0086, 0.0120, 0.0096, 0.0112, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-07 14:15:54,119 INFO [train2.py:809] (3/4) Epoch 3, batch 1350, loss[ctc_loss=0.1735, att_loss=0.2893, loss=0.2661, over 16410.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006306, over 44.00 utterances.], tot_loss[ctc_loss=0.2139, att_loss=0.3088, loss=0.2898, over 3266004.15 frames. utt_duration=1248 frames, utt_pad_proportion=0.05494, over 10481.68 utterances.], batch size: 44, lr: 3.20e-02, grad_scale: 8.0 2023-03-07 14:16:10,883 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3329, 2.7212, 3.9620, 2.2733, 3.4289, 4.4319, 4.4428, 2.9528], device='cuda:3'), covar=tensor([0.0313, 0.1199, 0.0552, 0.1410, 0.0910, 0.0157, 0.0278, 0.1337], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0147, 0.0125, 0.0145, 0.0157, 0.0086, 0.0102, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 14:16:32,206 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:17:17,854 INFO [train2.py:809] (3/4) Epoch 3, batch 1400, loss[ctc_loss=0.1938, att_loss=0.3102, loss=0.2869, over 17090.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01675, over 56.00 utterances.], tot_loss[ctc_loss=0.2131, att_loss=0.3085, loss=0.2894, over 3260557.31 frames. utt_duration=1226 frames, utt_pad_proportion=0.06336, over 10653.33 utterances.], batch size: 56, lr: 3.20e-02, grad_scale: 8.0 2023-03-07 14:17:18,105 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:18:16,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 3.704e+02 4.754e+02 6.457e+02 1.261e+03, threshold=9.507e+02, percent-clipped=3.0 2023-03-07 14:18:33,982 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:18:37,142 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:18:41,733 INFO [train2.py:809] (3/4) Epoch 3, batch 1450, loss[ctc_loss=0.1872, att_loss=0.3022, loss=0.2792, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005633, over 47.00 utterances.], tot_loss[ctc_loss=0.2129, att_loss=0.3085, loss=0.2894, over 3261631.27 frames. utt_duration=1245 frames, utt_pad_proportion=0.05696, over 10491.73 utterances.], batch size: 47, lr: 3.19e-02, grad_scale: 8.0 2023-03-07 14:19:11,566 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0932, 2.0140, 3.0985, 4.3642, 4.3287, 4.5424, 2.7464, 1.9931], device='cuda:3'), covar=tensor([0.0332, 0.3240, 0.1162, 0.0505, 0.0140, 0.0141, 0.2347, 0.2799], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0162, 0.0153, 0.0104, 0.0079, 0.0093, 0.0163, 0.0152], device='cuda:3'), out_proj_covar=tensor([9.8695e-05, 1.5200e-04, 1.4874e-04, 1.1767e-04, 8.1914e-05, 8.9313e-05, 1.5972e-04, 1.4241e-04], device='cuda:3') 2023-03-07 14:19:17,029 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 2023-03-07 14:19:20,463 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5498, 4.1979, 4.4940, 4.2149, 1.8352, 4.2219, 2.4112, 3.0368], device='cuda:3'), covar=tensor([0.0354, 0.0268, 0.0653, 0.0391, 0.4223, 0.0223, 0.1800, 0.1068], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0087, 0.0184, 0.0116, 0.0225, 0.0093, 0.0177, 0.0160], device='cuda:3'), out_proj_covar=tensor([8.2000e-05, 7.7140e-05, 1.4917e-04, 8.9123e-05, 1.6921e-04, 7.8592e-05, 1.4019e-04, 1.2824e-04], device='cuda:3') 2023-03-07 14:19:58,121 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:20:06,077 INFO [train2.py:809] (3/4) Epoch 3, batch 1500, loss[ctc_loss=0.263, att_loss=0.3277, loss=0.3148, over 16109.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006684, over 42.00 utterances.], tot_loss[ctc_loss=0.211, att_loss=0.3075, loss=0.2882, over 3267349.11 frames. utt_duration=1262 frames, utt_pad_proportion=0.0508, over 10365.97 utterances.], batch size: 42, lr: 3.18e-02, grad_scale: 8.0 2023-03-07 14:20:15,973 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:20:45,731 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0882, 4.4202, 4.1645, 4.6490, 4.7848, 4.6095, 3.9861, 1.8274], device='cuda:3'), covar=tensor([0.0432, 0.0413, 0.0411, 0.0165, 0.1043, 0.0277, 0.0542, 0.4636], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0107, 0.0102, 0.0103, 0.0196, 0.0125, 0.0093, 0.0244], device='cuda:3'), out_proj_covar=tensor([1.0050e-04, 8.4212e-05, 8.4863e-05, 8.6034e-05, 1.7553e-04, 9.7937e-05, 8.2593e-05, 1.9421e-04], device='cuda:3') 2023-03-07 14:21:03,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.770e+02 4.105e+02 5.004e+02 6.164e+02 1.572e+03, threshold=1.001e+03, percent-clipped=6.0 2023-03-07 14:21:14,583 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-07 14:21:27,818 INFO [train2.py:809] (3/4) Epoch 3, batch 1550, loss[ctc_loss=0.2432, att_loss=0.3315, loss=0.3139, over 17030.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01099, over 53.00 utterances.], tot_loss[ctc_loss=0.2123, att_loss=0.3082, loss=0.289, over 3262059.19 frames. utt_duration=1231 frames, utt_pad_proportion=0.06066, over 10612.12 utterances.], batch size: 53, lr: 3.18e-02, grad_scale: 8.0 2023-03-07 14:22:14,736 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7081, 5.9470, 5.2292, 5.8600, 5.4682, 5.3449, 5.2708, 5.2575], device='cuda:3'), covar=tensor([0.1045, 0.0824, 0.0760, 0.0598, 0.0607, 0.1112, 0.2365, 0.2138], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0297, 0.0245, 0.0227, 0.0219, 0.0306, 0.0327, 0.0317], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 14:22:50,227 INFO [train2.py:809] (3/4) Epoch 3, batch 1600, loss[ctc_loss=0.199, att_loss=0.3013, loss=0.2808, over 16402.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006835, over 44.00 utterances.], tot_loss[ctc_loss=0.2112, att_loss=0.3075, loss=0.2883, over 3259526.81 frames. utt_duration=1239 frames, utt_pad_proportion=0.05805, over 10537.32 utterances.], batch size: 44, lr: 3.17e-02, grad_scale: 8.0 2023-03-07 14:22:54,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2023-03-07 14:23:12,987 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:23:13,064 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3792, 1.8028, 2.4346, 1.1542, 1.2485, 2.8850, 1.5051, 1.0276], device='cuda:3'), covar=tensor([0.0575, 0.0803, 0.0628, 0.3933, 0.2226, 0.0367, 0.2950, 0.4089], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0055, 0.0052, 0.0083, 0.0068, 0.0057, 0.0069, 0.0092], device='cuda:3'), out_proj_covar=tensor([3.9372e-05, 3.6735e-05, 3.6579e-05, 4.8573e-05, 4.2632e-05, 3.2328e-05, 4.3318e-05, 5.9265e-05], device='cuda:3') 2023-03-07 14:23:14,685 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:23:29,312 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-07 14:23:42,804 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3356, 2.6181, 4.3086, 3.5069, 3.1432, 3.9975, 4.0979, 4.1553], device='cuda:3'), covar=tensor([0.0100, 0.1382, 0.0166, 0.1551, 0.2739, 0.0595, 0.0275, 0.0369], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0208, 0.0116, 0.0260, 0.0312, 0.0154, 0.0094, 0.0112], device='cuda:3'), out_proj_covar=tensor([8.2870e-05, 1.5255e-04, 8.6492e-05, 2.0370e-04, 2.2469e-04, 1.2520e-04, 7.6956e-05, 9.2371e-05], device='cuda:3') 2023-03-07 14:23:49,431 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.439e+02 4.014e+02 4.915e+02 6.528e+02 1.408e+03, threshold=9.829e+02, percent-clipped=8.0 2023-03-07 14:24:00,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-03-07 14:24:13,623 INFO [train2.py:809] (3/4) Epoch 3, batch 1650, loss[ctc_loss=0.1999, att_loss=0.3138, loss=0.291, over 16973.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007188, over 50.00 utterances.], tot_loss[ctc_loss=0.2134, att_loss=0.3095, loss=0.2902, over 3266047.97 frames. utt_duration=1189 frames, utt_pad_proportion=0.06925, over 11004.86 utterances.], batch size: 50, lr: 3.16e-02, grad_scale: 8.0 2023-03-07 14:24:30,313 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4359, 1.9699, 2.0125, 1.2711, 1.0078, 3.0829, 2.0257, 1.3758], device='cuda:3'), covar=tensor([0.0598, 0.0776, 0.0845, 0.3626, 0.2332, 0.0225, 0.1689, 0.3531], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0054, 0.0052, 0.0081, 0.0065, 0.0054, 0.0067, 0.0089], device='cuda:3'), out_proj_covar=tensor([3.8982e-05, 3.5971e-05, 3.5945e-05, 4.7988e-05, 4.1177e-05, 3.1059e-05, 4.2300e-05, 5.7329e-05], device='cuda:3') 2023-03-07 14:24:33,196 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:24:52,634 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:24:57,358 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5485, 3.8876, 3.8304, 3.8004, 4.0105, 3.8635, 3.8533, 3.7502], device='cuda:3'), covar=tensor([0.1245, 0.0604, 0.0302, 0.0526, 0.0344, 0.0414, 0.0289, 0.0355], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0177, 0.0120, 0.0142, 0.0188, 0.0208, 0.0161, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003], device='cuda:3') 2023-03-07 14:24:57,556 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:25:18,000 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:25:35,210 INFO [train2.py:809] (3/4) Epoch 3, batch 1700, loss[ctc_loss=0.2357, att_loss=0.325, loss=0.3071, over 16483.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006489, over 46.00 utterances.], tot_loss[ctc_loss=0.2135, att_loss=0.3098, loss=0.2906, over 3260913.51 frames. utt_duration=1194 frames, utt_pad_proportion=0.06909, over 10936.83 utterances.], batch size: 46, lr: 3.16e-02, grad_scale: 8.0 2023-03-07 14:25:35,545 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:25:56,183 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0238, 4.9081, 4.6915, 3.9172, 4.8860, 4.1636, 4.2253, 2.5949], device='cuda:3'), covar=tensor([0.0109, 0.0079, 0.0275, 0.0359, 0.0079, 0.0140, 0.0222, 0.1306], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0040, 0.0035, 0.0059, 0.0038, 0.0045, 0.0052, 0.0088], device='cuda:3'), out_proj_covar=tensor([9.5449e-05, 1.1618e-04, 1.1096e-04, 1.4886e-04, 1.0334e-04, 1.4041e-04, 1.3401e-04, 2.0998e-04], device='cuda:3') 2023-03-07 14:26:07,314 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2255, 4.5521, 4.8054, 4.9594, 4.3548, 5.1517, 4.6269, 5.2093], device='cuda:3'), covar=tensor([0.0505, 0.0590, 0.0499, 0.0486, 0.2032, 0.0523, 0.0726, 0.0483], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0251, 0.0242, 0.0283, 0.0428, 0.0232, 0.0206, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 14:26:09,530 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:26:33,374 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.300e+02 3.813e+02 4.618e+02 5.851e+02 1.256e+03, threshold=9.236e+02, percent-clipped=5.0 2023-03-07 14:26:54,264 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:26:57,505 INFO [train2.py:809] (3/4) Epoch 3, batch 1750, loss[ctc_loss=0.2387, att_loss=0.3307, loss=0.3123, over 17303.00 frames. utt_duration=1100 frames, utt_pad_proportion=0.03741, over 63.00 utterances.], tot_loss[ctc_loss=0.2139, att_loss=0.3101, loss=0.2909, over 3264405.94 frames. utt_duration=1195 frames, utt_pad_proportion=0.06801, over 10939.74 utterances.], batch size: 63, lr: 3.15e-02, grad_scale: 8.0 2023-03-07 14:26:57,952 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:27:37,581 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:27:37,695 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1831, 5.0265, 5.0951, 5.0064, 1.8608, 1.9737, 5.1065, 3.9485], device='cuda:3'), covar=tensor([0.0593, 0.0191, 0.0118, 0.0234, 0.9975, 0.3186, 0.0170, 0.2114], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0134, 0.0154, 0.0176, 0.0406, 0.0289, 0.0155, 0.0236], device='cuda:3'), out_proj_covar=tensor([1.2912e-04, 7.1109e-05, 7.9759e-05, 8.6836e-05, 1.9977e-04, 1.4555e-04, 7.8681e-05, 1.3429e-04], device='cuda:3') 2023-03-07 14:28:11,535 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9672, 4.2864, 4.5557, 4.7091, 1.7062, 4.2866, 2.3311, 2.9795], device='cuda:3'), covar=tensor([0.0304, 0.0207, 0.0540, 0.0171, 0.4178, 0.0194, 0.1673, 0.1040], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0082, 0.0189, 0.0107, 0.0228, 0.0091, 0.0179, 0.0159], device='cuda:3'), out_proj_covar=tensor([8.1659e-05, 7.4860e-05, 1.5352e-04, 8.4711e-05, 1.7233e-04, 8.0395e-05, 1.4210e-04, 1.2849e-04], device='cuda:3') 2023-03-07 14:28:19,376 INFO [train2.py:809] (3/4) Epoch 3, batch 1800, loss[ctc_loss=0.1983, att_loss=0.3082, loss=0.2862, over 16638.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004781, over 47.00 utterances.], tot_loss[ctc_loss=0.2113, att_loss=0.3083, loss=0.2889, over 3256821.13 frames. utt_duration=1228 frames, utt_pad_proportion=0.06137, over 10620.50 utterances.], batch size: 47, lr: 3.14e-02, grad_scale: 8.0 2023-03-07 14:28:21,133 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:29:11,675 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-03-07 14:29:17,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 4.029e+02 5.153e+02 6.488e+02 1.407e+03, threshold=1.031e+03, percent-clipped=9.0 2023-03-07 14:29:17,901 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:29:41,462 INFO [train2.py:809] (3/4) Epoch 3, batch 1850, loss[ctc_loss=0.2131, att_loss=0.3237, loss=0.3016, over 17403.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03122, over 63.00 utterances.], tot_loss[ctc_loss=0.2112, att_loss=0.3077, loss=0.2884, over 3252244.86 frames. utt_duration=1223 frames, utt_pad_proportion=0.06255, over 10650.35 utterances.], batch size: 63, lr: 3.14e-02, grad_scale: 8.0 2023-03-07 14:30:07,929 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.8205, 1.7276, 1.7564, 1.7234, 1.0545, 3.0113, 1.3872, 1.2282], device='cuda:3'), covar=tensor([0.0744, 0.1019, 0.0792, 0.2626, 0.2772, 0.0207, 0.2646, 0.2611], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0050, 0.0050, 0.0073, 0.0063, 0.0050, 0.0063, 0.0082], device='cuda:3'), out_proj_covar=tensor([3.7105e-05, 3.3462e-05, 3.3319e-05, 4.4095e-05, 4.0374e-05, 2.8318e-05, 3.9542e-05, 5.3342e-05], device='cuda:3') 2023-03-07 14:30:28,354 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3959, 4.5569, 4.2989, 4.7430, 4.7910, 4.5489, 4.2180, 2.0236], device='cuda:3'), covar=tensor([0.0244, 0.0291, 0.0333, 0.0166, 0.2038, 0.0255, 0.0401, 0.4093], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0108, 0.0104, 0.0105, 0.0203, 0.0126, 0.0092, 0.0247], device='cuda:3'), out_proj_covar=tensor([1.0213e-04, 8.7650e-05, 8.6853e-05, 8.9699e-05, 1.8292e-04, 9.9570e-05, 8.2759e-05, 1.9816e-04], device='cuda:3') 2023-03-07 14:31:02,969 INFO [train2.py:809] (3/4) Epoch 3, batch 1900, loss[ctc_loss=0.2659, att_loss=0.3444, loss=0.3287, over 14614.00 frames. utt_duration=404.6 frames, utt_pad_proportion=0.2964, over 145.00 utterances.], tot_loss[ctc_loss=0.2124, att_loss=0.3081, loss=0.2889, over 3249695.75 frames. utt_duration=1195 frames, utt_pad_proportion=0.07152, over 10891.72 utterances.], batch size: 145, lr: 3.13e-02, grad_scale: 8.0 2023-03-07 14:32:00,232 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.657e+02 4.209e+02 5.043e+02 6.488e+02 1.113e+03, threshold=1.009e+03, percent-clipped=3.0 2023-03-07 14:32:23,702 INFO [train2.py:809] (3/4) Epoch 3, batch 1950, loss[ctc_loss=0.1849, att_loss=0.3068, loss=0.2824, over 16453.00 frames. utt_duration=1432 frames, utt_pad_proportion=0.007307, over 46.00 utterances.], tot_loss[ctc_loss=0.2109, att_loss=0.3076, loss=0.2882, over 3257194.65 frames. utt_duration=1208 frames, utt_pad_proportion=0.06495, over 10801.90 utterances.], batch size: 46, lr: 3.13e-02, grad_scale: 8.0 2023-03-07 14:32:58,029 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:33:44,091 INFO [train2.py:809] (3/4) Epoch 3, batch 2000, loss[ctc_loss=0.2172, att_loss=0.3216, loss=0.3007, over 17041.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009565, over 52.00 utterances.], tot_loss[ctc_loss=0.2094, att_loss=0.3066, loss=0.2871, over 3252973.89 frames. utt_duration=1212 frames, utt_pad_proportion=0.06625, over 10746.50 utterances.], batch size: 52, lr: 3.12e-02, grad_scale: 8.0 2023-03-07 14:34:42,584 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8546, 5.0688, 5.3663, 5.6752, 5.0528, 5.8051, 5.0683, 5.7790], device='cuda:3'), covar=tensor([0.0487, 0.0517, 0.0550, 0.0379, 0.1641, 0.0584, 0.0411, 0.0534], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0254, 0.0243, 0.0287, 0.0440, 0.0232, 0.0212, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 14:34:46,865 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 4.346e+02 5.150e+02 6.489e+02 1.727e+03, threshold=1.030e+03, percent-clipped=5.0 2023-03-07 14:35:03,045 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:35:10,651 INFO [train2.py:809] (3/4) Epoch 3, batch 2050, loss[ctc_loss=0.2048, att_loss=0.3135, loss=0.2918, over 17027.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007628, over 51.00 utterances.], tot_loss[ctc_loss=0.2084, att_loss=0.306, loss=0.2865, over 3265338.68 frames. utt_duration=1227 frames, utt_pad_proportion=0.06025, over 10654.48 utterances.], batch size: 51, lr: 3.11e-02, grad_scale: 8.0 2023-03-07 14:35:43,985 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:35:45,430 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7569, 5.2721, 4.8144, 4.8815, 5.3307, 5.0965, 5.0628, 4.8176], device='cuda:3'), covar=tensor([0.1008, 0.0321, 0.0290, 0.0556, 0.0239, 0.0306, 0.0244, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0184, 0.0122, 0.0149, 0.0195, 0.0211, 0.0165, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 14:36:21,362 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:36:23,559 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-07 14:36:32,074 INFO [train2.py:809] (3/4) Epoch 3, batch 2100, loss[ctc_loss=0.3164, att_loss=0.3595, loss=0.3508, over 14442.00 frames. utt_duration=399.8 frames, utt_pad_proportion=0.3046, over 145.00 utterances.], tot_loss[ctc_loss=0.209, att_loss=0.3063, loss=0.2869, over 3274554.52 frames. utt_duration=1219 frames, utt_pad_proportion=0.0598, over 10754.69 utterances.], batch size: 145, lr: 3.11e-02, grad_scale: 8.0 2023-03-07 14:36:34,021 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:37:21,242 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:37:22,986 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 14:37:28,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.690e+02 4.150e+02 4.955e+02 6.007e+02 1.097e+03, threshold=9.911e+02, percent-clipped=1.0 2023-03-07 14:37:50,922 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:37:52,385 INFO [train2.py:809] (3/4) Epoch 3, batch 2150, loss[ctc_loss=0.1705, att_loss=0.2787, loss=0.2571, over 16123.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.005868, over 42.00 utterances.], tot_loss[ctc_loss=0.2105, att_loss=0.307, loss=0.2877, over 3269658.66 frames. utt_duration=1182 frames, utt_pad_proportion=0.0702, over 11082.68 utterances.], batch size: 42, lr: 3.10e-02, grad_scale: 8.0 2023-03-07 14:37:58,710 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:38:49,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-03-07 14:39:12,831 INFO [train2.py:809] (3/4) Epoch 3, batch 2200, loss[ctc_loss=0.1568, att_loss=0.2681, loss=0.2458, over 15756.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009625, over 38.00 utterances.], tot_loss[ctc_loss=0.211, att_loss=0.3075, loss=0.2882, over 3268286.19 frames. utt_duration=1160 frames, utt_pad_proportion=0.07581, over 11281.38 utterances.], batch size: 38, lr: 3.09e-02, grad_scale: 8.0 2023-03-07 14:40:09,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 4.147e+02 5.296e+02 6.795e+02 2.410e+03, threshold=1.059e+03, percent-clipped=6.0 2023-03-07 14:40:12,823 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-07 14:40:33,273 INFO [train2.py:809] (3/4) Epoch 3, batch 2250, loss[ctc_loss=0.1891, att_loss=0.2848, loss=0.2656, over 15500.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008753, over 36.00 utterances.], tot_loss[ctc_loss=0.2082, att_loss=0.3053, loss=0.2859, over 3260881.99 frames. utt_duration=1196 frames, utt_pad_proportion=0.06885, over 10920.20 utterances.], batch size: 36, lr: 3.09e-02, grad_scale: 8.0 2023-03-07 14:41:07,806 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:41:52,774 INFO [train2.py:809] (3/4) Epoch 3, batch 2300, loss[ctc_loss=0.1935, att_loss=0.299, loss=0.2779, over 16631.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004871, over 47.00 utterances.], tot_loss[ctc_loss=0.2073, att_loss=0.3054, loss=0.2858, over 3275184.70 frames. utt_duration=1231 frames, utt_pad_proportion=0.05767, over 10652.36 utterances.], batch size: 47, lr: 3.08e-02, grad_scale: 8.0 2023-03-07 14:42:23,466 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:42:49,025 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 3.999e+02 4.937e+02 6.652e+02 1.498e+03, threshold=9.874e+02, percent-clipped=4.0 2023-03-07 14:43:04,919 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:43:12,499 INFO [train2.py:809] (3/4) Epoch 3, batch 2350, loss[ctc_loss=0.181, att_loss=0.2918, loss=0.2697, over 17107.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01577, over 56.00 utterances.], tot_loss[ctc_loss=0.2073, att_loss=0.3059, loss=0.2862, over 3283904.11 frames. utt_duration=1233 frames, utt_pad_proportion=0.05466, over 10666.38 utterances.], batch size: 56, lr: 3.08e-02, grad_scale: 16.0 2023-03-07 14:44:13,725 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1166, 2.0434, 3.1858, 4.2030, 4.2157, 4.2863, 2.7021, 1.6923], device='cuda:3'), covar=tensor([0.0269, 0.3025, 0.1063, 0.0479, 0.0186, 0.0137, 0.1959, 0.3104], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0173, 0.0164, 0.0112, 0.0087, 0.0094, 0.0172, 0.0161], device='cuda:3'), out_proj_covar=tensor([1.1115e-04, 1.6459e-04, 1.6168e-04, 1.2702e-04, 9.1294e-05, 9.2480e-05, 1.7001e-04, 1.5395e-04], device='cuda:3') 2023-03-07 14:44:21,304 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:44:21,665 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5286, 2.6208, 4.9071, 3.6895, 2.8936, 4.3660, 4.4551, 4.4115], device='cuda:3'), covar=tensor([0.0130, 0.1624, 0.0140, 0.1414, 0.3295, 0.0467, 0.0293, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0216, 0.0117, 0.0270, 0.0328, 0.0164, 0.0102, 0.0115], device='cuda:3'), out_proj_covar=tensor([9.3618e-05, 1.6031e-04, 9.3276e-05, 2.1121e-04, 2.3920e-04, 1.3222e-04, 8.3469e-05, 9.5973e-05], device='cuda:3') 2023-03-07 14:44:32,344 INFO [train2.py:809] (3/4) Epoch 3, batch 2400, loss[ctc_loss=0.2144, att_loss=0.3193, loss=0.2983, over 16283.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006477, over 43.00 utterances.], tot_loss[ctc_loss=0.2082, att_loss=0.3063, loss=0.2867, over 3282666.42 frames. utt_duration=1242 frames, utt_pad_proportion=0.05203, over 10580.64 utterances.], batch size: 43, lr: 3.07e-02, grad_scale: 16.0 2023-03-07 14:45:14,417 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:45:20,850 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:45:28,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.513e+02 3.987e+02 4.935e+02 6.057e+02 1.238e+03, threshold=9.870e+02, percent-clipped=1.0 2023-03-07 14:45:36,609 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:45:39,827 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4651, 4.2569, 4.1438, 4.5676, 4.6417, 4.2495, 3.7672, 2.1833], device='cuda:3'), covar=tensor([0.0259, 0.0526, 0.0390, 0.0103, 0.1111, 0.0285, 0.0576, 0.3648], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0114, 0.0107, 0.0108, 0.0211, 0.0125, 0.0099, 0.0247], device='cuda:3'), out_proj_covar=tensor([1.0759e-04, 9.1519e-05, 9.0719e-05, 9.4106e-05, 1.9176e-04, 1.0155e-04, 8.8688e-05, 2.0316e-04], device='cuda:3') 2023-03-07 14:45:50,453 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 14:45:51,945 INFO [train2.py:809] (3/4) Epoch 3, batch 2450, loss[ctc_loss=0.2343, att_loss=0.3316, loss=0.3121, over 16767.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005952, over 48.00 utterances.], tot_loss[ctc_loss=0.2079, att_loss=0.3063, loss=0.2866, over 3285693.06 frames. utt_duration=1221 frames, utt_pad_proportion=0.05693, over 10778.11 utterances.], batch size: 48, lr: 3.06e-02, grad_scale: 16.0 2023-03-07 14:46:26,538 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-03-07 14:46:38,266 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:46:49,282 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:47:12,261 INFO [train2.py:809] (3/4) Epoch 3, batch 2500, loss[ctc_loss=0.1764, att_loss=0.2836, loss=0.2622, over 15989.00 frames. utt_duration=1600 frames, utt_pad_proportion=0.007827, over 40.00 utterances.], tot_loss[ctc_loss=0.2064, att_loss=0.3055, loss=0.2857, over 3277594.53 frames. utt_duration=1233 frames, utt_pad_proportion=0.05696, over 10646.45 utterances.], batch size: 40, lr: 3.06e-02, grad_scale: 16.0 2023-03-07 14:47:14,181 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:47:37,327 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7738, 3.9448, 3.1977, 4.0330, 3.8889, 3.7764, 3.4244, 2.3773], device='cuda:3'), covar=tensor([0.0237, 0.0264, 0.0338, 0.0139, 0.0910, 0.0214, 0.0422, 0.3083], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0112, 0.0111, 0.0218, 0.0130, 0.0102, 0.0257], device='cuda:3'), out_proj_covar=tensor([1.1243e-04, 9.3995e-05, 9.4381e-05, 9.7059e-05, 1.9845e-04, 1.0615e-04, 9.1952e-05, 2.1184e-04], device='cuda:3') 2023-03-07 14:48:09,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.331e+02 4.171e+02 5.202e+02 6.320e+02 1.110e+03, threshold=1.040e+03, percent-clipped=6.0 2023-03-07 14:48:27,734 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:48:33,424 INFO [train2.py:809] (3/4) Epoch 3, batch 2550, loss[ctc_loss=0.1891, att_loss=0.3041, loss=0.2811, over 16954.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008313, over 50.00 utterances.], tot_loss[ctc_loss=0.204, att_loss=0.3036, loss=0.2837, over 3271991.27 frames. utt_duration=1247 frames, utt_pad_proportion=0.05637, over 10506.19 utterances.], batch size: 50, lr: 3.05e-02, grad_scale: 16.0 2023-03-07 14:49:52,839 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4139, 4.8889, 4.6946, 5.0686, 4.4419, 4.7281, 5.2665, 4.9727], device='cuda:3'), covar=tensor([0.0334, 0.0229, 0.0466, 0.0147, 0.0375, 0.0191, 0.0224, 0.0154], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0121, 0.0147, 0.0090, 0.0136, 0.0100, 0.0117, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 14:49:54,182 INFO [train2.py:809] (3/4) Epoch 3, batch 2600, loss[ctc_loss=0.1707, att_loss=0.2905, loss=0.2665, over 16701.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.00565, over 46.00 utterances.], tot_loss[ctc_loss=0.2035, att_loss=0.3038, loss=0.2837, over 3274894.12 frames. utt_duration=1240 frames, utt_pad_proportion=0.05694, over 10579.13 utterances.], batch size: 46, lr: 3.05e-02, grad_scale: 16.0 2023-03-07 14:50:50,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.348e+02 3.926e+02 4.748e+02 6.116e+02 1.074e+03, threshold=9.496e+02, percent-clipped=2.0 2023-03-07 14:51:13,368 INFO [train2.py:809] (3/4) Epoch 3, batch 2650, loss[ctc_loss=0.1855, att_loss=0.3112, loss=0.286, over 16876.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007141, over 49.00 utterances.], tot_loss[ctc_loss=0.2031, att_loss=0.3034, loss=0.2833, over 3260785.17 frames. utt_duration=1243 frames, utt_pad_proportion=0.0598, over 10508.92 utterances.], batch size: 49, lr: 3.04e-02, grad_scale: 16.0 2023-03-07 14:52:32,126 INFO [train2.py:809] (3/4) Epoch 3, batch 2700, loss[ctc_loss=0.2154, att_loss=0.3291, loss=0.3064, over 16621.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005932, over 47.00 utterances.], tot_loss[ctc_loss=0.204, att_loss=0.3043, loss=0.2842, over 3264170.95 frames. utt_duration=1229 frames, utt_pad_proportion=0.06052, over 10637.17 utterances.], batch size: 47, lr: 3.03e-02, grad_scale: 16.0 2023-03-07 14:52:47,895 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-03-07 14:53:14,754 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:53:28,795 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.793e+02 4.160e+02 5.458e+02 6.683e+02 1.246e+03, threshold=1.092e+03, percent-clipped=5.0 2023-03-07 14:53:51,037 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:53:52,188 INFO [train2.py:809] (3/4) Epoch 3, batch 2750, loss[ctc_loss=0.2222, att_loss=0.3208, loss=0.3011, over 17280.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01313, over 55.00 utterances.], tot_loss[ctc_loss=0.2025, att_loss=0.3025, loss=0.2825, over 3245982.77 frames. utt_duration=1253 frames, utt_pad_proportion=0.05743, over 10377.21 utterances.], batch size: 55, lr: 3.03e-02, grad_scale: 16.0 2023-03-07 14:53:56,203 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0115, 5.3094, 5.5934, 5.7508, 5.1679, 5.8631, 5.0753, 5.9383], device='cuda:3'), covar=tensor([0.0480, 0.0480, 0.0399, 0.0451, 0.2115, 0.0628, 0.0452, 0.0529], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0257, 0.0248, 0.0300, 0.0451, 0.0240, 0.0217, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 14:54:18,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2023-03-07 14:54:32,775 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:54:34,620 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:55:06,809 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:55:08,219 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:55:12,969 INFO [train2.py:809] (3/4) Epoch 3, batch 2800, loss[ctc_loss=0.2425, att_loss=0.3366, loss=0.3178, over 17019.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007831, over 51.00 utterances.], tot_loss[ctc_loss=0.2028, att_loss=0.3032, loss=0.2831, over 3259151.32 frames. utt_duration=1262 frames, utt_pad_proportion=0.05336, over 10342.83 utterances.], batch size: 51, lr: 3.02e-02, grad_scale: 16.0 2023-03-07 14:55:22,651 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:55:39,858 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4717, 4.9911, 4.7684, 4.9441, 4.3581, 4.7391, 5.2516, 4.9543], device='cuda:3'), covar=tensor([0.0327, 0.0294, 0.0425, 0.0173, 0.0371, 0.0191, 0.0205, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0121, 0.0144, 0.0091, 0.0133, 0.0100, 0.0118, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 14:56:10,551 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.287e+02 4.175e+02 5.006e+02 5.883e+02 1.707e+03, threshold=1.001e+03, percent-clipped=3.0 2023-03-07 14:56:12,586 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:56:20,292 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:56:33,910 INFO [train2.py:809] (3/4) Epoch 3, batch 2850, loss[ctc_loss=0.1964, att_loss=0.3065, loss=0.2845, over 16461.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007377, over 46.00 utterances.], tot_loss[ctc_loss=0.2024, att_loss=0.3031, loss=0.2829, over 3263013.90 frames. utt_duration=1245 frames, utt_pad_proportion=0.05727, over 10498.43 utterances.], batch size: 46, lr: 3.02e-02, grad_scale: 16.0 2023-03-07 14:57:00,146 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 14:57:22,075 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1671, 4.3901, 4.8465, 4.8299, 2.1672, 4.6131, 2.6190, 2.7358], device='cuda:3'), covar=tensor([0.0214, 0.0139, 0.0520, 0.0187, 0.3432, 0.0183, 0.1583, 0.1131], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0084, 0.0194, 0.0104, 0.0215, 0.0093, 0.0180, 0.0166], device='cuda:3'), out_proj_covar=tensor([8.5298e-05, 7.6697e-05, 1.6085e-04, 8.5278e-05, 1.6900e-04, 8.0492e-05, 1.4491e-04, 1.3539e-04], device='cuda:3') 2023-03-07 14:57:54,307 INFO [train2.py:809] (3/4) Epoch 3, batch 2900, loss[ctc_loss=0.1743, att_loss=0.2638, loss=0.2459, over 15494.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009233, over 36.00 utterances.], tot_loss[ctc_loss=0.2041, att_loss=0.3044, loss=0.2843, over 3265193.82 frames. utt_duration=1222 frames, utt_pad_proportion=0.0633, over 10703.81 utterances.], batch size: 36, lr: 3.01e-02, grad_scale: 4.0 2023-03-07 14:58:13,351 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1776, 4.8955, 5.0670, 4.6288, 2.0868, 2.4819, 5.2144, 3.7620], device='cuda:3'), covar=tensor([0.0543, 0.0186, 0.0125, 0.0398, 0.7565, 0.2633, 0.0153, 0.2040], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0144, 0.0163, 0.0186, 0.0408, 0.0301, 0.0149, 0.0260], device='cuda:3'), out_proj_covar=tensor([1.3722e-04, 7.6882e-05, 8.7819e-05, 9.0696e-05, 2.0259e-04, 1.5127e-04, 7.8041e-05, 1.4560e-04], device='cuda:3') 2023-03-07 14:58:21,043 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 14:58:54,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.371e+02 4.021e+02 4.974e+02 6.236e+02 1.292e+03, threshold=9.948e+02, percent-clipped=6.0 2023-03-07 14:59:14,720 INFO [train2.py:809] (3/4) Epoch 3, batch 2950, loss[ctc_loss=0.2009, att_loss=0.2737, loss=0.2591, over 15792.00 frames. utt_duration=1664 frames, utt_pad_proportion=0.007286, over 38.00 utterances.], tot_loss[ctc_loss=0.2033, att_loss=0.3031, loss=0.2831, over 3268027.08 frames. utt_duration=1228 frames, utt_pad_proportion=0.06023, over 10655.96 utterances.], batch size: 38, lr: 3.01e-02, grad_scale: 4.0 2023-03-07 14:59:28,499 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6927, 5.1114, 4.8859, 5.1133, 4.5740, 4.8897, 5.3744, 5.0483], device='cuda:3'), covar=tensor([0.0287, 0.0250, 0.0385, 0.0113, 0.0340, 0.0154, 0.0201, 0.0125], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0120, 0.0141, 0.0088, 0.0131, 0.0099, 0.0115, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-07 14:59:59,143 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 15:00:00,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4591, 4.7105, 5.0173, 5.1495, 4.7061, 5.3362, 4.8335, 5.4705], device='cuda:3'), covar=tensor([0.0525, 0.0643, 0.0460, 0.0531, 0.1781, 0.0701, 0.0833, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0265, 0.0253, 0.0303, 0.0450, 0.0247, 0.0225, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 15:00:02,159 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3806, 4.9416, 4.5066, 4.8896, 4.9679, 4.6204, 4.2957, 4.8023], device='cuda:3'), covar=tensor([0.0106, 0.0121, 0.0089, 0.0111, 0.0080, 0.0103, 0.0315, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0042, 0.0044, 0.0032, 0.0032, 0.0039, 0.0059, 0.0053], device='cuda:3'), out_proj_covar=tensor([1.1572e-04, 1.1594e-04, 1.4153e-04, 9.2420e-05, 8.7801e-05, 1.1483e-04, 1.6349e-04, 1.4846e-04], device='cuda:3') 2023-03-07 15:00:34,062 INFO [train2.py:809] (3/4) Epoch 3, batch 3000, loss[ctc_loss=0.2088, att_loss=0.3189, loss=0.2969, over 17305.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01166, over 55.00 utterances.], tot_loss[ctc_loss=0.202, att_loss=0.302, loss=0.282, over 3267470.58 frames. utt_duration=1246 frames, utt_pad_proportion=0.05638, over 10498.95 utterances.], batch size: 55, lr: 3.00e-02, grad_scale: 4.0 2023-03-07 15:00:34,063 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 15:00:43,966 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2296, 3.2312, 2.8935, 3.0836, 3.4212, 3.0969, 2.0356, 3.6420], device='cuda:3'), covar=tensor([0.1554, 0.0598, 0.1529, 0.0745, 0.0710, 0.1030, 0.1409, 0.0286], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0092, 0.0144, 0.0117, 0.0104, 0.0138, 0.0122, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:3') 2023-03-07 15:00:47,743 INFO [train2.py:843] (3/4) Epoch 3, validation: ctc_loss=0.1004, att_loss=0.2657, loss=0.2327, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 15:00:47,743 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 15:01:48,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.705e+02 4.292e+02 5.307e+02 6.516e+02 1.303e+03, threshold=1.061e+03, percent-clipped=4.0 2023-03-07 15:02:09,268 INFO [train2.py:809] (3/4) Epoch 3, batch 3050, loss[ctc_loss=0.2005, att_loss=0.3052, loss=0.2842, over 16745.00 frames. utt_duration=1397 frames, utt_pad_proportion=0.007033, over 48.00 utterances.], tot_loss[ctc_loss=0.2015, att_loss=0.3017, loss=0.2816, over 3270407.24 frames. utt_duration=1240 frames, utt_pad_proportion=0.05742, over 10565.54 utterances.], batch size: 48, lr: 2.99e-02, grad_scale: 4.0 2023-03-07 15:02:17,817 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1040, 5.2314, 5.5539, 5.8235, 5.3337, 6.0253, 5.2373, 6.2049], device='cuda:3'), covar=tensor([0.0506, 0.0529, 0.0415, 0.0515, 0.1738, 0.0700, 0.0327, 0.0336], device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0262, 0.0251, 0.0297, 0.0445, 0.0239, 0.0220, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 15:02:32,778 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 2023-03-07 15:02:48,203 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7359, 5.1330, 4.9086, 5.1798, 4.6032, 4.9135, 5.3972, 5.0759], device='cuda:3'), covar=tensor([0.0268, 0.0248, 0.0353, 0.0128, 0.0373, 0.0129, 0.0201, 0.0153], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0123, 0.0145, 0.0092, 0.0137, 0.0101, 0.0117, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:03:23,385 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:03:29,047 INFO [train2.py:809] (3/4) Epoch 3, batch 3100, loss[ctc_loss=0.181, att_loss=0.2908, loss=0.2688, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007094, over 40.00 utterances.], tot_loss[ctc_loss=0.2026, att_loss=0.3025, loss=0.2825, over 3274706.61 frames. utt_duration=1250 frames, utt_pad_proportion=0.05414, over 10490.85 utterances.], batch size: 40, lr: 2.99e-02, grad_scale: 4.0 2023-03-07 15:03:55,829 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8215, 4.8318, 4.6750, 3.0684, 4.6305, 4.1504, 3.8463, 2.5642], device='cuda:3'), covar=tensor([0.0125, 0.0061, 0.0225, 0.0600, 0.0097, 0.0143, 0.0249, 0.1267], device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0042, 0.0038, 0.0068, 0.0041, 0.0051, 0.0057, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:04:06,740 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1252, 4.4900, 4.2477, 4.6608, 4.6684, 4.4958, 3.9587, 4.2942], device='cuda:3'), covar=tensor([0.0110, 0.0157, 0.0129, 0.0109, 0.0122, 0.0108, 0.0373, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0043, 0.0044, 0.0033, 0.0033, 0.0041, 0.0061, 0.0054], device='cuda:3'), out_proj_covar=tensor([1.1822e-04, 1.2085e-04, 1.4176e-04, 9.7228e-05, 9.2427e-05, 1.1935e-04, 1.6890e-04, 1.5323e-04], device='cuda:3') 2023-03-07 15:04:20,589 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:04:29,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.231e+02 3.861e+02 4.979e+02 6.138e+02 1.576e+03, threshold=9.958e+02, percent-clipped=5.0 2023-03-07 15:04:35,995 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 15:04:40,485 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:04:50,176 INFO [train2.py:809] (3/4) Epoch 3, batch 3150, loss[ctc_loss=0.2626, att_loss=0.3358, loss=0.3211, over 17070.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008528, over 53.00 utterances.], tot_loss[ctc_loss=0.2019, att_loss=0.3026, loss=0.2825, over 3281764.91 frames. utt_duration=1247 frames, utt_pad_proportion=0.05224, over 10543.41 utterances.], batch size: 53, lr: 2.98e-02, grad_scale: 4.0 2023-03-07 15:04:57,424 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7609, 4.7099, 4.8093, 3.4290, 4.7866, 4.1696, 4.3782, 2.2790], device='cuda:3'), covar=tensor([0.0164, 0.0102, 0.0238, 0.0614, 0.0087, 0.0155, 0.0209, 0.1706], device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0043, 0.0039, 0.0070, 0.0042, 0.0052, 0.0058, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:05:08,098 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:05:53,262 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:06:11,074 INFO [train2.py:809] (3/4) Epoch 3, batch 3200, loss[ctc_loss=0.1374, att_loss=0.251, loss=0.2283, over 15342.00 frames. utt_duration=1755 frames, utt_pad_proportion=0.0119, over 35.00 utterances.], tot_loss[ctc_loss=0.2015, att_loss=0.303, loss=0.2827, over 3278476.19 frames. utt_duration=1230 frames, utt_pad_proportion=0.05773, over 10676.35 utterances.], batch size: 35, lr: 2.98e-02, grad_scale: 8.0 2023-03-07 15:06:19,903 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:07:06,296 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7034, 5.0029, 4.8770, 4.9025, 5.2595, 4.9991, 4.8601, 4.6927], device='cuda:3'), covar=tensor([0.1141, 0.0441, 0.0211, 0.0551, 0.0221, 0.0284, 0.0265, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0308, 0.0192, 0.0129, 0.0159, 0.0207, 0.0227, 0.0178, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:07:10,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.395e+02 3.760e+02 4.677e+02 5.867e+02 1.602e+03, threshold=9.354e+02, percent-clipped=2.0 2023-03-07 15:07:15,787 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9870, 4.7644, 4.7804, 3.4010, 4.8172, 4.2720, 4.4393, 2.6454], device='cuda:3'), covar=tensor([0.0143, 0.0104, 0.0326, 0.0656, 0.0097, 0.0188, 0.0212, 0.1560], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0045, 0.0040, 0.0073, 0.0044, 0.0053, 0.0060, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 15:07:31,879 INFO [train2.py:809] (3/4) Epoch 3, batch 3250, loss[ctc_loss=0.1937, att_loss=0.3055, loss=0.2831, over 17336.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03667, over 63.00 utterances.], tot_loss[ctc_loss=0.2007, att_loss=0.3023, loss=0.282, over 3274727.69 frames. utt_duration=1228 frames, utt_pad_proportion=0.06038, over 10681.87 utterances.], batch size: 63, lr: 2.97e-02, grad_scale: 8.0 2023-03-07 15:07:57,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2023-03-07 15:07:58,391 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:08:07,921 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 15:08:52,867 INFO [train2.py:809] (3/4) Epoch 3, batch 3300, loss[ctc_loss=0.1914, att_loss=0.3031, loss=0.2808, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006599, over 48.00 utterances.], tot_loss[ctc_loss=0.2002, att_loss=0.3018, loss=0.2815, over 3270849.70 frames. utt_duration=1217 frames, utt_pad_proportion=0.06364, over 10766.67 utterances.], batch size: 48, lr: 2.97e-02, grad_scale: 8.0 2023-03-07 15:09:14,154 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3911, 4.3826, 4.3183, 4.5345, 4.8480, 4.2460, 4.1220, 1.9172], device='cuda:3'), covar=tensor([0.0230, 0.0426, 0.0320, 0.0151, 0.1099, 0.0281, 0.0596, 0.3653], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0111, 0.0106, 0.0107, 0.0215, 0.0128, 0.0102, 0.0234], device='cuda:3'), out_proj_covar=tensor([1.0975e-04, 9.2316e-05, 8.8538e-05, 9.3567e-05, 1.9591e-04, 1.0792e-04, 9.0495e-05, 1.9962e-04], device='cuda:3') 2023-03-07 15:09:52,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.248e+02 3.797e+02 4.904e+02 6.059e+02 1.982e+03, threshold=9.808e+02, percent-clipped=6.0 2023-03-07 15:10:12,979 INFO [train2.py:809] (3/4) Epoch 3, batch 3350, loss[ctc_loss=0.2063, att_loss=0.3005, loss=0.2816, over 16490.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006082, over 46.00 utterances.], tot_loss[ctc_loss=0.1983, att_loss=0.3002, loss=0.2798, over 3264680.49 frames. utt_duration=1234 frames, utt_pad_proportion=0.05997, over 10594.81 utterances.], batch size: 46, lr: 2.96e-02, grad_scale: 8.0 2023-03-07 15:11:34,096 INFO [train2.py:809] (3/4) Epoch 3, batch 3400, loss[ctc_loss=0.1871, att_loss=0.2823, loss=0.2633, over 15884.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009238, over 39.00 utterances.], tot_loss[ctc_loss=0.1983, att_loss=0.3, loss=0.2796, over 3261794.45 frames. utt_duration=1245 frames, utt_pad_proportion=0.05904, over 10490.09 utterances.], batch size: 39, lr: 2.96e-02, grad_scale: 8.0 2023-03-07 15:12:25,242 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:12:34,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.436e+02 3.676e+02 4.769e+02 6.355e+02 1.428e+03, threshold=9.539e+02, percent-clipped=8.0 2023-03-07 15:12:55,340 INFO [train2.py:809] (3/4) Epoch 3, batch 3450, loss[ctc_loss=0.1845, att_loss=0.2855, loss=0.2653, over 16033.00 frames. utt_duration=1605 frames, utt_pad_proportion=0.00584, over 40.00 utterances.], tot_loss[ctc_loss=0.1988, att_loss=0.3006, loss=0.2803, over 3261065.53 frames. utt_duration=1237 frames, utt_pad_proportion=0.06153, over 10560.09 utterances.], batch size: 40, lr: 2.95e-02, grad_scale: 8.0 2023-03-07 15:12:57,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8486, 5.9590, 5.3947, 5.9478, 5.6389, 5.4380, 5.4423, 5.3214], device='cuda:3'), covar=tensor([0.0838, 0.0802, 0.0646, 0.0617, 0.0507, 0.1284, 0.2138, 0.2335], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0320, 0.0259, 0.0254, 0.0231, 0.0318, 0.0350, 0.0335], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:13:12,845 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:13:41,675 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:14:15,451 INFO [train2.py:809] (3/4) Epoch 3, batch 3500, loss[ctc_loss=0.1482, att_loss=0.2572, loss=0.2354, over 15490.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.00944, over 36.00 utterances.], tot_loss[ctc_loss=0.1974, att_loss=0.3, loss=0.2795, over 3261488.71 frames. utt_duration=1246 frames, utt_pad_proportion=0.05977, over 10481.31 utterances.], batch size: 36, lr: 2.95e-02, grad_scale: 8.0 2023-03-07 15:14:26,719 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1538, 4.7293, 5.0445, 3.5106, 4.9181, 4.1411, 4.5705, 2.6150], device='cuda:3'), covar=tensor([0.0109, 0.0096, 0.0250, 0.0601, 0.0100, 0.0179, 0.0201, 0.1398], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0045, 0.0039, 0.0074, 0.0043, 0.0053, 0.0061, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 15:14:29,628 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:14:37,087 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-07 15:14:52,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-07 15:15:14,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 3.561e+02 4.473e+02 5.619e+02 1.545e+03, threshold=8.947e+02, percent-clipped=3.0 2023-03-07 15:15:36,470 INFO [train2.py:809] (3/4) Epoch 3, batch 3550, loss[ctc_loss=0.2148, att_loss=0.325, loss=0.303, over 17025.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01108, over 53.00 utterances.], tot_loss[ctc_loss=0.1964, att_loss=0.2998, loss=0.2791, over 3265645.12 frames. utt_duration=1269 frames, utt_pad_proportion=0.0541, over 10303.45 utterances.], batch size: 53, lr: 2.94e-02, grad_scale: 8.0 2023-03-07 15:15:53,513 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:15:56,852 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:16:12,287 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 15:16:54,935 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 2023-03-07 15:16:56,925 INFO [train2.py:809] (3/4) Epoch 3, batch 3600, loss[ctc_loss=0.2238, att_loss=0.3289, loss=0.3079, over 16477.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005875, over 46.00 utterances.], tot_loss[ctc_loss=0.1969, att_loss=0.3002, loss=0.2795, over 3265574.09 frames. utt_duration=1267 frames, utt_pad_proportion=0.05405, over 10323.47 utterances.], batch size: 46, lr: 2.93e-02, grad_scale: 8.0 2023-03-07 15:17:29,070 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 15:17:36,077 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:17:56,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.467e+02 4.035e+02 5.369e+02 6.421e+02 1.595e+03, threshold=1.074e+03, percent-clipped=5.0 2023-03-07 15:18:18,106 INFO [train2.py:809] (3/4) Epoch 3, batch 3650, loss[ctc_loss=0.184, att_loss=0.3036, loss=0.2797, over 16775.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005993, over 48.00 utterances.], tot_loss[ctc_loss=0.1963, att_loss=0.3, loss=0.2793, over 3273396.20 frames. utt_duration=1267 frames, utt_pad_proportion=0.05073, over 10346.48 utterances.], batch size: 48, lr: 2.93e-02, grad_scale: 8.0 2023-03-07 15:18:43,942 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:19:14,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.78 vs. limit=5.0 2023-03-07 15:19:38,109 INFO [train2.py:809] (3/4) Epoch 3, batch 3700, loss[ctc_loss=0.171, att_loss=0.2841, loss=0.2615, over 16266.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.008057, over 43.00 utterances.], tot_loss[ctc_loss=0.1968, att_loss=0.2994, loss=0.2789, over 3259834.86 frames. utt_duration=1254 frames, utt_pad_proportion=0.05684, over 10410.76 utterances.], batch size: 43, lr: 2.92e-02, grad_scale: 8.0 2023-03-07 15:19:57,875 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-07 15:20:22,387 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:20:27,186 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:20:37,361 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.856e+02 3.737e+02 4.615e+02 5.715e+02 1.108e+03, threshold=9.229e+02, percent-clipped=1.0 2023-03-07 15:20:55,619 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4893, 4.9402, 4.6604, 5.0247, 4.4573, 4.9041, 5.2653, 5.0425], device='cuda:3'), covar=tensor([0.0335, 0.0280, 0.0502, 0.0159, 0.0379, 0.0139, 0.0178, 0.0116], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0130, 0.0156, 0.0097, 0.0143, 0.0105, 0.0127, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:20:57,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-03-07 15:20:58,426 INFO [train2.py:809] (3/4) Epoch 3, batch 3750, loss[ctc_loss=0.1817, att_loss=0.28, loss=0.2604, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009144, over 39.00 utterances.], tot_loss[ctc_loss=0.1972, att_loss=0.2995, loss=0.2791, over 3267454.75 frames. utt_duration=1250 frames, utt_pad_proportion=0.05506, over 10465.64 utterances.], batch size: 39, lr: 2.92e-02, grad_scale: 8.0 2023-03-07 15:22:04,584 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:22:09,702 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6531, 5.8227, 5.1869, 5.8493, 5.5662, 5.1805, 5.2340, 5.2770], device='cuda:3'), covar=tensor([0.0990, 0.0734, 0.0672, 0.0580, 0.0481, 0.1235, 0.1936, 0.1834], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0312, 0.0252, 0.0245, 0.0225, 0.0319, 0.0333, 0.0328], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:22:18,743 INFO [train2.py:809] (3/4) Epoch 3, batch 3800, loss[ctc_loss=0.1644, att_loss=0.2778, loss=0.2551, over 14534.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.04203, over 32.00 utterances.], tot_loss[ctc_loss=0.1976, att_loss=0.3001, loss=0.2796, over 3276771.51 frames. utt_duration=1243 frames, utt_pad_proportion=0.0535, over 10554.30 utterances.], batch size: 32, lr: 2.91e-02, grad_scale: 8.0 2023-03-07 15:23:08,839 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.22 vs. limit=5.0 2023-03-07 15:23:18,123 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 4.239e+02 5.484e+02 6.646e+02 1.603e+03, threshold=1.097e+03, percent-clipped=8.0 2023-03-07 15:23:33,644 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:23:37,905 INFO [train2.py:809] (3/4) Epoch 3, batch 3850, loss[ctc_loss=0.1824, att_loss=0.3021, loss=0.2782, over 17440.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04495, over 69.00 utterances.], tot_loss[ctc_loss=0.1988, att_loss=0.3013, loss=0.2808, over 3282681.67 frames. utt_duration=1208 frames, utt_pad_proportion=0.06075, over 10879.52 utterances.], batch size: 69, lr: 2.91e-02, grad_scale: 8.0 2023-03-07 15:23:55,920 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:24:20,877 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:24:55,939 INFO [train2.py:809] (3/4) Epoch 3, batch 3900, loss[ctc_loss=0.2136, att_loss=0.3095, loss=0.2903, over 16677.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007182, over 46.00 utterances.], tot_loss[ctc_loss=0.1978, att_loss=0.3011, loss=0.2804, over 3283152.21 frames. utt_duration=1220 frames, utt_pad_proportion=0.05892, over 10779.75 utterances.], batch size: 46, lr: 2.90e-02, grad_scale: 8.0 2023-03-07 15:25:08,645 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:25:09,894 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:25:25,499 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:25:53,043 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 3.699e+02 4.427e+02 5.534e+02 1.749e+03, threshold=8.855e+02, percent-clipped=3.0 2023-03-07 15:25:54,897 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:26:13,114 INFO [train2.py:809] (3/4) Epoch 3, batch 3950, loss[ctc_loss=0.1817, att_loss=0.2977, loss=0.2745, over 16187.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006285, over 41.00 utterances.], tot_loss[ctc_loss=0.1969, att_loss=0.3003, loss=0.2796, over 3284149.34 frames. utt_duration=1240 frames, utt_pad_proportion=0.05431, over 10606.48 utterances.], batch size: 41, lr: 2.90e-02, grad_scale: 8.0 2023-03-07 15:26:14,903 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:26:57,730 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:27:27,004 INFO [train2.py:809] (3/4) Epoch 4, batch 0, loss[ctc_loss=0.2179, att_loss=0.3169, loss=0.2971, over 17381.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03478, over 63.00 utterances.], tot_loss[ctc_loss=0.2179, att_loss=0.3169, loss=0.2971, over 17381.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03478, over 63.00 utterances.], batch size: 63, lr: 2.71e-02, grad_scale: 8.0 2023-03-07 15:27:27,004 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 15:27:39,230 INFO [train2.py:843] (3/4) Epoch 4, validation: ctc_loss=0.101, att_loss=0.2624, loss=0.2301, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 15:27:39,231 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 15:27:55,707 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.5748, 1.7554, 2.9694, 1.8646, 2.5040, 2.6296, 1.6581, 1.4094], device='cuda:3'), covar=tensor([0.1233, 0.1396, 0.0553, 0.2407, 0.1085, 0.1001, 0.1641, 0.4498], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0056, 0.0054, 0.0067, 0.0049, 0.0061, 0.0055, 0.0079], device='cuda:3'), out_proj_covar=tensor([3.7347e-05, 3.3772e-05, 3.2977e-05, 4.2943e-05, 3.4709e-05, 3.6938e-05, 3.5264e-05, 5.5208e-05], device='cuda:3') 2023-03-07 15:28:16,824 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9295, 2.3374, 3.8160, 3.4906, 3.0811, 3.6817, 3.4800, 3.7681], device='cuda:3'), covar=tensor([0.0160, 0.1452, 0.0155, 0.0945, 0.1838, 0.0396, 0.0248, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0224, 0.0120, 0.0277, 0.0326, 0.0170, 0.0110, 0.0116], device='cuda:3'), out_proj_covar=tensor([9.9754e-05, 1.6961e-04, 9.6918e-05, 2.1904e-04, 2.4444e-04, 1.3768e-04, 9.2328e-05, 9.8953e-05], device='cuda:3') 2023-03-07 15:28:24,351 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:28:28,829 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:28:39,322 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:29:03,781 INFO [train2.py:809] (3/4) Epoch 4, batch 50, loss[ctc_loss=0.2052, att_loss=0.3192, loss=0.2964, over 17024.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.01156, over 53.00 utterances.], tot_loss[ctc_loss=0.2013, att_loss=0.3029, loss=0.2826, over 740323.65 frames. utt_duration=1216 frames, utt_pad_proportion=0.05957, over 2438.96 utterances.], batch size: 53, lr: 2.70e-02, grad_scale: 8.0 2023-03-07 15:29:08,345 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.811e+02 4.015e+02 4.845e+02 5.994e+02 1.313e+03, threshold=9.689e+02, percent-clipped=6.0 2023-03-07 15:29:13,807 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:30:10,884 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:30:24,239 INFO [train2.py:809] (3/4) Epoch 4, batch 100, loss[ctc_loss=0.1606, att_loss=0.2823, loss=0.258, over 16391.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.008145, over 44.00 utterances.], tot_loss[ctc_loss=0.1946, att_loss=0.2991, loss=0.2782, over 1301736.80 frames. utt_duration=1230 frames, utt_pad_proportion=0.05509, over 4238.81 utterances.], batch size: 44, lr: 2.70e-02, grad_scale: 8.0 2023-03-07 15:30:27,596 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:31:01,992 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:31:44,680 INFO [train2.py:809] (3/4) Epoch 4, batch 150, loss[ctc_loss=0.1672, att_loss=0.2869, loss=0.263, over 15944.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007078, over 41.00 utterances.], tot_loss[ctc_loss=0.1947, att_loss=0.2995, loss=0.2785, over 1748224.34 frames. utt_duration=1201 frames, utt_pad_proportion=0.05872, over 5832.18 utterances.], batch size: 41, lr: 2.69e-02, grad_scale: 8.0 2023-03-07 15:31:49,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.353e+02 4.093e+02 5.170e+02 6.831e+02 1.205e+03, threshold=1.034e+03, percent-clipped=5.0 2023-03-07 15:32:38,708 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:32:50,940 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7966, 5.9692, 5.3416, 5.9741, 5.6231, 5.3257, 5.4380, 5.3162], device='cuda:3'), covar=tensor([0.0906, 0.0697, 0.0622, 0.0548, 0.0551, 0.1059, 0.1619, 0.1759], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0330, 0.0271, 0.0255, 0.0239, 0.0330, 0.0360, 0.0345], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:33:04,151 INFO [train2.py:809] (3/4) Epoch 4, batch 200, loss[ctc_loss=0.1944, att_loss=0.2991, loss=0.2782, over 16173.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006596, over 41.00 utterances.], tot_loss[ctc_loss=0.1931, att_loss=0.2984, loss=0.2773, over 2085600.63 frames. utt_duration=1222 frames, utt_pad_proportion=0.05473, over 6835.72 utterances.], batch size: 41, lr: 2.69e-02, grad_scale: 8.0 2023-03-07 15:33:34,666 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:33:52,979 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8753, 5.3407, 5.1221, 5.3636, 4.7742, 5.2340, 5.6072, 5.3809], device='cuda:3'), covar=tensor([0.0294, 0.0179, 0.0323, 0.0115, 0.0319, 0.0124, 0.0178, 0.0104], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0135, 0.0163, 0.0102, 0.0150, 0.0109, 0.0132, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 15:33:59,125 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:34:20,168 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2162, 5.0443, 5.0851, 4.1547, 2.0924, 2.7177, 5.2447, 3.9899], device='cuda:3'), covar=tensor([0.0543, 0.0135, 0.0113, 0.0659, 0.7300, 0.2427, 0.0110, 0.1629], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0144, 0.0164, 0.0188, 0.0399, 0.0305, 0.0154, 0.0266], device='cuda:3'), out_proj_covar=tensor([1.4112e-04, 7.3877e-05, 8.7511e-05, 9.1253e-05, 2.0052e-04, 1.5276e-04, 7.9484e-05, 1.4874e-04], device='cuda:3') 2023-03-07 15:34:21,570 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:34:23,025 INFO [train2.py:809] (3/4) Epoch 4, batch 250, loss[ctc_loss=0.1792, att_loss=0.2975, loss=0.2739, over 17009.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.009266, over 51.00 utterances.], tot_loss[ctc_loss=0.1911, att_loss=0.2963, loss=0.2752, over 2345315.42 frames. utt_duration=1286 frames, utt_pad_proportion=0.04332, over 7305.93 utterances.], batch size: 51, lr: 2.68e-02, grad_scale: 8.0 2023-03-07 15:34:28,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.253e+02 3.783e+02 4.490e+02 5.232e+02 8.659e+02, threshold=8.979e+02, percent-clipped=0.0 2023-03-07 15:35:15,673 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:35:42,723 INFO [train2.py:809] (3/4) Epoch 4, batch 300, loss[ctc_loss=0.1453, att_loss=0.2469, loss=0.2266, over 15620.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.009753, over 37.00 utterances.], tot_loss[ctc_loss=0.1923, att_loss=0.2976, loss=0.2765, over 2557904.96 frames. utt_duration=1238 frames, utt_pad_proportion=0.05241, over 8273.35 utterances.], batch size: 37, lr: 2.68e-02, grad_scale: 8.0 2023-03-07 15:35:46,517 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-07 15:36:19,418 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:36:42,522 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:37:01,520 INFO [train2.py:809] (3/4) Epoch 4, batch 350, loss[ctc_loss=0.2199, att_loss=0.3103, loss=0.2922, over 16638.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004632, over 47.00 utterances.], tot_loss[ctc_loss=0.1908, att_loss=0.2969, loss=0.2757, over 2718322.08 frames. utt_duration=1247 frames, utt_pad_proportion=0.05089, over 8727.63 utterances.], batch size: 47, lr: 2.67e-02, grad_scale: 8.0 2023-03-07 15:37:03,267 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:37:06,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 4.027e+02 4.741e+02 6.835e+02 1.897e+03, threshold=9.483e+02, percent-clipped=6.0 2023-03-07 15:37:46,471 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5129, 1.0868, 1.9352, 1.2863, 1.7605, 1.1952, 1.6206, 1.6447], device='cuda:3'), covar=tensor([0.0460, 0.1497, 0.1233, 0.1006, 0.0670, 0.1328, 0.1065, 0.0826], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0092, 0.0092, 0.0075, 0.0074, 0.0080, 0.0092, 0.0095], device='cuda:3'), out_proj_covar=tensor([3.9428e-05, 5.0572e-05, 4.8295e-05, 4.2262e-05, 4.1809e-05, 4.2990e-05, 4.2976e-05, 4.3023e-05], device='cuda:3') 2023-03-07 15:37:58,246 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:37:59,317 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-07 15:37:59,873 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:38:20,465 INFO [train2.py:809] (3/4) Epoch 4, batch 400, loss[ctc_loss=0.1348, att_loss=0.2636, loss=0.2379, over 16136.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005313, over 42.00 utterances.], tot_loss[ctc_loss=0.1895, att_loss=0.2967, loss=0.2753, over 2841804.19 frames. utt_duration=1228 frames, utt_pad_proportion=0.05655, over 9268.31 utterances.], batch size: 42, lr: 2.67e-02, grad_scale: 8.0 2023-03-07 15:38:23,735 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:38:52,274 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4287, 4.4107, 4.3373, 4.4708, 4.9512, 4.5706, 4.1425, 1.9692], device='cuda:3'), covar=tensor([0.0265, 0.0448, 0.0298, 0.0110, 0.0933, 0.0251, 0.0418, 0.3690], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0114, 0.0108, 0.0110, 0.0234, 0.0136, 0.0103, 0.0248], device='cuda:3'), out_proj_covar=tensor([1.1800e-04, 9.7008e-05, 9.4572e-05, 1.0334e-04, 2.1404e-04, 1.1925e-04, 9.6710e-05, 2.1530e-04], device='cuda:3') 2023-03-07 15:39:39,638 INFO [train2.py:809] (3/4) Epoch 4, batch 450, loss[ctc_loss=0.2091, att_loss=0.2898, loss=0.2736, over 16006.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.006789, over 40.00 utterances.], tot_loss[ctc_loss=0.1884, att_loss=0.2959, loss=0.2744, over 2946317.73 frames. utt_duration=1252 frames, utt_pad_proportion=0.04875, over 9421.45 utterances.], batch size: 40, lr: 2.66e-02, grad_scale: 8.0 2023-03-07 15:39:39,754 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:39:40,121 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7251, 2.3279, 5.1151, 3.8114, 3.5117, 4.3344, 4.6382, 4.5003], device='cuda:3'), covar=tensor([0.0163, 0.1795, 0.0249, 0.1277, 0.2177, 0.0364, 0.0230, 0.0343], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0223, 0.0118, 0.0282, 0.0329, 0.0172, 0.0110, 0.0122], device='cuda:3'), out_proj_covar=tensor([9.9515e-05, 1.6999e-04, 9.6221e-05, 2.2240e-04, 2.4802e-04, 1.3905e-04, 9.3105e-05, 1.0337e-04], device='cuda:3') 2023-03-07 15:39:44,236 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.731e+02 4.738e+02 5.685e+02 9.673e+02, threshold=9.477e+02, percent-clipped=1.0 2023-03-07 15:40:18,280 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5437, 2.8330, 3.8059, 2.5823, 3.6194, 4.6405, 4.4243, 3.3097], device='cuda:3'), covar=tensor([0.0286, 0.1426, 0.0592, 0.1361, 0.0782, 0.0209, 0.0379, 0.1172], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0177, 0.0161, 0.0166, 0.0181, 0.0127, 0.0128, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:40:25,800 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:40:48,280 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:40:58,418 INFO [train2.py:809] (3/4) Epoch 4, batch 500, loss[ctc_loss=0.2101, att_loss=0.3204, loss=0.2984, over 17128.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01392, over 56.00 utterances.], tot_loss[ctc_loss=0.1895, att_loss=0.2968, loss=0.2754, over 3027466.82 frames. utt_duration=1239 frames, utt_pad_proportion=0.04935, over 9784.60 utterances.], batch size: 56, lr: 2.66e-02, grad_scale: 8.0 2023-03-07 15:41:28,703 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:42:09,410 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2778, 2.7190, 3.5510, 2.1017, 3.1678, 4.2034, 4.1069, 3.2072], device='cuda:3'), covar=tensor([0.0333, 0.1494, 0.0892, 0.1640, 0.1094, 0.0368, 0.0415, 0.1159], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0176, 0.0159, 0.0168, 0.0180, 0.0129, 0.0126, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:42:15,279 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:42:16,490 INFO [train2.py:809] (3/4) Epoch 4, batch 550, loss[ctc_loss=0.1845, att_loss=0.2791, loss=0.2601, over 16188.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006362, over 41.00 utterances.], tot_loss[ctc_loss=0.1894, att_loss=0.2965, loss=0.2751, over 3087799.97 frames. utt_duration=1258 frames, utt_pad_proportion=0.04457, over 9826.94 utterances.], batch size: 41, lr: 2.65e-02, grad_scale: 8.0 2023-03-07 15:42:21,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.239e+02 3.520e+02 4.552e+02 5.513e+02 1.151e+03, threshold=9.104e+02, percent-clipped=4.0 2023-03-07 15:42:23,587 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:42:43,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:43:31,177 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:43:35,782 INFO [train2.py:809] (3/4) Epoch 4, batch 600, loss[ctc_loss=0.1913, att_loss=0.3046, loss=0.282, over 16485.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005499, over 46.00 utterances.], tot_loss[ctc_loss=0.1885, att_loss=0.2957, loss=0.2743, over 3127545.64 frames. utt_duration=1251 frames, utt_pad_proportion=0.04866, over 10010.60 utterances.], batch size: 46, lr: 2.65e-02, grad_scale: 8.0 2023-03-07 15:44:13,049 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:44:55,222 INFO [train2.py:809] (3/4) Epoch 4, batch 650, loss[ctc_loss=0.1676, att_loss=0.2789, loss=0.2567, over 16406.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007495, over 44.00 utterances.], tot_loss[ctc_loss=0.1889, att_loss=0.2961, loss=0.2747, over 3162833.30 frames. utt_duration=1223 frames, utt_pad_proportion=0.05596, over 10361.14 utterances.], batch size: 44, lr: 2.65e-02, grad_scale: 8.0 2023-03-07 15:44:57,691 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:45:00,416 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 3.647e+02 4.227e+02 5.409e+02 1.351e+03, threshold=8.454e+02, percent-clipped=4.0 2023-03-07 15:45:06,949 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2831, 3.8051, 3.1002, 3.2812, 3.9007, 3.5237, 2.4741, 4.3939], device='cuda:3'), covar=tensor([0.1254, 0.0345, 0.1187, 0.0768, 0.0468, 0.0697, 0.1083, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0102, 0.0157, 0.0126, 0.0120, 0.0151, 0.0131, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:45:29,073 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:45:30,867 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7999, 3.9451, 3.2158, 3.5093, 3.9836, 3.7495, 2.3823, 4.7970], device='cuda:3'), covar=tensor([0.0941, 0.0324, 0.1137, 0.0664, 0.0454, 0.0677, 0.1223, 0.0141], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0102, 0.0157, 0.0124, 0.0120, 0.0150, 0.0130, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 15:45:53,192 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:46:13,055 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:46:14,656 INFO [train2.py:809] (3/4) Epoch 4, batch 700, loss[ctc_loss=0.1771, att_loss=0.2808, loss=0.26, over 15760.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.00939, over 38.00 utterances.], tot_loss[ctc_loss=0.1886, att_loss=0.2956, loss=0.2742, over 3180433.35 frames. utt_duration=1225 frames, utt_pad_proportion=0.05742, over 10397.19 utterances.], batch size: 38, lr: 2.64e-02, grad_scale: 8.0 2023-03-07 15:46:45,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-03-07 15:47:09,144 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:47:34,090 INFO [train2.py:809] (3/4) Epoch 4, batch 750, loss[ctc_loss=0.1814, att_loss=0.301, loss=0.2771, over 17005.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.008618, over 51.00 utterances.], tot_loss[ctc_loss=0.1871, att_loss=0.294, loss=0.2727, over 3201138.23 frames. utt_duration=1251 frames, utt_pad_proportion=0.05118, over 10245.41 utterances.], batch size: 51, lr: 2.64e-02, grad_scale: 8.0 2023-03-07 15:47:38,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 3.712e+02 4.325e+02 6.002e+02 1.306e+03, threshold=8.650e+02, percent-clipped=6.0 2023-03-07 15:47:51,196 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-07 15:48:19,071 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:48:52,752 INFO [train2.py:809] (3/4) Epoch 4, batch 800, loss[ctc_loss=0.2117, att_loss=0.3228, loss=0.3006, over 17125.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01382, over 56.00 utterances.], tot_loss[ctc_loss=0.1856, att_loss=0.2932, loss=0.2717, over 3218404.46 frames. utt_duration=1276 frames, utt_pad_proportion=0.04658, over 10101.79 utterances.], batch size: 56, lr: 2.63e-02, grad_scale: 8.0 2023-03-07 15:49:08,750 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:49:21,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-07 15:49:35,441 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:50:11,406 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:50:12,734 INFO [train2.py:809] (3/4) Epoch 4, batch 850, loss[ctc_loss=0.1858, att_loss=0.2807, loss=0.2617, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006479, over 40.00 utterances.], tot_loss[ctc_loss=0.188, att_loss=0.2946, loss=0.2733, over 3223471.65 frames. utt_duration=1213 frames, utt_pad_proportion=0.06327, over 10643.90 utterances.], batch size: 40, lr: 2.63e-02, grad_scale: 8.0 2023-03-07 15:50:17,379 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.255e+02 3.893e+02 4.938e+02 6.078e+02 1.799e+03, threshold=9.877e+02, percent-clipped=7.0 2023-03-07 15:50:46,145 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 15:51:32,705 INFO [train2.py:809] (3/4) Epoch 4, batch 900, loss[ctc_loss=0.1744, att_loss=0.2952, loss=0.2711, over 16776.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005153, over 48.00 utterances.], tot_loss[ctc_loss=0.1874, att_loss=0.2948, loss=0.2733, over 3236054.77 frames. utt_duration=1224 frames, utt_pad_proportion=0.05953, over 10588.30 utterances.], batch size: 48, lr: 2.62e-02, grad_scale: 16.0 2023-03-07 15:52:53,758 INFO [train2.py:809] (3/4) Epoch 4, batch 950, loss[ctc_loss=0.1417, att_loss=0.2762, loss=0.2493, over 16628.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005304, over 47.00 utterances.], tot_loss[ctc_loss=0.1862, att_loss=0.2941, loss=0.2725, over 3247625.50 frames. utt_duration=1227 frames, utt_pad_proportion=0.05849, over 10604.30 utterances.], batch size: 47, lr: 2.62e-02, grad_scale: 16.0 2023-03-07 15:52:58,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.008e+02 3.388e+02 4.239e+02 5.070e+02 1.281e+03, threshold=8.479e+02, percent-clipped=6.0 2023-03-07 15:53:24,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-07 15:54:15,082 INFO [train2.py:809] (3/4) Epoch 4, batch 1000, loss[ctc_loss=0.1617, att_loss=0.272, loss=0.2499, over 16184.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006577, over 41.00 utterances.], tot_loss[ctc_loss=0.1868, att_loss=0.294, loss=0.2725, over 3252242.89 frames. utt_duration=1238 frames, utt_pad_proportion=0.05656, over 10521.77 utterances.], batch size: 41, lr: 2.61e-02, grad_scale: 8.0 2023-03-07 15:54:38,436 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8093, 2.9281, 5.1403, 4.0106, 3.5160, 4.5653, 4.7798, 4.8004], device='cuda:3'), covar=tensor([0.0103, 0.1480, 0.0249, 0.1129, 0.2099, 0.0256, 0.0176, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0224, 0.0121, 0.0278, 0.0328, 0.0169, 0.0110, 0.0120], device='cuda:3'), out_proj_covar=tensor([9.8376e-05, 1.7142e-04, 9.9357e-05, 2.2026e-04, 2.4871e-04, 1.3848e-04, 9.3684e-05, 1.0356e-04], device='cuda:3') 2023-03-07 15:55:37,208 INFO [train2.py:809] (3/4) Epoch 4, batch 1050, loss[ctc_loss=0.2271, att_loss=0.3084, loss=0.2922, over 15946.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007031, over 41.00 utterances.], tot_loss[ctc_loss=0.1877, att_loss=0.2944, loss=0.273, over 3256319.73 frames. utt_duration=1246 frames, utt_pad_proportion=0.05575, over 10466.56 utterances.], batch size: 41, lr: 2.61e-02, grad_scale: 8.0 2023-03-07 15:55:43,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.497e+02 3.995e+02 4.964e+02 6.656e+02 1.651e+03, threshold=9.928e+02, percent-clipped=9.0 2023-03-07 15:56:58,336 INFO [train2.py:809] (3/4) Epoch 4, batch 1100, loss[ctc_loss=0.2467, att_loss=0.3282, loss=0.3119, over 14450.00 frames. utt_duration=397.4 frames, utt_pad_proportion=0.3064, over 146.00 utterances.], tot_loss[ctc_loss=0.187, att_loss=0.2938, loss=0.2724, over 3252449.10 frames. utt_duration=1218 frames, utt_pad_proportion=0.06404, over 10695.47 utterances.], batch size: 146, lr: 2.61e-02, grad_scale: 8.0 2023-03-07 15:58:18,891 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:58:20,172 INFO [train2.py:809] (3/4) Epoch 4, batch 1150, loss[ctc_loss=0.1309, att_loss=0.2562, loss=0.2311, over 15946.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007415, over 41.00 utterances.], tot_loss[ctc_loss=0.1874, att_loss=0.2938, loss=0.2725, over 3257440.04 frames. utt_duration=1216 frames, utt_pad_proportion=0.06452, over 10727.61 utterances.], batch size: 41, lr: 2.60e-02, grad_scale: 8.0 2023-03-07 15:58:26,385 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.646e+02 4.698e+02 5.581e+02 1.233e+03, threshold=9.396e+02, percent-clipped=4.0 2023-03-07 15:58:46,218 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 15:59:08,250 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4156, 5.2029, 5.1115, 4.2310, 2.1174, 2.9202, 5.3536, 3.6770], device='cuda:3'), covar=tensor([0.0417, 0.0173, 0.0191, 0.0800, 0.7281, 0.2435, 0.0263, 0.2226], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0146, 0.0165, 0.0185, 0.0386, 0.0310, 0.0152, 0.0280], device='cuda:3'), out_proj_covar=tensor([1.3989e-04, 7.4416e-05, 8.7554e-05, 8.9749e-05, 1.9606e-04, 1.5426e-04, 7.8012e-05, 1.5293e-04], device='cuda:3') 2023-03-07 15:59:09,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2986, 5.0964, 5.0863, 4.1875, 1.9011, 2.6522, 5.2220, 3.6109], device='cuda:3'), covar=tensor([0.0404, 0.0138, 0.0146, 0.0684, 0.7667, 0.2698, 0.0176, 0.2172], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0145, 0.0165, 0.0185, 0.0386, 0.0309, 0.0152, 0.0279], device='cuda:3'), out_proj_covar=tensor([1.3962e-04, 7.4285e-05, 8.7394e-05, 8.9639e-05, 1.9569e-04, 1.5404e-04, 7.7865e-05, 1.5261e-04], device='cuda:3') 2023-03-07 15:59:35,982 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 15:59:40,595 INFO [train2.py:809] (3/4) Epoch 4, batch 1200, loss[ctc_loss=0.197, att_loss=0.312, loss=0.289, over 17017.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008166, over 51.00 utterances.], tot_loss[ctc_loss=0.1866, att_loss=0.2935, loss=0.2721, over 3259272.40 frames. utt_duration=1221 frames, utt_pad_proportion=0.06172, over 10686.54 utterances.], batch size: 51, lr: 2.60e-02, grad_scale: 8.0 2023-03-07 16:01:01,002 INFO [train2.py:809] (3/4) Epoch 4, batch 1250, loss[ctc_loss=0.1357, att_loss=0.2452, loss=0.2233, over 15882.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.008869, over 39.00 utterances.], tot_loss[ctc_loss=0.1849, att_loss=0.292, loss=0.2706, over 3254012.25 frames. utt_duration=1242 frames, utt_pad_proportion=0.05827, over 10490.54 utterances.], batch size: 39, lr: 2.59e-02, grad_scale: 8.0 2023-03-07 16:01:07,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.553e+02 3.747e+02 4.896e+02 5.785e+02 2.345e+03, threshold=9.791e+02, percent-clipped=4.0 2023-03-07 16:01:40,932 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7880, 2.2349, 3.4473, 2.1134, 3.2690, 4.0606, 4.0214, 2.6509], device='cuda:3'), covar=tensor([0.0590, 0.2136, 0.0774, 0.1988, 0.0933, 0.0450, 0.0403, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0184, 0.0168, 0.0168, 0.0181, 0.0146, 0.0134, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 16:02:21,492 INFO [train2.py:809] (3/4) Epoch 4, batch 1300, loss[ctc_loss=0.1657, att_loss=0.2871, loss=0.2628, over 16874.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007097, over 49.00 utterances.], tot_loss[ctc_loss=0.184, att_loss=0.2914, loss=0.27, over 3256570.71 frames. utt_duration=1234 frames, utt_pad_proportion=0.06117, over 10572.92 utterances.], batch size: 49, lr: 2.59e-02, grad_scale: 8.0 2023-03-07 16:03:41,844 INFO [train2.py:809] (3/4) Epoch 4, batch 1350, loss[ctc_loss=0.2019, att_loss=0.2817, loss=0.2657, over 15749.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.008929, over 38.00 utterances.], tot_loss[ctc_loss=0.186, att_loss=0.2927, loss=0.2714, over 3260775.15 frames. utt_duration=1207 frames, utt_pad_proportion=0.06617, over 10815.52 utterances.], batch size: 38, lr: 2.58e-02, grad_scale: 8.0 2023-03-07 16:03:44,590 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6606, 4.9895, 4.6277, 4.4031, 5.4198, 5.0620, 4.4919, 2.9815], device='cuda:3'), covar=tensor([0.0230, 0.0242, 0.0197, 0.0326, 0.0894, 0.0139, 0.0307, 0.2306], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0111, 0.0106, 0.0109, 0.0237, 0.0130, 0.0102, 0.0244], device='cuda:3'), out_proj_covar=tensor([1.1644e-04, 9.8340e-05, 9.6028e-05, 1.0355e-04, 2.1855e-04, 1.1823e-04, 9.7139e-05, 2.1581e-04], device='cuda:3') 2023-03-07 16:03:48,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 3.582e+02 4.373e+02 5.653e+02 1.802e+03, threshold=8.745e+02, percent-clipped=3.0 2023-03-07 16:03:56,253 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-07 16:04:48,050 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1503, 2.3778, 3.7485, 2.2796, 3.0909, 4.2090, 4.1504, 2.8847], device='cuda:3'), covar=tensor([0.0424, 0.1833, 0.0689, 0.1600, 0.1107, 0.0461, 0.0414, 0.1548], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0185, 0.0168, 0.0170, 0.0183, 0.0143, 0.0132, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 16:04:54,737 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2023-03-07 16:05:02,320 INFO [train2.py:809] (3/4) Epoch 4, batch 1400, loss[ctc_loss=0.1435, att_loss=0.2685, loss=0.2435, over 16776.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005889, over 48.00 utterances.], tot_loss[ctc_loss=0.1852, att_loss=0.2925, loss=0.2711, over 3263286.78 frames. utt_duration=1207 frames, utt_pad_proportion=0.06502, over 10829.67 utterances.], batch size: 48, lr: 2.58e-02, grad_scale: 8.0 2023-03-07 16:05:35,160 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1138, 5.0796, 4.9540, 3.3942, 4.9859, 4.1028, 4.6453, 2.6210], device='cuda:3'), covar=tensor([0.0111, 0.0074, 0.0237, 0.0634, 0.0088, 0.0180, 0.0187, 0.1392], device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0051, 0.0045, 0.0081, 0.0049, 0.0060, 0.0069, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2023-03-07 16:05:44,172 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.77 vs. limit=5.0 2023-03-07 16:06:11,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-07 16:06:22,103 INFO [train2.py:809] (3/4) Epoch 4, batch 1450, loss[ctc_loss=0.152, att_loss=0.2616, loss=0.2396, over 15614.00 frames. utt_duration=1689 frames, utt_pad_proportion=0.01026, over 37.00 utterances.], tot_loss[ctc_loss=0.185, att_loss=0.2928, loss=0.2713, over 3273135.47 frames. utt_duration=1234 frames, utt_pad_proportion=0.05643, over 10626.35 utterances.], batch size: 37, lr: 2.58e-02, grad_scale: 8.0 2023-03-07 16:06:28,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.701e+02 3.999e+02 4.753e+02 5.960e+02 1.214e+03, threshold=9.507e+02, percent-clipped=6.0 2023-03-07 16:06:38,009 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8991, 5.3900, 4.9268, 5.4390, 4.7246, 5.0300, 5.6249, 5.3107], device='cuda:3'), covar=tensor([0.0321, 0.0182, 0.0521, 0.0125, 0.0380, 0.0154, 0.0161, 0.0136], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0144, 0.0179, 0.0113, 0.0162, 0.0114, 0.0138, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 16:06:47,420 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:07:42,520 INFO [train2.py:809] (3/4) Epoch 4, batch 1500, loss[ctc_loss=0.1831, att_loss=0.2958, loss=0.2732, over 16551.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005642, over 45.00 utterances.], tot_loss[ctc_loss=0.1842, att_loss=0.2926, loss=0.2709, over 3274429.92 frames. utt_duration=1239 frames, utt_pad_proportion=0.05452, over 10587.85 utterances.], batch size: 45, lr: 2.57e-02, grad_scale: 8.0 2023-03-07 16:08:04,753 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:08:16,727 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2023-03-07 16:09:03,077 INFO [train2.py:809] (3/4) Epoch 4, batch 1550, loss[ctc_loss=0.1613, att_loss=0.2667, loss=0.2457, over 15487.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009488, over 36.00 utterances.], tot_loss[ctc_loss=0.1856, att_loss=0.2937, loss=0.2721, over 3267813.22 frames. utt_duration=1204 frames, utt_pad_proportion=0.06528, over 10866.06 utterances.], batch size: 36, lr: 2.57e-02, grad_scale: 8.0 2023-03-07 16:09:09,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 3.814e+02 4.608e+02 5.940e+02 1.228e+03, threshold=9.215e+02, percent-clipped=5.0 2023-03-07 16:09:13,287 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2023-03-07 16:09:22,332 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-03-07 16:10:23,310 INFO [train2.py:809] (3/4) Epoch 4, batch 1600, loss[ctc_loss=0.213, att_loss=0.2796, loss=0.2663, over 14522.00 frames. utt_duration=1817 frames, utt_pad_proportion=0.0291, over 32.00 utterances.], tot_loss[ctc_loss=0.1857, att_loss=0.2934, loss=0.2718, over 3262717.58 frames. utt_duration=1208 frames, utt_pad_proportion=0.0662, over 10815.54 utterances.], batch size: 32, lr: 2.56e-02, grad_scale: 8.0 2023-03-07 16:10:28,788 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-07 16:11:17,943 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-07 16:11:42,752 INFO [train2.py:809] (3/4) Epoch 4, batch 1650, loss[ctc_loss=0.1548, att_loss=0.2531, loss=0.2334, over 14497.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.0408, over 32.00 utterances.], tot_loss[ctc_loss=0.1854, att_loss=0.2939, loss=0.2722, over 3276499.45 frames. utt_duration=1229 frames, utt_pad_proportion=0.05767, over 10673.01 utterances.], batch size: 32, lr: 2.56e-02, grad_scale: 8.0 2023-03-07 16:11:48,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.288e+02 3.919e+02 5.311e+02 6.584e+02 2.912e+03, threshold=1.062e+03, percent-clipped=12.0 2023-03-07 16:13:01,505 INFO [train2.py:809] (3/4) Epoch 4, batch 1700, loss[ctc_loss=0.1302, att_loss=0.2494, loss=0.2256, over 15653.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.007647, over 37.00 utterances.], tot_loss[ctc_loss=0.1837, att_loss=0.2924, loss=0.2707, over 3276811.22 frames. utt_duration=1248 frames, utt_pad_proportion=0.05194, over 10516.55 utterances.], batch size: 37, lr: 2.55e-02, grad_scale: 8.0 2023-03-07 16:13:57,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9893, 5.2057, 5.5573, 5.6502, 5.1595, 5.9567, 5.1168, 5.9837], device='cuda:3'), covar=tensor([0.0567, 0.0622, 0.0513, 0.0520, 0.1867, 0.0609, 0.0521, 0.0444], device='cuda:3'), in_proj_covar=tensor([0.0430, 0.0278, 0.0285, 0.0335, 0.0497, 0.0285, 0.0247, 0.0307], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 16:14:03,027 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-07 16:14:20,719 INFO [train2.py:809] (3/4) Epoch 4, batch 1750, loss[ctc_loss=0.2038, att_loss=0.3085, loss=0.2875, over 17046.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009876, over 53.00 utterances.], tot_loss[ctc_loss=0.1839, att_loss=0.2927, loss=0.271, over 3283595.20 frames. utt_duration=1237 frames, utt_pad_proportion=0.05202, over 10631.43 utterances.], batch size: 53, lr: 2.55e-02, grad_scale: 8.0 2023-03-07 16:14:27,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.465e+02 3.709e+02 4.336e+02 5.273e+02 1.516e+03, threshold=8.672e+02, percent-clipped=2.0 2023-03-07 16:15:15,167 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:15:39,467 INFO [train2.py:809] (3/4) Epoch 4, batch 1800, loss[ctc_loss=0.1522, att_loss=0.2851, loss=0.2586, over 16555.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.004745, over 45.00 utterances.], tot_loss[ctc_loss=0.1832, att_loss=0.2926, loss=0.2707, over 3287259.54 frames. utt_duration=1251 frames, utt_pad_proportion=0.04713, over 10524.79 utterances.], batch size: 45, lr: 2.55e-02, grad_scale: 8.0 2023-03-07 16:16:51,921 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 16:16:59,426 INFO [train2.py:809] (3/4) Epoch 4, batch 1850, loss[ctc_loss=0.183, att_loss=0.2857, loss=0.2651, over 16185.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006485, over 41.00 utterances.], tot_loss[ctc_loss=0.1824, att_loss=0.2915, loss=0.2697, over 3282049.25 frames. utt_duration=1249 frames, utt_pad_proportion=0.04981, over 10522.04 utterances.], batch size: 41, lr: 2.54e-02, grad_scale: 8.0 2023-03-07 16:17:05,503 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.641e+02 3.804e+02 4.653e+02 6.004e+02 1.107e+03, threshold=9.305e+02, percent-clipped=7.0 2023-03-07 16:17:28,405 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2023-03-07 16:18:19,883 INFO [train2.py:809] (3/4) Epoch 4, batch 1900, loss[ctc_loss=0.2282, att_loss=0.3037, loss=0.2886, over 15880.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009658, over 39.00 utterances.], tot_loss[ctc_loss=0.1827, att_loss=0.2925, loss=0.2706, over 3289060.33 frames. utt_duration=1236 frames, utt_pad_proportion=0.05189, over 10660.22 utterances.], batch size: 39, lr: 2.54e-02, grad_scale: 8.0 2023-03-07 16:18:56,478 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4865, 4.9812, 4.5166, 5.1357, 4.4113, 4.9009, 5.2230, 4.9831], device='cuda:3'), covar=tensor([0.0356, 0.0312, 0.0634, 0.0173, 0.0407, 0.0156, 0.0228, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0149, 0.0184, 0.0116, 0.0163, 0.0118, 0.0143, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 16:19:39,438 INFO [train2.py:809] (3/4) Epoch 4, batch 1950, loss[ctc_loss=0.1784, att_loss=0.3, loss=0.2757, over 17065.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.00905, over 53.00 utterances.], tot_loss[ctc_loss=0.1827, att_loss=0.2926, loss=0.2706, over 3287086.36 frames. utt_duration=1246 frames, utt_pad_proportion=0.05009, over 10566.00 utterances.], batch size: 53, lr: 2.53e-02, grad_scale: 8.0 2023-03-07 16:19:45,526 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.392e+02 3.636e+02 4.382e+02 5.668e+02 1.213e+03, threshold=8.764e+02, percent-clipped=2.0 2023-03-07 16:21:00,024 INFO [train2.py:809] (3/4) Epoch 4, batch 2000, loss[ctc_loss=0.1697, att_loss=0.3036, loss=0.2769, over 17280.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.0123, over 55.00 utterances.], tot_loss[ctc_loss=0.182, att_loss=0.2922, loss=0.2701, over 3290115.96 frames. utt_duration=1253 frames, utt_pad_proportion=0.04816, over 10515.85 utterances.], batch size: 55, lr: 2.53e-02, grad_scale: 8.0 2023-03-07 16:22:23,890 INFO [train2.py:809] (3/4) Epoch 4, batch 2050, loss[ctc_loss=0.2257, att_loss=0.3175, loss=0.2992, over 17239.00 frames. utt_duration=874.5 frames, utt_pad_proportion=0.08433, over 79.00 utterances.], tot_loss[ctc_loss=0.1807, att_loss=0.2913, loss=0.2691, over 3287605.19 frames. utt_duration=1256 frames, utt_pad_proportion=0.04754, over 10479.95 utterances.], batch size: 79, lr: 2.53e-02, grad_scale: 8.0 2023-03-07 16:22:30,194 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.271e+02 3.689e+02 4.518e+02 6.015e+02 1.676e+03, threshold=9.035e+02, percent-clipped=8.0 2023-03-07 16:23:11,626 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-03-07 16:23:25,420 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3996, 4.9306, 4.6346, 4.6102, 4.9195, 4.7072, 4.3689, 4.8555], device='cuda:3'), covar=tensor([0.0103, 0.0112, 0.0091, 0.0120, 0.0093, 0.0080, 0.0294, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0049, 0.0050, 0.0037, 0.0037, 0.0045, 0.0067, 0.0063], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 16:23:43,382 INFO [train2.py:809] (3/4) Epoch 4, batch 2100, loss[ctc_loss=0.138, att_loss=0.2529, loss=0.2299, over 15771.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.00784, over 38.00 utterances.], tot_loss[ctc_loss=0.1826, att_loss=0.2927, loss=0.2706, over 3282413.38 frames. utt_duration=1225 frames, utt_pad_proportion=0.05744, over 10731.23 utterances.], batch size: 38, lr: 2.52e-02, grad_scale: 8.0 2023-03-07 16:24:14,238 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:24:47,824 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 16:25:03,072 INFO [train2.py:809] (3/4) Epoch 4, batch 2150, loss[ctc_loss=0.1465, att_loss=0.2521, loss=0.2309, over 15885.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009316, over 39.00 utterances.], tot_loss[ctc_loss=0.181, att_loss=0.2916, loss=0.2695, over 3277534.37 frames. utt_duration=1229 frames, utt_pad_proportion=0.05784, over 10684.61 utterances.], batch size: 39, lr: 2.52e-02, grad_scale: 8.0 2023-03-07 16:25:03,428 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:25:09,250 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 3.723e+02 4.436e+02 6.346e+02 3.042e+03, threshold=8.872e+02, percent-clipped=8.0 2023-03-07 16:25:10,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 16:25:30,701 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 16:25:51,833 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:26:23,570 INFO [train2.py:809] (3/4) Epoch 4, batch 2200, loss[ctc_loss=0.1458, att_loss=0.2507, loss=0.2297, over 15361.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01111, over 35.00 utterances.], tot_loss[ctc_loss=0.1808, att_loss=0.2915, loss=0.2694, over 3276929.57 frames. utt_duration=1226 frames, utt_pad_proportion=0.05971, over 10707.70 utterances.], batch size: 35, lr: 2.51e-02, grad_scale: 8.0 2023-03-07 16:26:41,880 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:27:09,492 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 16:27:44,782 INFO [train2.py:809] (3/4) Epoch 4, batch 2250, loss[ctc_loss=0.1861, att_loss=0.3044, loss=0.2808, over 16608.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.00629, over 47.00 utterances.], tot_loss[ctc_loss=0.1811, att_loss=0.2918, loss=0.2697, over 3277488.31 frames. utt_duration=1237 frames, utt_pad_proportion=0.05745, over 10614.28 utterances.], batch size: 47, lr: 2.51e-02, grad_scale: 8.0 2023-03-07 16:27:50,860 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.200e+02 3.354e+02 4.208e+02 5.126e+02 1.091e+03, threshold=8.417e+02, percent-clipped=2.0 2023-03-07 16:29:04,197 INFO [train2.py:809] (3/4) Epoch 4, batch 2300, loss[ctc_loss=0.166, att_loss=0.2858, loss=0.2618, over 16693.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005918, over 46.00 utterances.], tot_loss[ctc_loss=0.181, att_loss=0.2917, loss=0.2696, over 3268178.90 frames. utt_duration=1239 frames, utt_pad_proportion=0.0585, over 10567.12 utterances.], batch size: 46, lr: 2.51e-02, grad_scale: 8.0 2023-03-07 16:30:24,650 INFO [train2.py:809] (3/4) Epoch 4, batch 2350, loss[ctc_loss=0.1608, att_loss=0.2861, loss=0.2611, over 16758.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006999, over 48.00 utterances.], tot_loss[ctc_loss=0.1791, att_loss=0.2902, loss=0.268, over 3266579.80 frames. utt_duration=1240 frames, utt_pad_proportion=0.05934, over 10552.24 utterances.], batch size: 48, lr: 2.50e-02, grad_scale: 8.0 2023-03-07 16:30:31,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 3.860e+02 4.546e+02 5.797e+02 1.491e+03, threshold=9.092e+02, percent-clipped=6.0 2023-03-07 16:31:44,319 INFO [train2.py:809] (3/4) Epoch 4, batch 2400, loss[ctc_loss=0.1758, att_loss=0.3013, loss=0.2762, over 16755.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006485, over 48.00 utterances.], tot_loss[ctc_loss=0.1796, att_loss=0.2902, loss=0.2681, over 3263701.93 frames. utt_duration=1235 frames, utt_pad_proportion=0.06217, over 10587.38 utterances.], batch size: 48, lr: 2.50e-02, grad_scale: 8.0 2023-03-07 16:32:48,052 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:33:03,729 INFO [train2.py:809] (3/4) Epoch 4, batch 2450, loss[ctc_loss=0.1706, att_loss=0.2948, loss=0.27, over 17328.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02149, over 59.00 utterances.], tot_loss[ctc_loss=0.1802, att_loss=0.2903, loss=0.2683, over 3261781.26 frames. utt_duration=1230 frames, utt_pad_proportion=0.06344, over 10620.73 utterances.], batch size: 59, lr: 2.49e-02, grad_scale: 8.0 2023-03-07 16:33:09,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 3.936e+02 4.992e+02 6.745e+02 1.784e+03, threshold=9.985e+02, percent-clipped=4.0 2023-03-07 16:33:20,214 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:33:30,093 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7638, 1.9520, 1.7361, 0.8805, 1.5461, 1.2731, 1.4963, 1.2750], device='cuda:3'), covar=tensor([0.0320, 0.0852, 0.1292, 0.1165, 0.0788, 0.0898, 0.0850, 0.1013], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0084, 0.0093, 0.0076, 0.0075, 0.0073, 0.0084, 0.0089], device='cuda:3'), out_proj_covar=tensor([3.7665e-05, 4.3015e-05, 4.7541e-05, 4.1511e-05, 3.8108e-05, 4.0803e-05, 3.9782e-05, 4.2967e-05], device='cuda:3') 2023-03-07 16:33:44,302 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:34:04,203 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.22 vs. limit=5.0 2023-03-07 16:34:04,975 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:34:23,282 INFO [train2.py:809] (3/4) Epoch 4, batch 2500, loss[ctc_loss=0.1974, att_loss=0.3128, loss=0.2897, over 17329.00 frames. utt_duration=1006 frames, utt_pad_proportion=0.05085, over 69.00 utterances.], tot_loss[ctc_loss=0.181, att_loss=0.2914, loss=0.2693, over 3267117.48 frames. utt_duration=1203 frames, utt_pad_proportion=0.06731, over 10874.49 utterances.], batch size: 69, lr: 2.49e-02, grad_scale: 8.0 2023-03-07 16:34:33,354 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:34:57,598 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:35:00,403 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 16:35:17,975 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:35:28,920 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3764, 4.6220, 4.7685, 4.9111, 2.1142, 4.6080, 2.6098, 1.5847], device='cuda:3'), covar=tensor([0.0196, 0.0166, 0.0600, 0.0224, 0.3130, 0.0197, 0.1705, 0.2153], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0094, 0.0228, 0.0113, 0.0231, 0.0102, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([9.4551e-05, 8.9865e-05, 1.9515e-04, 1.0020e-04, 1.9271e-04, 9.4999e-05, 1.7790e-04, 1.6384e-04], device='cuda:3') 2023-03-07 16:35:33,499 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1080, 4.5015, 4.6169, 4.7827, 5.0681, 5.0676, 4.6797, 2.3478], device='cuda:3'), covar=tensor([0.0133, 0.0361, 0.0223, 0.0185, 0.1106, 0.0129, 0.0227, 0.2928], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0119, 0.0114, 0.0115, 0.0256, 0.0131, 0.0104, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 16:35:42,396 INFO [train2.py:809] (3/4) Epoch 4, batch 2550, loss[ctc_loss=0.1559, att_loss=0.2564, loss=0.2363, over 15403.00 frames. utt_duration=1762 frames, utt_pad_proportion=0.009079, over 35.00 utterances.], tot_loss[ctc_loss=0.1804, att_loss=0.2908, loss=0.2687, over 3273459.59 frames. utt_duration=1238 frames, utt_pad_proportion=0.05752, over 10591.68 utterances.], batch size: 35, lr: 2.49e-02, grad_scale: 8.0 2023-03-07 16:35:48,914 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.665e+02 3.493e+02 4.541e+02 5.557e+02 1.503e+03, threshold=9.083e+02, percent-clipped=2.0 2023-03-07 16:36:19,390 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7912, 5.2200, 5.0130, 5.0248, 5.2588, 5.1938, 5.0595, 4.7432], device='cuda:3'), covar=tensor([0.1005, 0.0350, 0.0201, 0.0468, 0.0251, 0.0248, 0.0205, 0.0244], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0212, 0.0146, 0.0189, 0.0238, 0.0258, 0.0195, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 16:36:50,523 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-07 16:36:54,432 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:37:02,065 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 16:37:02,527 INFO [train2.py:809] (3/4) Epoch 4, batch 2600, loss[ctc_loss=0.1846, att_loss=0.2995, loss=0.2765, over 16862.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.00814, over 49.00 utterances.], tot_loss[ctc_loss=0.1792, att_loss=0.2897, loss=0.2676, over 3273061.30 frames. utt_duration=1254 frames, utt_pad_proportion=0.05397, over 10454.49 utterances.], batch size: 49, lr: 2.48e-02, grad_scale: 8.0 2023-03-07 16:38:02,274 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-07 16:38:21,813 INFO [train2.py:809] (3/4) Epoch 4, batch 2650, loss[ctc_loss=0.1712, att_loss=0.2702, loss=0.2504, over 16019.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006045, over 40.00 utterances.], tot_loss[ctc_loss=0.1791, att_loss=0.29, loss=0.2678, over 3275919.84 frames. utt_duration=1268 frames, utt_pad_proportion=0.04964, over 10347.09 utterances.], batch size: 40, lr: 2.48e-02, grad_scale: 8.0 2023-03-07 16:38:24,135 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6608, 5.0522, 4.7020, 5.1327, 4.6193, 4.9163, 5.3867, 5.1061], device='cuda:3'), covar=tensor([0.0445, 0.0275, 0.0643, 0.0159, 0.0422, 0.0178, 0.0191, 0.0154], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0151, 0.0191, 0.0121, 0.0168, 0.0121, 0.0145, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 16:38:25,763 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1681, 2.5607, 3.3817, 2.3680, 3.1966, 4.3864, 4.0108, 2.8289], device='cuda:3'), covar=tensor([0.0387, 0.1717, 0.0951, 0.1564, 0.1037, 0.0366, 0.0599, 0.1593], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0195, 0.0182, 0.0176, 0.0193, 0.0163, 0.0142, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 16:38:28,424 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.304e+02 3.793e+02 4.783e+02 6.009e+02 1.165e+03, threshold=9.567e+02, percent-clipped=2.0 2023-03-07 16:38:35,108 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7172, 2.8190, 5.1293, 4.1520, 3.4678, 4.5184, 4.8825, 4.7740], device='cuda:3'), covar=tensor([0.0145, 0.1483, 0.0245, 0.0998, 0.1940, 0.0276, 0.0154, 0.0218], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0230, 0.0121, 0.0286, 0.0316, 0.0176, 0.0106, 0.0130], device='cuda:3'), out_proj_covar=tensor([1.0506e-04, 1.7804e-04, 1.0065e-04, 2.2701e-04, 2.4609e-04, 1.4420e-04, 8.9906e-05, 1.0891e-04], device='cuda:3') 2023-03-07 16:39:41,961 INFO [train2.py:809] (3/4) Epoch 4, batch 2700, loss[ctc_loss=0.1391, att_loss=0.2787, loss=0.2508, over 16271.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007813, over 43.00 utterances.], tot_loss[ctc_loss=0.1797, att_loss=0.2904, loss=0.2683, over 3274487.46 frames. utt_duration=1246 frames, utt_pad_proportion=0.05549, over 10525.87 utterances.], batch size: 43, lr: 2.48e-02, grad_scale: 8.0 2023-03-07 16:41:01,175 INFO [train2.py:809] (3/4) Epoch 4, batch 2750, loss[ctc_loss=0.1587, att_loss=0.2879, loss=0.2621, over 16247.00 frames. utt_duration=1513 frames, utt_pad_proportion=0.008671, over 43.00 utterances.], tot_loss[ctc_loss=0.1786, att_loss=0.2893, loss=0.2672, over 3276798.34 frames. utt_duration=1269 frames, utt_pad_proportion=0.04915, over 10344.50 utterances.], batch size: 43, lr: 2.47e-02, grad_scale: 8.0 2023-03-07 16:41:07,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.333e+02 3.898e+02 4.978e+02 5.848e+02 1.042e+03, threshold=9.955e+02, percent-clipped=2.0 2023-03-07 16:41:42,286 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:42:02,321 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:42:23,123 INFO [train2.py:809] (3/4) Epoch 4, batch 2800, loss[ctc_loss=0.1423, att_loss=0.2662, loss=0.2415, over 16013.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007446, over 40.00 utterances.], tot_loss[ctc_loss=0.1781, att_loss=0.2887, loss=0.2666, over 3263928.80 frames. utt_duration=1245 frames, utt_pad_proportion=0.05834, over 10501.08 utterances.], batch size: 40, lr: 2.47e-02, grad_scale: 8.0 2023-03-07 16:42:24,944 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8618, 5.2526, 5.0800, 5.1584, 5.3367, 5.2526, 5.0299, 4.8950], device='cuda:3'), covar=tensor([0.0892, 0.0319, 0.0185, 0.0357, 0.0215, 0.0229, 0.0208, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0202, 0.0143, 0.0183, 0.0229, 0.0251, 0.0191, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 16:42:32,806 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:42:48,941 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:43:00,439 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:43:00,643 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 16:43:06,255 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 16:43:41,706 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:43:44,569 INFO [train2.py:809] (3/4) Epoch 4, batch 2850, loss[ctc_loss=0.1892, att_loss=0.299, loss=0.2771, over 16883.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.0067, over 49.00 utterances.], tot_loss[ctc_loss=0.1769, att_loss=0.288, loss=0.2658, over 3272077.97 frames. utt_duration=1269 frames, utt_pad_proportion=0.05043, over 10323.97 utterances.], batch size: 49, lr: 2.46e-02, grad_scale: 8.0 2023-03-07 16:43:50,719 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 3.258e+02 4.280e+02 5.701e+02 1.148e+03, threshold=8.559e+02, percent-clipped=5.0 2023-03-07 16:43:50,891 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:44:18,604 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 16:44:48,687 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:45:04,280 INFO [train2.py:809] (3/4) Epoch 4, batch 2900, loss[ctc_loss=0.2003, att_loss=0.3083, loss=0.2867, over 17389.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03381, over 63.00 utterances.], tot_loss[ctc_loss=0.1779, att_loss=0.2889, loss=0.2667, over 3269648.44 frames. utt_duration=1259 frames, utt_pad_proportion=0.05376, over 10399.55 utterances.], batch size: 63, lr: 2.46e-02, grad_scale: 8.0 2023-03-07 16:45:35,959 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2215, 5.0099, 4.9008, 3.3424, 1.9677, 2.6616, 4.9264, 3.8386], device='cuda:3'), covar=tensor([0.0505, 0.0156, 0.0174, 0.1258, 0.6213, 0.2282, 0.0169, 0.1838], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0150, 0.0179, 0.0176, 0.0370, 0.0313, 0.0159, 0.0287], device='cuda:3'), out_proj_covar=tensor([1.3958e-04, 7.2116e-05, 8.7020e-05, 8.2792e-05, 1.8829e-04, 1.5236e-04, 7.5470e-05, 1.5337e-04], device='cuda:3') 2023-03-07 16:45:55,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-07 16:46:16,396 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5907, 4.9623, 4.4959, 5.0519, 4.4733, 4.8600, 5.1690, 4.9135], device='cuda:3'), covar=tensor([0.0359, 0.0259, 0.0698, 0.0157, 0.0416, 0.0182, 0.0247, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0154, 0.0196, 0.0121, 0.0171, 0.0120, 0.0147, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 16:46:25,731 INFO [train2.py:809] (3/4) Epoch 4, batch 2950, loss[ctc_loss=0.1634, att_loss=0.2952, loss=0.2688, over 16998.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.008763, over 51.00 utterances.], tot_loss[ctc_loss=0.1762, att_loss=0.2881, loss=0.2657, over 3268653.28 frames. utt_duration=1291 frames, utt_pad_proportion=0.04621, over 10143.16 utterances.], batch size: 51, lr: 2.46e-02, grad_scale: 8.0 2023-03-07 16:46:32,081 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 3.902e+02 4.882e+02 6.670e+02 1.203e+03, threshold=9.763e+02, percent-clipped=9.0 2023-03-07 16:46:45,780 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0411, 4.4922, 4.2164, 4.5283, 4.5210, 4.2343, 3.8856, 4.3385], device='cuda:3'), covar=tensor([0.0109, 0.0101, 0.0097, 0.0075, 0.0072, 0.0094, 0.0304, 0.0160], device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0048, 0.0050, 0.0037, 0.0036, 0.0046, 0.0067, 0.0063], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 16:46:53,682 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.6246, 1.7548, 2.5617, 2.6124, 2.6512, 2.8396, 1.9315, 1.2255], device='cuda:3'), covar=tensor([0.1329, 0.2689, 0.0974, 0.2959, 0.1378, 0.4512, 0.1614, 1.0248], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0067, 0.0065, 0.0081, 0.0065, 0.0077, 0.0068, 0.0103], device='cuda:3'), out_proj_covar=tensor([4.4052e-05, 4.1490e-05, 3.9746e-05, 5.4701e-05, 4.2199e-05, 5.4333e-05, 4.5115e-05, 7.4979e-05], device='cuda:3') 2023-03-07 16:47:46,082 INFO [train2.py:809] (3/4) Epoch 4, batch 3000, loss[ctc_loss=0.1647, att_loss=0.2803, loss=0.2572, over 16527.00 frames. utt_duration=1470 frames, utt_pad_proportion=0.006426, over 45.00 utterances.], tot_loss[ctc_loss=0.1748, att_loss=0.2873, loss=0.2648, over 3276007.70 frames. utt_duration=1293 frames, utt_pad_proportion=0.04285, over 10147.09 utterances.], batch size: 45, lr: 2.45e-02, grad_scale: 16.0 2023-03-07 16:47:46,082 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 16:48:00,418 INFO [train2.py:843] (3/4) Epoch 4, validation: ctc_loss=0.08782, att_loss=0.2564, loss=0.2227, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 16:48:00,418 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 16:49:20,421 INFO [train2.py:809] (3/4) Epoch 4, batch 3050, loss[ctc_loss=0.1532, att_loss=0.284, loss=0.2579, over 16758.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006323, over 48.00 utterances.], tot_loss[ctc_loss=0.175, att_loss=0.2874, loss=0.2649, over 3282694.03 frames. utt_duration=1307 frames, utt_pad_proportion=0.03775, over 10061.48 utterances.], batch size: 48, lr: 2.45e-02, grad_scale: 16.0 2023-03-07 16:49:26,540 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 3.805e+02 4.579e+02 6.197e+02 1.095e+03, threshold=9.158e+02, percent-clipped=2.0 2023-03-07 16:49:47,191 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-07 16:50:40,669 INFO [train2.py:809] (3/4) Epoch 4, batch 3100, loss[ctc_loss=0.1711, att_loss=0.3047, loss=0.278, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005618, over 47.00 utterances.], tot_loss[ctc_loss=0.1744, att_loss=0.2874, loss=0.2648, over 3283427.47 frames. utt_duration=1309 frames, utt_pad_proportion=0.03765, over 10043.68 utterances.], batch size: 47, lr: 2.45e-02, grad_scale: 16.0 2023-03-07 16:50:46,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-07 16:51:06,829 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:51:51,002 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:52:01,855 INFO [train2.py:809] (3/4) Epoch 4, batch 3150, loss[ctc_loss=0.177, att_loss=0.3032, loss=0.2779, over 17425.00 frames. utt_duration=883.8 frames, utt_pad_proportion=0.07357, over 79.00 utterances.], tot_loss[ctc_loss=0.1749, att_loss=0.2879, loss=0.2653, over 3285204.63 frames. utt_duration=1287 frames, utt_pad_proportion=0.04065, over 10220.20 utterances.], batch size: 79, lr: 2.44e-02, grad_scale: 16.0 2023-03-07 16:52:07,991 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.047e+02 3.829e+02 4.318e+02 5.478e+02 1.412e+03, threshold=8.637e+02, percent-clipped=4.0 2023-03-07 16:52:23,556 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:53:05,042 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:53:21,397 INFO [train2.py:809] (3/4) Epoch 4, batch 3200, loss[ctc_loss=0.2019, att_loss=0.3098, loss=0.2882, over 17272.00 frames. utt_duration=1003 frames, utt_pad_proportion=0.05495, over 69.00 utterances.], tot_loss[ctc_loss=0.1757, att_loss=0.2882, loss=0.2657, over 3288658.43 frames. utt_duration=1282 frames, utt_pad_proportion=0.03945, over 10269.76 utterances.], batch size: 69, lr: 2.44e-02, grad_scale: 16.0 2023-03-07 16:54:21,479 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:54:28,792 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 16:54:42,166 INFO [train2.py:809] (3/4) Epoch 4, batch 3250, loss[ctc_loss=0.1787, att_loss=0.2894, loss=0.2672, over 17171.00 frames. utt_duration=695.3 frames, utt_pad_proportion=0.1276, over 99.00 utterances.], tot_loss[ctc_loss=0.176, att_loss=0.2887, loss=0.2661, over 3285665.71 frames. utt_duration=1271 frames, utt_pad_proportion=0.04344, over 10355.44 utterances.], batch size: 99, lr: 2.44e-02, grad_scale: 16.0 2023-03-07 16:54:48,521 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.387e+02 3.575e+02 4.506e+02 5.643e+02 1.593e+03, threshold=9.012e+02, percent-clipped=6.0 2023-03-07 16:56:01,882 INFO [train2.py:809] (3/4) Epoch 4, batch 3300, loss[ctc_loss=0.1302, att_loss=0.2543, loss=0.2295, over 13650.00 frames. utt_duration=1822 frames, utt_pad_proportion=0.07014, over 30.00 utterances.], tot_loss[ctc_loss=0.1783, att_loss=0.2905, loss=0.268, over 3280809.30 frames. utt_duration=1210 frames, utt_pad_proportion=0.06037, over 10855.53 utterances.], batch size: 30, lr: 2.43e-02, grad_scale: 16.0 2023-03-07 16:56:06,919 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 16:56:47,204 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3743, 5.0069, 4.9400, 3.2351, 1.8934, 2.6220, 4.9484, 3.7044], device='cuda:3'), covar=tensor([0.0402, 0.0167, 0.0185, 0.1594, 0.7315, 0.2634, 0.0142, 0.1785], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0154, 0.0177, 0.0177, 0.0367, 0.0314, 0.0160, 0.0288], device='cuda:3'), out_proj_covar=tensor([1.3943e-04, 7.4209e-05, 8.6783e-05, 8.2875e-05, 1.8655e-04, 1.5253e-04, 7.5898e-05, 1.5296e-04], device='cuda:3') 2023-03-07 16:56:51,861 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8649, 4.8311, 4.6096, 4.9368, 5.0497, 4.7933, 4.5543, 2.0421], device='cuda:3'), covar=tensor([0.0256, 0.0290, 0.0266, 0.0183, 0.1750, 0.0219, 0.0324, 0.3385], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0119, 0.0116, 0.0119, 0.0267, 0.0127, 0.0109, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 16:57:22,461 INFO [train2.py:809] (3/4) Epoch 4, batch 3350, loss[ctc_loss=0.1393, att_loss=0.2495, loss=0.2275, over 14502.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.03542, over 32.00 utterances.], tot_loss[ctc_loss=0.1777, att_loss=0.2899, loss=0.2674, over 3275189.22 frames. utt_duration=1218 frames, utt_pad_proportion=0.06002, over 10771.79 utterances.], batch size: 32, lr: 2.43e-02, grad_scale: 16.0 2023-03-07 16:57:28,537 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 3.835e+02 4.773e+02 5.805e+02 1.964e+03, threshold=9.546e+02, percent-clipped=8.0 2023-03-07 16:58:42,387 INFO [train2.py:809] (3/4) Epoch 4, batch 3400, loss[ctc_loss=0.1817, att_loss=0.2917, loss=0.2697, over 16626.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005215, over 47.00 utterances.], tot_loss[ctc_loss=0.1749, att_loss=0.2876, loss=0.2651, over 3261436.73 frames. utt_duration=1249 frames, utt_pad_proportion=0.05534, over 10454.56 utterances.], batch size: 47, lr: 2.42e-02, grad_scale: 16.0 2023-03-07 16:59:51,089 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:00:02,646 INFO [train2.py:809] (3/4) Epoch 4, batch 3450, loss[ctc_loss=0.1913, att_loss=0.3088, loss=0.2853, over 17055.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009485, over 53.00 utterances.], tot_loss[ctc_loss=0.1741, att_loss=0.2877, loss=0.265, over 3271029.69 frames. utt_duration=1275 frames, utt_pad_proportion=0.04736, over 10270.31 utterances.], batch size: 53, lr: 2.42e-02, grad_scale: 8.0 2023-03-07 17:00:10,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.330e+02 3.236e+02 3.892e+02 5.378e+02 1.574e+03, threshold=7.784e+02, percent-clipped=3.0 2023-03-07 17:00:20,246 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3066, 4.4086, 4.0982, 4.4705, 4.5161, 4.4206, 4.0524, 2.0775], device='cuda:3'), covar=tensor([0.0189, 0.0214, 0.0212, 0.0095, 0.0968, 0.0153, 0.0303, 0.3330], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0116, 0.0113, 0.0117, 0.0260, 0.0123, 0.0107, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 17:00:55,200 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8190, 6.0225, 5.3698, 5.9557, 5.6628, 5.3711, 5.3744, 5.3759], device='cuda:3'), covar=tensor([0.1012, 0.0751, 0.0738, 0.0590, 0.0685, 0.1239, 0.2125, 0.1898], device='cuda:3'), in_proj_covar=tensor([0.0314, 0.0348, 0.0273, 0.0277, 0.0256, 0.0346, 0.0376, 0.0354], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:00:58,492 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4580, 5.0295, 4.7120, 4.6998, 5.0669, 4.8950, 4.6627, 4.5280], device='cuda:3'), covar=tensor([0.1212, 0.0463, 0.0255, 0.0604, 0.0307, 0.0303, 0.0283, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0210, 0.0150, 0.0191, 0.0237, 0.0260, 0.0198, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:01:07,809 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:01:22,736 INFO [train2.py:809] (3/4) Epoch 4, batch 3500, loss[ctc_loss=0.1391, att_loss=0.2595, loss=0.2354, over 16133.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005818, over 42.00 utterances.], tot_loss[ctc_loss=0.1753, att_loss=0.2886, loss=0.2659, over 3278726.44 frames. utt_duration=1257 frames, utt_pad_proportion=0.04937, over 10446.83 utterances.], batch size: 42, lr: 2.42e-02, grad_scale: 8.0 2023-03-07 17:02:28,178 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5655, 3.8123, 3.6601, 3.8744, 3.8884, 3.7548, 3.3576, 3.7857], device='cuda:3'), covar=tensor([0.0110, 0.0121, 0.0106, 0.0080, 0.0074, 0.0099, 0.0376, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0051, 0.0053, 0.0038, 0.0038, 0.0047, 0.0069, 0.0065], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:02:43,135 INFO [train2.py:809] (3/4) Epoch 4, batch 3550, loss[ctc_loss=0.1881, att_loss=0.3043, loss=0.2811, over 16413.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006906, over 44.00 utterances.], tot_loss[ctc_loss=0.1745, att_loss=0.2878, loss=0.2651, over 3279913.38 frames. utt_duration=1258 frames, utt_pad_proportion=0.04967, over 10438.66 utterances.], batch size: 44, lr: 2.41e-02, grad_scale: 8.0 2023-03-07 17:02:50,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.279e+02 3.812e+02 4.413e+02 5.065e+02 1.151e+03, threshold=8.826e+02, percent-clipped=3.0 2023-03-07 17:02:59,328 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 17:04:00,642 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 17:04:03,538 INFO [train2.py:809] (3/4) Epoch 4, batch 3600, loss[ctc_loss=0.1497, att_loss=0.2625, loss=0.2399, over 16175.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006596, over 41.00 utterances.], tot_loss[ctc_loss=0.1748, att_loss=0.2877, loss=0.2651, over 3281689.26 frames. utt_duration=1266 frames, utt_pad_proportion=0.04789, over 10384.42 utterances.], batch size: 41, lr: 2.41e-02, grad_scale: 8.0 2023-03-07 17:05:24,928 INFO [train2.py:809] (3/4) Epoch 4, batch 3650, loss[ctc_loss=0.1487, att_loss=0.291, loss=0.2626, over 17283.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02468, over 59.00 utterances.], tot_loss[ctc_loss=0.1734, att_loss=0.287, loss=0.2643, over 3280411.18 frames. utt_duration=1277 frames, utt_pad_proportion=0.04435, over 10287.10 utterances.], batch size: 59, lr: 2.41e-02, grad_scale: 8.0 2023-03-07 17:05:32,893 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.822e+02 4.406e+02 5.174e+02 1.169e+03, threshold=8.811e+02, percent-clipped=2.0 2023-03-07 17:06:05,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-07 17:06:06,864 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6824, 3.3710, 3.6035, 3.6302, 2.3012, 3.7694, 2.6521, 2.1962], device='cuda:3'), covar=tensor([0.0213, 0.0211, 0.0678, 0.0399, 0.2448, 0.0241, 0.1346, 0.1468], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0094, 0.0232, 0.0114, 0.0223, 0.0095, 0.0210, 0.0193], device='cuda:3'), out_proj_covar=tensor([9.1396e-05, 8.9960e-05, 2.0117e-04, 1.0191e-04, 1.9148e-04, 8.9249e-05, 1.7855e-04, 1.6574e-04], device='cuda:3') 2023-03-07 17:06:40,486 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5186, 2.1560, 2.8653, 4.2347, 4.1333, 4.2250, 2.7913, 1.6990], device='cuda:3'), covar=tensor([0.0620, 0.2927, 0.1517, 0.0477, 0.0357, 0.0219, 0.2000, 0.3209], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0190, 0.0184, 0.0137, 0.0125, 0.0110, 0.0190, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:06:44,838 INFO [train2.py:809] (3/4) Epoch 4, batch 3700, loss[ctc_loss=0.161, att_loss=0.2865, loss=0.2614, over 16127.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006372, over 42.00 utterances.], tot_loss[ctc_loss=0.173, att_loss=0.2866, loss=0.2639, over 3267491.95 frames. utt_duration=1281 frames, utt_pad_proportion=0.04697, over 10214.75 utterances.], batch size: 42, lr: 2.40e-02, grad_scale: 8.0 2023-03-07 17:08:05,619 INFO [train2.py:809] (3/4) Epoch 4, batch 3750, loss[ctc_loss=0.1357, att_loss=0.2495, loss=0.2267, over 15364.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.009293, over 35.00 utterances.], tot_loss[ctc_loss=0.1727, att_loss=0.2867, loss=0.2639, over 3260542.13 frames. utt_duration=1272 frames, utt_pad_proportion=0.05125, over 10262.40 utterances.], batch size: 35, lr: 2.40e-02, grad_scale: 8.0 2023-03-07 17:08:13,130 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.429e+02 3.564e+02 4.340e+02 5.643e+02 1.443e+03, threshold=8.680e+02, percent-clipped=5.0 2023-03-07 17:08:36,542 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8920, 3.7826, 3.9640, 2.7531, 2.3782, 2.6060, 3.3779, 3.4247], device='cuda:3'), covar=tensor([0.0450, 0.0358, 0.0273, 0.1417, 0.5447, 0.2425, 0.0366, 0.1697], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0162, 0.0180, 0.0174, 0.0372, 0.0312, 0.0162, 0.0291], device='cuda:3'), out_proj_covar=tensor([1.4041e-04, 7.8087e-05, 8.7560e-05, 8.0432e-05, 1.8798e-04, 1.5084e-04, 7.6440e-05, 1.5241e-04], device='cuda:3') 2023-03-07 17:09:25,257 INFO [train2.py:809] (3/4) Epoch 4, batch 3800, loss[ctc_loss=0.1559, att_loss=0.2942, loss=0.2665, over 17287.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01199, over 55.00 utterances.], tot_loss[ctc_loss=0.1742, att_loss=0.2877, loss=0.265, over 3250834.81 frames. utt_duration=1238 frames, utt_pad_proportion=0.0629, over 10515.34 utterances.], batch size: 55, lr: 2.40e-02, grad_scale: 8.0 2023-03-07 17:09:41,118 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7163, 5.1852, 5.0457, 5.1202, 5.2336, 5.1132, 4.9340, 4.6932], device='cuda:3'), covar=tensor([0.0979, 0.0422, 0.0201, 0.0348, 0.0270, 0.0256, 0.0203, 0.0293], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0211, 0.0149, 0.0190, 0.0235, 0.0258, 0.0197, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:10:43,433 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3594, 5.1644, 4.9234, 3.3385, 1.9008, 2.6051, 5.0910, 3.7560], device='cuda:3'), covar=tensor([0.0468, 0.0161, 0.0237, 0.1697, 0.7035, 0.2593, 0.0189, 0.1940], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0159, 0.0183, 0.0176, 0.0369, 0.0313, 0.0161, 0.0294], device='cuda:3'), out_proj_covar=tensor([1.3913e-04, 7.6234e-05, 8.9302e-05, 8.1642e-05, 1.8645e-04, 1.5101e-04, 7.5697e-05, 1.5333e-04], device='cuda:3') 2023-03-07 17:10:46,094 INFO [train2.py:809] (3/4) Epoch 4, batch 3850, loss[ctc_loss=0.1734, att_loss=0.2697, loss=0.2505, over 15878.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.00929, over 39.00 utterances.], tot_loss[ctc_loss=0.1726, att_loss=0.2865, loss=0.2637, over 3251597.71 frames. utt_duration=1240 frames, utt_pad_proportion=0.0621, over 10500.32 utterances.], batch size: 39, lr: 2.39e-02, grad_scale: 8.0 2023-03-07 17:10:53,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 3.550e+02 4.549e+02 5.409e+02 1.209e+03, threshold=9.099e+02, percent-clipped=5.0 2023-03-07 17:12:00,186 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:12:02,975 INFO [train2.py:809] (3/4) Epoch 4, batch 3900, loss[ctc_loss=0.1563, att_loss=0.2891, loss=0.2625, over 17320.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01087, over 55.00 utterances.], tot_loss[ctc_loss=0.1728, att_loss=0.2866, loss=0.2639, over 3244455.79 frames. utt_duration=1199 frames, utt_pad_proportion=0.07206, over 10840.30 utterances.], batch size: 55, lr: 2.39e-02, grad_scale: 8.0 2023-03-07 17:12:21,808 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2691, 4.4875, 4.1961, 4.5861, 2.1333, 4.3467, 2.4568, 2.0719], device='cuda:3'), covar=tensor([0.0203, 0.0188, 0.1001, 0.0250, 0.3274, 0.0218, 0.2065, 0.1986], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0093, 0.0228, 0.0110, 0.0221, 0.0094, 0.0211, 0.0186], device='cuda:3'), out_proj_covar=tensor([9.0818e-05, 8.9488e-05, 1.9814e-04, 9.7806e-05, 1.9015e-04, 8.8334e-05, 1.7951e-04, 1.6054e-04], device='cuda:3') 2023-03-07 17:12:49,858 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:13:13,703 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:13:19,704 INFO [train2.py:809] (3/4) Epoch 4, batch 3950, loss[ctc_loss=0.1939, att_loss=0.3011, loss=0.2796, over 16969.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.0071, over 50.00 utterances.], tot_loss[ctc_loss=0.1723, att_loss=0.2855, loss=0.2629, over 3237193.11 frames. utt_duration=1229 frames, utt_pad_proportion=0.06784, over 10551.70 utterances.], batch size: 50, lr: 2.39e-02, grad_scale: 8.0 2023-03-07 17:13:27,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.270e+02 3.629e+02 4.356e+02 5.723e+02 1.612e+03, threshold=8.711e+02, percent-clipped=3.0 2023-03-07 17:14:39,237 INFO [train2.py:809] (3/4) Epoch 5, batch 0, loss[ctc_loss=0.1231, att_loss=0.2585, loss=0.2314, over 16007.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007265, over 40.00 utterances.], tot_loss[ctc_loss=0.1231, att_loss=0.2585, loss=0.2314, over 16007.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007265, over 40.00 utterances.], batch size: 40, lr: 2.22e-02, grad_scale: 8.0 2023-03-07 17:14:39,238 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 17:14:51,959 INFO [train2.py:843] (3/4) Epoch 5, validation: ctc_loss=0.08303, att_loss=0.2543, loss=0.22, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 17:14:51,960 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 17:15:06,130 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:15:28,157 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1660, 4.8953, 5.0287, 3.0423, 4.6854, 4.5442, 4.1501, 2.7020], device='cuda:3'), covar=tensor([0.0108, 0.0090, 0.0154, 0.0893, 0.0090, 0.0140, 0.0277, 0.1348], device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0056, 0.0045, 0.0090, 0.0051, 0.0064, 0.0073, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:16:13,042 INFO [train2.py:809] (3/4) Epoch 5, batch 50, loss[ctc_loss=0.159, att_loss=0.2907, loss=0.2643, over 16327.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006399, over 45.00 utterances.], tot_loss[ctc_loss=0.1675, att_loss=0.2848, loss=0.2614, over 743520.67 frames. utt_duration=1245 frames, utt_pad_proportion=0.04705, over 2391.01 utterances.], batch size: 45, lr: 2.22e-02, grad_scale: 8.0 2023-03-07 17:16:41,396 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:16:51,838 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 3.417e+02 4.421e+02 5.634e+02 1.181e+03, threshold=8.842e+02, percent-clipped=4.0 2023-03-07 17:17:12,875 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-03-07 17:17:23,381 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5043, 5.7549, 5.0605, 5.6621, 5.3212, 5.2485, 5.1134, 5.1087], device='cuda:3'), covar=tensor([0.1149, 0.0847, 0.0784, 0.0648, 0.0643, 0.1112, 0.2199, 0.2280], device='cuda:3'), in_proj_covar=tensor([0.0327, 0.0369, 0.0287, 0.0285, 0.0263, 0.0351, 0.0394, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:17:36,502 INFO [train2.py:809] (3/4) Epoch 5, batch 100, loss[ctc_loss=0.1839, att_loss=0.2994, loss=0.2763, over 17279.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02484, over 59.00 utterances.], tot_loss[ctc_loss=0.1662, att_loss=0.2842, loss=0.2606, over 1307941.72 frames. utt_duration=1268 frames, utt_pad_proportion=0.04342, over 4132.30 utterances.], batch size: 59, lr: 2.21e-02, grad_scale: 8.0 2023-03-07 17:18:18,156 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:18:19,582 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6343, 2.8545, 3.8308, 2.8771, 3.6149, 4.7083, 4.5461, 3.5197], device='cuda:3'), covar=tensor([0.0309, 0.1520, 0.0658, 0.1318, 0.0701, 0.0340, 0.0388, 0.1149], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0196, 0.0183, 0.0180, 0.0202, 0.0174, 0.0147, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:18:55,040 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3338, 5.0318, 5.2150, 3.1856, 5.0446, 4.4259, 4.3421, 2.6962], device='cuda:3'), covar=tensor([0.0062, 0.0069, 0.0099, 0.0826, 0.0052, 0.0142, 0.0234, 0.1235], device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0057, 0.0045, 0.0091, 0.0052, 0.0064, 0.0074, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:18:55,113 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:18:56,276 INFO [train2.py:809] (3/4) Epoch 5, batch 150, loss[ctc_loss=0.1283, att_loss=0.2594, loss=0.2331, over 15999.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007605, over 40.00 utterances.], tot_loss[ctc_loss=0.1674, att_loss=0.2856, loss=0.262, over 1749257.83 frames. utt_duration=1269 frames, utt_pad_proportion=0.04202, over 5521.02 utterances.], batch size: 40, lr: 2.21e-02, grad_scale: 8.0 2023-03-07 17:19:31,234 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.158e+02 4.095e+02 5.546e+02 1.058e+03, threshold=8.189e+02, percent-clipped=5.0 2023-03-07 17:19:45,569 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8208, 5.2359, 4.7377, 5.2300, 4.7607, 5.0151, 5.4912, 5.2165], device='cuda:3'), covar=tensor([0.0317, 0.0193, 0.0612, 0.0156, 0.0373, 0.0132, 0.0148, 0.0125], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0160, 0.0211, 0.0135, 0.0177, 0.0129, 0.0151, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-07 17:20:01,593 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 17:20:16,135 INFO [train2.py:809] (3/4) Epoch 5, batch 200, loss[ctc_loss=0.1642, att_loss=0.2819, loss=0.2584, over 16536.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006557, over 45.00 utterances.], tot_loss[ctc_loss=0.1697, att_loss=0.2864, loss=0.263, over 2080975.79 frames. utt_duration=1205 frames, utt_pad_proportion=0.06101, over 6914.05 utterances.], batch size: 45, lr: 2.21e-02, grad_scale: 8.0 2023-03-07 17:20:32,675 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:20:37,383 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:21:07,394 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-07 17:21:35,465 INFO [train2.py:809] (3/4) Epoch 5, batch 250, loss[ctc_loss=0.1771, att_loss=0.301, loss=0.2762, over 17408.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04642, over 69.00 utterances.], tot_loss[ctc_loss=0.1697, att_loss=0.2859, loss=0.2627, over 2344912.83 frames. utt_duration=1210 frames, utt_pad_proportion=0.0629, over 7758.74 utterances.], batch size: 69, lr: 2.20e-02, grad_scale: 8.0 2023-03-07 17:22:11,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.315e+02 3.409e+02 3.964e+02 4.827e+02 7.562e+02, threshold=7.928e+02, percent-clipped=0.0 2023-03-07 17:22:15,451 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:22:57,400 INFO [train2.py:809] (3/4) Epoch 5, batch 300, loss[ctc_loss=0.1641, att_loss=0.2785, loss=0.2556, over 16396.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007213, over 44.00 utterances.], tot_loss[ctc_loss=0.1674, att_loss=0.2842, loss=0.2608, over 2557448.78 frames. utt_duration=1229 frames, utt_pad_proportion=0.05651, over 8335.15 utterances.], batch size: 44, lr: 2.20e-02, grad_scale: 8.0 2023-03-07 17:23:02,216 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:24:18,575 INFO [train2.py:809] (3/4) Epoch 5, batch 350, loss[ctc_loss=0.1538, att_loss=0.2661, loss=0.2437, over 15883.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008095, over 39.00 utterances.], tot_loss[ctc_loss=0.1656, att_loss=0.2835, loss=0.2599, over 2722630.03 frames. utt_duration=1263 frames, utt_pad_proportion=0.04727, over 8636.30 utterances.], batch size: 39, lr: 2.20e-02, grad_scale: 8.0 2023-03-07 17:24:56,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.423e+02 4.365e+02 5.343e+02 1.409e+03, threshold=8.730e+02, percent-clipped=9.0 2023-03-07 17:25:22,781 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2473, 3.7815, 2.9549, 3.4243, 3.7612, 3.4726, 2.6377, 4.4002], device='cuda:3'), covar=tensor([0.1386, 0.0452, 0.1376, 0.0641, 0.0595, 0.0753, 0.1113, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0123, 0.0178, 0.0143, 0.0148, 0.0165, 0.0149, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:25:34,758 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-07 17:25:41,307 INFO [train2.py:809] (3/4) Epoch 5, batch 400, loss[ctc_loss=0.1529, att_loss=0.2903, loss=0.2628, over 17061.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.0078, over 52.00 utterances.], tot_loss[ctc_loss=0.166, att_loss=0.2831, loss=0.2597, over 2841611.16 frames. utt_duration=1256 frames, utt_pad_proportion=0.04989, over 9058.33 utterances.], batch size: 52, lr: 2.20e-02, grad_scale: 8.0 2023-03-07 17:26:07,466 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6288, 4.7939, 5.1261, 5.2378, 4.8907, 5.4379, 4.9599, 5.6062], device='cuda:3'), covar=tensor([0.0552, 0.0663, 0.0495, 0.0694, 0.1809, 0.1032, 0.0597, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0312, 0.0309, 0.0373, 0.0536, 0.0318, 0.0258, 0.0332], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 17:26:15,243 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:26:26,691 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4626, 2.2075, 4.9315, 3.7995, 2.9707, 4.5326, 4.6788, 4.6241], device='cuda:3'), covar=tensor([0.0182, 0.1782, 0.0201, 0.1225, 0.2240, 0.0238, 0.0141, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0233, 0.0118, 0.0293, 0.0313, 0.0174, 0.0107, 0.0128], device='cuda:3'), out_proj_covar=tensor([1.0838e-04, 1.8222e-04, 1.0146e-04, 2.3275e-04, 2.4761e-04, 1.4683e-04, 9.2096e-05, 1.1050e-04], device='cuda:3') 2023-03-07 17:26:37,592 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:27:02,335 INFO [train2.py:809] (3/4) Epoch 5, batch 450, loss[ctc_loss=0.1391, att_loss=0.27, loss=0.2438, over 16558.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005087, over 45.00 utterances.], tot_loss[ctc_loss=0.1663, att_loss=0.2838, loss=0.2603, over 2934792.33 frames. utt_duration=1260 frames, utt_pad_proportion=0.04937, over 9326.19 utterances.], batch size: 45, lr: 2.19e-02, grad_scale: 8.0 2023-03-07 17:27:08,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.25 vs. limit=2.0 2023-03-07 17:27:40,288 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.521e+02 3.844e+02 4.538e+02 5.770e+02 9.247e+02, threshold=9.077e+02, percent-clipped=1.0 2023-03-07 17:27:48,482 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6968, 3.7534, 3.1032, 3.4189, 3.8874, 3.4928, 2.8960, 4.3691], device='cuda:3'), covar=tensor([0.1064, 0.0425, 0.1214, 0.0573, 0.0493, 0.0648, 0.0843, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0120, 0.0170, 0.0139, 0.0144, 0.0161, 0.0145, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:28:17,583 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:28:24,060 INFO [train2.py:809] (3/4) Epoch 5, batch 500, loss[ctc_loss=0.1682, att_loss=0.2812, loss=0.2586, over 16404.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007314, over 44.00 utterances.], tot_loss[ctc_loss=0.1656, att_loss=0.2826, loss=0.2592, over 3010697.58 frames. utt_duration=1280 frames, utt_pad_proportion=0.04554, over 9418.00 utterances.], batch size: 44, lr: 2.19e-02, grad_scale: 8.0 2023-03-07 17:28:32,541 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:29:00,020 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4952, 1.2800, 1.7811, 2.7197, 2.1776, 1.4889, 1.6237, 1.9406], device='cuda:3'), covar=tensor([0.1951, 0.2688, 0.2103, 0.0889, 0.0764, 0.2571, 0.1849, 0.1897], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0095, 0.0098, 0.0079, 0.0081, 0.0086, 0.0099, 0.0088], device='cuda:3'), out_proj_covar=tensor([4.1280e-05, 5.0791e-05, 5.0998e-05, 4.3218e-05, 3.8379e-05, 4.5665e-05, 4.8806e-05, 4.4870e-05], device='cuda:3') 2023-03-07 17:29:45,249 INFO [train2.py:809] (3/4) Epoch 5, batch 550, loss[ctc_loss=0.1632, att_loss=0.302, loss=0.2742, over 17290.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01259, over 55.00 utterances.], tot_loss[ctc_loss=0.1658, att_loss=0.2829, loss=0.2595, over 3076989.89 frames. utt_duration=1295 frames, utt_pad_proportion=0.03986, over 9513.11 utterances.], batch size: 55, lr: 2.19e-02, grad_scale: 8.0 2023-03-07 17:30:16,800 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:30:22,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 3.300e+02 4.154e+02 5.468e+02 1.131e+03, threshold=8.309e+02, percent-clipped=7.0 2023-03-07 17:31:06,243 INFO [train2.py:809] (3/4) Epoch 5, batch 600, loss[ctc_loss=0.1503, att_loss=0.2558, loss=0.2347, over 15637.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009178, over 37.00 utterances.], tot_loss[ctc_loss=0.1661, att_loss=0.2831, loss=0.2597, over 3119954.86 frames. utt_duration=1307 frames, utt_pad_proportion=0.0383, over 9556.47 utterances.], batch size: 37, lr: 2.18e-02, grad_scale: 8.0 2023-03-07 17:31:06,629 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.6612, 2.9761, 3.0859, 2.5521, 2.7956, 2.7861, 2.2076, 1.4871], device='cuda:3'), covar=tensor([0.1607, 0.0982, 0.0855, 0.2762, 0.3451, 0.1677, 0.1288, 0.7729], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0064, 0.0068, 0.0086, 0.0065, 0.0079, 0.0068, 0.0102], device='cuda:3'), out_proj_covar=tensor([4.7186e-05, 4.1191e-05, 4.3946e-05, 5.9618e-05, 4.4333e-05, 5.8055e-05, 4.6907e-05, 7.5848e-05], device='cuda:3') 2023-03-07 17:31:12,936 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:32:27,517 INFO [train2.py:809] (3/4) Epoch 5, batch 650, loss[ctc_loss=0.1724, att_loss=0.3028, loss=0.2767, over 16470.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006312, over 46.00 utterances.], tot_loss[ctc_loss=0.165, att_loss=0.2822, loss=0.2587, over 3148305.52 frames. utt_duration=1293 frames, utt_pad_proportion=0.04355, over 9748.79 utterances.], batch size: 46, lr: 2.18e-02, grad_scale: 8.0 2023-03-07 17:32:29,902 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:32:36,236 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.88 vs. limit=2.0 2023-03-07 17:32:46,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 17:33:07,236 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 3.372e+02 4.098e+02 5.133e+02 8.733e+02, threshold=8.196e+02, percent-clipped=2.0 2023-03-07 17:33:51,402 INFO [train2.py:809] (3/4) Epoch 5, batch 700, loss[ctc_loss=0.1872, att_loss=0.3094, loss=0.2849, over 17330.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03687, over 63.00 utterances.], tot_loss[ctc_loss=0.1651, att_loss=0.2824, loss=0.259, over 3182591.33 frames. utt_duration=1287 frames, utt_pad_proportion=0.04324, over 9900.08 utterances.], batch size: 63, lr: 2.18e-02, grad_scale: 8.0 2023-03-07 17:34:26,843 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:34:36,398 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9610, 5.3113, 5.1748, 4.9683, 5.3918, 5.2874, 5.0053, 4.9006], device='cuda:3'), covar=tensor([0.1043, 0.0371, 0.0185, 0.0610, 0.0265, 0.0228, 0.0246, 0.0245], device='cuda:3'), in_proj_covar=tensor([0.0368, 0.0215, 0.0153, 0.0192, 0.0236, 0.0257, 0.0201, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:35:14,451 INFO [train2.py:809] (3/4) Epoch 5, batch 750, loss[ctc_loss=0.139, att_loss=0.2595, loss=0.2354, over 15958.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.006051, over 41.00 utterances.], tot_loss[ctc_loss=0.1667, att_loss=0.2834, loss=0.26, over 3184797.92 frames. utt_duration=1229 frames, utt_pad_proportion=0.06083, over 10378.31 utterances.], batch size: 41, lr: 2.17e-02, grad_scale: 8.0 2023-03-07 17:35:43,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.30 vs. limit=5.0 2023-03-07 17:35:45,802 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:35:50,584 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6811, 5.1147, 4.5803, 5.1715, 4.4399, 4.7630, 5.2862, 5.0558], device='cuda:3'), covar=tensor([0.0370, 0.0272, 0.0623, 0.0161, 0.0481, 0.0222, 0.0182, 0.0150], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0163, 0.0214, 0.0136, 0.0184, 0.0133, 0.0156, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-07 17:35:51,950 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.143e+02 3.355e+02 3.924e+02 5.306e+02 2.112e+03, threshold=7.847e+02, percent-clipped=6.0 2023-03-07 17:36:06,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-07 17:36:21,560 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:36:28,612 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 17:36:36,909 INFO [train2.py:809] (3/4) Epoch 5, batch 800, loss[ctc_loss=0.1416, att_loss=0.2708, loss=0.245, over 16188.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.005219, over 41.00 utterances.], tot_loss[ctc_loss=0.1659, att_loss=0.2823, loss=0.259, over 3202270.04 frames. utt_duration=1260 frames, utt_pad_proportion=0.05348, over 10179.45 utterances.], batch size: 41, lr: 2.17e-02, grad_scale: 8.0 2023-03-07 17:36:45,935 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:37:12,433 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-07 17:37:59,756 INFO [train2.py:809] (3/4) Epoch 5, batch 850, loss[ctc_loss=0.1572, att_loss=0.2919, loss=0.265, over 17047.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009963, over 53.00 utterances.], tot_loss[ctc_loss=0.1645, att_loss=0.2817, loss=0.2583, over 3217104.96 frames. utt_duration=1250 frames, utt_pad_proportion=0.056, over 10304.55 utterances.], batch size: 53, lr: 2.17e-02, grad_scale: 8.0 2023-03-07 17:38:04,517 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:38:30,526 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:38:36,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.471e+02 4.444e+02 5.932e+02 1.311e+03, threshold=8.889e+02, percent-clipped=13.0 2023-03-07 17:39:22,340 INFO [train2.py:809] (3/4) Epoch 5, batch 900, loss[ctc_loss=0.2096, att_loss=0.3158, loss=0.2945, over 17453.00 frames. utt_duration=885.2 frames, utt_pad_proportion=0.06817, over 79.00 utterances.], tot_loss[ctc_loss=0.165, att_loss=0.2824, loss=0.2589, over 3227247.79 frames. utt_duration=1230 frames, utt_pad_proportion=0.06059, over 10505.47 utterances.], batch size: 79, lr: 2.16e-02, grad_scale: 8.0 2023-03-07 17:39:34,956 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2309, 2.8872, 3.5458, 2.5621, 3.2126, 4.4139, 4.0933, 3.2015], device='cuda:3'), covar=tensor([0.0366, 0.1481, 0.1009, 0.1360, 0.1069, 0.0442, 0.0503, 0.1220], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0195, 0.0192, 0.0187, 0.0200, 0.0184, 0.0155, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:39:48,822 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:40:09,448 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2978, 3.6942, 2.9432, 3.3050, 3.7018, 3.3715, 2.5452, 4.1864], device='cuda:3'), covar=tensor([0.1244, 0.0351, 0.1236, 0.0665, 0.0535, 0.0702, 0.1048, 0.0338], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0123, 0.0174, 0.0144, 0.0150, 0.0166, 0.0146, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:40:13,279 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:40:42,553 INFO [train2.py:809] (3/4) Epoch 5, batch 950, loss[ctc_loss=0.1741, att_loss=0.2776, loss=0.2569, over 16276.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007326, over 43.00 utterances.], tot_loss[ctc_loss=0.1657, att_loss=0.2828, loss=0.2594, over 3240122.20 frames. utt_duration=1247 frames, utt_pad_proportion=0.05538, over 10408.83 utterances.], batch size: 43, lr: 2.16e-02, grad_scale: 8.0 2023-03-07 17:41:01,523 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9483, 6.0464, 5.3812, 5.9606, 5.6966, 5.4223, 5.5857, 5.3797], device='cuda:3'), covar=tensor([0.0930, 0.0739, 0.0775, 0.0614, 0.0553, 0.1089, 0.2038, 0.2186], device='cuda:3'), in_proj_covar=tensor([0.0323, 0.0361, 0.0294, 0.0281, 0.0265, 0.0346, 0.0389, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 17:41:03,296 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2134, 5.5561, 5.1239, 5.5671, 5.0962, 5.2085, 5.6973, 5.4419], device='cuda:3'), covar=tensor([0.0243, 0.0181, 0.0504, 0.0114, 0.0331, 0.0123, 0.0149, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0163, 0.0212, 0.0134, 0.0184, 0.0134, 0.0153, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-07 17:41:14,650 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:41:18,945 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 3.605e+02 4.438e+02 5.826e+02 1.823e+03, threshold=8.877e+02, percent-clipped=6.0 2023-03-07 17:41:52,640 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:42:04,883 INFO [train2.py:809] (3/4) Epoch 5, batch 1000, loss[ctc_loss=0.1466, att_loss=0.2714, loss=0.2464, over 16280.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006751, over 43.00 utterances.], tot_loss[ctc_loss=0.1656, att_loss=0.2828, loss=0.2593, over 3232970.32 frames. utt_duration=1244 frames, utt_pad_proportion=0.05911, over 10411.28 utterances.], batch size: 43, lr: 2.16e-02, grad_scale: 8.0 2023-03-07 17:42:54,668 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:43:22,944 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6572, 2.2902, 3.4248, 4.4533, 4.1505, 4.3459, 2.7894, 1.9035], device='cuda:3'), covar=tensor([0.0437, 0.2497, 0.1022, 0.0432, 0.0482, 0.0174, 0.1597, 0.2605], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0192, 0.0187, 0.0143, 0.0131, 0.0114, 0.0181, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:43:25,863 INFO [train2.py:809] (3/4) Epoch 5, batch 1050, loss[ctc_loss=0.1733, att_loss=0.2914, loss=0.2678, over 16911.00 frames. utt_duration=685.1 frames, utt_pad_proportion=0.1383, over 99.00 utterances.], tot_loss[ctc_loss=0.1654, att_loss=0.2822, loss=0.2588, over 3235098.73 frames. utt_duration=1250 frames, utt_pad_proportion=0.05725, over 10367.99 utterances.], batch size: 99, lr: 2.16e-02, grad_scale: 8.0 2023-03-07 17:43:39,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2023-03-07 17:43:59,537 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.3916, 2.3184, 2.7717, 2.3156, 2.8509, 2.5992, 2.2513, 1.3233], device='cuda:3'), covar=tensor([0.1555, 0.1488, 0.1228, 0.2543, 0.1128, 0.3422, 0.0906, 0.7693], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0067, 0.0071, 0.0086, 0.0067, 0.0085, 0.0066, 0.0109], device='cuda:3'), out_proj_covar=tensor([4.9676e-05, 4.4020e-05, 4.6502e-05, 6.0598e-05, 4.6782e-05, 6.2671e-05, 4.6797e-05, 7.9783e-05], device='cuda:3') 2023-03-07 17:44:02,518 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 3.274e+02 4.192e+02 5.406e+02 1.067e+03, threshold=8.384e+02, percent-clipped=5.0 2023-03-07 17:44:09,852 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-07 17:44:13,099 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2656, 2.7452, 3.5003, 2.7252, 3.3490, 4.4607, 4.3068, 3.1108], device='cuda:3'), covar=tensor([0.0387, 0.1462, 0.0813, 0.1288, 0.0891, 0.0393, 0.0432, 0.1293], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0193, 0.0190, 0.0184, 0.0197, 0.0187, 0.0157, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:44:32,635 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:44:47,450 INFO [train2.py:809] (3/4) Epoch 5, batch 1100, loss[ctc_loss=0.1534, att_loss=0.2891, loss=0.262, over 17036.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007016, over 51.00 utterances.], tot_loss[ctc_loss=0.1647, att_loss=0.2821, loss=0.2586, over 3240663.78 frames. utt_duration=1247 frames, utt_pad_proportion=0.05626, over 10405.77 utterances.], batch size: 51, lr: 2.15e-02, grad_scale: 8.0 2023-03-07 17:44:47,799 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6882, 2.4051, 3.5623, 4.3320, 4.1156, 4.2721, 2.5879, 1.7604], device='cuda:3'), covar=tensor([0.0461, 0.2584, 0.0956, 0.0536, 0.0567, 0.0267, 0.2024, 0.2741], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0189, 0.0185, 0.0142, 0.0132, 0.0115, 0.0182, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:44:52,495 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3545, 4.8249, 5.2533, 4.8688, 4.2238, 5.1585, 4.7334, 5.2032], device='cuda:3'), covar=tensor([0.1062, 0.0943, 0.0591, 0.1277, 0.3615, 0.1301, 0.0899, 0.1084], device='cuda:3'), in_proj_covar=tensor([0.0473, 0.0311, 0.0319, 0.0373, 0.0534, 0.0318, 0.0272, 0.0338], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 17:45:50,931 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:46:09,757 INFO [train2.py:809] (3/4) Epoch 5, batch 1150, loss[ctc_loss=0.1805, att_loss=0.3092, loss=0.2835, over 17067.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008215, over 52.00 utterances.], tot_loss[ctc_loss=0.1651, att_loss=0.2821, loss=0.2587, over 3248431.76 frames. utt_duration=1236 frames, utt_pad_proportion=0.05804, over 10525.41 utterances.], batch size: 52, lr: 2.15e-02, grad_scale: 8.0 2023-03-07 17:46:45,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 3.410e+02 4.345e+02 5.323e+02 1.056e+03, threshold=8.690e+02, percent-clipped=4.0 2023-03-07 17:46:53,157 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4051, 3.3589, 2.7825, 3.1251, 3.5208, 3.3052, 2.4596, 3.8107], device='cuda:3'), covar=tensor([0.1095, 0.0349, 0.1136, 0.0586, 0.0436, 0.0674, 0.0970, 0.0314], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0126, 0.0176, 0.0144, 0.0152, 0.0165, 0.0148, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:47:05,890 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3590, 2.5830, 3.6453, 2.8390, 3.3672, 4.4717, 4.2487, 3.2315], device='cuda:3'), covar=tensor([0.0369, 0.1717, 0.0719, 0.1404, 0.0959, 0.0526, 0.0371, 0.1313], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0193, 0.0189, 0.0183, 0.0197, 0.0186, 0.0152, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:47:27,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-03-07 17:47:31,225 INFO [train2.py:809] (3/4) Epoch 5, batch 1200, loss[ctc_loss=0.1913, att_loss=0.316, loss=0.2911, over 17293.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01077, over 55.00 utterances.], tot_loss[ctc_loss=0.1658, att_loss=0.2821, loss=0.2588, over 3252638.71 frames. utt_duration=1232 frames, utt_pad_proportion=0.0586, over 10569.13 utterances.], batch size: 55, lr: 2.15e-02, grad_scale: 8.0 2023-03-07 17:48:19,610 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0329, 4.6630, 4.6030, 4.7346, 5.3125, 5.1275, 4.8424, 2.3898], device='cuda:3'), covar=tensor([0.0195, 0.0445, 0.0308, 0.0294, 0.0898, 0.0172, 0.0252, 0.2741], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0122, 0.0116, 0.0124, 0.0276, 0.0132, 0.0112, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 17:48:52,438 INFO [train2.py:809] (3/4) Epoch 5, batch 1250, loss[ctc_loss=0.1939, att_loss=0.2847, loss=0.2665, over 16538.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005706, over 45.00 utterances.], tot_loss[ctc_loss=0.1657, att_loss=0.2822, loss=0.2589, over 3261571.77 frames. utt_duration=1235 frames, utt_pad_proportion=0.05743, over 10574.42 utterances.], batch size: 45, lr: 2.14e-02, grad_scale: 8.0 2023-03-07 17:48:57,273 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:49:19,604 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.5636, 2.6177, 3.0124, 2.8666, 3.0272, 2.9394, 2.3721, 1.5956], device='cuda:3'), covar=tensor([0.1524, 0.1748, 0.1635, 0.2170, 0.1287, 0.2588, 0.1380, 0.8245], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0066, 0.0071, 0.0085, 0.0066, 0.0082, 0.0065, 0.0110], device='cuda:3'), out_proj_covar=tensor([4.9005e-05, 4.3560e-05, 4.6598e-05, 5.9703e-05, 4.6372e-05, 6.0906e-05, 4.6733e-05, 7.9842e-05], device='cuda:3') 2023-03-07 17:49:29,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.117e+02 3.900e+02 4.861e+02 6.296e+02 1.076e+03, threshold=9.721e+02, percent-clipped=3.0 2023-03-07 17:49:55,876 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 17:50:02,394 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7413, 5.2038, 4.6734, 5.2433, 4.6204, 4.8932, 5.4307, 5.1364], device='cuda:3'), covar=tensor([0.0352, 0.0225, 0.0681, 0.0139, 0.0412, 0.0154, 0.0177, 0.0139], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0166, 0.0220, 0.0139, 0.0187, 0.0135, 0.0161, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-07 17:50:14,610 INFO [train2.py:809] (3/4) Epoch 5, batch 1300, loss[ctc_loss=0.1646, att_loss=0.2922, loss=0.2667, over 16953.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008254, over 50.00 utterances.], tot_loss[ctc_loss=0.1653, att_loss=0.2816, loss=0.2583, over 3257621.56 frames. utt_duration=1233 frames, utt_pad_proportion=0.06042, over 10576.70 utterances.], batch size: 50, lr: 2.14e-02, grad_scale: 8.0 2023-03-07 17:50:36,869 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:50:56,151 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:51:35,792 INFO [train2.py:809] (3/4) Epoch 5, batch 1350, loss[ctc_loss=0.1867, att_loss=0.3075, loss=0.2834, over 16871.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00739, over 49.00 utterances.], tot_loss[ctc_loss=0.1656, att_loss=0.2821, loss=0.2588, over 3257756.97 frames. utt_duration=1227 frames, utt_pad_proportion=0.06249, over 10637.46 utterances.], batch size: 49, lr: 2.14e-02, grad_scale: 8.0 2023-03-07 17:52:12,514 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 3.130e+02 4.012e+02 5.410e+02 1.134e+03, threshold=8.024e+02, percent-clipped=3.0 2023-03-07 17:52:24,637 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0336, 4.6532, 4.5976, 4.8139, 1.9834, 4.7179, 2.5404, 2.1990], device='cuda:3'), covar=tensor([0.0280, 0.0146, 0.0700, 0.0248, 0.3003, 0.0170, 0.1866, 0.1877], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0096, 0.0232, 0.0112, 0.0228, 0.0097, 0.0215, 0.0200], device='cuda:3'), out_proj_covar=tensor([9.8418e-05, 9.4302e-05, 2.0298e-04, 9.9117e-05, 1.9810e-04, 9.1113e-05, 1.8429e-04, 1.7250e-04], device='cuda:3') 2023-03-07 17:52:38,871 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.2469, 2.2091, 2.8224, 2.4413, 2.5556, 2.8594, 2.4218, 1.1959], device='cuda:3'), covar=tensor([0.1790, 0.2133, 0.1501, 0.4498, 0.1844, 0.7647, 0.1492, 1.2289], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0067, 0.0069, 0.0087, 0.0065, 0.0085, 0.0063, 0.0108], device='cuda:3'), out_proj_covar=tensor([4.8423e-05, 4.4479e-05, 4.5577e-05, 6.1134e-05, 4.5608e-05, 6.2661e-05, 4.5978e-05, 7.9737e-05], device='cuda:3') 2023-03-07 17:52:57,031 INFO [train2.py:809] (3/4) Epoch 5, batch 1400, loss[ctc_loss=0.1462, att_loss=0.2653, loss=0.2415, over 16174.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006596, over 41.00 utterances.], tot_loss[ctc_loss=0.1645, att_loss=0.2817, loss=0.2583, over 3270170.39 frames. utt_duration=1246 frames, utt_pad_proportion=0.05467, over 10512.19 utterances.], batch size: 41, lr: 2.14e-02, grad_scale: 8.0 2023-03-07 17:53:38,645 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2576, 4.8474, 4.6187, 4.5935, 4.9061, 4.5846, 3.9273, 4.6549], device='cuda:3'), covar=tensor([0.0145, 0.0171, 0.0115, 0.0139, 0.0089, 0.0098, 0.0415, 0.0350], device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0052, 0.0055, 0.0039, 0.0039, 0.0048, 0.0070, 0.0067], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 17:53:38,740 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:54:13,842 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 17:54:18,024 INFO [train2.py:809] (3/4) Epoch 5, batch 1450, loss[ctc_loss=0.1645, att_loss=0.2729, loss=0.2513, over 15628.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009326, over 37.00 utterances.], tot_loss[ctc_loss=0.1653, att_loss=0.2828, loss=0.2593, over 3270633.28 frames. utt_duration=1236 frames, utt_pad_proportion=0.05613, over 10593.43 utterances.], batch size: 37, lr: 2.13e-02, grad_scale: 8.0 2023-03-07 17:54:55,335 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 3.556e+02 4.184e+02 5.156e+02 1.348e+03, threshold=8.369e+02, percent-clipped=7.0 2023-03-07 17:55:19,616 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:55:28,921 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3363, 3.7541, 4.1527, 3.7332, 3.5261, 4.2801, 4.2383, 4.2612], device='cuda:3'), covar=tensor([0.1440, 0.1746, 0.1168, 0.1919, 0.3408, 0.1348, 0.1673, 0.1305], device='cuda:3'), in_proj_covar=tensor([0.0495, 0.0323, 0.0333, 0.0394, 0.0550, 0.0330, 0.0278, 0.0346], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 17:55:39,526 INFO [train2.py:809] (3/4) Epoch 5, batch 1500, loss[ctc_loss=0.178, att_loss=0.2908, loss=0.2683, over 17054.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008766, over 52.00 utterances.], tot_loss[ctc_loss=0.1652, att_loss=0.2829, loss=0.2593, over 3267843.47 frames. utt_duration=1214 frames, utt_pad_proportion=0.0635, over 10779.31 utterances.], batch size: 52, lr: 2.13e-02, grad_scale: 8.0 2023-03-07 17:55:52,476 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:56:59,736 INFO [train2.py:809] (3/4) Epoch 5, batch 1550, loss[ctc_loss=0.2042, att_loss=0.3126, loss=0.2909, over 16951.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.008503, over 50.00 utterances.], tot_loss[ctc_loss=0.1654, att_loss=0.2825, loss=0.2591, over 3269739.93 frames. utt_duration=1231 frames, utt_pad_proportion=0.05949, over 10638.26 utterances.], batch size: 50, lr: 2.13e-02, grad_scale: 8.0 2023-03-07 17:57:36,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.215e+02 3.523e+02 4.297e+02 5.219e+02 1.408e+03, threshold=8.593e+02, percent-clipped=2.0 2023-03-07 17:58:01,486 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 17:58:20,675 INFO [train2.py:809] (3/4) Epoch 5, batch 1600, loss[ctc_loss=0.1406, att_loss=0.259, loss=0.2353, over 15962.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.006576, over 41.00 utterances.], tot_loss[ctc_loss=0.1648, att_loss=0.2817, loss=0.2584, over 3268991.31 frames. utt_duration=1240 frames, utt_pad_proportion=0.05717, over 10561.30 utterances.], batch size: 41, lr: 2.12e-02, grad_scale: 8.0 2023-03-07 17:58:34,919 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 17:59:03,264 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:59:20,708 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 17:59:41,752 INFO [train2.py:809] (3/4) Epoch 5, batch 1650, loss[ctc_loss=0.1677, att_loss=0.2782, loss=0.2561, over 16272.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006873, over 43.00 utterances.], tot_loss[ctc_loss=0.1639, att_loss=0.2819, loss=0.2583, over 3273194.08 frames. utt_duration=1250 frames, utt_pad_proportion=0.05347, over 10487.84 utterances.], batch size: 43, lr: 2.12e-02, grad_scale: 8.0 2023-03-07 18:00:04,583 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7065, 5.0670, 4.5776, 5.2360, 4.5551, 4.9040, 5.2505, 5.1076], device='cuda:3'), covar=tensor([0.0344, 0.0356, 0.0768, 0.0155, 0.0427, 0.0195, 0.0238, 0.0154], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0168, 0.0217, 0.0139, 0.0187, 0.0137, 0.0162, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:00:19,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.498e+02 4.244e+02 5.016e+02 1.623e+03, threshold=8.488e+02, percent-clipped=3.0 2023-03-07 18:00:22,224 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:00:54,270 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0559, 1.5030, 1.9744, 2.5741, 2.1971, 2.2756, 2.0875, 2.2907], device='cuda:3'), covar=tensor([0.0574, 0.2596, 0.2426, 0.1209, 0.0890, 0.1284, 0.1727, 0.1217], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0104, 0.0085, 0.0086, 0.0080, 0.0102, 0.0089], device='cuda:3'), out_proj_covar=tensor([3.9582e-05, 5.1727e-05, 5.3367e-05, 4.3442e-05, 4.0696e-05, 4.3868e-05, 5.0872e-05, 4.7164e-05], device='cuda:3') 2023-03-07 18:01:04,844 INFO [train2.py:809] (3/4) Epoch 5, batch 1700, loss[ctc_loss=0.1391, att_loss=0.2722, loss=0.2456, over 16145.00 frames. utt_duration=1539 frames, utt_pad_proportion=0.004987, over 42.00 utterances.], tot_loss[ctc_loss=0.1637, att_loss=0.2819, loss=0.2583, over 3276803.78 frames. utt_duration=1241 frames, utt_pad_proportion=0.05454, over 10575.91 utterances.], batch size: 42, lr: 2.12e-02, grad_scale: 8.0 2023-03-07 18:02:26,020 INFO [train2.py:809] (3/4) Epoch 5, batch 1750, loss[ctc_loss=0.182, att_loss=0.303, loss=0.2788, over 17246.00 frames. utt_duration=1171 frames, utt_pad_proportion=0.02684, over 59.00 utterances.], tot_loss[ctc_loss=0.1624, att_loss=0.2811, loss=0.2574, over 3282320.32 frames. utt_duration=1261 frames, utt_pad_proportion=0.04825, over 10426.05 utterances.], batch size: 59, lr: 2.12e-02, grad_scale: 8.0 2023-03-07 18:02:31,220 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8966, 4.8350, 4.7542, 4.7621, 5.3055, 5.1717, 4.7809, 2.2996], device='cuda:3'), covar=tensor([0.0186, 0.0239, 0.0180, 0.0223, 0.0828, 0.0129, 0.0231, 0.2809], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0118, 0.0115, 0.0125, 0.0278, 0.0128, 0.0110, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 18:02:47,899 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:03:02,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.276e+02 3.443e+02 4.314e+02 5.475e+02 1.397e+03, threshold=8.628e+02, percent-clipped=3.0 2023-03-07 18:03:17,851 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:03:37,969 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:03:46,849 INFO [train2.py:809] (3/4) Epoch 5, batch 1800, loss[ctc_loss=0.1688, att_loss=0.2862, loss=0.2627, over 17442.00 frames. utt_duration=1109 frames, utt_pad_proportion=0.03071, over 63.00 utterances.], tot_loss[ctc_loss=0.1617, att_loss=0.2809, loss=0.2571, over 3273862.28 frames. utt_duration=1271 frames, utt_pad_proportion=0.04734, over 10311.98 utterances.], batch size: 63, lr: 2.11e-02, grad_scale: 8.0 2023-03-07 18:03:51,608 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 18:04:27,117 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:05:07,330 INFO [train2.py:809] (3/4) Epoch 5, batch 1850, loss[ctc_loss=0.1694, att_loss=0.2758, loss=0.2545, over 16018.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006769, over 40.00 utterances.], tot_loss[ctc_loss=0.1619, att_loss=0.2805, loss=0.2567, over 3262116.39 frames. utt_duration=1250 frames, utt_pad_proportion=0.05657, over 10450.99 utterances.], batch size: 40, lr: 2.11e-02, grad_scale: 8.0 2023-03-07 18:05:09,328 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2569, 4.9153, 4.7824, 2.8665, 1.9207, 2.4140, 4.8350, 3.6770], device='cuda:3'), covar=tensor([0.0443, 0.0169, 0.0214, 0.2176, 0.6731, 0.2953, 0.0159, 0.1855], device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0166, 0.0192, 0.0179, 0.0361, 0.0323, 0.0173, 0.0307], device='cuda:3'), out_proj_covar=tensor([1.3758e-04, 7.5192e-05, 8.8594e-05, 8.3852e-05, 1.8012e-04, 1.5208e-04, 7.8566e-05, 1.5363e-04], device='cuda:3') 2023-03-07 18:05:15,561 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:05:45,055 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.153e+02 3.254e+02 3.960e+02 5.162e+02 1.019e+03, threshold=7.921e+02, percent-clipped=3.0 2023-03-07 18:06:28,245 INFO [train2.py:809] (3/4) Epoch 5, batch 1900, loss[ctc_loss=0.1522, att_loss=0.258, loss=0.2368, over 14486.00 frames. utt_duration=1812 frames, utt_pad_proportion=0.03668, over 32.00 utterances.], tot_loss[ctc_loss=0.1624, att_loss=0.2808, loss=0.2571, over 3263055.28 frames. utt_duration=1264 frames, utt_pad_proportion=0.05349, over 10341.21 utterances.], batch size: 32, lr: 2.11e-02, grad_scale: 8.0 2023-03-07 18:06:42,551 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:07:49,566 INFO [train2.py:809] (3/4) Epoch 5, batch 1950, loss[ctc_loss=0.1502, att_loss=0.2781, loss=0.2526, over 16532.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006692, over 45.00 utterances.], tot_loss[ctc_loss=0.1618, att_loss=0.2807, loss=0.2569, over 3276497.62 frames. utt_duration=1282 frames, utt_pad_proportion=0.04576, over 10234.79 utterances.], batch size: 45, lr: 2.11e-02, grad_scale: 8.0 2023-03-07 18:07:51,287 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6451, 5.1306, 5.1544, 5.0853, 5.1932, 5.2432, 4.9550, 4.9316], device='cuda:3'), covar=tensor([0.1301, 0.0489, 0.0215, 0.0598, 0.0462, 0.0305, 0.0310, 0.0305], device='cuda:3'), in_proj_covar=tensor([0.0380, 0.0231, 0.0156, 0.0202, 0.0251, 0.0275, 0.0211, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 18:08:00,637 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:08:24,123 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9241, 5.3368, 4.8204, 5.4067, 4.8096, 5.0595, 5.5384, 5.3384], device='cuda:3'), covar=tensor([0.0400, 0.0268, 0.0632, 0.0138, 0.0388, 0.0161, 0.0198, 0.0123], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0170, 0.0220, 0.0139, 0.0189, 0.0138, 0.0161, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:08:27,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 3.619e+02 4.276e+02 5.445e+02 1.197e+03, threshold=8.551e+02, percent-clipped=4.0 2023-03-07 18:09:06,161 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6666, 6.0242, 5.2957, 6.0293, 5.6081, 5.3315, 5.3249, 5.3751], device='cuda:3'), covar=tensor([0.1480, 0.0942, 0.0820, 0.0575, 0.0623, 0.1524, 0.2779, 0.2186], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0384, 0.0301, 0.0305, 0.0281, 0.0369, 0.0413, 0.0384], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 18:09:07,888 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7514, 5.0774, 4.6642, 5.2317, 4.6146, 4.9465, 5.3376, 5.1113], device='cuda:3'), covar=tensor([0.0320, 0.0226, 0.0644, 0.0121, 0.0404, 0.0153, 0.0214, 0.0149], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0170, 0.0218, 0.0138, 0.0187, 0.0137, 0.0161, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:09:10,810 INFO [train2.py:809] (3/4) Epoch 5, batch 2000, loss[ctc_loss=0.1824, att_loss=0.2976, loss=0.2745, over 17350.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02102, over 59.00 utterances.], tot_loss[ctc_loss=0.1614, att_loss=0.2808, loss=0.257, over 3270774.20 frames. utt_duration=1279 frames, utt_pad_proportion=0.04817, over 10244.14 utterances.], batch size: 59, lr: 2.10e-02, grad_scale: 8.0 2023-03-07 18:09:50,814 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:10:32,015 INFO [train2.py:809] (3/4) Epoch 5, batch 2050, loss[ctc_loss=0.1476, att_loss=0.2776, loss=0.2516, over 16553.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005567, over 45.00 utterances.], tot_loss[ctc_loss=0.1601, att_loss=0.2801, loss=0.2561, over 3273093.25 frames. utt_duration=1277 frames, utt_pad_proportion=0.04873, over 10265.16 utterances.], batch size: 45, lr: 2.10e-02, grad_scale: 8.0 2023-03-07 18:11:14,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.268e+02 3.678e+02 4.390e+02 5.847e+02 1.003e+03, threshold=8.779e+02, percent-clipped=6.0 2023-03-07 18:11:29,698 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:11:34,415 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:11:57,537 INFO [train2.py:809] (3/4) Epoch 5, batch 2100, loss[ctc_loss=0.1764, att_loss=0.2988, loss=0.2743, over 17109.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01471, over 56.00 utterances.], tot_loss[ctc_loss=0.1615, att_loss=0.2806, loss=0.2568, over 3268474.47 frames. utt_duration=1258 frames, utt_pad_proportion=0.05461, over 10401.57 utterances.], batch size: 56, lr: 2.10e-02, grad_scale: 8.0 2023-03-07 18:12:02,540 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 18:12:30,800 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:12:46,834 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:13:17,729 INFO [train2.py:809] (3/4) Epoch 5, batch 2150, loss[ctc_loss=0.1342, att_loss=0.2601, loss=0.2349, over 16532.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006737, over 45.00 utterances.], tot_loss[ctc_loss=0.1613, att_loss=0.2801, loss=0.2563, over 3259267.26 frames. utt_duration=1274 frames, utt_pad_proportion=0.05193, over 10241.77 utterances.], batch size: 45, lr: 2.09e-02, grad_scale: 8.0 2023-03-07 18:13:17,955 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:13:19,445 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 18:13:51,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-07 18:13:54,648 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.249e+02 3.704e+02 4.287e+02 5.437e+02 1.158e+03, threshold=8.574e+02, percent-clipped=4.0 2023-03-07 18:14:38,379 INFO [train2.py:809] (3/4) Epoch 5, batch 2200, loss[ctc_loss=0.1279, att_loss=0.2396, loss=0.2173, over 16168.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.00698, over 41.00 utterances.], tot_loss[ctc_loss=0.1611, att_loss=0.2804, loss=0.2565, over 3262443.12 frames. utt_duration=1263 frames, utt_pad_proportion=0.05374, over 10348.27 utterances.], batch size: 41, lr: 2.09e-02, grad_scale: 8.0 2023-03-07 18:15:13,676 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6556, 5.0422, 4.5093, 5.0213, 4.5442, 4.8319, 5.1772, 4.9634], device='cuda:3'), covar=tensor([0.0327, 0.0281, 0.0716, 0.0162, 0.0374, 0.0168, 0.0209, 0.0158], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0166, 0.0212, 0.0136, 0.0181, 0.0133, 0.0158, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:15:35,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 2023-03-07 18:15:39,778 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3426, 4.4967, 4.6292, 4.8206, 2.2752, 4.6935, 2.4171, 2.0294], device='cuda:3'), covar=tensor([0.0189, 0.0156, 0.0675, 0.0117, 0.2902, 0.0118, 0.1910, 0.1988], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0101, 0.0241, 0.0110, 0.0232, 0.0095, 0.0217, 0.0205], device='cuda:3'), out_proj_covar=tensor([1.0285e-04, 9.9311e-05, 2.1051e-04, 9.6485e-05, 2.0297e-04, 9.0113e-05, 1.8684e-04, 1.7890e-04], device='cuda:3') 2023-03-07 18:15:41,254 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:15:42,907 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:16:00,106 INFO [train2.py:809] (3/4) Epoch 5, batch 2250, loss[ctc_loss=0.1271, att_loss=0.2419, loss=0.219, over 15654.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008307, over 37.00 utterances.], tot_loss[ctc_loss=0.1596, att_loss=0.2793, loss=0.2553, over 3265217.39 frames. utt_duration=1263 frames, utt_pad_proportion=0.05235, over 10355.32 utterances.], batch size: 37, lr: 2.09e-02, grad_scale: 8.0 2023-03-07 18:16:38,767 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.354e+02 3.428e+02 4.207e+02 5.393e+02 1.191e+03, threshold=8.413e+02, percent-clipped=8.0 2023-03-07 18:17:05,809 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6235, 3.9043, 3.8958, 4.0480, 3.9989, 3.8313, 3.2091, 3.8084], device='cuda:3'), covar=tensor([0.0104, 0.0131, 0.0086, 0.0058, 0.0059, 0.0096, 0.0421, 0.0197], device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0054, 0.0055, 0.0038, 0.0037, 0.0047, 0.0071, 0.0066], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 18:17:20,591 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:17:21,715 INFO [train2.py:809] (3/4) Epoch 5, batch 2300, loss[ctc_loss=0.1554, att_loss=0.2902, loss=0.2633, over 17073.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007794, over 52.00 utterances.], tot_loss[ctc_loss=0.1592, att_loss=0.2794, loss=0.2554, over 3270335.91 frames. utt_duration=1241 frames, utt_pad_proportion=0.05547, over 10551.52 utterances.], batch size: 52, lr: 2.09e-02, grad_scale: 8.0 2023-03-07 18:17:22,132 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:17:28,521 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3114, 2.3875, 1.9723, 1.8001, 1.8579, 1.5453, 1.6580, 3.0282], device='cuda:3'), covar=tensor([0.0423, 0.1105, 0.2051, 0.1543, 0.0889, 0.1258, 0.1649, 0.0918], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0100, 0.0105, 0.0089, 0.0086, 0.0083, 0.0103, 0.0089], device='cuda:3'), out_proj_covar=tensor([4.0692e-05, 5.3535e-05, 5.4815e-05, 4.5441e-05, 4.0742e-05, 4.5596e-05, 5.2500e-05, 4.8857e-05], device='cuda:3') 2023-03-07 18:18:31,625 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9686, 6.1658, 5.5532, 6.0566, 5.7655, 5.5527, 5.6220, 5.3224], device='cuda:3'), covar=tensor([0.1017, 0.0731, 0.0692, 0.0655, 0.0637, 0.1055, 0.1924, 0.2185], device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0375, 0.0292, 0.0299, 0.0277, 0.0355, 0.0402, 0.0377], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 18:18:43,284 INFO [train2.py:809] (3/4) Epoch 5, batch 2350, loss[ctc_loss=0.1457, att_loss=0.2647, loss=0.2409, over 15876.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009736, over 39.00 utterances.], tot_loss[ctc_loss=0.1597, att_loss=0.2796, loss=0.2556, over 3269721.22 frames. utt_duration=1254 frames, utt_pad_proportion=0.05372, over 10444.90 utterances.], batch size: 39, lr: 2.08e-02, grad_scale: 8.0 2023-03-07 18:18:45,097 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0846, 4.5618, 4.3780, 4.6862, 4.6498, 4.3522, 3.4693, 4.3815], device='cuda:3'), covar=tensor([0.0113, 0.0183, 0.0111, 0.0071, 0.0080, 0.0099, 0.0501, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0054, 0.0056, 0.0038, 0.0038, 0.0047, 0.0071, 0.0067], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 18:19:21,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.258e+02 3.468e+02 4.143e+02 4.951e+02 1.211e+03, threshold=8.286e+02, percent-clipped=3.0 2023-03-07 18:19:27,745 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6719, 5.0976, 4.9185, 4.8759, 5.1119, 4.9722, 4.7702, 4.5881], device='cuda:3'), covar=tensor([0.1108, 0.0362, 0.0216, 0.0497, 0.0268, 0.0303, 0.0275, 0.0356], device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0233, 0.0163, 0.0206, 0.0254, 0.0279, 0.0212, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 18:19:32,553 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:19:40,595 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9191, 5.3219, 5.1857, 5.2209, 5.3786, 5.2291, 5.0582, 4.9034], device='cuda:3'), covar=tensor([0.0989, 0.0354, 0.0191, 0.0332, 0.0227, 0.0253, 0.0215, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0386, 0.0231, 0.0162, 0.0203, 0.0252, 0.0276, 0.0209, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 18:19:50,024 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5400, 3.6144, 2.9790, 3.1857, 3.7785, 3.3199, 2.4671, 4.1358], device='cuda:3'), covar=tensor([0.1140, 0.0452, 0.1087, 0.0724, 0.0495, 0.0706, 0.1031, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0129, 0.0177, 0.0143, 0.0157, 0.0168, 0.0149, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 18:20:04,250 INFO [train2.py:809] (3/4) Epoch 5, batch 2400, loss[ctc_loss=0.1381, att_loss=0.2733, loss=0.2462, over 16539.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005676, over 45.00 utterances.], tot_loss[ctc_loss=0.1603, att_loss=0.2803, loss=0.2563, over 3274027.02 frames. utt_duration=1239 frames, utt_pad_proportion=0.05541, over 10578.83 utterances.], batch size: 45, lr: 2.08e-02, grad_scale: 16.0 2023-03-07 18:20:37,116 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.1795, 3.0000, 3.3233, 2.1729, 2.8624, 3.3252, 3.1622, 1.3911], device='cuda:3'), covar=tensor([0.1364, 0.1265, 0.1439, 0.5211, 0.1960, 0.1691, 0.0684, 1.1700], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0072, 0.0071, 0.0095, 0.0071, 0.0090, 0.0065, 0.0115], device='cuda:3'), out_proj_covar=tensor([5.0729e-05, 4.7873e-05, 4.8613e-05, 6.7796e-05, 4.9343e-05, 6.7410e-05, 4.7972e-05, 8.5954e-05], device='cuda:3') 2023-03-07 18:20:38,640 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:21:26,052 INFO [train2.py:809] (3/4) Epoch 5, batch 2450, loss[ctc_loss=0.1974, att_loss=0.2978, loss=0.2778, over 16611.00 frames. utt_duration=679.5 frames, utt_pad_proportion=0.1484, over 98.00 utterances.], tot_loss[ctc_loss=0.1602, att_loss=0.2796, loss=0.2557, over 3269395.23 frames. utt_duration=1246 frames, utt_pad_proportion=0.05445, over 10508.92 utterances.], batch size: 98, lr: 2.08e-02, grad_scale: 16.0 2023-03-07 18:21:26,380 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:21:56,556 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:22:04,373 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.376e+02 4.245e+02 5.566e+02 1.055e+03, threshold=8.490e+02, percent-clipped=7.0 2023-03-07 18:22:28,797 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-07 18:22:44,602 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:22:47,529 INFO [train2.py:809] (3/4) Epoch 5, batch 2500, loss[ctc_loss=0.134, att_loss=0.2761, loss=0.2477, over 17305.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02417, over 59.00 utterances.], tot_loss[ctc_loss=0.1599, att_loss=0.2795, loss=0.2556, over 3267562.10 frames. utt_duration=1223 frames, utt_pad_proportion=0.0613, over 10703.75 utterances.], batch size: 59, lr: 2.08e-02, grad_scale: 16.0 2023-03-07 18:23:01,574 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7255, 5.2546, 4.6548, 5.2981, 4.5967, 4.9088, 5.4284, 5.1606], device='cuda:3'), covar=tensor([0.0368, 0.0232, 0.0725, 0.0144, 0.0388, 0.0207, 0.0169, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0175, 0.0225, 0.0143, 0.0192, 0.0142, 0.0165, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:23:58,082 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1402, 1.8168, 2.2364, 1.7152, 2.7646, 1.8404, 1.6993, 3.3513], device='cuda:3'), covar=tensor([0.0720, 0.2643, 0.2740, 0.1454, 0.0714, 0.1644, 0.1898, 0.0587], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0103, 0.0105, 0.0088, 0.0085, 0.0085, 0.0102, 0.0087], device='cuda:3'), out_proj_covar=tensor([4.2349e-05, 5.5299e-05, 5.5279e-05, 4.5429e-05, 4.0717e-05, 4.6478e-05, 5.2299e-05, 4.8056e-05], device='cuda:3') 2023-03-07 18:24:03,326 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0282, 1.7769, 1.9361, 1.6235, 2.6041, 1.8970, 1.5724, 3.1665], device='cuda:3'), covar=tensor([0.0594, 0.2661, 0.2737, 0.1844, 0.0775, 0.1381, 0.2004, 0.0577], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0103, 0.0105, 0.0088, 0.0085, 0.0086, 0.0102, 0.0087], device='cuda:3'), out_proj_covar=tensor([4.2425e-05, 5.5361e-05, 5.5397e-05, 4.5523e-05, 4.0794e-05, 4.6543e-05, 5.2404e-05, 4.8129e-05], device='cuda:3') 2023-03-07 18:24:09,075 INFO [train2.py:809] (3/4) Epoch 5, batch 2550, loss[ctc_loss=0.2344, att_loss=0.3247, loss=0.3067, over 14350.00 frames. utt_duration=392 frames, utt_pad_proportion=0.3122, over 147.00 utterances.], tot_loss[ctc_loss=0.1598, att_loss=0.2795, loss=0.2556, over 3270811.67 frames. utt_duration=1211 frames, utt_pad_proportion=0.06381, over 10813.77 utterances.], batch size: 147, lr: 2.07e-02, grad_scale: 16.0 2023-03-07 18:24:16,259 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:24:20,293 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 18:24:44,672 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 18:24:45,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 3.434e+02 4.162e+02 5.436e+02 1.075e+03, threshold=8.324e+02, percent-clipped=2.0 2023-03-07 18:25:20,136 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:25:21,725 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:25:23,468 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.6807, 3.0505, 3.1294, 2.1070, 2.8315, 2.8736, 3.3307, 1.8239], device='cuda:3'), covar=tensor([0.1773, 0.0969, 0.6704, 0.5035, 0.0950, 0.2897, 0.0529, 0.7518], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0066, 0.0068, 0.0089, 0.0066, 0.0084, 0.0059, 0.0106], device='cuda:3'), out_proj_covar=tensor([4.8119e-05, 4.4757e-05, 4.6959e-05, 6.4247e-05, 4.6381e-05, 6.3222e-05, 4.3871e-05, 7.9649e-05], device='cuda:3') 2023-03-07 18:25:29,424 INFO [train2.py:809] (3/4) Epoch 5, batch 2600, loss[ctc_loss=0.1676, att_loss=0.2931, loss=0.268, over 17139.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.0132, over 56.00 utterances.], tot_loss[ctc_loss=0.1596, att_loss=0.2796, loss=0.2556, over 3273873.60 frames. utt_duration=1215 frames, utt_pad_proportion=0.06196, over 10794.70 utterances.], batch size: 56, lr: 2.07e-02, grad_scale: 16.0 2023-03-07 18:25:37,561 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 18:25:54,382 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:26:05,450 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1262, 1.6187, 2.2854, 1.9454, 2.0797, 1.9190, 1.7797, 2.9011], device='cuda:3'), covar=tensor([0.0931, 0.4464, 0.4691, 0.2569, 0.2386, 0.2870, 0.2867, 0.1415], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0103, 0.0108, 0.0091, 0.0089, 0.0088, 0.0101, 0.0090], device='cuda:3'), out_proj_covar=tensor([4.2526e-05, 5.5364e-05, 5.6727e-05, 4.6602e-05, 4.2327e-05, 4.7791e-05, 5.2389e-05, 4.9254e-05], device='cuda:3') 2023-03-07 18:26:22,959 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 18:26:50,081 INFO [train2.py:809] (3/4) Epoch 5, batch 2650, loss[ctc_loss=0.1585, att_loss=0.2693, loss=0.2471, over 16319.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006673, over 45.00 utterances.], tot_loss[ctc_loss=0.1599, att_loss=0.2798, loss=0.2558, over 3275634.74 frames. utt_duration=1218 frames, utt_pad_proportion=0.05921, over 10767.09 utterances.], batch size: 45, lr: 2.07e-02, grad_scale: 16.0 2023-03-07 18:27:20,029 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9194, 5.1146, 5.5037, 5.3896, 5.1620, 5.8424, 5.0324, 5.9223], device='cuda:3'), covar=tensor([0.0538, 0.0699, 0.0563, 0.0712, 0.1872, 0.0788, 0.0580, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0506, 0.0328, 0.0340, 0.0393, 0.0556, 0.0338, 0.0290, 0.0356], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 18:27:27,353 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 3.747e+02 4.549e+02 5.816e+02 1.798e+03, threshold=9.097e+02, percent-clipped=9.0 2023-03-07 18:27:38,434 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:28:10,688 INFO [train2.py:809] (3/4) Epoch 5, batch 2700, loss[ctc_loss=0.1234, att_loss=0.2322, loss=0.2104, over 14567.00 frames. utt_duration=1822 frames, utt_pad_proportion=0.03622, over 32.00 utterances.], tot_loss[ctc_loss=0.1594, att_loss=0.2798, loss=0.2557, over 3280468.13 frames. utt_duration=1247 frames, utt_pad_proportion=0.0514, over 10536.12 utterances.], batch size: 32, lr: 2.07e-02, grad_scale: 16.0 2023-03-07 18:28:42,117 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9540, 4.9671, 4.9710, 2.6697, 4.8261, 4.2471, 4.3468, 2.6321], device='cuda:3'), covar=tensor([0.0121, 0.0071, 0.0148, 0.1114, 0.0084, 0.0195, 0.0268, 0.1298], device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0060, 0.0048, 0.0097, 0.0058, 0.0069, 0.0078, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 18:28:56,338 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:29:13,361 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:29:24,848 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8215, 4.7660, 4.7148, 4.8567, 5.1017, 5.0601, 4.5856, 2.1301], device='cuda:3'), covar=tensor([0.0231, 0.0272, 0.0186, 0.0219, 0.1000, 0.0208, 0.0253, 0.3112], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0120, 0.0119, 0.0124, 0.0285, 0.0128, 0.0116, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 18:29:31,904 INFO [train2.py:809] (3/4) Epoch 5, batch 2750, loss[ctc_loss=0.1586, att_loss=0.2897, loss=0.2634, over 17093.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01507, over 56.00 utterances.], tot_loss[ctc_loss=0.1597, att_loss=0.2802, loss=0.2561, over 3278262.06 frames. utt_duration=1250 frames, utt_pad_proportion=0.0519, over 10504.07 utterances.], batch size: 56, lr: 2.06e-02, grad_scale: 16.0 2023-03-07 18:30:10,719 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.386e+02 3.574e+02 4.434e+02 5.832e+02 1.352e+03, threshold=8.867e+02, percent-clipped=7.0 2023-03-07 18:30:29,970 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1484, 4.7777, 4.8880, 2.7383, 2.0112, 2.6771, 4.5873, 3.7733], device='cuda:3'), covar=tensor([0.0519, 0.0173, 0.0148, 0.2374, 0.5895, 0.2528, 0.0223, 0.1538], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0171, 0.0195, 0.0178, 0.0357, 0.0324, 0.0176, 0.0307], device='cuda:3'), out_proj_covar=tensor([1.4060e-04, 7.5785e-05, 8.8575e-05, 8.2697e-05, 1.7641e-04, 1.4931e-04, 7.7908e-05, 1.5157e-04], device='cuda:3') 2023-03-07 18:30:41,033 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-07 18:30:51,277 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:30:52,488 INFO [train2.py:809] (3/4) Epoch 5, batch 2800, loss[ctc_loss=0.1578, att_loss=0.295, loss=0.2675, over 17022.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.00832, over 51.00 utterances.], tot_loss[ctc_loss=0.161, att_loss=0.2811, loss=0.2571, over 3280939.47 frames. utt_duration=1231 frames, utt_pad_proportion=0.05534, over 10671.42 utterances.], batch size: 51, lr: 2.06e-02, grad_scale: 8.0 2023-03-07 18:31:02,291 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-07 18:31:48,736 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7524, 5.1960, 4.9608, 5.0007, 5.2066, 5.1755, 4.8518, 4.7115], device='cuda:3'), covar=tensor([0.1155, 0.0380, 0.0262, 0.0460, 0.0314, 0.0296, 0.0280, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0389, 0.0233, 0.0164, 0.0207, 0.0259, 0.0283, 0.0217, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 18:31:51,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-07 18:32:14,578 INFO [train2.py:809] (3/4) Epoch 5, batch 2850, loss[ctc_loss=0.1357, att_loss=0.2511, loss=0.228, over 15489.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009206, over 36.00 utterances.], tot_loss[ctc_loss=0.1589, att_loss=0.279, loss=0.255, over 3271822.44 frames. utt_duration=1246 frames, utt_pad_proportion=0.05513, over 10519.94 utterances.], batch size: 36, lr: 2.06e-02, grad_scale: 8.0 2023-03-07 18:32:15,324 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-03-07 18:32:44,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 18:32:54,813 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 3.458e+02 4.600e+02 5.611e+02 1.235e+03, threshold=9.200e+02, percent-clipped=3.0 2023-03-07 18:33:27,981 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:33:29,511 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:33:29,674 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6813, 2.8043, 5.1393, 3.8999, 3.2113, 4.6280, 4.6876, 4.7984], device='cuda:3'), covar=tensor([0.0188, 0.1624, 0.0166, 0.1217, 0.2032, 0.0240, 0.0159, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0241, 0.0119, 0.0292, 0.0310, 0.0187, 0.0107, 0.0135], device='cuda:3'), out_proj_covar=tensor([1.1634e-04, 1.9224e-04, 1.0119e-04, 2.3526e-04, 2.5180e-04, 1.5733e-04, 9.5131e-05, 1.1863e-04], device='cuda:3') 2023-03-07 18:33:37,330 INFO [train2.py:809] (3/4) Epoch 5, batch 2900, loss[ctc_loss=0.1547, att_loss=0.2566, loss=0.2362, over 11961.00 frames. utt_duration=1842 frames, utt_pad_proportion=0.167, over 26.00 utterances.], tot_loss[ctc_loss=0.1573, att_loss=0.278, loss=0.2539, over 3267350.84 frames. utt_duration=1244 frames, utt_pad_proportion=0.05613, over 10519.64 utterances.], batch size: 26, lr: 2.06e-02, grad_scale: 8.0 2023-03-07 18:33:53,703 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:34:21,717 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 18:34:44,643 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:34:46,081 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:34:56,887 INFO [train2.py:809] (3/4) Epoch 5, batch 2950, loss[ctc_loss=0.1505, att_loss=0.2859, loss=0.2588, over 16991.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009978, over 51.00 utterances.], tot_loss[ctc_loss=0.1562, att_loss=0.2773, loss=0.2531, over 3269510.79 frames. utt_duration=1273 frames, utt_pad_proportion=0.04892, over 10289.55 utterances.], batch size: 51, lr: 2.05e-02, grad_scale: 8.0 2023-03-07 18:35:11,869 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6908, 2.1742, 5.0074, 3.6987, 3.1911, 4.6100, 4.7625, 4.8041], device='cuda:3'), covar=tensor([0.0102, 0.1965, 0.0106, 0.1394, 0.2058, 0.0175, 0.0114, 0.0125], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0246, 0.0124, 0.0301, 0.0320, 0.0191, 0.0111, 0.0138], device='cuda:3'), out_proj_covar=tensor([1.1975e-04, 1.9649e-04, 1.0535e-04, 2.4202e-04, 2.5973e-04, 1.6145e-04, 9.8903e-05, 1.2081e-04], device='cuda:3') 2023-03-07 18:35:34,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.226e+02 3.285e+02 4.050e+02 4.764e+02 8.688e+02, threshold=8.101e+02, percent-clipped=0.0 2023-03-07 18:35:46,990 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-03-07 18:36:03,877 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 18:36:08,031 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:36:17,499 INFO [train2.py:809] (3/4) Epoch 5, batch 3000, loss[ctc_loss=0.1423, att_loss=0.2597, loss=0.2362, over 15955.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006393, over 41.00 utterances.], tot_loss[ctc_loss=0.1577, att_loss=0.2782, loss=0.2541, over 3264868.27 frames. utt_duration=1251 frames, utt_pad_proportion=0.05754, over 10450.24 utterances.], batch size: 41, lr: 2.05e-02, grad_scale: 8.0 2023-03-07 18:36:17,500 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 18:36:31,697 INFO [train2.py:843] (3/4) Epoch 5, validation: ctc_loss=0.07531, att_loss=0.2513, loss=0.2161, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 18:36:31,697 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 18:36:34,325 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-07 18:37:33,296 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-07 18:37:52,826 INFO [train2.py:809] (3/4) Epoch 5, batch 3050, loss[ctc_loss=0.1417, att_loss=0.269, loss=0.2435, over 15944.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.006938, over 41.00 utterances.], tot_loss[ctc_loss=0.1587, att_loss=0.2788, loss=0.2548, over 3262641.07 frames. utt_duration=1205 frames, utt_pad_proportion=0.06899, over 10843.57 utterances.], batch size: 41, lr: 2.05e-02, grad_scale: 8.0 2023-03-07 18:38:01,045 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:38:31,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.565e+02 3.659e+02 4.471e+02 5.591e+02 1.197e+03, threshold=8.941e+02, percent-clipped=8.0 2023-03-07 18:39:04,995 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:39:07,603 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-07 18:39:13,976 INFO [train2.py:809] (3/4) Epoch 5, batch 3100, loss[ctc_loss=0.2457, att_loss=0.3085, loss=0.296, over 16201.00 frames. utt_duration=1582 frames, utt_pad_proportion=0.004924, over 41.00 utterances.], tot_loss[ctc_loss=0.1584, att_loss=0.2791, loss=0.2549, over 3268461.20 frames. utt_duration=1222 frames, utt_pad_proportion=0.06301, over 10707.65 utterances.], batch size: 41, lr: 2.05e-02, grad_scale: 8.0 2023-03-07 18:40:27,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-03-07 18:40:36,262 INFO [train2.py:809] (3/4) Epoch 5, batch 3150, loss[ctc_loss=0.1574, att_loss=0.2931, loss=0.2659, over 17071.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007852, over 52.00 utterances.], tot_loss[ctc_loss=0.1592, att_loss=0.28, loss=0.2558, over 3269028.86 frames. utt_duration=1204 frames, utt_pad_proportion=0.06761, over 10874.78 utterances.], batch size: 52, lr: 2.04e-02, grad_scale: 8.0 2023-03-07 18:41:14,954 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.661e+02 4.363e+02 5.729e+02 1.353e+03, threshold=8.726e+02, percent-clipped=4.0 2023-03-07 18:41:46,870 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 18:41:48,445 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8073, 4.5066, 4.3394, 4.8293, 4.9602, 4.7832, 4.5675, 1.9675], device='cuda:3'), covar=tensor([0.0236, 0.0341, 0.0382, 0.0124, 0.1201, 0.0221, 0.0263, 0.3472], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0126, 0.0123, 0.0125, 0.0297, 0.0136, 0.0119, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 18:41:57,273 INFO [train2.py:809] (3/4) Epoch 5, batch 3200, loss[ctc_loss=0.1106, att_loss=0.2426, loss=0.2162, over 16019.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006738, over 40.00 utterances.], tot_loss[ctc_loss=0.1584, att_loss=0.2795, loss=0.2553, over 3273828.40 frames. utt_duration=1199 frames, utt_pad_proportion=0.06699, over 10931.75 utterances.], batch size: 40, lr: 2.04e-02, grad_scale: 8.0 2023-03-07 18:42:14,088 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:42:43,190 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 18:43:11,780 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8267, 5.0670, 5.3781, 5.3143, 5.0865, 5.7645, 5.1050, 5.8532], device='cuda:3'), covar=tensor([0.0587, 0.0616, 0.0583, 0.0744, 0.1785, 0.0685, 0.0518, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0326, 0.0342, 0.0405, 0.0563, 0.0341, 0.0288, 0.0358], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 18:43:19,470 INFO [train2.py:809] (3/4) Epoch 5, batch 3250, loss[ctc_loss=0.1539, att_loss=0.2827, loss=0.2569, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005424, over 47.00 utterances.], tot_loss[ctc_loss=0.1573, att_loss=0.2785, loss=0.2543, over 3271498.82 frames. utt_duration=1220 frames, utt_pad_proportion=0.06333, over 10739.71 utterances.], batch size: 47, lr: 2.04e-02, grad_scale: 8.0 2023-03-07 18:43:26,706 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 18:43:29,653 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3965, 4.8514, 4.6962, 4.9652, 5.1215, 4.6784, 4.1230, 4.7951], device='cuda:3'), covar=tensor([0.0111, 0.0119, 0.0085, 0.0075, 0.0059, 0.0087, 0.0354, 0.0176], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0055, 0.0059, 0.0040, 0.0041, 0.0051, 0.0073, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-07 18:43:32,676 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:43:58,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.144e+02 3.305e+02 4.293e+02 5.572e+02 8.816e+02, threshold=8.587e+02, percent-clipped=1.0 2023-03-07 18:44:01,776 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 18:44:02,584 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-07 18:44:31,736 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1801, 4.4191, 4.6257, 3.7551, 2.3703, 4.5783, 2.9291, 2.1949], device='cuda:3'), covar=tensor([0.0216, 0.0155, 0.0670, 0.0691, 0.2636, 0.0199, 0.1592, 0.1766], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0101, 0.0247, 0.0117, 0.0228, 0.0103, 0.0221, 0.0208], device='cuda:3'), out_proj_covar=tensor([1.0319e-04, 9.9678e-05, 2.1772e-04, 1.0319e-04, 2.0251e-04, 9.7476e-05, 1.9211e-04, 1.8244e-04], device='cuda:3') 2023-03-07 18:44:40,684 INFO [train2.py:809] (3/4) Epoch 5, batch 3300, loss[ctc_loss=0.1707, att_loss=0.2927, loss=0.2683, over 17356.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03551, over 63.00 utterances.], tot_loss[ctc_loss=0.1569, att_loss=0.2777, loss=0.2536, over 3264112.93 frames. utt_duration=1230 frames, utt_pad_proportion=0.06126, over 10624.73 utterances.], batch size: 63, lr: 2.04e-02, grad_scale: 8.0 2023-03-07 18:44:40,866 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3934, 5.5972, 4.8639, 5.6011, 5.2438, 4.9780, 5.0561, 4.9556], device='cuda:3'), covar=tensor([0.1139, 0.0820, 0.0847, 0.0598, 0.0689, 0.1280, 0.2135, 0.2387], device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0377, 0.0295, 0.0297, 0.0280, 0.0362, 0.0397, 0.0378], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 18:46:01,522 INFO [train2.py:809] (3/4) Epoch 5, batch 3350, loss[ctc_loss=0.1526, att_loss=0.2856, loss=0.259, over 17402.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.04777, over 69.00 utterances.], tot_loss[ctc_loss=0.1569, att_loss=0.2778, loss=0.2537, over 3267121.78 frames. utt_duration=1230 frames, utt_pad_proportion=0.0609, over 10636.48 utterances.], batch size: 69, lr: 2.03e-02, grad_scale: 8.0 2023-03-07 18:46:01,757 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:46:39,662 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.279e+02 3.326e+02 4.027e+02 5.244e+02 1.066e+03, threshold=8.053e+02, percent-clipped=2.0 2023-03-07 18:47:13,625 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:47:22,652 INFO [train2.py:809] (3/4) Epoch 5, batch 3400, loss[ctc_loss=0.1773, att_loss=0.3041, loss=0.2788, over 17037.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.008787, over 52.00 utterances.], tot_loss[ctc_loss=0.1575, att_loss=0.2783, loss=0.2542, over 3267521.87 frames. utt_duration=1213 frames, utt_pad_proportion=0.0647, over 10785.23 utterances.], batch size: 52, lr: 2.03e-02, grad_scale: 8.0 2023-03-07 18:47:53,616 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 18:48:30,579 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:48:42,955 INFO [train2.py:809] (3/4) Epoch 5, batch 3450, loss[ctc_loss=0.1483, att_loss=0.26, loss=0.2377, over 15761.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.00786, over 38.00 utterances.], tot_loss[ctc_loss=0.1566, att_loss=0.2779, loss=0.2537, over 3269412.65 frames. utt_duration=1233 frames, utt_pad_proportion=0.05935, over 10620.23 utterances.], batch size: 38, lr: 2.03e-02, grad_scale: 8.0 2023-03-07 18:48:56,568 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:49:21,715 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 3.490e+02 4.079e+02 5.311e+02 1.453e+03, threshold=8.158e+02, percent-clipped=3.0 2023-03-07 18:50:03,982 INFO [train2.py:809] (3/4) Epoch 5, batch 3500, loss[ctc_loss=0.1476, att_loss=0.2528, loss=0.2318, over 15385.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.01028, over 35.00 utterances.], tot_loss[ctc_loss=0.1568, att_loss=0.2778, loss=0.2536, over 3270187.54 frames. utt_duration=1265 frames, utt_pad_proportion=0.05107, over 10352.93 utterances.], batch size: 35, lr: 2.03e-02, grad_scale: 8.0 2023-03-07 18:50:34,951 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:50:46,103 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3096, 4.5050, 4.5694, 4.8556, 2.4602, 4.7212, 2.5836, 2.0498], device='cuda:3'), covar=tensor([0.0191, 0.0135, 0.0686, 0.0130, 0.2578, 0.0172, 0.1850, 0.1878], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0100, 0.0247, 0.0111, 0.0229, 0.0100, 0.0225, 0.0208], device='cuda:3'), out_proj_covar=tensor([1.0673e-04, 9.9583e-05, 2.1808e-04, 9.9294e-05, 2.0374e-04, 9.6602e-05, 1.9589e-04, 1.8284e-04], device='cuda:3') 2023-03-07 18:51:24,360 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 18:51:25,709 INFO [train2.py:809] (3/4) Epoch 5, batch 3550, loss[ctc_loss=0.1576, att_loss=0.2849, loss=0.2594, over 16957.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007918, over 50.00 utterances.], tot_loss[ctc_loss=0.1583, att_loss=0.2792, loss=0.255, over 3263871.23 frames. utt_duration=1225 frames, utt_pad_proportion=0.06104, over 10670.01 utterances.], batch size: 50, lr: 2.02e-02, grad_scale: 8.0 2023-03-07 18:51:30,591 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6953, 5.1088, 4.4733, 5.2382, 4.4357, 4.9254, 5.2395, 5.0110], device='cuda:3'), covar=tensor([0.0411, 0.0270, 0.0899, 0.0164, 0.0515, 0.0225, 0.0285, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0177, 0.0232, 0.0148, 0.0194, 0.0146, 0.0166, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 18:52:02,969 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.286e+02 3.371e+02 4.237e+02 4.987e+02 1.005e+03, threshold=8.474e+02, percent-clipped=3.0 2023-03-07 18:52:06,540 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2820, 5.2400, 5.2478, 3.3327, 5.1842, 4.2574, 4.8539, 2.7451], device='cuda:3'), covar=tensor([0.0120, 0.0059, 0.0136, 0.0835, 0.0065, 0.0184, 0.0188, 0.1365], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0060, 0.0050, 0.0097, 0.0057, 0.0070, 0.0080, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 18:52:46,422 INFO [train2.py:809] (3/4) Epoch 5, batch 3600, loss[ctc_loss=0.1342, att_loss=0.2657, loss=0.2394, over 16322.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006734, over 45.00 utterances.], tot_loss[ctc_loss=0.1583, att_loss=0.2788, loss=0.2547, over 3261249.56 frames. utt_duration=1217 frames, utt_pad_proportion=0.06442, over 10732.46 utterances.], batch size: 45, lr: 2.02e-02, grad_scale: 8.0 2023-03-07 18:52:49,059 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-07 18:53:15,211 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-03-07 18:54:07,429 INFO [train2.py:809] (3/4) Epoch 5, batch 3650, loss[ctc_loss=0.2038, att_loss=0.3057, loss=0.2853, over 17282.00 frames. utt_duration=876.9 frames, utt_pad_proportion=0.08078, over 79.00 utterances.], tot_loss[ctc_loss=0.1584, att_loss=0.2789, loss=0.2548, over 3258960.13 frames. utt_duration=1218 frames, utt_pad_proportion=0.06415, over 10711.82 utterances.], batch size: 79, lr: 2.02e-02, grad_scale: 8.0 2023-03-07 18:54:07,765 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:54:44,974 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.264e+02 3.543e+02 4.665e+02 6.112e+02 1.723e+03, threshold=9.329e+02, percent-clipped=10.0 2023-03-07 18:54:59,255 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:55:25,704 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:55:28,660 INFO [train2.py:809] (3/4) Epoch 5, batch 3700, loss[ctc_loss=0.1631, att_loss=0.2943, loss=0.2681, over 17347.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03371, over 63.00 utterances.], tot_loss[ctc_loss=0.1576, att_loss=0.2786, loss=0.2544, over 3268568.07 frames. utt_duration=1216 frames, utt_pad_proportion=0.0625, over 10763.49 utterances.], batch size: 63, lr: 2.02e-02, grad_scale: 8.0 2023-03-07 18:56:39,224 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:56:49,596 INFO [train2.py:809] (3/4) Epoch 5, batch 3750, loss[ctc_loss=0.152, att_loss=0.2781, loss=0.2529, over 16679.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006339, over 46.00 utterances.], tot_loss[ctc_loss=0.1576, att_loss=0.2789, loss=0.2547, over 3279148.65 frames. utt_duration=1208 frames, utt_pad_proportion=0.06083, over 10873.66 utterances.], batch size: 46, lr: 2.01e-02, grad_scale: 8.0 2023-03-07 18:57:27,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.299e+02 3.451e+02 3.922e+02 5.029e+02 1.290e+03, threshold=7.843e+02, percent-clipped=3.0 2023-03-07 18:57:32,147 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-07 18:58:11,650 INFO [train2.py:809] (3/4) Epoch 5, batch 3800, loss[ctc_loss=0.1478, att_loss=0.2788, loss=0.2526, over 16117.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006237, over 42.00 utterances.], tot_loss[ctc_loss=0.156, att_loss=0.2782, loss=0.2538, over 3273117.52 frames. utt_duration=1230 frames, utt_pad_proportion=0.05822, over 10661.31 utterances.], batch size: 42, lr: 2.01e-02, grad_scale: 8.0 2023-03-07 18:58:21,395 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0624, 5.2970, 5.5070, 5.5251, 5.3118, 5.9857, 5.0328, 6.0757], device='cuda:3'), covar=tensor([0.0473, 0.0604, 0.0605, 0.0833, 0.1450, 0.0798, 0.0506, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0326, 0.0355, 0.0418, 0.0574, 0.0355, 0.0298, 0.0364], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 18:58:33,854 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 18:59:30,531 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 18:59:31,759 INFO [train2.py:809] (3/4) Epoch 5, batch 3850, loss[ctc_loss=0.1494, att_loss=0.2671, loss=0.2435, over 16173.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006688, over 41.00 utterances.], tot_loss[ctc_loss=0.1563, att_loss=0.2781, loss=0.2538, over 3266756.48 frames. utt_duration=1226 frames, utt_pad_proportion=0.06132, over 10669.52 utterances.], batch size: 41, lr: 2.01e-02, grad_scale: 8.0 2023-03-07 19:00:09,266 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.273e+02 3.504e+02 4.097e+02 4.952e+02 1.319e+03, threshold=8.195e+02, percent-clipped=3.0 2023-03-07 19:00:28,703 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-07 19:00:44,704 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 19:00:48,870 INFO [train2.py:809] (3/4) Epoch 5, batch 3900, loss[ctc_loss=0.168, att_loss=0.3008, loss=0.2743, over 16862.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008692, over 49.00 utterances.], tot_loss[ctc_loss=0.1572, att_loss=0.2783, loss=0.2541, over 3272001.55 frames. utt_duration=1235 frames, utt_pad_proportion=0.0577, over 10613.83 utterances.], batch size: 49, lr: 2.01e-02, grad_scale: 8.0 2023-03-07 19:02:05,453 INFO [train2.py:809] (3/4) Epoch 5, batch 3950, loss[ctc_loss=0.1763, att_loss=0.2994, loss=0.2748, over 16272.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007706, over 43.00 utterances.], tot_loss[ctc_loss=0.1566, att_loss=0.2775, loss=0.2534, over 3260171.52 frames. utt_duration=1252 frames, utt_pad_proportion=0.0557, over 10430.17 utterances.], batch size: 43, lr: 2.00e-02, grad_scale: 8.0 2023-03-07 19:02:38,740 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8671, 5.2840, 4.6528, 5.2561, 4.6728, 5.0420, 5.4201, 5.1901], device='cuda:3'), covar=tensor([0.0326, 0.0183, 0.0654, 0.0144, 0.0416, 0.0145, 0.0177, 0.0119], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0182, 0.0236, 0.0154, 0.0199, 0.0147, 0.0172, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 19:02:43,149 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.585e+02 3.417e+02 4.181e+02 5.217e+02 1.635e+03, threshold=8.363e+02, percent-clipped=3.0 2023-03-07 19:03:26,309 INFO [train2.py:809] (3/4) Epoch 6, batch 0, loss[ctc_loss=0.185, att_loss=0.2983, loss=0.2757, over 16958.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007412, over 50.00 utterances.], tot_loss[ctc_loss=0.185, att_loss=0.2983, loss=0.2757, over 16958.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007412, over 50.00 utterances.], batch size: 50, lr: 1.87e-02, grad_scale: 8.0 2023-03-07 19:03:26,309 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 19:03:38,950 INFO [train2.py:843] (3/4) Epoch 6, validation: ctc_loss=0.07525, att_loss=0.2502, loss=0.2152, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 19:03:38,951 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 19:04:58,945 INFO [train2.py:809] (3/4) Epoch 6, batch 50, loss[ctc_loss=0.1821, att_loss=0.2969, loss=0.2739, over 17341.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03631, over 63.00 utterances.], tot_loss[ctc_loss=0.1575, att_loss=0.2797, loss=0.2553, over 740674.75 frames. utt_duration=1134 frames, utt_pad_proportion=0.07474, over 2615.87 utterances.], batch size: 63, lr: 1.87e-02, grad_scale: 8.0 2023-03-07 19:05:05,355 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:06:01,869 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2023-03-07 19:06:05,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 3.240e+02 3.986e+02 5.094e+02 1.358e+03, threshold=7.972e+02, percent-clipped=5.0 2023-03-07 19:06:17,289 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6432, 5.2118, 4.9317, 5.0943, 5.1390, 4.6840, 4.0991, 5.0797], device='cuda:3'), covar=tensor([0.0081, 0.0083, 0.0066, 0.0057, 0.0067, 0.0107, 0.0364, 0.0134], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0052, 0.0057, 0.0040, 0.0040, 0.0050, 0.0072, 0.0068], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-07 19:06:22,314 INFO [train2.py:809] (3/4) Epoch 6, batch 100, loss[ctc_loss=0.1844, att_loss=0.2922, loss=0.2706, over 17158.00 frames. utt_duration=694.8 frames, utt_pad_proportion=0.1249, over 99.00 utterances.], tot_loss[ctc_loss=0.1544, att_loss=0.277, loss=0.2525, over 1304123.60 frames. utt_duration=1181 frames, utt_pad_proportion=0.06673, over 4424.09 utterances.], batch size: 99, lr: 1.86e-02, grad_scale: 8.0 2023-03-07 19:06:22,685 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:06:47,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 19:06:49,316 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8636, 4.9760, 4.9952, 2.5356, 4.6246, 4.1949, 4.1202, 2.2633], device='cuda:3'), covar=tensor([0.0199, 0.0076, 0.0136, 0.1307, 0.0101, 0.0189, 0.0319, 0.2100], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0064, 0.0053, 0.0099, 0.0060, 0.0071, 0.0080, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 19:07:09,802 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:07:41,947 INFO [train2.py:809] (3/4) Epoch 6, batch 150, loss[ctc_loss=0.1209, att_loss=0.2331, loss=0.2106, over 15620.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.009896, over 37.00 utterances.], tot_loss[ctc_loss=0.1521, att_loss=0.2754, loss=0.2507, over 1739818.24 frames. utt_duration=1234 frames, utt_pad_proportion=0.05611, over 5647.77 utterances.], batch size: 37, lr: 1.86e-02, grad_scale: 8.0 2023-03-07 19:08:00,477 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:08:12,700 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:08:27,282 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:08:45,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 3.260e+02 3.980e+02 4.904e+02 1.174e+03, threshold=7.960e+02, percent-clipped=3.0 2023-03-07 19:08:49,518 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:09:02,829 INFO [train2.py:809] (3/4) Epoch 6, batch 200, loss[ctc_loss=0.1387, att_loss=0.2741, loss=0.247, over 16634.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004901, over 47.00 utterances.], tot_loss[ctc_loss=0.1519, att_loss=0.2757, loss=0.2509, over 2087527.98 frames. utt_duration=1232 frames, utt_pad_proportion=0.0529, over 6786.82 utterances.], batch size: 47, lr: 1.86e-02, grad_scale: 8.0 2023-03-07 19:09:51,024 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:10:23,260 INFO [train2.py:809] (3/4) Epoch 6, batch 250, loss[ctc_loss=0.2206, att_loss=0.3125, loss=0.2941, over 16643.00 frames. utt_duration=1418 frames, utt_pad_proportion=0.004258, over 47.00 utterances.], tot_loss[ctc_loss=0.1518, att_loss=0.276, loss=0.2512, over 2351165.64 frames. utt_duration=1209 frames, utt_pad_proportion=0.05807, over 7786.04 utterances.], batch size: 47, lr: 1.86e-02, grad_scale: 8.0 2023-03-07 19:10:28,320 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:11:17,505 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:11:27,955 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.978e+02 3.813e+02 4.681e+02 1.334e+03, threshold=7.626e+02, percent-clipped=7.0 2023-03-07 19:11:43,059 INFO [train2.py:809] (3/4) Epoch 6, batch 300, loss[ctc_loss=0.1506, att_loss=0.2898, loss=0.2619, over 16463.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006463, over 46.00 utterances.], tot_loss[ctc_loss=0.1513, att_loss=0.2751, loss=0.2503, over 2544043.62 frames. utt_duration=1217 frames, utt_pad_proportion=0.06094, over 8373.40 utterances.], batch size: 46, lr: 1.86e-02, grad_scale: 4.0 2023-03-07 19:12:54,599 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:13:01,865 INFO [train2.py:809] (3/4) Epoch 6, batch 350, loss[ctc_loss=0.1407, att_loss=0.2808, loss=0.2528, over 17111.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.0138, over 56.00 utterances.], tot_loss[ctc_loss=0.1508, att_loss=0.2751, loss=0.2503, over 2709808.94 frames. utt_duration=1248 frames, utt_pad_proportion=0.05109, over 8694.02 utterances.], batch size: 56, lr: 1.85e-02, grad_scale: 4.0 2023-03-07 19:13:08,374 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:13:18,608 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5799, 4.7855, 4.7374, 4.9932, 2.3118, 4.9274, 2.7127, 2.4556], device='cuda:3'), covar=tensor([0.0146, 0.0143, 0.0697, 0.0167, 0.2774, 0.0154, 0.1789, 0.1889], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0100, 0.0256, 0.0113, 0.0231, 0.0106, 0.0230, 0.0209], device='cuda:3'), out_proj_covar=tensor([1.0603e-04, 9.9616e-05, 2.2611e-04, 1.0250e-04, 2.0675e-04, 9.9970e-05, 2.0125e-04, 1.8498e-04], device='cuda:3') 2023-03-07 19:14:07,909 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.145e+02 3.150e+02 3.613e+02 4.726e+02 1.759e+03, threshold=7.227e+02, percent-clipped=5.0 2023-03-07 19:14:22,592 INFO [train2.py:809] (3/4) Epoch 6, batch 400, loss[ctc_loss=0.2127, att_loss=0.314, loss=0.2937, over 17315.00 frames. utt_duration=878.3 frames, utt_pad_proportion=0.08034, over 79.00 utterances.], tot_loss[ctc_loss=0.151, att_loss=0.275, loss=0.2502, over 2822215.23 frames. utt_duration=1218 frames, utt_pad_proportion=0.06195, over 9276.29 utterances.], batch size: 79, lr: 1.85e-02, grad_scale: 8.0 2023-03-07 19:14:26,379 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:15:10,554 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-03-07 19:15:42,857 INFO [train2.py:809] (3/4) Epoch 6, batch 450, loss[ctc_loss=0.1649, att_loss=0.289, loss=0.2642, over 16320.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006779, over 45.00 utterances.], tot_loss[ctc_loss=0.1521, att_loss=0.2751, loss=0.2505, over 2917559.73 frames. utt_duration=1206 frames, utt_pad_proportion=0.06727, over 9689.83 utterances.], batch size: 45, lr: 1.85e-02, grad_scale: 8.0 2023-03-07 19:15:51,796 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1984, 5.2024, 4.9902, 2.9919, 2.1198, 2.8167, 4.8836, 3.8402], device='cuda:3'), covar=tensor([0.0560, 0.0190, 0.0312, 0.2649, 0.6036, 0.2653, 0.0301, 0.2036], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0183, 0.0206, 0.0182, 0.0362, 0.0333, 0.0187, 0.0324], device='cuda:3'), out_proj_covar=tensor([1.4330e-04, 7.7393e-05, 9.1990e-05, 8.3547e-05, 1.7563e-04, 1.5038e-04, 8.0581e-05, 1.5569e-04], device='cuda:3') 2023-03-07 19:15:53,073 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:16:12,230 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:16:48,764 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 3.385e+02 4.079e+02 4.609e+02 1.123e+03, threshold=8.157e+02, percent-clipped=3.0 2023-03-07 19:17:00,110 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:17:02,922 INFO [train2.py:809] (3/4) Epoch 6, batch 500, loss[ctc_loss=0.1176, att_loss=0.2465, loss=0.2207, over 16293.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006351, over 43.00 utterances.], tot_loss[ctc_loss=0.1505, att_loss=0.2744, loss=0.2496, over 3002017.50 frames. utt_duration=1227 frames, utt_pad_proportion=0.06027, over 9798.47 utterances.], batch size: 43, lr: 1.85e-02, grad_scale: 8.0 2023-03-07 19:17:42,563 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:17:48,969 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:18:19,467 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:18:22,352 INFO [train2.py:809] (3/4) Epoch 6, batch 550, loss[ctc_loss=0.1242, att_loss=0.2467, loss=0.2222, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008386, over 39.00 utterances.], tot_loss[ctc_loss=0.1498, att_loss=0.2734, loss=0.2487, over 3047395.97 frames. utt_duration=1210 frames, utt_pad_proportion=0.06849, over 10084.66 utterances.], batch size: 39, lr: 1.84e-02, grad_scale: 8.0 2023-03-07 19:18:37,148 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:19:28,296 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.263e+02 3.939e+02 5.001e+02 1.057e+03, threshold=7.878e+02, percent-clipped=3.0 2023-03-07 19:19:31,667 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9290, 5.2986, 4.6715, 5.3663, 4.7055, 4.9830, 5.5049, 5.2068], device='cuda:3'), covar=tensor([0.0359, 0.0188, 0.0683, 0.0126, 0.0388, 0.0185, 0.0135, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0186, 0.0242, 0.0155, 0.0202, 0.0150, 0.0176, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 19:19:43,049 INFO [train2.py:809] (3/4) Epoch 6, batch 600, loss[ctc_loss=0.1515, att_loss=0.2741, loss=0.2496, over 15945.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007337, over 41.00 utterances.], tot_loss[ctc_loss=0.1501, att_loss=0.2734, loss=0.2487, over 3090990.28 frames. utt_duration=1190 frames, utt_pad_proportion=0.07335, over 10400.64 utterances.], batch size: 41, lr: 1.84e-02, grad_scale: 8.0 2023-03-07 19:20:16,276 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9120, 5.1828, 5.4050, 5.4861, 5.2732, 5.9345, 5.0629, 5.9417], device='cuda:3'), covar=tensor([0.0636, 0.0638, 0.0543, 0.0749, 0.1743, 0.0570, 0.0549, 0.0554], device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0342, 0.0364, 0.0425, 0.0587, 0.0361, 0.0300, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 19:20:47,283 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:21:01,021 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-07 19:21:03,189 INFO [train2.py:809] (3/4) Epoch 6, batch 650, loss[ctc_loss=0.1712, att_loss=0.307, loss=0.2799, over 17053.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009775, over 53.00 utterances.], tot_loss[ctc_loss=0.1487, att_loss=0.2726, loss=0.2478, over 3137353.50 frames. utt_duration=1221 frames, utt_pad_proportion=0.06319, over 10291.26 utterances.], batch size: 53, lr: 1.84e-02, grad_scale: 8.0 2023-03-07 19:21:32,208 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:22:08,669 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.894e+02 3.503e+02 4.491e+02 1.231e+03, threshold=7.006e+02, percent-clipped=2.0 2023-03-07 19:22:09,165 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7699, 1.7341, 1.9911, 1.3473, 3.1647, 2.0424, 1.5921, 2.2099], device='cuda:3'), covar=tensor([0.0214, 0.2777, 0.2072, 0.1683, 0.0439, 0.1472, 0.2376, 0.1129], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0098, 0.0092, 0.0087, 0.0079, 0.0079, 0.0101, 0.0078], device='cuda:3'), out_proj_covar=tensor([4.0376e-05, 5.3352e-05, 5.1309e-05, 4.4019e-05, 3.8392e-05, 4.5751e-05, 5.2914e-05, 4.4391e-05], device='cuda:3') 2023-03-07 19:22:23,251 INFO [train2.py:809] (3/4) Epoch 6, batch 700, loss[ctc_loss=0.1302, att_loss=0.258, loss=0.2325, over 16178.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.007129, over 41.00 utterances.], tot_loss[ctc_loss=0.1503, att_loss=0.2737, loss=0.249, over 3164007.18 frames. utt_duration=1198 frames, utt_pad_proportion=0.07044, over 10576.43 utterances.], batch size: 41, lr: 1.84e-02, grad_scale: 8.0 2023-03-07 19:23:09,548 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:23:42,152 INFO [train2.py:809] (3/4) Epoch 6, batch 750, loss[ctc_loss=0.1409, att_loss=0.2503, loss=0.2284, over 15757.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.008143, over 38.00 utterances.], tot_loss[ctc_loss=0.1514, att_loss=0.2747, loss=0.25, over 3193209.51 frames. utt_duration=1219 frames, utt_pad_proportion=0.06373, over 10492.36 utterances.], batch size: 38, lr: 1.84e-02, grad_scale: 8.0 2023-03-07 19:23:51,701 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:24:47,042 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.294e+02 3.454e+02 4.265e+02 5.138e+02 1.250e+03, threshold=8.530e+02, percent-clipped=7.0 2023-03-07 19:25:01,489 INFO [train2.py:809] (3/4) Epoch 6, batch 800, loss[ctc_loss=0.1516, att_loss=0.259, loss=0.2375, over 15469.00 frames. utt_duration=1721 frames, utt_pad_proportion=0.01005, over 36.00 utterances.], tot_loss[ctc_loss=0.15, att_loss=0.2738, loss=0.249, over 3212686.78 frames. utt_duration=1225 frames, utt_pad_proportion=0.06033, over 10501.46 utterances.], batch size: 36, lr: 1.83e-02, grad_scale: 8.0 2023-03-07 19:25:07,754 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:25:28,442 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4751, 5.0003, 4.5877, 4.8432, 4.9263, 4.7237, 3.6103, 4.8710], device='cuda:3'), covar=tensor([0.0099, 0.0123, 0.0092, 0.0087, 0.0076, 0.0100, 0.0530, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0055, 0.0061, 0.0043, 0.0042, 0.0053, 0.0076, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 19:25:39,422 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:25:41,081 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:25:54,746 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1130, 4.8116, 4.5376, 4.6899, 5.1522, 5.0599, 4.5948, 2.1207], device='cuda:3'), covar=tensor([0.0198, 0.0250, 0.0220, 0.0203, 0.1052, 0.0208, 0.0259, 0.3078], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0123, 0.0118, 0.0125, 0.0288, 0.0130, 0.0115, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 19:26:18,417 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:26:21,152 INFO [train2.py:809] (3/4) Epoch 6, batch 850, loss[ctc_loss=0.16, att_loss=0.2913, loss=0.265, over 17118.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01517, over 56.00 utterances.], tot_loss[ctc_loss=0.1508, att_loss=0.2747, loss=0.2499, over 3229007.24 frames. utt_duration=1226 frames, utt_pad_proportion=0.05886, over 10545.35 utterances.], batch size: 56, lr: 1.83e-02, grad_scale: 8.0 2023-03-07 19:26:24,704 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7456, 4.0532, 3.2313, 3.7610, 4.0932, 3.8372, 3.0336, 4.7629], device='cuda:3'), covar=tensor([0.1151, 0.0377, 0.1092, 0.0573, 0.0471, 0.0556, 0.0839, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0137, 0.0183, 0.0151, 0.0167, 0.0176, 0.0155, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 19:26:27,701 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:26:57,842 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:27:26,710 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 3.098e+02 3.888e+02 4.985e+02 1.005e+03, threshold=7.777e+02, percent-clipped=2.0 2023-03-07 19:27:35,417 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:27:41,728 INFO [train2.py:809] (3/4) Epoch 6, batch 900, loss[ctc_loss=0.1859, att_loss=0.3006, loss=0.2776, over 17372.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03472, over 63.00 utterances.], tot_loss[ctc_loss=0.1517, att_loss=0.2757, loss=0.2509, over 3235748.34 frames. utt_duration=1190 frames, utt_pad_proportion=0.06875, over 10887.72 utterances.], batch size: 63, lr: 1.83e-02, grad_scale: 8.0 2023-03-07 19:28:45,618 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:29:01,656 INFO [train2.py:809] (3/4) Epoch 6, batch 950, loss[ctc_loss=0.1926, att_loss=0.3125, loss=0.2885, over 17423.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01299, over 57.00 utterances.], tot_loss[ctc_loss=0.15, att_loss=0.2743, loss=0.2495, over 3239930.64 frames. utt_duration=1216 frames, utt_pad_proportion=0.06469, over 10668.59 utterances.], batch size: 57, lr: 1.83e-02, grad_scale: 8.0 2023-03-07 19:29:18,749 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4108, 5.2665, 5.2263, 3.3487, 5.2422, 4.7200, 4.8704, 3.6465], device='cuda:3'), covar=tensor([0.0079, 0.0063, 0.0156, 0.0843, 0.0049, 0.0131, 0.0182, 0.0889], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0065, 0.0053, 0.0098, 0.0058, 0.0072, 0.0081, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 19:30:01,786 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:30:06,282 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 3.096e+02 4.099e+02 5.928e+02 9.906e+02, threshold=8.198e+02, percent-clipped=5.0 2023-03-07 19:30:21,097 INFO [train2.py:809] (3/4) Epoch 6, batch 1000, loss[ctc_loss=0.1586, att_loss=0.2879, loss=0.262, over 17295.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02489, over 59.00 utterances.], tot_loss[ctc_loss=0.1501, att_loss=0.2745, loss=0.2496, over 3242953.69 frames. utt_duration=1229 frames, utt_pad_proportion=0.06201, over 10566.98 utterances.], batch size: 59, lr: 1.83e-02, grad_scale: 8.0 2023-03-07 19:30:53,751 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-07 19:30:59,406 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:31:40,996 INFO [train2.py:809] (3/4) Epoch 6, batch 1050, loss[ctc_loss=0.1346, att_loss=0.2562, loss=0.2319, over 15382.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01054, over 35.00 utterances.], tot_loss[ctc_loss=0.1491, att_loss=0.2741, loss=0.2491, over 3247408.48 frames. utt_duration=1231 frames, utt_pad_proportion=0.06272, over 10562.41 utterances.], batch size: 35, lr: 1.82e-02, grad_scale: 8.0 2023-03-07 19:31:48,110 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-07 19:32:10,320 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5484, 2.0185, 4.8686, 3.9397, 2.9156, 4.4499, 4.7749, 4.8068], device='cuda:3'), covar=tensor([0.0197, 0.2342, 0.0172, 0.1107, 0.2306, 0.0224, 0.0106, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0240, 0.0120, 0.0292, 0.0293, 0.0180, 0.0104, 0.0132], device='cuda:3'), out_proj_covar=tensor([1.1488e-04, 1.9455e-04, 1.0406e-04, 2.3566e-04, 2.4405e-04, 1.5414e-04, 9.4098e-05, 1.1702e-04], device='cuda:3') 2023-03-07 19:32:47,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 3.142e+02 3.611e+02 4.263e+02 9.074e+02, threshold=7.223e+02, percent-clipped=1.0 2023-03-07 19:33:02,164 INFO [train2.py:809] (3/4) Epoch 6, batch 1100, loss[ctc_loss=0.1362, att_loss=0.2669, loss=0.2408, over 16546.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005972, over 45.00 utterances.], tot_loss[ctc_loss=0.1472, att_loss=0.273, loss=0.2478, over 3253457.87 frames. utt_duration=1243 frames, utt_pad_proportion=0.05855, over 10479.82 utterances.], batch size: 45, lr: 1.82e-02, grad_scale: 8.0 2023-03-07 19:33:34,223 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7009, 5.2200, 4.5895, 5.0763, 4.5863, 4.8811, 5.3866, 5.0747], device='cuda:3'), covar=tensor([0.0439, 0.0198, 0.0788, 0.0245, 0.0445, 0.0208, 0.0180, 0.0147], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0184, 0.0237, 0.0161, 0.0198, 0.0154, 0.0177, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 19:33:40,406 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:33:52,761 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:34:22,325 INFO [train2.py:809] (3/4) Epoch 6, batch 1150, loss[ctc_loss=0.1746, att_loss=0.2928, loss=0.2691, over 17387.00 frames. utt_duration=881.7 frames, utt_pad_proportion=0.07385, over 79.00 utterances.], tot_loss[ctc_loss=0.1478, att_loss=0.2729, loss=0.2479, over 3261984.27 frames. utt_duration=1260 frames, utt_pad_proportion=0.05199, over 10366.83 utterances.], batch size: 79, lr: 1.82e-02, grad_scale: 8.0 2023-03-07 19:34:28,580 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:34:57,182 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:35:27,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.084e+02 3.236e+02 4.040e+02 5.073e+02 1.159e+03, threshold=8.079e+02, percent-clipped=5.0 2023-03-07 19:35:30,959 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:35:39,359 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:35:41,984 INFO [train2.py:809] (3/4) Epoch 6, batch 1200, loss[ctc_loss=0.1834, att_loss=0.2963, loss=0.2737, over 16484.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005544, over 46.00 utterances.], tot_loss[ctc_loss=0.1483, att_loss=0.2732, loss=0.2482, over 3266402.59 frames. utt_duration=1255 frames, utt_pad_proportion=0.05167, over 10419.26 utterances.], batch size: 46, lr: 1.82e-02, grad_scale: 8.0 2023-03-07 19:35:45,287 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:36:11,176 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.5650, 2.9816, 3.1101, 2.1190, 3.2378, 3.1363, 2.7929, 1.8645], device='cuda:3'), covar=tensor([0.1839, 0.0978, 0.3275, 0.7767, 0.1369, 0.3516, 0.0828, 0.9079], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0071, 0.0072, 0.0104, 0.0068, 0.0098, 0.0064, 0.0115], device='cuda:3'), out_proj_covar=tensor([5.4791e-05, 5.0228e-05, 5.3997e-05, 7.6920e-05, 5.1721e-05, 7.4845e-05, 4.8730e-05, 8.7472e-05], device='cuda:3') 2023-03-07 19:36:18,611 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7820, 5.1737, 5.1019, 5.0175, 5.2413, 5.1693, 4.9286, 4.6575], device='cuda:3'), covar=tensor([0.1101, 0.0453, 0.0212, 0.0448, 0.0292, 0.0290, 0.0298, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0227, 0.0166, 0.0206, 0.0260, 0.0291, 0.0222, 0.0250], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 19:37:01,554 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-03-07 19:37:02,264 INFO [train2.py:809] (3/4) Epoch 6, batch 1250, loss[ctc_loss=0.09918, att_loss=0.2366, loss=0.2091, over 14512.00 frames. utt_duration=1816 frames, utt_pad_proportion=0.03776, over 32.00 utterances.], tot_loss[ctc_loss=0.147, att_loss=0.2721, loss=0.2471, over 3260747.98 frames. utt_duration=1252 frames, utt_pad_proportion=0.05583, over 10431.77 utterances.], batch size: 32, lr: 1.82e-02, grad_scale: 8.0 2023-03-07 19:37:13,803 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2023-03-07 19:37:17,135 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:37:53,808 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 19:38:08,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.094e+02 3.230e+02 3.853e+02 4.858e+02 1.560e+03, threshold=7.707e+02, percent-clipped=4.0 2023-03-07 19:38:22,121 INFO [train2.py:809] (3/4) Epoch 6, batch 1300, loss[ctc_loss=0.1261, att_loss=0.259, loss=0.2325, over 16285.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007249, over 43.00 utterances.], tot_loss[ctc_loss=0.1476, att_loss=0.2721, loss=0.2472, over 3254951.78 frames. utt_duration=1211 frames, utt_pad_proportion=0.06762, over 10765.42 utterances.], batch size: 43, lr: 1.81e-02, grad_scale: 8.0 2023-03-07 19:38:56,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8158, 5.2774, 4.4124, 5.4609, 4.6775, 5.0700, 5.3625, 5.1504], device='cuda:3'), covar=tensor([0.0421, 0.0297, 0.1014, 0.0165, 0.0457, 0.0182, 0.0258, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0186, 0.0241, 0.0160, 0.0198, 0.0154, 0.0176, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 19:38:59,932 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:39:21,630 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2023-03-07 19:39:29,511 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-07 19:39:42,382 INFO [train2.py:809] (3/4) Epoch 6, batch 1350, loss[ctc_loss=0.1634, att_loss=0.2913, loss=0.2657, over 16930.00 frames. utt_duration=685.4 frames, utt_pad_proportion=0.139, over 99.00 utterances.], tot_loss[ctc_loss=0.1478, att_loss=0.2727, loss=0.2477, over 3265539.31 frames. utt_duration=1237 frames, utt_pad_proportion=0.05843, over 10575.09 utterances.], batch size: 99, lr: 1.81e-02, grad_scale: 8.0 2023-03-07 19:40:17,788 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:40:48,695 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.207e+02 3.220e+02 3.890e+02 4.663e+02 1.481e+03, threshold=7.781e+02, percent-clipped=4.0 2023-03-07 19:40:50,647 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:41:02,395 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-07 19:41:02,795 INFO [train2.py:809] (3/4) Epoch 6, batch 1400, loss[ctc_loss=0.1656, att_loss=0.2953, loss=0.2694, over 17138.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01406, over 56.00 utterances.], tot_loss[ctc_loss=0.1491, att_loss=0.2741, loss=0.2491, over 3264791.62 frames. utt_duration=1208 frames, utt_pad_proportion=0.06721, over 10823.58 utterances.], batch size: 56, lr: 1.81e-02, grad_scale: 8.0 2023-03-07 19:41:16,035 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:42:23,225 INFO [train2.py:809] (3/4) Epoch 6, batch 1450, loss[ctc_loss=0.1324, att_loss=0.2541, loss=0.2298, over 15904.00 frames. utt_duration=1633 frames, utt_pad_proportion=0.008023, over 39.00 utterances.], tot_loss[ctc_loss=0.1492, att_loss=0.2745, loss=0.2494, over 3268535.90 frames. utt_duration=1215 frames, utt_pad_proportion=0.06476, over 10775.67 utterances.], batch size: 39, lr: 1.81e-02, grad_scale: 8.0 2023-03-07 19:42:28,312 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:42:32,969 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 19:42:54,069 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:43:16,641 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.25 vs. limit=5.0 2023-03-07 19:43:24,932 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:43:29,215 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.143e+02 3.096e+02 3.926e+02 4.783e+02 1.162e+03, threshold=7.851e+02, percent-clipped=3.0 2023-03-07 19:43:43,054 INFO [train2.py:809] (3/4) Epoch 6, batch 1500, loss[ctc_loss=0.1381, att_loss=0.2721, loss=0.2453, over 16535.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006467, over 45.00 utterances.], tot_loss[ctc_loss=0.1494, att_loss=0.2749, loss=0.2498, over 3274286.66 frames. utt_duration=1215 frames, utt_pad_proportion=0.06205, over 10796.27 utterances.], batch size: 45, lr: 1.81e-02, grad_scale: 8.0 2023-03-07 19:44:19,163 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7405, 3.7539, 4.0117, 4.0494, 3.9974, 4.0167, 3.7938, 3.7283], device='cuda:3'), covar=tensor([0.0961, 0.0898, 0.0214, 0.0348, 0.0330, 0.0362, 0.0303, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0228, 0.0171, 0.0209, 0.0268, 0.0300, 0.0225, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 19:45:01,846 INFO [train2.py:809] (3/4) Epoch 6, batch 1550, loss[ctc_loss=0.1604, att_loss=0.2984, loss=0.2708, over 17351.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03566, over 63.00 utterances.], tot_loss[ctc_loss=0.1485, att_loss=0.2739, loss=0.2488, over 3271697.46 frames. utt_duration=1208 frames, utt_pad_proportion=0.06479, over 10846.29 utterances.], batch size: 63, lr: 1.80e-02, grad_scale: 8.0 2023-03-07 19:45:08,984 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:46:07,491 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.008e+02 3.352e+02 3.791e+02 4.457e+02 9.611e+02, threshold=7.583e+02, percent-clipped=1.0 2023-03-07 19:46:08,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-03-07 19:46:21,678 INFO [train2.py:809] (3/4) Epoch 6, batch 1600, loss[ctc_loss=0.1369, att_loss=0.278, loss=0.2497, over 17221.00 frames. utt_duration=873.5 frames, utt_pad_proportion=0.08435, over 79.00 utterances.], tot_loss[ctc_loss=0.147, att_loss=0.2737, loss=0.2484, over 3280335.27 frames. utt_duration=1228 frames, utt_pad_proportion=0.05679, over 10694.28 utterances.], batch size: 79, lr: 1.80e-02, grad_scale: 8.0 2023-03-07 19:47:41,028 INFO [train2.py:809] (3/4) Epoch 6, batch 1650, loss[ctc_loss=0.1514, att_loss=0.2825, loss=0.2563, over 16950.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008283, over 50.00 utterances.], tot_loss[ctc_loss=0.1467, att_loss=0.2731, loss=0.2478, over 3275442.91 frames. utt_duration=1232 frames, utt_pad_proportion=0.05719, over 10650.57 utterances.], batch size: 50, lr: 1.80e-02, grad_scale: 8.0 2023-03-07 19:48:12,576 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3979, 2.2941, 2.5656, 4.3516, 4.0960, 4.1694, 2.8698, 1.7182], device='cuda:3'), covar=tensor([0.0587, 0.2415, 0.1750, 0.0416, 0.0441, 0.0229, 0.1468, 0.2550], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0201, 0.0192, 0.0157, 0.0149, 0.0125, 0.0196, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 19:48:20,307 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:48:47,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 3.087e+02 4.072e+02 5.275e+02 1.721e+03, threshold=8.145e+02, percent-clipped=7.0 2023-03-07 19:48:50,763 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:49:01,651 INFO [train2.py:809] (3/4) Epoch 6, batch 1700, loss[ctc_loss=0.1238, att_loss=0.2704, loss=0.2411, over 16882.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006509, over 49.00 utterances.], tot_loss[ctc_loss=0.1456, att_loss=0.2722, loss=0.2469, over 3275637.94 frames. utt_duration=1245 frames, utt_pad_proportion=0.05472, over 10538.63 utterances.], batch size: 49, lr: 1.80e-02, grad_scale: 8.0 2023-03-07 19:49:09,646 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-07 19:49:55,163 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2205, 4.5635, 4.5934, 4.6325, 2.2945, 4.7080, 2.7035, 1.9530], device='cuda:3'), covar=tensor([0.0232, 0.0196, 0.0745, 0.0188, 0.2420, 0.0160, 0.1739, 0.1978], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0098, 0.0249, 0.0108, 0.0224, 0.0102, 0.0222, 0.0207], device='cuda:3'), out_proj_covar=tensor([1.0710e-04, 9.8340e-05, 2.2046e-04, 9.8530e-05, 2.0307e-04, 9.7003e-05, 1.9543e-04, 1.8401e-04], device='cuda:3') 2023-03-07 19:50:00,520 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:50:19,181 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:50:22,110 INFO [train2.py:809] (3/4) Epoch 6, batch 1750, loss[ctc_loss=0.2029, att_loss=0.3135, loss=0.2914, over 14310.00 frames. utt_duration=393.7 frames, utt_pad_proportion=0.3141, over 146.00 utterances.], tot_loss[ctc_loss=0.1468, att_loss=0.2733, loss=0.248, over 3274193.85 frames. utt_duration=1204 frames, utt_pad_proportion=0.06546, over 10887.76 utterances.], batch size: 146, lr: 1.80e-02, grad_scale: 8.0 2023-03-07 19:50:29,378 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:50:44,694 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:51:23,930 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:51:28,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 3.408e+02 4.076e+02 5.243e+02 8.515e+02, threshold=8.152e+02, percent-clipped=2.0 2023-03-07 19:51:42,603 INFO [train2.py:809] (3/4) Epoch 6, batch 1800, loss[ctc_loss=0.172, att_loss=0.2837, loss=0.2613, over 16130.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.006064, over 42.00 utterances.], tot_loss[ctc_loss=0.147, att_loss=0.273, loss=0.2478, over 3252419.70 frames. utt_duration=1198 frames, utt_pad_proportion=0.07137, over 10870.20 utterances.], batch size: 42, lr: 1.79e-02, grad_scale: 8.0 2023-03-07 19:52:39,986 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:52:43,440 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.8145, 3.1105, 3.4545, 2.2166, 3.1854, 3.1240, 2.9348, 2.0526], device='cuda:3'), covar=tensor([0.1901, 0.0841, 0.1279, 0.5651, 0.1385, 0.4556, 0.0907, 0.8576], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0074, 0.0079, 0.0113, 0.0069, 0.0107, 0.0069, 0.0125], device='cuda:3'), out_proj_covar=tensor([5.7840e-05, 5.3511e-05, 5.9691e-05, 8.3626e-05, 5.4863e-05, 8.1610e-05, 5.2867e-05, 9.5102e-05], device='cuda:3') 2023-03-07 19:52:51,284 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:53:02,475 INFO [train2.py:809] (3/4) Epoch 6, batch 1850, loss[ctc_loss=0.1036, att_loss=0.2423, loss=0.2146, over 15975.00 frames. utt_duration=1599 frames, utt_pad_proportion=0.00871, over 40.00 utterances.], tot_loss[ctc_loss=0.1467, att_loss=0.2733, loss=0.248, over 3258365.36 frames. utt_duration=1200 frames, utt_pad_proportion=0.07017, over 10875.17 utterances.], batch size: 40, lr: 1.79e-02, grad_scale: 8.0 2023-03-07 19:53:09,227 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:53:19,162 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.67 vs. limit=5.0 2023-03-07 19:53:26,390 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2040, 5.0489, 5.2125, 3.0767, 4.8603, 4.2102, 4.2564, 2.5277], device='cuda:3'), covar=tensor([0.0093, 0.0072, 0.0117, 0.0942, 0.0080, 0.0204, 0.0270, 0.1460], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0065, 0.0053, 0.0098, 0.0058, 0.0075, 0.0080, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 19:54:09,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.296e+02 3.946e+02 5.007e+02 1.050e+03, threshold=7.891e+02, percent-clipped=3.0 2023-03-07 19:54:24,820 INFO [train2.py:809] (3/4) Epoch 6, batch 1900, loss[ctc_loss=0.1625, att_loss=0.2871, loss=0.2622, over 16403.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007495, over 44.00 utterances.], tot_loss[ctc_loss=0.1453, att_loss=0.2725, loss=0.247, over 3262889.45 frames. utt_duration=1227 frames, utt_pad_proportion=0.06336, over 10650.36 utterances.], batch size: 44, lr: 1.79e-02, grad_scale: 8.0 2023-03-07 19:54:28,011 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:54:30,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2023-03-07 19:54:31,357 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:55:14,667 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:55:41,335 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2023-03-07 19:55:47,728 INFO [train2.py:809] (3/4) Epoch 6, batch 1950, loss[ctc_loss=0.1191, att_loss=0.2475, loss=0.2219, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007239, over 43.00 utterances.], tot_loss[ctc_loss=0.1448, att_loss=0.2722, loss=0.2467, over 3267096.07 frames. utt_duration=1253 frames, utt_pad_proportion=0.055, over 10442.23 utterances.], batch size: 43, lr: 1.79e-02, grad_scale: 8.0 2023-03-07 19:56:30,689 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 19:56:54,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 3.406e+02 4.124e+02 5.166e+02 8.214e+02, threshold=8.249e+02, percent-clipped=4.0 2023-03-07 19:56:54,813 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:57:09,082 INFO [train2.py:809] (3/4) Epoch 6, batch 2000, loss[ctc_loss=0.1549, att_loss=0.2629, loss=0.2413, over 16023.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006858, over 40.00 utterances.], tot_loss[ctc_loss=0.145, att_loss=0.2727, loss=0.2472, over 3272816.91 frames. utt_duration=1244 frames, utt_pad_proportion=0.05597, over 10538.71 utterances.], batch size: 40, lr: 1.79e-02, grad_scale: 8.0 2023-03-07 19:57:59,924 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:58:13,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-07 19:58:19,825 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-03-07 19:58:27,710 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:58:29,229 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:58:30,578 INFO [train2.py:809] (3/4) Epoch 6, batch 2050, loss[ctc_loss=0.1224, att_loss=0.2578, loss=0.2307, over 16966.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.006974, over 50.00 utterances.], tot_loss[ctc_loss=0.1447, att_loss=0.2728, loss=0.2472, over 3276255.67 frames. utt_duration=1247 frames, utt_pad_proportion=0.05214, over 10522.70 utterances.], batch size: 50, lr: 1.78e-02, grad_scale: 8.0 2023-03-07 19:58:30,970 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:58:53,173 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:59:17,785 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9124, 1.3013, 1.8696, 1.6213, 2.8964, 1.7941, 1.7695, 2.4726], device='cuda:3'), covar=tensor([0.0677, 0.4633, 0.2814, 0.2418, 0.0810, 0.2453, 0.2881, 0.1605], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0101, 0.0103, 0.0088, 0.0084, 0.0083, 0.0099, 0.0076], device='cuda:3'), out_proj_covar=tensor([4.0416e-05, 5.5695e-05, 5.5450e-05, 4.5603e-05, 4.0826e-05, 4.7853e-05, 5.4017e-05, 4.4448e-05], device='cuda:3') 2023-03-07 19:59:42,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 3.148e+02 3.962e+02 4.824e+02 9.666e+02, threshold=7.923e+02, percent-clipped=1.0 2023-03-07 19:59:51,298 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 19:59:57,614 INFO [train2.py:809] (3/4) Epoch 6, batch 2100, loss[ctc_loss=0.1706, att_loss=0.2949, loss=0.2701, over 17070.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008716, over 53.00 utterances.], tot_loss[ctc_loss=0.1454, att_loss=0.2729, loss=0.2474, over 3270532.10 frames. utt_duration=1235 frames, utt_pad_proportion=0.05724, over 10601.76 utterances.], batch size: 53, lr: 1.78e-02, grad_scale: 8.0 2023-03-07 20:00:15,245 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:00:16,532 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:00:48,027 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1926, 5.1593, 5.0718, 2.4367, 1.9039, 2.5863, 4.8846, 3.7619], device='cuda:3'), covar=tensor([0.0549, 0.0166, 0.0223, 0.3179, 0.6250, 0.3012, 0.0281, 0.1939], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0179, 0.0206, 0.0180, 0.0353, 0.0333, 0.0195, 0.0325], device='cuda:3'), out_proj_covar=tensor([1.4873e-04, 7.5669e-05, 9.2676e-05, 8.1132e-05, 1.6944e-04, 1.4901e-04, 8.4356e-05, 1.5372e-04], device='cuda:3') 2023-03-07 20:01:20,721 INFO [train2.py:809] (3/4) Epoch 6, batch 2150, loss[ctc_loss=0.1628, att_loss=0.279, loss=0.2558, over 16891.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006244, over 49.00 utterances.], tot_loss[ctc_loss=0.1459, att_loss=0.2732, loss=0.2478, over 3277845.75 frames. utt_duration=1249 frames, utt_pad_proportion=0.05266, over 10510.75 utterances.], batch size: 49, lr: 1.78e-02, grad_scale: 8.0 2023-03-07 20:01:59,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-07 20:02:28,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.159e+02 3.198e+02 3.917e+02 4.536e+02 8.271e+02, threshold=7.834e+02, percent-clipped=2.0 2023-03-07 20:02:42,305 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:02:43,576 INFO [train2.py:809] (3/4) Epoch 6, batch 2200, loss[ctc_loss=0.2487, att_loss=0.3108, loss=0.2984, over 13910.00 frames. utt_duration=385.1 frames, utt_pad_proportion=0.3314, over 145.00 utterances.], tot_loss[ctc_loss=0.1448, att_loss=0.272, loss=0.2466, over 3278519.07 frames. utt_duration=1250 frames, utt_pad_proportion=0.0519, over 10502.39 utterances.], batch size: 145, lr: 1.78e-02, grad_scale: 8.0 2023-03-07 20:02:46,331 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-07 20:02:59,635 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:04:05,341 INFO [train2.py:809] (3/4) Epoch 6, batch 2250, loss[ctc_loss=0.1729, att_loss=0.3046, loss=0.2782, over 17359.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03541, over 63.00 utterances.], tot_loss[ctc_loss=0.1438, att_loss=0.2714, loss=0.2459, over 3279804.35 frames. utt_duration=1263 frames, utt_pad_proportion=0.04829, over 10396.75 utterances.], batch size: 63, lr: 1.78e-02, grad_scale: 8.0 2023-03-07 20:04:14,643 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:04:38,892 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:05:03,103 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:05:11,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.999e+02 3.625e+02 4.758e+02 1.652e+03, threshold=7.250e+02, percent-clipped=4.0 2023-03-07 20:05:17,982 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:05:25,451 INFO [train2.py:809] (3/4) Epoch 6, batch 2300, loss[ctc_loss=0.1545, att_loss=0.2671, loss=0.2445, over 16420.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006528, over 44.00 utterances.], tot_loss[ctc_loss=0.145, att_loss=0.272, loss=0.2466, over 3273038.35 frames. utt_duration=1254 frames, utt_pad_proportion=0.05191, over 10455.84 utterances.], batch size: 44, lr: 1.77e-02, grad_scale: 16.0 2023-03-07 20:05:53,155 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:05:53,184 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:06:10,772 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3939, 5.1990, 5.0783, 2.5136, 2.0037, 2.6632, 4.9576, 3.7799], device='cuda:3'), covar=tensor([0.0456, 0.0194, 0.0259, 0.3467, 0.6160, 0.2742, 0.0260, 0.2071], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0179, 0.0211, 0.0182, 0.0359, 0.0335, 0.0194, 0.0330], device='cuda:3'), out_proj_covar=tensor([1.4898e-04, 7.6551e-05, 9.5104e-05, 8.2170e-05, 1.7142e-04, 1.4955e-04, 8.3874e-05, 1.5580e-04], device='cuda:3') 2023-03-07 20:06:15,985 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:06:44,806 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:06:46,003 INFO [train2.py:809] (3/4) Epoch 6, batch 2350, loss[ctc_loss=0.1416, att_loss=0.2686, loss=0.2432, over 16556.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.00452, over 45.00 utterances.], tot_loss[ctc_loss=0.1446, att_loss=0.2718, loss=0.2464, over 3277722.68 frames. utt_duration=1266 frames, utt_pad_proportion=0.04741, over 10370.95 utterances.], batch size: 45, lr: 1.77e-02, grad_scale: 16.0 2023-03-07 20:06:56,051 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:07:21,236 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5581, 2.7134, 5.0164, 3.8875, 3.0677, 4.6175, 4.8593, 4.8112], device='cuda:3'), covar=tensor([0.0212, 0.1747, 0.0181, 0.1139, 0.2089, 0.0227, 0.0105, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0244, 0.0123, 0.0301, 0.0295, 0.0184, 0.0106, 0.0138], device='cuda:3'), out_proj_covar=tensor([1.2010e-04, 1.9905e-04, 1.0901e-04, 2.4391e-04, 2.4821e-04, 1.5937e-04, 9.5220e-05, 1.2318e-04], device='cuda:3') 2023-03-07 20:07:30,965 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:07:32,241 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:07:49,350 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4474, 2.3255, 3.1930, 4.1535, 4.0579, 4.0658, 2.8339, 2.0595], device='cuda:3'), covar=tensor([0.0630, 0.2636, 0.1268, 0.0523, 0.0488, 0.0291, 0.1613, 0.2438], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0202, 0.0196, 0.0164, 0.0146, 0.0125, 0.0193, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 20:07:52,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 3.128e+02 3.825e+02 4.605e+02 9.373e+02, threshold=7.650e+02, percent-clipped=4.0 2023-03-07 20:07:57,165 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1046, 6.2398, 5.6383, 6.1257, 5.9179, 5.5363, 5.7265, 5.4734], device='cuda:3'), covar=tensor([0.0970, 0.0817, 0.0681, 0.0619, 0.0615, 0.1248, 0.2206, 0.2022], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0397, 0.0311, 0.0327, 0.0289, 0.0372, 0.0436, 0.0396], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 20:08:01,641 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:08:06,059 INFO [train2.py:809] (3/4) Epoch 6, batch 2400, loss[ctc_loss=0.1088, att_loss=0.2364, loss=0.2109, over 14080.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.05202, over 31.00 utterances.], tot_loss[ctc_loss=0.1443, att_loss=0.2709, loss=0.2456, over 3266941.83 frames. utt_duration=1280 frames, utt_pad_proportion=0.04676, over 10220.87 utterances.], batch size: 31, lr: 1.77e-02, grad_scale: 16.0 2023-03-07 20:08:15,571 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:09:26,765 INFO [train2.py:809] (3/4) Epoch 6, batch 2450, loss[ctc_loss=0.1227, att_loss=0.247, loss=0.2221, over 15654.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008307, over 37.00 utterances.], tot_loss[ctc_loss=0.1444, att_loss=0.2714, loss=0.246, over 3276154.26 frames. utt_duration=1275 frames, utt_pad_proportion=0.04584, over 10290.45 utterances.], batch size: 37, lr: 1.77e-02, grad_scale: 16.0 2023-03-07 20:09:51,333 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:10:34,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 3.249e+02 3.944e+02 4.920e+02 7.510e+02, threshold=7.889e+02, percent-clipped=0.0 2023-03-07 20:10:47,136 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:10:48,448 INFO [train2.py:809] (3/4) Epoch 6, batch 2500, loss[ctc_loss=0.146, att_loss=0.2462, loss=0.2262, over 15362.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01181, over 35.00 utterances.], tot_loss[ctc_loss=0.1448, att_loss=0.2716, loss=0.2462, over 3270805.50 frames. utt_duration=1241 frames, utt_pad_proportion=0.05676, over 10558.49 utterances.], batch size: 35, lr: 1.77e-02, grad_scale: 16.0 2023-03-07 20:11:09,034 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5120, 2.6524, 3.6126, 2.7080, 3.6327, 4.5871, 4.3597, 3.0320], device='cuda:3'), covar=tensor([0.0349, 0.1798, 0.1028, 0.1387, 0.0910, 0.0508, 0.0444, 0.1465], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0211, 0.0213, 0.0194, 0.0217, 0.0225, 0.0174, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 20:11:15,033 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-03-07 20:11:30,436 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:11:55,113 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.6661, 2.5825, 2.8577, 2.2238, 2.7939, 2.6238, 2.8072, 1.5077], device='cuda:3'), covar=tensor([0.1406, 0.1422, 0.2205, 0.6767, 0.2344, 0.2824, 0.0721, 1.0300], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0074, 0.0079, 0.0111, 0.0070, 0.0107, 0.0066, 0.0125], device='cuda:3'), out_proj_covar=tensor([5.8012e-05, 5.4198e-05, 6.1411e-05, 8.3618e-05, 5.5884e-05, 8.2085e-05, 5.0685e-05, 9.5393e-05], device='cuda:3') 2023-03-07 20:12:04,179 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:12:08,536 INFO [train2.py:809] (3/4) Epoch 6, batch 2550, loss[ctc_loss=0.1523, att_loss=0.2981, loss=0.2689, over 17040.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.00736, over 51.00 utterances.], tot_loss[ctc_loss=0.1452, att_loss=0.2719, loss=0.2466, over 3277852.51 frames. utt_duration=1234 frames, utt_pad_proportion=0.05564, over 10634.04 utterances.], batch size: 51, lr: 1.76e-02, grad_scale: 16.0 2023-03-07 20:12:34,600 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:12:43,105 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.5537, 2.7834, 3.0520, 2.3312, 3.0181, 3.0848, 2.8168, 1.8034], device='cuda:3'), covar=tensor([0.1914, 0.1487, 0.2111, 0.6139, 0.1718, 0.3600, 0.0890, 1.0434], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0074, 0.0081, 0.0112, 0.0070, 0.0109, 0.0066, 0.0126], device='cuda:3'), out_proj_covar=tensor([5.8251e-05, 5.4516e-05, 6.2219e-05, 8.4196e-05, 5.6056e-05, 8.3216e-05, 5.1320e-05, 9.6198e-05], device='cuda:3') 2023-03-07 20:13:07,431 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:13:07,891 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-07 20:13:14,820 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 3.216e+02 3.975e+02 4.594e+02 1.058e+03, threshold=7.951e+02, percent-clipped=4.0 2023-03-07 20:13:21,108 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7930, 5.0840, 4.6099, 5.2153, 4.5806, 4.8771, 5.3486, 5.1411], device='cuda:3'), covar=tensor([0.0465, 0.0276, 0.0766, 0.0170, 0.0454, 0.0207, 0.0214, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0189, 0.0248, 0.0166, 0.0206, 0.0160, 0.0182, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 20:13:21,355 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1858, 5.0893, 5.0391, 2.5646, 2.0480, 2.7907, 4.9754, 3.8105], device='cuda:3'), covar=tensor([0.0457, 0.0151, 0.0236, 0.3249, 0.5972, 0.2591, 0.0198, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0179, 0.0212, 0.0179, 0.0358, 0.0332, 0.0194, 0.0330], device='cuda:3'), out_proj_covar=tensor([1.4567e-04, 7.5690e-05, 9.4908e-05, 8.0941e-05, 1.7051e-04, 1.4814e-04, 8.2849e-05, 1.5520e-04], device='cuda:3') 2023-03-07 20:13:28,846 INFO [train2.py:809] (3/4) Epoch 6, batch 2600, loss[ctc_loss=0.1311, att_loss=0.2559, loss=0.2309, over 16618.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005887, over 47.00 utterances.], tot_loss[ctc_loss=0.1458, att_loss=0.2723, loss=0.247, over 3265457.98 frames. utt_duration=1203 frames, utt_pad_proportion=0.06627, over 10867.09 utterances.], batch size: 47, lr: 1.76e-02, grad_scale: 16.0 2023-03-07 20:13:48,529 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:14:05,104 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2244, 5.1011, 5.1664, 3.0756, 5.0109, 4.3237, 4.4811, 2.4741], device='cuda:3'), covar=tensor([0.0084, 0.0074, 0.0129, 0.0880, 0.0069, 0.0183, 0.0241, 0.1447], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0067, 0.0055, 0.0100, 0.0060, 0.0078, 0.0082, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 20:14:12,259 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3195, 3.7793, 2.9291, 3.2848, 3.8439, 3.3520, 2.2136, 4.2647], device='cuda:3'), covar=tensor([0.1263, 0.0276, 0.1052, 0.0663, 0.0459, 0.0638, 0.1155, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0141, 0.0187, 0.0154, 0.0173, 0.0184, 0.0158, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 20:14:25,010 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:14:51,055 INFO [train2.py:809] (3/4) Epoch 6, batch 2650, loss[ctc_loss=0.1454, att_loss=0.2824, loss=0.255, over 17281.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01237, over 55.00 utterances.], tot_loss[ctc_loss=0.1449, att_loss=0.2715, loss=0.2462, over 3270324.74 frames. utt_duration=1218 frames, utt_pad_proportion=0.06146, over 10756.28 utterances.], batch size: 55, lr: 1.76e-02, grad_scale: 16.0 2023-03-07 20:14:52,732 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:15:28,348 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:15:59,449 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 3.146e+02 3.715e+02 4.789e+02 1.567e+03, threshold=7.430e+02, percent-clipped=3.0 2023-03-07 20:16:11,933 INFO [train2.py:809] (3/4) Epoch 6, batch 2700, loss[ctc_loss=0.1233, att_loss=0.2598, loss=0.2325, over 15989.00 frames. utt_duration=1600 frames, utt_pad_proportion=0.008442, over 40.00 utterances.], tot_loss[ctc_loss=0.1436, att_loss=0.271, loss=0.2455, over 3272790.38 frames. utt_duration=1250 frames, utt_pad_proportion=0.05344, over 10484.35 utterances.], batch size: 40, lr: 1.76e-02, grad_scale: 8.0 2023-03-07 20:16:21,398 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:16:42,436 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:17:26,049 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-07 20:17:32,381 INFO [train2.py:809] (3/4) Epoch 6, batch 2750, loss[ctc_loss=0.141, att_loss=0.2707, loss=0.2448, over 16974.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007012, over 50.00 utterances.], tot_loss[ctc_loss=0.1449, att_loss=0.2714, loss=0.2461, over 3274733.89 frames. utt_duration=1235 frames, utt_pad_proportion=0.05759, over 10620.05 utterances.], batch size: 50, lr: 1.76e-02, grad_scale: 8.0 2023-03-07 20:17:38,673 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:18:21,024 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 20:18:40,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 3.125e+02 3.816e+02 4.658e+02 9.758e+02, threshold=7.632e+02, percent-clipped=3.0 2023-03-07 20:18:52,607 INFO [train2.py:809] (3/4) Epoch 6, batch 2800, loss[ctc_loss=0.1387, att_loss=0.2656, loss=0.2402, over 16373.00 frames. utt_duration=1490 frames, utt_pad_proportion=0.008589, over 44.00 utterances.], tot_loss[ctc_loss=0.1443, att_loss=0.2711, loss=0.2457, over 3271508.78 frames. utt_duration=1245 frames, utt_pad_proportion=0.05515, over 10522.82 utterances.], batch size: 44, lr: 1.76e-02, grad_scale: 8.0 2023-03-07 20:19:06,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 20:19:26,389 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:19:34,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-07 20:20:11,900 INFO [train2.py:809] (3/4) Epoch 6, batch 2850, loss[ctc_loss=0.1658, att_loss=0.2731, loss=0.2516, over 16273.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007767, over 43.00 utterances.], tot_loss[ctc_loss=0.1436, att_loss=0.2705, loss=0.2451, over 3275213.35 frames. utt_duration=1272 frames, utt_pad_proportion=0.04804, over 10310.89 utterances.], batch size: 43, lr: 1.75e-02, grad_scale: 8.0 2023-03-07 20:20:23,341 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7290, 5.9270, 5.3168, 5.8470, 5.6451, 5.3440, 5.3402, 5.1700], device='cuda:3'), covar=tensor([0.1218, 0.0824, 0.0845, 0.0633, 0.0616, 0.1161, 0.2392, 0.2323], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0401, 0.0306, 0.0321, 0.0290, 0.0369, 0.0435, 0.0390], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 20:20:38,197 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:20:51,868 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:21:20,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 3.039e+02 3.742e+02 4.818e+02 9.718e+02, threshold=7.483e+02, percent-clipped=5.0 2023-03-07 20:21:33,093 INFO [train2.py:809] (3/4) Epoch 6, batch 2900, loss[ctc_loss=0.1563, att_loss=0.2905, loss=0.2636, over 17139.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01378, over 56.00 utterances.], tot_loss[ctc_loss=0.1431, att_loss=0.2709, loss=0.2454, over 3285818.96 frames. utt_duration=1282 frames, utt_pad_proportion=0.04262, over 10263.69 utterances.], batch size: 56, lr: 1.75e-02, grad_scale: 8.0 2023-03-07 20:21:41,179 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:21:49,523 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3853, 2.6278, 3.4869, 2.5277, 3.4787, 4.4951, 4.2008, 3.0655], device='cuda:3'), covar=tensor([0.0375, 0.1677, 0.1055, 0.1545, 0.0973, 0.0463, 0.0532, 0.1424], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0211, 0.0214, 0.0196, 0.0220, 0.0225, 0.0180, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 20:21:52,560 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:21:55,480 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:22:30,795 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:22:53,778 INFO [train2.py:809] (3/4) Epoch 6, batch 2950, loss[ctc_loss=0.1226, att_loss=0.2485, loss=0.2233, over 16173.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006309, over 41.00 utterances.], tot_loss[ctc_loss=0.1417, att_loss=0.2701, loss=0.2444, over 3286584.09 frames. utt_duration=1297 frames, utt_pad_proportion=0.03853, over 10146.29 utterances.], batch size: 41, lr: 1.75e-02, grad_scale: 8.0 2023-03-07 20:22:55,655 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:23:10,206 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:23:20,754 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:23:32,176 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:24:02,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 3.285e+02 3.915e+02 4.917e+02 1.190e+03, threshold=7.831e+02, percent-clipped=3.0 2023-03-07 20:24:05,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.27 vs. limit=5.0 2023-03-07 20:24:08,447 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5335, 2.6739, 3.3913, 4.2255, 3.9609, 4.2007, 2.8031, 1.9692], device='cuda:3'), covar=tensor([0.0547, 0.2138, 0.1088, 0.0543, 0.0579, 0.0211, 0.1529, 0.2504], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0198, 0.0193, 0.0165, 0.0147, 0.0122, 0.0188, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 20:24:12,924 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:24:14,354 INFO [train2.py:809] (3/4) Epoch 6, batch 3000, loss[ctc_loss=0.1645, att_loss=0.2963, loss=0.27, over 17418.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.03011, over 63.00 utterances.], tot_loss[ctc_loss=0.1422, att_loss=0.2703, loss=0.2447, over 3281104.92 frames. utt_duration=1270 frames, utt_pad_proportion=0.04712, over 10345.69 utterances.], batch size: 63, lr: 1.75e-02, grad_scale: 8.0 2023-03-07 20:24:14,354 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 20:24:28,227 INFO [train2.py:843] (3/4) Epoch 6, validation: ctc_loss=0.06806, att_loss=0.2473, loss=0.2115, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 20:24:28,228 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 20:25:01,614 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:25:37,461 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2368, 4.5208, 4.4701, 4.7666, 2.3113, 4.5211, 2.2402, 1.5509], device='cuda:3'), covar=tensor([0.0206, 0.0137, 0.0686, 0.0118, 0.2253, 0.0140, 0.1796, 0.1984], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0098, 0.0243, 0.0106, 0.0220, 0.0098, 0.0217, 0.0199], device='cuda:3'), out_proj_covar=tensor([1.0680e-04, 9.7802e-05, 2.1638e-04, 9.6297e-05, 2.0135e-04, 9.4335e-05, 1.9247e-04, 1.7844e-04], device='cuda:3') 2023-03-07 20:25:47,896 INFO [train2.py:809] (3/4) Epoch 6, batch 3050, loss[ctc_loss=0.1284, att_loss=0.2667, loss=0.239, over 16403.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007404, over 44.00 utterances.], tot_loss[ctc_loss=0.1428, att_loss=0.2706, loss=0.245, over 3280093.09 frames. utt_duration=1259 frames, utt_pad_proportion=0.0503, over 10434.28 utterances.], batch size: 44, lr: 1.75e-02, grad_scale: 8.0 2023-03-07 20:26:27,898 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 20:26:42,806 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1404, 5.0668, 5.0844, 2.5661, 2.0968, 3.0457, 4.8521, 3.9752], device='cuda:3'), covar=tensor([0.0576, 0.0174, 0.0216, 0.3222, 0.5739, 0.2270, 0.0290, 0.1649], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0186, 0.0210, 0.0176, 0.0354, 0.0336, 0.0200, 0.0329], device='cuda:3'), out_proj_covar=tensor([1.4643e-04, 7.8098e-05, 9.3578e-05, 8.0327e-05, 1.6797e-04, 1.4890e-04, 8.4784e-05, 1.5449e-04], device='cuda:3') 2023-03-07 20:26:44,128 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2091, 5.1793, 5.3016, 3.0980, 5.0037, 4.2923, 4.3186, 2.4557], device='cuda:3'), covar=tensor([0.0148, 0.0102, 0.0089, 0.1077, 0.0085, 0.0176, 0.0358, 0.1945], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0067, 0.0054, 0.0098, 0.0059, 0.0076, 0.0081, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 20:26:55,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 3.380e+02 4.169e+02 5.152e+02 1.227e+03, threshold=8.338e+02, percent-clipped=8.0 2023-03-07 20:26:55,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3573, 1.6076, 2.2175, 1.4943, 3.0683, 2.5997, 1.7909, 1.1026], device='cuda:3'), covar=tensor([0.0288, 0.2907, 0.2949, 0.1959, 0.0480, 0.1090, 0.2054, 0.2238], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0093, 0.0081, 0.0077, 0.0078, 0.0090, 0.0075], device='cuda:3'), out_proj_covar=tensor([3.8540e-05, 5.1841e-05, 5.1930e-05, 4.3091e-05, 3.8229e-05, 4.5441e-05, 5.0199e-05, 4.4461e-05], device='cuda:3') 2023-03-07 20:27:07,728 INFO [train2.py:809] (3/4) Epoch 6, batch 3100, loss[ctc_loss=0.1663, att_loss=0.2821, loss=0.2589, over 16472.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006985, over 46.00 utterances.], tot_loss[ctc_loss=0.1442, att_loss=0.2711, loss=0.2457, over 3269707.87 frames. utt_duration=1233 frames, utt_pad_proportion=0.05923, over 10619.48 utterances.], batch size: 46, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:27:42,245 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:28:28,020 INFO [train2.py:809] (3/4) Epoch 6, batch 3150, loss[ctc_loss=0.1098, att_loss=0.2296, loss=0.2056, over 15523.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007396, over 36.00 utterances.], tot_loss[ctc_loss=0.1438, att_loss=0.2706, loss=0.2452, over 3266269.63 frames. utt_duration=1247 frames, utt_pad_proportion=0.05699, over 10491.92 utterances.], batch size: 36, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:28:59,007 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:29:30,211 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:29:36,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 3.296e+02 3.877e+02 4.841e+02 1.237e+03, threshold=7.753e+02, percent-clipped=8.0 2023-03-07 20:29:48,527 INFO [train2.py:809] (3/4) Epoch 6, batch 3200, loss[ctc_loss=0.1553, att_loss=0.3002, loss=0.2712, over 17329.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02317, over 59.00 utterances.], tot_loss[ctc_loss=0.1438, att_loss=0.2712, loss=0.2457, over 3272024.65 frames. utt_duration=1246 frames, utt_pad_proportion=0.05627, over 10520.73 utterances.], batch size: 59, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:30:36,878 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:30:46,595 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.7148, 3.2566, 3.3830, 2.5679, 3.1973, 3.1889, 3.1071, 1.5629], device='cuda:3'), covar=tensor([0.2063, 0.1236, 0.2781, 0.6000, 0.7527, 0.7326, 0.0927, 1.1221], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0077, 0.0082, 0.0119, 0.0074, 0.0109, 0.0069, 0.0127], device='cuda:3'), out_proj_covar=tensor([6.0640e-05, 5.7099e-05, 6.4299e-05, 8.9131e-05, 5.8610e-05, 8.4954e-05, 5.3640e-05, 9.8158e-05], device='cuda:3') 2023-03-07 20:31:07,696 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:31:08,933 INFO [train2.py:809] (3/4) Epoch 6, batch 3250, loss[ctc_loss=0.182, att_loss=0.2959, loss=0.2731, over 17008.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008385, over 51.00 utterances.], tot_loss[ctc_loss=0.1423, att_loss=0.2701, loss=0.2445, over 3267915.46 frames. utt_duration=1281 frames, utt_pad_proportion=0.04862, over 10217.64 utterances.], batch size: 51, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:31:26,778 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:32:12,273 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.5228, 3.1221, 3.3188, 2.5804, 2.9562, 3.1662, 2.9674, 1.7022], device='cuda:3'), covar=tensor([0.1450, 0.1297, 0.2101, 0.4980, 0.2617, 0.4686, 0.0953, 0.8027], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0076, 0.0080, 0.0118, 0.0073, 0.0107, 0.0068, 0.0123], device='cuda:3'), out_proj_covar=tensor([6.0285e-05, 5.6367e-05, 6.2660e-05, 8.8404e-05, 5.8377e-05, 8.3622e-05, 5.2687e-05, 9.6151e-05], device='cuda:3') 2023-03-07 20:32:18,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.115e+02 3.273e+02 3.918e+02 4.726e+02 1.280e+03, threshold=7.837e+02, percent-clipped=1.0 2023-03-07 20:32:26,186 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:32:30,542 INFO [train2.py:809] (3/4) Epoch 6, batch 3300, loss[ctc_loss=0.1295, att_loss=0.2391, loss=0.2172, over 15609.00 frames. utt_duration=1689 frames, utt_pad_proportion=0.01123, over 37.00 utterances.], tot_loss[ctc_loss=0.142, att_loss=0.2707, loss=0.245, over 3278842.69 frames. utt_duration=1277 frames, utt_pad_proportion=0.04647, over 10278.74 utterances.], batch size: 37, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:32:45,796 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:32:58,933 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0152, 4.8131, 4.9563, 4.7648, 5.2414, 5.1913, 4.7064, 2.2545], device='cuda:3'), covar=tensor([0.0170, 0.0262, 0.0136, 0.0218, 0.1074, 0.0154, 0.0263, 0.2350], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0124, 0.0117, 0.0120, 0.0298, 0.0129, 0.0113, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 20:33:18,783 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-03-07 20:33:51,762 INFO [train2.py:809] (3/4) Epoch 6, batch 3350, loss[ctc_loss=0.1186, att_loss=0.2732, loss=0.2423, over 17128.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01467, over 56.00 utterances.], tot_loss[ctc_loss=0.1427, att_loss=0.2704, loss=0.2448, over 3269958.43 frames. utt_duration=1256 frames, utt_pad_proportion=0.05485, over 10426.46 utterances.], batch size: 56, lr: 1.74e-02, grad_scale: 8.0 2023-03-07 20:34:05,073 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:34:24,922 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:34:32,808 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:34:59,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 3.226e+02 3.955e+02 4.943e+02 9.794e+02, threshold=7.910e+02, percent-clipped=2.0 2023-03-07 20:35:11,505 INFO [train2.py:809] (3/4) Epoch 6, batch 3400, loss[ctc_loss=0.0952, att_loss=0.2277, loss=0.2012, over 15889.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008973, over 39.00 utterances.], tot_loss[ctc_loss=0.1441, att_loss=0.2708, loss=0.2454, over 3264248.17 frames. utt_duration=1237 frames, utt_pad_proportion=0.06, over 10569.35 utterances.], batch size: 39, lr: 1.73e-02, grad_scale: 8.0 2023-03-07 20:35:49,211 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:36:31,952 INFO [train2.py:809] (3/4) Epoch 6, batch 3450, loss[ctc_loss=0.1357, att_loss=0.2825, loss=0.2531, over 17060.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.008491, over 53.00 utterances.], tot_loss[ctc_loss=0.1442, att_loss=0.2711, loss=0.2457, over 3261527.35 frames. utt_duration=1225 frames, utt_pad_proportion=0.06383, over 10661.22 utterances.], batch size: 53, lr: 1.73e-02, grad_scale: 8.0 2023-03-07 20:37:39,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.288e+02 3.391e+02 4.040e+02 5.091e+02 1.147e+03, threshold=8.079e+02, percent-clipped=5.0 2023-03-07 20:37:46,486 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0778, 5.0276, 4.9901, 2.4705, 1.9002, 2.4372, 4.6629, 3.7441], device='cuda:3'), covar=tensor([0.0561, 0.0161, 0.0177, 0.3331, 0.6136, 0.3057, 0.0357, 0.1900], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0188, 0.0212, 0.0179, 0.0354, 0.0338, 0.0199, 0.0334], device='cuda:3'), out_proj_covar=tensor([1.4798e-04, 7.8933e-05, 9.4457e-05, 8.2109e-05, 1.6805e-04, 1.4950e-04, 8.3410e-05, 1.5619e-04], device='cuda:3') 2023-03-07 20:37:52,187 INFO [train2.py:809] (3/4) Epoch 6, batch 3500, loss[ctc_loss=0.1399, att_loss=0.2688, loss=0.243, over 15926.00 frames. utt_duration=1555 frames, utt_pad_proportion=0.008135, over 41.00 utterances.], tot_loss[ctc_loss=0.1453, att_loss=0.2715, loss=0.2463, over 3255463.21 frames. utt_duration=1179 frames, utt_pad_proportion=0.07797, over 11057.23 utterances.], batch size: 41, lr: 1.73e-02, grad_scale: 8.0 2023-03-07 20:38:40,730 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:39:03,011 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:39:12,998 INFO [train2.py:809] (3/4) Epoch 6, batch 3550, loss[ctc_loss=0.09828, att_loss=0.2412, loss=0.2126, over 16180.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006243, over 41.00 utterances.], tot_loss[ctc_loss=0.1447, att_loss=0.2711, loss=0.2458, over 3260932.96 frames. utt_duration=1193 frames, utt_pad_proportion=0.07363, over 10947.97 utterances.], batch size: 41, lr: 1.73e-02, grad_scale: 8.0 2023-03-07 20:39:30,686 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:39:40,596 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0248, 6.1803, 5.7033, 6.1299, 5.8926, 5.5970, 5.5970, 5.5618], device='cuda:3'), covar=tensor([0.0908, 0.0690, 0.0532, 0.0525, 0.0647, 0.0976, 0.2051, 0.1755], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0395, 0.0303, 0.0325, 0.0294, 0.0371, 0.0434, 0.0392], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 20:39:42,260 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:39:57,395 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:40:19,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 3.330e+02 3.992e+02 4.756e+02 1.054e+03, threshold=7.983e+02, percent-clipped=2.0 2023-03-07 20:40:31,492 INFO [train2.py:809] (3/4) Epoch 6, batch 3600, loss[ctc_loss=0.1643, att_loss=0.2928, loss=0.2671, over 17420.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.03309, over 63.00 utterances.], tot_loss[ctc_loss=0.1449, att_loss=0.2712, loss=0.2459, over 3257616.24 frames. utt_duration=1224 frames, utt_pad_proportion=0.06589, over 10662.38 utterances.], batch size: 63, lr: 1.73e-02, grad_scale: 8.0 2023-03-07 20:40:46,969 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:41:18,606 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 20:41:51,578 INFO [train2.py:809] (3/4) Epoch 6, batch 3650, loss[ctc_loss=0.1018, att_loss=0.2348, loss=0.2082, over 15509.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008258, over 36.00 utterances.], tot_loss[ctc_loss=0.1444, att_loss=0.2707, loss=0.2455, over 3256544.43 frames. utt_duration=1229 frames, utt_pad_proportion=0.06339, over 10614.61 utterances.], batch size: 36, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:41:57,522 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:42:11,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.98 vs. limit=5.0 2023-03-07 20:42:16,990 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:42:43,155 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8580, 6.0289, 5.3975, 5.9577, 5.6878, 5.3482, 5.4845, 5.2387], device='cuda:3'), covar=tensor([0.1090, 0.0799, 0.0749, 0.0623, 0.0697, 0.1191, 0.1929, 0.2195], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0399, 0.0304, 0.0322, 0.0289, 0.0363, 0.0427, 0.0391], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 20:42:59,617 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.210e+02 3.073e+02 3.816e+02 5.187e+02 1.095e+03, threshold=7.631e+02, percent-clipped=5.0 2023-03-07 20:43:09,943 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-07 20:43:12,933 INFO [train2.py:809] (3/4) Epoch 6, batch 3700, loss[ctc_loss=0.1445, att_loss=0.259, loss=0.2361, over 15517.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.007272, over 36.00 utterances.], tot_loss[ctc_loss=0.1457, att_loss=0.2713, loss=0.2462, over 3247921.18 frames. utt_duration=1203 frames, utt_pad_proportion=0.07141, over 10809.01 utterances.], batch size: 36, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:44:32,075 INFO [train2.py:809] (3/4) Epoch 6, batch 3750, loss[ctc_loss=0.1161, att_loss=0.2419, loss=0.2168, over 16002.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007083, over 40.00 utterances.], tot_loss[ctc_loss=0.1452, att_loss=0.2712, loss=0.246, over 3253721.87 frames. utt_duration=1216 frames, utt_pad_proportion=0.0674, over 10715.16 utterances.], batch size: 40, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:45:38,746 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 3.181e+02 3.944e+02 5.174e+02 1.956e+03, threshold=7.887e+02, percent-clipped=9.0 2023-03-07 20:45:52,112 INFO [train2.py:809] (3/4) Epoch 6, batch 3800, loss[ctc_loss=0.1465, att_loss=0.2745, loss=0.2489, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006287, over 45.00 utterances.], tot_loss[ctc_loss=0.1455, att_loss=0.2714, loss=0.2462, over 3256698.78 frames. utt_duration=1206 frames, utt_pad_proportion=0.0691, over 10815.90 utterances.], batch size: 45, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:46:13,308 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7507, 5.0497, 4.9858, 4.8681, 5.1598, 5.0813, 4.8003, 4.5593], device='cuda:3'), covar=tensor([0.0851, 0.0359, 0.0237, 0.0508, 0.0230, 0.0250, 0.0269, 0.0280], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0232, 0.0178, 0.0220, 0.0276, 0.0304, 0.0228, 0.0260], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 20:47:02,219 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:47:04,357 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 20:47:11,176 INFO [train2.py:809] (3/4) Epoch 6, batch 3850, loss[ctc_loss=0.1132, att_loss=0.2477, loss=0.2208, over 15890.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009066, over 39.00 utterances.], tot_loss[ctc_loss=0.1452, att_loss=0.2717, loss=0.2464, over 3268462.62 frames. utt_duration=1207 frames, utt_pad_proportion=0.06596, over 10844.29 utterances.], batch size: 39, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:47:18,235 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9655, 5.2848, 5.2510, 5.1311, 5.3716, 5.2906, 5.0180, 4.7630], device='cuda:3'), covar=tensor([0.0770, 0.0291, 0.0145, 0.0398, 0.0204, 0.0215, 0.0260, 0.0239], device='cuda:3'), in_proj_covar=tensor([0.0392, 0.0229, 0.0176, 0.0218, 0.0275, 0.0303, 0.0227, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 20:47:30,831 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9414, 2.0046, 1.9131, 1.0889, 3.2582, 1.9558, 1.3954, 2.7330], device='cuda:3'), covar=tensor([0.0223, 0.1773, 0.2326, 0.1894, 0.0514, 0.1461, 0.2692, 0.0815], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0086, 0.0092, 0.0079, 0.0079, 0.0077, 0.0088, 0.0069], device='cuda:3'), out_proj_covar=tensor([3.8353e-05, 5.0866e-05, 5.1706e-05, 4.2746e-05, 3.8461e-05, 4.5002e-05, 4.9817e-05, 4.1851e-05], device='cuda:3') 2023-03-07 20:48:14,079 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7673, 1.8199, 1.8103, 1.2688, 3.3418, 2.1567, 1.2981, 2.7803], device='cuda:3'), covar=tensor([0.0292, 0.2093, 0.2566, 0.1452, 0.0602, 0.0917, 0.2204, 0.0689], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0086, 0.0092, 0.0078, 0.0078, 0.0076, 0.0087, 0.0068], device='cuda:3'), out_proj_covar=tensor([3.8297e-05, 5.0697e-05, 5.1835e-05, 4.2208e-05, 3.8119e-05, 4.4264e-05, 4.9287e-05, 4.1238e-05], device='cuda:3') 2023-03-07 20:48:18,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 3.226e+02 4.048e+02 5.028e+02 1.314e+03, threshold=8.096e+02, percent-clipped=6.0 2023-03-07 20:48:18,526 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:48:31,079 INFO [train2.py:809] (3/4) Epoch 6, batch 3900, loss[ctc_loss=0.1466, att_loss=0.2883, loss=0.26, over 17093.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01669, over 56.00 utterances.], tot_loss[ctc_loss=0.1447, att_loss=0.2717, loss=0.2463, over 3264217.15 frames. utt_duration=1203 frames, utt_pad_proportion=0.06882, over 10868.91 utterances.], batch size: 56, lr: 1.72e-02, grad_scale: 8.0 2023-03-07 20:49:08,486 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 20:49:12,464 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 2023-03-07 20:49:48,967 INFO [train2.py:809] (3/4) Epoch 6, batch 3950, loss[ctc_loss=0.09978, att_loss=0.2403, loss=0.2122, over 16187.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.005768, over 41.00 utterances.], tot_loss[ctc_loss=0.1427, att_loss=0.2699, loss=0.2444, over 3264173.79 frames. utt_duration=1245 frames, utt_pad_proportion=0.05877, over 10503.99 utterances.], batch size: 41, lr: 1.71e-02, grad_scale: 8.0 2023-03-07 20:49:54,040 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:50:12,164 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:50:21,041 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7931, 2.2087, 2.4574, 3.2336, 3.1914, 3.3628, 2.5465, 2.0645], device='cuda:3'), covar=tensor([0.0614, 0.2051, 0.1222, 0.0685, 0.0539, 0.0279, 0.1428, 0.1937], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0197, 0.0195, 0.0166, 0.0144, 0.0126, 0.0189, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 20:51:06,496 INFO [train2.py:809] (3/4) Epoch 7, batch 0, loss[ctc_loss=0.1345, att_loss=0.2742, loss=0.2463, over 17293.00 frames. utt_duration=877.2 frames, utt_pad_proportion=0.0805, over 79.00 utterances.], tot_loss[ctc_loss=0.1345, att_loss=0.2742, loss=0.2463, over 17293.00 frames. utt_duration=877.2 frames, utt_pad_proportion=0.0805, over 79.00 utterances.], batch size: 79, lr: 1.61e-02, grad_scale: 8.0 2023-03-07 20:51:06,496 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 20:51:19,279 INFO [train2.py:843] (3/4) Epoch 7, validation: ctc_loss=0.06772, att_loss=0.2471, loss=0.2112, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 20:51:19,280 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 20:51:28,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9573, 4.8388, 4.8348, 4.7349, 5.1852, 4.9542, 4.6960, 2.2651], device='cuda:3'), covar=tensor([0.0185, 0.0205, 0.0160, 0.0146, 0.0952, 0.0146, 0.0241, 0.2528], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0122, 0.0118, 0.0123, 0.0288, 0.0124, 0.0112, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 20:51:34,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.227e+02 4.296e+02 5.385e+02 1.552e+03, threshold=8.591e+02, percent-clipped=5.0 2023-03-07 20:51:48,668 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:52:07,302 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:52:07,364 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7498, 5.9192, 5.3551, 5.8927, 5.6135, 5.2774, 5.3023, 5.2075], device='cuda:3'), covar=tensor([0.1146, 0.0999, 0.0780, 0.0756, 0.0725, 0.1515, 0.2903, 0.2427], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0412, 0.0310, 0.0331, 0.0300, 0.0372, 0.0448, 0.0397], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 20:52:38,298 INFO [train2.py:809] (3/4) Epoch 7, batch 50, loss[ctc_loss=0.1211, att_loss=0.2741, loss=0.2435, over 16773.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006436, over 48.00 utterances.], tot_loss[ctc_loss=0.1389, att_loss=0.2698, loss=0.2436, over 733586.91 frames. utt_duration=1258 frames, utt_pad_proportion=0.0566, over 2334.99 utterances.], batch size: 48, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:54:02,452 INFO [train2.py:809] (3/4) Epoch 7, batch 100, loss[ctc_loss=0.128, att_loss=0.2633, loss=0.2362, over 17037.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.006754, over 51.00 utterances.], tot_loss[ctc_loss=0.138, att_loss=0.2671, loss=0.2413, over 1297684.81 frames. utt_duration=1230 frames, utt_pad_proportion=0.0585, over 4224.23 utterances.], batch size: 51, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:54:18,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.045e+02 3.704e+02 4.782e+02 7.393e+02, threshold=7.408e+02, percent-clipped=0.0 2023-03-07 20:54:57,397 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:55:03,660 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6938, 1.8215, 5.1271, 3.9138, 3.1359, 4.6883, 4.9005, 4.8506], device='cuda:3'), covar=tensor([0.0164, 0.1956, 0.0073, 0.1044, 0.2020, 0.0181, 0.0069, 0.0144], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0241, 0.0122, 0.0301, 0.0297, 0.0186, 0.0108, 0.0144], device='cuda:3'), out_proj_covar=tensor([1.3030e-04, 1.9783e-04, 1.0920e-04, 2.4501e-04, 2.5240e-04, 1.6271e-04, 9.5777e-05, 1.2973e-04], device='cuda:3') 2023-03-07 20:55:08,171 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0923, 5.1436, 5.0920, 2.3085, 1.8482, 2.3310, 4.8662, 3.7794], device='cuda:3'), covar=tensor([0.0577, 0.0170, 0.0198, 0.3239, 0.6664, 0.3157, 0.0271, 0.1822], device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0192, 0.0216, 0.0179, 0.0354, 0.0336, 0.0204, 0.0331], device='cuda:3'), out_proj_covar=tensor([1.4735e-04, 7.8369e-05, 9.4913e-05, 8.2427e-05, 1.6716e-04, 1.4769e-04, 8.5091e-05, 1.5354e-04], device='cuda:3') 2023-03-07 20:55:17,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-07 20:55:23,559 INFO [train2.py:809] (3/4) Epoch 7, batch 150, loss[ctc_loss=0.1115, att_loss=0.2291, loss=0.2056, over 15798.00 frames. utt_duration=1664 frames, utt_pad_proportion=0.006862, over 38.00 utterances.], tot_loss[ctc_loss=0.1403, att_loss=0.2685, loss=0.2429, over 1732914.16 frames. utt_duration=1182 frames, utt_pad_proportion=0.0725, over 5873.13 utterances.], batch size: 38, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:56:01,443 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9387, 5.1556, 5.5150, 5.5200, 5.2597, 5.8561, 5.0823, 6.0125], device='cuda:3'), covar=tensor([0.0608, 0.0628, 0.0528, 0.0807, 0.1889, 0.0783, 0.0504, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0568, 0.0355, 0.0378, 0.0448, 0.0612, 0.0389, 0.0315, 0.0394], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 20:56:13,945 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.8390, 3.0574, 3.1484, 2.6820, 2.9695, 2.9920, 3.1000, 2.1683], device='cuda:3'), covar=tensor([0.0939, 0.1176, 0.2966, 0.3546, 0.2888, 0.5262, 0.0865, 0.6528], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0080, 0.0085, 0.0123, 0.0076, 0.0112, 0.0071, 0.0128], device='cuda:3'), out_proj_covar=tensor([6.2433e-05, 5.9153e-05, 6.7543e-05, 9.3479e-05, 6.0414e-05, 8.7711e-05, 5.4741e-05, 9.9985e-05], device='cuda:3') 2023-03-07 20:56:34,765 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:56:42,639 INFO [train2.py:809] (3/4) Epoch 7, batch 200, loss[ctc_loss=0.1312, att_loss=0.2741, loss=0.2455, over 16698.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005133, over 46.00 utterances.], tot_loss[ctc_loss=0.1418, att_loss=0.2708, loss=0.245, over 2078841.48 frames. utt_duration=1144 frames, utt_pad_proportion=0.07865, over 7278.82 utterances.], batch size: 46, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:56:56,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.983e+02 2.953e+02 3.642e+02 4.338e+02 9.354e+02, threshold=7.285e+02, percent-clipped=1.0 2023-03-07 20:57:47,338 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 20:57:54,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-07 20:58:02,069 INFO [train2.py:809] (3/4) Epoch 7, batch 250, loss[ctc_loss=0.07874, att_loss=0.2165, loss=0.189, over 15630.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008962, over 37.00 utterances.], tot_loss[ctc_loss=0.1389, att_loss=0.268, loss=0.2422, over 2338927.76 frames. utt_duration=1192 frames, utt_pad_proportion=0.06947, over 7858.63 utterances.], batch size: 37, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:58:55,826 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-03-07 20:59:02,755 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 20:59:09,086 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9074, 3.9905, 3.4816, 4.0142, 4.0247, 3.8742, 3.6667, 2.4923], device='cuda:3'), covar=tensor([0.0238, 0.0234, 0.0328, 0.0158, 0.0909, 0.0226, 0.0384, 0.1948], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0126, 0.0122, 0.0124, 0.0294, 0.0126, 0.0118, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 20:59:22,693 INFO [train2.py:809] (3/4) Epoch 7, batch 300, loss[ctc_loss=0.2877, att_loss=0.347, loss=0.3352, over 13889.00 frames. utt_duration=381.9 frames, utt_pad_proportion=0.3336, over 146.00 utterances.], tot_loss[ctc_loss=0.1408, att_loss=0.2694, loss=0.2437, over 2535818.50 frames. utt_duration=1156 frames, utt_pad_proportion=0.08255, over 8789.02 utterances.], batch size: 146, lr: 1.60e-02, grad_scale: 8.0 2023-03-07 20:59:36,411 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.956e+02 3.932e+02 4.844e+02 1.089e+03, threshold=7.865e+02, percent-clipped=8.0 2023-03-07 20:59:39,193 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-07 21:00:17,456 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1596, 4.5329, 4.1826, 4.6230, 2.5531, 4.3132, 2.5889, 1.4273], device='cuda:3'), covar=tensor([0.0266, 0.0127, 0.0884, 0.0177, 0.2152, 0.0214, 0.1700, 0.2218], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0102, 0.0250, 0.0108, 0.0221, 0.0099, 0.0221, 0.0205], device='cuda:3'), out_proj_covar=tensor([1.1463e-04, 1.0180e-04, 2.2404e-04, 9.8551e-05, 2.0438e-04, 9.6843e-05, 1.9830e-04, 1.8436e-04], device='cuda:3') 2023-03-07 21:00:42,739 INFO [train2.py:809] (3/4) Epoch 7, batch 350, loss[ctc_loss=0.1175, att_loss=0.267, loss=0.2371, over 15942.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007295, over 41.00 utterances.], tot_loss[ctc_loss=0.1404, att_loss=0.2696, loss=0.2437, over 2700951.85 frames. utt_duration=1178 frames, utt_pad_proportion=0.07596, over 9183.17 utterances.], batch size: 41, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:01:01,574 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7409, 4.5783, 4.5766, 4.7606, 5.1195, 4.6898, 4.4416, 2.0293], device='cuda:3'), covar=tensor([0.0255, 0.0384, 0.0271, 0.0153, 0.1025, 0.0242, 0.0446, 0.2795], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0128, 0.0124, 0.0125, 0.0298, 0.0128, 0.0119, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 21:02:03,067 INFO [train2.py:809] (3/4) Epoch 7, batch 400, loss[ctc_loss=0.1059, att_loss=0.2281, loss=0.2036, over 15632.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009653, over 37.00 utterances.], tot_loss[ctc_loss=0.1389, att_loss=0.2688, loss=0.2428, over 2836908.56 frames. utt_duration=1209 frames, utt_pad_proportion=0.06418, over 9393.97 utterances.], batch size: 37, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:02:06,665 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4510, 2.2623, 4.9796, 3.7622, 2.8451, 4.3422, 4.6692, 4.6957], device='cuda:3'), covar=tensor([0.0232, 0.2008, 0.0119, 0.1231, 0.2357, 0.0275, 0.0120, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0240, 0.0124, 0.0299, 0.0292, 0.0183, 0.0107, 0.0141], device='cuda:3'), out_proj_covar=tensor([1.2973e-04, 1.9714e-04, 1.1018e-04, 2.4409e-04, 2.4915e-04, 1.5984e-04, 9.5447e-05, 1.2697e-04], device='cuda:3') 2023-03-07 21:02:16,567 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.848e+02 3.496e+02 4.281e+02 6.983e+02, threshold=6.993e+02, percent-clipped=0.0 2023-03-07 21:03:03,254 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6177, 5.0450, 4.3159, 5.0459, 4.5109, 4.8328, 5.1410, 4.9404], device='cuda:3'), covar=tensor([0.0447, 0.0231, 0.0822, 0.0198, 0.0396, 0.0203, 0.0232, 0.0168], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0199, 0.0259, 0.0178, 0.0217, 0.0164, 0.0186, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 21:03:22,773 INFO [train2.py:809] (3/4) Epoch 7, batch 450, loss[ctc_loss=0.1406, att_loss=0.2764, loss=0.2492, over 16324.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006536, over 45.00 utterances.], tot_loss[ctc_loss=0.1401, att_loss=0.2697, loss=0.2438, over 2931706.02 frames. utt_duration=1179 frames, utt_pad_proportion=0.07357, over 9957.71 utterances.], batch size: 45, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:04:26,025 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:04:42,795 INFO [train2.py:809] (3/4) Epoch 7, batch 500, loss[ctc_loss=0.1285, att_loss=0.2671, loss=0.2394, over 17117.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01516, over 56.00 utterances.], tot_loss[ctc_loss=0.1398, att_loss=0.2693, loss=0.2434, over 3013766.64 frames. utt_duration=1196 frames, utt_pad_proportion=0.06673, over 10096.24 utterances.], batch size: 56, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:04:56,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 3.293e+02 3.980e+02 5.188e+02 9.157e+02, threshold=7.960e+02, percent-clipped=6.0 2023-03-07 21:06:02,364 INFO [train2.py:809] (3/4) Epoch 7, batch 550, loss[ctc_loss=0.1787, att_loss=0.2941, loss=0.271, over 16946.00 frames. utt_duration=693.3 frames, utt_pad_proportion=0.1312, over 98.00 utterances.], tot_loss[ctc_loss=0.1401, att_loss=0.2688, loss=0.2431, over 3060426.96 frames. utt_duration=1196 frames, utt_pad_proportion=0.06963, over 10248.74 utterances.], batch size: 98, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:07:23,381 INFO [train2.py:809] (3/4) Epoch 7, batch 600, loss[ctc_loss=0.1525, att_loss=0.2657, loss=0.243, over 15860.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.008464, over 39.00 utterances.], tot_loss[ctc_loss=0.1399, att_loss=0.2691, loss=0.2432, over 3114968.02 frames. utt_duration=1229 frames, utt_pad_proportion=0.05903, over 10152.87 utterances.], batch size: 39, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:07:37,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.166e+02 3.070e+02 3.787e+02 4.448e+02 1.341e+03, threshold=7.574e+02, percent-clipped=2.0 2023-03-07 21:07:40,656 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3570, 4.7694, 4.6718, 4.8398, 4.8645, 4.4837, 3.5295, 4.6930], device='cuda:3'), covar=tensor([0.0111, 0.0140, 0.0088, 0.0121, 0.0100, 0.0102, 0.0530, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0058, 0.0066, 0.0045, 0.0046, 0.0055, 0.0078, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:08:17,983 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4941, 2.7160, 3.5275, 2.7230, 3.5517, 4.6373, 4.2684, 3.3230], device='cuda:3'), covar=tensor([0.0433, 0.1769, 0.1130, 0.1589, 0.0963, 0.0525, 0.0517, 0.1342], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0211, 0.0214, 0.0191, 0.0215, 0.0235, 0.0180, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 21:08:33,812 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:08:43,729 INFO [train2.py:809] (3/4) Epoch 7, batch 650, loss[ctc_loss=0.1192, att_loss=0.2564, loss=0.229, over 16116.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006818, over 42.00 utterances.], tot_loss[ctc_loss=0.1382, att_loss=0.2681, loss=0.2421, over 3144776.03 frames. utt_duration=1232 frames, utt_pad_proportion=0.06102, over 10222.49 utterances.], batch size: 42, lr: 1.59e-02, grad_scale: 8.0 2023-03-07 21:08:56,541 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0994, 4.9692, 4.8528, 4.7697, 5.4255, 5.2785, 4.7241, 2.3169], device='cuda:3'), covar=tensor([0.0173, 0.0246, 0.0210, 0.0265, 0.0745, 0.0126, 0.0284, 0.2447], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0122, 0.0122, 0.0119, 0.0293, 0.0124, 0.0115, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 21:09:08,953 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:09:10,670 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0627, 5.1848, 5.0537, 2.3867, 1.9104, 2.5945, 4.5831, 3.6502], device='cuda:3'), covar=tensor([0.0595, 0.0171, 0.0199, 0.3751, 0.6470, 0.2708, 0.0512, 0.2048], device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0189, 0.0213, 0.0176, 0.0354, 0.0338, 0.0205, 0.0331], device='cuda:3'), out_proj_covar=tensor([1.4685e-04, 7.6889e-05, 9.2981e-05, 8.1076e-05, 1.6618e-04, 1.4760e-04, 8.5731e-05, 1.5307e-04], device='cuda:3') 2023-03-07 21:10:04,223 INFO [train2.py:809] (3/4) Epoch 7, batch 700, loss[ctc_loss=0.1233, att_loss=0.2598, loss=0.2325, over 16397.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007152, over 44.00 utterances.], tot_loss[ctc_loss=0.1376, att_loss=0.2679, loss=0.2419, over 3181357.48 frames. utt_duration=1235 frames, utt_pad_proportion=0.05618, over 10319.94 utterances.], batch size: 44, lr: 1.58e-02, grad_scale: 8.0 2023-03-07 21:10:12,542 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:10:18,242 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.863e+02 3.503e+02 4.304e+02 9.533e+02, threshold=7.005e+02, percent-clipped=2.0 2023-03-07 21:10:45,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.68 vs. limit=5.0 2023-03-07 21:10:46,441 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:11:23,264 INFO [train2.py:809] (3/4) Epoch 7, batch 750, loss[ctc_loss=0.1578, att_loss=0.2874, loss=0.2615, over 16976.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.00691, over 50.00 utterances.], tot_loss[ctc_loss=0.1386, att_loss=0.2688, loss=0.2428, over 3205458.84 frames. utt_duration=1227 frames, utt_pad_proportion=0.0577, over 10459.56 utterances.], batch size: 50, lr: 1.58e-02, grad_scale: 16.0 2023-03-07 21:12:27,415 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:12:43,247 INFO [train2.py:809] (3/4) Epoch 7, batch 800, loss[ctc_loss=0.1181, att_loss=0.2668, loss=0.237, over 16334.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006065, over 45.00 utterances.], tot_loss[ctc_loss=0.1394, att_loss=0.269, loss=0.243, over 3212988.60 frames. utt_duration=1203 frames, utt_pad_proportion=0.06644, over 10696.48 utterances.], batch size: 45, lr: 1.58e-02, grad_scale: 16.0 2023-03-07 21:12:57,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.992e+02 3.996e+02 4.981e+02 1.398e+03, threshold=7.991e+02, percent-clipped=9.0 2023-03-07 21:13:43,783 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:14:03,648 INFO [train2.py:809] (3/4) Epoch 7, batch 850, loss[ctc_loss=0.1189, att_loss=0.2452, loss=0.2199, over 15996.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.007884, over 40.00 utterances.], tot_loss[ctc_loss=0.1379, att_loss=0.268, loss=0.242, over 3225554.38 frames. utt_duration=1214 frames, utt_pad_proportion=0.06437, over 10640.86 utterances.], batch size: 40, lr: 1.58e-02, grad_scale: 16.0 2023-03-07 21:14:08,856 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2887, 5.1966, 5.0466, 2.6492, 2.0761, 2.5992, 4.5914, 3.7228], device='cuda:3'), covar=tensor([0.0503, 0.0148, 0.0197, 0.3241, 0.5967, 0.2642, 0.0455, 0.1937], device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0187, 0.0214, 0.0177, 0.0353, 0.0338, 0.0205, 0.0331], device='cuda:3'), out_proj_covar=tensor([1.4797e-04, 7.5852e-05, 9.4240e-05, 8.1263e-05, 1.6563e-04, 1.4777e-04, 8.4611e-05, 1.5299e-04], device='cuda:3') 2023-03-07 21:14:22,664 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:15:24,353 INFO [train2.py:809] (3/4) Epoch 7, batch 900, loss[ctc_loss=0.1454, att_loss=0.2849, loss=0.257, over 16767.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005745, over 48.00 utterances.], tot_loss[ctc_loss=0.1381, att_loss=0.2682, loss=0.2422, over 3233800.61 frames. utt_duration=1190 frames, utt_pad_proportion=0.07077, over 10880.85 utterances.], batch size: 48, lr: 1.58e-02, grad_scale: 16.0 2023-03-07 21:15:38,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 2.848e+02 3.280e+02 4.214e+02 1.122e+03, threshold=6.561e+02, percent-clipped=2.0 2023-03-07 21:16:00,315 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:16:06,641 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7354, 2.1697, 4.9613, 3.6951, 3.0637, 4.7347, 4.9785, 4.8821], device='cuda:3'), covar=tensor([0.0167, 0.1854, 0.0159, 0.1082, 0.1878, 0.0168, 0.0078, 0.0163], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0237, 0.0124, 0.0296, 0.0291, 0.0180, 0.0107, 0.0138], device='cuda:3'), out_proj_covar=tensor([1.3003e-04, 1.9507e-04, 1.1043e-04, 2.4254e-04, 2.4825e-04, 1.5723e-04, 9.5542e-05, 1.2508e-04], device='cuda:3') 2023-03-07 21:16:36,260 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:16:44,175 INFO [train2.py:809] (3/4) Epoch 7, batch 950, loss[ctc_loss=0.1275, att_loss=0.258, loss=0.2319, over 16162.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.007424, over 41.00 utterances.], tot_loss[ctc_loss=0.1359, att_loss=0.2665, loss=0.2404, over 3240641.79 frames. utt_duration=1232 frames, utt_pad_proportion=0.0605, over 10532.25 utterances.], batch size: 41, lr: 1.58e-02, grad_scale: 16.0 2023-03-07 21:17:24,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 21:18:03,788 INFO [train2.py:809] (3/4) Epoch 7, batch 1000, loss[ctc_loss=0.1604, att_loss=0.2867, loss=0.2614, over 17140.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01318, over 56.00 utterances.], tot_loss[ctc_loss=0.1363, att_loss=0.2668, loss=0.2407, over 3242810.44 frames. utt_duration=1246 frames, utt_pad_proportion=0.05682, over 10421.47 utterances.], batch size: 56, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:18:04,027 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:18:13,389 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:18:17,632 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.930e+02 3.688e+02 4.379e+02 1.333e+03, threshold=7.376e+02, percent-clipped=8.0 2023-03-07 21:18:37,653 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:19:22,727 INFO [train2.py:809] (3/4) Epoch 7, batch 1050, loss[ctc_loss=0.1274, att_loss=0.2517, loss=0.2268, over 16284.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006914, over 43.00 utterances.], tot_loss[ctc_loss=0.1368, att_loss=0.2672, loss=0.2411, over 3246203.76 frames. utt_duration=1235 frames, utt_pad_proportion=0.05883, over 10523.09 utterances.], batch size: 43, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:20:03,413 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1211, 5.0725, 4.9135, 2.8412, 4.8326, 4.3028, 4.1382, 2.3632], device='cuda:3'), covar=tensor([0.0111, 0.0067, 0.0184, 0.1025, 0.0092, 0.0196, 0.0320, 0.1414], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0068, 0.0058, 0.0098, 0.0062, 0.0078, 0.0083, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 21:20:25,610 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-07 21:20:29,715 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:20:43,049 INFO [train2.py:809] (3/4) Epoch 7, batch 1100, loss[ctc_loss=0.1634, att_loss=0.2852, loss=0.2609, over 17026.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007526, over 51.00 utterances.], tot_loss[ctc_loss=0.136, att_loss=0.2669, loss=0.2407, over 3254660.64 frames. utt_duration=1225 frames, utt_pad_proportion=0.05991, over 10640.03 utterances.], batch size: 51, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:20:57,051 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.910e+02 3.574e+02 4.609e+02 1.124e+03, threshold=7.148e+02, percent-clipped=6.0 2023-03-07 21:22:03,894 INFO [train2.py:809] (3/4) Epoch 7, batch 1150, loss[ctc_loss=0.1519, att_loss=0.2838, loss=0.2574, over 16757.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.007087, over 48.00 utterances.], tot_loss[ctc_loss=0.1361, att_loss=0.2668, loss=0.2407, over 3252281.49 frames. utt_duration=1231 frames, utt_pad_proportion=0.06136, over 10580.80 utterances.], batch size: 48, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:22:08,926 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 21:23:24,556 INFO [train2.py:809] (3/4) Epoch 7, batch 1200, loss[ctc_loss=0.1547, att_loss=0.2612, loss=0.2399, over 15637.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009384, over 37.00 utterances.], tot_loss[ctc_loss=0.1358, att_loss=0.2668, loss=0.2406, over 3256785.51 frames. utt_duration=1211 frames, utt_pad_proportion=0.06524, over 10768.20 utterances.], batch size: 37, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:23:38,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.863e+02 3.570e+02 4.681e+02 1.368e+03, threshold=7.141e+02, percent-clipped=7.0 2023-03-07 21:23:52,669 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:24:05,387 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6644, 5.0817, 4.4984, 5.0801, 4.5021, 4.8629, 5.2532, 5.0161], device='cuda:3'), covar=tensor([0.0539, 0.0231, 0.0974, 0.0229, 0.0558, 0.0234, 0.0231, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0203, 0.0263, 0.0179, 0.0217, 0.0165, 0.0191, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 21:24:17,577 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 21:24:34,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-03-07 21:24:44,724 INFO [train2.py:809] (3/4) Epoch 7, batch 1250, loss[ctc_loss=0.1118, att_loss=0.2493, loss=0.2218, over 15998.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008452, over 40.00 utterances.], tot_loss[ctc_loss=0.1369, att_loss=0.2676, loss=0.2414, over 3250215.07 frames. utt_duration=1185 frames, utt_pad_proportion=0.07348, over 10982.56 utterances.], batch size: 40, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:25:21,382 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1507, 4.8783, 4.9201, 2.5617, 4.8026, 4.2473, 4.1306, 2.7138], device='cuda:3'), covar=tensor([0.0076, 0.0123, 0.0183, 0.1150, 0.0101, 0.0207, 0.0344, 0.1365], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0068, 0.0059, 0.0098, 0.0063, 0.0079, 0.0083, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 21:26:05,284 INFO [train2.py:809] (3/4) Epoch 7, batch 1300, loss[ctc_loss=0.154, att_loss=0.2842, loss=0.2582, over 16339.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005563, over 45.00 utterances.], tot_loss[ctc_loss=0.137, att_loss=0.2677, loss=0.2415, over 3256591.77 frames. utt_duration=1196 frames, utt_pad_proportion=0.07014, over 10909.31 utterances.], batch size: 45, lr: 1.57e-02, grad_scale: 16.0 2023-03-07 21:26:05,540 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:26:06,849 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:26:19,176 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.854e+02 3.559e+02 4.625e+02 8.622e+02, threshold=7.119e+02, percent-clipped=5.0 2023-03-07 21:26:19,660 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2174, 4.6771, 4.5806, 4.8484, 2.2634, 4.3873, 2.6692, 2.1687], device='cuda:3'), covar=tensor([0.0282, 0.0181, 0.0819, 0.0142, 0.2617, 0.0234, 0.1913, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0103, 0.0253, 0.0109, 0.0225, 0.0105, 0.0227, 0.0203], device='cuda:3'), out_proj_covar=tensor([1.1652e-04, 1.0340e-04, 2.2704e-04, 9.9568e-05, 2.0824e-04, 1.0270e-04, 2.0354e-04, 1.8434e-04], device='cuda:3') 2023-03-07 21:26:39,780 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:26:40,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-03-07 21:27:22,556 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:27:25,584 INFO [train2.py:809] (3/4) Epoch 7, batch 1350, loss[ctc_loss=0.1532, att_loss=0.281, loss=0.2554, over 16952.00 frames. utt_duration=686.5 frames, utt_pad_proportion=0.1343, over 99.00 utterances.], tot_loss[ctc_loss=0.1364, att_loss=0.2675, loss=0.2413, over 3262373.07 frames. utt_duration=1223 frames, utt_pad_proportion=0.06145, over 10679.58 utterances.], batch size: 99, lr: 1.56e-02, grad_scale: 16.0 2023-03-07 21:27:56,368 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:28:45,922 INFO [train2.py:809] (3/4) Epoch 7, batch 1400, loss[ctc_loss=0.1341, att_loss=0.2546, loss=0.2305, over 16399.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006956, over 44.00 utterances.], tot_loss[ctc_loss=0.1355, att_loss=0.2672, loss=0.2408, over 3267511.90 frames. utt_duration=1218 frames, utt_pad_proportion=0.06161, over 10741.39 utterances.], batch size: 44, lr: 1.56e-02, grad_scale: 16.0 2023-03-07 21:29:01,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.920e+02 3.478e+02 4.487e+02 1.591e+03, threshold=6.957e+02, percent-clipped=5.0 2023-03-07 21:29:14,321 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8951, 4.6940, 4.6343, 4.7397, 5.1587, 4.9583, 4.6234, 2.2125], device='cuda:3'), covar=tensor([0.0191, 0.0246, 0.0206, 0.0154, 0.0943, 0.0188, 0.0292, 0.2683], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0123, 0.0121, 0.0120, 0.0304, 0.0127, 0.0116, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 21:30:03,020 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 21:30:05,929 INFO [train2.py:809] (3/4) Epoch 7, batch 1450, loss[ctc_loss=0.1378, att_loss=0.2722, loss=0.2453, over 16960.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008152, over 50.00 utterances.], tot_loss[ctc_loss=0.1346, att_loss=0.2663, loss=0.24, over 3273002.82 frames. utt_duration=1249 frames, utt_pad_proportion=0.05331, over 10494.24 utterances.], batch size: 50, lr: 1.56e-02, grad_scale: 8.0 2023-03-07 21:30:50,045 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-07 21:31:25,457 INFO [train2.py:809] (3/4) Epoch 7, batch 1500, loss[ctc_loss=0.09737, att_loss=0.2332, loss=0.2061, over 15776.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008447, over 38.00 utterances.], tot_loss[ctc_loss=0.1359, att_loss=0.2677, loss=0.2413, over 3276059.78 frames. utt_duration=1248 frames, utt_pad_proportion=0.0534, over 10510.57 utterances.], batch size: 38, lr: 1.56e-02, grad_scale: 8.0 2023-03-07 21:31:40,547 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.966e+02 3.623e+02 4.720e+02 8.485e+02, threshold=7.246e+02, percent-clipped=4.0 2023-03-07 21:31:52,977 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:32:04,499 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 21:32:44,993 INFO [train2.py:809] (3/4) Epoch 7, batch 1550, loss[ctc_loss=0.1264, att_loss=0.2466, loss=0.2225, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008798, over 37.00 utterances.], tot_loss[ctc_loss=0.1361, att_loss=0.2676, loss=0.2413, over 3280621.09 frames. utt_duration=1276 frames, utt_pad_proportion=0.04591, over 10294.52 utterances.], batch size: 37, lr: 1.56e-02, grad_scale: 8.0 2023-03-07 21:33:10,111 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:33:11,874 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2558, 4.8763, 4.5042, 4.7924, 4.8653, 4.4918, 3.4536, 4.6339], device='cuda:3'), covar=tensor([0.0116, 0.0085, 0.0104, 0.0111, 0.0066, 0.0095, 0.0527, 0.0180], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0056, 0.0068, 0.0045, 0.0045, 0.0055, 0.0078, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:33:42,988 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 21:34:04,337 INFO [train2.py:809] (3/4) Epoch 7, batch 1600, loss[ctc_loss=0.1561, att_loss=0.2939, loss=0.2663, over 17270.00 frames. utt_duration=1172 frames, utt_pad_proportion=0.02412, over 59.00 utterances.], tot_loss[ctc_loss=0.1358, att_loss=0.2673, loss=0.241, over 3274393.57 frames. utt_duration=1258 frames, utt_pad_proportion=0.05188, over 10420.01 utterances.], batch size: 59, lr: 1.56e-02, grad_scale: 8.0 2023-03-07 21:34:06,174 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:34:07,761 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2671, 2.3341, 3.2384, 4.0348, 3.8950, 3.8737, 2.6467, 1.9391], device='cuda:3'), covar=tensor([0.0594, 0.2534, 0.0987, 0.0681, 0.0446, 0.0351, 0.1688, 0.2556], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0199, 0.0188, 0.0172, 0.0153, 0.0130, 0.0186, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 21:34:19,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.911e+02 3.522e+02 4.081e+02 9.255e+02, threshold=7.044e+02, percent-clipped=3.0 2023-03-07 21:35:20,318 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-03-07 21:35:22,695 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:35:24,112 INFO [train2.py:809] (3/4) Epoch 7, batch 1650, loss[ctc_loss=0.126, att_loss=0.272, loss=0.2428, over 16632.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004856, over 47.00 utterances.], tot_loss[ctc_loss=0.136, att_loss=0.2675, loss=0.2412, over 3274889.30 frames. utt_duration=1227 frames, utt_pad_proportion=0.05867, over 10684.83 utterances.], batch size: 47, lr: 1.56e-02, grad_scale: 8.0 2023-03-07 21:36:44,062 INFO [train2.py:809] (3/4) Epoch 7, batch 1700, loss[ctc_loss=0.1512, att_loss=0.2769, loss=0.2518, over 17027.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.00754, over 51.00 utterances.], tot_loss[ctc_loss=0.1364, att_loss=0.2679, loss=0.2416, over 3278494.41 frames. utt_duration=1214 frames, utt_pad_proportion=0.0613, over 10818.76 utterances.], batch size: 51, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:36:59,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 2.859e+02 3.554e+02 4.579e+02 8.259e+02, threshold=7.107e+02, percent-clipped=6.0 2023-03-07 21:38:00,692 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 21:38:03,530 INFO [train2.py:809] (3/4) Epoch 7, batch 1750, loss[ctc_loss=0.1589, att_loss=0.2912, loss=0.2647, over 16474.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006101, over 46.00 utterances.], tot_loss[ctc_loss=0.1359, att_loss=0.267, loss=0.2408, over 3259510.02 frames. utt_duration=1226 frames, utt_pad_proportion=0.0638, over 10647.15 utterances.], batch size: 46, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:39:18,793 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:39:25,097 INFO [train2.py:809] (3/4) Epoch 7, batch 1800, loss[ctc_loss=0.1385, att_loss=0.2794, loss=0.2512, over 16761.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.00566, over 48.00 utterances.], tot_loss[ctc_loss=0.1344, att_loss=0.2665, loss=0.2401, over 3262815.09 frames. utt_duration=1223 frames, utt_pad_proportion=0.06349, over 10687.69 utterances.], batch size: 48, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:39:40,904 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.893e+02 3.531e+02 4.524e+02 7.224e+02, threshold=7.063e+02, percent-clipped=1.0 2023-03-07 21:40:19,667 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1501, 5.0559, 4.9616, 2.4073, 1.9249, 2.6750, 4.7018, 3.6892], device='cuda:3'), covar=tensor([0.0558, 0.0178, 0.0243, 0.3376, 0.6412, 0.2655, 0.0304, 0.1977], device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0196, 0.0217, 0.0182, 0.0363, 0.0339, 0.0206, 0.0341], device='cuda:3'), out_proj_covar=tensor([1.4954e-04, 7.7847e-05, 9.5895e-05, 8.4178e-05, 1.6843e-04, 1.4737e-04, 8.4035e-05, 1.5545e-04], device='cuda:3') 2023-03-07 21:40:46,309 INFO [train2.py:809] (3/4) Epoch 7, batch 1850, loss[ctc_loss=0.126, att_loss=0.2582, loss=0.2317, over 16343.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005563, over 45.00 utterances.], tot_loss[ctc_loss=0.1338, att_loss=0.266, loss=0.2396, over 3259534.99 frames. utt_duration=1217 frames, utt_pad_proportion=0.06703, over 10722.52 utterances.], batch size: 45, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:41:04,476 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-07 21:41:37,052 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 21:41:41,137 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.76 vs. limit=5.0 2023-03-07 21:42:06,980 INFO [train2.py:809] (3/4) Epoch 7, batch 1900, loss[ctc_loss=0.1531, att_loss=0.2748, loss=0.2505, over 16541.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006092, over 45.00 utterances.], tot_loss[ctc_loss=0.1341, att_loss=0.2659, loss=0.2396, over 3258319.27 frames. utt_duration=1222 frames, utt_pad_proportion=0.06702, over 10679.60 utterances.], batch size: 45, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:42:22,403 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.923e+02 3.414e+02 4.315e+02 6.616e+02, threshold=6.827e+02, percent-clipped=0.0 2023-03-07 21:43:27,371 INFO [train2.py:809] (3/4) Epoch 7, batch 1950, loss[ctc_loss=0.1228, att_loss=0.2848, loss=0.2524, over 17432.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01144, over 57.00 utterances.], tot_loss[ctc_loss=0.135, att_loss=0.2665, loss=0.2402, over 3266682.58 frames. utt_duration=1233 frames, utt_pad_proportion=0.06255, over 10612.07 utterances.], batch size: 57, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:43:35,793 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 21:44:01,831 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 21:44:48,787 INFO [train2.py:809] (3/4) Epoch 7, batch 2000, loss[ctc_loss=0.1146, att_loss=0.2553, loss=0.2271, over 16325.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006354, over 45.00 utterances.], tot_loss[ctc_loss=0.1355, att_loss=0.2677, loss=0.2413, over 3269582.22 frames. utt_duration=1215 frames, utt_pad_proportion=0.06665, over 10773.19 utterances.], batch size: 45, lr: 1.55e-02, grad_scale: 8.0 2023-03-07 21:45:04,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.979e+02 3.606e+02 4.355e+02 1.257e+03, threshold=7.211e+02, percent-clipped=2.0 2023-03-07 21:45:14,777 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 21:45:43,824 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2459, 4.8580, 4.6357, 4.8590, 4.8780, 4.4747, 3.5345, 4.6005], device='cuda:3'), covar=tensor([0.0108, 0.0097, 0.0083, 0.0075, 0.0076, 0.0111, 0.0556, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0068, 0.0045, 0.0047, 0.0056, 0.0080, 0.0077], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:46:09,513 INFO [train2.py:809] (3/4) Epoch 7, batch 2050, loss[ctc_loss=0.2013, att_loss=0.3014, loss=0.2814, over 13887.00 frames. utt_duration=384.7 frames, utt_pad_proportion=0.3309, over 145.00 utterances.], tot_loss[ctc_loss=0.1361, att_loss=0.2679, loss=0.2415, over 3269592.82 frames. utt_duration=1210 frames, utt_pad_proportion=0.06737, over 10824.39 utterances.], batch size: 145, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:47:33,897 INFO [train2.py:809] (3/4) Epoch 7, batch 2100, loss[ctc_loss=0.1544, att_loss=0.2872, loss=0.2606, over 17376.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.0342, over 63.00 utterances.], tot_loss[ctc_loss=0.1361, att_loss=0.2677, loss=0.2414, over 3272849.54 frames. utt_duration=1217 frames, utt_pad_proportion=0.06506, over 10768.48 utterances.], batch size: 63, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:47:49,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.961e+02 3.539e+02 4.278e+02 1.039e+03, threshold=7.077e+02, percent-clipped=2.0 2023-03-07 21:48:38,042 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5667, 4.0196, 3.1427, 3.5227, 4.0370, 3.6206, 2.4287, 4.4984], device='cuda:3'), covar=tensor([0.1336, 0.0434, 0.1133, 0.0746, 0.0642, 0.0706, 0.1259, 0.0475], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0156, 0.0190, 0.0161, 0.0189, 0.0190, 0.0164, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:48:54,478 INFO [train2.py:809] (3/4) Epoch 7, batch 2150, loss[ctc_loss=0.1474, att_loss=0.2847, loss=0.2573, over 17473.00 frames. utt_duration=1014 frames, utt_pad_proportion=0.0442, over 69.00 utterances.], tot_loss[ctc_loss=0.136, att_loss=0.2676, loss=0.2413, over 3272856.37 frames. utt_duration=1220 frames, utt_pad_proportion=0.06404, over 10740.05 utterances.], batch size: 69, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:49:44,664 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 21:50:13,403 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:50:14,579 INFO [train2.py:809] (3/4) Epoch 7, batch 2200, loss[ctc_loss=0.1241, att_loss=0.2714, loss=0.2419, over 17273.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01249, over 55.00 utterances.], tot_loss[ctc_loss=0.1347, att_loss=0.2668, loss=0.2404, over 3270034.54 frames. utt_duration=1239 frames, utt_pad_proportion=0.05753, over 10567.15 utterances.], batch size: 55, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:50:30,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.999e+02 3.731e+02 5.039e+02 1.380e+03, threshold=7.462e+02, percent-clipped=10.0 2023-03-07 21:51:01,661 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 21:51:04,789 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4939, 4.9372, 4.7626, 5.0993, 5.0958, 4.7388, 3.4010, 4.8638], device='cuda:3'), covar=tensor([0.0094, 0.0091, 0.0103, 0.0059, 0.0075, 0.0083, 0.0593, 0.0199], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0055, 0.0067, 0.0044, 0.0046, 0.0055, 0.0078, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:51:34,424 INFO [train2.py:809] (3/4) Epoch 7, batch 2250, loss[ctc_loss=0.1372, att_loss=0.2556, loss=0.2319, over 16130.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.00556, over 42.00 utterances.], tot_loss[ctc_loss=0.1346, att_loss=0.2669, loss=0.2404, over 3277499.22 frames. utt_duration=1252 frames, utt_pad_proportion=0.05156, over 10480.81 utterances.], batch size: 42, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:51:50,754 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:51:59,362 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:52:41,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6751, 5.1528, 4.9622, 5.3275, 5.2519, 4.9245, 3.9678, 5.1810], device='cuda:3'), covar=tensor([0.0086, 0.0106, 0.0080, 0.0052, 0.0083, 0.0085, 0.0423, 0.0128], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0067, 0.0043, 0.0046, 0.0055, 0.0078, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 21:52:55,165 INFO [train2.py:809] (3/4) Epoch 7, batch 2300, loss[ctc_loss=0.1022, att_loss=0.267, loss=0.234, over 16681.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007004, over 46.00 utterances.], tot_loss[ctc_loss=0.1348, att_loss=0.2675, loss=0.2409, over 3274004.68 frames. utt_duration=1248 frames, utt_pad_proportion=0.05265, over 10505.26 utterances.], batch size: 46, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:53:10,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.838e+02 3.640e+02 4.798e+02 9.404e+02, threshold=7.281e+02, percent-clipped=6.0 2023-03-07 21:53:13,041 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 21:53:37,091 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:54:11,200 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:54:15,533 INFO [train2.py:809] (3/4) Epoch 7, batch 2350, loss[ctc_loss=0.1453, att_loss=0.2861, loss=0.258, over 17433.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.03119, over 63.00 utterances.], tot_loss[ctc_loss=0.1353, att_loss=0.2676, loss=0.2411, over 3276975.22 frames. utt_duration=1240 frames, utt_pad_proportion=0.05541, over 10587.77 utterances.], batch size: 63, lr: 1.54e-02, grad_scale: 8.0 2023-03-07 21:55:19,721 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4207, 4.7144, 4.2099, 5.0816, 2.3683, 4.6162, 2.6275, 2.3563], device='cuda:3'), covar=tensor([0.0246, 0.0138, 0.0985, 0.0122, 0.2355, 0.0209, 0.1710, 0.1627], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0103, 0.0249, 0.0109, 0.0218, 0.0104, 0.0226, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-07 21:55:36,003 INFO [train2.py:809] (3/4) Epoch 7, batch 2400, loss[ctc_loss=0.1304, att_loss=0.2577, loss=0.2323, over 15870.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009015, over 39.00 utterances.], tot_loss[ctc_loss=0.1351, att_loss=0.2673, loss=0.2408, over 3277451.33 frames. utt_duration=1231 frames, utt_pad_proportion=0.05695, over 10659.86 utterances.], batch size: 39, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 21:55:49,051 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:55:52,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 3.013e+02 3.656e+02 4.621e+02 8.184e+02, threshold=7.313e+02, percent-clipped=1.0 2023-03-07 21:56:22,360 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9233, 1.7273, 1.9989, 1.6462, 2.9654, 2.1417, 2.2079, 1.9334], device='cuda:3'), covar=tensor([0.0588, 0.3826, 0.3927, 0.1866, 0.0637, 0.1871, 0.2418, 0.1993], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0090, 0.0091, 0.0082, 0.0079, 0.0079, 0.0090, 0.0069], device='cuda:3'), out_proj_covar=tensor([4.0184e-05, 5.4187e-05, 5.2606e-05, 4.5077e-05, 3.9950e-05, 4.6555e-05, 5.3008e-05, 4.3863e-05], device='cuda:3') 2023-03-07 21:56:56,979 INFO [train2.py:809] (3/4) Epoch 7, batch 2450, loss[ctc_loss=0.1545, att_loss=0.271, loss=0.2477, over 16337.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005806, over 45.00 utterances.], tot_loss[ctc_loss=0.1341, att_loss=0.2663, loss=0.2399, over 3280076.30 frames. utt_duration=1244 frames, utt_pad_proportion=0.05296, over 10556.91 utterances.], batch size: 45, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 21:58:18,163 INFO [train2.py:809] (3/4) Epoch 7, batch 2500, loss[ctc_loss=0.1487, att_loss=0.2868, loss=0.2592, over 17308.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01154, over 55.00 utterances.], tot_loss[ctc_loss=0.1331, att_loss=0.266, loss=0.2395, over 3285013.63 frames. utt_duration=1243 frames, utt_pad_proportion=0.05171, over 10582.65 utterances.], batch size: 55, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 21:58:26,180 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7577, 1.7261, 1.8238, 1.6400, 2.9245, 2.1504, 1.8554, 2.0746], device='cuda:3'), covar=tensor([0.0602, 0.3185, 0.2468, 0.0845, 0.0553, 0.1205, 0.1783, 0.1050], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0090, 0.0088, 0.0082, 0.0077, 0.0077, 0.0090, 0.0068], device='cuda:3'), out_proj_covar=tensor([4.0026e-05, 5.3839e-05, 5.1627e-05, 4.4988e-05, 3.9623e-05, 4.5846e-05, 5.3073e-05, 4.3376e-05], device='cuda:3') 2023-03-07 21:58:33,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.816e+02 3.513e+02 4.106e+02 7.996e+02, threshold=7.027e+02, percent-clipped=1.0 2023-03-07 21:58:36,090 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 21:59:14,955 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1450, 5.2701, 5.2600, 2.6172, 4.9836, 4.5216, 4.7177, 2.1414], device='cuda:3'), covar=tensor([0.0194, 0.0083, 0.0161, 0.1359, 0.0094, 0.0213, 0.0285, 0.2318], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0072, 0.0059, 0.0099, 0.0063, 0.0081, 0.0085, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 21:59:38,004 INFO [train2.py:809] (3/4) Epoch 7, batch 2550, loss[ctc_loss=0.1386, att_loss=0.2716, loss=0.245, over 16336.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005776, over 45.00 utterances.], tot_loss[ctc_loss=0.1326, att_loss=0.2652, loss=0.2387, over 3279203.99 frames. utt_duration=1254 frames, utt_pad_proportion=0.04884, over 10469.03 utterances.], batch size: 45, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 21:59:46,031 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:00:14,069 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:00:58,231 INFO [train2.py:809] (3/4) Epoch 7, batch 2600, loss[ctc_loss=0.142, att_loss=0.2709, loss=0.2452, over 15969.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.006031, over 41.00 utterances.], tot_loss[ctc_loss=0.1333, att_loss=0.2655, loss=0.2391, over 3280637.74 frames. utt_duration=1252 frames, utt_pad_proportion=0.04955, over 10492.15 utterances.], batch size: 41, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 22:00:58,412 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7079, 6.0291, 5.2124, 5.8593, 5.4851, 5.2653, 5.4659, 5.2659], device='cuda:3'), covar=tensor([0.1546, 0.0959, 0.1093, 0.0785, 0.0970, 0.1481, 0.2163, 0.2497], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0435, 0.0327, 0.0340, 0.0312, 0.0386, 0.0446, 0.0411], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 22:01:14,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 3.263e+02 3.835e+02 5.237e+02 1.388e+03, threshold=7.671e+02, percent-clipped=11.0 2023-03-07 22:01:16,498 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:01:31,414 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:02:15,357 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7625, 5.1500, 4.9773, 4.9435, 5.1871, 5.1290, 4.8913, 4.5940], device='cuda:3'), covar=tensor([0.1249, 0.0468, 0.0268, 0.0481, 0.0270, 0.0318, 0.0283, 0.0357], device='cuda:3'), in_proj_covar=tensor([0.0420, 0.0252, 0.0197, 0.0234, 0.0290, 0.0317, 0.0238, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 22:02:18,264 INFO [train2.py:809] (3/4) Epoch 7, batch 2650, loss[ctc_loss=0.124, att_loss=0.2725, loss=0.2428, over 17303.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.0111, over 55.00 utterances.], tot_loss[ctc_loss=0.133, att_loss=0.2653, loss=0.2388, over 3278562.38 frames. utt_duration=1261 frames, utt_pad_proportion=0.04782, over 10409.89 utterances.], batch size: 55, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 22:02:32,525 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 22:03:37,579 INFO [train2.py:809] (3/4) Epoch 7, batch 2700, loss[ctc_loss=0.1948, att_loss=0.3052, loss=0.2831, over 17446.00 frames. utt_duration=1110 frames, utt_pad_proportion=0.02675, over 63.00 utterances.], tot_loss[ctc_loss=0.1343, att_loss=0.2663, loss=0.2399, over 3279753.49 frames. utt_duration=1230 frames, utt_pad_proportion=0.05368, over 10675.15 utterances.], batch size: 63, lr: 1.53e-02, grad_scale: 8.0 2023-03-07 22:03:40,984 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9613, 5.2445, 5.1200, 5.0696, 5.2795, 5.3390, 5.1053, 4.7663], device='cuda:3'), covar=tensor([0.0942, 0.0366, 0.0243, 0.0408, 0.0256, 0.0207, 0.0243, 0.0292], device='cuda:3'), in_proj_covar=tensor([0.0425, 0.0255, 0.0199, 0.0236, 0.0294, 0.0319, 0.0240, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 22:03:42,446 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:03:42,580 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4524, 3.0073, 3.6242, 2.7789, 3.4977, 4.5752, 4.3210, 3.3407], device='cuda:3'), covar=tensor([0.0443, 0.1388, 0.0887, 0.1343, 0.0954, 0.0687, 0.0544, 0.1179], device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0210, 0.0211, 0.0191, 0.0216, 0.0239, 0.0185, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 22:03:43,098 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-03-07 22:03:53,739 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 3.023e+02 3.591e+02 4.363e+02 1.241e+03, threshold=7.183e+02, percent-clipped=4.0 2023-03-07 22:04:35,794 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4519, 2.4196, 5.0060, 3.8186, 3.0266, 4.3855, 4.8836, 4.6992], device='cuda:3'), covar=tensor([0.0279, 0.1756, 0.0158, 0.1328, 0.2059, 0.0268, 0.0108, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0246, 0.0127, 0.0307, 0.0289, 0.0183, 0.0111, 0.0143], device='cuda:3'), out_proj_covar=tensor([1.3423e-04, 2.0321e-04, 1.1398e-04, 2.5131e-04, 2.5002e-04, 1.6243e-04, 9.9801e-05, 1.3072e-04], device='cuda:3') 2023-03-07 22:04:56,686 INFO [train2.py:809] (3/4) Epoch 7, batch 2750, loss[ctc_loss=0.1474, att_loss=0.2913, loss=0.2625, over 17315.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01119, over 55.00 utterances.], tot_loss[ctc_loss=0.1333, att_loss=0.2658, loss=0.2393, over 3285863.85 frames. utt_duration=1256 frames, utt_pad_proportion=0.04685, over 10476.63 utterances.], batch size: 55, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:05:10,907 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2176, 5.0981, 4.9938, 2.7159, 1.9683, 2.6538, 4.7958, 3.6526], device='cuda:3'), covar=tensor([0.0564, 0.0198, 0.0255, 0.3092, 0.6285, 0.2887, 0.0352, 0.2190], device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0199, 0.0221, 0.0182, 0.0362, 0.0345, 0.0213, 0.0345], device='cuda:3'), out_proj_covar=tensor([1.5262e-04, 7.9098e-05, 9.7661e-05, 8.4611e-05, 1.6713e-04, 1.4898e-04, 8.6849e-05, 1.5642e-04], device='cuda:3') 2023-03-07 22:05:26,798 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0932, 4.8229, 4.7969, 2.3858, 2.0023, 2.5194, 4.1285, 3.5471], device='cuda:3'), covar=tensor([0.0583, 0.0150, 0.0186, 0.3343, 0.6179, 0.2960, 0.0596, 0.2017], device='cuda:3'), in_proj_covar=tensor([0.0312, 0.0198, 0.0221, 0.0182, 0.0362, 0.0345, 0.0213, 0.0344], device='cuda:3'), out_proj_covar=tensor([1.5219e-04, 7.9020e-05, 9.7655e-05, 8.4624e-05, 1.6691e-04, 1.4864e-04, 8.6553e-05, 1.5611e-04], device='cuda:3') 2023-03-07 22:05:38,842 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5973, 4.6228, 4.5772, 4.5493, 4.9189, 4.5728, 4.5822, 2.1232], device='cuda:3'), covar=tensor([0.0239, 0.0334, 0.0265, 0.0156, 0.0850, 0.0211, 0.0284, 0.2618], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0123, 0.0123, 0.0121, 0.0297, 0.0125, 0.0114, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 22:06:05,578 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-07 22:06:07,216 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-07 22:06:15,390 INFO [train2.py:809] (3/4) Epoch 7, batch 2800, loss[ctc_loss=0.1386, att_loss=0.2902, loss=0.2599, over 17121.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01494, over 56.00 utterances.], tot_loss[ctc_loss=0.1338, att_loss=0.2661, loss=0.2397, over 3280647.28 frames. utt_duration=1238 frames, utt_pad_proportion=0.05161, over 10613.27 utterances.], batch size: 56, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:06:20,356 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8305, 4.6789, 4.3647, 4.4689, 2.7115, 4.5129, 2.5186, 1.7766], device='cuda:3'), covar=tensor([0.0313, 0.0162, 0.0856, 0.0187, 0.2066, 0.0188, 0.1794, 0.2017], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0103, 0.0247, 0.0111, 0.0218, 0.0102, 0.0224, 0.0203], device='cuda:3'), out_proj_covar=tensor([1.1729e-04, 1.0425e-04, 2.2297e-04, 1.0343e-04, 2.0348e-04, 9.9219e-05, 2.0119e-04, 1.8427e-04], device='cuda:3') 2023-03-07 22:06:31,174 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 3.082e+02 3.761e+02 4.456e+02 1.465e+03, threshold=7.521e+02, percent-clipped=2.0 2023-03-07 22:07:02,421 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-07 22:07:03,605 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-03-07 22:07:34,540 INFO [train2.py:809] (3/4) Epoch 7, batch 2850, loss[ctc_loss=0.1207, att_loss=0.239, loss=0.2153, over 15781.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007353, over 38.00 utterances.], tot_loss[ctc_loss=0.1333, att_loss=0.266, loss=0.2395, over 3285354.85 frames. utt_duration=1250 frames, utt_pad_proportion=0.04706, over 10524.12 utterances.], batch size: 38, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:07:42,522 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:08:01,766 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:08:40,819 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2372, 4.8200, 4.5603, 4.4226, 2.2553, 4.7538, 2.4900, 1.6487], device='cuda:3'), covar=tensor([0.0213, 0.0116, 0.0655, 0.0305, 0.2482, 0.0126, 0.1827, 0.2025], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0103, 0.0251, 0.0114, 0.0223, 0.0102, 0.0227, 0.0203], device='cuda:3'), out_proj_covar=tensor([1.1820e-04, 1.0502e-04, 2.2644e-04, 1.0641e-04, 2.0809e-04, 9.9130e-05, 2.0432e-04, 1.8479e-04], device='cuda:3') 2023-03-07 22:08:54,865 INFO [train2.py:809] (3/4) Epoch 7, batch 2900, loss[ctc_loss=0.1487, att_loss=0.2958, loss=0.2663, over 17042.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009507, over 52.00 utterances.], tot_loss[ctc_loss=0.133, att_loss=0.266, loss=0.2394, over 3289207.44 frames. utt_duration=1250 frames, utt_pad_proportion=0.04672, over 10539.76 utterances.], batch size: 52, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:08:59,637 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:09:11,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 3.082e+02 3.745e+02 4.484e+02 1.236e+03, threshold=7.491e+02, percent-clipped=3.0 2023-03-07 22:09:30,292 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:09:58,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 2023-03-07 22:10:16,147 INFO [train2.py:809] (3/4) Epoch 7, batch 2950, loss[ctc_loss=0.1047, att_loss=0.2384, loss=0.2117, over 15397.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.008988, over 35.00 utterances.], tot_loss[ctc_loss=0.132, att_loss=0.2652, loss=0.2386, over 3283435.62 frames. utt_duration=1263 frames, utt_pad_proportion=0.04682, over 10407.61 utterances.], batch size: 35, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:10:47,880 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:11:11,090 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-07 22:11:36,461 INFO [train2.py:809] (3/4) Epoch 7, batch 3000, loss[ctc_loss=0.1428, att_loss=0.2593, loss=0.236, over 16024.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006351, over 40.00 utterances.], tot_loss[ctc_loss=0.1312, att_loss=0.2647, loss=0.238, over 3276558.23 frames. utt_duration=1271 frames, utt_pad_proportion=0.04674, over 10323.76 utterances.], batch size: 40, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:11:36,461 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 22:11:48,543 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([1.8653, 2.1528, 1.9862, 1.9689, 2.1127, 1.7139, 2.3138, 1.3762], device='cuda:3'), covar=tensor([0.0941, 0.2064, 0.2286, 0.5428, 0.2266, 0.2691, 0.0846, 0.7579], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0082, 0.0087, 0.0136, 0.0078, 0.0123, 0.0072, 0.0127], device='cuda:3'), out_proj_covar=tensor([6.4157e-05, 6.4115e-05, 7.2936e-05, 1.0406e-04, 6.5044e-05, 9.6752e-05, 5.7915e-05, 1.0191e-04], device='cuda:3') 2023-03-07 22:11:50,141 INFO [train2.py:843] (3/4) Epoch 7, validation: ctc_loss=0.06224, att_loss=0.2434, loss=0.2072, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 22:11:50,142 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 22:11:52,098 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:11:55,129 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:12:06,321 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.833e+02 3.331e+02 4.099e+02 1.001e+03, threshold=6.663e+02, percent-clipped=1.0 2023-03-07 22:13:05,217 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:13:05,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 2023-03-07 22:13:06,694 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9735, 3.9951, 3.4437, 3.8243, 4.1020, 3.7492, 2.6546, 4.6067], device='cuda:3'), covar=tensor([0.1029, 0.0367, 0.0916, 0.0571, 0.0496, 0.0654, 0.1012, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0157, 0.0190, 0.0161, 0.0192, 0.0191, 0.0164, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 22:13:09,496 INFO [train2.py:809] (3/4) Epoch 7, batch 3050, loss[ctc_loss=0.08931, att_loss=0.2218, loss=0.1953, over 15616.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.009959, over 37.00 utterances.], tot_loss[ctc_loss=0.1314, att_loss=0.2648, loss=0.2382, over 3278192.44 frames. utt_duration=1264 frames, utt_pad_proportion=0.04866, over 10383.44 utterances.], batch size: 37, lr: 1.52e-02, grad_scale: 8.0 2023-03-07 22:13:11,201 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:13:11,384 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0393, 4.6422, 4.4070, 4.6151, 4.6951, 4.3818, 3.4023, 4.4551], device='cuda:3'), covar=tensor([0.0123, 0.0090, 0.0106, 0.0087, 0.0084, 0.0102, 0.0552, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0068, 0.0044, 0.0045, 0.0056, 0.0080, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 22:13:29,165 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 22:13:54,207 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-07 22:14:29,789 INFO [train2.py:809] (3/4) Epoch 7, batch 3100, loss[ctc_loss=0.1009, att_loss=0.2555, loss=0.2246, over 16326.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006718, over 45.00 utterances.], tot_loss[ctc_loss=0.1323, att_loss=0.2654, loss=0.2388, over 3280294.38 frames. utt_duration=1275 frames, utt_pad_proportion=0.04597, over 10302.32 utterances.], batch size: 45, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:14:43,250 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:14:44,900 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7883, 2.6051, 5.1580, 4.0772, 3.1408, 4.8800, 5.0970, 4.8334], device='cuda:3'), covar=tensor([0.0176, 0.1716, 0.0177, 0.1044, 0.1902, 0.0178, 0.0090, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0242, 0.0125, 0.0302, 0.0284, 0.0178, 0.0107, 0.0142], device='cuda:3'), out_proj_covar=tensor([1.3301e-04, 2.0076e-04, 1.1196e-04, 2.4693e-04, 2.4595e-04, 1.5786e-04, 9.7134e-05, 1.2904e-04], device='cuda:3') 2023-03-07 22:14:45,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.079e+02 3.057e+02 3.764e+02 4.844e+02 1.301e+03, threshold=7.527e+02, percent-clipped=9.0 2023-03-07 22:14:54,058 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3360, 3.1517, 3.1233, 2.7926, 3.2441, 3.0478, 3.4014, 2.0224], device='cuda:3'), covar=tensor([0.1148, 0.1399, 0.3182, 0.6068, 0.1755, 0.4145, 0.0722, 1.1724], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0082, 0.0086, 0.0139, 0.0077, 0.0125, 0.0072, 0.0131], device='cuda:3'), out_proj_covar=tensor([6.3712e-05, 6.4202e-05, 7.2682e-05, 1.0588e-04, 6.4731e-05, 9.7699e-05, 5.8120e-05, 1.0429e-04], device='cuda:3') 2023-03-07 22:15:49,016 INFO [train2.py:809] (3/4) Epoch 7, batch 3150, loss[ctc_loss=0.1705, att_loss=0.2918, loss=0.2676, over 17456.00 frames. utt_duration=1014 frames, utt_pad_proportion=0.04364, over 69.00 utterances.], tot_loss[ctc_loss=0.133, att_loss=0.2658, loss=0.2393, over 3272539.39 frames. utt_duration=1270 frames, utt_pad_proportion=0.04847, over 10316.24 utterances.], batch size: 69, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:16:17,124 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:16:53,655 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5220, 2.3756, 5.0549, 3.7914, 2.8677, 4.5104, 4.8909, 4.7164], device='cuda:3'), covar=tensor([0.0214, 0.1834, 0.0169, 0.1154, 0.2198, 0.0266, 0.0089, 0.0207], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0245, 0.0126, 0.0304, 0.0286, 0.0180, 0.0109, 0.0142], device='cuda:3'), out_proj_covar=tensor([1.3560e-04, 2.0298e-04, 1.1283e-04, 2.4891e-04, 2.4818e-04, 1.5974e-04, 9.8598e-05, 1.2974e-04], device='cuda:3') 2023-03-07 22:17:09,174 INFO [train2.py:809] (3/4) Epoch 7, batch 3200, loss[ctc_loss=0.1092, att_loss=0.2257, loss=0.2024, over 15632.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009199, over 37.00 utterances.], tot_loss[ctc_loss=0.1319, att_loss=0.2651, loss=0.2385, over 3273491.34 frames. utt_duration=1279 frames, utt_pad_proportion=0.04659, over 10248.73 utterances.], batch size: 37, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:17:25,180 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.918e+02 3.542e+02 4.352e+02 6.869e+02, threshold=7.084e+02, percent-clipped=0.0 2023-03-07 22:17:32,945 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:18:14,720 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2793, 5.1175, 5.1534, 2.8685, 4.9015, 4.5984, 4.4863, 2.7967], device='cuda:3'), covar=tensor([0.0089, 0.0089, 0.0148, 0.0940, 0.0104, 0.0150, 0.0257, 0.1157], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0072, 0.0060, 0.0098, 0.0064, 0.0081, 0.0083, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 22:18:21,609 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:18:29,007 INFO [train2.py:809] (3/4) Epoch 7, batch 3250, loss[ctc_loss=0.1225, att_loss=0.261, loss=0.2333, over 16549.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005777, over 45.00 utterances.], tot_loss[ctc_loss=0.1316, att_loss=0.2646, loss=0.238, over 3279639.55 frames. utt_duration=1276 frames, utt_pad_proportion=0.0446, over 10291.15 utterances.], batch size: 45, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:18:30,227 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-07 22:19:26,873 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-07 22:19:48,584 INFO [train2.py:809] (3/4) Epoch 7, batch 3300, loss[ctc_loss=0.1253, att_loss=0.2516, loss=0.2263, over 16267.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007437, over 43.00 utterances.], tot_loss[ctc_loss=0.1321, att_loss=0.2651, loss=0.2385, over 3283076.99 frames. utt_duration=1283 frames, utt_pad_proportion=0.0426, over 10244.28 utterances.], batch size: 43, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:19:59,216 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:20:05,127 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.634e+02 2.966e+02 3.461e+02 4.502e+02 1.067e+03, threshold=6.921e+02, percent-clipped=5.0 2023-03-07 22:20:30,714 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5386, 4.8254, 4.7834, 4.7123, 4.9221, 4.9178, 4.6920, 4.3262], device='cuda:3'), covar=tensor([0.1071, 0.0503, 0.0250, 0.0485, 0.0318, 0.0283, 0.0226, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0413, 0.0246, 0.0190, 0.0234, 0.0288, 0.0304, 0.0236, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 22:21:08,399 INFO [train2.py:809] (3/4) Epoch 7, batch 3350, loss[ctc_loss=0.1049, att_loss=0.2344, loss=0.2085, over 15926.00 frames. utt_duration=1555 frames, utt_pad_proportion=0.008166, over 41.00 utterances.], tot_loss[ctc_loss=0.1313, att_loss=0.2644, loss=0.2378, over 3281121.10 frames. utt_duration=1263 frames, utt_pad_proportion=0.04838, over 10401.15 utterances.], batch size: 41, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:21:20,231 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 22:22:22,586 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2858, 5.1632, 5.0357, 2.2229, 1.8815, 2.8488, 4.3917, 3.8134], device='cuda:3'), covar=tensor([0.0492, 0.0158, 0.0236, 0.4002, 0.6153, 0.2421, 0.0606, 0.1900], device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0196, 0.0221, 0.0183, 0.0357, 0.0344, 0.0217, 0.0345], device='cuda:3'), out_proj_covar=tensor([1.4940e-04, 7.8457e-05, 9.8838e-05, 8.5158e-05, 1.6463e-04, 1.4725e-04, 8.8389e-05, 1.5602e-04], device='cuda:3') 2023-03-07 22:22:28,311 INFO [train2.py:809] (3/4) Epoch 7, batch 3400, loss[ctc_loss=0.2117, att_loss=0.3044, loss=0.2859, over 13560.00 frames. utt_duration=375.5 frames, utt_pad_proportion=0.3469, over 145.00 utterances.], tot_loss[ctc_loss=0.1309, att_loss=0.2639, loss=0.2373, over 3273964.37 frames. utt_duration=1263 frames, utt_pad_proportion=0.051, over 10379.82 utterances.], batch size: 145, lr: 1.51e-02, grad_scale: 8.0 2023-03-07 22:22:33,627 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:22:43,940 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.932e+02 3.516e+02 4.279e+02 1.181e+03, threshold=7.032e+02, percent-clipped=5.0 2023-03-07 22:23:14,995 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9496, 6.1028, 5.4055, 5.9831, 5.7491, 5.3747, 5.6195, 5.3340], device='cuda:3'), covar=tensor([0.0966, 0.0915, 0.0902, 0.0725, 0.0633, 0.1241, 0.2076, 0.2222], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0441, 0.0333, 0.0348, 0.0319, 0.0391, 0.0462, 0.0420], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 22:23:25,612 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9277, 5.3700, 4.8420, 5.4105, 4.7480, 5.1107, 5.5697, 5.2921], device='cuda:3'), covar=tensor([0.0456, 0.0233, 0.0707, 0.0175, 0.0474, 0.0150, 0.0167, 0.0144], device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0209, 0.0267, 0.0193, 0.0217, 0.0164, 0.0195, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 22:23:45,930 INFO [train2.py:809] (3/4) Epoch 7, batch 3450, loss[ctc_loss=0.107, att_loss=0.2636, loss=0.2323, over 16963.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007787, over 50.00 utterances.], tot_loss[ctc_loss=0.1311, att_loss=0.2637, loss=0.2372, over 3271283.99 frames. utt_duration=1267 frames, utt_pad_proportion=0.05115, over 10336.37 utterances.], batch size: 50, lr: 1.51e-02, grad_scale: 16.0 2023-03-07 22:24:40,153 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 2023-03-07 22:25:06,081 INFO [train2.py:809] (3/4) Epoch 7, batch 3500, loss[ctc_loss=0.09174, att_loss=0.2351, loss=0.2065, over 16389.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007515, over 44.00 utterances.], tot_loss[ctc_loss=0.1309, att_loss=0.264, loss=0.2374, over 3277312.61 frames. utt_duration=1278 frames, utt_pad_proportion=0.04621, over 10268.80 utterances.], batch size: 44, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:25:22,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 3.129e+02 3.937e+02 5.112e+02 1.151e+03, threshold=7.873e+02, percent-clipped=8.0 2023-03-07 22:25:52,728 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-07 22:26:26,695 INFO [train2.py:809] (3/4) Epoch 7, batch 3550, loss[ctc_loss=0.1483, att_loss=0.2816, loss=0.255, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01149, over 55.00 utterances.], tot_loss[ctc_loss=0.1302, att_loss=0.2637, loss=0.237, over 3279441.67 frames. utt_duration=1294 frames, utt_pad_proportion=0.04142, over 10146.37 utterances.], batch size: 55, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:27:46,245 INFO [train2.py:809] (3/4) Epoch 7, batch 3600, loss[ctc_loss=0.1119, att_loss=0.2703, loss=0.2387, over 16960.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007947, over 50.00 utterances.], tot_loss[ctc_loss=0.1304, att_loss=0.2642, loss=0.2375, over 3277631.77 frames. utt_duration=1303 frames, utt_pad_proportion=0.04055, over 10076.33 utterances.], batch size: 50, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:27:48,090 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:28:02,318 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.030e+02 2.874e+02 3.597e+02 4.562e+02 1.127e+03, threshold=7.194e+02, percent-clipped=4.0 2023-03-07 22:28:03,204 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3167, 5.0108, 5.1394, 5.1332, 5.0005, 5.2245, 5.0213, 4.7313], device='cuda:3'), covar=tensor([0.1844, 0.0741, 0.0293, 0.0552, 0.0710, 0.0384, 0.0306, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0412, 0.0251, 0.0191, 0.0234, 0.0289, 0.0311, 0.0241, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-07 22:28:32,990 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9049, 5.1444, 5.4181, 5.4046, 5.2695, 5.8644, 5.2044, 5.9799], device='cuda:3'), covar=tensor([0.0615, 0.0624, 0.0603, 0.0852, 0.1651, 0.0713, 0.0476, 0.0591], device='cuda:3'), in_proj_covar=tensor([0.0596, 0.0363, 0.0406, 0.0469, 0.0629, 0.0409, 0.0336, 0.0415], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 22:29:06,203 INFO [train2.py:809] (3/4) Epoch 7, batch 3650, loss[ctc_loss=0.1074, att_loss=0.2486, loss=0.2204, over 16315.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007007, over 45.00 utterances.], tot_loss[ctc_loss=0.1296, att_loss=0.2635, loss=0.2367, over 3265795.73 frames. utt_duration=1290 frames, utt_pad_proportion=0.04757, over 10137.90 utterances.], batch size: 45, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:29:18,200 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:30:27,717 INFO [train2.py:809] (3/4) Epoch 7, batch 3700, loss[ctc_loss=0.08139, att_loss=0.2109, loss=0.185, over 15777.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008243, over 38.00 utterances.], tot_loss[ctc_loss=0.1302, att_loss=0.2636, loss=0.2369, over 3260663.49 frames. utt_duration=1261 frames, utt_pad_proportion=0.05697, over 10357.43 utterances.], batch size: 38, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:30:33,200 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:30:36,301 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 22:30:44,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.774e+02 3.024e+02 4.004e+02 4.895e+02 1.157e+03, threshold=8.007e+02, percent-clipped=6.0 2023-03-07 22:31:47,384 INFO [train2.py:809] (3/4) Epoch 7, batch 3750, loss[ctc_loss=0.1555, att_loss=0.2796, loss=0.2548, over 17353.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03384, over 63.00 utterances.], tot_loss[ctc_loss=0.1303, att_loss=0.2636, loss=0.2369, over 3264045.08 frames. utt_duration=1260 frames, utt_pad_proportion=0.05609, over 10376.56 utterances.], batch size: 63, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:31:49,603 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:32:16,149 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1087, 5.4135, 4.9425, 5.4183, 4.8829, 5.1537, 5.5513, 5.3318], device='cuda:3'), covar=tensor([0.0386, 0.0234, 0.0661, 0.0213, 0.0387, 0.0160, 0.0170, 0.0143], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0204, 0.0262, 0.0188, 0.0215, 0.0162, 0.0192, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-07 22:33:07,743 INFO [train2.py:809] (3/4) Epoch 7, batch 3800, loss[ctc_loss=0.1381, att_loss=0.2773, loss=0.2495, over 17047.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009303, over 52.00 utterances.], tot_loss[ctc_loss=0.13, att_loss=0.264, loss=0.2372, over 3273891.95 frames. utt_duration=1258 frames, utt_pad_proportion=0.05351, over 10422.82 utterances.], batch size: 52, lr: 1.50e-02, grad_scale: 16.0 2023-03-07 22:33:25,170 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 3.032e+02 3.559e+02 4.389e+02 6.627e+02, threshold=7.117e+02, percent-clipped=0.0 2023-03-07 22:33:27,194 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 22:34:04,629 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8995, 5.2001, 5.4574, 5.4203, 5.2021, 5.8330, 5.1812, 5.9357], device='cuda:3'), covar=tensor([0.0697, 0.0711, 0.0620, 0.1088, 0.1890, 0.0867, 0.0493, 0.0610], device='cuda:3'), in_proj_covar=tensor([0.0593, 0.0362, 0.0406, 0.0472, 0.0634, 0.0414, 0.0339, 0.0417], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 22:34:28,357 INFO [train2.py:809] (3/4) Epoch 7, batch 3850, loss[ctc_loss=0.127, att_loss=0.2441, loss=0.2206, over 16013.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006463, over 40.00 utterances.], tot_loss[ctc_loss=0.1288, att_loss=0.263, loss=0.2361, over 3261362.89 frames. utt_duration=1255 frames, utt_pad_proportion=0.05792, over 10404.77 utterances.], batch size: 40, lr: 1.49e-02, grad_scale: 16.0 2023-03-07 22:35:03,360 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 22:35:46,403 INFO [train2.py:809] (3/4) Epoch 7, batch 3900, loss[ctc_loss=0.1234, att_loss=0.2634, loss=0.2354, over 17037.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.008962, over 52.00 utterances.], tot_loss[ctc_loss=0.1291, att_loss=0.2631, loss=0.2363, over 3258954.04 frames. utt_duration=1257 frames, utt_pad_proportion=0.05606, over 10380.53 utterances.], batch size: 52, lr: 1.49e-02, grad_scale: 16.0 2023-03-07 22:35:48,269 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:36:01,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 3.007e+02 3.697e+02 4.485e+02 8.676e+02, threshold=7.394e+02, percent-clipped=2.0 2023-03-07 22:37:02,156 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:37:03,594 INFO [train2.py:809] (3/4) Epoch 7, batch 3950, loss[ctc_loss=0.1221, att_loss=0.2649, loss=0.2364, over 17298.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02472, over 59.00 utterances.], tot_loss[ctc_loss=0.1286, att_loss=0.2626, loss=0.2358, over 3266781.92 frames. utt_duration=1280 frames, utt_pad_proportion=0.04887, over 10217.95 utterances.], batch size: 59, lr: 1.49e-02, grad_scale: 16.0 2023-03-07 22:38:22,294 INFO [train2.py:809] (3/4) Epoch 8, batch 0, loss[ctc_loss=0.1026, att_loss=0.2409, loss=0.2133, over 16126.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006279, over 42.00 utterances.], tot_loss[ctc_loss=0.1026, att_loss=0.2409, loss=0.2133, over 16126.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006279, over 42.00 utterances.], batch size: 42, lr: 1.40e-02, grad_scale: 8.0 2023-03-07 22:38:22,294 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 22:38:34,609 INFO [train2.py:843] (3/4) Epoch 8, validation: ctc_loss=0.06098, att_loss=0.2435, loss=0.207, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 22:38:34,610 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 22:39:19,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.808e+02 3.476e+02 4.336e+02 8.495e+02, threshold=6.952e+02, percent-clipped=4.0 2023-03-07 22:39:31,996 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0591, 2.6497, 3.5558, 2.4220, 3.2939, 4.2521, 4.0494, 2.8987], device='cuda:3'), covar=tensor([0.0397, 0.1627, 0.0883, 0.1572, 0.0946, 0.0619, 0.0518, 0.1414], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0213, 0.0217, 0.0197, 0.0223, 0.0249, 0.0191, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 22:39:54,574 INFO [train2.py:809] (3/4) Epoch 8, batch 50, loss[ctc_loss=0.1006, att_loss=0.2482, loss=0.2187, over 16404.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.00751, over 44.00 utterances.], tot_loss[ctc_loss=0.1268, att_loss=0.2618, loss=0.2348, over 747859.76 frames. utt_duration=1302 frames, utt_pad_proportion=0.02826, over 2300.95 utterances.], batch size: 44, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:41:14,413 INFO [train2.py:809] (3/4) Epoch 8, batch 100, loss[ctc_loss=0.1054, att_loss=0.237, loss=0.2107, over 15877.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009736, over 39.00 utterances.], tot_loss[ctc_loss=0.1288, att_loss=0.2632, loss=0.2363, over 1313538.81 frames. utt_duration=1303 frames, utt_pad_proportion=0.0323, over 4038.17 utterances.], batch size: 39, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:41:27,045 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8715, 1.4956, 1.9442, 1.1178, 3.3105, 2.8061, 0.8384, 1.2364], device='cuda:3'), covar=tensor([0.0834, 0.4069, 0.2941, 0.2801, 0.0591, 0.0950, 0.4694, 0.2835], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0088, 0.0088, 0.0080, 0.0077, 0.0074, 0.0089, 0.0070], device='cuda:3'), out_proj_covar=tensor([4.2125e-05, 5.4393e-05, 5.1854e-05, 4.5086e-05, 3.8825e-05, 4.4667e-05, 5.4000e-05, 4.4718e-05], device='cuda:3') 2023-03-07 22:41:57,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.79 vs. limit=5.0 2023-03-07 22:42:02,729 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.962e+02 3.730e+02 4.479e+02 1.131e+03, threshold=7.459e+02, percent-clipped=8.0 2023-03-07 22:42:14,159 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2195, 4.8165, 4.5599, 4.7142, 2.2809, 4.6559, 2.4377, 1.9209], device='cuda:3'), covar=tensor([0.0220, 0.0100, 0.0637, 0.0147, 0.2264, 0.0151, 0.1845, 0.1939], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0096, 0.0250, 0.0112, 0.0219, 0.0099, 0.0224, 0.0202], device='cuda:3'), out_proj_covar=tensor([1.1862e-04, 1.0043e-04, 2.2668e-04, 1.0504e-04, 2.0656e-04, 9.7455e-05, 2.0246e-04, 1.8493e-04], device='cuda:3') 2023-03-07 22:42:39,310 INFO [train2.py:809] (3/4) Epoch 8, batch 150, loss[ctc_loss=0.1285, att_loss=0.2817, loss=0.251, over 17052.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008119, over 52.00 utterances.], tot_loss[ctc_loss=0.1289, att_loss=0.2651, loss=0.2379, over 1751414.60 frames. utt_duration=1248 frames, utt_pad_proportion=0.04802, over 5620.37 utterances.], batch size: 52, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:43:31,960 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:43:33,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3204, 4.2699, 4.2352, 4.3530, 4.6206, 4.5113, 4.0295, 2.2235], device='cuda:3'), covar=tensor([0.0233, 0.0444, 0.0296, 0.0172, 0.1290, 0.0175, 0.0330, 0.2515], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0122, 0.0120, 0.0123, 0.0296, 0.0122, 0.0110, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 22:43:38,256 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2623, 5.5934, 4.9853, 5.5332, 5.2371, 4.9798, 5.0123, 4.9680], device='cuda:3'), covar=tensor([0.1461, 0.0871, 0.0817, 0.0672, 0.0727, 0.1219, 0.2403, 0.1927], device='cuda:3'), in_proj_covar=tensor([0.0387, 0.0438, 0.0340, 0.0348, 0.0318, 0.0393, 0.0465, 0.0420], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 22:44:00,312 INFO [train2.py:809] (3/4) Epoch 8, batch 200, loss[ctc_loss=0.1029, att_loss=0.2355, loss=0.209, over 14576.00 frames. utt_duration=1824 frames, utt_pad_proportion=0.03924, over 32.00 utterances.], tot_loss[ctc_loss=0.1281, att_loss=0.2641, loss=0.2369, over 2084416.50 frames. utt_duration=1241 frames, utt_pad_proportion=0.05049, over 6729.15 utterances.], batch size: 32, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:44:28,033 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.0246, 3.3237, 3.2638, 2.7592, 3.2119, 3.0805, 2.9300, 2.1435], device='cuda:3'), covar=tensor([0.1462, 0.1288, 0.2329, 0.7502, 0.2661, 0.4236, 0.1260, 1.1223], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0088, 0.0091, 0.0148, 0.0082, 0.0132, 0.0082, 0.0142], device='cuda:3'), out_proj_covar=tensor([6.8150e-05, 6.9711e-05, 7.6967e-05, 1.1391e-04, 6.9410e-05, 1.0490e-04, 6.5901e-05, 1.1291e-04], device='cuda:3') 2023-03-07 22:44:44,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.763e+02 3.553e+02 4.276e+02 6.955e+02, threshold=7.106e+02, percent-clipped=0.0 2023-03-07 22:45:08,551 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5385, 2.7334, 3.6240, 2.6505, 3.4531, 4.6029, 4.2791, 3.2850], device='cuda:3'), covar=tensor([0.0394, 0.1873, 0.1028, 0.1648, 0.1067, 0.0578, 0.0647, 0.1239], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0214, 0.0221, 0.0198, 0.0221, 0.0252, 0.0195, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 22:45:08,686 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5722, 2.3257, 5.0340, 4.0097, 3.0017, 4.4530, 4.8260, 4.7045], device='cuda:3'), covar=tensor([0.0203, 0.1803, 0.0142, 0.0927, 0.1842, 0.0201, 0.0099, 0.0206], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0244, 0.0124, 0.0304, 0.0282, 0.0183, 0.0107, 0.0141], device='cuda:3'), out_proj_covar=tensor([1.3531e-04, 2.0237e-04, 1.1096e-04, 2.4904e-04, 2.4639e-04, 1.6280e-04, 9.6553e-05, 1.2859e-04], device='cuda:3') 2023-03-07 22:45:21,332 INFO [train2.py:809] (3/4) Epoch 8, batch 250, loss[ctc_loss=0.1827, att_loss=0.2926, loss=0.2706, over 14576.00 frames. utt_duration=400.8 frames, utt_pad_proportion=0.3006, over 146.00 utterances.], tot_loss[ctc_loss=0.1281, att_loss=0.2638, loss=0.2366, over 2340274.85 frames. utt_duration=1222 frames, utt_pad_proportion=0.05907, over 7670.39 utterances.], batch size: 146, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:46:41,308 INFO [train2.py:809] (3/4) Epoch 8, batch 300, loss[ctc_loss=0.1402, att_loss=0.2506, loss=0.2285, over 16021.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006997, over 40.00 utterances.], tot_loss[ctc_loss=0.129, att_loss=0.2639, loss=0.2369, over 2549527.69 frames. utt_duration=1231 frames, utt_pad_proportion=0.05739, over 8296.42 utterances.], batch size: 40, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:46:54,324 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-07 22:47:25,882 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3919, 2.3385, 3.1847, 4.1830, 3.8969, 3.9713, 2.5535, 2.1023], device='cuda:3'), covar=tensor([0.0645, 0.2670, 0.1272, 0.0663, 0.0630, 0.0389, 0.2084, 0.2671], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0198, 0.0191, 0.0174, 0.0163, 0.0134, 0.0189, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 22:47:27,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.948e+02 3.357e+02 4.007e+02 1.001e+03, threshold=6.714e+02, percent-clipped=3.0 2023-03-07 22:47:51,064 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:48:03,884 INFO [train2.py:809] (3/4) Epoch 8, batch 350, loss[ctc_loss=0.0996, att_loss=0.2285, loss=0.2027, over 15651.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008497, over 37.00 utterances.], tot_loss[ctc_loss=0.1283, att_loss=0.2636, loss=0.2365, over 2718970.27 frames. utt_duration=1261 frames, utt_pad_proportion=0.04745, over 8636.67 utterances.], batch size: 37, lr: 1.40e-02, grad_scale: 4.0 2023-03-07 22:48:33,655 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:49:25,166 INFO [train2.py:809] (3/4) Epoch 8, batch 400, loss[ctc_loss=0.1448, att_loss=0.2783, loss=0.2516, over 17402.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03387, over 63.00 utterances.], tot_loss[ctc_loss=0.1273, att_loss=0.2633, loss=0.2361, over 2845378.60 frames. utt_duration=1261 frames, utt_pad_proportion=0.04805, over 9033.40 utterances.], batch size: 63, lr: 1.40e-02, grad_scale: 8.0 2023-03-07 22:49:25,534 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:49:30,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:50:01,565 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-07 22:50:09,511 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.601e+02 3.135e+02 4.187e+02 9.825e+02, threshold=6.270e+02, percent-clipped=5.0 2023-03-07 22:50:11,514 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={3} 2023-03-07 22:50:45,305 INFO [train2.py:809] (3/4) Epoch 8, batch 450, loss[ctc_loss=0.137, att_loss=0.2625, loss=0.2374, over 16301.00 frames. utt_duration=1518 frames, utt_pad_proportion=0.00594, over 43.00 utterances.], tot_loss[ctc_loss=0.1259, att_loss=0.2617, loss=0.2346, over 2932351.03 frames. utt_duration=1282 frames, utt_pad_proportion=0.04451, over 9158.12 utterances.], batch size: 43, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:51:03,328 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:51:37,062 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 22:52:04,595 INFO [train2.py:809] (3/4) Epoch 8, batch 500, loss[ctc_loss=0.1213, att_loss=0.2721, loss=0.242, over 16880.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006979, over 49.00 utterances.], tot_loss[ctc_loss=0.1266, att_loss=0.2626, loss=0.2354, over 3009651.78 frames. utt_duration=1270 frames, utt_pad_proportion=0.04785, over 9488.45 utterances.], batch size: 49, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:52:04,798 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9833, 5.1571, 5.4536, 5.4906, 5.2967, 5.9736, 5.0891, 6.0759], device='cuda:3'), covar=tensor([0.0630, 0.0684, 0.0661, 0.0854, 0.1753, 0.0804, 0.0544, 0.0493], device='cuda:3'), in_proj_covar=tensor([0.0608, 0.0367, 0.0421, 0.0477, 0.0643, 0.0427, 0.0345, 0.0424], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-07 22:52:49,177 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.823e+02 3.661e+02 4.475e+02 7.657e+02, threshold=7.322e+02, percent-clipped=5.0 2023-03-07 22:52:54,253 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 22:53:25,909 INFO [train2.py:809] (3/4) Epoch 8, batch 550, loss[ctc_loss=0.1169, att_loss=0.2598, loss=0.2312, over 16390.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007666, over 44.00 utterances.], tot_loss[ctc_loss=0.1263, att_loss=0.2624, loss=0.2352, over 3073168.33 frames. utt_duration=1274 frames, utt_pad_proportion=0.04553, over 9663.24 utterances.], batch size: 44, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:54:43,139 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:54:44,264 INFO [train2.py:809] (3/4) Epoch 8, batch 600, loss[ctc_loss=0.1554, att_loss=0.2851, loss=0.2591, over 17095.00 frames. utt_duration=685.2 frames, utt_pad_proportion=0.1316, over 100.00 utterances.], tot_loss[ctc_loss=0.1257, att_loss=0.2619, loss=0.2347, over 3118777.99 frames. utt_duration=1302 frames, utt_pad_proportion=0.03931, over 9589.15 utterances.], batch size: 100, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:54:52,837 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:55:27,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.636e+02 3.439e+02 4.247e+02 1.142e+03, threshold=6.879e+02, percent-clipped=2.0 2023-03-07 22:55:57,580 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5526, 4.8523, 4.3426, 4.9852, 4.3439, 4.7051, 4.9608, 4.8222], device='cuda:3'), covar=tensor([0.0465, 0.0240, 0.0794, 0.0178, 0.0405, 0.0225, 0.0253, 0.0165], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0216, 0.0280, 0.0203, 0.0226, 0.0174, 0.0203, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 22:56:03,435 INFO [train2.py:809] (3/4) Epoch 8, batch 650, loss[ctc_loss=0.1255, att_loss=0.2855, loss=0.2535, over 17419.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.03205, over 63.00 utterances.], tot_loss[ctc_loss=0.1266, att_loss=0.2621, loss=0.235, over 3147362.10 frames. utt_duration=1276 frames, utt_pad_proportion=0.04766, over 9876.51 utterances.], batch size: 63, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:56:19,712 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:56:21,103 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7666, 5.0889, 4.4970, 5.1775, 4.5431, 4.8541, 5.2646, 5.0611], device='cuda:3'), covar=tensor([0.0501, 0.0329, 0.0930, 0.0218, 0.0498, 0.0255, 0.0213, 0.0204], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0217, 0.0282, 0.0204, 0.0228, 0.0175, 0.0203, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 22:56:29,214 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:57:21,176 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 22:57:24,654 INFO [train2.py:809] (3/4) Epoch 8, batch 700, loss[ctc_loss=0.1317, att_loss=0.2759, loss=0.247, over 16956.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007991, over 50.00 utterances.], tot_loss[ctc_loss=0.1281, att_loss=0.2635, loss=0.2364, over 3181282.97 frames. utt_duration=1252 frames, utt_pad_proportion=0.05106, over 10175.24 utterances.], batch size: 50, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:57:39,089 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4184, 2.6012, 3.2433, 4.5382, 4.0532, 4.2277, 2.9176, 2.0074], device='cuda:3'), covar=tensor([0.0633, 0.2270, 0.1304, 0.0434, 0.0523, 0.0269, 0.1577, 0.2817], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0197, 0.0190, 0.0175, 0.0165, 0.0134, 0.0190, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 22:58:02,981 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 22:58:08,929 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.921e+02 3.565e+02 4.512e+02 9.961e+02, threshold=7.130e+02, percent-clipped=4.0 2023-03-07 22:58:44,786 INFO [train2.py:809] (3/4) Epoch 8, batch 750, loss[ctc_loss=0.152, att_loss=0.2755, loss=0.2508, over 16870.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.008222, over 49.00 utterances.], tot_loss[ctc_loss=0.1286, att_loss=0.2638, loss=0.2368, over 3208886.48 frames. utt_duration=1247 frames, utt_pad_proportion=0.05185, over 10301.92 utterances.], batch size: 49, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 22:58:51,919 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9498, 5.3276, 4.7192, 5.3955, 4.6870, 4.9654, 5.4978, 5.2815], device='cuda:3'), covar=tensor([0.0462, 0.0229, 0.0947, 0.0185, 0.0415, 0.0219, 0.0165, 0.0143], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0212, 0.0280, 0.0201, 0.0226, 0.0174, 0.0199, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 22:58:55,032 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:00:04,389 INFO [train2.py:809] (3/4) Epoch 8, batch 800, loss[ctc_loss=0.1365, att_loss=0.2642, loss=0.2387, over 16964.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007714, over 50.00 utterances.], tot_loss[ctc_loss=0.1275, att_loss=0.263, loss=0.2359, over 3225115.95 frames. utt_duration=1270 frames, utt_pad_proportion=0.04787, over 10168.44 utterances.], batch size: 50, lr: 1.39e-02, grad_scale: 8.0 2023-03-07 23:00:47,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.816e+02 3.750e+02 4.467e+02 1.351e+03, threshold=7.500e+02, percent-clipped=6.0 2023-03-07 23:01:24,435 INFO [train2.py:809] (3/4) Epoch 8, batch 850, loss[ctc_loss=0.1154, att_loss=0.2651, loss=0.2352, over 17433.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.03139, over 63.00 utterances.], tot_loss[ctc_loss=0.1272, att_loss=0.2628, loss=0.2357, over 3234473.87 frames. utt_duration=1242 frames, utt_pad_proportion=0.05525, over 10430.65 utterances.], batch size: 63, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:01:58,202 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0334, 5.3982, 5.2229, 5.3329, 5.4621, 5.4877, 5.2022, 4.9250], device='cuda:3'), covar=tensor([0.1069, 0.0469, 0.0249, 0.0504, 0.0268, 0.0244, 0.0253, 0.0320], device='cuda:3'), in_proj_covar=tensor([0.0429, 0.0262, 0.0205, 0.0246, 0.0304, 0.0330, 0.0250, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:02:23,660 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0201, 4.0316, 3.9467, 2.5825, 3.8622, 3.7141, 3.5641, 2.5268], device='cuda:3'), covar=tensor([0.0107, 0.0101, 0.0160, 0.1025, 0.0099, 0.0292, 0.0305, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0073, 0.0062, 0.0099, 0.0064, 0.0082, 0.0084, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:02:43,216 INFO [train2.py:809] (3/4) Epoch 8, batch 900, loss[ctc_loss=0.1268, att_loss=0.2456, loss=0.2219, over 15617.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.01068, over 37.00 utterances.], tot_loss[ctc_loss=0.1268, att_loss=0.2626, loss=0.2354, over 3237866.40 frames. utt_duration=1246 frames, utt_pad_proportion=0.05734, over 10406.76 utterances.], batch size: 37, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:03:07,213 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2965, 5.1363, 5.1286, 2.2104, 1.8031, 2.5025, 3.9848, 3.8926], device='cuda:3'), covar=tensor([0.0496, 0.0227, 0.0255, 0.3348, 0.6413, 0.3008, 0.0855, 0.1800], device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0195, 0.0214, 0.0177, 0.0345, 0.0328, 0.0209, 0.0335], device='cuda:3'), out_proj_covar=tensor([1.4693e-04, 7.6287e-05, 9.3935e-05, 8.1544e-05, 1.5746e-04, 1.3969e-04, 8.4958e-05, 1.4991e-04], device='cuda:3') 2023-03-07 23:03:27,129 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.976e+02 3.721e+02 4.391e+02 9.491e+02, threshold=7.443e+02, percent-clipped=5.0 2023-03-07 23:03:45,719 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-03-07 23:03:59,496 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3754, 1.9116, 1.8555, 1.3838, 3.3453, 1.7060, 1.6838, 1.2793], device='cuda:3'), covar=tensor([0.1585, 0.2931, 0.2552, 0.2238, 0.0424, 0.2059, 0.3207, 0.2790], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0079, 0.0083, 0.0074, 0.0071, 0.0071, 0.0080, 0.0065], device='cuda:3'), out_proj_covar=tensor([4.0104e-05, 5.0068e-05, 4.9575e-05, 4.2616e-05, 3.7298e-05, 4.3312e-05, 4.9441e-05, 4.1775e-05], device='cuda:3') 2023-03-07 23:04:03,682 INFO [train2.py:809] (3/4) Epoch 8, batch 950, loss[ctc_loss=0.1014, att_loss=0.2337, loss=0.2072, over 15650.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.007948, over 37.00 utterances.], tot_loss[ctc_loss=0.1275, att_loss=0.2635, loss=0.2363, over 3249259.56 frames. utt_duration=1224 frames, utt_pad_proportion=0.06111, over 10632.18 utterances.], batch size: 37, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:04:11,996 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:04:21,429 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:04:42,086 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:05:21,492 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:05:24,289 INFO [train2.py:809] (3/4) Epoch 8, batch 1000, loss[ctc_loss=0.1256, att_loss=0.2478, loss=0.2234, over 15860.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01075, over 39.00 utterances.], tot_loss[ctc_loss=0.1268, att_loss=0.2626, loss=0.2354, over 3248807.15 frames. utt_duration=1248 frames, utt_pad_proportion=0.05639, over 10426.56 utterances.], batch size: 39, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:05:25,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-07 23:05:50,769 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0873, 5.1272, 4.9350, 2.4308, 1.9440, 2.7702, 3.9447, 3.8329], device='cuda:3'), covar=tensor([0.0625, 0.0172, 0.0235, 0.3569, 0.6200, 0.2520, 0.1050, 0.1951], device='cuda:3'), in_proj_covar=tensor([0.0316, 0.0200, 0.0222, 0.0183, 0.0356, 0.0339, 0.0216, 0.0347], device='cuda:3'), out_proj_covar=tensor([1.5134e-04, 7.8332e-05, 9.8272e-05, 8.4838e-05, 1.6242e-04, 1.4415e-04, 8.7860e-05, 1.5530e-04], device='cuda:3') 2023-03-07 23:06:01,453 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={2} 2023-03-07 23:06:01,599 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5470, 2.5137, 5.0437, 3.6524, 2.8345, 4.4268, 4.7496, 4.6748], device='cuda:3'), covar=tensor([0.0236, 0.1732, 0.0136, 0.1381, 0.2095, 0.0215, 0.0099, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0239, 0.0121, 0.0299, 0.0277, 0.0180, 0.0104, 0.0139], device='cuda:3'), out_proj_covar=tensor([1.3311e-04, 1.9815e-04, 1.0850e-04, 2.4511e-04, 2.4311e-04, 1.5989e-04, 9.4102e-05, 1.2916e-04], device='cuda:3') 2023-03-07 23:06:07,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.707e+02 2.634e+02 3.393e+02 4.335e+02 1.819e+03, threshold=6.785e+02, percent-clipped=6.0 2023-03-07 23:06:20,126 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:06:31,397 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8008, 4.8286, 4.7542, 4.7381, 5.2005, 4.9510, 4.5296, 2.2698], device='cuda:3'), covar=tensor([0.0189, 0.0281, 0.0264, 0.0214, 0.1117, 0.0175, 0.0352, 0.2385], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0123, 0.0126, 0.0125, 0.0310, 0.0124, 0.0117, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 23:06:37,056 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:06:43,111 INFO [train2.py:809] (3/4) Epoch 8, batch 1050, loss[ctc_loss=0.1257, att_loss=0.2754, loss=0.2455, over 17317.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02365, over 59.00 utterances.], tot_loss[ctc_loss=0.1278, att_loss=0.263, loss=0.236, over 3257133.49 frames. utt_duration=1231 frames, utt_pad_proportion=0.05862, over 10600.57 utterances.], batch size: 59, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:06:53,378 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:07:18,035 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 23:07:46,827 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2238, 3.4964, 3.4501, 2.9854, 3.6082, 3.5031, 3.3063, 2.0525], device='cuda:3'), covar=tensor([0.1327, 0.1082, 0.2058, 0.5496, 0.0855, 0.3004, 0.0695, 0.9621], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0088, 0.0091, 0.0144, 0.0081, 0.0134, 0.0079, 0.0138], device='cuda:3'), out_proj_covar=tensor([6.8336e-05, 6.9978e-05, 7.7938e-05, 1.1187e-04, 6.9611e-05, 1.0673e-04, 6.5083e-05, 1.1043e-04], device='cuda:3') 2023-03-07 23:08:03,727 INFO [train2.py:809] (3/4) Epoch 8, batch 1100, loss[ctc_loss=0.0925, att_loss=0.2466, loss=0.2158, over 16121.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006541, over 42.00 utterances.], tot_loss[ctc_loss=0.1272, att_loss=0.263, loss=0.2358, over 3269642.29 frames. utt_duration=1240 frames, utt_pad_proportion=0.05394, over 10558.41 utterances.], batch size: 42, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:08:10,622 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:08:20,266 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9149, 4.8660, 4.7522, 2.7858, 4.7188, 4.2995, 4.1467, 2.3152], device='cuda:3'), covar=tensor([0.0111, 0.0083, 0.0222, 0.0971, 0.0079, 0.0188, 0.0320, 0.1471], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0074, 0.0062, 0.0098, 0.0064, 0.0082, 0.0084, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:08:28,624 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:08:48,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.718e+02 3.264e+02 3.994e+02 7.593e+02, threshold=6.528e+02, percent-clipped=2.0 2023-03-07 23:09:18,213 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-07 23:09:25,645 INFO [train2.py:809] (3/4) Epoch 8, batch 1150, loss[ctc_loss=0.1078, att_loss=0.256, loss=0.2264, over 16337.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005609, over 45.00 utterances.], tot_loss[ctc_loss=0.126, att_loss=0.2622, loss=0.235, over 3269653.31 frames. utt_duration=1247 frames, utt_pad_proportion=0.05367, over 10503.09 utterances.], batch size: 45, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:09:32,983 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-07 23:09:36,941 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7394, 2.6711, 5.1291, 4.0496, 3.1837, 4.7229, 4.9607, 4.9143], device='cuda:3'), covar=tensor([0.0203, 0.1626, 0.0163, 0.0935, 0.1767, 0.0193, 0.0095, 0.0168], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0239, 0.0122, 0.0298, 0.0277, 0.0180, 0.0105, 0.0140], device='cuda:3'), out_proj_covar=tensor([1.3386e-04, 1.9932e-04, 1.0940e-04, 2.4520e-04, 2.4360e-04, 1.5939e-04, 9.4565e-05, 1.2966e-04], device='cuda:3') 2023-03-07 23:10:06,621 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:10:46,211 INFO [train2.py:809] (3/4) Epoch 8, batch 1200, loss[ctc_loss=0.195, att_loss=0.3014, loss=0.2801, over 13699.00 frames. utt_duration=379.2 frames, utt_pad_proportion=0.3416, over 145.00 utterances.], tot_loss[ctc_loss=0.1253, att_loss=0.2616, loss=0.2343, over 3267352.97 frames. utt_duration=1236 frames, utt_pad_proportion=0.05816, over 10590.11 utterances.], batch size: 145, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:11:30,217 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.794e+02 3.105e+02 3.921e+02 8.857e+02, threshold=6.210e+02, percent-clipped=2.0 2023-03-07 23:11:56,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2266, 3.9500, 3.1135, 3.7364, 4.1162, 3.8344, 2.8804, 4.6026], device='cuda:3'), covar=tensor([0.1021, 0.0472, 0.1277, 0.0526, 0.0669, 0.0644, 0.1004, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0160, 0.0194, 0.0163, 0.0201, 0.0196, 0.0169, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:12:06,184 INFO [train2.py:809] (3/4) Epoch 8, batch 1250, loss[ctc_loss=0.09921, att_loss=0.2441, loss=0.2151, over 16530.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006872, over 45.00 utterances.], tot_loss[ctc_loss=0.1262, att_loss=0.2624, loss=0.2352, over 3267465.51 frames. utt_duration=1216 frames, utt_pad_proportion=0.06191, over 10761.47 utterances.], batch size: 45, lr: 1.38e-02, grad_scale: 8.0 2023-03-07 23:12:14,243 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:12:23,536 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:13:25,931 INFO [train2.py:809] (3/4) Epoch 8, batch 1300, loss[ctc_loss=0.1316, att_loss=0.2577, loss=0.2325, over 15999.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.006452, over 40.00 utterances.], tot_loss[ctc_loss=0.1258, att_loss=0.2616, loss=0.2345, over 3257976.76 frames. utt_duration=1234 frames, utt_pad_proportion=0.059, over 10577.79 utterances.], batch size: 40, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:13:30,547 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:13:39,857 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:14:09,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.738e+02 3.321e+02 4.360e+02 1.126e+03, threshold=6.642e+02, percent-clipped=8.0 2023-03-07 23:14:13,598 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:14:44,864 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:14:46,056 INFO [train2.py:809] (3/4) Epoch 8, batch 1350, loss[ctc_loss=0.1136, att_loss=0.24, loss=0.2148, over 14151.00 frames. utt_duration=1827 frames, utt_pad_proportion=0.04782, over 31.00 utterances.], tot_loss[ctc_loss=0.126, att_loss=0.2616, loss=0.2345, over 3263881.49 frames. utt_duration=1254 frames, utt_pad_proportion=0.05251, over 10421.30 utterances.], batch size: 31, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:15:28,206 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7836, 6.0195, 5.3316, 5.9393, 5.7071, 5.3124, 5.4361, 5.1896], device='cuda:3'), covar=tensor([0.1098, 0.0918, 0.0849, 0.0703, 0.0579, 0.1413, 0.2505, 0.2396], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0440, 0.0337, 0.0345, 0.0323, 0.0390, 0.0465, 0.0416], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 23:16:06,780 INFO [train2.py:809] (3/4) Epoch 8, batch 1400, loss[ctc_loss=0.2028, att_loss=0.3088, loss=0.2876, over 13920.00 frames. utt_duration=380.2 frames, utt_pad_proportion=0.333, over 147.00 utterances.], tot_loss[ctc_loss=0.1255, att_loss=0.2615, loss=0.2343, over 3265566.97 frames. utt_duration=1261 frames, utt_pad_proportion=0.05031, over 10370.66 utterances.], batch size: 147, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:16:23,187 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:16:52,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.810e+02 3.377e+02 4.369e+02 8.474e+02, threshold=6.754e+02, percent-clipped=4.0 2023-03-07 23:17:27,562 INFO [train2.py:809] (3/4) Epoch 8, batch 1450, loss[ctc_loss=0.1231, att_loss=0.2454, loss=0.221, over 11856.00 frames. utt_duration=1826 frames, utt_pad_proportion=0.04412, over 26.00 utterances.], tot_loss[ctc_loss=0.125, att_loss=0.2612, loss=0.234, over 3262980.49 frames. utt_duration=1267 frames, utt_pad_proportion=0.04919, over 10309.59 utterances.], batch size: 26, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:17:48,693 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-07 23:18:00,635 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:18:18,195 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0054, 5.1033, 5.0271, 2.5184, 4.8850, 4.5770, 4.3371, 2.0169], device='cuda:3'), covar=tensor([0.0187, 0.0087, 0.0149, 0.1342, 0.0090, 0.0154, 0.0374, 0.2229], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0073, 0.0060, 0.0097, 0.0063, 0.0081, 0.0083, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:18:47,770 INFO [train2.py:809] (3/4) Epoch 8, batch 1500, loss[ctc_loss=0.1246, att_loss=0.2723, loss=0.2428, over 16771.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006392, over 48.00 utterances.], tot_loss[ctc_loss=0.1252, att_loss=0.2615, loss=0.2342, over 3264168.97 frames. utt_duration=1239 frames, utt_pad_proportion=0.0574, over 10546.84 utterances.], batch size: 48, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:19:30,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.592e+02 3.615e+02 4.521e+02 1.886e+03, threshold=7.229e+02, percent-clipped=4.0 2023-03-07 23:19:39,586 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6455, 4.5057, 4.5119, 4.5330, 5.0763, 4.8434, 4.3718, 2.1303], device='cuda:3'), covar=tensor([0.0245, 0.0366, 0.0240, 0.0174, 0.1033, 0.0194, 0.0314, 0.2599], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0123, 0.0128, 0.0126, 0.0309, 0.0123, 0.0114, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-07 23:20:07,839 INFO [train2.py:809] (3/4) Epoch 8, batch 1550, loss[ctc_loss=0.1369, att_loss=0.2463, loss=0.2244, over 15493.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009137, over 36.00 utterances.], tot_loss[ctc_loss=0.1254, att_loss=0.2612, loss=0.2341, over 3266574.86 frames. utt_duration=1274 frames, utt_pad_proportion=0.04914, over 10266.41 utterances.], batch size: 36, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:21:28,331 INFO [train2.py:809] (3/4) Epoch 8, batch 1600, loss[ctc_loss=0.1126, att_loss=0.2582, loss=0.2291, over 16629.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005095, over 47.00 utterances.], tot_loss[ctc_loss=0.1248, att_loss=0.261, loss=0.2338, over 3267550.09 frames. utt_duration=1262 frames, utt_pad_proportion=0.05357, over 10372.68 utterances.], batch size: 47, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:22:11,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.839e+02 3.566e+02 4.248e+02 8.470e+02, threshold=7.133e+02, percent-clipped=2.0 2023-03-07 23:22:16,282 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:22:48,443 INFO [train2.py:809] (3/4) Epoch 8, batch 1650, loss[ctc_loss=0.08929, att_loss=0.2223, loss=0.1957, over 12779.00 frames. utt_duration=1827 frames, utt_pad_proportion=0.1203, over 28.00 utterances.], tot_loss[ctc_loss=0.1248, att_loss=0.2615, loss=0.2342, over 3264579.45 frames. utt_duration=1249 frames, utt_pad_proportion=0.05745, over 10467.63 utterances.], batch size: 28, lr: 1.37e-02, grad_scale: 8.0 2023-03-07 23:23:10,606 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:23:33,457 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:24:08,552 INFO [train2.py:809] (3/4) Epoch 8, batch 1700, loss[ctc_loss=0.1304, att_loss=0.2639, loss=0.2372, over 16128.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.005606, over 42.00 utterances.], tot_loss[ctc_loss=0.1235, att_loss=0.2609, loss=0.2335, over 3261413.91 frames. utt_duration=1285 frames, utt_pad_proportion=0.04907, over 10162.49 utterances.], batch size: 42, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:24:11,302 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-07 23:24:16,290 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:24:47,870 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:24:52,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-07 23:24:52,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.662e+02 3.188e+02 4.272e+02 1.218e+03, threshold=6.376e+02, percent-clipped=1.0 2023-03-07 23:25:28,580 INFO [train2.py:809] (3/4) Epoch 8, batch 1750, loss[ctc_loss=0.1212, att_loss=0.263, loss=0.2346, over 16331.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006141, over 45.00 utterances.], tot_loss[ctc_loss=0.1255, att_loss=0.2626, loss=0.2352, over 3269620.09 frames. utt_duration=1233 frames, utt_pad_proportion=0.06033, over 10616.60 utterances.], batch size: 45, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:25:33,782 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9144, 2.5645, 3.4596, 2.5614, 3.2459, 4.1668, 3.8876, 2.9459], device='cuda:3'), covar=tensor([0.0490, 0.1816, 0.1028, 0.1474, 0.1052, 0.0741, 0.0711, 0.1376], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0217, 0.0223, 0.0198, 0.0224, 0.0260, 0.0196, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-07 23:26:01,605 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:26:08,399 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:26:28,325 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0997, 5.0327, 5.0227, 2.2286, 1.8194, 2.8245, 4.2743, 3.5579], device='cuda:3'), covar=tensor([0.0630, 0.0169, 0.0177, 0.3613, 0.6360, 0.2526, 0.0603, 0.2175], device='cuda:3'), in_proj_covar=tensor([0.0316, 0.0201, 0.0226, 0.0186, 0.0357, 0.0339, 0.0215, 0.0344], device='cuda:3'), out_proj_covar=tensor([1.5050e-04, 7.9329e-05, 1.0104e-04, 8.6875e-05, 1.6193e-04, 1.4405e-04, 8.7004e-05, 1.5377e-04], device='cuda:3') 2023-03-07 23:26:48,814 INFO [train2.py:809] (3/4) Epoch 8, batch 1800, loss[ctc_loss=0.1562, att_loss=0.2886, loss=0.2621, over 16892.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006215, over 49.00 utterances.], tot_loss[ctc_loss=0.1255, att_loss=0.2621, loss=0.2348, over 3268634.98 frames. utt_duration=1252 frames, utt_pad_proportion=0.05562, over 10452.84 utterances.], batch size: 49, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:27:19,037 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:27:33,855 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.769e+02 3.006e+02 3.671e+02 4.359e+02 7.069e+02, threshold=7.341e+02, percent-clipped=3.0 2023-03-07 23:27:47,157 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 23:28:08,901 INFO [train2.py:809] (3/4) Epoch 8, batch 1850, loss[ctc_loss=0.1336, att_loss=0.2499, loss=0.2267, over 15954.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007057, over 41.00 utterances.], tot_loss[ctc_loss=0.1252, att_loss=0.2614, loss=0.2342, over 3269996.91 frames. utt_duration=1263 frames, utt_pad_proportion=0.05103, over 10369.68 utterances.], batch size: 41, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:28:15,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5110, 2.3767, 3.2982, 4.2656, 4.0329, 4.0037, 2.7511, 1.9224], device='cuda:3'), covar=tensor([0.0690, 0.2669, 0.1069, 0.0584, 0.0562, 0.0351, 0.1642, 0.2712], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0193, 0.0181, 0.0173, 0.0164, 0.0132, 0.0185, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 23:28:55,950 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2839, 5.1803, 5.1073, 2.9732, 4.9749, 4.7202, 4.4058, 2.6316], device='cuda:3'), covar=tensor([0.0111, 0.0068, 0.0161, 0.0973, 0.0087, 0.0143, 0.0313, 0.1378], device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0074, 0.0061, 0.0099, 0.0064, 0.0083, 0.0085, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:29:29,189 INFO [train2.py:809] (3/4) Epoch 8, batch 1900, loss[ctc_loss=0.1077, att_loss=0.2437, loss=0.2165, over 14570.00 frames. utt_duration=1823 frames, utt_pad_proportion=0.03263, over 32.00 utterances.], tot_loss[ctc_loss=0.124, att_loss=0.2606, loss=0.2333, over 3267229.75 frames. utt_duration=1282 frames, utt_pad_proportion=0.04785, over 10202.35 utterances.], batch size: 32, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:29:53,375 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:30:06,152 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8589, 6.1143, 5.6472, 5.9144, 5.7961, 5.4920, 5.5314, 5.3961], device='cuda:3'), covar=tensor([0.1134, 0.0775, 0.0636, 0.0674, 0.0636, 0.1145, 0.2109, 0.1932], device='cuda:3'), in_proj_covar=tensor([0.0387, 0.0444, 0.0338, 0.0351, 0.0324, 0.0395, 0.0466, 0.0421], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-07 23:30:14,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.848e+02 3.489e+02 4.183e+02 9.529e+02, threshold=6.977e+02, percent-clipped=3.0 2023-03-07 23:30:36,207 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6421, 5.0308, 4.8223, 4.7795, 5.0596, 5.0244, 4.8169, 4.4338], device='cuda:3'), covar=tensor([0.1050, 0.0436, 0.0301, 0.0532, 0.0290, 0.0307, 0.0244, 0.0353], device='cuda:3'), in_proj_covar=tensor([0.0430, 0.0263, 0.0208, 0.0247, 0.0312, 0.0332, 0.0253, 0.0286], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:30:49,806 INFO [train2.py:809] (3/4) Epoch 8, batch 1950, loss[ctc_loss=0.1096, att_loss=0.2377, loss=0.2121, over 16180.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006397, over 41.00 utterances.], tot_loss[ctc_loss=0.1232, att_loss=0.2597, loss=0.2324, over 3266251.82 frames. utt_duration=1309 frames, utt_pad_proportion=0.04175, over 9989.97 utterances.], batch size: 41, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:31:25,705 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-07 23:31:31,479 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-07 23:31:32,165 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:32:10,087 INFO [train2.py:809] (3/4) Epoch 8, batch 2000, loss[ctc_loss=0.1147, att_loss=0.2283, loss=0.2056, over 15740.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.009181, over 38.00 utterances.], tot_loss[ctc_loss=0.1223, att_loss=0.2589, loss=0.2316, over 3261769.31 frames. utt_duration=1304 frames, utt_pad_proportion=0.04495, over 10018.15 utterances.], batch size: 38, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:32:12,129 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7415, 2.5198, 3.9174, 3.5149, 2.9230, 3.8126, 3.6739, 3.8339], device='cuda:3'), covar=tensor([0.0212, 0.1250, 0.0106, 0.0862, 0.1584, 0.0271, 0.0140, 0.0218], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0245, 0.0125, 0.0304, 0.0286, 0.0186, 0.0110, 0.0148], device='cuda:3'), out_proj_covar=tensor([1.4054e-04, 2.0466e-04, 1.1105e-04, 2.5084e-04, 2.5183e-04, 1.6499e-04, 9.9600e-05, 1.3738e-04], device='cuda:3') 2023-03-07 23:32:18,053 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:32:28,928 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4478, 3.5149, 2.9778, 3.2589, 3.6438, 3.3200, 2.4252, 3.9316], device='cuda:3'), covar=tensor([0.1118, 0.0446, 0.0913, 0.0636, 0.0604, 0.0661, 0.1019, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0160, 0.0193, 0.0164, 0.0204, 0.0194, 0.0168, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:32:29,683 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-07 23:32:41,922 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:32:54,604 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.836e+02 3.233e+02 3.942e+02 8.276e+02, threshold=6.465e+02, percent-clipped=1.0 2023-03-07 23:33:30,286 INFO [train2.py:809] (3/4) Epoch 8, batch 2050, loss[ctc_loss=0.1313, att_loss=0.2761, loss=0.2472, over 17062.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008403, over 52.00 utterances.], tot_loss[ctc_loss=0.1223, att_loss=0.2588, loss=0.2315, over 3261078.52 frames. utt_duration=1286 frames, utt_pad_proportion=0.04791, over 10155.92 utterances.], batch size: 52, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:33:34,866 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:34:03,374 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3497, 4.8966, 4.7133, 4.8662, 4.7681, 4.5686, 3.8294, 4.7277], device='cuda:3'), covar=tensor([0.0114, 0.0123, 0.0109, 0.0104, 0.0126, 0.0115, 0.0445, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0061, 0.0071, 0.0046, 0.0049, 0.0058, 0.0083, 0.0082], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-07 23:34:16,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0456, 4.8760, 4.2784, 4.4935, 2.2218, 4.7818, 2.1297, 1.7577], device='cuda:3'), covar=tensor([0.0298, 0.0095, 0.0781, 0.0216, 0.2317, 0.0131, 0.1894, 0.1757], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0098, 0.0250, 0.0110, 0.0220, 0.0099, 0.0226, 0.0198], device='cuda:3'), out_proj_covar=tensor([1.2230e-04, 1.0188e-04, 2.2800e-04, 1.0500e-04, 2.0845e-04, 9.7805e-05, 2.0514e-04, 1.8132e-04], device='cuda:3') 2023-03-07 23:34:51,161 INFO [train2.py:809] (3/4) Epoch 8, batch 2100, loss[ctc_loss=0.1079, att_loss=0.2528, loss=0.2238, over 16280.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007402, over 43.00 utterances.], tot_loss[ctc_loss=0.1213, att_loss=0.2586, loss=0.2311, over 3265212.75 frames. utt_duration=1287 frames, utt_pad_proportion=0.04645, over 10156.93 utterances.], batch size: 43, lr: 1.36e-02, grad_scale: 8.0 2023-03-07 23:35:43,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.671e+02 3.420e+02 4.365e+02 1.034e+03, threshold=6.841e+02, percent-clipped=5.0 2023-03-07 23:35:47,043 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 23:36:11,985 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5448, 1.7071, 1.7416, 1.5977, 2.6182, 1.8021, 1.5527, 2.2434], device='cuda:3'), covar=tensor([0.0531, 0.2862, 0.2376, 0.1773, 0.0761, 0.1729, 0.2709, 0.1330], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0083, 0.0086, 0.0076, 0.0074, 0.0070, 0.0081, 0.0064], device='cuda:3'), out_proj_covar=tensor([4.1060e-05, 5.2021e-05, 5.2259e-05, 4.4268e-05, 4.0067e-05, 4.4696e-05, 5.0748e-05, 4.2130e-05], device='cuda:3') 2023-03-07 23:36:16,106 INFO [train2.py:809] (3/4) Epoch 8, batch 2150, loss[ctc_loss=0.1067, att_loss=0.2404, loss=0.2137, over 10573.00 frames. utt_duration=1840 frames, utt_pad_proportion=0.2269, over 23.00 utterances.], tot_loss[ctc_loss=0.1226, att_loss=0.2597, loss=0.2323, over 3269885.21 frames. utt_duration=1271 frames, utt_pad_proportion=0.04765, over 10303.26 utterances.], batch size: 23, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:36:34,200 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1705, 5.0682, 4.9461, 2.5923, 1.9716, 2.7208, 3.9057, 3.6314], device='cuda:3'), covar=tensor([0.0569, 0.0153, 0.0196, 0.2978, 0.6082, 0.2731, 0.0889, 0.1999], device='cuda:3'), in_proj_covar=tensor([0.0318, 0.0200, 0.0226, 0.0188, 0.0353, 0.0339, 0.0216, 0.0344], device='cuda:3'), out_proj_covar=tensor([1.5200e-04, 7.9584e-05, 1.0125e-04, 8.7611e-05, 1.6065e-04, 1.4387e-04, 8.7122e-05, 1.5419e-04], device='cuda:3') 2023-03-07 23:37:11,487 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6616, 2.4989, 4.9796, 3.9152, 3.0511, 4.6615, 4.8333, 4.7257], device='cuda:3'), covar=tensor([0.0193, 0.1735, 0.0139, 0.1027, 0.1944, 0.0186, 0.0089, 0.0197], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0241, 0.0125, 0.0301, 0.0283, 0.0182, 0.0108, 0.0145], device='cuda:3'), out_proj_covar=tensor([1.3771e-04, 2.0123e-04, 1.1110e-04, 2.4856e-04, 2.4979e-04, 1.6156e-04, 9.8091e-05, 1.3490e-04], device='cuda:3') 2023-03-07 23:37:36,071 INFO [train2.py:809] (3/4) Epoch 8, batch 2200, loss[ctc_loss=0.146, att_loss=0.2564, loss=0.2344, over 16387.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007651, over 44.00 utterances.], tot_loss[ctc_loss=0.123, att_loss=0.2603, loss=0.2329, over 3271694.36 frames. utt_duration=1267 frames, utt_pad_proportion=0.04962, over 10339.86 utterances.], batch size: 44, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:38:22,433 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.675e+02 3.232e+02 4.079e+02 9.910e+02, threshold=6.464e+02, percent-clipped=1.0 2023-03-07 23:38:37,644 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:38:42,110 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4481, 2.3768, 3.2860, 4.3383, 3.9991, 4.1366, 2.5599, 1.8667], device='cuda:3'), covar=tensor([0.0714, 0.2824, 0.1186, 0.0646, 0.0646, 0.0343, 0.2163, 0.2905], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0194, 0.0182, 0.0174, 0.0165, 0.0134, 0.0187, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 23:38:55,633 INFO [train2.py:809] (3/4) Epoch 8, batch 2250, loss[ctc_loss=0.1515, att_loss=0.2684, loss=0.245, over 17283.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02451, over 59.00 utterances.], tot_loss[ctc_loss=0.123, att_loss=0.2602, loss=0.2328, over 3266516.64 frames. utt_duration=1264 frames, utt_pad_proportion=0.0531, over 10350.37 utterances.], batch size: 59, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:39:11,471 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2915, 3.3581, 3.4215, 2.9788, 3.3973, 3.4231, 3.4400, 1.9959], device='cuda:3'), covar=tensor([0.1254, 0.1648, 0.2975, 0.5224, 0.2400, 0.4576, 0.0961, 1.0009], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0090, 0.0096, 0.0154, 0.0086, 0.0142, 0.0084, 0.0146], device='cuda:3'), out_proj_covar=tensor([7.3455e-05, 7.3890e-05, 8.2815e-05, 1.2032e-04, 7.4471e-05, 1.1381e-04, 6.9428e-05, 1.1688e-04], device='cuda:3') 2023-03-07 23:39:28,583 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2404, 5.4036, 5.8220, 5.6944, 5.5575, 6.1631, 5.2866, 6.2275], device='cuda:3'), covar=tensor([0.0575, 0.0599, 0.0541, 0.0798, 0.1726, 0.0756, 0.0494, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0614, 0.0371, 0.0427, 0.0484, 0.0661, 0.0430, 0.0339, 0.0420], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 23:39:28,596 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:40:13,340 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:40:14,496 INFO [train2.py:809] (3/4) Epoch 8, batch 2300, loss[ctc_loss=0.1084, att_loss=0.2626, loss=0.2318, over 16950.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.007763, over 50.00 utterances.], tot_loss[ctc_loss=0.1229, att_loss=0.2607, loss=0.2331, over 3272319.61 frames. utt_duration=1253 frames, utt_pad_proportion=0.05381, over 10456.06 utterances.], batch size: 50, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:40:46,675 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:40:50,453 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:40:53,592 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2933, 4.7457, 4.1826, 4.7000, 2.3876, 4.6143, 2.2852, 1.6936], device='cuda:3'), covar=tensor([0.0237, 0.0097, 0.0708, 0.0145, 0.2173, 0.0123, 0.1708, 0.1680], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0098, 0.0253, 0.0110, 0.0221, 0.0099, 0.0225, 0.0198], device='cuda:3'), out_proj_covar=tensor([1.2064e-04, 1.0252e-04, 2.3072e-04, 1.0432e-04, 2.0946e-04, 9.8080e-05, 2.0493e-04, 1.8175e-04], device='cuda:3') 2023-03-07 23:41:00,842 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.649e+02 3.337e+02 4.141e+02 1.009e+03, threshold=6.675e+02, percent-clipped=1.0 2023-03-07 23:41:06,283 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-03-07 23:41:34,431 INFO [train2.py:809] (3/4) Epoch 8, batch 2350, loss[ctc_loss=0.1077, att_loss=0.2651, loss=0.2337, over 17250.00 frames. utt_duration=1171 frames, utt_pad_proportion=0.02659, over 59.00 utterances.], tot_loss[ctc_loss=0.1236, att_loss=0.2613, loss=0.2338, over 3277232.84 frames. utt_duration=1234 frames, utt_pad_proportion=0.0569, over 10634.46 utterances.], batch size: 59, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:42:02,892 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:42:28,182 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:42:42,268 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1536, 5.2041, 5.1208, 2.5474, 2.0011, 2.6071, 4.3335, 3.6243], device='cuda:3'), covar=tensor([0.0593, 0.0265, 0.0254, 0.3263, 0.6128, 0.2752, 0.0577, 0.2246], device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0196, 0.0220, 0.0186, 0.0346, 0.0332, 0.0212, 0.0337], device='cuda:3'), out_proj_covar=tensor([1.4963e-04, 7.7879e-05, 9.7441e-05, 8.6275e-05, 1.5754e-04, 1.4126e-04, 8.5352e-05, 1.5077e-04], device='cuda:3') 2023-03-07 23:42:54,198 INFO [train2.py:809] (3/4) Epoch 8, batch 2400, loss[ctc_loss=0.1266, att_loss=0.2606, loss=0.2338, over 16179.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006366, over 41.00 utterances.], tot_loss[ctc_loss=0.1229, att_loss=0.2613, loss=0.2336, over 3285378.78 frames. utt_duration=1257 frames, utt_pad_proportion=0.04934, over 10464.44 utterances.], batch size: 41, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:43:00,647 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1672, 4.8845, 4.4600, 4.8505, 2.6025, 4.6700, 2.3635, 1.7087], device='cuda:3'), covar=tensor([0.0265, 0.0098, 0.0720, 0.0140, 0.2159, 0.0139, 0.1825, 0.1906], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0098, 0.0254, 0.0110, 0.0221, 0.0100, 0.0225, 0.0199], device='cuda:3'), out_proj_covar=tensor([1.2162e-04, 1.0272e-04, 2.3147e-04, 1.0487e-04, 2.0965e-04, 9.8896e-05, 2.0516e-04, 1.8231e-04], device='cuda:3') 2023-03-07 23:43:40,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.833e+02 3.441e+02 4.475e+02 1.318e+03, threshold=6.883e+02, percent-clipped=7.0 2023-03-07 23:43:44,649 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={1} 2023-03-07 23:44:06,812 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-07 23:44:13,410 INFO [train2.py:809] (3/4) Epoch 8, batch 2450, loss[ctc_loss=0.1335, att_loss=0.2501, loss=0.2268, over 15778.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008243, over 38.00 utterances.], tot_loss[ctc_loss=0.1234, att_loss=0.2621, loss=0.2343, over 3281126.22 frames. utt_duration=1259 frames, utt_pad_proportion=0.0502, over 10439.81 utterances.], batch size: 38, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:44:19,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-07 23:44:28,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0 2023-03-07 23:44:50,960 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:44:57,225 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-07 23:45:00,782 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:45:34,119 INFO [train2.py:809] (3/4) Epoch 8, batch 2500, loss[ctc_loss=0.1559, att_loss=0.256, loss=0.236, over 15673.00 frames. utt_duration=1696 frames, utt_pad_proportion=0.006222, over 37.00 utterances.], tot_loss[ctc_loss=0.1238, att_loss=0.2615, loss=0.2339, over 3261540.51 frames. utt_duration=1227 frames, utt_pad_proportion=0.06332, over 10644.94 utterances.], batch size: 37, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:46:16,191 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2023-03-07 23:46:21,433 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.650e+02 3.302e+02 4.215e+02 1.043e+03, threshold=6.603e+02, percent-clipped=3.0 2023-03-07 23:46:30,159 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:46:54,580 INFO [train2.py:809] (3/4) Epoch 8, batch 2550, loss[ctc_loss=0.09312, att_loss=0.237, loss=0.2083, over 16169.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006278, over 41.00 utterances.], tot_loss[ctc_loss=0.1232, att_loss=0.2616, loss=0.2339, over 3272136.94 frames. utt_duration=1238 frames, utt_pad_proportion=0.05762, over 10583.39 utterances.], batch size: 41, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:47:28,451 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:48:04,597 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:48:04,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5940, 4.9238, 4.2743, 5.1572, 4.4074, 4.8985, 5.1203, 4.8978], device='cuda:3'), covar=tensor([0.0573, 0.0415, 0.1099, 0.0240, 0.0524, 0.0210, 0.0308, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0220, 0.0285, 0.0205, 0.0232, 0.0179, 0.0205, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-07 23:48:13,627 INFO [train2.py:809] (3/4) Epoch 8, batch 2600, loss[ctc_loss=0.1092, att_loss=0.2243, loss=0.2013, over 15640.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009083, over 37.00 utterances.], tot_loss[ctc_loss=0.1224, att_loss=0.2607, loss=0.2331, over 3272389.20 frames. utt_duration=1271 frames, utt_pad_proportion=0.05, over 10311.61 utterances.], batch size: 37, lr: 1.35e-02, grad_scale: 8.0 2023-03-07 23:48:29,062 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-07 23:48:44,716 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:49:00,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.127e+02 2.736e+02 3.358e+02 3.851e+02 8.673e+02, threshold=6.715e+02, percent-clipped=1.0 2023-03-07 23:49:33,582 INFO [train2.py:809] (3/4) Epoch 8, batch 2650, loss[ctc_loss=0.09828, att_loss=0.246, loss=0.2164, over 16375.00 frames. utt_duration=1490 frames, utt_pad_proportion=0.009037, over 44.00 utterances.], tot_loss[ctc_loss=0.1231, att_loss=0.2615, loss=0.2338, over 3286260.96 frames. utt_duration=1273 frames, utt_pad_proportion=0.04508, over 10336.12 utterances.], batch size: 44, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:50:18,950 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:50:52,904 INFO [train2.py:809] (3/4) Epoch 8, batch 2700, loss[ctc_loss=0.1509, att_loss=0.246, loss=0.227, over 15376.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.0109, over 35.00 utterances.], tot_loss[ctc_loss=0.1243, att_loss=0.2616, loss=0.2341, over 3282363.62 frames. utt_duration=1260 frames, utt_pad_proportion=0.04817, over 10434.95 utterances.], batch size: 35, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:51:39,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.974e+02 3.669e+02 4.665e+02 1.299e+03, threshold=7.338e+02, percent-clipped=5.0 2023-03-07 23:52:12,703 INFO [train2.py:809] (3/4) Epoch 8, batch 2750, loss[ctc_loss=0.1184, att_loss=0.2372, loss=0.2135, over 15507.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008242, over 36.00 utterances.], tot_loss[ctc_loss=0.1235, att_loss=0.2615, loss=0.2339, over 3286625.07 frames. utt_duration=1279 frames, utt_pad_proportion=0.04274, over 10293.88 utterances.], batch size: 36, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:52:19,269 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6872, 1.9880, 1.6792, 1.8639, 2.8481, 1.8445, 1.9926, 2.5708], device='cuda:3'), covar=tensor([0.0414, 0.3182, 0.2932, 0.1300, 0.0584, 0.1647, 0.2050, 0.1280], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0083, 0.0088, 0.0076, 0.0073, 0.0069, 0.0078, 0.0066], device='cuda:3'), out_proj_covar=tensor([4.1279e-05, 5.2103e-05, 5.3177e-05, 4.4830e-05, 4.0005e-05, 4.4163e-05, 4.9306e-05, 4.3433e-05], device='cuda:3') 2023-03-07 23:52:37,411 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-07 23:52:42,300 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:53:33,180 INFO [train2.py:809] (3/4) Epoch 8, batch 2800, loss[ctc_loss=0.1075, att_loss=0.2449, loss=0.2174, over 16264.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007559, over 43.00 utterances.], tot_loss[ctc_loss=0.1232, att_loss=0.2616, loss=0.2339, over 3290655.77 frames. utt_duration=1266 frames, utt_pad_proportion=0.04376, over 10406.21 utterances.], batch size: 43, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:54:20,568 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.22 vs. limit=5.0 2023-03-07 23:54:21,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.729e+02 3.286e+02 4.018e+02 8.527e+02, threshold=6.573e+02, percent-clipped=3.0 2023-03-07 23:54:21,493 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:54:21,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={0} 2023-03-07 23:54:53,845 INFO [train2.py:809] (3/4) Epoch 8, batch 2850, loss[ctc_loss=0.1161, att_loss=0.2696, loss=0.2389, over 16769.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006614, over 48.00 utterances.], tot_loss[ctc_loss=0.1221, att_loss=0.2602, loss=0.2326, over 3280647.40 frames. utt_duration=1288 frames, utt_pad_proportion=0.04205, over 10197.79 utterances.], batch size: 48, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:54:58,829 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0348, 5.0145, 4.8802, 2.7439, 4.7891, 4.4280, 4.1528, 2.3830], device='cuda:3'), covar=tensor([0.0095, 0.0063, 0.0172, 0.1024, 0.0085, 0.0168, 0.0324, 0.1543], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0075, 0.0061, 0.0099, 0.0066, 0.0083, 0.0086, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-07 23:55:54,933 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6246, 3.7075, 3.6643, 2.9491, 3.5657, 3.7998, 3.7706, 2.1274], device='cuda:3'), covar=tensor([0.1056, 0.1363, 0.2635, 0.7495, 0.6994, 0.4024, 0.1099, 1.1351], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0092, 0.0097, 0.0156, 0.0087, 0.0143, 0.0083, 0.0149], device='cuda:3'), out_proj_covar=tensor([7.3785e-05, 7.5569e-05, 8.4239e-05, 1.2310e-04, 7.6259e-05, 1.1471e-04, 6.9908e-05, 1.1912e-04], device='cuda:3') 2023-03-07 23:56:04,133 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:56:13,170 INFO [train2.py:809] (3/4) Epoch 8, batch 2900, loss[ctc_loss=0.09546, att_loss=0.2372, loss=0.2088, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009222, over 39.00 utterances.], tot_loss[ctc_loss=0.1228, att_loss=0.2608, loss=0.2332, over 3279314.10 frames. utt_duration=1269 frames, utt_pad_proportion=0.04816, over 10352.30 utterances.], batch size: 39, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:56:13,548 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0938, 4.8069, 4.1243, 4.6505, 2.1471, 4.5078, 2.4871, 2.1067], device='cuda:3'), covar=tensor([0.0280, 0.0090, 0.1015, 0.0199, 0.2754, 0.0158, 0.1881, 0.1827], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0100, 0.0257, 0.0110, 0.0226, 0.0101, 0.0228, 0.0200], device='cuda:3'), out_proj_covar=tensor([1.2445e-04, 1.0503e-04, 2.3460e-04, 1.0553e-04, 2.1341e-04, 9.9987e-05, 2.0851e-04, 1.8408e-04], device='cuda:3') 2023-03-07 23:56:13,613 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7561, 2.5353, 5.0485, 3.8050, 3.0818, 4.5408, 4.8318, 4.7412], device='cuda:3'), covar=tensor([0.0168, 0.1919, 0.0114, 0.1127, 0.1958, 0.0223, 0.0099, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0242, 0.0125, 0.0305, 0.0284, 0.0183, 0.0110, 0.0143], device='cuda:3'), out_proj_covar=tensor([1.3744e-04, 2.0245e-04, 1.1077e-04, 2.5160e-04, 2.5113e-04, 1.6220e-04, 9.9559e-05, 1.3445e-04], device='cuda:3') 2023-03-07 23:56:40,926 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7824, 5.0472, 5.3557, 5.2585, 5.1849, 5.7853, 5.0119, 5.8550], device='cuda:3'), covar=tensor([0.0648, 0.0579, 0.0620, 0.0862, 0.1838, 0.0763, 0.0569, 0.0517], device='cuda:3'), in_proj_covar=tensor([0.0616, 0.0373, 0.0427, 0.0485, 0.0660, 0.0435, 0.0350, 0.0428], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-07 23:56:59,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.826e+02 3.398e+02 4.214e+02 7.798e+02, threshold=6.796e+02, percent-clipped=3.0 2023-03-07 23:57:20,147 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:57:32,269 INFO [train2.py:809] (3/4) Epoch 8, batch 2950, loss[ctc_loss=0.1089, att_loss=0.2678, loss=0.236, over 17038.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01047, over 53.00 utterances.], tot_loss[ctc_loss=0.1249, att_loss=0.2622, loss=0.2347, over 3274167.76 frames. utt_duration=1235 frames, utt_pad_proportion=0.05814, over 10613.57 utterances.], batch size: 53, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:57:41,959 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5111, 3.0229, 3.5929, 4.6164, 4.2781, 4.2941, 2.9127, 2.1425], device='cuda:3'), covar=tensor([0.0633, 0.1949, 0.1010, 0.0424, 0.0587, 0.0288, 0.1582, 0.2565], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0200, 0.0188, 0.0178, 0.0174, 0.0140, 0.0191, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-07 23:58:17,848 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:58:52,462 INFO [train2.py:809] (3/4) Epoch 8, batch 3000, loss[ctc_loss=0.143, att_loss=0.2815, loss=0.2538, over 17360.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03526, over 63.00 utterances.], tot_loss[ctc_loss=0.1236, att_loss=0.2611, loss=0.2336, over 3264421.03 frames. utt_duration=1236 frames, utt_pad_proportion=0.06011, over 10579.59 utterances.], batch size: 63, lr: 1.34e-02, grad_scale: 8.0 2023-03-07 23:58:52,463 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-07 23:59:06,432 INFO [train2.py:843] (3/4) Epoch 8, validation: ctc_loss=0.05812, att_loss=0.2422, loss=0.2054, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-07 23:59:06,433 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-07 23:59:32,940 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:59:48,143 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set() 2023-03-07 23:59:53,216 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 2.876e+02 3.540e+02 4.400e+02 7.249e+02, threshold=7.079e+02, percent-clipped=3.0 2023-03-08 00:00:07,458 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:00:21,875 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:00:26,210 INFO [train2.py:809] (3/4) Epoch 8, batch 3050, loss[ctc_loss=0.08808, att_loss=0.252, loss=0.2192, over 16333.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005806, over 45.00 utterances.], tot_loss[ctc_loss=0.1222, att_loss=0.2604, loss=0.2328, over 3270183.77 frames. utt_duration=1262 frames, utt_pad_proportion=0.05171, over 10380.17 utterances.], batch size: 45, lr: 1.34e-02, grad_scale: 8.0 2023-03-08 00:01:10,387 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:01:26,455 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-08 00:01:44,875 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:01:46,066 INFO [train2.py:809] (3/4) Epoch 8, batch 3100, loss[ctc_loss=0.1365, att_loss=0.2758, loss=0.248, over 16981.00 frames. utt_duration=687.7 frames, utt_pad_proportion=0.136, over 99.00 utterances.], tot_loss[ctc_loss=0.1219, att_loss=0.2607, loss=0.2329, over 3270534.02 frames. utt_duration=1253 frames, utt_pad_proportion=0.054, over 10451.32 utterances.], batch size: 99, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:01:59,092 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:02:13,204 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 00:02:24,738 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:02:24,884 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 00:02:32,585 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.541e+02 3.093e+02 3.936e+02 7.180e+02, threshold=6.186e+02, percent-clipped=1.0 2023-03-08 00:02:32,950 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:02:44,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.93 vs. limit=5.0 2023-03-08 00:03:05,876 INFO [train2.py:809] (3/4) Epoch 8, batch 3150, loss[ctc_loss=0.09972, att_loss=0.2475, loss=0.2179, over 16418.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006033, over 44.00 utterances.], tot_loss[ctc_loss=0.123, att_loss=0.2611, loss=0.2335, over 3268560.07 frames. utt_duration=1219 frames, utt_pad_proportion=0.06328, over 10738.70 utterances.], batch size: 44, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:03:49,189 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:03:55,439 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1605, 5.3242, 5.6903, 5.5970, 5.4981, 6.1008, 5.2884, 6.1277], device='cuda:3'), covar=tensor([0.0581, 0.0545, 0.0596, 0.0816, 0.1705, 0.0804, 0.0449, 0.0637], device='cuda:3'), in_proj_covar=tensor([0.0618, 0.0380, 0.0435, 0.0491, 0.0665, 0.0433, 0.0351, 0.0433], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-08 00:03:57,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-08 00:04:01,730 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 00:04:25,395 INFO [train2.py:809] (3/4) Epoch 8, batch 3200, loss[ctc_loss=0.1721, att_loss=0.2944, loss=0.2699, over 13997.00 frames. utt_duration=385 frames, utt_pad_proportion=0.3292, over 146.00 utterances.], tot_loss[ctc_loss=0.1232, att_loss=0.2606, loss=0.2332, over 3266473.84 frames. utt_duration=1220 frames, utt_pad_proportion=0.06411, over 10726.06 utterances.], batch size: 146, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:04:49,936 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2243, 4.5723, 4.6337, 4.4159, 2.1910, 4.5808, 2.3481, 1.9069], device='cuda:3'), covar=tensor([0.0257, 0.0147, 0.0667, 0.0205, 0.2476, 0.0136, 0.1911, 0.1862], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0101, 0.0250, 0.0109, 0.0219, 0.0099, 0.0223, 0.0199], device='cuda:3'), out_proj_covar=tensor([1.2342e-04, 1.0468e-04, 2.2914e-04, 1.0446e-04, 2.0743e-04, 9.8305e-05, 2.0400e-04, 1.8279e-04], device='cuda:3') 2023-03-08 00:05:11,435 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.783e+02 3.320e+02 3.859e+02 7.988e+02, threshold=6.640e+02, percent-clipped=5.0 2023-03-08 00:05:23,231 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-03-08 00:05:44,553 INFO [train2.py:809] (3/4) Epoch 8, batch 3250, loss[ctc_loss=0.1319, att_loss=0.2789, loss=0.2495, over 16764.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005997, over 48.00 utterances.], tot_loss[ctc_loss=0.122, att_loss=0.2599, loss=0.2323, over 3270101.73 frames. utt_duration=1245 frames, utt_pad_proportion=0.05647, over 10518.43 utterances.], batch size: 48, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:07:03,811 INFO [train2.py:809] (3/4) Epoch 8, batch 3300, loss[ctc_loss=0.1037, att_loss=0.2428, loss=0.215, over 16250.00 frames. utt_duration=1513 frames, utt_pad_proportion=0.009001, over 43.00 utterances.], tot_loss[ctc_loss=0.1222, att_loss=0.2602, loss=0.2326, over 3269483.10 frames. utt_duration=1251 frames, utt_pad_proportion=0.05461, over 10470.31 utterances.], batch size: 43, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:07:51,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.610e+02 3.147e+02 4.010e+02 7.285e+02, threshold=6.294e+02, percent-clipped=3.0 2023-03-08 00:07:57,139 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-08 00:08:24,608 INFO [train2.py:809] (3/4) Epoch 8, batch 3350, loss[ctc_loss=0.1196, att_loss=0.2668, loss=0.2373, over 17074.00 frames. utt_duration=691.5 frames, utt_pad_proportion=0.1302, over 99.00 utterances.], tot_loss[ctc_loss=0.122, att_loss=0.26, loss=0.2324, over 3263328.09 frames. utt_duration=1238 frames, utt_pad_proportion=0.05938, over 10556.73 utterances.], batch size: 99, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:09:00,709 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:09:35,256 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:09:44,165 INFO [train2.py:809] (3/4) Epoch 8, batch 3400, loss[ctc_loss=0.1087, att_loss=0.2628, loss=0.232, over 16759.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006278, over 48.00 utterances.], tot_loss[ctc_loss=0.1231, att_loss=0.2609, loss=0.2334, over 3268412.72 frames. utt_duration=1227 frames, utt_pad_proportion=0.06026, over 10665.84 utterances.], batch size: 48, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:09:49,555 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:10:23,078 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1084, 5.0312, 5.0409, 3.1324, 4.9179, 4.6475, 4.5479, 3.1699], device='cuda:3'), covar=tensor([0.0119, 0.0092, 0.0182, 0.0877, 0.0092, 0.0152, 0.0256, 0.1079], device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0075, 0.0063, 0.0100, 0.0066, 0.0085, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 00:10:23,088 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:10:30,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.871e+02 3.632e+02 4.520e+02 1.169e+03, threshold=7.263e+02, percent-clipped=5.0 2023-03-08 00:10:55,147 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8015, 6.0461, 5.4517, 5.9248, 5.7099, 5.3622, 5.4873, 5.3500], device='cuda:3'), covar=tensor([0.1230, 0.1041, 0.0889, 0.0737, 0.0874, 0.1324, 0.2324, 0.2276], device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0454, 0.0345, 0.0359, 0.0329, 0.0398, 0.0468, 0.0423], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 00:11:04,306 INFO [train2.py:809] (3/4) Epoch 8, batch 3450, loss[ctc_loss=0.08777, att_loss=0.2314, loss=0.2027, over 16414.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006049, over 44.00 utterances.], tot_loss[ctc_loss=0.1235, att_loss=0.2609, loss=0.2334, over 3264226.55 frames. utt_duration=1216 frames, utt_pad_proportion=0.06553, over 10749.20 utterances.], batch size: 44, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:11:40,438 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:11:52,639 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 00:11:59,183 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9411, 4.4066, 4.0145, 4.2968, 2.3855, 4.3291, 2.6506, 1.6252], device='cuda:3'), covar=tensor([0.0319, 0.0113, 0.0876, 0.0163, 0.2316, 0.0133, 0.1774, 0.2098], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0102, 0.0255, 0.0111, 0.0224, 0.0103, 0.0230, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:12:25,032 INFO [train2.py:809] (3/4) Epoch 8, batch 3500, loss[ctc_loss=0.08926, att_loss=0.2312, loss=0.2028, over 15630.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009167, over 37.00 utterances.], tot_loss[ctc_loss=0.1237, att_loss=0.2609, loss=0.2335, over 3257020.11 frames. utt_duration=1224 frames, utt_pad_proportion=0.06654, over 10654.92 utterances.], batch size: 37, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:12:42,236 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4720, 2.2456, 3.1072, 4.3416, 3.9089, 4.0628, 2.8060, 2.0299], device='cuda:3'), covar=tensor([0.0707, 0.2744, 0.1298, 0.0560, 0.0719, 0.0362, 0.1635, 0.2704], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0197, 0.0183, 0.0176, 0.0169, 0.0135, 0.0183, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:12:54,443 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7407, 3.8465, 3.1504, 3.6554, 4.0442, 3.6230, 2.7153, 4.3376], device='cuda:3'), covar=tensor([0.1051, 0.0374, 0.1030, 0.0571, 0.0487, 0.0572, 0.0980, 0.0385], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0165, 0.0193, 0.0162, 0.0207, 0.0196, 0.0169, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 00:13:02,211 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:13:03,801 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6756, 2.9996, 3.7580, 3.2045, 3.2926, 4.6909, 4.4894, 3.6050], device='cuda:3'), covar=tensor([0.0402, 0.1652, 0.1022, 0.1207, 0.1235, 0.0724, 0.0425, 0.1070], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0218, 0.0227, 0.0197, 0.0229, 0.0268, 0.0201, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-08 00:13:12,730 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.623e+02 3.276e+02 4.275e+02 9.794e+02, threshold=6.553e+02, percent-clipped=4.0 2023-03-08 00:13:41,417 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5044, 2.8349, 3.7340, 3.1105, 3.2672, 4.6409, 4.4280, 3.5808], device='cuda:3'), covar=tensor([0.0410, 0.1685, 0.0933, 0.1162, 0.1398, 0.0729, 0.0518, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0219, 0.0226, 0.0197, 0.0230, 0.0267, 0.0201, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2023-03-08 00:13:47,410 INFO [train2.py:809] (3/4) Epoch 8, batch 3550, loss[ctc_loss=0.1195, att_loss=0.265, loss=0.2359, over 17449.00 frames. utt_duration=884.9 frames, utt_pad_proportion=0.07439, over 79.00 utterances.], tot_loss[ctc_loss=0.1225, att_loss=0.2603, loss=0.2328, over 3264488.11 frames. utt_duration=1243 frames, utt_pad_proportion=0.05981, over 10520.49 utterances.], batch size: 79, lr: 1.33e-02, grad_scale: 8.0 2023-03-08 00:14:40,240 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:14:57,479 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.25 vs. limit=5.0 2023-03-08 00:15:08,417 INFO [train2.py:809] (3/4) Epoch 8, batch 3600, loss[ctc_loss=0.1179, att_loss=0.2673, loss=0.2374, over 17037.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009652, over 52.00 utterances.], tot_loss[ctc_loss=0.122, att_loss=0.2604, loss=0.2327, over 3272775.70 frames. utt_duration=1243 frames, utt_pad_proportion=0.05805, over 10545.97 utterances.], batch size: 52, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:15:08,784 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6221, 3.6073, 2.9298, 3.4608, 3.7268, 3.4065, 2.6266, 4.1136], device='cuda:3'), covar=tensor([0.1092, 0.0425, 0.1083, 0.0547, 0.0644, 0.0636, 0.0954, 0.0456], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0167, 0.0195, 0.0164, 0.0207, 0.0197, 0.0170, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 00:15:15,105 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 00:15:54,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.609e+02 2.998e+02 4.010e+02 6.869e+02, threshold=5.995e+02, percent-clipped=2.0 2023-03-08 00:16:29,577 INFO [train2.py:809] (3/4) Epoch 8, batch 3650, loss[ctc_loss=0.1325, att_loss=0.2592, loss=0.2338, over 16006.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.00805, over 40.00 utterances.], tot_loss[ctc_loss=0.1224, att_loss=0.2607, loss=0.233, over 3264479.82 frames. utt_duration=1207 frames, utt_pad_proportion=0.06868, over 10833.93 utterances.], batch size: 40, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:16:54,277 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:17:04,962 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:17:14,047 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0526, 5.3171, 5.5481, 5.6227, 5.4672, 5.9447, 5.1460, 6.0709], device='cuda:3'), covar=tensor([0.0676, 0.0734, 0.0601, 0.1000, 0.1784, 0.0931, 0.0558, 0.0666], device='cuda:3'), in_proj_covar=tensor([0.0637, 0.0386, 0.0439, 0.0503, 0.0681, 0.0438, 0.0354, 0.0437], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 00:17:29,045 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0096, 5.0406, 5.0384, 2.3850, 1.9889, 2.4043, 3.8268, 3.7121], device='cuda:3'), covar=tensor([0.0725, 0.0157, 0.0184, 0.3525, 0.5940, 0.3166, 0.0997, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0326, 0.0211, 0.0231, 0.0189, 0.0359, 0.0345, 0.0220, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.5476e-04, 8.2108e-05, 1.0149e-04, 8.7413e-05, 1.6246e-04, 1.4594e-04, 8.7754e-05, 1.5570e-04], device='cuda:3') 2023-03-08 00:17:40,273 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:17:50,150 INFO [train2.py:809] (3/4) Epoch 8, batch 3700, loss[ctc_loss=0.1055, att_loss=0.2442, loss=0.2165, over 15633.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009542, over 37.00 utterances.], tot_loss[ctc_loss=0.1218, att_loss=0.2601, loss=0.2324, over 3262096.21 frames. utt_duration=1227 frames, utt_pad_proportion=0.06542, over 10648.88 utterances.], batch size: 37, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:17:55,128 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:18:22,278 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:18:22,433 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0822, 5.4730, 4.8358, 5.5218, 4.8121, 5.1024, 5.5443, 5.4071], device='cuda:3'), covar=tensor([0.0451, 0.0224, 0.0913, 0.0144, 0.0447, 0.0174, 0.0196, 0.0130], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0226, 0.0287, 0.0214, 0.0236, 0.0183, 0.0208, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-08 00:18:35,885 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.776e+02 3.358e+02 4.277e+02 8.903e+02, threshold=6.717e+02, percent-clipped=4.0 2023-03-08 00:18:54,900 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-08 00:18:57,105 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:19:10,108 INFO [train2.py:809] (3/4) Epoch 8, batch 3750, loss[ctc_loss=0.1005, att_loss=0.2292, loss=0.2035, over 13170.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.03598, over 29.00 utterances.], tot_loss[ctc_loss=0.1211, att_loss=0.2592, loss=0.2316, over 3253327.84 frames. utt_duration=1249 frames, utt_pad_proportion=0.0622, over 10432.97 utterances.], batch size: 29, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:19:11,680 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:19:57,365 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 00:20:29,568 INFO [train2.py:809] (3/4) Epoch 8, batch 3800, loss[ctc_loss=0.1054, att_loss=0.2521, loss=0.2227, over 16273.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007782, over 43.00 utterances.], tot_loss[ctc_loss=0.1215, att_loss=0.2592, loss=0.2317, over 3257908.49 frames. utt_duration=1251 frames, utt_pad_proportion=0.05906, over 10432.89 utterances.], batch size: 43, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:20:56,778 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:21:13,685 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:21:14,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.740e+02 3.298e+02 4.235e+02 1.019e+03, threshold=6.595e+02, percent-clipped=5.0 2023-03-08 00:21:49,310 INFO [train2.py:809] (3/4) Epoch 8, batch 3850, loss[ctc_loss=0.1033, att_loss=0.2312, loss=0.2056, over 15493.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009089, over 36.00 utterances.], tot_loss[ctc_loss=0.1204, att_loss=0.2582, loss=0.2306, over 3256896.79 frames. utt_duration=1279 frames, utt_pad_proportion=0.05178, over 10200.38 utterances.], batch size: 36, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:21:58,015 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4262, 2.5830, 3.2666, 4.4641, 4.0177, 4.0497, 2.7436, 2.0698], device='cuda:3'), covar=tensor([0.0693, 0.2400, 0.1170, 0.0515, 0.0636, 0.0368, 0.1682, 0.2558], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0199, 0.0184, 0.0179, 0.0172, 0.0141, 0.0188, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:22:32,484 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:22:32,673 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:23:05,945 INFO [train2.py:809] (3/4) Epoch 8, batch 3900, loss[ctc_loss=0.1046, att_loss=0.2649, loss=0.2328, over 17329.00 frames. utt_duration=1262 frames, utt_pad_proportion=0.01042, over 55.00 utterances.], tot_loss[ctc_loss=0.1207, att_loss=0.2585, loss=0.2309, over 3264021.87 frames. utt_duration=1284 frames, utt_pad_proportion=0.04896, over 10179.63 utterances.], batch size: 55, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:23:07,801 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:23:50,996 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.828e+02 2.672e+02 3.345e+02 4.002e+02 8.198e+02, threshold=6.690e+02, percent-clipped=4.0 2023-03-08 00:24:01,910 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:24:22,890 INFO [train2.py:809] (3/4) Epoch 8, batch 3950, loss[ctc_loss=0.1336, att_loss=0.2798, loss=0.2506, over 17285.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01295, over 55.00 utterances.], tot_loss[ctc_loss=0.1202, att_loss=0.2595, loss=0.2316, over 3270499.47 frames. utt_duration=1297 frames, utt_pad_proportion=0.04303, over 10100.55 utterances.], batch size: 55, lr: 1.32e-02, grad_scale: 8.0 2023-03-08 00:24:38,344 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:24:41,468 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:25:39,114 INFO [train2.py:809] (3/4) Epoch 9, batch 0, loss[ctc_loss=0.1368, att_loss=0.2723, loss=0.2452, over 17369.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03388, over 63.00 utterances.], tot_loss[ctc_loss=0.1368, att_loss=0.2723, loss=0.2452, over 17369.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03388, over 63.00 utterances.], batch size: 63, lr: 1.25e-02, grad_scale: 8.0 2023-03-08 00:25:39,114 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 00:25:51,858 INFO [train2.py:843] (3/4) Epoch 9, validation: ctc_loss=0.05701, att_loss=0.2418, loss=0.2048, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 00:25:51,859 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 00:26:05,417 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:26:12,283 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.72 vs. limit=5.0 2023-03-08 00:26:14,997 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:27:04,504 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:27:05,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.740e+02 3.193e+02 4.001e+02 8.553e+02, threshold=6.386e+02, percent-clipped=3.0 2023-03-08 00:27:12,469 INFO [train2.py:809] (3/4) Epoch 9, batch 50, loss[ctc_loss=0.1171, att_loss=0.2665, loss=0.2367, over 16962.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.006959, over 50.00 utterances.], tot_loss[ctc_loss=0.1202, att_loss=0.2603, loss=0.2323, over 735525.90 frames. utt_duration=1230 frames, utt_pad_proportion=0.05968, over 2395.26 utterances.], batch size: 50, lr: 1.25e-02, grad_scale: 16.0 2023-03-08 00:27:43,180 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:28:00,018 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:28:31,525 INFO [train2.py:809] (3/4) Epoch 9, batch 100, loss[ctc_loss=0.1011, att_loss=0.2569, loss=0.2258, over 16628.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005125, over 47.00 utterances.], tot_loss[ctc_loss=0.1211, att_loss=0.2586, loss=0.2311, over 1286059.29 frames. utt_duration=1239 frames, utt_pad_proportion=0.06566, over 4157.19 utterances.], batch size: 47, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:28:41,637 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:29:41,558 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:29:48,727 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 3.075e+02 3.496e+02 4.231e+02 7.573e+02, threshold=6.993e+02, percent-clipped=4.0 2023-03-08 00:29:49,555 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-08 00:29:54,995 INFO [train2.py:809] (3/4) Epoch 9, batch 150, loss[ctc_loss=0.1172, att_loss=0.2766, loss=0.2447, over 17384.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.0339, over 63.00 utterances.], tot_loss[ctc_loss=0.1195, att_loss=0.2587, loss=0.2309, over 1727678.67 frames. utt_duration=1284 frames, utt_pad_proportion=0.04896, over 5387.86 utterances.], batch size: 63, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:30:15,032 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4469, 3.7374, 3.7668, 3.0036, 3.4057, 3.7341, 3.5965, 2.4429], device='cuda:3'), covar=tensor([0.1206, 0.1272, 0.1657, 0.5677, 0.3933, 0.2798, 0.0725, 0.7387], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0098, 0.0102, 0.0166, 0.0089, 0.0151, 0.0086, 0.0153], device='cuda:3'), out_proj_covar=tensor([7.9631e-05, 8.1565e-05, 8.9138e-05, 1.3141e-04, 7.8432e-05, 1.2162e-04, 7.3243e-05, 1.2269e-04], device='cuda:3') 2023-03-08 00:30:59,449 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:31:07,686 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:31:15,196 INFO [train2.py:809] (3/4) Epoch 9, batch 200, loss[ctc_loss=0.1154, att_loss=0.2731, loss=0.2416, over 17395.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03326, over 63.00 utterances.], tot_loss[ctc_loss=0.1187, att_loss=0.2579, loss=0.23, over 2063542.90 frames. utt_duration=1255 frames, utt_pad_proportion=0.05921, over 6584.83 utterances.], batch size: 63, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:31:43,443 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1078, 5.1531, 5.0447, 2.6005, 2.0139, 2.6739, 3.7053, 3.8573], device='cuda:3'), covar=tensor([0.0628, 0.0174, 0.0218, 0.3108, 0.5563, 0.2670, 0.1113, 0.1800], device='cuda:3'), in_proj_covar=tensor([0.0320, 0.0202, 0.0228, 0.0186, 0.0350, 0.0337, 0.0216, 0.0344], device='cuda:3'), out_proj_covar=tensor([1.5131e-04, 7.8451e-05, 1.0013e-04, 8.5382e-05, 1.5815e-04, 1.4211e-04, 8.6249e-05, 1.5141e-04], device='cuda:3') 2023-03-08 00:31:49,482 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5969, 3.7583, 3.0852, 3.5929, 3.9030, 3.5782, 2.9347, 4.3584], device='cuda:3'), covar=tensor([0.1111, 0.0524, 0.0991, 0.0570, 0.0557, 0.0612, 0.0788, 0.0529], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0173, 0.0204, 0.0170, 0.0213, 0.0205, 0.0177, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 00:32:23,996 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:32:28,423 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.737e+02 3.348e+02 3.979e+02 1.144e+03, threshold=6.695e+02, percent-clipped=2.0 2023-03-08 00:32:34,682 INFO [train2.py:809] (3/4) Epoch 9, batch 250, loss[ctc_loss=0.1348, att_loss=0.2785, loss=0.2498, over 17098.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01548, over 56.00 utterances.], tot_loss[ctc_loss=0.1188, att_loss=0.2578, loss=0.23, over 2330605.54 frames. utt_duration=1276 frames, utt_pad_proportion=0.05243, over 7316.90 utterances.], batch size: 56, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:33:13,086 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:33:17,811 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:33:25,564 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3089, 4.5054, 3.7671, 4.7225, 4.0087, 4.2606, 4.4982, 4.4235], device='cuda:3'), covar=tensor([0.0488, 0.0365, 0.1236, 0.0223, 0.0508, 0.0459, 0.0483, 0.0257], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0224, 0.0286, 0.0215, 0.0236, 0.0184, 0.0209, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-08 00:33:54,277 INFO [train2.py:809] (3/4) Epoch 9, batch 300, loss[ctc_loss=0.07622, att_loss=0.2095, loss=0.1828, over 15649.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008576, over 37.00 utterances.], tot_loss[ctc_loss=0.1186, att_loss=0.2571, loss=0.2294, over 2533464.13 frames. utt_duration=1230 frames, utt_pad_proportion=0.06413, over 8250.10 utterances.], batch size: 37, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:34:02,870 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8204, 5.1626, 4.5860, 5.3125, 4.5853, 4.9720, 5.3282, 5.0396], device='cuda:3'), covar=tensor([0.0493, 0.0228, 0.0861, 0.0186, 0.0456, 0.0205, 0.0192, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0223, 0.0283, 0.0215, 0.0235, 0.0183, 0.0208, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-08 00:34:09,517 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:34:14,334 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2033, 4.6906, 4.6137, 4.7913, 2.3364, 4.3878, 2.6568, 1.6592], device='cuda:3'), covar=tensor([0.0260, 0.0139, 0.0675, 0.0126, 0.2378, 0.0201, 0.1846, 0.2160], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0104, 0.0254, 0.0110, 0.0224, 0.0105, 0.0232, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:34:25,427 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1426, 4.8109, 4.9635, 2.2236, 1.8812, 2.2449, 3.3929, 3.4337], device='cuda:3'), covar=tensor([0.0776, 0.0350, 0.0374, 0.3295, 0.6485, 0.3492, 0.1306, 0.2814], device='cuda:3'), in_proj_covar=tensor([0.0317, 0.0202, 0.0228, 0.0184, 0.0351, 0.0335, 0.0216, 0.0345], device='cuda:3'), out_proj_covar=tensor([1.4998e-04, 7.8500e-05, 1.0022e-04, 8.4581e-05, 1.5816e-04, 1.4122e-04, 8.6060e-05, 1.5149e-04], device='cuda:3') 2023-03-08 00:34:26,649 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2829, 4.4920, 4.5144, 4.5625, 4.6076, 4.5613, 4.3426, 4.1634], device='cuda:3'), covar=tensor([0.0977, 0.0585, 0.0285, 0.0360, 0.0325, 0.0313, 0.0271, 0.0341], device='cuda:3'), in_proj_covar=tensor([0.0430, 0.0272, 0.0212, 0.0256, 0.0314, 0.0337, 0.0255, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 00:34:34,216 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 00:35:08,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.524e+02 3.099e+02 4.059e+02 8.225e+02, threshold=6.197e+02, percent-clipped=2.0 2023-03-08 00:35:14,583 INFO [train2.py:809] (3/4) Epoch 9, batch 350, loss[ctc_loss=0.1555, att_loss=0.2832, loss=0.2576, over 16946.00 frames. utt_duration=686.1 frames, utt_pad_proportion=0.138, over 99.00 utterances.], tot_loss[ctc_loss=0.1174, att_loss=0.2571, loss=0.2292, over 2696616.56 frames. utt_duration=1233 frames, utt_pad_proportion=0.06109, over 8757.70 utterances.], batch size: 99, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:35:21,887 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3957, 4.4047, 4.6407, 4.5112, 5.0496, 4.4789, 4.3887, 2.0360], device='cuda:3'), covar=tensor([0.0302, 0.0310, 0.0231, 0.0237, 0.1120, 0.0277, 0.0297, 0.2669], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0125, 0.0128, 0.0133, 0.0318, 0.0126, 0.0114, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 00:35:36,501 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3087, 3.6750, 3.6235, 2.9887, 3.4447, 3.6052, 3.4978, 2.1735], device='cuda:3'), covar=tensor([0.1381, 0.1071, 0.2863, 0.6630, 0.4308, 0.5137, 0.0853, 1.1160], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0097, 0.0103, 0.0168, 0.0089, 0.0153, 0.0087, 0.0154], device='cuda:3'), out_proj_covar=tensor([7.9670e-05, 8.0445e-05, 9.0029e-05, 1.3300e-04, 7.8541e-05, 1.2256e-04, 7.4129e-05, 1.2330e-04], device='cuda:3') 2023-03-08 00:35:37,877 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:35:42,816 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4564, 4.7080, 4.7714, 5.1789, 2.6628, 4.7079, 3.0985, 2.4224], device='cuda:3'), covar=tensor([0.0231, 0.0144, 0.0564, 0.0099, 0.1796, 0.0143, 0.1413, 0.1510], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0103, 0.0250, 0.0106, 0.0219, 0.0102, 0.0229, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:36:35,198 INFO [train2.py:809] (3/4) Epoch 9, batch 400, loss[ctc_loss=0.113, att_loss=0.2516, loss=0.2239, over 16482.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005725, over 46.00 utterances.], tot_loss[ctc_loss=0.1189, att_loss=0.2587, loss=0.2307, over 2823407.08 frames. utt_duration=1204 frames, utt_pad_proportion=0.06796, over 9390.61 utterances.], batch size: 46, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:36:36,900 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:37:32,241 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:37:48,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.567e+02 3.123e+02 3.914e+02 5.946e+02, threshold=6.245e+02, percent-clipped=0.0 2023-03-08 00:37:54,622 INFO [train2.py:809] (3/4) Epoch 9, batch 450, loss[ctc_loss=0.1062, att_loss=0.237, loss=0.2108, over 15610.00 frames. utt_duration=1689 frames, utt_pad_proportion=0.01095, over 37.00 utterances.], tot_loss[ctc_loss=0.1193, att_loss=0.2594, loss=0.2314, over 2936416.75 frames. utt_duration=1226 frames, utt_pad_proportion=0.05796, over 9596.09 utterances.], batch size: 37, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:38:59,124 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:39:02,278 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3745, 1.9107, 2.1607, 1.9391, 2.9754, 1.8150, 1.7460, 1.6050], device='cuda:3'), covar=tensor([0.0980, 0.4888, 0.2711, 0.1937, 0.0978, 0.2796, 0.3552, 0.3231], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0086, 0.0085, 0.0075, 0.0072, 0.0071, 0.0084, 0.0066], device='cuda:3'), out_proj_covar=tensor([4.2976e-05, 5.4672e-05, 5.3212e-05, 4.5807e-05, 4.0902e-05, 4.6336e-05, 5.2489e-05, 4.4238e-05], device='cuda:3') 2023-03-08 00:39:14,594 INFO [train2.py:809] (3/4) Epoch 9, batch 500, loss[ctc_loss=0.1044, att_loss=0.2362, loss=0.2099, over 16198.00 frames. utt_duration=1582 frames, utt_pad_proportion=0.00517, over 41.00 utterances.], tot_loss[ctc_loss=0.1184, att_loss=0.2584, loss=0.2304, over 3010834.81 frames. utt_duration=1250 frames, utt_pad_proportion=0.05134, over 9648.89 utterances.], batch size: 41, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:40:15,787 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:40:28,732 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.692e+02 3.196e+02 3.958e+02 9.983e+02, threshold=6.392e+02, percent-clipped=1.0 2023-03-08 00:40:35,579 INFO [train2.py:809] (3/4) Epoch 9, batch 550, loss[ctc_loss=0.1751, att_loss=0.2879, loss=0.2653, over 17395.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.04736, over 69.00 utterances.], tot_loss[ctc_loss=0.1169, att_loss=0.2571, loss=0.2291, over 3059356.94 frames. utt_duration=1253 frames, utt_pad_proportion=0.05325, over 9774.42 utterances.], batch size: 69, lr: 1.24e-02, grad_scale: 16.0 2023-03-08 00:41:04,783 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3907, 4.6911, 4.6990, 5.0316, 2.7541, 4.5434, 2.7207, 1.9622], device='cuda:3'), covar=tensor([0.0291, 0.0208, 0.0689, 0.0120, 0.1752, 0.0179, 0.1566, 0.1724], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0104, 0.0254, 0.0108, 0.0220, 0.0103, 0.0231, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:41:06,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-03-08 00:41:13,838 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:41:55,743 INFO [train2.py:809] (3/4) Epoch 9, batch 600, loss[ctc_loss=0.09062, att_loss=0.251, loss=0.219, over 16996.00 frames. utt_duration=1334 frames, utt_pad_proportion=0.00858, over 51.00 utterances.], tot_loss[ctc_loss=0.1178, att_loss=0.2577, loss=0.2297, over 3109922.68 frames. utt_duration=1236 frames, utt_pad_proportion=0.0558, over 10074.05 utterances.], batch size: 51, lr: 1.23e-02, grad_scale: 16.0 2023-03-08 00:42:10,461 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:42:29,841 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:42:40,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 2023-03-08 00:43:07,804 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.723e+02 3.556e+02 4.400e+02 8.502e+02, threshold=7.112e+02, percent-clipped=9.0 2023-03-08 00:43:14,504 INFO [train2.py:809] (3/4) Epoch 9, batch 650, loss[ctc_loss=0.16, att_loss=0.2815, loss=0.2572, over 16955.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007427, over 50.00 utterances.], tot_loss[ctc_loss=0.1189, att_loss=0.2577, loss=0.2299, over 3143232.27 frames. utt_duration=1250 frames, utt_pad_proportion=0.05438, over 10068.77 utterances.], batch size: 50, lr: 1.23e-02, grad_scale: 16.0 2023-03-08 00:43:25,742 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:43:36,648 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:44:33,112 INFO [train2.py:809] (3/4) Epoch 9, batch 700, loss[ctc_loss=0.1478, att_loss=0.2761, loss=0.2504, over 17549.00 frames. utt_duration=1018 frames, utt_pad_proportion=0.04008, over 69.00 utterances.], tot_loss[ctc_loss=0.1194, att_loss=0.2578, loss=0.2301, over 3170278.08 frames. utt_duration=1249 frames, utt_pad_proportion=0.05473, over 10165.81 utterances.], batch size: 69, lr: 1.23e-02, grad_scale: 16.0 2023-03-08 00:44:35,679 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:44:52,285 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:45:31,393 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:45:47,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.817e+02 3.384e+02 3.932e+02 1.069e+03, threshold=6.768e+02, percent-clipped=2.0 2023-03-08 00:45:51,659 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:45:53,047 INFO [train2.py:809] (3/4) Epoch 9, batch 750, loss[ctc_loss=0.1555, att_loss=0.2865, loss=0.2603, over 16115.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006803, over 42.00 utterances.], tot_loss[ctc_loss=0.1197, att_loss=0.2583, loss=0.2306, over 3186411.26 frames. utt_duration=1229 frames, utt_pad_proportion=0.06022, over 10382.32 utterances.], batch size: 42, lr: 1.23e-02, grad_scale: 16.0 2023-03-08 00:46:22,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2023-03-08 00:46:47,263 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:47:09,921 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6170, 5.8615, 5.1867, 5.7136, 5.3798, 5.1053, 5.2160, 5.2349], device='cuda:3'), covar=tensor([0.1331, 0.0875, 0.0935, 0.0709, 0.0928, 0.1393, 0.2781, 0.2129], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0444, 0.0347, 0.0361, 0.0334, 0.0395, 0.0481, 0.0419], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 00:47:12,887 INFO [train2.py:809] (3/4) Epoch 9, batch 800, loss[ctc_loss=0.1133, att_loss=0.2631, loss=0.2331, over 17054.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008752, over 52.00 utterances.], tot_loss[ctc_loss=0.1184, att_loss=0.2572, loss=0.2294, over 3200458.30 frames. utt_duration=1233 frames, utt_pad_proportion=0.06196, over 10394.99 utterances.], batch size: 52, lr: 1.23e-02, grad_scale: 16.0 2023-03-08 00:47:30,185 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-08 00:47:30,331 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-08 00:48:28,947 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.761e+02 3.783e+02 4.545e+02 8.141e+02, threshold=7.567e+02, percent-clipped=8.0 2023-03-08 00:48:33,639 INFO [train2.py:809] (3/4) Epoch 9, batch 850, loss[ctc_loss=0.1517, att_loss=0.2907, loss=0.2629, over 17360.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02036, over 59.00 utterances.], tot_loss[ctc_loss=0.1182, att_loss=0.2577, loss=0.2298, over 3218872.47 frames. utt_duration=1253 frames, utt_pad_proportion=0.05557, over 10288.97 utterances.], batch size: 59, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:49:54,782 INFO [train2.py:809] (3/4) Epoch 9, batch 900, loss[ctc_loss=0.137, att_loss=0.2574, loss=0.2333, over 16124.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006249, over 42.00 utterances.], tot_loss[ctc_loss=0.1182, att_loss=0.258, loss=0.2301, over 3234089.18 frames. utt_duration=1252 frames, utt_pad_proportion=0.05535, over 10348.77 utterances.], batch size: 42, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:50:08,196 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-08 00:50:51,165 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-08 00:51:11,047 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.630e+02 3.149e+02 3.930e+02 7.487e+02, threshold=6.299e+02, percent-clipped=0.0 2023-03-08 00:51:15,683 INFO [train2.py:809] (3/4) Epoch 9, batch 950, loss[ctc_loss=0.1274, att_loss=0.2646, loss=0.2372, over 16954.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.00821, over 50.00 utterances.], tot_loss[ctc_loss=0.1193, att_loss=0.2588, loss=0.2309, over 3237762.65 frames. utt_duration=1218 frames, utt_pad_proportion=0.06264, over 10643.51 utterances.], batch size: 50, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:52:35,535 INFO [train2.py:809] (3/4) Epoch 9, batch 1000, loss[ctc_loss=0.1555, att_loss=0.287, loss=0.2607, over 17095.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01576, over 56.00 utterances.], tot_loss[ctc_loss=0.1203, att_loss=0.2597, loss=0.2318, over 3244097.97 frames. utt_duration=1194 frames, utt_pad_proportion=0.06842, over 10881.01 utterances.], batch size: 56, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:52:57,956 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 00:53:22,721 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4524, 2.5090, 2.9858, 4.3558, 4.1071, 4.0942, 2.8373, 1.7400], device='cuda:3'), covar=tensor([0.0660, 0.2412, 0.1334, 0.0601, 0.0584, 0.0334, 0.1546, 0.2902], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0208, 0.0193, 0.0187, 0.0177, 0.0142, 0.0192, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:53:49,490 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.0809, 3.5062, 3.4209, 2.8684, 3.4668, 3.5894, 3.4207, 2.3995], device='cuda:3'), covar=tensor([0.1323, 0.1670, 0.3665, 0.7592, 0.1230, 0.3232, 0.1081, 0.9305], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0102, 0.0106, 0.0174, 0.0092, 0.0157, 0.0090, 0.0156], device='cuda:3'), out_proj_covar=tensor([8.2292e-05, 8.4926e-05, 9.3588e-05, 1.3788e-04, 8.1698e-05, 1.2665e-04, 7.7133e-05, 1.2547e-04], device='cuda:3') 2023-03-08 00:53:50,523 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.656e+02 3.223e+02 3.962e+02 9.648e+02, threshold=6.446e+02, percent-clipped=5.0 2023-03-08 00:53:55,324 INFO [train2.py:809] (3/4) Epoch 9, batch 1050, loss[ctc_loss=0.1043, att_loss=0.2617, loss=0.2303, over 16609.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.005457, over 47.00 utterances.], tot_loss[ctc_loss=0.1189, att_loss=0.259, loss=0.231, over 3251986.81 frames. utt_duration=1213 frames, utt_pad_proportion=0.06327, over 10735.80 utterances.], batch size: 47, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:54:16,103 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7083, 3.8359, 3.8428, 2.4371, 2.3301, 2.7353, 2.5549, 3.4071], device='cuda:3'), covar=tensor([0.0631, 0.0280, 0.0356, 0.2521, 0.4180, 0.2101, 0.1560, 0.1439], device='cuda:3'), in_proj_covar=tensor([0.0322, 0.0206, 0.0234, 0.0187, 0.0353, 0.0335, 0.0222, 0.0347], device='cuda:3'), out_proj_covar=tensor([1.5222e-04, 7.8458e-05, 1.0252e-04, 8.5847e-05, 1.5875e-04, 1.4073e-04, 8.8594e-05, 1.5185e-04], device='cuda:3') 2023-03-08 00:54:34,513 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-08 00:55:10,098 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:55:15,879 INFO [train2.py:809] (3/4) Epoch 9, batch 1100, loss[ctc_loss=0.1267, att_loss=0.2658, loss=0.238, over 16972.00 frames. utt_duration=687.2 frames, utt_pad_proportion=0.1367, over 99.00 utterances.], tot_loss[ctc_loss=0.1177, att_loss=0.2579, loss=0.2298, over 3256538.10 frames. utt_duration=1230 frames, utt_pad_proportion=0.05859, over 10605.72 utterances.], batch size: 99, lr: 1.23e-02, grad_scale: 8.0 2023-03-08 00:56:31,654 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 3.040e+02 3.595e+02 4.615e+02 9.742e+02, threshold=7.190e+02, percent-clipped=7.0 2023-03-08 00:56:36,387 INFO [train2.py:809] (3/4) Epoch 9, batch 1150, loss[ctc_loss=0.09445, att_loss=0.2294, loss=0.2024, over 15489.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009344, over 36.00 utterances.], tot_loss[ctc_loss=0.1177, att_loss=0.2578, loss=0.2297, over 3257453.04 frames. utt_duration=1213 frames, utt_pad_proportion=0.06421, over 10755.13 utterances.], batch size: 36, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 00:56:47,502 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:57:54,928 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:57:56,105 INFO [train2.py:809] (3/4) Epoch 9, batch 1200, loss[ctc_loss=0.1331, att_loss=0.2795, loss=0.2502, over 17399.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03211, over 63.00 utterances.], tot_loss[ctc_loss=0.1172, att_loss=0.2577, loss=0.2296, over 3261056.67 frames. utt_duration=1222 frames, utt_pad_proportion=0.06223, over 10687.15 utterances.], batch size: 63, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 00:58:28,651 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0664, 4.6509, 4.1546, 4.5273, 2.3484, 4.4291, 2.7697, 1.6698], device='cuda:3'), covar=tensor([0.0278, 0.0113, 0.0959, 0.0183, 0.2256, 0.0189, 0.1788, 0.2021], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0102, 0.0253, 0.0109, 0.0220, 0.0105, 0.0228, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 00:59:12,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.465e+02 3.116e+02 3.703e+02 8.772e+02, threshold=6.232e+02, percent-clipped=2.0 2023-03-08 00:59:17,165 INFO [train2.py:809] (3/4) Epoch 9, batch 1250, loss[ctc_loss=0.1102, att_loss=0.2548, loss=0.2259, over 17385.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03377, over 63.00 utterances.], tot_loss[ctc_loss=0.1166, att_loss=0.2569, loss=0.2289, over 3263972.96 frames. utt_duration=1240 frames, utt_pad_proportion=0.05749, over 10542.97 utterances.], batch size: 63, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 00:59:32,983 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 00:59:39,521 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6101, 5.0099, 4.8044, 5.0408, 5.0547, 4.7450, 3.7238, 4.9527], device='cuda:3'), covar=tensor([0.0108, 0.0087, 0.0107, 0.0075, 0.0078, 0.0107, 0.0527, 0.0196], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0066, 0.0078, 0.0051, 0.0054, 0.0064, 0.0087, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:00:14,114 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:00:36,808 INFO [train2.py:809] (3/4) Epoch 9, batch 1300, loss[ctc_loss=0.112, att_loss=0.2449, loss=0.2184, over 15880.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009627, over 39.00 utterances.], tot_loss[ctc_loss=0.1161, att_loss=0.2564, loss=0.2283, over 3254770.08 frames. utt_duration=1235 frames, utt_pad_proportion=0.06075, over 10552.39 utterances.], batch size: 39, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:01:35,947 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2875, 5.1636, 4.9639, 2.5473, 1.9815, 2.8722, 3.7803, 3.8513], device='cuda:3'), covar=tensor([0.0542, 0.0228, 0.0268, 0.3773, 0.6233, 0.2719, 0.1311, 0.1824], device='cuda:3'), in_proj_covar=tensor([0.0318, 0.0206, 0.0230, 0.0185, 0.0349, 0.0337, 0.0219, 0.0345], device='cuda:3'), out_proj_covar=tensor([1.5030e-04, 7.9416e-05, 1.0097e-04, 8.4653e-05, 1.5711e-04, 1.4122e-04, 8.7003e-05, 1.5069e-04], device='cuda:3') 2023-03-08 01:01:43,375 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8022, 2.2452, 2.4323, 3.2564, 3.1811, 3.3775, 2.5984, 2.1420], device='cuda:3'), covar=tensor([0.0531, 0.1707, 0.1166, 0.0684, 0.0773, 0.0265, 0.1111, 0.1573], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0198, 0.0184, 0.0176, 0.0170, 0.0136, 0.0185, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 01:01:51,109 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:01:52,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.474e+02 3.066e+02 3.872e+02 7.426e+02, threshold=6.132e+02, percent-clipped=4.0 2023-03-08 01:01:56,794 INFO [train2.py:809] (3/4) Epoch 9, batch 1350, loss[ctc_loss=0.1371, att_loss=0.237, loss=0.217, over 15753.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009531, over 38.00 utterances.], tot_loss[ctc_loss=0.1165, att_loss=0.2566, loss=0.2286, over 3260395.83 frames. utt_duration=1235 frames, utt_pad_proportion=0.06017, over 10575.51 utterances.], batch size: 38, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:02:30,173 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9276, 4.6910, 4.7754, 4.8188, 4.6544, 4.8015, 4.5988, 4.3639], device='cuda:3'), covar=tensor([0.1971, 0.0802, 0.0299, 0.0492, 0.0683, 0.0419, 0.0381, 0.0412], device='cuda:3'), in_proj_covar=tensor([0.0432, 0.0271, 0.0214, 0.0255, 0.0321, 0.0340, 0.0260, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:03:16,721 INFO [train2.py:809] (3/4) Epoch 9, batch 1400, loss[ctc_loss=0.1418, att_loss=0.2837, loss=0.2553, over 17341.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03554, over 63.00 utterances.], tot_loss[ctc_loss=0.1168, att_loss=0.2571, loss=0.229, over 3264284.35 frames. utt_duration=1234 frames, utt_pad_proportion=0.06009, over 10596.18 utterances.], batch size: 63, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:04:32,132 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.521e+02 2.872e+02 3.748e+02 7.311e+02, threshold=5.745e+02, percent-clipped=4.0 2023-03-08 01:04:37,048 INFO [train2.py:809] (3/4) Epoch 9, batch 1450, loss[ctc_loss=0.09964, att_loss=0.2575, loss=0.2259, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006737, over 45.00 utterances.], tot_loss[ctc_loss=0.1156, att_loss=0.2563, loss=0.2281, over 3267658.67 frames. utt_duration=1267 frames, utt_pad_proportion=0.05183, over 10329.09 utterances.], batch size: 45, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:04:40,490 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:05:56,599 INFO [train2.py:809] (3/4) Epoch 9, batch 1500, loss[ctc_loss=0.1133, att_loss=0.2685, loss=0.2374, over 17034.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01069, over 53.00 utterances.], tot_loss[ctc_loss=0.1161, att_loss=0.257, loss=0.2288, over 3278235.09 frames. utt_duration=1282 frames, utt_pad_proportion=0.04541, over 10239.79 utterances.], batch size: 53, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:07:10,428 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8401, 6.0724, 5.4569, 5.9037, 5.7070, 5.3066, 5.4922, 5.3730], device='cuda:3'), covar=tensor([0.1111, 0.0778, 0.0830, 0.0645, 0.0655, 0.1326, 0.2113, 0.2042], device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0456, 0.0350, 0.0365, 0.0333, 0.0397, 0.0484, 0.0430], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 01:07:11,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 2.556e+02 3.254e+02 4.022e+02 8.037e+02, threshold=6.507e+02, percent-clipped=3.0 2023-03-08 01:07:16,447 INFO [train2.py:809] (3/4) Epoch 9, batch 1550, loss[ctc_loss=0.123, att_loss=0.2642, loss=0.236, over 16280.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006736, over 43.00 utterances.], tot_loss[ctc_loss=0.1163, att_loss=0.2572, loss=0.229, over 3281876.22 frames. utt_duration=1286 frames, utt_pad_proportion=0.04349, over 10218.05 utterances.], batch size: 43, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:07:24,569 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:08:33,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-08 01:08:37,543 INFO [train2.py:809] (3/4) Epoch 9, batch 1600, loss[ctc_loss=0.1177, att_loss=0.251, loss=0.2244, over 16396.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007994, over 44.00 utterances.], tot_loss[ctc_loss=0.1156, att_loss=0.2566, loss=0.2284, over 3282609.77 frames. utt_duration=1265 frames, utt_pad_proportion=0.04796, over 10389.53 utterances.], batch size: 44, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:08:40,894 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:09:43,894 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:09:53,212 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.746e+02 3.273e+02 4.611e+02 8.460e+02, threshold=6.545e+02, percent-clipped=8.0 2023-03-08 01:09:57,953 INFO [train2.py:809] (3/4) Epoch 9, batch 1650, loss[ctc_loss=0.1448, att_loss=0.2816, loss=0.2542, over 17336.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02255, over 59.00 utterances.], tot_loss[ctc_loss=0.1157, att_loss=0.2564, loss=0.2282, over 3282526.06 frames. utt_duration=1260 frames, utt_pad_proportion=0.04944, over 10431.03 utterances.], batch size: 59, lr: 1.22e-02, grad_scale: 8.0 2023-03-08 01:09:59,842 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4372, 2.4725, 3.3287, 4.3645, 3.8168, 3.9203, 2.8978, 2.1289], device='cuda:3'), covar=tensor([0.0685, 0.2512, 0.1041, 0.0532, 0.0704, 0.0392, 0.1635, 0.2496], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0202, 0.0188, 0.0183, 0.0174, 0.0140, 0.0189, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 01:10:19,003 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:11:18,160 INFO [train2.py:809] (3/4) Epoch 9, batch 1700, loss[ctc_loss=0.1286, att_loss=0.282, loss=0.2514, over 17039.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.007579, over 52.00 utterances.], tot_loss[ctc_loss=0.1167, att_loss=0.2571, loss=0.229, over 3267092.98 frames. utt_duration=1210 frames, utt_pad_proportion=0.06455, over 10814.44 utterances.], batch size: 52, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:11:20,478 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.88 vs. limit=5.0 2023-03-08 01:12:35,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.664e+02 3.328e+02 4.191e+02 7.435e+02, threshold=6.656e+02, percent-clipped=3.0 2023-03-08 01:12:39,100 INFO [train2.py:809] (3/4) Epoch 9, batch 1750, loss[ctc_loss=0.08211, att_loss=0.2311, loss=0.2013, over 16283.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007082, over 43.00 utterances.], tot_loss[ctc_loss=0.1154, att_loss=0.2563, loss=0.2281, over 3270763.53 frames. utt_duration=1231 frames, utt_pad_proportion=0.05885, over 10636.81 utterances.], batch size: 43, lr: 1.21e-02, grad_scale: 4.0 2023-03-08 01:12:42,497 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:13:17,014 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-03-08 01:13:36,462 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-08 01:13:58,114 INFO [train2.py:809] (3/4) Epoch 9, batch 1800, loss[ctc_loss=0.1304, att_loss=0.2706, loss=0.2426, over 16481.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006549, over 46.00 utterances.], tot_loss[ctc_loss=0.1163, att_loss=0.2566, loss=0.2285, over 3274917.17 frames. utt_duration=1237 frames, utt_pad_proportion=0.05613, over 10601.06 utterances.], batch size: 46, lr: 1.21e-02, grad_scale: 4.0 2023-03-08 01:13:58,226 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:14:09,868 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:14:45,944 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2023-03-08 01:15:14,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.773e+02 3.319e+02 4.110e+02 8.879e+02, threshold=6.639e+02, percent-clipped=6.0 2023-03-08 01:15:17,757 INFO [train2.py:809] (3/4) Epoch 9, batch 1850, loss[ctc_loss=0.1375, att_loss=0.2733, loss=0.2461, over 16773.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006141, over 48.00 utterances.], tot_loss[ctc_loss=0.117, att_loss=0.257, loss=0.229, over 3265712.72 frames. utt_duration=1231 frames, utt_pad_proportion=0.06058, over 10624.75 utterances.], batch size: 48, lr: 1.21e-02, grad_scale: 4.0 2023-03-08 01:15:26,853 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:15:47,525 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 01:16:37,509 INFO [train2.py:809] (3/4) Epoch 9, batch 1900, loss[ctc_loss=0.1007, att_loss=0.2562, loss=0.2251, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005185, over 47.00 utterances.], tot_loss[ctc_loss=0.1172, att_loss=0.2572, loss=0.2292, over 3269633.26 frames. utt_duration=1238 frames, utt_pad_proportion=0.05788, over 10580.98 utterances.], batch size: 47, lr: 1.21e-02, grad_scale: 4.0 2023-03-08 01:16:42,310 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:17:34,794 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.86 vs. limit=5.0 2023-03-08 01:17:44,863 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:17:55,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.711e+02 3.222e+02 4.216e+02 1.081e+03, threshold=6.444e+02, percent-clipped=7.0 2023-03-08 01:17:58,269 INFO [train2.py:809] (3/4) Epoch 9, batch 1950, loss[ctc_loss=0.1179, att_loss=0.2584, loss=0.2303, over 15958.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006871, over 41.00 utterances.], tot_loss[ctc_loss=0.1172, att_loss=0.2574, loss=0.2294, over 3276070.46 frames. utt_duration=1253 frames, utt_pad_proportion=0.05124, over 10469.84 utterances.], batch size: 41, lr: 1.21e-02, grad_scale: 4.0 2023-03-08 01:18:11,658 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:19:01,162 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:19:17,451 INFO [train2.py:809] (3/4) Epoch 9, batch 2000, loss[ctc_loss=0.104, att_loss=0.2482, loss=0.2194, over 16314.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007174, over 45.00 utterances.], tot_loss[ctc_loss=0.1166, att_loss=0.2571, loss=0.229, over 3272700.25 frames. utt_duration=1246 frames, utt_pad_proportion=0.05185, over 10516.72 utterances.], batch size: 45, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:19:42,760 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-08 01:20:34,522 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.627e+02 3.033e+02 3.756e+02 1.034e+03, threshold=6.065e+02, percent-clipped=4.0 2023-03-08 01:20:37,625 INFO [train2.py:809] (3/4) Epoch 9, batch 2050, loss[ctc_loss=0.1527, att_loss=0.2566, loss=0.2358, over 15496.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009201, over 36.00 utterances.], tot_loss[ctc_loss=0.1159, att_loss=0.2568, loss=0.2286, over 3267924.82 frames. utt_duration=1235 frames, utt_pad_proportion=0.05757, over 10597.48 utterances.], batch size: 36, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:21:08,877 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:21:28,839 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6471, 2.1088, 5.0228, 3.9436, 3.0683, 4.4506, 4.8025, 4.7200], device='cuda:3'), covar=tensor([0.0206, 0.2051, 0.0146, 0.1051, 0.1881, 0.0209, 0.0105, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0248, 0.0129, 0.0306, 0.0283, 0.0182, 0.0113, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 01:21:58,313 INFO [train2.py:809] (3/4) Epoch 9, batch 2100, loss[ctc_loss=0.1188, att_loss=0.2688, loss=0.2388, over 16760.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006777, over 48.00 utterances.], tot_loss[ctc_loss=0.1156, att_loss=0.2569, loss=0.2287, over 3272061.45 frames. utt_duration=1265 frames, utt_pad_proportion=0.04931, over 10356.18 utterances.], batch size: 48, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:22:33,466 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6204, 2.8128, 3.5476, 3.0294, 3.3828, 4.7192, 4.4894, 3.4551], device='cuda:3'), covar=tensor([0.0312, 0.1716, 0.1168, 0.1306, 0.1213, 0.0619, 0.0510, 0.1135], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0221, 0.0233, 0.0204, 0.0235, 0.0276, 0.0206, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-08 01:22:33,544 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0468, 4.4609, 4.8224, 5.1601, 2.3252, 4.6505, 2.7540, 1.5175], device='cuda:3'), covar=tensor([0.0318, 0.0182, 0.0580, 0.0091, 0.2086, 0.0185, 0.1624, 0.1900], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0102, 0.0251, 0.0105, 0.0218, 0.0101, 0.0223, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 01:22:50,653 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:22:59,981 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9050, 5.3244, 5.2040, 5.2616, 5.4034, 5.3320, 5.0175, 4.7763], device='cuda:3'), covar=tensor([0.1209, 0.0512, 0.0257, 0.0448, 0.0265, 0.0320, 0.0293, 0.0331], device='cuda:3'), in_proj_covar=tensor([0.0435, 0.0276, 0.0218, 0.0257, 0.0323, 0.0347, 0.0263, 0.0299], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:23:00,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2704, 5.1906, 5.0652, 2.9863, 4.9153, 4.6944, 4.6061, 3.1322], device='cuda:3'), covar=tensor([0.0092, 0.0088, 0.0155, 0.0990, 0.0097, 0.0155, 0.0243, 0.1205], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0078, 0.0067, 0.0103, 0.0067, 0.0090, 0.0089, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:23:18,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.606e+02 3.129e+02 4.184e+02 1.323e+03, threshold=6.258e+02, percent-clipped=9.0 2023-03-08 01:23:22,291 INFO [train2.py:809] (3/4) Epoch 9, batch 2150, loss[ctc_loss=0.1198, att_loss=0.2657, loss=0.2365, over 17488.00 frames. utt_duration=887.3 frames, utt_pad_proportion=0.07094, over 79.00 utterances.], tot_loss[ctc_loss=0.1165, att_loss=0.2574, loss=0.2292, over 3277847.01 frames. utt_duration=1249 frames, utt_pad_proportion=0.05222, over 10511.45 utterances.], batch size: 79, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:23:43,587 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 01:24:02,275 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:24:33,437 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4463, 1.8604, 2.1794, 1.2876, 2.6510, 2.7005, 2.1969, 3.3574], device='cuda:3'), covar=tensor([0.0533, 0.3700, 0.2573, 0.2541, 0.1013, 0.0992, 0.2175, 0.0580], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0080, 0.0071, 0.0072, 0.0067, 0.0078, 0.0061], device='cuda:3'), out_proj_covar=tensor([4.2822e-05, 5.2376e-05, 5.1384e-05, 4.4525e-05, 4.1658e-05, 4.4065e-05, 5.0610e-05, 4.1359e-05], device='cuda:3') 2023-03-08 01:24:42,988 INFO [train2.py:809] (3/4) Epoch 9, batch 2200, loss[ctc_loss=0.1432, att_loss=0.276, loss=0.2494, over 17034.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007738, over 51.00 utterances.], tot_loss[ctc_loss=0.1172, att_loss=0.258, loss=0.2299, over 3283141.17 frames. utt_duration=1257 frames, utt_pad_proportion=0.04921, over 10462.96 utterances.], batch size: 51, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:25:31,251 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2023-03-08 01:25:38,261 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:25:58,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.581e+02 3.160e+02 3.967e+02 1.179e+03, threshold=6.319e+02, percent-clipped=5.0 2023-03-08 01:26:01,794 INFO [train2.py:809] (3/4) Epoch 9, batch 2250, loss[ctc_loss=0.1138, att_loss=0.2693, loss=0.2382, over 17329.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02129, over 59.00 utterances.], tot_loss[ctc_loss=0.1162, att_loss=0.2572, loss=0.229, over 3278815.92 frames. utt_duration=1267 frames, utt_pad_proportion=0.04806, over 10361.47 utterances.], batch size: 59, lr: 1.21e-02, grad_scale: 8.0 2023-03-08 01:26:06,654 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3669, 4.9461, 4.7386, 4.8479, 4.8448, 4.5778, 3.3909, 4.9137], device='cuda:3'), covar=tensor([0.0112, 0.0098, 0.0107, 0.0083, 0.0108, 0.0103, 0.0626, 0.0180], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0065, 0.0077, 0.0048, 0.0052, 0.0062, 0.0084, 0.0083], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:26:14,040 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:26:42,354 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:27:21,330 INFO [train2.py:809] (3/4) Epoch 9, batch 2300, loss[ctc_loss=0.1093, att_loss=0.251, loss=0.2227, over 15893.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008957, over 39.00 utterances.], tot_loss[ctc_loss=0.1169, att_loss=0.2578, loss=0.2296, over 3277350.90 frames. utt_duration=1261 frames, utt_pad_proportion=0.05014, over 10406.55 utterances.], batch size: 39, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:27:23,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6726, 4.9995, 5.2620, 5.1529, 5.0799, 5.6785, 4.9944, 5.7141], device='cuda:3'), covar=tensor([0.0753, 0.0676, 0.0682, 0.1060, 0.1928, 0.0784, 0.0720, 0.0675], device='cuda:3'), in_proj_covar=tensor([0.0659, 0.0400, 0.0454, 0.0525, 0.0698, 0.0462, 0.0379, 0.0451], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:27:30,967 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:28:20,189 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:28:31,125 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2598, 4.4295, 4.7100, 5.2598, 2.4656, 4.9193, 2.4935, 1.7644], device='cuda:3'), covar=tensor([0.0229, 0.0167, 0.0635, 0.0060, 0.1985, 0.0104, 0.1759, 0.1834], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0105, 0.0256, 0.0107, 0.0222, 0.0103, 0.0228, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 01:28:39,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.559e+02 3.119e+02 3.946e+02 1.160e+03, threshold=6.239e+02, percent-clipped=6.0 2023-03-08 01:28:40,970 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0729, 4.7239, 4.4923, 4.7881, 4.6938, 4.4761, 3.2381, 4.6415], device='cuda:3'), covar=tensor([0.0143, 0.0114, 0.0128, 0.0080, 0.0111, 0.0125, 0.0698, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0066, 0.0079, 0.0050, 0.0053, 0.0063, 0.0086, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:28:42,208 INFO [train2.py:809] (3/4) Epoch 9, batch 2350, loss[ctc_loss=0.11, att_loss=0.2649, loss=0.2339, over 17311.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.03543, over 63.00 utterances.], tot_loss[ctc_loss=0.1167, att_loss=0.2582, loss=0.2299, over 3281082.52 frames. utt_duration=1258 frames, utt_pad_proportion=0.05089, over 10443.25 utterances.], batch size: 63, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:29:25,562 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8949, 5.1784, 5.4350, 5.3551, 5.2860, 5.8718, 5.0656, 5.9449], device='cuda:3'), covar=tensor([0.0628, 0.0735, 0.0580, 0.0920, 0.1859, 0.0749, 0.0647, 0.0549], device='cuda:3'), in_proj_covar=tensor([0.0661, 0.0401, 0.0456, 0.0528, 0.0699, 0.0467, 0.0376, 0.0453], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:30:03,497 INFO [train2.py:809] (3/4) Epoch 9, batch 2400, loss[ctc_loss=0.1426, att_loss=0.2691, loss=0.2438, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007509, over 50.00 utterances.], tot_loss[ctc_loss=0.1177, att_loss=0.2583, loss=0.2302, over 3279884.88 frames. utt_duration=1246 frames, utt_pad_proportion=0.05356, over 10546.14 utterances.], batch size: 50, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:30:21,805 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6343, 2.1870, 5.0151, 3.7002, 2.9485, 4.4430, 4.8128, 4.7346], device='cuda:3'), covar=tensor([0.0228, 0.2017, 0.0152, 0.1191, 0.2001, 0.0211, 0.0110, 0.0206], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0252, 0.0131, 0.0311, 0.0285, 0.0185, 0.0113, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 01:30:45,154 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:31:21,639 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.539e+02 3.254e+02 3.979e+02 8.947e+02, threshold=6.508e+02, percent-clipped=3.0 2023-03-08 01:31:24,806 INFO [train2.py:809] (3/4) Epoch 9, batch 2450, loss[ctc_loss=0.117, att_loss=0.2705, loss=0.2398, over 16945.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.008751, over 50.00 utterances.], tot_loss[ctc_loss=0.1173, att_loss=0.2579, loss=0.2298, over 3274173.85 frames. utt_duration=1255 frames, utt_pad_proportion=0.05204, over 10446.42 utterances.], batch size: 50, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:31:47,047 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:32:45,725 INFO [train2.py:809] (3/4) Epoch 9, batch 2500, loss[ctc_loss=0.1161, att_loss=0.2545, loss=0.2268, over 16268.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.008026, over 43.00 utterances.], tot_loss[ctc_loss=0.117, att_loss=0.2581, loss=0.2299, over 3283904.31 frames. utt_duration=1261 frames, utt_pad_proportion=0.04828, over 10428.16 utterances.], batch size: 43, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:33:03,132 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:33:35,230 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:34:03,745 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.575e+02 3.136e+02 3.976e+02 7.316e+02, threshold=6.272e+02, percent-clipped=1.0 2023-03-08 01:34:06,905 INFO [train2.py:809] (3/4) Epoch 9, batch 2550, loss[ctc_loss=0.1867, att_loss=0.2913, loss=0.2704, over 13723.00 frames. utt_duration=379.8 frames, utt_pad_proportion=0.3406, over 145.00 utterances.], tot_loss[ctc_loss=0.1175, att_loss=0.2583, loss=0.2302, over 3280232.63 frames. utt_duration=1230 frames, utt_pad_proportion=0.05905, over 10684.10 utterances.], batch size: 145, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:34:07,336 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7459, 1.7371, 1.9260, 1.8242, 2.4038, 1.8549, 1.4872, 3.0493], device='cuda:3'), covar=tensor([0.0836, 0.4148, 0.3790, 0.1713, 0.1306, 0.1676, 0.3316, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0085, 0.0085, 0.0073, 0.0074, 0.0069, 0.0081, 0.0062], device='cuda:3'), out_proj_covar=tensor([4.4379e-05, 5.4615e-05, 5.3948e-05, 4.5878e-05, 4.3235e-05, 4.5746e-05, 5.2681e-05, 4.2040e-05], device='cuda:3') 2023-03-08 01:35:28,480 INFO [train2.py:809] (3/4) Epoch 9, batch 2600, loss[ctc_loss=0.0956, att_loss=0.2643, loss=0.2306, over 16630.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.00502, over 47.00 utterances.], tot_loss[ctc_loss=0.1169, att_loss=0.2581, loss=0.2299, over 3285833.00 frames. utt_duration=1241 frames, utt_pad_proportion=0.05448, over 10602.06 utterances.], batch size: 47, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:36:04,163 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.93 vs. limit=5.0 2023-03-08 01:36:19,431 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:36:21,215 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0965, 1.5709, 2.1927, 1.8758, 2.5654, 2.1461, 1.8528, 3.2558], device='cuda:3'), covar=tensor([0.0964, 0.5124, 0.4202, 0.2056, 0.1557, 0.1742, 0.3011, 0.0875], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0084, 0.0084, 0.0072, 0.0074, 0.0068, 0.0081, 0.0060], device='cuda:3'), out_proj_covar=tensor([4.3862e-05, 5.4420e-05, 5.3653e-05, 4.5628e-05, 4.3215e-05, 4.5574e-05, 5.2374e-05, 4.1464e-05], device='cuda:3') 2023-03-08 01:36:46,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.544e+02 3.202e+02 3.743e+02 1.159e+03, threshold=6.403e+02, percent-clipped=3.0 2023-03-08 01:36:49,752 INFO [train2.py:809] (3/4) Epoch 9, batch 2650, loss[ctc_loss=0.1167, att_loss=0.2424, loss=0.2172, over 15883.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009331, over 39.00 utterances.], tot_loss[ctc_loss=0.1159, att_loss=0.257, loss=0.2288, over 3270762.60 frames. utt_duration=1253 frames, utt_pad_proportion=0.05358, over 10451.12 utterances.], batch size: 39, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:38:10,642 INFO [train2.py:809] (3/4) Epoch 9, batch 2700, loss[ctc_loss=0.106, att_loss=0.2485, loss=0.22, over 16533.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.005931, over 45.00 utterances.], tot_loss[ctc_loss=0.1167, att_loss=0.2578, loss=0.2296, over 3271319.45 frames. utt_duration=1243 frames, utt_pad_proportion=0.05506, over 10536.35 utterances.], batch size: 45, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:38:32,022 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7408, 3.7178, 3.1464, 3.3774, 3.8241, 3.5595, 2.4915, 4.2768], device='cuda:3'), covar=tensor([0.0923, 0.0391, 0.0896, 0.0580, 0.0544, 0.0565, 0.0936, 0.0339], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0171, 0.0197, 0.0167, 0.0214, 0.0204, 0.0173, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:38:47,926 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:38:51,079 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:38:55,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 01:39:22,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7628, 3.6088, 2.9697, 3.4076, 3.8699, 3.5257, 2.6061, 4.2432], device='cuda:3'), covar=tensor([0.1050, 0.0464, 0.1177, 0.0606, 0.0587, 0.0688, 0.0960, 0.0503], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0173, 0.0199, 0.0168, 0.0216, 0.0206, 0.0175, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:39:27,954 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 2.571e+02 3.320e+02 3.912e+02 1.043e+03, threshold=6.640e+02, percent-clipped=3.0 2023-03-08 01:39:30,982 INFO [train2.py:809] (3/4) Epoch 9, batch 2750, loss[ctc_loss=0.1113, att_loss=0.2286, loss=0.2051, over 14435.00 frames. utt_duration=1806 frames, utt_pad_proportion=0.03552, over 32.00 utterances.], tot_loss[ctc_loss=0.1171, att_loss=0.2582, loss=0.23, over 3280985.60 frames. utt_duration=1259 frames, utt_pad_proportion=0.04856, over 10435.73 utterances.], batch size: 32, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:39:34,447 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:39:41,359 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:39:42,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-08 01:40:08,110 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:40:17,840 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5511, 3.7785, 3.0682, 3.4501, 3.8998, 3.5297, 2.3277, 4.0715], device='cuda:3'), covar=tensor([0.1270, 0.0403, 0.1128, 0.0719, 0.0648, 0.0707, 0.1280, 0.0710], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0171, 0.0197, 0.0165, 0.0214, 0.0204, 0.0173, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:40:25,661 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:40:27,098 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7434, 5.1943, 4.5925, 5.2445, 4.5411, 4.9078, 5.3271, 5.0413], device='cuda:3'), covar=tensor([0.0570, 0.0214, 0.0859, 0.0214, 0.0463, 0.0203, 0.0218, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0235, 0.0300, 0.0229, 0.0244, 0.0190, 0.0221, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 01:40:50,936 INFO [train2.py:809] (3/4) Epoch 9, batch 2800, loss[ctc_loss=0.1136, att_loss=0.2614, loss=0.2318, over 17287.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01205, over 55.00 utterances.], tot_loss[ctc_loss=0.1168, att_loss=0.2582, loss=0.2299, over 3283181.70 frames. utt_duration=1261 frames, utt_pad_proportion=0.04768, over 10427.92 utterances.], batch size: 55, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:41:13,007 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 01:41:17,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 01:41:19,700 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:41:40,081 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:41:55,870 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 01:42:07,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.689e+02 3.286e+02 4.031e+02 7.048e+02, threshold=6.572e+02, percent-clipped=5.0 2023-03-08 01:42:11,090 INFO [train2.py:809] (3/4) Epoch 9, batch 2850, loss[ctc_loss=0.1155, att_loss=0.2621, loss=0.2328, over 16455.00 frames. utt_duration=1432 frames, utt_pad_proportion=0.007307, over 46.00 utterances.], tot_loss[ctc_loss=0.1167, att_loss=0.258, loss=0.2297, over 3286473.76 frames. utt_duration=1249 frames, utt_pad_proportion=0.04952, over 10533.70 utterances.], batch size: 46, lr: 1.20e-02, grad_scale: 8.0 2023-03-08 01:42:56,380 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:43:31,238 INFO [train2.py:809] (3/4) Epoch 9, batch 2900, loss[ctc_loss=0.1133, att_loss=0.2657, loss=0.2352, over 17408.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.0327, over 63.00 utterances.], tot_loss[ctc_loss=0.1152, att_loss=0.2568, loss=0.2284, over 3284354.41 frames. utt_duration=1272 frames, utt_pad_proportion=0.04446, over 10338.67 utterances.], batch size: 63, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:43:33,231 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 01:44:22,089 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:44:48,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.823e+02 3.310e+02 4.171e+02 9.231e+02, threshold=6.619e+02, percent-clipped=4.0 2023-03-08 01:44:51,955 INFO [train2.py:809] (3/4) Epoch 9, batch 2950, loss[ctc_loss=0.1193, att_loss=0.269, loss=0.2391, over 16880.00 frames. utt_duration=683.6 frames, utt_pad_proportion=0.1423, over 99.00 utterances.], tot_loss[ctc_loss=0.1153, att_loss=0.2565, loss=0.2282, over 3277334.93 frames. utt_duration=1263 frames, utt_pad_proportion=0.04886, over 10393.68 utterances.], batch size: 99, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:45:40,214 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:45:40,375 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7750, 5.1490, 4.6581, 5.2442, 4.5967, 4.8576, 5.3047, 5.0455], device='cuda:3'), covar=tensor([0.0604, 0.0273, 0.0818, 0.0227, 0.0455, 0.0249, 0.0278, 0.0217], device='cuda:3'), in_proj_covar=tensor([0.0312, 0.0242, 0.0304, 0.0233, 0.0247, 0.0193, 0.0224, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 01:46:13,923 INFO [train2.py:809] (3/4) Epoch 9, batch 3000, loss[ctc_loss=0.111, att_loss=0.2323, loss=0.208, over 15426.00 frames. utt_duration=1764 frames, utt_pad_proportion=0.007633, over 35.00 utterances.], tot_loss[ctc_loss=0.1159, att_loss=0.257, loss=0.2287, over 3268311.96 frames. utt_duration=1233 frames, utt_pad_proportion=0.05907, over 10617.93 utterances.], batch size: 35, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:46:13,923 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 01:46:28,755 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.3344, 6.5090, 6.0519, 6.3278, 6.2039, 5.9399, 6.0510, 5.8002], device='cuda:3'), covar=tensor([0.1067, 0.0756, 0.0857, 0.0637, 0.0770, 0.0958, 0.1717, 0.1742], device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0473, 0.0364, 0.0374, 0.0343, 0.0408, 0.0494, 0.0440], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 01:46:32,376 INFO [train2.py:843] (3/4) Epoch 9, validation: ctc_loss=0.05401, att_loss=0.2408, loss=0.2035, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 01:46:32,377 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 01:46:41,761 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-08 01:46:57,425 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9183, 5.1914, 5.4817, 5.4100, 5.3278, 5.8994, 5.1574, 5.9722], device='cuda:3'), covar=tensor([0.0670, 0.0674, 0.0622, 0.1052, 0.1740, 0.0833, 0.0700, 0.0628], device='cuda:3'), in_proj_covar=tensor([0.0666, 0.0401, 0.0460, 0.0524, 0.0702, 0.0478, 0.0381, 0.0450], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:47:49,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.760e+02 3.196e+02 3.997e+02 8.726e+02, threshold=6.392e+02, percent-clipped=4.0 2023-03-08 01:47:52,528 INFO [train2.py:809] (3/4) Epoch 9, batch 3050, loss[ctc_loss=0.1172, att_loss=0.2727, loss=0.2416, over 16626.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005349, over 47.00 utterances.], tot_loss[ctc_loss=0.1161, att_loss=0.2576, loss=0.2293, over 3278291.82 frames. utt_duration=1249 frames, utt_pad_proportion=0.05179, over 10511.47 utterances.], batch size: 47, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:48:40,029 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:48:46,369 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.0875, 3.4361, 3.2483, 2.5423, 3.2869, 3.2965, 3.1399, 2.1603], device='cuda:3'), covar=tensor([0.1636, 0.2360, 0.3564, 0.9123, 0.1719, 0.5043, 0.1893, 0.9810], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0107, 0.0113, 0.0178, 0.0096, 0.0163, 0.0093, 0.0160], device='cuda:3'), out_proj_covar=tensor([8.2812e-05, 9.1049e-05, 1.0019e-04, 1.4257e-04, 8.6416e-05, 1.3354e-04, 8.0563e-05, 1.3015e-04], device='cuda:3') 2023-03-08 01:49:12,666 INFO [train2.py:809] (3/4) Epoch 9, batch 3100, loss[ctc_loss=0.08772, att_loss=0.2272, loss=0.1993, over 15641.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009115, over 37.00 utterances.], tot_loss[ctc_loss=0.1156, att_loss=0.2571, loss=0.2288, over 3277463.64 frames. utt_duration=1250 frames, utt_pad_proportion=0.05098, over 10500.67 utterances.], batch size: 37, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:49:25,035 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 01:49:32,537 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:49:56,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-08 01:50:30,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.700e+02 3.373e+02 4.220e+02 7.587e+02, threshold=6.745e+02, percent-clipped=2.0 2023-03-08 01:50:34,013 INFO [train2.py:809] (3/4) Epoch 9, batch 3150, loss[ctc_loss=0.1024, att_loss=0.2506, loss=0.221, over 16296.00 frames. utt_duration=1518 frames, utt_pad_proportion=0.006214, over 43.00 utterances.], tot_loss[ctc_loss=0.1141, att_loss=0.2557, loss=0.2274, over 3272142.53 frames. utt_duration=1271 frames, utt_pad_proportion=0.04815, over 10312.63 utterances.], batch size: 43, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:51:48,851 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 01:51:54,779 INFO [train2.py:809] (3/4) Epoch 9, batch 3200, loss[ctc_loss=0.08953, att_loss=0.2528, loss=0.2201, over 16766.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.00573, over 48.00 utterances.], tot_loss[ctc_loss=0.1142, att_loss=0.2564, loss=0.228, over 3285990.35 frames. utt_duration=1287 frames, utt_pad_proportion=0.04113, over 10223.70 utterances.], batch size: 48, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:52:24,783 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-08 01:52:31,374 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:52:57,546 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6291, 2.3322, 3.2921, 2.4598, 3.1254, 3.9968, 3.7794, 2.8995], device='cuda:3'), covar=tensor([0.0601, 0.2072, 0.1147, 0.1582, 0.1221, 0.0852, 0.0677, 0.1470], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0227, 0.0240, 0.0205, 0.0235, 0.0283, 0.0211, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 01:53:13,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.431e+02 2.913e+02 3.671e+02 9.523e+02, threshold=5.826e+02, percent-clipped=2.0 2023-03-08 01:53:16,524 INFO [train2.py:809] (3/4) Epoch 9, batch 3250, loss[ctc_loss=0.1007, att_loss=0.24, loss=0.2121, over 16275.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007463, over 43.00 utterances.], tot_loss[ctc_loss=0.1148, att_loss=0.2565, loss=0.2282, over 3279888.89 frames. utt_duration=1252 frames, utt_pad_proportion=0.05135, over 10495.07 utterances.], batch size: 43, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:54:10,630 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:54:37,146 INFO [train2.py:809] (3/4) Epoch 9, batch 3300, loss[ctc_loss=0.1236, att_loss=0.2636, loss=0.2356, over 16323.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006536, over 45.00 utterances.], tot_loss[ctc_loss=0.1151, att_loss=0.2567, loss=0.2284, over 3286965.32 frames. utt_duration=1252 frames, utt_pad_proportion=0.04775, over 10517.62 utterances.], batch size: 45, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:54:56,481 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7999, 5.1564, 5.0949, 5.0207, 5.1986, 5.1633, 4.8305, 4.6407], device='cuda:3'), covar=tensor([0.1138, 0.0500, 0.0259, 0.0624, 0.0271, 0.0309, 0.0308, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0437, 0.0273, 0.0224, 0.0262, 0.0325, 0.0349, 0.0268, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 01:55:55,870 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.550e+02 3.164e+02 3.869e+02 8.354e+02, threshold=6.328e+02, percent-clipped=5.0 2023-03-08 01:55:58,989 INFO [train2.py:809] (3/4) Epoch 9, batch 3350, loss[ctc_loss=0.1193, att_loss=0.2697, loss=0.2396, over 16978.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.006793, over 50.00 utterances.], tot_loss[ctc_loss=0.1151, att_loss=0.2571, loss=0.2287, over 3283815.49 frames. utt_duration=1231 frames, utt_pad_proportion=0.05392, over 10684.25 utterances.], batch size: 50, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:56:43,788 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-03-08 01:56:46,164 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:56:59,206 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-08 01:57:09,319 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-08 01:57:10,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-08 01:57:11,733 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1846, 5.1896, 5.1245, 2.3338, 1.9607, 2.9596, 3.8824, 3.7260], device='cuda:3'), covar=tensor([0.0581, 0.0238, 0.0218, 0.4430, 0.5882, 0.2503, 0.1299, 0.2122], device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0218, 0.0234, 0.0200, 0.0361, 0.0343, 0.0231, 0.0360], device='cuda:3'), out_proj_covar=tensor([1.5658e-04, 8.3668e-05, 1.0213e-04, 9.1624e-05, 1.6028e-04, 1.4233e-04, 9.1772e-05, 1.5566e-04], device='cuda:3') 2023-03-08 01:57:18,718 INFO [train2.py:809] (3/4) Epoch 9, batch 3400, loss[ctc_loss=0.1134, att_loss=0.2223, loss=0.2005, over 15484.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009174, over 36.00 utterances.], tot_loss[ctc_loss=0.1158, att_loss=0.2568, loss=0.2286, over 3274713.07 frames. utt_duration=1224 frames, utt_pad_proportion=0.05898, over 10717.96 utterances.], batch size: 36, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:57:18,896 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0865, 6.2761, 5.6422, 6.1220, 5.9246, 5.5527, 5.6527, 5.5114], device='cuda:3'), covar=tensor([0.0989, 0.0770, 0.0767, 0.0702, 0.0684, 0.1201, 0.1814, 0.2044], device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0478, 0.0361, 0.0370, 0.0345, 0.0407, 0.0496, 0.0442], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 01:57:32,781 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 01:57:38,886 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:58:03,245 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:58:07,558 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-08 01:58:36,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.749e+02 3.298e+02 4.168e+02 1.046e+03, threshold=6.595e+02, percent-clipped=7.0 2023-03-08 01:58:39,566 INFO [train2.py:809] (3/4) Epoch 9, batch 3450, loss[ctc_loss=0.1068, att_loss=0.2577, loss=0.2275, over 16399.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007797, over 44.00 utterances.], tot_loss[ctc_loss=0.1153, att_loss=0.2566, loss=0.2284, over 3275288.02 frames. utt_duration=1224 frames, utt_pad_proportion=0.05926, over 10715.01 utterances.], batch size: 44, lr: 1.19e-02, grad_scale: 8.0 2023-03-08 01:58:50,422 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:58:56,841 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 01:59:21,320 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 01:59:47,135 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1721, 4.5831, 4.5510, 4.7493, 2.4436, 4.3642, 2.3587, 1.5328], device='cuda:3'), covar=tensor([0.0308, 0.0123, 0.0613, 0.0134, 0.1992, 0.0179, 0.1832, 0.2002], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0107, 0.0259, 0.0108, 0.0222, 0.0106, 0.0228, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 01:59:55,008 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 02:00:01,143 INFO [train2.py:809] (3/4) Epoch 9, batch 3500, loss[ctc_loss=0.1012, att_loss=0.2644, loss=0.2317, over 16674.00 frames. utt_duration=1451 frames, utt_pad_proportion=0.006592, over 46.00 utterances.], tot_loss[ctc_loss=0.1154, att_loss=0.2568, loss=0.2285, over 3273188.43 frames. utt_duration=1226 frames, utt_pad_proportion=0.05966, over 10689.27 utterances.], batch size: 46, lr: 1.18e-02, grad_scale: 8.0 2023-03-08 02:00:09,018 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:00:34,574 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-03-08 02:00:59,474 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:01:12,463 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:01:14,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-08 02:01:18,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.480e+02 3.195e+02 3.962e+02 8.803e+02, threshold=6.389e+02, percent-clipped=2.0 2023-03-08 02:01:21,704 INFO [train2.py:809] (3/4) Epoch 9, batch 3550, loss[ctc_loss=0.1146, att_loss=0.2365, loss=0.2121, over 14894.00 frames. utt_duration=1807 frames, utt_pad_proportion=0.03167, over 33.00 utterances.], tot_loss[ctc_loss=0.1154, att_loss=0.2569, loss=0.2286, over 3270377.79 frames. utt_duration=1253 frames, utt_pad_proportion=0.05413, over 10455.44 utterances.], batch size: 33, lr: 1.18e-02, grad_scale: 8.0 2023-03-08 02:01:46,271 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:02:06,796 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:02:41,174 INFO [train2.py:809] (3/4) Epoch 9, batch 3600, loss[ctc_loss=0.1084, att_loss=0.267, loss=0.2353, over 16772.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006155, over 48.00 utterances.], tot_loss[ctc_loss=0.1175, att_loss=0.2586, loss=0.2304, over 3262476.40 frames. utt_duration=1229 frames, utt_pad_proportion=0.06021, over 10634.66 utterances.], batch size: 48, lr: 1.18e-02, grad_scale: 8.0 2023-03-08 02:03:40,929 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8303, 5.2228, 4.1692, 5.4078, 4.6636, 5.0272, 5.2754, 5.1355], device='cuda:3'), covar=tensor([0.0563, 0.0365, 0.1359, 0.0229, 0.0442, 0.0225, 0.0309, 0.0208], device='cuda:3'), in_proj_covar=tensor([0.0311, 0.0243, 0.0307, 0.0234, 0.0245, 0.0195, 0.0225, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 02:03:50,960 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3457, 1.7084, 2.0082, 1.4229, 2.2633, 1.7998, 1.8455, 1.6253], device='cuda:3'), covar=tensor([0.0783, 0.4403, 0.3778, 0.2215, 0.1353, 0.1570, 0.2506, 0.1833], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0089, 0.0090, 0.0077, 0.0077, 0.0072, 0.0084, 0.0067], device='cuda:3'), out_proj_covar=tensor([4.5804e-05, 5.7603e-05, 5.8009e-05, 4.8842e-05, 4.5952e-05, 4.8120e-05, 5.5197e-05, 4.6215e-05], device='cuda:3') 2023-03-08 02:03:58,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.714e+02 3.545e+02 4.370e+02 8.238e+02, threshold=7.091e+02, percent-clipped=8.0 2023-03-08 02:04:01,314 INFO [train2.py:809] (3/4) Epoch 9, batch 3650, loss[ctc_loss=0.09466, att_loss=0.2311, loss=0.2038, over 15370.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01052, over 35.00 utterances.], tot_loss[ctc_loss=0.1177, att_loss=0.2593, loss=0.231, over 3276055.45 frames. utt_duration=1234 frames, utt_pad_proportion=0.0544, over 10634.00 utterances.], batch size: 35, lr: 1.18e-02, grad_scale: 8.0 2023-03-08 02:04:10,658 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8226, 3.9186, 3.2842, 3.4914, 4.0102, 3.6336, 2.8069, 4.3609], device='cuda:3'), covar=tensor([0.1098, 0.0358, 0.1037, 0.0592, 0.0530, 0.0639, 0.0941, 0.0447], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0173, 0.0202, 0.0167, 0.0218, 0.0204, 0.0177, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 02:04:18,685 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:05:22,722 INFO [train2.py:809] (3/4) Epoch 9, batch 3700, loss[ctc_loss=0.1186, att_loss=0.25, loss=0.2237, over 16271.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.00763, over 43.00 utterances.], tot_loss[ctc_loss=0.1159, att_loss=0.2578, loss=0.2295, over 3272010.28 frames. utt_duration=1254 frames, utt_pad_proportion=0.0519, over 10450.06 utterances.], batch size: 43, lr: 1.18e-02, grad_scale: 8.0 2023-03-08 02:05:57,192 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:05:57,625 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.27 vs. limit=5.0 2023-03-08 02:06:40,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.486e+02 3.166e+02 4.039e+02 7.024e+02, threshold=6.332e+02, percent-clipped=0.0 2023-03-08 02:06:43,372 INFO [train2.py:809] (3/4) Epoch 9, batch 3750, loss[ctc_loss=0.1249, att_loss=0.2546, loss=0.2286, over 15351.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01187, over 35.00 utterances.], tot_loss[ctc_loss=0.1155, att_loss=0.2571, loss=0.2288, over 3269795.52 frames. utt_duration=1248 frames, utt_pad_proportion=0.0547, over 10494.95 utterances.], batch size: 35, lr: 1.18e-02, grad_scale: 16.0 2023-03-08 02:08:04,806 INFO [train2.py:809] (3/4) Epoch 9, batch 3800, loss[ctc_loss=0.1033, att_loss=0.2316, loss=0.206, over 15760.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.008547, over 38.00 utterances.], tot_loss[ctc_loss=0.1147, att_loss=0.2563, loss=0.228, over 3265636.37 frames. utt_duration=1251 frames, utt_pad_proportion=0.05406, over 10456.35 utterances.], batch size: 38, lr: 1.18e-02, grad_scale: 16.0 2023-03-08 02:08:18,171 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.18 vs. limit=5.0 2023-03-08 02:08:39,896 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2416, 3.9690, 3.3479, 3.8897, 4.1033, 3.9466, 3.1787, 4.4881], device='cuda:3'), covar=tensor([0.0882, 0.0410, 0.0941, 0.0447, 0.0507, 0.0581, 0.0746, 0.0441], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0175, 0.0205, 0.0169, 0.0221, 0.0207, 0.0178, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 02:08:56,079 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:09:23,353 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.475e+02 3.126e+02 3.794e+02 8.902e+02, threshold=6.253e+02, percent-clipped=4.0 2023-03-08 02:09:26,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 02:09:26,475 INFO [train2.py:809] (3/4) Epoch 9, batch 3850, loss[ctc_loss=0.118, att_loss=0.2635, loss=0.2344, over 16767.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005938, over 48.00 utterances.], tot_loss[ctc_loss=0.1144, att_loss=0.256, loss=0.2277, over 3264894.06 frames. utt_duration=1238 frames, utt_pad_proportion=0.0597, over 10565.22 utterances.], batch size: 48, lr: 1.18e-02, grad_scale: 16.0 2023-03-08 02:09:42,265 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:10:10,752 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:10:45,206 INFO [train2.py:809] (3/4) Epoch 9, batch 3900, loss[ctc_loss=0.1182, att_loss=0.2738, loss=0.2427, over 16640.00 frames. utt_duration=1418 frames, utt_pad_proportion=0.004393, over 47.00 utterances.], tot_loss[ctc_loss=0.1134, att_loss=0.2556, loss=0.2271, over 3272447.99 frames. utt_duration=1256 frames, utt_pad_proportion=0.05402, over 10434.69 utterances.], batch size: 47, lr: 1.18e-02, grad_scale: 16.0 2023-03-08 02:10:47,123 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:11:26,174 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:11:50,473 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:11:53,458 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7896, 1.5125, 2.5568, 1.5750, 2.5084, 2.5766, 2.8635, 2.3996], device='cuda:3'), covar=tensor([0.0534, 0.4745, 0.2704, 0.2552, 0.1089, 0.0990, 0.1318, 0.1394], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0085, 0.0089, 0.0076, 0.0075, 0.0069, 0.0080, 0.0064], device='cuda:3'), out_proj_covar=tensor([4.4676e-05, 5.5537e-05, 5.6898e-05, 4.8184e-05, 4.5065e-05, 4.6517e-05, 5.2994e-05, 4.4172e-05], device='cuda:3') 2023-03-08 02:12:00,556 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.688e+02 3.135e+02 3.909e+02 9.741e+02, threshold=6.269e+02, percent-clipped=3.0 2023-03-08 02:12:03,698 INFO [train2.py:809] (3/4) Epoch 9, batch 3950, loss[ctc_loss=0.09782, att_loss=0.2513, loss=0.2206, over 17053.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.007239, over 52.00 utterances.], tot_loss[ctc_loss=0.1137, att_loss=0.2559, loss=0.2275, over 3264176.59 frames. utt_duration=1241 frames, utt_pad_proportion=0.05906, over 10534.16 utterances.], batch size: 52, lr: 1.18e-02, grad_scale: 16.0 2023-03-08 02:12:22,932 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:12:24,431 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3847, 2.4950, 2.9985, 4.2076, 3.8623, 3.7774, 2.8052, 1.8139], device='cuda:3'), covar=tensor([0.0673, 0.2155, 0.1055, 0.0710, 0.0609, 0.0462, 0.1525, 0.2555], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0201, 0.0184, 0.0182, 0.0172, 0.0141, 0.0189, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:12:31,248 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.71 vs. limit=5.0 2023-03-08 02:12:38,955 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-08 02:13:22,878 INFO [train2.py:809] (3/4) Epoch 10, batch 0, loss[ctc_loss=0.1372, att_loss=0.2772, loss=0.2492, over 17078.00 frames. utt_duration=1291 frames, utt_pad_proportion=0.008063, over 53.00 utterances.], tot_loss[ctc_loss=0.1372, att_loss=0.2772, loss=0.2492, over 17078.00 frames. utt_duration=1291 frames, utt_pad_proportion=0.008063, over 53.00 utterances.], batch size: 53, lr: 1.12e-02, grad_scale: 16.0 2023-03-08 02:13:22,878 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 02:13:35,371 INFO [train2.py:843] (3/4) Epoch 10, validation: ctc_loss=0.0538, att_loss=0.2416, loss=0.204, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 02:13:35,372 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 02:13:43,518 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1480, 5.1322, 5.0184, 2.2751, 1.8663, 2.8111, 3.1813, 3.7991], device='cuda:3'), covar=tensor([0.0596, 0.0210, 0.0183, 0.4049, 0.6753, 0.2542, 0.1748, 0.1858], device='cuda:3'), in_proj_covar=tensor([0.0331, 0.0215, 0.0233, 0.0197, 0.0358, 0.0337, 0.0230, 0.0351], device='cuda:3'), out_proj_covar=tensor([1.5396e-04, 8.2625e-05, 1.0091e-04, 8.9824e-05, 1.5849e-04, 1.3898e-04, 9.0879e-05, 1.5177e-04], device='cuda:3') 2023-03-08 02:14:04,762 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:14:25,926 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:14:44,623 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4382, 2.5324, 5.0758, 3.8296, 2.9059, 4.3605, 4.9640, 4.6123], device='cuda:3'), covar=tensor([0.0245, 0.1648, 0.0158, 0.1006, 0.1941, 0.0228, 0.0090, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0236, 0.0127, 0.0292, 0.0269, 0.0177, 0.0107, 0.0140], device='cuda:3'), out_proj_covar=tensor([1.2982e-04, 1.9962e-04, 1.1520e-04, 2.4447e-04, 2.4150e-04, 1.5939e-04, 9.9017e-05, 1.3539e-04], device='cuda:3') 2023-03-08 02:14:46,068 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1627, 4.6962, 4.7485, 4.7998, 2.7652, 4.9456, 3.0301, 1.8665], device='cuda:3'), covar=tensor([0.0330, 0.0150, 0.0561, 0.0118, 0.1810, 0.0107, 0.1322, 0.1730], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0108, 0.0252, 0.0108, 0.0221, 0.0101, 0.0224, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:14:54,635 INFO [train2.py:809] (3/4) Epoch 10, batch 50, loss[ctc_loss=0.09709, att_loss=0.2435, loss=0.2142, over 16776.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.006007, over 48.00 utterances.], tot_loss[ctc_loss=0.1125, att_loss=0.2535, loss=0.2253, over 738806.24 frames. utt_duration=1279 frames, utt_pad_proportion=0.04761, over 2312.75 utterances.], batch size: 48, lr: 1.12e-02, grad_scale: 16.0 2023-03-08 02:15:18,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.508e+02 2.998e+02 4.058e+02 7.580e+02, threshold=5.995e+02, percent-clipped=4.0 2023-03-08 02:15:18,529 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1167, 4.7229, 4.4030, 4.6619, 2.3628, 4.8234, 2.7583, 1.9256], device='cuda:3'), covar=tensor([0.0338, 0.0129, 0.0809, 0.0133, 0.2380, 0.0106, 0.1590, 0.1858], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0107, 0.0252, 0.0108, 0.0221, 0.0101, 0.0224, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:16:15,455 INFO [train2.py:809] (3/4) Epoch 10, batch 100, loss[ctc_loss=0.1362, att_loss=0.2528, loss=0.2295, over 15508.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.00821, over 36.00 utterances.], tot_loss[ctc_loss=0.1152, att_loss=0.2561, loss=0.2279, over 1305846.94 frames. utt_duration=1217 frames, utt_pad_proportion=0.05738, over 4298.09 utterances.], batch size: 36, lr: 1.12e-02, grad_scale: 16.0 2023-03-08 02:16:37,958 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9424, 5.2007, 5.4179, 5.3945, 5.2768, 5.8883, 5.1568, 5.9219], device='cuda:3'), covar=tensor([0.0580, 0.0663, 0.0621, 0.0918, 0.1915, 0.0769, 0.0576, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0672, 0.0406, 0.0465, 0.0533, 0.0708, 0.0471, 0.0386, 0.0462], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 02:17:36,813 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 02:17:41,613 INFO [train2.py:809] (3/4) Epoch 10, batch 150, loss[ctc_loss=0.08877, att_loss=0.237, loss=0.2074, over 14483.00 frames. utt_duration=1812 frames, utt_pad_proportion=0.03524, over 32.00 utterances.], tot_loss[ctc_loss=0.1149, att_loss=0.2565, loss=0.2282, over 1740108.79 frames. utt_duration=1215 frames, utt_pad_proportion=0.06139, over 5737.51 utterances.], batch size: 32, lr: 1.12e-02, grad_scale: 16.0 2023-03-08 02:18:03,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.601e+02 3.072e+02 3.945e+02 7.612e+02, threshold=6.144e+02, percent-clipped=5.0 2023-03-08 02:18:22,449 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:18:41,296 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-08 02:18:52,126 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 02:19:00,382 INFO [train2.py:809] (3/4) Epoch 10, batch 200, loss[ctc_loss=0.1066, att_loss=0.2582, loss=0.2279, over 16544.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006092, over 45.00 utterances.], tot_loss[ctc_loss=0.1128, att_loss=0.2545, loss=0.2262, over 2079979.19 frames. utt_duration=1260 frames, utt_pad_proportion=0.05233, over 6610.08 utterances.], batch size: 45, lr: 1.12e-02, grad_scale: 16.0 2023-03-08 02:19:38,215 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:20:13,426 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3481, 5.6231, 5.0149, 5.4907, 5.1438, 4.9071, 5.0395, 4.8702], device='cuda:3'), covar=tensor([0.1334, 0.1003, 0.0879, 0.0795, 0.1030, 0.1372, 0.2471, 0.2532], device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0490, 0.0362, 0.0374, 0.0355, 0.0414, 0.0499, 0.0448], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 02:20:19,272 INFO [train2.py:809] (3/4) Epoch 10, batch 250, loss[ctc_loss=0.09684, att_loss=0.2519, loss=0.2209, over 16951.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007544, over 50.00 utterances.], tot_loss[ctc_loss=0.1134, att_loss=0.2558, loss=0.2273, over 2354604.85 frames. utt_duration=1260 frames, utt_pad_proportion=0.04804, over 7483.69 utterances.], batch size: 50, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:20:41,720 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.78 vs. limit=5.0 2023-03-08 02:20:42,222 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.693e+02 3.162e+02 4.102e+02 9.052e+02, threshold=6.324e+02, percent-clipped=7.0 2023-03-08 02:20:56,478 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:20:59,762 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2718, 2.4544, 3.0991, 2.4925, 3.0060, 3.4030, 3.3434, 2.6716], device='cuda:3'), covar=tensor([0.0500, 0.1473, 0.1030, 0.1118, 0.0988, 0.1053, 0.0639, 0.1251], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0220, 0.0236, 0.0200, 0.0232, 0.0280, 0.0208, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-08 02:21:39,367 INFO [train2.py:809] (3/4) Epoch 10, batch 300, loss[ctc_loss=0.176, att_loss=0.2886, loss=0.2661, over 14247.00 frames. utt_duration=389.1 frames, utt_pad_proportion=0.3174, over 147.00 utterances.], tot_loss[ctc_loss=0.1128, att_loss=0.2561, loss=0.2274, over 2562715.15 frames. utt_duration=1252 frames, utt_pad_proportion=0.05045, over 8198.52 utterances.], batch size: 147, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:22:01,545 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:22:23,350 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:22:31,049 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:22:34,230 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0826, 5.1013, 4.9431, 2.8884, 4.8014, 4.6979, 4.1917, 2.3917], device='cuda:3'), covar=tensor([0.0113, 0.0079, 0.0215, 0.0982, 0.0087, 0.0162, 0.0317, 0.1563], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0077, 0.0067, 0.0099, 0.0064, 0.0089, 0.0087, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 02:23:01,541 INFO [train2.py:809] (3/4) Epoch 10, batch 350, loss[ctc_loss=0.09882, att_loss=0.2327, loss=0.2059, over 15781.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.00829, over 38.00 utterances.], tot_loss[ctc_loss=0.1117, att_loss=0.2544, loss=0.2259, over 2704496.37 frames. utt_duration=1264 frames, utt_pad_proportion=0.05349, over 8566.50 utterances.], batch size: 38, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:23:24,552 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.621e+02 3.186e+02 4.132e+02 7.508e+02, threshold=6.371e+02, percent-clipped=3.0 2023-03-08 02:23:50,210 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:24:03,323 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:24:21,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-08 02:24:24,487 INFO [train2.py:809] (3/4) Epoch 10, batch 400, loss[ctc_loss=0.1405, att_loss=0.2678, loss=0.2423, over 17019.00 frames. utt_duration=689.1 frames, utt_pad_proportion=0.1365, over 99.00 utterances.], tot_loss[ctc_loss=0.1116, att_loss=0.2546, loss=0.226, over 2831140.22 frames. utt_duration=1228 frames, utt_pad_proportion=0.06115, over 9229.81 utterances.], batch size: 99, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:25:45,287 INFO [train2.py:809] (3/4) Epoch 10, batch 450, loss[ctc_loss=0.1082, att_loss=0.2574, loss=0.2275, over 16761.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006204, over 48.00 utterances.], tot_loss[ctc_loss=0.1111, att_loss=0.2542, loss=0.2256, over 2928831.86 frames. utt_duration=1235 frames, utt_pad_proportion=0.05984, over 9495.80 utterances.], batch size: 48, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:26:06,980 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.409e+02 2.979e+02 3.960e+02 1.144e+03, threshold=5.958e+02, percent-clipped=2.0 2023-03-08 02:27:03,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-08 02:27:03,697 INFO [train2.py:809] (3/4) Epoch 10, batch 500, loss[ctc_loss=0.08281, att_loss=0.2356, loss=0.2051, over 15974.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005814, over 41.00 utterances.], tot_loss[ctc_loss=0.1107, att_loss=0.2542, loss=0.2255, over 3007004.70 frames. utt_duration=1264 frames, utt_pad_proportion=0.05158, over 9527.75 utterances.], batch size: 41, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:27:04,493 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8660, 6.0945, 5.5855, 5.8753, 5.7478, 5.3681, 5.5679, 5.2181], device='cuda:3'), covar=tensor([0.1027, 0.0725, 0.0684, 0.0719, 0.0702, 0.1202, 0.1607, 0.2053], device='cuda:3'), in_proj_covar=tensor([0.0420, 0.0481, 0.0360, 0.0376, 0.0353, 0.0409, 0.0496, 0.0445], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 02:27:28,480 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-03-08 02:28:22,569 INFO [train2.py:809] (3/4) Epoch 10, batch 550, loss[ctc_loss=0.07628, att_loss=0.2195, loss=0.1909, over 15999.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.0079, over 40.00 utterances.], tot_loss[ctc_loss=0.1105, att_loss=0.2536, loss=0.225, over 3063471.97 frames. utt_duration=1276 frames, utt_pad_proportion=0.04884, over 9615.42 utterances.], batch size: 40, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:28:35,158 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5298, 3.7292, 3.6386, 3.0224, 3.7208, 3.6417, 3.5968, 2.0634], device='cuda:3'), covar=tensor([0.1378, 0.1553, 0.2871, 0.8321, 0.4450, 0.3022, 0.1013, 1.1768], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0109, 0.0115, 0.0179, 0.0097, 0.0163, 0.0096, 0.0165], device='cuda:3'), out_proj_covar=tensor([8.3710e-05, 9.4089e-05, 1.0255e-04, 1.4470e-04, 8.8069e-05, 1.3434e-04, 8.3482e-05, 1.3402e-04], device='cuda:3') 2023-03-08 02:28:45,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.390e+02 3.015e+02 3.600e+02 8.483e+02, threshold=6.031e+02, percent-clipped=3.0 2023-03-08 02:28:59,596 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:29:41,952 INFO [train2.py:809] (3/4) Epoch 10, batch 600, loss[ctc_loss=0.1119, att_loss=0.2402, loss=0.2145, over 15958.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006871, over 41.00 utterances.], tot_loss[ctc_loss=0.11, att_loss=0.2534, loss=0.2247, over 3110303.13 frames. utt_duration=1277 frames, utt_pad_proportion=0.0492, over 9753.04 utterances.], batch size: 41, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:29:55,768 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7016, 2.5674, 3.3387, 2.4516, 3.3207, 3.9757, 3.7669, 2.9657], device='cuda:3'), covar=tensor([0.0520, 0.1689, 0.0990, 0.1395, 0.0847, 0.0768, 0.0657, 0.1230], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0220, 0.0237, 0.0199, 0.0230, 0.0280, 0.0210, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2023-03-08 02:30:00,500 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3219, 4.6195, 4.7064, 4.8848, 2.8548, 4.7714, 3.0173, 2.4211], device='cuda:3'), covar=tensor([0.0235, 0.0153, 0.0561, 0.0134, 0.1543, 0.0138, 0.1333, 0.1451], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0105, 0.0247, 0.0107, 0.0216, 0.0101, 0.0223, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:30:03,592 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:30:15,930 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:31:02,501 INFO [train2.py:809] (3/4) Epoch 10, batch 650, loss[ctc_loss=0.131, att_loss=0.2744, loss=0.2457, over 16991.00 frames. utt_duration=1361 frames, utt_pad_proportion=0.006121, over 50.00 utterances.], tot_loss[ctc_loss=0.1102, att_loss=0.2532, loss=0.2246, over 3144519.33 frames. utt_duration=1278 frames, utt_pad_proportion=0.04818, over 9852.91 utterances.], batch size: 50, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:31:20,058 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:31:24,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.550e+02 3.093e+02 3.802e+02 8.175e+02, threshold=6.187e+02, percent-clipped=1.0 2023-03-08 02:31:53,765 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:32:06,661 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5334, 2.1773, 5.0318, 3.8802, 2.8646, 4.3448, 4.8911, 4.6412], device='cuda:3'), covar=tensor([0.0239, 0.1964, 0.0187, 0.1018, 0.1975, 0.0233, 0.0097, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0238, 0.0128, 0.0296, 0.0271, 0.0179, 0.0109, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 02:32:21,151 INFO [train2.py:809] (3/4) Epoch 10, batch 700, loss[ctc_loss=0.135, att_loss=0.2709, loss=0.2437, over 17313.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.03688, over 63.00 utterances.], tot_loss[ctc_loss=0.1111, att_loss=0.2544, loss=0.2258, over 3180873.19 frames. utt_duration=1276 frames, utt_pad_proportion=0.04615, over 9986.54 utterances.], batch size: 63, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:32:28,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-08 02:32:33,016 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0451, 4.3715, 4.0239, 4.6955, 2.4626, 4.3215, 2.7401, 1.7591], device='cuda:3'), covar=tensor([0.0304, 0.0129, 0.0909, 0.0134, 0.2098, 0.0175, 0.1698, 0.2033], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0106, 0.0251, 0.0107, 0.0220, 0.0102, 0.0225, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:33:41,080 INFO [train2.py:809] (3/4) Epoch 10, batch 750, loss[ctc_loss=0.1246, att_loss=0.2684, loss=0.2397, over 17063.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008287, over 52.00 utterances.], tot_loss[ctc_loss=0.1105, att_loss=0.2541, loss=0.2254, over 3205294.52 frames. utt_duration=1295 frames, utt_pad_proportion=0.04218, over 9910.68 utterances.], batch size: 52, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:33:43,682 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:33:48,483 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:33:55,280 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:34:04,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.573e+02 3.133e+02 3.667e+02 8.255e+02, threshold=6.267e+02, percent-clipped=5.0 2023-03-08 02:35:01,774 INFO [train2.py:809] (3/4) Epoch 10, batch 800, loss[ctc_loss=0.1121, att_loss=0.2646, loss=0.2341, over 17300.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02378, over 59.00 utterances.], tot_loss[ctc_loss=0.111, att_loss=0.2539, loss=0.2253, over 3212351.41 frames. utt_duration=1264 frames, utt_pad_proportion=0.05154, over 10180.20 utterances.], batch size: 59, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:35:21,866 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:35:26,436 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:35:32,478 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:36:15,424 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:36:18,469 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3584, 4.2229, 4.3132, 4.3942, 4.6420, 4.5011, 4.3609, 2.1963], device='cuda:3'), covar=tensor([0.0211, 0.0395, 0.0275, 0.0166, 0.1171, 0.0155, 0.0230, 0.2178], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0126, 0.0130, 0.0133, 0.0323, 0.0120, 0.0116, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 02:36:21,096 INFO [train2.py:809] (3/4) Epoch 10, batch 850, loss[ctc_loss=0.1179, att_loss=0.2587, loss=0.2306, over 16709.00 frames. utt_duration=683.4 frames, utt_pad_proportion=0.1436, over 98.00 utterances.], tot_loss[ctc_loss=0.1114, att_loss=0.2536, loss=0.2252, over 3224054.88 frames. utt_duration=1275 frames, utt_pad_proportion=0.0492, over 10122.66 utterances.], batch size: 98, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:36:33,418 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7962, 5.0371, 5.2697, 5.2334, 5.1464, 5.6790, 5.0648, 5.8411], device='cuda:3'), covar=tensor([0.0659, 0.0731, 0.0728, 0.0950, 0.1612, 0.0920, 0.0659, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0678, 0.0415, 0.0478, 0.0536, 0.0706, 0.0473, 0.0386, 0.0467], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 02:36:43,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.594e+02 3.302e+02 4.019e+02 1.127e+03, threshold=6.604e+02, percent-clipped=9.0 2023-03-08 02:37:41,457 INFO [train2.py:809] (3/4) Epoch 10, batch 900, loss[ctc_loss=0.0931, att_loss=0.2235, loss=0.1974, over 15498.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008801, over 36.00 utterances.], tot_loss[ctc_loss=0.1123, att_loss=0.2549, loss=0.2264, over 3237589.58 frames. utt_duration=1277 frames, utt_pad_proportion=0.0494, over 10154.69 utterances.], batch size: 36, lr: 1.11e-02, grad_scale: 16.0 2023-03-08 02:37:53,271 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 02:39:01,056 INFO [train2.py:809] (3/4) Epoch 10, batch 950, loss[ctc_loss=0.1047, att_loss=0.2496, loss=0.2207, over 15948.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007228, over 41.00 utterances.], tot_loss[ctc_loss=0.1122, att_loss=0.2553, loss=0.2267, over 3246842.22 frames. utt_duration=1274 frames, utt_pad_proportion=0.05027, over 10208.71 utterances.], batch size: 41, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:39:03,636 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 02:39:23,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.661e+02 3.552e+02 4.499e+02 1.125e+03, threshold=7.103e+02, percent-clipped=5.0 2023-03-08 02:39:40,420 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.24 vs. limit=5.0 2023-03-08 02:39:40,653 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-03-08 02:39:46,323 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3923, 2.6802, 3.5992, 2.7942, 3.4825, 4.6007, 4.3141, 3.2484], device='cuda:3'), covar=tensor([0.0395, 0.1927, 0.1265, 0.1537, 0.1109, 0.0565, 0.0539, 0.1362], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0228, 0.0243, 0.0202, 0.0239, 0.0291, 0.0214, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 02:39:54,066 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:40:22,145 INFO [train2.py:809] (3/4) Epoch 10, batch 1000, loss[ctc_loss=0.1047, att_loss=0.249, loss=0.2202, over 16171.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.00678, over 41.00 utterances.], tot_loss[ctc_loss=0.1111, att_loss=0.2552, loss=0.2264, over 3249163.21 frames. utt_duration=1248 frames, utt_pad_proportion=0.05611, over 10423.20 utterances.], batch size: 41, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:40:30,553 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1979, 4.6609, 4.3999, 4.7441, 2.4763, 4.6003, 2.5385, 1.9813], device='cuda:3'), covar=tensor([0.0278, 0.0137, 0.0778, 0.0191, 0.2007, 0.0154, 0.1704, 0.1718], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0106, 0.0247, 0.0108, 0.0216, 0.0101, 0.0224, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:40:41,862 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 02:41:11,562 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:41:43,595 INFO [train2.py:809] (3/4) Epoch 10, batch 1050, loss[ctc_loss=0.1127, att_loss=0.2599, loss=0.2305, over 16781.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.00583, over 48.00 utterances.], tot_loss[ctc_loss=0.1118, att_loss=0.2555, loss=0.2268, over 3248639.64 frames. utt_duration=1242 frames, utt_pad_proportion=0.06009, over 10474.85 utterances.], batch size: 48, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:41:55,897 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:41:57,619 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4365, 2.6010, 4.9684, 4.0442, 2.8381, 4.5109, 4.9838, 4.8015], device='cuda:3'), covar=tensor([0.0231, 0.1701, 0.0214, 0.0927, 0.2019, 0.0207, 0.0096, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0241, 0.0130, 0.0294, 0.0272, 0.0180, 0.0110, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 02:42:06,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.544e+02 2.981e+02 3.460e+02 1.238e+03, threshold=5.962e+02, percent-clipped=1.0 2023-03-08 02:43:05,654 INFO [train2.py:809] (3/4) Epoch 10, batch 1100, loss[ctc_loss=0.09842, att_loss=0.264, loss=0.2309, over 17302.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01212, over 55.00 utterances.], tot_loss[ctc_loss=0.1105, att_loss=0.255, loss=0.2261, over 3250639.46 frames. utt_duration=1246 frames, utt_pad_proportion=0.06031, over 10450.10 utterances.], batch size: 55, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:43:12,935 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:43:17,541 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:43:22,064 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:43:28,268 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:43:34,590 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:44:25,908 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-08 02:44:26,443 INFO [train2.py:809] (3/4) Epoch 10, batch 1150, loss[ctc_loss=0.1245, att_loss=0.2731, loss=0.2434, over 16890.00 frames. utt_duration=683.9 frames, utt_pad_proportion=0.1419, over 99.00 utterances.], tot_loss[ctc_loss=0.1101, att_loss=0.2541, loss=0.2253, over 3247807.87 frames. utt_duration=1253 frames, utt_pad_proportion=0.06022, over 10383.52 utterances.], batch size: 99, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:44:26,827 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:44:48,827 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.521e+02 2.891e+02 3.394e+02 5.463e+02, threshold=5.783e+02, percent-clipped=0.0 2023-03-08 02:44:49,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 02:44:50,883 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:45:46,917 INFO [train2.py:809] (3/4) Epoch 10, batch 1200, loss[ctc_loss=0.09177, att_loss=0.2448, loss=0.2142, over 16548.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005882, over 45.00 utterances.], tot_loss[ctc_loss=0.1093, att_loss=0.2531, loss=0.2243, over 3253498.33 frames. utt_duration=1273 frames, utt_pad_proportion=0.05431, over 10232.07 utterances.], batch size: 45, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:45:50,264 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 02:46:04,804 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:46:19,058 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1032, 5.1464, 5.0371, 2.4180, 1.9229, 2.7973, 3.3698, 3.7781], device='cuda:3'), covar=tensor([0.0639, 0.0205, 0.0198, 0.4056, 0.6155, 0.2737, 0.1709, 0.1922], device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0219, 0.0236, 0.0202, 0.0356, 0.0341, 0.0229, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.5387e-04, 8.3810e-05, 1.0197e-04, 9.1592e-05, 1.5749e-04, 1.4028e-04, 9.0227e-05, 1.5182e-04], device='cuda:3') 2023-03-08 02:46:32,880 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6854, 2.7645, 3.5205, 4.5796, 3.9984, 4.1059, 2.9236, 2.1502], device='cuda:3'), covar=tensor([0.0451, 0.1962, 0.0963, 0.0394, 0.0607, 0.0272, 0.1404, 0.2169], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0200, 0.0187, 0.0187, 0.0178, 0.0142, 0.0186, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:47:06,742 INFO [train2.py:809] (3/4) Epoch 10, batch 1250, loss[ctc_loss=0.1226, att_loss=0.2702, loss=0.2407, over 17049.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009071, over 52.00 utterances.], tot_loss[ctc_loss=0.1092, att_loss=0.253, loss=0.2243, over 3259777.08 frames. utt_duration=1293 frames, utt_pad_proportion=0.04779, over 10094.29 utterances.], batch size: 52, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:47:14,641 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7449, 5.9448, 5.2548, 5.7829, 5.5722, 5.1835, 5.4142, 5.2572], device='cuda:3'), covar=tensor([0.1098, 0.0926, 0.0907, 0.0723, 0.0748, 0.1551, 0.2229, 0.2157], device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0485, 0.0366, 0.0375, 0.0349, 0.0408, 0.0498, 0.0451], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 02:47:19,266 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 02:47:28,076 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9717, 5.2291, 5.2961, 5.2445, 5.3307, 5.2944, 4.9808, 4.7305], device='cuda:3'), covar=tensor([0.1078, 0.0567, 0.0239, 0.0436, 0.0303, 0.0299, 0.0309, 0.0325], device='cuda:3'), in_proj_covar=tensor([0.0441, 0.0281, 0.0228, 0.0262, 0.0329, 0.0351, 0.0278, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 02:47:29,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.541e+02 3.066e+02 3.675e+02 8.173e+02, threshold=6.132e+02, percent-clipped=8.0 2023-03-08 02:47:50,659 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1362, 5.2000, 5.0277, 2.4450, 2.0621, 2.8527, 3.7057, 3.8732], device='cuda:3'), covar=tensor([0.0652, 0.0229, 0.0208, 0.4144, 0.5928, 0.2622, 0.1394, 0.1811], device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0217, 0.0233, 0.0198, 0.0351, 0.0336, 0.0227, 0.0349], device='cuda:3'), out_proj_covar=tensor([1.5169e-04, 8.2815e-05, 1.0056e-04, 9.0457e-05, 1.5566e-04, 1.3823e-04, 8.9349e-05, 1.5011e-04], device='cuda:3') 2023-03-08 02:48:28,602 INFO [train2.py:809] (3/4) Epoch 10, batch 1300, loss[ctc_loss=0.09081, att_loss=0.2232, loss=0.1967, over 15519.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007476, over 36.00 utterances.], tot_loss[ctc_loss=0.1086, att_loss=0.2532, loss=0.2243, over 3271576.43 frames. utt_duration=1308 frames, utt_pad_proportion=0.04111, over 10013.58 utterances.], batch size: 36, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:48:40,488 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 02:48:57,906 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4493, 4.7769, 5.0737, 4.9438, 4.9170, 5.4114, 4.8714, 5.5457], device='cuda:3'), covar=tensor([0.0691, 0.0679, 0.0725, 0.1041, 0.1736, 0.0884, 0.0853, 0.0593], device='cuda:3'), in_proj_covar=tensor([0.0671, 0.0412, 0.0471, 0.0529, 0.0705, 0.0475, 0.0385, 0.0463], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 02:49:00,607 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 02:49:49,922 INFO [train2.py:809] (3/4) Epoch 10, batch 1350, loss[ctc_loss=0.1109, att_loss=0.2728, loss=0.2405, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005185, over 47.00 utterances.], tot_loss[ctc_loss=0.1087, att_loss=0.2532, loss=0.2243, over 3276015.19 frames. utt_duration=1312 frames, utt_pad_proportion=0.03932, over 9996.11 utterances.], batch size: 47, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:50:12,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.454e+02 3.108e+02 4.121e+02 7.457e+02, threshold=6.215e+02, percent-clipped=5.0 2023-03-08 02:51:02,248 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4592, 2.4924, 3.1724, 4.2911, 3.7721, 3.9218, 2.7676, 2.0319], device='cuda:3'), covar=tensor([0.0591, 0.2440, 0.1070, 0.0650, 0.0749, 0.0385, 0.1802, 0.2576], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0204, 0.0190, 0.0189, 0.0179, 0.0144, 0.0191, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:51:11,272 INFO [train2.py:809] (3/4) Epoch 10, batch 1400, loss[ctc_loss=0.1095, att_loss=0.2564, loss=0.227, over 16618.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005857, over 47.00 utterances.], tot_loss[ctc_loss=0.1084, att_loss=0.2529, loss=0.224, over 3270893.96 frames. utt_duration=1304 frames, utt_pad_proportion=0.04264, over 10042.38 utterances.], batch size: 47, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:51:23,183 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:51:27,900 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:51:32,389 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:51:33,995 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:51:34,060 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:52:19,852 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1850, 4.6273, 4.5125, 4.7594, 2.4815, 4.7939, 2.5462, 2.0002], device='cuda:3'), covar=tensor([0.0269, 0.0188, 0.0792, 0.0160, 0.2247, 0.0139, 0.1968, 0.2072], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0106, 0.0252, 0.0109, 0.0218, 0.0102, 0.0226, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 02:52:31,805 INFO [train2.py:809] (3/4) Epoch 10, batch 1450, loss[ctc_loss=0.1275, att_loss=0.2792, loss=0.2489, over 17042.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.007907, over 52.00 utterances.], tot_loss[ctc_loss=0.1091, att_loss=0.2534, loss=0.2245, over 3273521.73 frames. utt_duration=1294 frames, utt_pad_proportion=0.04395, over 10130.60 utterances.], batch size: 52, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:52:39,598 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:52:44,934 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:52:48,263 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:52:51,392 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:52:54,387 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.516e+02 3.005e+02 3.782e+02 1.128e+03, threshold=6.010e+02, percent-clipped=1.0 2023-03-08 02:53:11,956 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:53:51,708 INFO [train2.py:809] (3/4) Epoch 10, batch 1500, loss[ctc_loss=0.1184, att_loss=0.2744, loss=0.2432, over 16630.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005125, over 47.00 utterances.], tot_loss[ctc_loss=0.1098, att_loss=0.2537, loss=0.2249, over 3268191.61 frames. utt_duration=1270 frames, utt_pad_proportion=0.05258, over 10305.71 utterances.], batch size: 47, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:53:53,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-08 02:53:55,193 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:54:01,987 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:54:08,289 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:54:17,642 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9226, 5.2623, 4.7079, 5.3435, 4.6909, 5.0564, 5.3950, 5.2069], device='cuda:3'), covar=tensor([0.0494, 0.0287, 0.0800, 0.0215, 0.0463, 0.0188, 0.0206, 0.0140], device='cuda:3'), in_proj_covar=tensor([0.0320, 0.0251, 0.0309, 0.0240, 0.0252, 0.0196, 0.0234, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 02:54:19,342 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0765, 3.7219, 3.1376, 3.5249, 3.9334, 3.5173, 2.6090, 4.3884], device='cuda:3'), covar=tensor([0.0907, 0.0497, 0.0975, 0.0649, 0.0571, 0.0671, 0.0972, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0175, 0.0198, 0.0168, 0.0221, 0.0204, 0.0175, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 02:54:48,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 02:55:12,298 INFO [train2.py:809] (3/4) Epoch 10, batch 1550, loss[ctc_loss=0.07696, att_loss=0.2208, loss=0.192, over 14598.00 frames. utt_duration=1826 frames, utt_pad_proportion=0.03477, over 32.00 utterances.], tot_loss[ctc_loss=0.1108, att_loss=0.2548, loss=0.226, over 3274253.81 frames. utt_duration=1236 frames, utt_pad_proportion=0.05878, over 10605.63 utterances.], batch size: 32, lr: 1.10e-02, grad_scale: 16.0 2023-03-08 02:55:12,439 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:55:34,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.658e+02 3.084e+02 3.993e+02 7.629e+02, threshold=6.167e+02, percent-clipped=4.0 2023-03-08 02:55:45,987 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 02:56:32,393 INFO [train2.py:809] (3/4) Epoch 10, batch 1600, loss[ctc_loss=0.1099, att_loss=0.249, loss=0.2212, over 16688.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.005907, over 46.00 utterances.], tot_loss[ctc_loss=0.1112, att_loss=0.2549, loss=0.2261, over 3276063.12 frames. utt_duration=1245 frames, utt_pad_proportion=0.05507, over 10537.00 utterances.], batch size: 46, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 02:56:43,957 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 02:57:17,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-08 02:57:53,647 INFO [train2.py:809] (3/4) Epoch 10, batch 1650, loss[ctc_loss=0.1051, att_loss=0.2427, loss=0.2152, over 16120.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.005991, over 42.00 utterances.], tot_loss[ctc_loss=0.1111, att_loss=0.2545, loss=0.2259, over 3268790.90 frames. utt_duration=1212 frames, utt_pad_proportion=0.06436, over 10798.28 utterances.], batch size: 42, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 02:58:01,289 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 02:58:14,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.486e+02 2.891e+02 3.571e+02 1.016e+03, threshold=5.782e+02, percent-clipped=2.0 2023-03-08 02:59:12,749 INFO [train2.py:809] (3/4) Epoch 10, batch 1700, loss[ctc_loss=0.09259, att_loss=0.2474, loss=0.2164, over 16534.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.005811, over 45.00 utterances.], tot_loss[ctc_loss=0.111, att_loss=0.2551, loss=0.2262, over 3281287.62 frames. utt_duration=1219 frames, utt_pad_proportion=0.05956, over 10782.10 utterances.], batch size: 45, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 02:59:14,050 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-08 02:59:16,131 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3769, 4.6079, 4.1663, 4.6964, 4.1916, 4.3233, 4.7321, 4.5915], device='cuda:3'), covar=tensor([0.0512, 0.0291, 0.0809, 0.0239, 0.0423, 0.0360, 0.0209, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0246, 0.0308, 0.0238, 0.0251, 0.0195, 0.0230, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 02:59:33,331 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:00:23,735 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8701, 3.0252, 3.5195, 4.6937, 4.1417, 4.2018, 3.0297, 2.1126], device='cuda:3'), covar=tensor([0.0498, 0.2158, 0.1042, 0.0473, 0.0793, 0.0343, 0.1485, 0.2605], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0212, 0.0192, 0.0188, 0.0183, 0.0145, 0.0193, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 03:00:33,881 INFO [train2.py:809] (3/4) Epoch 10, batch 1750, loss[ctc_loss=0.1217, att_loss=0.2786, loss=0.2472, over 16963.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.006974, over 50.00 utterances.], tot_loss[ctc_loss=0.11, att_loss=0.2541, loss=0.2252, over 3274266.01 frames. utt_duration=1214 frames, utt_pad_proportion=0.06312, over 10798.34 utterances.], batch size: 50, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:00:49,735 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:00:51,059 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:00:55,647 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.443e+02 2.432e+02 2.940e+02 3.870e+02 6.231e+02, threshold=5.879e+02, percent-clipped=2.0 2023-03-08 03:01:05,305 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:01:53,789 INFO [train2.py:809] (3/4) Epoch 10, batch 1800, loss[ctc_loss=0.1016, att_loss=0.2592, loss=0.2277, over 17045.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.007806, over 53.00 utterances.], tot_loss[ctc_loss=0.1091, att_loss=0.2533, loss=0.2244, over 3275040.33 frames. utt_duration=1242 frames, utt_pad_proportion=0.05562, over 10561.25 utterances.], batch size: 53, lr: 1.09e-02, grad_scale: 32.0 2023-03-08 03:02:03,007 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:02:05,718 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:02:56,390 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1711, 5.1886, 5.0796, 2.4258, 2.0695, 2.8865, 3.0910, 3.8814], device='cuda:3'), covar=tensor([0.0656, 0.0189, 0.0204, 0.4036, 0.5949, 0.2540, 0.1960, 0.1700], device='cuda:3'), in_proj_covar=tensor([0.0335, 0.0218, 0.0235, 0.0201, 0.0354, 0.0336, 0.0231, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.5402e-04, 8.3185e-05, 1.0214e-04, 9.1767e-05, 1.5668e-04, 1.3751e-04, 9.0767e-05, 1.5212e-04], device='cuda:3') 2023-03-08 03:03:13,562 INFO [train2.py:809] (3/4) Epoch 10, batch 1850, loss[ctc_loss=0.1223, att_loss=0.2586, loss=0.2314, over 16574.00 frames. utt_duration=677.9 frames, utt_pad_proportion=0.1505, over 98.00 utterances.], tot_loss[ctc_loss=0.1095, att_loss=0.2534, loss=0.2246, over 3263284.95 frames. utt_duration=1227 frames, utt_pad_proportion=0.06065, over 10650.99 utterances.], batch size: 98, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:03:18,309 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1432, 5.4296, 5.4194, 5.3479, 5.5297, 5.5213, 5.2488, 5.0519], device='cuda:3'), covar=tensor([0.1020, 0.0425, 0.0198, 0.0418, 0.0235, 0.0246, 0.0292, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0435, 0.0278, 0.0229, 0.0266, 0.0326, 0.0346, 0.0273, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 03:03:19,738 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:03:36,152 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.414e+02 3.142e+02 4.365e+02 1.152e+03, threshold=6.284e+02, percent-clipped=6.0 2023-03-08 03:03:37,921 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:04:33,114 INFO [train2.py:809] (3/4) Epoch 10, batch 1900, loss[ctc_loss=0.1087, att_loss=0.2583, loss=0.2284, over 17020.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008684, over 51.00 utterances.], tot_loss[ctc_loss=0.1094, att_loss=0.2533, loss=0.2245, over 3269804.49 frames. utt_duration=1238 frames, utt_pad_proportion=0.05643, over 10576.97 utterances.], batch size: 51, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:05:33,009 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9158, 1.5415, 2.3207, 2.2942, 2.6719, 1.9811, 2.2032, 2.8790], device='cuda:3'), covar=tensor([0.1081, 0.5546, 0.3792, 0.1896, 0.1338, 0.1740, 0.2792, 0.1167], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0084, 0.0087, 0.0072, 0.0071, 0.0068, 0.0079, 0.0062], device='cuda:3'), out_proj_covar=tensor([4.5677e-05, 5.5449e-05, 5.6903e-05, 4.7291e-05, 4.4511e-05, 4.6645e-05, 5.3013e-05, 4.3241e-05], device='cuda:3') 2023-03-08 03:05:53,428 INFO [train2.py:809] (3/4) Epoch 10, batch 1950, loss[ctc_loss=0.1162, att_loss=0.2723, loss=0.2411, over 17021.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.01143, over 53.00 utterances.], tot_loss[ctc_loss=0.1094, att_loss=0.2537, loss=0.2248, over 3271688.94 frames. utt_duration=1241 frames, utt_pad_proportion=0.05304, over 10561.57 utterances.], batch size: 53, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:06:16,462 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.508e+02 3.059e+02 3.528e+02 6.330e+02, threshold=6.117e+02, percent-clipped=1.0 2023-03-08 03:07:12,865 INFO [train2.py:809] (3/4) Epoch 10, batch 2000, loss[ctc_loss=0.07477, att_loss=0.2205, loss=0.1914, over 12267.00 frames. utt_duration=1819 frames, utt_pad_proportion=0.0711, over 27.00 utterances.], tot_loss[ctc_loss=0.1089, att_loss=0.2534, loss=0.2245, over 3275941.20 frames. utt_duration=1259 frames, utt_pad_proportion=0.04721, over 10422.85 utterances.], batch size: 27, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:07:16,217 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0268, 5.0466, 4.8733, 2.9699, 4.7625, 4.4031, 4.1402, 2.4135], device='cuda:3'), covar=tensor([0.0118, 0.0089, 0.0195, 0.0952, 0.0104, 0.0205, 0.0344, 0.1626], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0078, 0.0068, 0.0099, 0.0066, 0.0089, 0.0088, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 03:07:43,182 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1077, 5.0910, 5.0009, 3.0546, 4.8756, 4.6873, 4.4407, 2.8178], device='cuda:3'), covar=tensor([0.0128, 0.0091, 0.0168, 0.0881, 0.0087, 0.0146, 0.0248, 0.1269], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0079, 0.0068, 0.0099, 0.0066, 0.0089, 0.0088, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 03:08:32,707 INFO [train2.py:809] (3/4) Epoch 10, batch 2050, loss[ctc_loss=0.09622, att_loss=0.2546, loss=0.2229, over 16621.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005588, over 47.00 utterances.], tot_loss[ctc_loss=0.1088, att_loss=0.2539, loss=0.2249, over 3286675.37 frames. utt_duration=1264 frames, utt_pad_proportion=0.04311, over 10409.42 utterances.], batch size: 47, lr: 1.09e-02, grad_scale: 16.0 2023-03-08 03:08:34,518 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8476, 5.1050, 4.6553, 5.2677, 4.5693, 4.8277, 5.2874, 5.0693], device='cuda:3'), covar=tensor([0.0568, 0.0292, 0.0790, 0.0222, 0.0455, 0.0256, 0.0210, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0321, 0.0253, 0.0319, 0.0242, 0.0259, 0.0199, 0.0236, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 03:08:37,708 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4559, 2.7390, 3.5449, 2.8702, 3.4001, 4.5949, 4.2204, 3.2513], device='cuda:3'), covar=tensor([0.0423, 0.1789, 0.1176, 0.1464, 0.1161, 0.0639, 0.0836, 0.1328], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0228, 0.0243, 0.0207, 0.0238, 0.0294, 0.0215, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:08:39,255 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5763, 2.7121, 3.2609, 4.4001, 3.9860, 4.0096, 2.7504, 1.6447], device='cuda:3'), covar=tensor([0.0594, 0.2357, 0.1101, 0.0590, 0.0804, 0.0441, 0.1708, 0.2649], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0204, 0.0187, 0.0185, 0.0180, 0.0143, 0.0187, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 03:08:55,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.394e+02 2.856e+02 3.386e+02 7.118e+02, threshold=5.711e+02, percent-clipped=3.0 2023-03-08 03:09:04,676 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:09:52,804 INFO [train2.py:809] (3/4) Epoch 10, batch 2100, loss[ctc_loss=0.1065, att_loss=0.2609, loss=0.23, over 17291.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01253, over 55.00 utterances.], tot_loss[ctc_loss=0.11, att_loss=0.2546, loss=0.2256, over 3282919.81 frames. utt_duration=1235 frames, utt_pad_proportion=0.05146, over 10648.27 utterances.], batch size: 55, lr: 1.09e-02, grad_scale: 8.0 2023-03-08 03:10:21,721 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:10:33,473 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:11:18,143 INFO [train2.py:809] (3/4) Epoch 10, batch 2150, loss[ctc_loss=0.1186, att_loss=0.238, loss=0.2141, over 16416.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006049, over 44.00 utterances.], tot_loss[ctc_loss=0.1105, att_loss=0.2547, loss=0.2259, over 3288291.06 frames. utt_duration=1238 frames, utt_pad_proportion=0.05002, over 10634.09 utterances.], batch size: 44, lr: 1.09e-02, grad_scale: 8.0 2023-03-08 03:11:18,463 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3147, 2.3753, 3.4354, 2.6691, 3.1432, 4.6419, 4.4033, 2.9246], device='cuda:3'), covar=tensor([0.0645, 0.2616, 0.1203, 0.1935, 0.1367, 0.0719, 0.0598, 0.2059], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0229, 0.0243, 0.0206, 0.0239, 0.0295, 0.0216, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:11:41,469 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2587, 5.2478, 5.1866, 2.4952, 1.9633, 2.4848, 3.7138, 3.9387], device='cuda:3'), covar=tensor([0.0595, 0.0236, 0.0216, 0.3522, 0.6336, 0.3136, 0.1318, 0.1607], device='cuda:3'), in_proj_covar=tensor([0.0330, 0.0217, 0.0233, 0.0200, 0.0354, 0.0335, 0.0227, 0.0351], device='cuda:3'), out_proj_covar=tensor([1.5146e-04, 8.2996e-05, 1.0077e-04, 9.0835e-05, 1.5626e-04, 1.3735e-04, 8.9333e-05, 1.5018e-04], device='cuda:3') 2023-03-08 03:11:42,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.860e+02 3.437e+02 3.901e+02 7.219e+02, threshold=6.874e+02, percent-clipped=5.0 2023-03-08 03:11:42,762 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:12:15,439 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:12:37,087 INFO [train2.py:809] (3/4) Epoch 10, batch 2200, loss[ctc_loss=0.1353, att_loss=0.2696, loss=0.2427, over 17424.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04663, over 69.00 utterances.], tot_loss[ctc_loss=0.1117, att_loss=0.2555, loss=0.2268, over 3283326.88 frames. utt_duration=1225 frames, utt_pad_proportion=0.05484, over 10737.79 utterances.], batch size: 69, lr: 1.09e-02, grad_scale: 8.0 2023-03-08 03:12:58,883 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:13:56,129 INFO [train2.py:809] (3/4) Epoch 10, batch 2250, loss[ctc_loss=0.09987, att_loss=0.2333, loss=0.2066, over 15523.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007444, over 36.00 utterances.], tot_loss[ctc_loss=0.1117, att_loss=0.2555, loss=0.2267, over 3285319.76 frames. utt_duration=1219 frames, utt_pad_proportion=0.05575, over 10794.81 utterances.], batch size: 36, lr: 1.09e-02, grad_scale: 8.0 2023-03-08 03:14:03,779 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6617, 5.9439, 5.3079, 5.7440, 5.5603, 5.1467, 5.3186, 5.0948], device='cuda:3'), covar=tensor([0.1268, 0.0825, 0.0797, 0.0704, 0.0840, 0.1446, 0.2276, 0.2424], device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0480, 0.0361, 0.0368, 0.0342, 0.0406, 0.0485, 0.0446], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 03:14:21,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.484e+02 3.268e+02 3.999e+02 7.414e+02, threshold=6.535e+02, percent-clipped=1.0 2023-03-08 03:15:04,951 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 03:15:14,535 INFO [train2.py:809] (3/4) Epoch 10, batch 2300, loss[ctc_loss=0.1376, att_loss=0.2716, loss=0.2448, over 17002.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009542, over 51.00 utterances.], tot_loss[ctc_loss=0.1118, att_loss=0.2554, loss=0.2267, over 3276773.88 frames. utt_duration=1211 frames, utt_pad_proportion=0.06031, over 10838.03 utterances.], batch size: 51, lr: 1.09e-02, grad_scale: 8.0 2023-03-08 03:15:21,365 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-08 03:15:48,212 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0209, 5.0243, 4.9518, 2.9591, 4.7474, 4.4597, 4.0738, 2.6134], device='cuda:3'), covar=tensor([0.0120, 0.0069, 0.0163, 0.0943, 0.0095, 0.0192, 0.0337, 0.1363], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0078, 0.0069, 0.0100, 0.0066, 0.0089, 0.0088, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 03:16:02,757 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-08 03:16:33,797 INFO [train2.py:809] (3/4) Epoch 10, batch 2350, loss[ctc_loss=0.0884, att_loss=0.2437, loss=0.2127, over 16172.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006596, over 41.00 utterances.], tot_loss[ctc_loss=0.1103, att_loss=0.2543, loss=0.2255, over 3275289.90 frames. utt_duration=1240 frames, utt_pad_proportion=0.0534, over 10579.46 utterances.], batch size: 41, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:16:59,114 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.404e+02 3.254e+02 4.140e+02 7.234e+02, threshold=6.508e+02, percent-clipped=3.0 2023-03-08 03:17:13,073 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6033, 5.1340, 4.9199, 5.0484, 4.9916, 4.7339, 3.5722, 5.0996], device='cuda:3'), covar=tensor([0.0103, 0.0112, 0.0110, 0.0082, 0.0089, 0.0124, 0.0688, 0.0189], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0070, 0.0083, 0.0051, 0.0056, 0.0066, 0.0088, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:17:13,931 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-03-08 03:17:53,628 INFO [train2.py:809] (3/4) Epoch 10, batch 2400, loss[ctc_loss=0.1112, att_loss=0.2353, loss=0.2105, over 16174.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007266, over 41.00 utterances.], tot_loss[ctc_loss=0.1113, att_loss=0.255, loss=0.2262, over 3273428.92 frames. utt_duration=1211 frames, utt_pad_proportion=0.0621, over 10828.35 utterances.], batch size: 41, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:19:14,660 INFO [train2.py:809] (3/4) Epoch 10, batch 2450, loss[ctc_loss=0.1141, att_loss=0.261, loss=0.2316, over 17353.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.05032, over 69.00 utterances.], tot_loss[ctc_loss=0.1106, att_loss=0.2546, loss=0.2258, over 3281885.25 frames. utt_duration=1226 frames, utt_pad_proportion=0.05568, over 10719.25 utterances.], batch size: 69, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:19:40,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.470e+02 3.021e+02 3.879e+02 8.501e+02, threshold=6.042e+02, percent-clipped=3.0 2023-03-08 03:20:05,111 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:20:34,413 INFO [train2.py:809] (3/4) Epoch 10, batch 2500, loss[ctc_loss=0.1087, att_loss=0.2365, loss=0.2109, over 16176.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.007113, over 41.00 utterances.], tot_loss[ctc_loss=0.1099, att_loss=0.2538, loss=0.225, over 3285169.14 frames. utt_duration=1251 frames, utt_pad_proportion=0.04893, over 10514.03 utterances.], batch size: 41, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:20:36,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 03:20:43,867 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:21:43,332 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-08 03:21:53,305 INFO [train2.py:809] (3/4) Epoch 10, batch 2550, loss[ctc_loss=0.1081, att_loss=0.2606, loss=0.2301, over 17043.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.008381, over 53.00 utterances.], tot_loss[ctc_loss=0.109, att_loss=0.2519, loss=0.2234, over 3265224.01 frames. utt_duration=1268 frames, utt_pad_proportion=0.05055, over 10310.45 utterances.], batch size: 53, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:22:18,520 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.719e+02 3.198e+02 3.977e+02 1.146e+03, threshold=6.397e+02, percent-clipped=5.0 2023-03-08 03:22:21,734 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:23:12,912 INFO [train2.py:809] (3/4) Epoch 10, batch 2600, loss[ctc_loss=0.07427, att_loss=0.2211, loss=0.1917, over 16185.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006454, over 41.00 utterances.], tot_loss[ctc_loss=0.1093, att_loss=0.252, loss=0.2234, over 3261684.65 frames. utt_duration=1252 frames, utt_pad_proportion=0.05597, over 10434.60 utterances.], batch size: 41, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:23:19,815 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-08 03:24:31,644 INFO [train2.py:809] (3/4) Epoch 10, batch 2650, loss[ctc_loss=0.1239, att_loss=0.2722, loss=0.2425, over 17313.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01069, over 55.00 utterances.], tot_loss[ctc_loss=0.109, att_loss=0.2519, loss=0.2234, over 3252380.43 frames. utt_duration=1231 frames, utt_pad_proportion=0.06388, over 10584.16 utterances.], batch size: 55, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:24:58,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.727e+02 3.206e+02 4.393e+02 9.975e+02, threshold=6.412e+02, percent-clipped=4.0 2023-03-08 03:25:51,260 INFO [train2.py:809] (3/4) Epoch 10, batch 2700, loss[ctc_loss=0.0922, att_loss=0.255, loss=0.2224, over 16634.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004811, over 47.00 utterances.], tot_loss[ctc_loss=0.1099, att_loss=0.2527, loss=0.2241, over 3250250.02 frames. utt_duration=1212 frames, utt_pad_proportion=0.06935, over 10737.16 utterances.], batch size: 47, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:26:01,434 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8456, 5.1354, 5.3824, 5.3528, 5.2678, 5.7925, 5.1863, 5.9128], device='cuda:3'), covar=tensor([0.0796, 0.0710, 0.0732, 0.1061, 0.2063, 0.0978, 0.0634, 0.0745], device='cuda:3'), in_proj_covar=tensor([0.0687, 0.0423, 0.0488, 0.0551, 0.0726, 0.0481, 0.0395, 0.0481], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:27:11,538 INFO [train2.py:809] (3/4) Epoch 10, batch 2750, loss[ctc_loss=0.1859, att_loss=0.2989, loss=0.2763, over 14173.00 frames. utt_duration=389.8 frames, utt_pad_proportion=0.3196, over 146.00 utterances.], tot_loss[ctc_loss=0.1092, att_loss=0.2526, loss=0.2239, over 3255440.22 frames. utt_duration=1224 frames, utt_pad_proportion=0.06592, over 10650.35 utterances.], batch size: 146, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:27:23,404 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3876, 4.0452, 3.4929, 3.8760, 4.1569, 3.8785, 3.1026, 4.5981], device='cuda:3'), covar=tensor([0.0807, 0.0393, 0.0959, 0.0585, 0.0576, 0.0588, 0.0771, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0176, 0.0201, 0.0168, 0.0226, 0.0206, 0.0175, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 03:27:32,195 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-03-08 03:27:38,143 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.648e+02 3.075e+02 3.913e+02 9.612e+02, threshold=6.151e+02, percent-clipped=3.0 2023-03-08 03:28:01,956 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:28:31,008 INFO [train2.py:809] (3/4) Epoch 10, batch 2800, loss[ctc_loss=0.0843, att_loss=0.2376, loss=0.207, over 16171.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006842, over 41.00 utterances.], tot_loss[ctc_loss=0.1088, att_loss=0.2523, loss=0.2236, over 3252507.03 frames. utt_duration=1254 frames, utt_pad_proportion=0.05957, over 10390.25 utterances.], batch size: 41, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:28:36,731 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6213, 5.0762, 4.9152, 5.0485, 5.1212, 4.7414, 3.8249, 4.9801], device='cuda:3'), covar=tensor([0.0092, 0.0104, 0.0084, 0.0072, 0.0079, 0.0105, 0.0518, 0.0201], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0071, 0.0084, 0.0052, 0.0056, 0.0068, 0.0090, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:28:39,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 03:29:08,976 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5878, 3.7361, 3.7639, 3.0456, 3.6950, 3.6961, 3.7107, 2.2823], device='cuda:3'), covar=tensor([0.1092, 0.1691, 0.1789, 0.6195, 0.1221, 0.6106, 0.0693, 0.8861], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0113, 0.0120, 0.0185, 0.0098, 0.0174, 0.0098, 0.0172], device='cuda:3'), out_proj_covar=tensor([8.9793e-05, 9.8611e-05, 1.0762e-04, 1.5046e-04, 9.0741e-05, 1.4418e-04, 8.6792e-05, 1.3987e-04], device='cuda:3') 2023-03-08 03:29:18,059 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:29:31,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-08 03:29:51,020 INFO [train2.py:809] (3/4) Epoch 10, batch 2850, loss[ctc_loss=0.07239, att_loss=0.2347, loss=0.2022, over 16114.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006422, over 42.00 utterances.], tot_loss[ctc_loss=0.1086, att_loss=0.2517, loss=0.2231, over 3257429.65 frames. utt_duration=1272 frames, utt_pad_proportion=0.05305, over 10256.75 utterances.], batch size: 42, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:30:11,790 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:30:17,872 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.597e+02 3.110e+02 4.075e+02 8.061e+02, threshold=6.220e+02, percent-clipped=4.0 2023-03-08 03:30:45,550 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-08 03:31:10,905 INFO [train2.py:809] (3/4) Epoch 10, batch 2900, loss[ctc_loss=0.1103, att_loss=0.2641, loss=0.2333, over 17280.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01333, over 55.00 utterances.], tot_loss[ctc_loss=0.1077, att_loss=0.2516, loss=0.2228, over 3263807.72 frames. utt_duration=1279 frames, utt_pad_proportion=0.04879, over 10219.30 utterances.], batch size: 55, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:32:30,745 INFO [train2.py:809] (3/4) Epoch 10, batch 2950, loss[ctc_loss=0.08514, att_loss=0.2469, loss=0.2145, over 16401.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007495, over 44.00 utterances.], tot_loss[ctc_loss=0.1087, att_loss=0.2521, loss=0.2234, over 3256012.26 frames. utt_duration=1279 frames, utt_pad_proportion=0.0506, over 10198.40 utterances.], batch size: 44, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:32:57,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.729e+02 3.322e+02 4.124e+02 1.749e+03, threshold=6.643e+02, percent-clipped=6.0 2023-03-08 03:32:59,449 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4482, 4.5877, 4.3758, 4.5273, 5.0138, 4.6146, 4.4862, 2.2584], device='cuda:3'), covar=tensor([0.0224, 0.0292, 0.0300, 0.0195, 0.0925, 0.0182, 0.0277, 0.2319], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0126, 0.0132, 0.0134, 0.0323, 0.0118, 0.0116, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 03:33:02,377 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:33:50,299 INFO [train2.py:809] (3/4) Epoch 10, batch 3000, loss[ctc_loss=0.113, att_loss=0.2704, loss=0.239, over 17142.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01378, over 56.00 utterances.], tot_loss[ctc_loss=0.1077, att_loss=0.252, loss=0.2232, over 3266381.06 frames. utt_duration=1289 frames, utt_pad_proportion=0.04482, over 10144.76 utterances.], batch size: 56, lr: 1.08e-02, grad_scale: 8.0 2023-03-08 03:33:50,299 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 03:34:06,507 INFO [train2.py:843] (3/4) Epoch 10, validation: ctc_loss=0.0536, att_loss=0.2406, loss=0.2032, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 03:34:06,508 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 03:34:25,620 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6396, 3.8039, 3.6973, 3.1358, 3.6863, 3.8429, 3.6455, 2.4581], device='cuda:3'), covar=tensor([0.1320, 0.1199, 0.2875, 0.6293, 0.1417, 0.3428, 0.0814, 0.9412], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0112, 0.0119, 0.0184, 0.0096, 0.0171, 0.0099, 0.0170], device='cuda:3'), out_proj_covar=tensor([8.9696e-05, 9.7569e-05, 1.0708e-04, 1.4928e-04, 8.9068e-05, 1.4188e-04, 8.7805e-05, 1.3830e-04], device='cuda:3') 2023-03-08 03:34:54,874 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:35:15,592 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0904, 4.5867, 4.5299, 4.4806, 4.6209, 4.4052, 2.8980, 4.4289], device='cuda:3'), covar=tensor([0.0114, 0.0118, 0.0106, 0.0093, 0.0097, 0.0104, 0.0879, 0.0210], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0069, 0.0082, 0.0051, 0.0055, 0.0066, 0.0088, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:35:26,806 INFO [train2.py:809] (3/4) Epoch 10, batch 3050, loss[ctc_loss=0.1679, att_loss=0.287, loss=0.2632, over 14137.00 frames. utt_duration=391.6 frames, utt_pad_proportion=0.3202, over 145.00 utterances.], tot_loss[ctc_loss=0.1075, att_loss=0.2516, loss=0.2228, over 3263401.83 frames. utt_duration=1272 frames, utt_pad_proportion=0.04932, over 10277.05 utterances.], batch size: 145, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:35:53,206 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.391e+02 2.901e+02 3.646e+02 9.557e+02, threshold=5.803e+02, percent-clipped=2.0 2023-03-08 03:36:30,283 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-03-08 03:36:47,159 INFO [train2.py:809] (3/4) Epoch 10, batch 3100, loss[ctc_loss=0.1015, att_loss=0.2512, loss=0.2212, over 16768.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006333, over 48.00 utterances.], tot_loss[ctc_loss=0.1074, att_loss=0.2522, loss=0.2232, over 3268517.72 frames. utt_duration=1261 frames, utt_pad_proportion=0.05189, over 10379.75 utterances.], batch size: 48, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:36:51,854 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 03:38:08,121 INFO [train2.py:809] (3/4) Epoch 10, batch 3150, loss[ctc_loss=0.1295, att_loss=0.2694, loss=0.2414, over 17048.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008984, over 52.00 utterances.], tot_loss[ctc_loss=0.1075, att_loss=0.2526, loss=0.2236, over 3260321.41 frames. utt_duration=1239 frames, utt_pad_proportion=0.05954, over 10537.06 utterances.], batch size: 52, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:38:28,470 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:38:29,017 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-08 03:38:34,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.629e+02 3.270e+02 4.032e+02 9.981e+02, threshold=6.541e+02, percent-clipped=6.0 2023-03-08 03:39:27,538 INFO [train2.py:809] (3/4) Epoch 10, batch 3200, loss[ctc_loss=0.09087, att_loss=0.2531, loss=0.2207, over 17332.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03496, over 63.00 utterances.], tot_loss[ctc_loss=0.1072, att_loss=0.2523, loss=0.2233, over 3254845.77 frames. utt_duration=1241 frames, utt_pad_proportion=0.06166, over 10505.25 utterances.], batch size: 63, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:39:44,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:40:34,906 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5922, 2.9599, 3.6365, 2.8125, 3.5499, 4.6443, 4.4099, 3.2849], device='cuda:3'), covar=tensor([0.0315, 0.1503, 0.0853, 0.1460, 0.0865, 0.0666, 0.0463, 0.1243], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0220, 0.0235, 0.0199, 0.0231, 0.0288, 0.0208, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:40:48,182 INFO [train2.py:809] (3/4) Epoch 10, batch 3250, loss[ctc_loss=0.1165, att_loss=0.2649, loss=0.2353, over 16756.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.007073, over 48.00 utterances.], tot_loss[ctc_loss=0.108, att_loss=0.2526, loss=0.2237, over 3254307.71 frames. utt_duration=1237 frames, utt_pad_proportion=0.06248, over 10534.20 utterances.], batch size: 48, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:41:13,727 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.584e+02 3.081e+02 4.035e+02 1.100e+03, threshold=6.162e+02, percent-clipped=4.0 2023-03-08 03:41:36,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-03-08 03:42:08,484 INFO [train2.py:809] (3/4) Epoch 10, batch 3300, loss[ctc_loss=0.1092, att_loss=0.2375, loss=0.2118, over 15641.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008016, over 37.00 utterances.], tot_loss[ctc_loss=0.1095, att_loss=0.254, loss=0.2251, over 3259071.70 frames. utt_duration=1210 frames, utt_pad_proportion=0.06817, over 10784.70 utterances.], batch size: 37, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:42:44,531 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3008, 2.6103, 3.5206, 2.8731, 3.3172, 4.3063, 4.1057, 3.2003], device='cuda:3'), covar=tensor([0.0352, 0.1813, 0.1043, 0.1410, 0.1041, 0.0773, 0.0514, 0.1233], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0225, 0.0239, 0.0203, 0.0235, 0.0294, 0.0212, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:42:48,933 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:42:58,392 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:43:29,379 INFO [train2.py:809] (3/4) Epoch 10, batch 3350, loss[ctc_loss=0.1381, att_loss=0.2724, loss=0.2455, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005229, over 47.00 utterances.], tot_loss[ctc_loss=0.1102, att_loss=0.2546, loss=0.2257, over 3255058.91 frames. utt_duration=1213 frames, utt_pad_proportion=0.06887, over 10743.77 utterances.], batch size: 47, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:43:47,663 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5680, 2.2691, 4.9784, 3.7809, 2.9071, 4.3442, 4.7289, 4.5247], device='cuda:3'), covar=tensor([0.0230, 0.1948, 0.0126, 0.1075, 0.1991, 0.0244, 0.0087, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0248, 0.0133, 0.0307, 0.0276, 0.0187, 0.0114, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 03:43:54,947 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.443e+02 2.910e+02 3.809e+02 9.404e+02, threshold=5.819e+02, percent-clipped=2.0 2023-03-08 03:44:28,162 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2023-03-08 03:44:37,706 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:44:49,735 INFO [train2.py:809] (3/4) Epoch 10, batch 3400, loss[ctc_loss=0.09786, att_loss=0.2592, loss=0.2269, over 17170.00 frames. utt_duration=1228 frames, utt_pad_proportion=0.01202, over 56.00 utterances.], tot_loss[ctc_loss=0.1093, att_loss=0.254, loss=0.225, over 3263016.52 frames. utt_duration=1233 frames, utt_pad_proportion=0.06195, over 10596.41 utterances.], batch size: 56, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:45:37,897 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:46:11,199 INFO [train2.py:809] (3/4) Epoch 10, batch 3450, loss[ctc_loss=0.1528, att_loss=0.2827, loss=0.2567, over 17285.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01276, over 55.00 utterances.], tot_loss[ctc_loss=0.1091, att_loss=0.2529, loss=0.2241, over 3259980.78 frames. utt_duration=1247 frames, utt_pad_proportion=0.05854, over 10465.64 utterances.], batch size: 55, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:46:36,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.326e+02 2.969e+02 3.835e+02 9.960e+02, threshold=5.938e+02, percent-clipped=3.0 2023-03-08 03:47:16,368 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:47:31,575 INFO [train2.py:809] (3/4) Epoch 10, batch 3500, loss[ctc_loss=0.1226, att_loss=0.2691, loss=0.2398, over 17021.00 frames. utt_duration=1311 frames, utt_pad_proportion=0.01001, over 52.00 utterances.], tot_loss[ctc_loss=0.109, att_loss=0.2532, loss=0.2244, over 3259330.21 frames. utt_duration=1239 frames, utt_pad_proportion=0.06038, over 10537.86 utterances.], batch size: 52, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:47:38,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-08 03:48:14,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.80 vs. limit=5.0 2023-03-08 03:48:52,043 INFO [train2.py:809] (3/4) Epoch 10, batch 3550, loss[ctc_loss=0.1285, att_loss=0.276, loss=0.2465, over 17321.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.0233, over 59.00 utterances.], tot_loss[ctc_loss=0.109, att_loss=0.253, loss=0.2242, over 3250367.89 frames. utt_duration=1247 frames, utt_pad_proportion=0.0603, over 10441.99 utterances.], batch size: 59, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:49:17,422 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.370e+02 2.861e+02 3.451e+02 5.590e+02, threshold=5.722e+02, percent-clipped=0.0 2023-03-08 03:49:54,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.46 vs. limit=5.0 2023-03-08 03:50:11,854 INFO [train2.py:809] (3/4) Epoch 10, batch 3600, loss[ctc_loss=0.1046, att_loss=0.2766, loss=0.2422, over 16774.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005434, over 48.00 utterances.], tot_loss[ctc_loss=0.1087, att_loss=0.253, loss=0.2241, over 3254299.73 frames. utt_duration=1275 frames, utt_pad_proportion=0.0519, over 10218.76 utterances.], batch size: 48, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:50:27,187 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:50:52,080 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:51:02,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 03:51:14,106 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6526, 2.5538, 4.8700, 3.7234, 2.8459, 4.2945, 4.7406, 4.5299], device='cuda:3'), covar=tensor([0.0173, 0.1740, 0.0141, 0.1173, 0.2057, 0.0231, 0.0096, 0.0198], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0249, 0.0133, 0.0308, 0.0275, 0.0188, 0.0114, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 03:51:32,918 INFO [train2.py:809] (3/4) Epoch 10, batch 3650, loss[ctc_loss=0.0964, att_loss=0.2483, loss=0.2179, over 17012.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008946, over 51.00 utterances.], tot_loss[ctc_loss=0.1089, att_loss=0.2536, loss=0.2247, over 3267373.04 frames. utt_duration=1258 frames, utt_pad_proportion=0.05307, over 10398.16 utterances.], batch size: 51, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:51:57,849 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 2.407e+02 3.049e+02 3.792e+02 1.207e+03, threshold=6.099e+02, percent-clipped=6.0 2023-03-08 03:52:04,626 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:52:09,224 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:52:31,784 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:52:53,643 INFO [train2.py:809] (3/4) Epoch 10, batch 3700, loss[ctc_loss=0.1132, att_loss=0.2564, loss=0.2277, over 17473.00 frames. utt_duration=1111 frames, utt_pad_proportion=0.02915, over 63.00 utterances.], tot_loss[ctc_loss=0.1085, att_loss=0.2539, loss=0.2248, over 3274778.24 frames. utt_duration=1242 frames, utt_pad_proportion=0.05526, over 10555.67 utterances.], batch size: 63, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:54:12,931 INFO [train2.py:809] (3/4) Epoch 10, batch 3750, loss[ctc_loss=0.09442, att_loss=0.264, loss=0.23, over 16872.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.008046, over 49.00 utterances.], tot_loss[ctc_loss=0.108, att_loss=0.2532, loss=0.2242, over 3271392.01 frames. utt_duration=1241 frames, utt_pad_proportion=0.05675, over 10556.56 utterances.], batch size: 49, lr: 1.07e-02, grad_scale: 8.0 2023-03-08 03:54:25,579 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-08 03:54:38,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.452e+02 2.346e+02 2.722e+02 3.425e+02 7.090e+02, threshold=5.444e+02, percent-clipped=1.0 2023-03-08 03:55:08,146 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 03:55:09,580 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:55:33,233 INFO [train2.py:809] (3/4) Epoch 10, batch 3800, loss[ctc_loss=0.1223, att_loss=0.2644, loss=0.236, over 17465.00 frames. utt_duration=885.9 frames, utt_pad_proportion=0.07142, over 79.00 utterances.], tot_loss[ctc_loss=0.1078, att_loss=0.2531, loss=0.2241, over 3272313.06 frames. utt_duration=1246 frames, utt_pad_proportion=0.05434, over 10519.47 utterances.], batch size: 79, lr: 1.06e-02, grad_scale: 8.0 2023-03-08 03:56:46,644 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 03:56:53,082 INFO [train2.py:809] (3/4) Epoch 10, batch 3850, loss[ctc_loss=0.08981, att_loss=0.2259, loss=0.1987, over 15894.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008183, over 39.00 utterances.], tot_loss[ctc_loss=0.1069, att_loss=0.2522, loss=0.2231, over 3275348.52 frames. utt_duration=1274 frames, utt_pad_proportion=0.04722, over 10295.50 utterances.], batch size: 39, lr: 1.06e-02, grad_scale: 8.0 2023-03-08 03:56:54,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-03-08 03:56:59,524 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2707, 5.2662, 5.1383, 2.9440, 5.0161, 4.8354, 4.4920, 2.9032], device='cuda:3'), covar=tensor([0.0140, 0.0107, 0.0215, 0.1132, 0.0107, 0.0175, 0.0333, 0.1492], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0082, 0.0074, 0.0102, 0.0069, 0.0093, 0.0093, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 03:57:07,155 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5804, 5.0194, 4.7661, 4.9299, 5.0569, 4.7446, 3.9124, 4.9028], device='cuda:3'), covar=tensor([0.0105, 0.0089, 0.0115, 0.0071, 0.0091, 0.0105, 0.0516, 0.0217], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0071, 0.0086, 0.0053, 0.0057, 0.0068, 0.0091, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:57:18,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.275e+02 2.627e+02 3.517e+02 7.871e+02, threshold=5.254e+02, percent-clipped=5.0 2023-03-08 03:58:10,237 INFO [train2.py:809] (3/4) Epoch 10, batch 3900, loss[ctc_loss=0.1689, att_loss=0.2857, loss=0.2623, over 13830.00 frames. utt_duration=383.2 frames, utt_pad_proportion=0.3336, over 145.00 utterances.], tot_loss[ctc_loss=0.1082, att_loss=0.2535, loss=0.2244, over 3273442.92 frames. utt_duration=1249 frames, utt_pad_proportion=0.05392, over 10496.30 utterances.], batch size: 145, lr: 1.06e-02, grad_scale: 8.0 2023-03-08 03:58:30,703 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:59:26,951 INFO [train2.py:809] (3/4) Epoch 10, batch 3950, loss[ctc_loss=0.1433, att_loss=0.2821, loss=0.2543, over 16886.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006744, over 49.00 utterances.], tot_loss[ctc_loss=0.1077, att_loss=0.2533, loss=0.2242, over 3278749.13 frames. utt_duration=1267 frames, utt_pad_proportion=0.04936, over 10367.18 utterances.], batch size: 49, lr: 1.06e-02, grad_scale: 8.0 2023-03-08 03:59:38,023 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6896, 5.2552, 4.9852, 5.1875, 5.2593, 5.0254, 3.9662, 5.1911], device='cuda:3'), covar=tensor([0.0096, 0.0077, 0.0090, 0.0066, 0.0080, 0.0070, 0.0504, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0069, 0.0083, 0.0051, 0.0055, 0.0066, 0.0089, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 03:59:44,056 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:59:50,030 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 03:59:51,184 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.608e+02 3.275e+02 3.896e+02 7.798e+02, threshold=6.551e+02, percent-clipped=9.0 2023-03-08 04:00:04,092 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:00:45,409 INFO [train2.py:809] (3/4) Epoch 11, batch 0, loss[ctc_loss=0.1087, att_loss=0.2628, loss=0.232, over 16464.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007422, over 46.00 utterances.], tot_loss[ctc_loss=0.1087, att_loss=0.2628, loss=0.232, over 16464.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007422, over 46.00 utterances.], batch size: 46, lr: 1.01e-02, grad_scale: 8.0 2023-03-08 04:00:45,409 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 04:00:57,591 INFO [train2.py:843] (3/4) Epoch 11, validation: ctc_loss=0.05063, att_loss=0.2383, loss=0.2008, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 04:00:57,592 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 04:01:02,352 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:01:58,660 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:02:18,274 INFO [train2.py:809] (3/4) Epoch 11, batch 50, loss[ctc_loss=0.09756, att_loss=0.2281, loss=0.202, over 15487.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009648, over 36.00 utterances.], tot_loss[ctc_loss=0.1053, att_loss=0.2521, loss=0.2227, over 736375.68 frames. utt_duration=1134 frames, utt_pad_proportion=0.08049, over 2600.42 utterances.], batch size: 36, lr: 1.01e-02, grad_scale: 8.0 2023-03-08 04:02:19,875 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:03:08,052 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.626e+02 3.075e+02 3.847e+02 1.291e+03, threshold=6.150e+02, percent-clipped=4.0 2023-03-08 04:03:37,055 INFO [train2.py:809] (3/4) Epoch 11, batch 100, loss[ctc_loss=0.09284, att_loss=0.2247, loss=0.1983, over 15388.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.009931, over 35.00 utterances.], tot_loss[ctc_loss=0.1046, att_loss=0.2507, loss=0.2215, over 1294470.71 frames. utt_duration=1236 frames, utt_pad_proportion=0.06289, over 4194.78 utterances.], batch size: 35, lr: 1.01e-02, grad_scale: 8.0 2023-03-08 04:03:40,299 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:03:58,311 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 04:04:51,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5989, 1.5677, 2.5177, 1.8124, 2.9008, 2.7095, 1.4805, 2.4214], device='cuda:3'), covar=tensor([0.0841, 0.7447, 0.3663, 0.3508, 0.1972, 0.2809, 0.5177, 0.1626], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0091, 0.0089, 0.0075, 0.0075, 0.0071, 0.0087, 0.0062], device='cuda:3'), out_proj_covar=tensor([4.8552e-05, 6.0457e-05, 6.0091e-05, 5.0651e-05, 4.8211e-05, 4.9243e-05, 5.7960e-05, 4.4608e-05], device='cuda:3') 2023-03-08 04:04:55,826 INFO [train2.py:809] (3/4) Epoch 11, batch 150, loss[ctc_loss=0.1067, att_loss=0.2608, loss=0.2299, over 17034.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007797, over 51.00 utterances.], tot_loss[ctc_loss=0.1045, att_loss=0.2495, loss=0.2205, over 1728730.47 frames. utt_duration=1286 frames, utt_pad_proportion=0.05004, over 5383.01 utterances.], batch size: 51, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:04:55,960 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:05:07,709 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 04:05:51,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.498e+02 3.139e+02 4.301e+02 8.550e+02, threshold=6.278e+02, percent-clipped=5.0 2023-03-08 04:06:20,489 INFO [train2.py:809] (3/4) Epoch 11, batch 200, loss[ctc_loss=0.09643, att_loss=0.2167, loss=0.1926, over 15376.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01091, over 35.00 utterances.], tot_loss[ctc_loss=0.1064, att_loss=0.2517, loss=0.2226, over 2073121.13 frames. utt_duration=1235 frames, utt_pad_proportion=0.05874, over 6721.09 utterances.], batch size: 35, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:06:40,171 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-08 04:06:46,020 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-08 04:07:32,190 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6386, 2.4067, 4.9818, 3.9139, 3.0270, 4.4056, 4.9417, 4.5944], device='cuda:3'), covar=tensor([0.0200, 0.1803, 0.0124, 0.1081, 0.1860, 0.0226, 0.0076, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0245, 0.0134, 0.0306, 0.0274, 0.0187, 0.0115, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 04:07:39,777 INFO [train2.py:809] (3/4) Epoch 11, batch 250, loss[ctc_loss=0.1053, att_loss=0.2615, loss=0.2302, over 17256.00 frames. utt_duration=1097 frames, utt_pad_proportion=0.04018, over 63.00 utterances.], tot_loss[ctc_loss=0.1065, att_loss=0.2516, loss=0.2226, over 2325341.45 frames. utt_duration=1228 frames, utt_pad_proportion=0.06593, over 7586.22 utterances.], batch size: 63, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:08:28,152 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:08:29,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.404e+02 2.494e+02 2.990e+02 3.866e+02 8.648e+02, threshold=5.979e+02, percent-clipped=3.0 2023-03-08 04:08:34,924 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:08:59,716 INFO [train2.py:809] (3/4) Epoch 11, batch 300, loss[ctc_loss=0.08653, att_loss=0.2272, loss=0.1991, over 16189.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.005599, over 41.00 utterances.], tot_loss[ctc_loss=0.1059, att_loss=0.251, loss=0.222, over 2525194.35 frames. utt_duration=1242 frames, utt_pad_proportion=0.06326, over 8142.52 utterances.], batch size: 41, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:09:45,568 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:09:52,693 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:10:05,827 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2096, 5.1768, 5.0836, 3.1492, 5.0495, 4.9005, 4.7128, 2.9646], device='cuda:3'), covar=tensor([0.0109, 0.0094, 0.0229, 0.0885, 0.0080, 0.0154, 0.0247, 0.1250], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0083, 0.0074, 0.0103, 0.0068, 0.0094, 0.0092, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 04:10:13,508 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-08 04:10:20,965 INFO [train2.py:809] (3/4) Epoch 11, batch 350, loss[ctc_loss=0.1126, att_loss=0.262, loss=0.2321, over 17070.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.00703, over 52.00 utterances.], tot_loss[ctc_loss=0.1046, att_loss=0.2502, loss=0.2211, over 2696351.72 frames. utt_duration=1275 frames, utt_pad_proportion=0.05255, over 8469.37 utterances.], batch size: 52, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:10:31,941 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:10:38,794 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-08 04:11:08,646 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:11:11,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.384e+02 2.929e+02 3.608e+02 1.665e+03, threshold=5.857e+02, percent-clipped=2.0 2023-03-08 04:11:20,257 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7211, 2.0427, 4.9260, 3.6726, 2.9314, 4.3111, 4.8454, 4.5364], device='cuda:3'), covar=tensor([0.0181, 0.2158, 0.0144, 0.1216, 0.2019, 0.0261, 0.0090, 0.0244], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0247, 0.0134, 0.0308, 0.0277, 0.0188, 0.0115, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 04:11:41,140 INFO [train2.py:809] (3/4) Epoch 11, batch 400, loss[ctc_loss=0.1118, att_loss=0.2377, loss=0.2125, over 15776.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.0084, over 38.00 utterances.], tot_loss[ctc_loss=0.1051, att_loss=0.2512, loss=0.222, over 2823828.38 frames. utt_duration=1263 frames, utt_pad_proportion=0.05433, over 8952.31 utterances.], batch size: 38, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:12:09,279 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8537, 5.0835, 5.3893, 5.3048, 5.2210, 5.7608, 5.1118, 5.8675], device='cuda:3'), covar=tensor([0.0524, 0.0649, 0.0602, 0.0963, 0.1627, 0.0812, 0.0607, 0.0566], device='cuda:3'), in_proj_covar=tensor([0.0693, 0.0419, 0.0483, 0.0550, 0.0727, 0.0487, 0.0395, 0.0477], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:12:09,458 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:12:36,229 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:12:47,333 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:13:01,947 INFO [train2.py:809] (3/4) Epoch 11, batch 450, loss[ctc_loss=0.08351, att_loss=0.248, loss=0.2151, over 16759.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.007058, over 48.00 utterances.], tot_loss[ctc_loss=0.1049, att_loss=0.2514, loss=0.2221, over 2925511.90 frames. utt_duration=1277 frames, utt_pad_proportion=0.04933, over 9172.52 utterances.], batch size: 48, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:13:12,963 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 04:13:52,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.417e+02 2.903e+02 3.339e+02 6.124e+02, threshold=5.806e+02, percent-clipped=1.0 2023-03-08 04:14:08,005 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:14:13,299 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:14:21,408 INFO [train2.py:809] (3/4) Epoch 11, batch 500, loss[ctc_loss=0.1048, att_loss=0.2634, loss=0.2317, over 17031.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.006555, over 51.00 utterances.], tot_loss[ctc_loss=0.1052, att_loss=0.2516, loss=0.2224, over 3006502.38 frames. utt_duration=1256 frames, utt_pad_proportion=0.05374, over 9582.81 utterances.], batch size: 51, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:14:29,167 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 04:14:45,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-08 04:15:33,681 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7010, 2.8196, 3.3680, 4.4069, 4.0349, 4.0073, 2.7247, 2.0341], device='cuda:3'), covar=tensor([0.0535, 0.2407, 0.1148, 0.0758, 0.0725, 0.0416, 0.1752, 0.2839], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0207, 0.0187, 0.0190, 0.0183, 0.0147, 0.0191, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 04:15:41,805 INFO [train2.py:809] (3/4) Epoch 11, batch 550, loss[ctc_loss=0.1031, att_loss=0.2465, loss=0.2178, over 16621.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.004994, over 47.00 utterances.], tot_loss[ctc_loss=0.1057, att_loss=0.2519, loss=0.2227, over 3065099.77 frames. utt_duration=1265 frames, utt_pad_proportion=0.05246, over 9706.96 utterances.], batch size: 47, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:15:46,935 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:16:13,532 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6713, 2.5089, 5.0866, 3.7068, 2.7807, 4.4467, 4.9292, 4.7207], device='cuda:3'), covar=tensor([0.0217, 0.1700, 0.0163, 0.1306, 0.2089, 0.0223, 0.0098, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0245, 0.0133, 0.0305, 0.0275, 0.0186, 0.0115, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 04:16:33,189 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.316e+02 2.892e+02 3.547e+02 1.294e+03, threshold=5.784e+02, percent-clipped=4.0 2023-03-08 04:16:38,054 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:17:02,143 INFO [train2.py:809] (3/4) Epoch 11, batch 600, loss[ctc_loss=0.08112, att_loss=0.2527, loss=0.2184, over 16467.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006674, over 46.00 utterances.], tot_loss[ctc_loss=0.1044, att_loss=0.2515, loss=0.2221, over 3115249.60 frames. utt_duration=1266 frames, utt_pad_proportion=0.05014, over 9852.35 utterances.], batch size: 46, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:17:08,528 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6401, 3.8008, 3.2710, 3.3741, 3.9228, 3.5497, 2.8457, 4.2782], device='cuda:3'), covar=tensor([0.1183, 0.0447, 0.0961, 0.0673, 0.0598, 0.0651, 0.0982, 0.0423], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0179, 0.0204, 0.0173, 0.0231, 0.0209, 0.0181, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 04:17:23,433 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-03-08 04:17:54,373 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:17:54,576 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:18:22,138 INFO [train2.py:809] (3/4) Epoch 11, batch 650, loss[ctc_loss=0.1063, att_loss=0.2422, loss=0.215, over 16397.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007888, over 44.00 utterances.], tot_loss[ctc_loss=0.1045, att_loss=0.2513, loss=0.2219, over 3143120.23 frames. utt_duration=1259 frames, utt_pad_proportion=0.05321, over 9999.20 utterances.], batch size: 44, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:18:28,822 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:18:30,189 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0051, 5.2804, 5.5900, 5.4400, 5.4138, 5.8838, 5.2333, 6.0531], device='cuda:3'), covar=tensor([0.0637, 0.0664, 0.0732, 0.0986, 0.1832, 0.1055, 0.0579, 0.0571], device='cuda:3'), in_proj_covar=tensor([0.0702, 0.0427, 0.0495, 0.0553, 0.0735, 0.0496, 0.0394, 0.0480], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:19:11,084 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:19:12,491 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.380e+02 3.105e+02 3.848e+02 6.391e+02, threshold=6.210e+02, percent-clipped=2.0 2023-03-08 04:19:40,520 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9416, 4.7881, 4.8052, 4.6984, 5.2245, 4.8835, 4.6564, 2.3344], device='cuda:3'), covar=tensor([0.0147, 0.0257, 0.0187, 0.0236, 0.1003, 0.0146, 0.0287, 0.2272], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0127, 0.0134, 0.0135, 0.0327, 0.0120, 0.0121, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 04:19:41,592 INFO [train2.py:809] (3/4) Epoch 11, batch 700, loss[ctc_loss=0.08459, att_loss=0.2179, loss=0.1913, over 15755.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009625, over 38.00 utterances.], tot_loss[ctc_loss=0.1039, att_loss=0.2506, loss=0.2213, over 3175654.09 frames. utt_duration=1285 frames, utt_pad_proportion=0.04606, over 9893.13 utterances.], batch size: 38, lr: 1.01e-02, grad_scale: 16.0 2023-03-08 04:20:01,469 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:20:04,702 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:20:04,777 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8060, 2.5333, 5.0822, 3.9341, 3.1407, 4.6136, 4.9613, 4.7964], device='cuda:3'), covar=tensor([0.0185, 0.1654, 0.0164, 0.0936, 0.1762, 0.0187, 0.0085, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0243, 0.0133, 0.0303, 0.0274, 0.0185, 0.0115, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 04:20:38,011 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:21:00,395 INFO [train2.py:809] (3/4) Epoch 11, batch 750, loss[ctc_loss=0.1127, att_loss=0.2668, loss=0.236, over 17014.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008123, over 51.00 utterances.], tot_loss[ctc_loss=0.1042, att_loss=0.2508, loss=0.2215, over 3194601.22 frames. utt_duration=1278 frames, utt_pad_proportion=0.04698, over 10008.96 utterances.], batch size: 51, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:21:02,284 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2811, 4.6275, 4.8755, 4.7764, 4.7569, 5.2393, 4.7953, 5.3137], device='cuda:3'), covar=tensor([0.0813, 0.0716, 0.0774, 0.1075, 0.1902, 0.0761, 0.1007, 0.0691], device='cuda:3'), in_proj_covar=tensor([0.0708, 0.0429, 0.0493, 0.0555, 0.0737, 0.0495, 0.0396, 0.0485], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:21:44,101 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.65 vs. limit=5.0 2023-03-08 04:21:51,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 2.391e+02 2.736e+02 3.441e+02 6.786e+02, threshold=5.472e+02, percent-clipped=2.0 2023-03-08 04:21:58,286 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 04:22:03,990 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:22:20,791 INFO [train2.py:809] (3/4) Epoch 11, batch 800, loss[ctc_loss=0.1639, att_loss=0.2754, loss=0.2531, over 14478.00 frames. utt_duration=395.4 frames, utt_pad_proportion=0.3063, over 147.00 utterances.], tot_loss[ctc_loss=0.1051, att_loss=0.2515, loss=0.2222, over 3217396.41 frames. utt_duration=1253 frames, utt_pad_proportion=0.05224, over 10286.01 utterances.], batch size: 147, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:22:41,711 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:23:38,278 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:23:41,161 INFO [train2.py:809] (3/4) Epoch 11, batch 850, loss[ctc_loss=0.1121, att_loss=0.2542, loss=0.2258, over 16337.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005746, over 45.00 utterances.], tot_loss[ctc_loss=0.1057, att_loss=0.2524, loss=0.223, over 3243781.07 frames. utt_duration=1248 frames, utt_pad_proportion=0.04939, over 10410.90 utterances.], batch size: 45, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:24:18,986 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:24:28,098 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-03-08 04:24:31,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.347e+02 2.849e+02 3.646e+02 6.696e+02, threshold=5.699e+02, percent-clipped=1.0 2023-03-08 04:25:00,584 INFO [train2.py:809] (3/4) Epoch 11, batch 900, loss[ctc_loss=0.112, att_loss=0.2505, loss=0.2228, over 16274.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007752, over 43.00 utterances.], tot_loss[ctc_loss=0.1045, att_loss=0.2507, loss=0.2214, over 3249016.81 frames. utt_duration=1275 frames, utt_pad_proportion=0.04418, over 10204.08 utterances.], batch size: 43, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:25:56,317 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7849, 5.0841, 4.5738, 5.2410, 4.5640, 4.9326, 5.2622, 5.0331], device='cuda:3'), covar=tensor([0.0534, 0.0348, 0.0900, 0.0281, 0.0470, 0.0215, 0.0214, 0.0184], device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0253, 0.0315, 0.0247, 0.0261, 0.0199, 0.0239, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 04:26:21,122 INFO [train2.py:809] (3/4) Epoch 11, batch 950, loss[ctc_loss=0.1399, att_loss=0.2835, loss=0.2548, over 17370.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.0348, over 63.00 utterances.], tot_loss[ctc_loss=0.1054, att_loss=0.2519, loss=0.2226, over 3259040.71 frames. utt_duration=1237 frames, utt_pad_proportion=0.0525, over 10553.78 utterances.], batch size: 63, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:27:12,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.301e+02 2.865e+02 3.345e+02 1.136e+03, threshold=5.731e+02, percent-clipped=4.0 2023-03-08 04:27:41,337 INFO [train2.py:809] (3/4) Epoch 11, batch 1000, loss[ctc_loss=0.1786, att_loss=0.2962, loss=0.2727, over 13325.00 frames. utt_duration=366.4 frames, utt_pad_proportion=0.3616, over 146.00 utterances.], tot_loss[ctc_loss=0.1055, att_loss=0.2515, loss=0.2223, over 3260906.10 frames. utt_duration=1226 frames, utt_pad_proportion=0.05728, over 10656.18 utterances.], batch size: 146, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:27:57,016 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:28:00,748 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-08 04:28:01,768 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:28:25,276 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:28:35,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-08 04:28:39,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:29:01,314 INFO [train2.py:809] (3/4) Epoch 11, batch 1050, loss[ctc_loss=0.09845, att_loss=0.236, loss=0.2085, over 15941.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007078, over 41.00 utterances.], tot_loss[ctc_loss=0.1049, att_loss=0.251, loss=0.2218, over 3257272.40 frames. utt_duration=1222 frames, utt_pad_proportion=0.06129, over 10678.16 utterances.], batch size: 41, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:29:18,982 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:29:20,765 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:29:52,389 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.443e+02 2.925e+02 3.809e+02 9.259e+02, threshold=5.849e+02, percent-clipped=3.0 2023-03-08 04:29:56,134 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:30:03,360 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:30:06,413 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:30:22,026 INFO [train2.py:809] (3/4) Epoch 11, batch 1100, loss[ctc_loss=0.142, att_loss=0.2867, loss=0.2578, over 17345.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.01969, over 59.00 utterances.], tot_loss[ctc_loss=0.1062, att_loss=0.2523, loss=0.2231, over 3271451.14 frames. utt_duration=1211 frames, utt_pad_proportion=0.06062, over 10823.43 utterances.], batch size: 59, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:30:59,257 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 04:31:23,534 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:31:28,425 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0382, 3.8901, 3.3128, 3.6971, 4.1076, 3.7710, 3.1821, 4.5167], device='cuda:3'), covar=tensor([0.1026, 0.0440, 0.1023, 0.0604, 0.0654, 0.0583, 0.0803, 0.0473], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0179, 0.0200, 0.0171, 0.0230, 0.0208, 0.0181, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 04:31:39,191 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:31:42,181 INFO [train2.py:809] (3/4) Epoch 11, batch 1150, loss[ctc_loss=0.1758, att_loss=0.2867, loss=0.2645, over 13963.00 frames. utt_duration=381.5 frames, utt_pad_proportion=0.3306, over 147.00 utterances.], tot_loss[ctc_loss=0.1069, att_loss=0.2524, loss=0.2233, over 3280774.79 frames. utt_duration=1197 frames, utt_pad_proportion=0.06167, over 10977.40 utterances.], batch size: 147, lr: 1.00e-02, grad_scale: 16.0 2023-03-08 04:31:56,123 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:32:12,919 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:32:34,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.552e+02 3.086e+02 3.791e+02 8.496e+02, threshold=6.172e+02, percent-clipped=3.0 2023-03-08 04:32:40,624 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-08 04:32:56,346 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:33:02,226 INFO [train2.py:809] (3/4) Epoch 11, batch 1200, loss[ctc_loss=0.08118, att_loss=0.2481, loss=0.2147, over 17035.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007767, over 51.00 utterances.], tot_loss[ctc_loss=0.1066, att_loss=0.2516, loss=0.2226, over 3284303.31 frames. utt_duration=1230 frames, utt_pad_proportion=0.05341, over 10695.54 utterances.], batch size: 51, lr: 9.99e-03, grad_scale: 16.0 2023-03-08 04:33:13,632 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:33:34,484 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:34:22,827 INFO [train2.py:809] (3/4) Epoch 11, batch 1250, loss[ctc_loss=0.1107, att_loss=0.2634, loss=0.2328, over 17125.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01392, over 56.00 utterances.], tot_loss[ctc_loss=0.1061, att_loss=0.2513, loss=0.2223, over 3269299.98 frames. utt_duration=1232 frames, utt_pad_proportion=0.05773, over 10631.52 utterances.], batch size: 56, lr: 9.99e-03, grad_scale: 16.0 2023-03-08 04:34:45,115 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:34:51,842 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:35:14,658 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.546e+02 2.278e+02 2.660e+02 3.579e+02 5.724e+02, threshold=5.320e+02, percent-clipped=1.0 2023-03-08 04:35:42,806 INFO [train2.py:809] (3/4) Epoch 11, batch 1300, loss[ctc_loss=0.11, att_loss=0.2361, loss=0.2109, over 15337.00 frames. utt_duration=1754 frames, utt_pad_proportion=0.01283, over 35.00 utterances.], tot_loss[ctc_loss=0.1063, att_loss=0.2514, loss=0.2224, over 3265626.32 frames. utt_duration=1232 frames, utt_pad_proportion=0.05905, over 10611.34 utterances.], batch size: 35, lr: 9.98e-03, grad_scale: 16.0 2023-03-08 04:35:58,614 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:36:23,572 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:36:31,784 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7354, 5.9105, 5.2592, 5.7175, 5.5646, 5.1540, 5.3121, 5.1169], device='cuda:3'), covar=tensor([0.1104, 0.0878, 0.0817, 0.0763, 0.0722, 0.1446, 0.2167, 0.2265], device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0512, 0.0372, 0.0386, 0.0360, 0.0422, 0.0524, 0.0460], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 04:36:49,368 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9620, 5.2139, 5.4874, 5.4359, 5.3156, 5.9256, 5.2442, 6.0432], device='cuda:3'), covar=tensor([0.0664, 0.0743, 0.0705, 0.1129, 0.2104, 0.0849, 0.0626, 0.0618], device='cuda:3'), in_proj_covar=tensor([0.0703, 0.0430, 0.0496, 0.0563, 0.0747, 0.0490, 0.0401, 0.0483], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:37:03,119 INFO [train2.py:809] (3/4) Epoch 11, batch 1350, loss[ctc_loss=0.1159, att_loss=0.2479, loss=0.2215, over 13650.00 frames. utt_duration=1822 frames, utt_pad_proportion=0.06922, over 30.00 utterances.], tot_loss[ctc_loss=0.1062, att_loss=0.2523, loss=0.2231, over 3278389.40 frames. utt_duration=1248 frames, utt_pad_proportion=0.05156, over 10523.85 utterances.], batch size: 30, lr: 9.98e-03, grad_scale: 16.0 2023-03-08 04:37:15,595 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:37:55,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.465e+02 2.945e+02 3.733e+02 9.191e+02, threshold=5.889e+02, percent-clipped=5.0 2023-03-08 04:37:56,864 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:38:23,863 INFO [train2.py:809] (3/4) Epoch 11, batch 1400, loss[ctc_loss=0.09517, att_loss=0.2306, loss=0.2035, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008766, over 37.00 utterances.], tot_loss[ctc_loss=0.1039, att_loss=0.2507, loss=0.2213, over 3276061.00 frames. utt_duration=1281 frames, utt_pad_proportion=0.04459, over 10242.58 utterances.], batch size: 37, lr: 9.97e-03, grad_scale: 16.0 2023-03-08 04:38:52,753 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 04:39:44,639 INFO [train2.py:809] (3/4) Epoch 11, batch 1450, loss[ctc_loss=0.1047, att_loss=0.2555, loss=0.2253, over 17228.00 frames. utt_duration=1170 frames, utt_pad_proportion=0.0269, over 59.00 utterances.], tot_loss[ctc_loss=0.1057, att_loss=0.2521, loss=0.2228, over 3280008.45 frames. utt_duration=1227 frames, utt_pad_proportion=0.05801, over 10709.22 utterances.], batch size: 59, lr: 9.96e-03, grad_scale: 16.0 2023-03-08 04:40:15,987 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:40:37,477 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.447e+02 2.872e+02 3.322e+02 6.594e+02, threshold=5.743e+02, percent-clipped=2.0 2023-03-08 04:40:45,291 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4834, 4.8782, 5.1118, 4.9805, 4.8879, 5.4571, 4.9603, 5.5523], device='cuda:3'), covar=tensor([0.0708, 0.0788, 0.0736, 0.1083, 0.1882, 0.0827, 0.0821, 0.0604], device='cuda:3'), in_proj_covar=tensor([0.0716, 0.0433, 0.0501, 0.0574, 0.0755, 0.0501, 0.0406, 0.0490], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:41:03,445 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-08 04:41:05,351 INFO [train2.py:809] (3/4) Epoch 11, batch 1500, loss[ctc_loss=0.07477, att_loss=0.2122, loss=0.1847, over 14465.00 frames. utt_duration=1810 frames, utt_pad_proportion=0.03631, over 32.00 utterances.], tot_loss[ctc_loss=0.1053, att_loss=0.252, loss=0.2227, over 3281138.11 frames. utt_duration=1241 frames, utt_pad_proportion=0.05428, over 10589.72 utterances.], batch size: 32, lr: 9.96e-03, grad_scale: 16.0 2023-03-08 04:41:30,413 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:41:33,409 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:41:36,585 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0444, 5.3138, 5.6264, 5.4734, 5.5075, 6.0366, 5.2000, 6.1182], device='cuda:3'), covar=tensor([0.0703, 0.0630, 0.0652, 0.1062, 0.1844, 0.0756, 0.0547, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0717, 0.0430, 0.0500, 0.0571, 0.0751, 0.0500, 0.0406, 0.0487], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:42:25,611 INFO [train2.py:809] (3/4) Epoch 11, batch 1550, loss[ctc_loss=0.1622, att_loss=0.2803, loss=0.2567, over 13981.00 frames. utt_duration=387.1 frames, utt_pad_proportion=0.3279, over 145.00 utterances.], tot_loss[ctc_loss=0.1047, att_loss=0.2513, loss=0.222, over 3273022.17 frames. utt_duration=1249 frames, utt_pad_proportion=0.05479, over 10494.41 utterances.], batch size: 145, lr: 9.95e-03, grad_scale: 16.0 2023-03-08 04:42:40,984 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2858, 2.4464, 4.4740, 3.6408, 3.0991, 4.1171, 4.1631, 4.1502], device='cuda:3'), covar=tensor([0.0191, 0.1736, 0.0106, 0.0953, 0.1562, 0.0243, 0.0147, 0.0289], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0247, 0.0135, 0.0307, 0.0278, 0.0188, 0.0119, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 04:42:47,299 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:43:19,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.586e+02 2.982e+02 3.761e+02 9.691e+02, threshold=5.963e+02, percent-clipped=6.0 2023-03-08 04:43:42,972 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0164, 5.2439, 5.5999, 5.5057, 5.4905, 6.0559, 5.2273, 6.1759], device='cuda:3'), covar=tensor([0.0635, 0.0709, 0.0659, 0.1043, 0.1727, 0.0693, 0.0534, 0.0471], device='cuda:3'), in_proj_covar=tensor([0.0713, 0.0425, 0.0493, 0.0563, 0.0742, 0.0495, 0.0402, 0.0480], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:43:45,720 INFO [train2.py:809] (3/4) Epoch 11, batch 1600, loss[ctc_loss=0.1091, att_loss=0.2693, loss=0.2373, over 17061.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009035, over 53.00 utterances.], tot_loss[ctc_loss=0.105, att_loss=0.2515, loss=0.2222, over 3282529.43 frames. utt_duration=1265 frames, utt_pad_proportion=0.04825, over 10388.43 utterances.], batch size: 53, lr: 9.95e-03, grad_scale: 8.0 2023-03-08 04:44:05,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6794, 5.9808, 5.4198, 5.7923, 5.6464, 5.1996, 5.2859, 5.1472], device='cuda:3'), covar=tensor([0.1314, 0.0880, 0.0766, 0.0721, 0.0803, 0.1423, 0.2617, 0.2366], device='cuda:3'), in_proj_covar=tensor([0.0441, 0.0505, 0.0369, 0.0383, 0.0359, 0.0417, 0.0517, 0.0460], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 04:44:17,931 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:44:21,015 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2997, 2.5643, 3.6235, 2.8381, 3.5418, 4.6306, 4.3595, 3.1493], device='cuda:3'), covar=tensor([0.0431, 0.1812, 0.0886, 0.1384, 0.0851, 0.0478, 0.0471, 0.1333], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0225, 0.0240, 0.0204, 0.0240, 0.0296, 0.0214, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 04:45:05,419 INFO [train2.py:809] (3/4) Epoch 11, batch 1650, loss[ctc_loss=0.1702, att_loss=0.2899, loss=0.266, over 14830.00 frames. utt_duration=404.9 frames, utt_pad_proportion=0.2897, over 147.00 utterances.], tot_loss[ctc_loss=0.1049, att_loss=0.2516, loss=0.2222, over 3275459.94 frames. utt_duration=1261 frames, utt_pad_proportion=0.05075, over 10401.36 utterances.], batch size: 147, lr: 9.94e-03, grad_scale: 8.0 2023-03-08 04:45:12,034 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0115, 3.9435, 3.3811, 3.6551, 4.1468, 3.6882, 3.1802, 4.4488], device='cuda:3'), covar=tensor([0.1013, 0.0391, 0.0954, 0.0564, 0.0591, 0.0643, 0.0757, 0.0456], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0180, 0.0203, 0.0173, 0.0234, 0.0214, 0.0183, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 04:45:28,381 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9930, 5.3350, 4.8013, 5.4101, 4.7545, 5.0064, 5.4779, 5.2523], device='cuda:3'), covar=tensor([0.0547, 0.0263, 0.0867, 0.0241, 0.0427, 0.0217, 0.0190, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0328, 0.0255, 0.0317, 0.0251, 0.0260, 0.0201, 0.0239, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 04:45:58,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.532e+02 2.866e+02 3.682e+02 7.453e+02, threshold=5.731e+02, percent-clipped=3.0 2023-03-08 04:45:58,664 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:46:25,296 INFO [train2.py:809] (3/4) Epoch 11, batch 1700, loss[ctc_loss=0.1079, att_loss=0.2625, loss=0.2316, over 17415.00 frames. utt_duration=883.3 frames, utt_pad_proportion=0.07513, over 79.00 utterances.], tot_loss[ctc_loss=0.1049, att_loss=0.2514, loss=0.2221, over 3265382.64 frames. utt_duration=1241 frames, utt_pad_proportion=0.05909, over 10534.69 utterances.], batch size: 79, lr: 9.93e-03, grad_scale: 8.0 2023-03-08 04:46:27,379 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0255, 5.1073, 4.9766, 2.5678, 2.0012, 2.9179, 2.7134, 3.9236], device='cuda:3'), covar=tensor([0.0696, 0.0202, 0.0210, 0.3996, 0.6084, 0.2566, 0.2576, 0.1669], device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0221, 0.0237, 0.0211, 0.0351, 0.0337, 0.0230, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.5105e-04, 8.2417e-05, 1.0126e-04, 9.4747e-05, 1.5383e-04, 1.3697e-04, 9.1151e-05, 1.5073e-04], device='cuda:3') 2023-03-08 04:46:55,399 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 04:47:16,353 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:47:46,032 INFO [train2.py:809] (3/4) Epoch 11, batch 1750, loss[ctc_loss=0.09402, att_loss=0.2565, loss=0.224, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007451, over 50.00 utterances.], tot_loss[ctc_loss=0.105, att_loss=0.2518, loss=0.2225, over 3272077.12 frames. utt_duration=1244 frames, utt_pad_proportion=0.05726, over 10532.15 utterances.], batch size: 50, lr: 9.93e-03, grad_scale: 8.0 2023-03-08 04:48:12,508 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:48:39,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.288e+02 2.927e+02 3.480e+02 6.791e+02, threshold=5.853e+02, percent-clipped=3.0 2023-03-08 04:49:06,235 INFO [train2.py:809] (3/4) Epoch 11, batch 1800, loss[ctc_loss=0.0921, att_loss=0.2423, loss=0.2122, over 16286.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.00634, over 43.00 utterances.], tot_loss[ctc_loss=0.1039, att_loss=0.251, loss=0.2216, over 3269652.84 frames. utt_duration=1251 frames, utt_pad_proportion=0.05648, over 10469.35 utterances.], batch size: 43, lr: 9.92e-03, grad_scale: 8.0 2023-03-08 04:49:25,328 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2587, 2.5817, 3.1475, 4.3812, 3.9481, 4.0039, 2.8126, 1.8853], device='cuda:3'), covar=tensor([0.0746, 0.2288, 0.1080, 0.0446, 0.0716, 0.0353, 0.1572, 0.2543], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0205, 0.0183, 0.0187, 0.0186, 0.0145, 0.0190, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 04:49:31,693 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:49:33,305 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:49:35,871 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-08 04:50:13,907 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-08 04:50:27,077 INFO [train2.py:809] (3/4) Epoch 11, batch 1850, loss[ctc_loss=0.08318, att_loss=0.2374, loss=0.2065, over 15943.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007171, over 41.00 utterances.], tot_loss[ctc_loss=0.1039, att_loss=0.251, loss=0.2216, over 3278920.32 frames. utt_duration=1257 frames, utt_pad_proportion=0.05237, over 10450.55 utterances.], batch size: 41, lr: 9.92e-03, grad_scale: 8.0 2023-03-08 04:50:48,920 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:50:49,122 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:51:12,150 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 04:51:20,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.415e+02 2.840e+02 3.452e+02 8.289e+02, threshold=5.680e+02, percent-clipped=4.0 2023-03-08 04:51:46,880 INFO [train2.py:809] (3/4) Epoch 11, batch 1900, loss[ctc_loss=0.09222, att_loss=0.2283, loss=0.2011, over 15792.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.008191, over 38.00 utterances.], tot_loss[ctc_loss=0.1038, att_loss=0.2507, loss=0.2213, over 3272945.57 frames. utt_duration=1275 frames, utt_pad_proportion=0.04907, over 10281.92 utterances.], batch size: 38, lr: 9.91e-03, grad_scale: 8.0 2023-03-08 04:51:48,955 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1365, 5.1407, 5.0307, 2.4417, 2.0924, 2.8177, 3.2626, 3.8330], device='cuda:3'), covar=tensor([0.0594, 0.0214, 0.0187, 0.3915, 0.5616, 0.2537, 0.1701, 0.1715], device='cuda:3'), in_proj_covar=tensor([0.0332, 0.0221, 0.0235, 0.0209, 0.0350, 0.0335, 0.0228, 0.0355], device='cuda:3'), out_proj_covar=tensor([1.5020e-04, 8.2758e-05, 1.0044e-04, 9.3969e-05, 1.5336e-04, 1.3620e-04, 9.0260e-05, 1.5035e-04], device='cuda:3') 2023-03-08 04:52:04,782 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:52:19,818 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:52:23,278 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-08 04:53:05,088 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 04:53:07,401 INFO [train2.py:809] (3/4) Epoch 11, batch 1950, loss[ctc_loss=0.08046, att_loss=0.2482, loss=0.2147, over 16867.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008398, over 49.00 utterances.], tot_loss[ctc_loss=0.1033, att_loss=0.2503, loss=0.2209, over 3274437.16 frames. utt_duration=1272 frames, utt_pad_proportion=0.04789, over 10310.23 utterances.], batch size: 49, lr: 9.91e-03, grad_scale: 8.0 2023-03-08 04:53:37,477 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:54:00,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 2.550e+02 3.116e+02 4.012e+02 6.949e+02, threshold=6.232e+02, percent-clipped=6.0 2023-03-08 04:54:27,440 INFO [train2.py:809] (3/4) Epoch 11, batch 2000, loss[ctc_loss=0.1108, att_loss=0.2665, loss=0.2353, over 17300.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02361, over 59.00 utterances.], tot_loss[ctc_loss=0.1032, att_loss=0.2506, loss=0.2211, over 3279743.02 frames. utt_duration=1259 frames, utt_pad_proportion=0.04956, over 10434.12 utterances.], batch size: 59, lr: 9.90e-03, grad_scale: 8.0 2023-03-08 04:55:04,210 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-08 04:55:29,619 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0341, 5.2854, 5.2681, 5.2274, 5.3595, 5.2965, 5.0087, 4.8385], device='cuda:3'), covar=tensor([0.1010, 0.0448, 0.0236, 0.0405, 0.0253, 0.0272, 0.0313, 0.0311], device='cuda:3'), in_proj_covar=tensor([0.0448, 0.0284, 0.0239, 0.0277, 0.0337, 0.0353, 0.0285, 0.0315], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 04:55:44,773 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-03-08 04:55:47,162 INFO [train2.py:809] (3/4) Epoch 11, batch 2050, loss[ctc_loss=0.0961, att_loss=0.2329, loss=0.2056, over 12363.00 frames. utt_duration=1833 frames, utt_pad_proportion=0.143, over 27.00 utterances.], tot_loss[ctc_loss=0.1041, att_loss=0.2507, loss=0.2214, over 3275639.09 frames. utt_duration=1265 frames, utt_pad_proportion=0.04893, over 10368.81 utterances.], batch size: 27, lr: 9.89e-03, grad_scale: 8.0 2023-03-08 04:56:17,851 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:56:40,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.492e+02 2.943e+02 3.583e+02 1.498e+03, threshold=5.886e+02, percent-clipped=5.0 2023-03-08 04:57:07,334 INFO [train2.py:809] (3/4) Epoch 11, batch 2100, loss[ctc_loss=0.122, att_loss=0.2732, loss=0.243, over 17320.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02197, over 59.00 utterances.], tot_loss[ctc_loss=0.1042, att_loss=0.2506, loss=0.2213, over 3281545.75 frames. utt_duration=1264 frames, utt_pad_proportion=0.04781, over 10397.08 utterances.], batch size: 59, lr: 9.89e-03, grad_scale: 8.0 2023-03-08 04:57:23,979 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 04:57:54,624 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 04:58:20,502 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1052, 5.2003, 5.1204, 2.5943, 2.0340, 3.0046, 2.8855, 4.0058], device='cuda:3'), covar=tensor([0.0640, 0.0308, 0.0212, 0.4250, 0.5623, 0.2429, 0.2298, 0.1564], device='cuda:3'), in_proj_covar=tensor([0.0321, 0.0215, 0.0230, 0.0202, 0.0338, 0.0324, 0.0223, 0.0343], device='cuda:3'), out_proj_covar=tensor([1.4525e-04, 8.0385e-05, 9.8518e-05, 9.0246e-05, 1.4800e-04, 1.3173e-04, 8.8146e-05, 1.4513e-04], device='cuda:3') 2023-03-08 04:58:27,971 INFO [train2.py:809] (3/4) Epoch 11, batch 2150, loss[ctc_loss=0.0852, att_loss=0.2603, loss=0.2253, over 17058.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009427, over 53.00 utterances.], tot_loss[ctc_loss=0.1035, att_loss=0.2497, loss=0.2204, over 3261775.09 frames. utt_duration=1258 frames, utt_pad_proportion=0.05633, over 10387.50 utterances.], batch size: 53, lr: 9.88e-03, grad_scale: 8.0 2023-03-08 04:58:43,251 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-03-08 04:59:07,019 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 04:59:08,305 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 04:59:25,317 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.469e+02 2.816e+02 3.466e+02 1.172e+03, threshold=5.632e+02, percent-clipped=3.0 2023-03-08 04:59:52,147 INFO [train2.py:809] (3/4) Epoch 11, batch 2200, loss[ctc_loss=0.1245, att_loss=0.2462, loss=0.2219, over 16107.00 frames. utt_duration=1535 frames, utt_pad_proportion=0.006823, over 42.00 utterances.], tot_loss[ctc_loss=0.1035, att_loss=0.2503, loss=0.2209, over 3270285.78 frames. utt_duration=1251 frames, utt_pad_proportion=0.05323, over 10470.12 utterances.], batch size: 42, lr: 9.88e-03, grad_scale: 8.0 2023-03-08 05:00:41,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 05:01:12,043 INFO [train2.py:809] (3/4) Epoch 11, batch 2250, loss[ctc_loss=0.08165, att_loss=0.2296, loss=0.2, over 16161.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.008048, over 41.00 utterances.], tot_loss[ctc_loss=0.1034, att_loss=0.2509, loss=0.2214, over 3273137.38 frames. utt_duration=1236 frames, utt_pad_proportion=0.05737, over 10606.40 utterances.], batch size: 41, lr: 9.87e-03, grad_scale: 8.0 2023-03-08 05:01:42,289 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8871, 2.4335, 3.9688, 3.4913, 2.8781, 3.7066, 3.6487, 3.7244], device='cuda:3'), covar=tensor([0.0204, 0.1477, 0.0106, 0.0854, 0.1481, 0.0259, 0.0146, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0241, 0.0132, 0.0301, 0.0275, 0.0184, 0.0118, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 05:01:46,929 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8635, 2.1153, 1.8980, 1.6198, 2.5266, 2.0970, 1.8808, 2.1786], device='cuda:3'), covar=tensor([0.0966, 0.4359, 0.4818, 0.2623, 0.1194, 0.1776, 0.2865, 0.1235], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0087, 0.0092, 0.0077, 0.0077, 0.0071, 0.0087, 0.0062], device='cuda:3'), out_proj_covar=tensor([5.0380e-05, 5.9585e-05, 6.2206e-05, 5.2400e-05, 5.0180e-05, 4.9889e-05, 5.8841e-05, 4.5251e-05], device='cuda:3') 2023-03-08 05:02:04,982 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.270e+02 2.801e+02 3.546e+02 6.959e+02, threshold=5.601e+02, percent-clipped=2.0 2023-03-08 05:02:08,364 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6664, 5.2100, 5.0006, 5.1251, 5.0802, 4.9087, 3.7884, 5.1859], device='cuda:3'), covar=tensor([0.0115, 0.0100, 0.0126, 0.0068, 0.0109, 0.0105, 0.0644, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0072, 0.0088, 0.0054, 0.0059, 0.0070, 0.0092, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:02:16,884 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6074, 2.6354, 4.8385, 3.7826, 2.9852, 4.1801, 4.4860, 4.4915], device='cuda:3'), covar=tensor([0.0137, 0.1656, 0.0091, 0.0925, 0.1789, 0.0267, 0.0123, 0.0187], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0240, 0.0132, 0.0300, 0.0274, 0.0183, 0.0118, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 05:02:26,342 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0185, 5.3728, 5.5298, 5.4457, 5.5220, 6.0011, 5.2317, 6.1293], device='cuda:3'), covar=tensor([0.0641, 0.0614, 0.0741, 0.0999, 0.1696, 0.0748, 0.0664, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0712, 0.0423, 0.0492, 0.0560, 0.0741, 0.0491, 0.0402, 0.0487], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:02:32,246 INFO [train2.py:809] (3/4) Epoch 11, batch 2300, loss[ctc_loss=0.1018, att_loss=0.2606, loss=0.2289, over 16818.00 frames. utt_duration=688 frames, utt_pad_proportion=0.1379, over 98.00 utterances.], tot_loss[ctc_loss=0.1036, att_loss=0.2513, loss=0.2217, over 3269999.03 frames. utt_duration=1188 frames, utt_pad_proportion=0.06931, over 11022.80 utterances.], batch size: 98, lr: 9.86e-03, grad_scale: 8.0 2023-03-08 05:03:16,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2635, 5.5458, 5.0057, 5.6298, 5.0248, 5.0864, 5.6660, 5.4659], device='cuda:3'), covar=tensor([0.0444, 0.0249, 0.0786, 0.0177, 0.0366, 0.0190, 0.0181, 0.0159], device='cuda:3'), in_proj_covar=tensor([0.0328, 0.0255, 0.0314, 0.0249, 0.0257, 0.0200, 0.0240, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 05:03:33,799 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4244, 4.8243, 4.6866, 4.7585, 4.8217, 4.5732, 3.4433, 4.7531], device='cuda:3'), covar=tensor([0.0101, 0.0118, 0.0104, 0.0075, 0.0084, 0.0107, 0.0684, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0072, 0.0088, 0.0054, 0.0059, 0.0071, 0.0092, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:03:42,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-03-08 05:03:52,294 INFO [train2.py:809] (3/4) Epoch 11, batch 2350, loss[ctc_loss=0.1909, att_loss=0.2903, loss=0.2704, over 14581.00 frames. utt_duration=400.9 frames, utt_pad_proportion=0.3028, over 146.00 utterances.], tot_loss[ctc_loss=0.1031, att_loss=0.2511, loss=0.2215, over 3264545.51 frames. utt_duration=1195 frames, utt_pad_proportion=0.06806, over 10940.69 utterances.], batch size: 146, lr: 9.86e-03, grad_scale: 8.0 2023-03-08 05:04:45,693 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.492e+02 2.954e+02 3.764e+02 8.189e+02, threshold=5.908e+02, percent-clipped=4.0 2023-03-08 05:05:02,221 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6800, 5.2377, 5.0246, 5.0292, 5.2234, 4.9519, 3.9189, 5.2294], device='cuda:3'), covar=tensor([0.0095, 0.0083, 0.0098, 0.0079, 0.0073, 0.0082, 0.0568, 0.0126], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0070, 0.0086, 0.0053, 0.0057, 0.0069, 0.0090, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:05:13,513 INFO [train2.py:809] (3/4) Epoch 11, batch 2400, loss[ctc_loss=0.1129, att_loss=0.2593, loss=0.2301, over 16297.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.00632, over 43.00 utterances.], tot_loss[ctc_loss=0.1032, att_loss=0.2509, loss=0.2214, over 3263222.95 frames. utt_duration=1212 frames, utt_pad_proportion=0.06556, over 10779.79 utterances.], batch size: 43, lr: 9.85e-03, grad_scale: 8.0 2023-03-08 05:05:52,554 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:06:34,469 INFO [train2.py:809] (3/4) Epoch 11, batch 2450, loss[ctc_loss=0.1249, att_loss=0.255, loss=0.229, over 16277.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007584, over 43.00 utterances.], tot_loss[ctc_loss=0.1047, att_loss=0.2522, loss=0.2227, over 3270885.34 frames. utt_duration=1200 frames, utt_pad_proportion=0.06605, over 10919.31 utterances.], batch size: 43, lr: 9.85e-03, grad_scale: 8.0 2023-03-08 05:07:00,003 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 05:07:09,506 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 05:07:26,991 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.356e+02 2.915e+02 3.557e+02 7.393e+02, threshold=5.830e+02, percent-clipped=4.0 2023-03-08 05:07:54,612 INFO [train2.py:809] (3/4) Epoch 11, batch 2500, loss[ctc_loss=0.1612, att_loss=0.2809, loss=0.2569, over 17023.00 frames. utt_duration=689.2 frames, utt_pad_proportion=0.1363, over 99.00 utterances.], tot_loss[ctc_loss=0.1042, att_loss=0.2518, loss=0.2223, over 3267112.19 frames. utt_duration=1209 frames, utt_pad_proportion=0.06354, over 10825.29 utterances.], batch size: 99, lr: 9.84e-03, grad_scale: 8.0 2023-03-08 05:08:26,079 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:08:59,976 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:09:14,576 INFO [train2.py:809] (3/4) Epoch 11, batch 2550, loss[ctc_loss=0.109, att_loss=0.2589, loss=0.2289, over 16476.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006207, over 46.00 utterances.], tot_loss[ctc_loss=0.1027, att_loss=0.2508, loss=0.2212, over 3271494.61 frames. utt_duration=1233 frames, utt_pad_proportion=0.05638, over 10625.26 utterances.], batch size: 46, lr: 9.84e-03, grad_scale: 8.0 2023-03-08 05:10:03,856 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0845, 5.2355, 5.0570, 2.3798, 1.9358, 2.9553, 2.5048, 3.9608], device='cuda:3'), covar=tensor([0.0649, 0.0242, 0.0192, 0.3798, 0.5852, 0.2470, 0.2406, 0.1570], device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0223, 0.0239, 0.0209, 0.0350, 0.0336, 0.0230, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.5061e-04, 8.3390e-05, 1.0266e-04, 9.3438e-05, 1.5281e-04, 1.3677e-04, 9.1092e-05, 1.4934e-04], device='cuda:3') 2023-03-08 05:10:06,487 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.376e+02 2.981e+02 3.759e+02 6.397e+02, threshold=5.961e+02, percent-clipped=1.0 2023-03-08 05:10:22,302 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8907, 5.0451, 4.8972, 2.4067, 1.7666, 2.9632, 2.3824, 3.7938], device='cuda:3'), covar=tensor([0.0853, 0.0205, 0.0249, 0.4312, 0.6725, 0.2481, 0.2862, 0.1767], device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0223, 0.0239, 0.0209, 0.0350, 0.0336, 0.0230, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.5047e-04, 8.3426e-05, 1.0268e-04, 9.3122e-05, 1.5269e-04, 1.3659e-04, 9.1093e-05, 1.4917e-04], device='cuda:3') 2023-03-08 05:10:34,788 INFO [train2.py:809] (3/4) Epoch 11, batch 2600, loss[ctc_loss=0.08907, att_loss=0.2422, loss=0.2116, over 16542.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006182, over 45.00 utterances.], tot_loss[ctc_loss=0.1028, att_loss=0.2511, loss=0.2214, over 3278445.12 frames. utt_duration=1222 frames, utt_pad_proportion=0.05704, over 10745.77 utterances.], batch size: 45, lr: 9.83e-03, grad_scale: 8.0 2023-03-08 05:10:38,292 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:11:35,164 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9788, 6.2250, 5.6134, 5.9177, 5.9057, 5.4908, 5.5853, 5.3538], device='cuda:3'), covar=tensor([0.1255, 0.0919, 0.0979, 0.0901, 0.0772, 0.1431, 0.2664, 0.2433], device='cuda:3'), in_proj_covar=tensor([0.0449, 0.0509, 0.0378, 0.0386, 0.0368, 0.0424, 0.0534, 0.0470], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 05:11:55,322 INFO [train2.py:809] (3/4) Epoch 11, batch 2650, loss[ctc_loss=0.1048, att_loss=0.2271, loss=0.2027, over 15774.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.007074, over 38.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.25, loss=0.2203, over 3267780.36 frames. utt_duration=1229 frames, utt_pad_proportion=0.05932, over 10650.22 utterances.], batch size: 38, lr: 9.82e-03, grad_scale: 8.0 2023-03-08 05:12:20,804 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1316, 4.0936, 3.9706, 3.9382, 4.5720, 4.1274, 4.0088, 2.2103], device='cuda:3'), covar=tensor([0.0223, 0.0455, 0.0392, 0.0164, 0.0890, 0.0221, 0.0357, 0.1999], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0130, 0.0137, 0.0140, 0.0327, 0.0119, 0.0121, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 05:12:23,805 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:12:47,510 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.235e+02 2.761e+02 3.286e+02 8.509e+02, threshold=5.523e+02, percent-clipped=3.0 2023-03-08 05:13:05,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-08 05:13:15,974 INFO [train2.py:809] (3/4) Epoch 11, batch 2700, loss[ctc_loss=0.09042, att_loss=0.2513, loss=0.2192, over 16878.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007958, over 49.00 utterances.], tot_loss[ctc_loss=0.1023, att_loss=0.2499, loss=0.2204, over 3266700.13 frames. utt_duration=1225 frames, utt_pad_proportion=0.0605, over 10682.66 utterances.], batch size: 49, lr: 9.82e-03, grad_scale: 8.0 2023-03-08 05:13:53,912 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:14:02,151 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:14:37,196 INFO [train2.py:809] (3/4) Epoch 11, batch 2750, loss[ctc_loss=0.08979, att_loss=0.2216, loss=0.1953, over 15635.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009431, over 37.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.2499, loss=0.2203, over 3264207.72 frames. utt_duration=1215 frames, utt_pad_proportion=0.0633, over 10757.17 utterances.], batch size: 37, lr: 9.81e-03, grad_scale: 8.0 2023-03-08 05:14:42,583 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-08 05:15:02,451 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 05:15:12,248 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:15:30,220 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.426e+02 2.955e+02 3.528e+02 6.434e+02, threshold=5.910e+02, percent-clipped=2.0 2023-03-08 05:15:57,767 INFO [train2.py:809] (3/4) Epoch 11, batch 2800, loss[ctc_loss=0.1079, att_loss=0.2516, loss=0.2228, over 16097.00 frames. utt_duration=1535 frames, utt_pad_proportion=0.007911, over 42.00 utterances.], tot_loss[ctc_loss=0.1022, att_loss=0.2501, loss=0.2205, over 3259492.75 frames. utt_duration=1209 frames, utt_pad_proportion=0.06557, over 10795.75 utterances.], batch size: 42, lr: 9.81e-03, grad_scale: 8.0 2023-03-08 05:16:06,436 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2023-03-08 05:16:15,476 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:16:20,522 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 05:17:17,562 INFO [train2.py:809] (3/4) Epoch 11, batch 2850, loss[ctc_loss=0.1054, att_loss=0.2357, loss=0.2096, over 15499.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.007772, over 36.00 utterances.], tot_loss[ctc_loss=0.1026, att_loss=0.2499, loss=0.2205, over 3256196.39 frames. utt_duration=1209 frames, utt_pad_proportion=0.0673, over 10790.69 utterances.], batch size: 36, lr: 9.80e-03, grad_scale: 8.0 2023-03-08 05:17:52,683 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:18:10,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.468e+02 2.915e+02 3.973e+02 1.035e+03, threshold=5.829e+02, percent-clipped=4.0 2023-03-08 05:18:33,602 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:18:38,144 INFO [train2.py:809] (3/4) Epoch 11, batch 2900, loss[ctc_loss=0.08293, att_loss=0.2507, loss=0.2171, over 16481.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005891, over 46.00 utterances.], tot_loss[ctc_loss=0.1014, att_loss=0.2492, loss=0.2197, over 3258263.89 frames. utt_duration=1226 frames, utt_pad_proportion=0.06231, over 10639.61 utterances.], batch size: 46, lr: 9.80e-03, grad_scale: 8.0 2023-03-08 05:18:56,514 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5522, 2.4440, 4.9565, 3.9902, 2.8257, 4.3554, 4.9154, 4.6309], device='cuda:3'), covar=tensor([0.0220, 0.1825, 0.0150, 0.0920, 0.2119, 0.0261, 0.0085, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0243, 0.0136, 0.0305, 0.0278, 0.0187, 0.0120, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 05:19:58,449 INFO [train2.py:809] (3/4) Epoch 11, batch 2950, loss[ctc_loss=0.1054, att_loss=0.2357, loss=0.2096, over 15999.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.007285, over 40.00 utterances.], tot_loss[ctc_loss=0.1015, att_loss=0.2495, loss=0.2199, over 3267005.69 frames. utt_duration=1222 frames, utt_pad_proportion=0.06033, over 10709.58 utterances.], batch size: 40, lr: 9.79e-03, grad_scale: 8.0 2023-03-08 05:20:18,543 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6511, 5.2409, 4.9477, 5.1538, 5.2683, 4.8021, 3.9462, 5.1269], device='cuda:3'), covar=tensor([0.0107, 0.0080, 0.0116, 0.0077, 0.0092, 0.0096, 0.0527, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0071, 0.0087, 0.0053, 0.0059, 0.0070, 0.0091, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:20:52,199 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.273e+02 2.678e+02 3.532e+02 7.141e+02, threshold=5.356e+02, percent-clipped=1.0 2023-03-08 05:21:19,018 INFO [train2.py:809] (3/4) Epoch 11, batch 3000, loss[ctc_loss=0.1075, att_loss=0.2516, loss=0.2228, over 16549.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005957, over 45.00 utterances.], tot_loss[ctc_loss=0.1005, att_loss=0.2489, loss=0.2192, over 3263842.85 frames. utt_duration=1251 frames, utt_pad_proportion=0.05345, over 10450.07 utterances.], batch size: 45, lr: 9.78e-03, grad_scale: 8.0 2023-03-08 05:21:19,018 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 05:21:32,741 INFO [train2.py:843] (3/4) Epoch 11, validation: ctc_loss=0.04985, att_loss=0.2382, loss=0.2006, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 05:21:32,742 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 05:22:01,153 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0296, 5.2899, 5.5666, 5.4472, 5.5225, 6.0087, 5.2306, 6.1244], device='cuda:3'), covar=tensor([0.0576, 0.0588, 0.0614, 0.1001, 0.1540, 0.0787, 0.0531, 0.0539], device='cuda:3'), in_proj_covar=tensor([0.0724, 0.0431, 0.0494, 0.0570, 0.0753, 0.0502, 0.0408, 0.0496], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:22:05,854 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:22:10,475 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:22:37,876 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:22:52,096 INFO [train2.py:809] (3/4) Epoch 11, batch 3050, loss[ctc_loss=0.1726, att_loss=0.2873, loss=0.2644, over 14048.00 frames. utt_duration=389 frames, utt_pad_proportion=0.3247, over 145.00 utterances.], tot_loss[ctc_loss=0.1012, att_loss=0.2496, loss=0.2199, over 3266411.69 frames. utt_duration=1258 frames, utt_pad_proportion=0.05149, over 10399.33 utterances.], batch size: 145, lr: 9.78e-03, grad_scale: 8.0 2023-03-08 05:23:43,378 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:23:44,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.391e+02 3.019e+02 3.669e+02 9.404e+02, threshold=6.038e+02, percent-clipped=6.0 2023-03-08 05:24:04,596 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-08 05:24:11,644 INFO [train2.py:809] (3/4) Epoch 11, batch 3100, loss[ctc_loss=0.09508, att_loss=0.251, loss=0.2198, over 16262.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.00765, over 43.00 utterances.], tot_loss[ctc_loss=0.101, att_loss=0.2494, loss=0.2197, over 3267695.45 frames. utt_duration=1277 frames, utt_pad_proportion=0.0467, over 10250.01 utterances.], batch size: 43, lr: 9.77e-03, grad_scale: 8.0 2023-03-08 05:24:15,753 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:25:32,676 INFO [train2.py:809] (3/4) Epoch 11, batch 3150, loss[ctc_loss=0.09245, att_loss=0.2469, loss=0.216, over 16401.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006264, over 44.00 utterances.], tot_loss[ctc_loss=0.1012, att_loss=0.2491, loss=0.2196, over 3265753.02 frames. utt_duration=1274 frames, utt_pad_proportion=0.04681, over 10263.23 utterances.], batch size: 44, lr: 9.77e-03, grad_scale: 8.0 2023-03-08 05:26:00,833 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:26:04,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-08 05:26:26,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.340e+02 2.878e+02 3.301e+02 6.036e+02, threshold=5.756e+02, percent-clipped=0.0 2023-03-08 05:26:47,928 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0032, 6.2438, 5.6649, 5.9755, 5.8242, 5.4142, 5.6803, 5.4913], device='cuda:3'), covar=tensor([0.1411, 0.0827, 0.0842, 0.0797, 0.0769, 0.1559, 0.2753, 0.2629], device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0505, 0.0377, 0.0386, 0.0362, 0.0423, 0.0527, 0.0463], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 05:26:49,605 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:26:53,986 INFO [train2.py:809] (3/4) Epoch 11, batch 3200, loss[ctc_loss=0.08883, att_loss=0.2398, loss=0.2096, over 16784.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005549, over 48.00 utterances.], tot_loss[ctc_loss=0.1016, att_loss=0.2492, loss=0.2197, over 3259846.68 frames. utt_duration=1290 frames, utt_pad_proportion=0.04481, over 10120.77 utterances.], batch size: 48, lr: 9.76e-03, grad_scale: 8.0 2023-03-08 05:28:06,592 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:28:14,935 INFO [train2.py:809] (3/4) Epoch 11, batch 3250, loss[ctc_loss=0.07461, att_loss=0.2388, loss=0.2059, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007269, over 43.00 utterances.], tot_loss[ctc_loss=0.1015, att_loss=0.2493, loss=0.2197, over 3266841.66 frames. utt_duration=1299 frames, utt_pad_proportion=0.04247, over 10071.49 utterances.], batch size: 43, lr: 9.76e-03, grad_scale: 8.0 2023-03-08 05:29:07,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.382e+02 2.901e+02 3.870e+02 8.262e+02, threshold=5.801e+02, percent-clipped=4.0 2023-03-08 05:29:35,238 INFO [train2.py:809] (3/4) Epoch 11, batch 3300, loss[ctc_loss=0.1491, att_loss=0.2813, loss=0.2549, over 17056.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009543, over 53.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.2493, loss=0.2198, over 3272477.81 frames. utt_duration=1294 frames, utt_pad_proportion=0.04085, over 10126.46 utterances.], batch size: 53, lr: 9.75e-03, grad_scale: 8.0 2023-03-08 05:30:06,512 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2242, 4.7252, 4.4864, 4.7142, 4.7748, 4.4128, 3.3133, 4.5654], device='cuda:3'), covar=tensor([0.0114, 0.0102, 0.0135, 0.0078, 0.0084, 0.0108, 0.0682, 0.0236], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0070, 0.0085, 0.0052, 0.0058, 0.0069, 0.0089, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:30:13,256 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:30:55,501 INFO [train2.py:809] (3/4) Epoch 11, batch 3350, loss[ctc_loss=0.09546, att_loss=0.2267, loss=0.2005, over 15640.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009099, over 37.00 utterances.], tot_loss[ctc_loss=0.1021, att_loss=0.2495, loss=0.22, over 3274171.34 frames. utt_duration=1291 frames, utt_pad_proportion=0.04177, over 10158.01 utterances.], batch size: 37, lr: 9.75e-03, grad_scale: 8.0 2023-03-08 05:31:15,566 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4542, 2.2307, 4.9230, 3.8633, 2.7971, 4.3146, 4.8121, 4.5683], device='cuda:3'), covar=tensor([0.0223, 0.2049, 0.0138, 0.1074, 0.2078, 0.0243, 0.0090, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0245, 0.0137, 0.0307, 0.0275, 0.0188, 0.0121, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 05:31:28,110 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3267, 4.8036, 4.5084, 4.9538, 2.6401, 4.5536, 2.5820, 1.7324], device='cuda:3'), covar=tensor([0.0321, 0.0166, 0.0701, 0.0119, 0.1797, 0.0203, 0.1586, 0.1814], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0112, 0.0257, 0.0109, 0.0219, 0.0109, 0.0223, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 05:31:31,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:31:39,582 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:31:49,462 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.322e+02 2.901e+02 3.775e+02 1.145e+03, threshold=5.802e+02, percent-clipped=9.0 2023-03-08 05:32:05,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-08 05:32:12,660 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:32:17,048 INFO [train2.py:809] (3/4) Epoch 11, batch 3400, loss[ctc_loss=0.07905, att_loss=0.2386, loss=0.2067, over 16479.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006011, over 46.00 utterances.], tot_loss[ctc_loss=0.1013, att_loss=0.2492, loss=0.2196, over 3271074.31 frames. utt_duration=1270 frames, utt_pad_proportion=0.04889, over 10316.27 utterances.], batch size: 46, lr: 9.74e-03, grad_scale: 8.0 2023-03-08 05:33:37,403 INFO [train2.py:809] (3/4) Epoch 11, batch 3450, loss[ctc_loss=0.1047, att_loss=0.2329, loss=0.2072, over 14493.00 frames. utt_duration=1813 frames, utt_pad_proportion=0.04876, over 32.00 utterances.], tot_loss[ctc_loss=0.1007, att_loss=0.2486, loss=0.219, over 3265741.80 frames. utt_duration=1286 frames, utt_pad_proportion=0.0456, over 10172.83 utterances.], batch size: 32, lr: 9.73e-03, grad_scale: 8.0 2023-03-08 05:34:03,794 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:34:29,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.188e+02 2.793e+02 3.558e+02 5.760e+02, threshold=5.586e+02, percent-clipped=0.0 2023-03-08 05:34:57,080 INFO [train2.py:809] (3/4) Epoch 11, batch 3500, loss[ctc_loss=0.1071, att_loss=0.2647, loss=0.2331, over 17433.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04348, over 69.00 utterances.], tot_loss[ctc_loss=0.1021, att_loss=0.2498, loss=0.2202, over 3268961.11 frames. utt_duration=1259 frames, utt_pad_proportion=0.05204, over 10398.15 utterances.], batch size: 69, lr: 9.73e-03, grad_scale: 8.0 2023-03-08 05:35:12,178 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 05:35:20,566 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:36:17,030 INFO [train2.py:809] (3/4) Epoch 11, batch 3550, loss[ctc_loss=0.09908, att_loss=0.2612, loss=0.2288, over 16766.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.00654, over 48.00 utterances.], tot_loss[ctc_loss=0.103, att_loss=0.2508, loss=0.2212, over 3266704.25 frames. utt_duration=1236 frames, utt_pad_proportion=0.05746, over 10588.09 utterances.], batch size: 48, lr: 9.72e-03, grad_scale: 4.0 2023-03-08 05:36:46,078 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5221, 4.9177, 4.7539, 4.9361, 4.9162, 4.4985, 3.4258, 4.7219], device='cuda:3'), covar=tensor([0.0113, 0.0101, 0.0116, 0.0067, 0.0080, 0.0132, 0.0711, 0.0206], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0071, 0.0087, 0.0053, 0.0059, 0.0070, 0.0091, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:36:55,853 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.43 vs. limit=5.0 2023-03-08 05:36:57,527 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1205, 5.0575, 4.9500, 2.9746, 4.8386, 4.5870, 4.3036, 2.8643], device='cuda:3'), covar=tensor([0.0122, 0.0083, 0.0209, 0.1051, 0.0086, 0.0190, 0.0330, 0.1338], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0085, 0.0078, 0.0107, 0.0071, 0.0097, 0.0095, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 05:37:11,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.366e+02 2.941e+02 3.430e+02 6.532e+02, threshold=5.882e+02, percent-clipped=4.0 2023-03-08 05:37:37,565 INFO [train2.py:809] (3/4) Epoch 11, batch 3600, loss[ctc_loss=0.1104, att_loss=0.2594, loss=0.2296, over 16889.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007091, over 49.00 utterances.], tot_loss[ctc_loss=0.1023, att_loss=0.2506, loss=0.2209, over 3269691.72 frames. utt_duration=1244 frames, utt_pad_proportion=0.05444, over 10522.23 utterances.], batch size: 49, lr: 9.72e-03, grad_scale: 8.0 2023-03-08 05:38:58,677 INFO [train2.py:809] (3/4) Epoch 11, batch 3650, loss[ctc_loss=0.08401, att_loss=0.2371, loss=0.2065, over 16410.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007027, over 44.00 utterances.], tot_loss[ctc_loss=0.1025, att_loss=0.2505, loss=0.2209, over 3264749.33 frames. utt_duration=1230 frames, utt_pad_proportion=0.06136, over 10630.84 utterances.], batch size: 44, lr: 9.71e-03, grad_scale: 8.0 2023-03-08 05:39:10,788 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-08 05:39:41,280 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:39:42,823 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:39:53,256 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.334e+02 2.934e+02 3.698e+02 7.886e+02, threshold=5.869e+02, percent-clipped=6.0 2023-03-08 05:40:14,889 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:40:19,100 INFO [train2.py:809] (3/4) Epoch 11, batch 3700, loss[ctc_loss=0.07159, att_loss=0.2315, loss=0.1995, over 15345.00 frames. utt_duration=1755 frames, utt_pad_proportion=0.01004, over 35.00 utterances.], tot_loss[ctc_loss=0.1033, att_loss=0.2507, loss=0.2212, over 3261123.75 frames. utt_duration=1219 frames, utt_pad_proportion=0.06414, over 10710.29 utterances.], batch size: 35, lr: 9.71e-03, grad_scale: 8.0 2023-03-08 05:40:26,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.83 vs. limit=5.0 2023-03-08 05:40:59,608 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:41:19,022 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:41:30,967 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:41:38,578 INFO [train2.py:809] (3/4) Epoch 11, batch 3750, loss[ctc_loss=0.09295, att_loss=0.2398, loss=0.2104, over 16406.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.005916, over 44.00 utterances.], tot_loss[ctc_loss=0.103, att_loss=0.2498, loss=0.2204, over 3256482.63 frames. utt_duration=1214 frames, utt_pad_proportion=0.06739, over 10746.56 utterances.], batch size: 44, lr: 9.70e-03, grad_scale: 8.0 2023-03-08 05:42:32,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.236e+02 2.829e+02 3.523e+02 7.340e+02, threshold=5.658e+02, percent-clipped=1.0 2023-03-08 05:42:58,804 INFO [train2.py:809] (3/4) Epoch 11, batch 3800, loss[ctc_loss=0.1298, att_loss=0.2742, loss=0.2454, over 17126.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.0139, over 56.00 utterances.], tot_loss[ctc_loss=0.1023, att_loss=0.2496, loss=0.2202, over 3255606.88 frames. utt_duration=1227 frames, utt_pad_proportion=0.06298, over 10625.20 utterances.], batch size: 56, lr: 9.70e-03, grad_scale: 8.0 2023-03-08 05:44:06,353 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5417, 2.7017, 3.5700, 2.9805, 3.5048, 4.6253, 4.3607, 3.3827], device='cuda:3'), covar=tensor([0.0357, 0.1786, 0.1060, 0.1275, 0.0986, 0.0779, 0.0462, 0.1295], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0228, 0.0246, 0.0203, 0.0237, 0.0303, 0.0216, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:44:18,443 INFO [train2.py:809] (3/4) Epoch 11, batch 3850, loss[ctc_loss=0.09721, att_loss=0.2397, loss=0.2112, over 16000.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007605, over 40.00 utterances.], tot_loss[ctc_loss=0.1024, att_loss=0.2497, loss=0.2203, over 3263510.77 frames. utt_duration=1235 frames, utt_pad_proportion=0.05884, over 10581.28 utterances.], batch size: 40, lr: 9.69e-03, grad_scale: 8.0 2023-03-08 05:44:18,776 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6310, 2.7108, 3.6843, 4.5371, 4.0298, 3.9606, 2.9723, 2.4887], device='cuda:3'), covar=tensor([0.0665, 0.2181, 0.0885, 0.0531, 0.0705, 0.0377, 0.1456, 0.2044], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0207, 0.0187, 0.0191, 0.0189, 0.0152, 0.0195, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 05:45:10,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.441e+02 3.005e+02 4.132e+02 9.940e+02, threshold=6.009e+02, percent-clipped=5.0 2023-03-08 05:45:35,983 INFO [train2.py:809] (3/4) Epoch 11, batch 3900, loss[ctc_loss=0.1033, att_loss=0.2325, loss=0.2066, over 15365.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01155, over 35.00 utterances.], tot_loss[ctc_loss=0.1025, att_loss=0.2502, loss=0.2207, over 3270171.06 frames. utt_duration=1250 frames, utt_pad_proportion=0.05423, over 10480.42 utterances.], batch size: 35, lr: 9.69e-03, grad_scale: 8.0 2023-03-08 05:46:33,705 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8722, 4.0073, 4.1044, 4.0776, 4.1099, 4.0946, 3.9070, 3.7988], device='cuda:3'), covar=tensor([0.1001, 0.0731, 0.0298, 0.0455, 0.0359, 0.0381, 0.0328, 0.0343], device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0288, 0.0247, 0.0283, 0.0346, 0.0358, 0.0285, 0.0319], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 05:46:51,873 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:46:53,045 INFO [train2.py:809] (3/4) Epoch 11, batch 3950, loss[ctc_loss=0.092, att_loss=0.2456, loss=0.2148, over 16292.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006518, over 43.00 utterances.], tot_loss[ctc_loss=0.1034, att_loss=0.2507, loss=0.2213, over 3262606.73 frames. utt_duration=1218 frames, utt_pad_proportion=0.06387, over 10728.27 utterances.], batch size: 43, lr: 9.68e-03, grad_scale: 8.0 2023-03-08 05:46:56,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-03-08 05:47:13,708 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1812, 4.6316, 4.7512, 4.7298, 2.6475, 4.3367, 2.7535, 1.6458], device='cuda:3'), covar=tensor([0.0360, 0.0137, 0.0509, 0.0114, 0.1763, 0.0167, 0.1400, 0.1829], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0113, 0.0256, 0.0109, 0.0220, 0.0109, 0.0222, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 05:48:13,297 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.409e+02 2.914e+02 3.960e+02 1.049e+03, threshold=5.829e+02, percent-clipped=5.0 2023-03-08 05:48:13,341 INFO [train2.py:809] (3/4) Epoch 12, batch 0, loss[ctc_loss=0.06716, att_loss=0.2086, loss=0.1803, over 15501.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008658, over 36.00 utterances.], tot_loss[ctc_loss=0.06716, att_loss=0.2086, loss=0.1803, over 15501.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008658, over 36.00 utterances.], batch size: 36, lr: 9.27e-03, grad_scale: 8.0 2023-03-08 05:48:13,341 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 05:48:20,696 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9737, 3.9790, 3.8250, 2.2502, 3.7747, 3.8625, 3.4304, 2.3804], device='cuda:3'), covar=tensor([0.0139, 0.0134, 0.0257, 0.1177, 0.0149, 0.0224, 0.0386, 0.1478], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0086, 0.0078, 0.0107, 0.0072, 0.0097, 0.0095, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 05:48:25,648 INFO [train2.py:843] (3/4) Epoch 12, validation: ctc_loss=0.04785, att_loss=0.2384, loss=0.2003, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 05:48:25,649 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 05:48:25,946 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3103, 2.2655, 3.5178, 2.6216, 3.2432, 4.6112, 4.4857, 2.6886], device='cuda:3'), covar=tensor([0.0558, 0.2318, 0.0929, 0.1801, 0.1049, 0.0522, 0.0434, 0.1996], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0225, 0.0245, 0.0203, 0.0235, 0.0301, 0.0215, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:49:09,181 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:49:45,229 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:49:47,998 INFO [train2.py:809] (3/4) Epoch 12, batch 50, loss[ctc_loss=0.1132, att_loss=0.2588, loss=0.2297, over 16260.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.00765, over 43.00 utterances.], tot_loss[ctc_loss=0.1008, att_loss=0.2497, loss=0.2199, over 738634.92 frames. utt_duration=1178 frames, utt_pad_proportion=0.07192, over 2510.52 utterances.], batch size: 43, lr: 9.26e-03, grad_scale: 8.0 2023-03-08 05:50:18,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.51 vs. limit=5.0 2023-03-08 05:50:48,046 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:51:09,635 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.447e+02 2.836e+02 3.349e+02 4.947e+02, threshold=5.672e+02, percent-clipped=0.0 2023-03-08 05:51:09,679 INFO [train2.py:809] (3/4) Epoch 12, batch 100, loss[ctc_loss=0.1097, att_loss=0.2668, loss=0.2354, over 17139.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01329, over 56.00 utterances.], tot_loss[ctc_loss=0.1014, att_loss=0.2508, loss=0.2209, over 1304407.05 frames. utt_duration=1227 frames, utt_pad_proportion=0.05737, over 4256.80 utterances.], batch size: 56, lr: 9.26e-03, grad_scale: 8.0 2023-03-08 05:51:20,104 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7873, 2.9025, 3.8117, 3.1859, 3.7432, 4.8170, 4.5995, 3.5936], device='cuda:3'), covar=tensor([0.0333, 0.1694, 0.0975, 0.1371, 0.0908, 0.0643, 0.0371, 0.1139], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0225, 0.0245, 0.0202, 0.0234, 0.0297, 0.0211, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:52:09,333 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:52:28,411 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:52:31,105 INFO [train2.py:809] (3/4) Epoch 12, batch 150, loss[ctc_loss=0.1075, att_loss=0.2661, loss=0.2344, over 16699.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005873, over 46.00 utterances.], tot_loss[ctc_loss=0.09902, att_loss=0.2486, loss=0.2187, over 1735162.79 frames. utt_duration=1203 frames, utt_pad_proportion=0.06665, over 5777.49 utterances.], batch size: 46, lr: 9.25e-03, grad_scale: 8.0 2023-03-08 05:53:52,685 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:53:55,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.379e+02 2.873e+02 3.449e+02 9.028e+02, threshold=5.746e+02, percent-clipped=3.0 2023-03-08 05:53:55,443 INFO [train2.py:809] (3/4) Epoch 12, batch 200, loss[ctc_loss=0.07595, att_loss=0.2394, loss=0.2067, over 16103.00 frames. utt_duration=1535 frames, utt_pad_proportion=0.006853, over 42.00 utterances.], tot_loss[ctc_loss=0.099, att_loss=0.2491, loss=0.2191, over 2078816.79 frames. utt_duration=1232 frames, utt_pad_proportion=0.06037, over 6758.59 utterances.], batch size: 42, lr: 9.25e-03, grad_scale: 8.0 2023-03-08 05:55:14,811 INFO [train2.py:809] (3/4) Epoch 12, batch 250, loss[ctc_loss=0.1526, att_loss=0.2754, loss=0.2508, over 14580.00 frames. utt_duration=400.9 frames, utt_pad_proportion=0.3028, over 146.00 utterances.], tot_loss[ctc_loss=0.09916, att_loss=0.249, loss=0.2191, over 2339937.44 frames. utt_duration=1250 frames, utt_pad_proportion=0.05804, over 7495.96 utterances.], batch size: 146, lr: 9.24e-03, grad_scale: 8.0 2023-03-08 05:56:00,215 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4283, 4.8923, 4.8746, 4.9210, 4.9189, 4.6000, 3.1447, 4.6522], device='cuda:3'), covar=tensor([0.0112, 0.0123, 0.0100, 0.0078, 0.0084, 0.0105, 0.0806, 0.0245], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0069, 0.0085, 0.0052, 0.0058, 0.0069, 0.0090, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 05:56:35,433 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.301e+02 2.863e+02 3.737e+02 8.031e+02, threshold=5.726e+02, percent-clipped=4.0 2023-03-08 05:56:35,478 INFO [train2.py:809] (3/4) Epoch 12, batch 300, loss[ctc_loss=0.101, att_loss=0.2237, loss=0.1992, over 15767.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.008887, over 38.00 utterances.], tot_loss[ctc_loss=0.09978, att_loss=0.2488, loss=0.219, over 2542777.48 frames. utt_duration=1240 frames, utt_pad_proportion=0.05962, over 8210.20 utterances.], batch size: 38, lr: 9.24e-03, grad_scale: 8.0 2023-03-08 05:57:10,235 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:57:53,912 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:57:56,723 INFO [train2.py:809] (3/4) Epoch 12, batch 350, loss[ctc_loss=0.07068, att_loss=0.2318, loss=0.1996, over 16419.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.005928, over 44.00 utterances.], tot_loss[ctc_loss=0.09924, att_loss=0.2488, loss=0.2189, over 2701481.64 frames. utt_duration=1231 frames, utt_pad_proportion=0.0627, over 8790.94 utterances.], batch size: 44, lr: 9.23e-03, grad_scale: 8.0 2023-03-08 05:59:10,320 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 05:59:17,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.261e+02 2.765e+02 3.364e+02 1.191e+03, threshold=5.531e+02, percent-clipped=3.0 2023-03-08 05:59:17,148 INFO [train2.py:809] (3/4) Epoch 12, batch 400, loss[ctc_loss=0.07551, att_loss=0.2232, loss=0.1936, over 16017.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006722, over 40.00 utterances.], tot_loss[ctc_loss=0.09919, att_loss=0.2487, loss=0.2188, over 2835780.69 frames. utt_duration=1251 frames, utt_pad_proportion=0.05454, over 9075.25 utterances.], batch size: 40, lr: 9.23e-03, grad_scale: 8.0 2023-03-08 06:00:24,816 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:00:36,682 INFO [train2.py:809] (3/4) Epoch 12, batch 450, loss[ctc_loss=0.1317, att_loss=0.2726, loss=0.2444, over 16471.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.007031, over 46.00 utterances.], tot_loss[ctc_loss=0.09998, att_loss=0.2488, loss=0.219, over 2925441.94 frames. utt_duration=1253 frames, utt_pad_proportion=0.05644, over 9353.39 utterances.], batch size: 46, lr: 9.22e-03, grad_scale: 8.0 2023-03-08 06:01:00,663 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-08 06:01:44,249 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:01:56,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.333e+02 2.897e+02 3.791e+02 1.244e+03, threshold=5.793e+02, percent-clipped=7.0 2023-03-08 06:01:56,118 INFO [train2.py:809] (3/4) Epoch 12, batch 500, loss[ctc_loss=0.09917, att_loss=0.2624, loss=0.2297, over 16470.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.007121, over 46.00 utterances.], tot_loss[ctc_loss=0.09993, att_loss=0.2491, loss=0.2193, over 3006275.50 frames. utt_duration=1255 frames, utt_pad_proportion=0.0526, over 9591.04 utterances.], batch size: 46, lr: 9.22e-03, grad_scale: 8.0 2023-03-08 06:02:50,295 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9567, 5.2741, 4.7555, 5.3430, 4.6925, 4.9452, 5.4166, 5.2444], device='cuda:3'), covar=tensor([0.0521, 0.0324, 0.0896, 0.0270, 0.0451, 0.0264, 0.0225, 0.0166], device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0264, 0.0318, 0.0261, 0.0271, 0.0207, 0.0250, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 06:03:15,765 INFO [train2.py:809] (3/4) Epoch 12, batch 550, loss[ctc_loss=0.08244, att_loss=0.2266, loss=0.1977, over 15501.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008801, over 36.00 utterances.], tot_loss[ctc_loss=0.09981, att_loss=0.2485, loss=0.2187, over 3068180.33 frames. utt_duration=1243 frames, utt_pad_proportion=0.05359, over 9883.75 utterances.], batch size: 36, lr: 9.21e-03, grad_scale: 8.0 2023-03-08 06:04:01,239 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 06:04:07,089 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2023-03-08 06:04:34,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.230e+02 2.908e+02 3.570e+02 8.469e+02, threshold=5.817e+02, percent-clipped=3.0 2023-03-08 06:04:35,036 INFO [train2.py:809] (3/4) Epoch 12, batch 600, loss[ctc_loss=0.1057, att_loss=0.263, loss=0.2315, over 17483.00 frames. utt_duration=886.8 frames, utt_pad_proportion=0.07239, over 79.00 utterances.], tot_loss[ctc_loss=0.1001, att_loss=0.249, loss=0.2192, over 3120551.15 frames. utt_duration=1255 frames, utt_pad_proportion=0.04828, over 9960.87 utterances.], batch size: 79, lr: 9.21e-03, grad_scale: 8.0 2023-03-08 06:05:08,619 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:05:39,080 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2394, 2.4804, 3.1345, 4.1765, 3.7331, 3.7030, 2.8146, 1.8812], device='cuda:3'), covar=tensor([0.0705, 0.2296, 0.0993, 0.0515, 0.0705, 0.0430, 0.1447, 0.2492], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0206, 0.0186, 0.0196, 0.0190, 0.0153, 0.0195, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:05:53,594 INFO [train2.py:809] (3/4) Epoch 12, batch 650, loss[ctc_loss=0.09513, att_loss=0.2355, loss=0.2074, over 15889.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008942, over 39.00 utterances.], tot_loss[ctc_loss=0.1008, att_loss=0.2499, loss=0.2201, over 3159632.08 frames. utt_duration=1243 frames, utt_pad_proportion=0.05084, over 10177.62 utterances.], batch size: 39, lr: 9.20e-03, grad_scale: 8.0 2023-03-08 06:06:13,790 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-03-08 06:06:23,477 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:07:12,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.522e+02 3.198e+02 3.829e+02 8.235e+02, threshold=6.395e+02, percent-clipped=2.0 2023-03-08 06:07:12,401 INFO [train2.py:809] (3/4) Epoch 12, batch 700, loss[ctc_loss=0.08668, att_loss=0.2305, loss=0.2017, over 15893.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008355, over 39.00 utterances.], tot_loss[ctc_loss=0.1023, att_loss=0.2502, loss=0.2207, over 3175193.45 frames. utt_duration=1202 frames, utt_pad_proportion=0.06592, over 10580.44 utterances.], batch size: 39, lr: 9.20e-03, grad_scale: 8.0 2023-03-08 06:07:20,663 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:07:57,874 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:08:08,613 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8674, 4.9277, 4.7744, 2.8789, 4.7206, 4.5329, 4.2370, 2.6591], device='cuda:3'), covar=tensor([0.0116, 0.0084, 0.0284, 0.1029, 0.0085, 0.0196, 0.0331, 0.1478], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0085, 0.0079, 0.0105, 0.0072, 0.0096, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:08:21,845 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:08:21,966 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0140, 4.5067, 4.4553, 4.7075, 2.4709, 4.4506, 2.5643, 1.6358], device='cuda:3'), covar=tensor([0.0331, 0.0151, 0.0681, 0.0113, 0.1936, 0.0158, 0.1602, 0.1911], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0114, 0.0255, 0.0110, 0.0216, 0.0108, 0.0222, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:08:29,717 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3833, 2.8499, 3.6313, 4.4183, 4.1441, 4.0259, 2.8756, 2.3024], device='cuda:3'), covar=tensor([0.0688, 0.2010, 0.0807, 0.0499, 0.0605, 0.0394, 0.1555, 0.2242], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0203, 0.0183, 0.0192, 0.0184, 0.0151, 0.0194, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:08:32,492 INFO [train2.py:809] (3/4) Epoch 12, batch 750, loss[ctc_loss=0.1082, att_loss=0.2644, loss=0.2332, over 16884.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006803, over 49.00 utterances.], tot_loss[ctc_loss=0.101, att_loss=0.2493, loss=0.2196, over 3192084.94 frames. utt_duration=1225 frames, utt_pad_proportion=0.06027, over 10436.65 utterances.], batch size: 49, lr: 9.19e-03, grad_scale: 8.0 2023-03-08 06:08:58,639 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:09:35,559 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:09:39,035 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:09:42,323 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:09:52,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.480e+02 2.940e+02 3.562e+02 6.927e+02, threshold=5.880e+02, percent-clipped=2.0 2023-03-08 06:09:52,709 INFO [train2.py:809] (3/4) Epoch 12, batch 800, loss[ctc_loss=0.1131, att_loss=0.2488, loss=0.2217, over 16546.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005912, over 45.00 utterances.], tot_loss[ctc_loss=0.1005, att_loss=0.2486, loss=0.219, over 3203809.25 frames. utt_duration=1257 frames, utt_pad_proportion=0.05464, over 10206.29 utterances.], batch size: 45, lr: 9.19e-03, grad_scale: 8.0 2023-03-08 06:09:54,523 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7666, 5.0296, 5.0521, 4.9600, 5.0930, 5.0814, 4.8031, 4.5520], device='cuda:3'), covar=tensor([0.1010, 0.0523, 0.0235, 0.0588, 0.0297, 0.0273, 0.0311, 0.0331], device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0294, 0.0253, 0.0292, 0.0352, 0.0361, 0.0291, 0.0327], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 06:10:26,566 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1469, 5.3031, 5.0598, 2.5279, 1.9814, 2.9708, 2.7499, 3.7780], device='cuda:3'), covar=tensor([0.0629, 0.0222, 0.0231, 0.4633, 0.5896, 0.2288, 0.2757, 0.1966], device='cuda:3'), in_proj_covar=tensor([0.0337, 0.0226, 0.0242, 0.0215, 0.0349, 0.0337, 0.0233, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.5074e-04, 8.3163e-05, 1.0372e-04, 9.5436e-05, 1.5162e-04, 1.3600e-04, 9.1690e-05, 1.4927e-04], device='cuda:3') 2023-03-08 06:10:55,652 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 06:10:58,498 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:11:12,234 INFO [train2.py:809] (3/4) Epoch 12, batch 850, loss[ctc_loss=0.108, att_loss=0.2558, loss=0.2262, over 17024.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007555, over 51.00 utterances.], tot_loss[ctc_loss=0.09999, att_loss=0.2484, loss=0.2187, over 3222823.89 frames. utt_duration=1272 frames, utt_pad_proportion=0.04912, over 10145.33 utterances.], batch size: 51, lr: 9.18e-03, grad_scale: 8.0 2023-03-08 06:12:30,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.371e+02 2.719e+02 3.452e+02 7.248e+02, threshold=5.437e+02, percent-clipped=1.0 2023-03-08 06:12:30,618 INFO [train2.py:809] (3/4) Epoch 12, batch 900, loss[ctc_loss=0.09214, att_loss=0.2472, loss=0.2162, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005573, over 47.00 utterances.], tot_loss[ctc_loss=0.1004, att_loss=0.249, loss=0.2193, over 3238901.57 frames. utt_duration=1272 frames, utt_pad_proportion=0.04708, over 10194.80 utterances.], batch size: 47, lr: 9.18e-03, grad_scale: 8.0 2023-03-08 06:13:11,853 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2022, 5.2794, 5.1381, 2.5226, 2.0268, 2.7533, 2.9700, 3.9126], device='cuda:3'), covar=tensor([0.0600, 0.0227, 0.0221, 0.4453, 0.5960, 0.2689, 0.2402, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0223, 0.0239, 0.0211, 0.0345, 0.0334, 0.0228, 0.0350], device='cuda:3'), out_proj_covar=tensor([1.4882e-04, 8.2140e-05, 1.0244e-04, 9.4033e-05, 1.5002e-04, 1.3474e-04, 9.0101e-05, 1.4758e-04], device='cuda:3') 2023-03-08 06:13:49,827 INFO [train2.py:809] (3/4) Epoch 12, batch 950, loss[ctc_loss=0.09674, att_loss=0.2501, loss=0.2195, over 16329.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006262, over 45.00 utterances.], tot_loss[ctc_loss=0.09998, att_loss=0.2496, loss=0.2197, over 3253617.39 frames. utt_duration=1273 frames, utt_pad_proportion=0.04561, over 10234.05 utterances.], batch size: 45, lr: 9.17e-03, grad_scale: 8.0 2023-03-08 06:13:50,148 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1125, 5.0403, 4.8858, 2.8213, 4.8410, 4.6623, 4.1426, 2.6207], device='cuda:3'), covar=tensor([0.0099, 0.0082, 0.0282, 0.1111, 0.0087, 0.0179, 0.0350, 0.1457], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0085, 0.0078, 0.0105, 0.0071, 0.0096, 0.0094, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:14:44,486 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1004, 4.0784, 3.9860, 4.1271, 4.4347, 4.0480, 4.0405, 2.3366], device='cuda:3'), covar=tensor([0.0245, 0.0353, 0.0339, 0.0199, 0.0998, 0.0240, 0.0268, 0.2004], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0132, 0.0136, 0.0144, 0.0330, 0.0119, 0.0119, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 06:14:45,850 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0413, 5.3110, 5.3716, 5.2505, 5.3730, 5.3292, 5.0684, 4.7510], device='cuda:3'), covar=tensor([0.0999, 0.0541, 0.0214, 0.0519, 0.0315, 0.0322, 0.0324, 0.0399], device='cuda:3'), in_proj_covar=tensor([0.0456, 0.0293, 0.0253, 0.0289, 0.0351, 0.0365, 0.0289, 0.0326], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 06:14:47,875 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-08 06:15:00,924 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7322, 4.6287, 4.5849, 4.6153, 5.1581, 4.5653, 4.6642, 2.3331], device='cuda:3'), covar=tensor([0.0174, 0.0289, 0.0249, 0.0296, 0.1046, 0.0207, 0.0244, 0.2151], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0132, 0.0136, 0.0144, 0.0329, 0.0119, 0.0119, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 06:15:09,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.480e+02 3.005e+02 3.455e+02 6.503e+02, threshold=6.009e+02, percent-clipped=2.0 2023-03-08 06:15:09,843 INFO [train2.py:809] (3/4) Epoch 12, batch 1000, loss[ctc_loss=0.1061, att_loss=0.2681, loss=0.2357, over 17064.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.007393, over 52.00 utterances.], tot_loss[ctc_loss=0.0997, att_loss=0.2499, loss=0.2198, over 3260475.26 frames. utt_duration=1244 frames, utt_pad_proportion=0.05199, over 10497.62 utterances.], batch size: 52, lr: 9.17e-03, grad_scale: 8.0 2023-03-08 06:16:28,102 INFO [train2.py:809] (3/4) Epoch 12, batch 1050, loss[ctc_loss=0.08838, att_loss=0.2417, loss=0.211, over 16882.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006759, over 49.00 utterances.], tot_loss[ctc_loss=0.1009, att_loss=0.2496, loss=0.2199, over 3259408.50 frames. utt_duration=1231 frames, utt_pad_proportion=0.05702, over 10607.57 utterances.], batch size: 49, lr: 9.16e-03, grad_scale: 8.0 2023-03-08 06:16:43,797 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2023-03-08 06:16:46,103 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:17:09,430 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2615, 3.6844, 3.2503, 3.5101, 4.0057, 3.6380, 3.0646, 4.3374], device='cuda:3'), covar=tensor([0.0853, 0.0535, 0.1011, 0.0623, 0.0713, 0.0661, 0.0785, 0.0543], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0207, 0.0177, 0.0242, 0.0214, 0.0185, 0.0256], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 06:17:22,336 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:17:47,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.358e+02 3.023e+02 3.796e+02 1.133e+03, threshold=6.045e+02, percent-clipped=5.0 2023-03-08 06:17:47,253 INFO [train2.py:809] (3/4) Epoch 12, batch 1100, loss[ctc_loss=0.1113, att_loss=0.2582, loss=0.2288, over 16334.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006095, over 45.00 utterances.], tot_loss[ctc_loss=0.1011, att_loss=0.2499, loss=0.2201, over 3268278.35 frames. utt_duration=1251 frames, utt_pad_proportion=0.05012, over 10458.45 utterances.], batch size: 45, lr: 9.16e-03, grad_scale: 8.0 2023-03-08 06:19:06,841 INFO [train2.py:809] (3/4) Epoch 12, batch 1150, loss[ctc_loss=0.07317, att_loss=0.2203, loss=0.1909, over 14557.00 frames. utt_duration=1821 frames, utt_pad_proportion=0.0449, over 32.00 utterances.], tot_loss[ctc_loss=0.1017, att_loss=0.2498, loss=0.2202, over 3256554.32 frames. utt_duration=1224 frames, utt_pad_proportion=0.06148, over 10658.38 utterances.], batch size: 32, lr: 9.15e-03, grad_scale: 8.0 2023-03-08 06:20:09,411 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-08 06:20:27,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.531e+02 3.116e+02 3.796e+02 1.058e+03, threshold=6.232e+02, percent-clipped=4.0 2023-03-08 06:20:27,149 INFO [train2.py:809] (3/4) Epoch 12, batch 1200, loss[ctc_loss=0.08648, att_loss=0.2493, loss=0.2167, over 16178.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006412, over 41.00 utterances.], tot_loss[ctc_loss=0.1018, att_loss=0.2499, loss=0.2203, over 3258635.17 frames. utt_duration=1205 frames, utt_pad_proportion=0.06727, over 10833.44 utterances.], batch size: 41, lr: 9.15e-03, grad_scale: 8.0 2023-03-08 06:21:07,161 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8597, 6.0111, 5.4936, 5.8030, 5.7544, 5.2655, 5.5406, 5.2679], device='cuda:3'), covar=tensor([0.1149, 0.0953, 0.0724, 0.0711, 0.0787, 0.1421, 0.2320, 0.2356], device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0521, 0.0390, 0.0390, 0.0379, 0.0438, 0.0530, 0.0468], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 06:21:22,303 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:21:38,648 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.20 vs. limit=5.0 2023-03-08 06:21:47,504 INFO [train2.py:809] (3/4) Epoch 12, batch 1250, loss[ctc_loss=0.1103, att_loss=0.2436, loss=0.217, over 11306.00 frames. utt_duration=1811 frames, utt_pad_proportion=0.03176, over 25.00 utterances.], tot_loss[ctc_loss=0.1016, att_loss=0.2499, loss=0.2202, over 3249006.20 frames. utt_duration=1203 frames, utt_pad_proportion=0.06701, over 10811.97 utterances.], batch size: 25, lr: 9.14e-03, grad_scale: 8.0 2023-03-08 06:22:02,294 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:22:58,411 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 06:23:05,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.248e+02 2.874e+02 3.449e+02 8.124e+02, threshold=5.748e+02, percent-clipped=2.0 2023-03-08 06:23:05,913 INFO [train2.py:809] (3/4) Epoch 12, batch 1300, loss[ctc_loss=0.1117, att_loss=0.2379, loss=0.2127, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006939, over 40.00 utterances.], tot_loss[ctc_loss=0.1012, att_loss=0.2492, loss=0.2196, over 3256442.83 frames. utt_duration=1227 frames, utt_pad_proportion=0.06133, over 10632.92 utterances.], batch size: 40, lr: 9.14e-03, grad_scale: 8.0 2023-03-08 06:23:37,511 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:23:40,684 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4658, 2.3476, 4.8614, 3.9756, 2.8683, 4.1793, 4.5404, 4.5288], device='cuda:3'), covar=tensor([0.0175, 0.1850, 0.0092, 0.0904, 0.1860, 0.0258, 0.0125, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0243, 0.0136, 0.0302, 0.0270, 0.0183, 0.0121, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 06:23:51,097 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3790, 2.5853, 3.3684, 4.2549, 3.8655, 3.9308, 2.7998, 2.1052], device='cuda:3'), covar=tensor([0.0715, 0.2319, 0.1035, 0.0561, 0.0762, 0.0365, 0.1533, 0.2458], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0209, 0.0187, 0.0196, 0.0191, 0.0152, 0.0196, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:24:24,487 INFO [train2.py:809] (3/4) Epoch 12, batch 1350, loss[ctc_loss=0.09262, att_loss=0.2535, loss=0.2213, over 17389.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03365, over 63.00 utterances.], tot_loss[ctc_loss=0.1016, att_loss=0.2504, loss=0.2207, over 3272055.94 frames. utt_duration=1219 frames, utt_pad_proportion=0.05853, over 10751.19 utterances.], batch size: 63, lr: 9.13e-03, grad_scale: 8.0 2023-03-08 06:24:27,017 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-08 06:24:37,414 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3438, 4.9489, 4.7541, 4.7823, 4.9132, 4.5577, 3.1031, 4.6222], device='cuda:3'), covar=tensor([0.0132, 0.0105, 0.0130, 0.0089, 0.0083, 0.0118, 0.0847, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0070, 0.0087, 0.0053, 0.0058, 0.0070, 0.0091, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:24:40,578 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0583, 5.0103, 4.9460, 3.0964, 4.7242, 4.5806, 4.2258, 3.0084], device='cuda:3'), covar=tensor([0.0139, 0.0087, 0.0239, 0.0966, 0.0107, 0.0215, 0.0370, 0.1236], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0085, 0.0079, 0.0105, 0.0072, 0.0097, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:24:42,130 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:25:18,416 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:25:35,559 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3916, 3.9291, 3.4072, 3.7182, 4.2354, 3.8929, 3.3640, 4.5797], device='cuda:3'), covar=tensor([0.0889, 0.0426, 0.0995, 0.0562, 0.0597, 0.0576, 0.0741, 0.0453], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0187, 0.0207, 0.0175, 0.0242, 0.0212, 0.0184, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 06:25:43,525 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.386e+02 2.782e+02 3.300e+02 6.204e+02, threshold=5.564e+02, percent-clipped=3.0 2023-03-08 06:25:43,570 INFO [train2.py:809] (3/4) Epoch 12, batch 1400, loss[ctc_loss=0.07753, att_loss=0.2297, loss=0.1993, over 15751.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.009955, over 38.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.2508, loss=0.221, over 3277832.44 frames. utt_duration=1225 frames, utt_pad_proportion=0.05566, over 10717.17 utterances.], batch size: 38, lr: 9.13e-03, grad_scale: 8.0 2023-03-08 06:25:58,488 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:26:24,953 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9869, 4.3790, 4.4934, 4.6463, 2.5953, 4.7345, 2.5612, 1.9886], device='cuda:3'), covar=tensor([0.0327, 0.0168, 0.0620, 0.0122, 0.1739, 0.0113, 0.1536, 0.1678], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0114, 0.0256, 0.0112, 0.0218, 0.0107, 0.0224, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:26:34,541 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:27:04,204 INFO [train2.py:809] (3/4) Epoch 12, batch 1450, loss[ctc_loss=0.06273, att_loss=0.2126, loss=0.1826, over 15637.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009336, over 37.00 utterances.], tot_loss[ctc_loss=0.1012, att_loss=0.2503, loss=0.2205, over 3272422.94 frames. utt_duration=1214 frames, utt_pad_proportion=0.06033, over 10792.07 utterances.], batch size: 37, lr: 9.12e-03, grad_scale: 8.0 2023-03-08 06:28:04,254 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8742, 6.1421, 5.5567, 5.9132, 5.8435, 5.3557, 5.4286, 5.2996], device='cuda:3'), covar=tensor([0.1209, 0.0990, 0.0905, 0.0804, 0.0766, 0.1365, 0.2679, 0.2752], device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0517, 0.0391, 0.0392, 0.0377, 0.0436, 0.0533, 0.0470], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 06:28:06,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-03-08 06:28:24,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 2.356e+02 2.763e+02 3.457e+02 6.962e+02, threshold=5.526e+02, percent-clipped=5.0 2023-03-08 06:28:24,658 INFO [train2.py:809] (3/4) Epoch 12, batch 1500, loss[ctc_loss=0.09431, att_loss=0.2454, loss=0.2152, over 16387.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007682, over 44.00 utterances.], tot_loss[ctc_loss=0.09984, att_loss=0.2489, loss=0.2191, over 3261986.96 frames. utt_duration=1217 frames, utt_pad_proportion=0.06274, over 10733.29 utterances.], batch size: 44, lr: 9.12e-03, grad_scale: 8.0 2023-03-08 06:28:36,331 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9937, 5.0024, 4.9348, 2.4982, 4.8315, 4.5897, 4.2007, 2.4305], device='cuda:3'), covar=tensor([0.0117, 0.0090, 0.0203, 0.1271, 0.0086, 0.0189, 0.0355, 0.1570], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0085, 0.0078, 0.0104, 0.0071, 0.0096, 0.0095, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:29:06,691 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 06:29:45,020 INFO [train2.py:809] (3/4) Epoch 12, batch 1550, loss[ctc_loss=0.1187, att_loss=0.2486, loss=0.2226, over 16262.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007513, over 43.00 utterances.], tot_loss[ctc_loss=0.09923, att_loss=0.2484, loss=0.2186, over 3256742.33 frames. utt_duration=1214 frames, utt_pad_proportion=0.06548, over 10745.73 utterances.], batch size: 43, lr: 9.11e-03, grad_scale: 8.0 2023-03-08 06:30:44,858 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 06:30:49,190 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 06:31:05,229 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 2.353e+02 2.825e+02 3.392e+02 6.329e+02, threshold=5.650e+02, percent-clipped=2.0 2023-03-08 06:31:05,273 INFO [train2.py:809] (3/4) Epoch 12, batch 1600, loss[ctc_loss=0.07667, att_loss=0.2376, loss=0.2054, over 15970.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005938, over 41.00 utterances.], tot_loss[ctc_loss=0.1003, att_loss=0.2496, loss=0.2197, over 3258047.22 frames. utt_duration=1213 frames, utt_pad_proportion=0.0645, over 10753.12 utterances.], batch size: 41, lr: 9.11e-03, grad_scale: 16.0 2023-03-08 06:31:26,661 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-08 06:31:28,672 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:32:11,304 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-08 06:32:24,618 INFO [train2.py:809] (3/4) Epoch 12, batch 1650, loss[ctc_loss=0.09349, att_loss=0.2403, loss=0.2109, over 16283.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006355, over 43.00 utterances.], tot_loss[ctc_loss=0.1, att_loss=0.249, loss=0.2192, over 3259651.59 frames. utt_duration=1237 frames, utt_pad_proportion=0.05964, over 10553.07 utterances.], batch size: 43, lr: 9.10e-03, grad_scale: 16.0 2023-03-08 06:33:20,987 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4680, 2.7751, 3.1807, 4.2277, 3.7598, 4.0239, 2.7796, 1.9248], device='cuda:3'), covar=tensor([0.0668, 0.2160, 0.1139, 0.0658, 0.0809, 0.0367, 0.1696, 0.2536], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0209, 0.0187, 0.0196, 0.0192, 0.0152, 0.0198, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 06:33:43,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.369e+02 2.877e+02 3.430e+02 6.900e+02, threshold=5.754e+02, percent-clipped=3.0 2023-03-08 06:33:43,949 INFO [train2.py:809] (3/4) Epoch 12, batch 1700, loss[ctc_loss=0.1138, att_loss=0.2399, loss=0.2147, over 15488.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009472, over 36.00 utterances.], tot_loss[ctc_loss=0.09986, att_loss=0.2482, loss=0.2185, over 3253301.99 frames. utt_duration=1249 frames, utt_pad_proportion=0.05914, over 10430.96 utterances.], batch size: 36, lr: 9.10e-03, grad_scale: 16.0 2023-03-08 06:34:03,400 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9526, 5.3332, 4.6908, 5.4048, 4.7272, 5.1281, 5.4484, 5.2058], device='cuda:3'), covar=tensor([0.0609, 0.0271, 0.0947, 0.0195, 0.0424, 0.0210, 0.0207, 0.0199], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0265, 0.0323, 0.0263, 0.0274, 0.0206, 0.0253, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 06:35:04,020 INFO [train2.py:809] (3/4) Epoch 12, batch 1750, loss[ctc_loss=0.09732, att_loss=0.2604, loss=0.2278, over 16874.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007899, over 49.00 utterances.], tot_loss[ctc_loss=0.1007, att_loss=0.2494, loss=0.2197, over 3260804.81 frames. utt_duration=1213 frames, utt_pad_proportion=0.06644, over 10767.00 utterances.], batch size: 49, lr: 9.09e-03, grad_scale: 16.0 2023-03-08 06:35:47,763 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1053, 5.1753, 5.0431, 2.3445, 1.9941, 2.9537, 3.2128, 4.0103], device='cuda:3'), covar=tensor([0.0720, 0.0260, 0.0223, 0.4907, 0.6177, 0.2629, 0.2314, 0.1639], device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0223, 0.0239, 0.0213, 0.0346, 0.0337, 0.0230, 0.0355], device='cuda:3'), out_proj_covar=tensor([1.5167e-04, 8.2708e-05, 1.0281e-04, 9.4189e-05, 1.5031e-04, 1.3580e-04, 9.0912e-05, 1.4931e-04], device='cuda:3') 2023-03-08 06:36:24,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.369e+02 2.874e+02 3.424e+02 1.036e+03, threshold=5.747e+02, percent-clipped=2.0 2023-03-08 06:36:24,230 INFO [train2.py:809] (3/4) Epoch 12, batch 1800, loss[ctc_loss=0.1147, att_loss=0.2699, loss=0.2389, over 17298.00 frames. utt_duration=1004 frames, utt_pad_proportion=0.05161, over 69.00 utterances.], tot_loss[ctc_loss=0.09987, att_loss=0.2485, loss=0.2188, over 3260196.04 frames. utt_duration=1239 frames, utt_pad_proportion=0.06015, over 10541.82 utterances.], batch size: 69, lr: 9.09e-03, grad_scale: 16.0 2023-03-08 06:36:33,247 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2504, 5.2165, 5.1087, 2.4842, 2.0410, 3.0048, 3.2421, 3.8924], device='cuda:3'), covar=tensor([0.0589, 0.0297, 0.0235, 0.4144, 0.5587, 0.2290, 0.2095, 0.1821], device='cuda:3'), in_proj_covar=tensor([0.0337, 0.0222, 0.0239, 0.0212, 0.0344, 0.0335, 0.0230, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.5089e-04, 8.2243e-05, 1.0242e-04, 9.3638e-05, 1.4933e-04, 1.3495e-04, 9.0690e-05, 1.4878e-04], device='cuda:3') 2023-03-08 06:37:43,872 INFO [train2.py:809] (3/4) Epoch 12, batch 1850, loss[ctc_loss=0.1062, att_loss=0.2478, loss=0.2195, over 16690.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.00626, over 46.00 utterances.], tot_loss[ctc_loss=0.09893, att_loss=0.2482, loss=0.2183, over 3262395.29 frames. utt_duration=1250 frames, utt_pad_proportion=0.05837, over 10456.35 utterances.], batch size: 46, lr: 9.08e-03, grad_scale: 8.0 2023-03-08 06:38:32,215 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:38:35,061 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 06:38:48,036 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 06:38:53,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-03-08 06:39:02,765 INFO [train2.py:809] (3/4) Epoch 12, batch 1900, loss[ctc_loss=0.1409, att_loss=0.2811, loss=0.2531, over 17053.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008839, over 52.00 utterances.], tot_loss[ctc_loss=0.09898, att_loss=0.2482, loss=0.2183, over 3261074.04 frames. utt_duration=1266 frames, utt_pad_proportion=0.05333, over 10318.24 utterances.], batch size: 52, lr: 9.08e-03, grad_scale: 8.0 2023-03-08 06:39:04,238 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.369e+02 2.834e+02 3.743e+02 8.835e+02, threshold=5.667e+02, percent-clipped=7.0 2023-03-08 06:39:26,591 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:40:02,991 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:40:07,783 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:40:21,062 INFO [train2.py:809] (3/4) Epoch 12, batch 1950, loss[ctc_loss=0.09314, att_loss=0.2365, loss=0.2078, over 15939.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007944, over 41.00 utterances.], tot_loss[ctc_loss=0.09965, att_loss=0.2488, loss=0.2189, over 3265904.71 frames. utt_duration=1271 frames, utt_pad_proportion=0.05107, over 10292.43 utterances.], batch size: 41, lr: 9.07e-03, grad_scale: 8.0 2023-03-08 06:40:24,528 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:40:41,623 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:41:41,126 INFO [train2.py:809] (3/4) Epoch 12, batch 2000, loss[ctc_loss=0.1051, att_loss=0.2523, loss=0.2229, over 16614.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005842, over 47.00 utterances.], tot_loss[ctc_loss=0.1005, att_loss=0.2499, loss=0.22, over 3273661.75 frames. utt_duration=1240 frames, utt_pad_proportion=0.05633, over 10570.90 utterances.], batch size: 47, lr: 9.07e-03, grad_scale: 8.0 2023-03-08 06:41:42,586 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.558e+02 3.002e+02 3.965e+02 8.183e+02, threshold=6.004e+02, percent-clipped=7.0 2023-03-08 06:42:02,221 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:42:41,090 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:42:58,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2023-03-08 06:43:00,500 INFO [train2.py:809] (3/4) Epoch 12, batch 2050, loss[ctc_loss=0.07611, att_loss=0.2209, loss=0.1919, over 16169.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007634, over 41.00 utterances.], tot_loss[ctc_loss=0.1013, att_loss=0.2501, loss=0.2203, over 3269717.71 frames. utt_duration=1226 frames, utt_pad_proportion=0.06023, over 10684.87 utterances.], batch size: 41, lr: 9.06e-03, grad_scale: 8.0 2023-03-08 06:43:08,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 2023-03-08 06:43:17,967 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-08 06:43:28,705 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9587, 6.1754, 5.6437, 5.9223, 5.8696, 5.4332, 5.6856, 5.4171], device='cuda:3'), covar=tensor([0.1175, 0.0836, 0.0756, 0.0711, 0.0716, 0.1529, 0.1819, 0.2003], device='cuda:3'), in_proj_covar=tensor([0.0442, 0.0504, 0.0382, 0.0384, 0.0371, 0.0421, 0.0521, 0.0453], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 06:44:18,307 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:44:21,028 INFO [train2.py:809] (3/4) Epoch 12, batch 2100, loss[ctc_loss=0.07305, att_loss=0.2329, loss=0.2009, over 16314.00 frames. utt_duration=1451 frames, utt_pad_proportion=0.007387, over 45.00 utterances.], tot_loss[ctc_loss=0.1011, att_loss=0.2499, loss=0.2202, over 3257828.19 frames. utt_duration=1192 frames, utt_pad_proportion=0.07233, over 10944.64 utterances.], batch size: 45, lr: 9.06e-03, grad_scale: 8.0 2023-03-08 06:44:22,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 2.414e+02 3.055e+02 3.692e+02 7.937e+02, threshold=6.110e+02, percent-clipped=4.0 2023-03-08 06:45:41,060 INFO [train2.py:809] (3/4) Epoch 12, batch 2150, loss[ctc_loss=0.07997, att_loss=0.2232, loss=0.1946, over 15786.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007694, over 38.00 utterances.], tot_loss[ctc_loss=0.1004, att_loss=0.2496, loss=0.2198, over 3264148.43 frames. utt_duration=1223 frames, utt_pad_proportion=0.06302, over 10690.14 utterances.], batch size: 38, lr: 9.05e-03, grad_scale: 8.0 2023-03-08 06:46:24,727 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5965, 5.1066, 4.8794, 4.9122, 4.9389, 4.7471, 3.6193, 4.9281], device='cuda:3'), covar=tensor([0.0119, 0.0119, 0.0118, 0.0078, 0.0115, 0.0117, 0.0655, 0.0204], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0072, 0.0088, 0.0054, 0.0060, 0.0071, 0.0093, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:46:37,279 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 06:47:04,791 INFO [train2.py:809] (3/4) Epoch 12, batch 2200, loss[ctc_loss=0.1055, att_loss=0.2537, loss=0.2241, over 17104.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01596, over 56.00 utterances.], tot_loss[ctc_loss=0.1008, att_loss=0.2498, loss=0.22, over 3269421.19 frames. utt_duration=1210 frames, utt_pad_proportion=0.06488, over 10824.40 utterances.], batch size: 56, lr: 9.05e-03, grad_scale: 8.0 2023-03-08 06:47:06,228 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.315e+02 2.803e+02 3.534e+02 8.357e+02, threshold=5.606e+02, percent-clipped=3.0 2023-03-08 06:47:37,443 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0483, 4.4104, 4.2830, 4.4225, 4.4244, 4.2166, 3.1501, 4.2874], device='cuda:3'), covar=tensor([0.0132, 0.0107, 0.0118, 0.0073, 0.0090, 0.0109, 0.0668, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0071, 0.0087, 0.0054, 0.0060, 0.0071, 0.0092, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:47:39,121 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5022, 5.0039, 4.8216, 4.9789, 5.0372, 4.7542, 3.5458, 4.8364], device='cuda:3'), covar=tensor([0.0119, 0.0086, 0.0098, 0.0062, 0.0075, 0.0092, 0.0636, 0.0172], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0071, 0.0087, 0.0054, 0.0060, 0.0071, 0.0092, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:47:53,444 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 06:48:02,429 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:48:15,525 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-08 06:48:23,934 INFO [train2.py:809] (3/4) Epoch 12, batch 2250, loss[ctc_loss=0.08538, att_loss=0.2443, loss=0.2125, over 15935.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.007746, over 41.00 utterances.], tot_loss[ctc_loss=0.1018, att_loss=0.25, loss=0.2203, over 3259490.04 frames. utt_duration=1173 frames, utt_pad_proportion=0.07785, over 11129.51 utterances.], batch size: 41, lr: 9.04e-03, grad_scale: 8.0 2023-03-08 06:49:01,146 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-08 06:49:06,666 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9715, 5.0254, 4.8555, 2.6083, 4.8323, 4.5790, 4.1889, 2.5129], device='cuda:3'), covar=tensor([0.0115, 0.0067, 0.0225, 0.1203, 0.0071, 0.0164, 0.0325, 0.1449], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0084, 0.0077, 0.0103, 0.0070, 0.0095, 0.0093, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:49:42,482 INFO [train2.py:809] (3/4) Epoch 12, batch 2300, loss[ctc_loss=0.09081, att_loss=0.2434, loss=0.2129, over 17012.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008283, over 51.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.2494, loss=0.2199, over 3251497.28 frames. utt_duration=1195 frames, utt_pad_proportion=0.07372, over 10894.76 utterances.], batch size: 51, lr: 9.04e-03, grad_scale: 8.0 2023-03-08 06:49:44,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.436e+02 3.067e+02 3.650e+02 8.781e+02, threshold=6.133e+02, percent-clipped=4.0 2023-03-08 06:49:55,622 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:51:01,257 INFO [train2.py:809] (3/4) Epoch 12, batch 2350, loss[ctc_loss=0.1207, att_loss=0.2619, loss=0.2337, over 17021.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007628, over 51.00 utterances.], tot_loss[ctc_loss=0.1017, att_loss=0.2497, loss=0.2201, over 3252363.45 frames. utt_duration=1180 frames, utt_pad_proportion=0.07637, over 11043.27 utterances.], batch size: 51, lr: 9.03e-03, grad_scale: 8.0 2023-03-08 06:51:39,052 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9940, 2.0222, 3.3562, 2.3147, 3.2248, 4.3017, 4.2568, 2.4596], device='cuda:3'), covar=tensor([0.0546, 0.2331, 0.0905, 0.1795, 0.0992, 0.0610, 0.0421, 0.1861], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0232, 0.0248, 0.0206, 0.0241, 0.0310, 0.0224, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:52:09,337 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:52:19,911 INFO [train2.py:809] (3/4) Epoch 12, batch 2400, loss[ctc_loss=0.09717, att_loss=0.2463, loss=0.2165, over 16294.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006473, over 43.00 utterances.], tot_loss[ctc_loss=0.1024, att_loss=0.2502, loss=0.2206, over 3257227.86 frames. utt_duration=1179 frames, utt_pad_proportion=0.07525, over 11065.39 utterances.], batch size: 43, lr: 9.03e-03, grad_scale: 8.0 2023-03-08 06:52:21,344 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.431e+02 2.918e+02 3.900e+02 7.255e+02, threshold=5.836e+02, percent-clipped=5.0 2023-03-08 06:53:39,243 INFO [train2.py:809] (3/4) Epoch 12, batch 2450, loss[ctc_loss=0.09886, att_loss=0.2551, loss=0.2239, over 17039.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007418, over 51.00 utterances.], tot_loss[ctc_loss=0.1018, att_loss=0.2497, loss=0.2201, over 3263123.56 frames. utt_duration=1179 frames, utt_pad_proportion=0.07319, over 11087.53 utterances.], batch size: 51, lr: 9.02e-03, grad_scale: 8.0 2023-03-08 06:53:46,243 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:53:58,786 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4971, 3.6603, 3.5689, 3.1243, 3.5252, 3.6232, 3.5515, 2.5911], device='cuda:3'), covar=tensor([0.1381, 0.1552, 0.2348, 0.6214, 0.4708, 0.3028, 0.1207, 0.8245], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0123, 0.0135, 0.0205, 0.0108, 0.0188, 0.0112, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2023-03-08 06:54:49,318 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-03-08 06:54:58,104 INFO [train2.py:809] (3/4) Epoch 12, batch 2500, loss[ctc_loss=0.1217, att_loss=0.266, loss=0.2371, over 17303.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01174, over 55.00 utterances.], tot_loss[ctc_loss=0.102, att_loss=0.2501, loss=0.2204, over 3272144.61 frames. utt_duration=1191 frames, utt_pad_proportion=0.06858, over 11001.73 utterances.], batch size: 55, lr: 9.02e-03, grad_scale: 8.0 2023-03-08 06:55:00,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.509e+02 2.157e+02 2.613e+02 3.362e+02 8.530e+02, threshold=5.227e+02, percent-clipped=3.0 2023-03-08 06:55:00,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1236, 5.0707, 4.8968, 2.8318, 4.7476, 4.6495, 4.2686, 2.6828], device='cuda:3'), covar=tensor([0.0110, 0.0096, 0.0270, 0.1135, 0.0104, 0.0174, 0.0345, 0.1444], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0084, 0.0076, 0.0102, 0.0070, 0.0094, 0.0093, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0002, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 06:55:15,732 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0731, 5.2735, 5.5469, 5.4048, 5.4647, 6.0076, 5.2698, 6.1064], device='cuda:3'), covar=tensor([0.0577, 0.0637, 0.0681, 0.1070, 0.1683, 0.0774, 0.0513, 0.0541], device='cuda:3'), in_proj_covar=tensor([0.0747, 0.0442, 0.0514, 0.0572, 0.0761, 0.0520, 0.0415, 0.0508], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 06:55:22,644 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:55:56,113 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:56:18,495 INFO [train2.py:809] (3/4) Epoch 12, batch 2550, loss[ctc_loss=0.1442, att_loss=0.2814, loss=0.254, over 16899.00 frames. utt_duration=1381 frames, utt_pad_proportion=0.005877, over 49.00 utterances.], tot_loss[ctc_loss=0.1019, att_loss=0.2495, loss=0.22, over 3261811.77 frames. utt_duration=1213 frames, utt_pad_proportion=0.06601, over 10770.97 utterances.], batch size: 49, lr: 9.01e-03, grad_scale: 8.0 2023-03-08 06:57:08,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.67 vs. limit=5.0 2023-03-08 06:57:12,240 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:57:38,532 INFO [train2.py:809] (3/4) Epoch 12, batch 2600, loss[ctc_loss=0.09327, att_loss=0.2629, loss=0.229, over 17315.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.0359, over 63.00 utterances.], tot_loss[ctc_loss=0.1011, att_loss=0.2493, loss=0.2197, over 3257720.01 frames. utt_duration=1195 frames, utt_pad_proportion=0.07168, over 10917.12 utterances.], batch size: 63, lr: 9.01e-03, grad_scale: 8.0 2023-03-08 06:57:39,865 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.587e+02 3.130e+02 3.974e+02 7.202e+02, threshold=6.259e+02, percent-clipped=6.0 2023-03-08 06:57:51,771 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:57:51,940 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7703, 2.3326, 5.0568, 3.9796, 3.0463, 4.3023, 5.0454, 4.7751], device='cuda:3'), covar=tensor([0.0178, 0.1908, 0.0179, 0.1041, 0.1854, 0.0239, 0.0079, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0247, 0.0140, 0.0309, 0.0273, 0.0186, 0.0127, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 06:58:56,052 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 06:58:58,112 INFO [train2.py:809] (3/4) Epoch 12, batch 2650, loss[ctc_loss=0.06444, att_loss=0.2121, loss=0.1826, over 15896.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.00877, over 39.00 utterances.], tot_loss[ctc_loss=0.1006, att_loss=0.2485, loss=0.2189, over 3252939.59 frames. utt_duration=1190 frames, utt_pad_proportion=0.07337, over 10946.34 utterances.], batch size: 39, lr: 9.00e-03, grad_scale: 8.0 2023-03-08 06:59:08,161 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 06:59:20,163 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-03-08 07:00:07,671 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:00:18,193 INFO [train2.py:809] (3/4) Epoch 12, batch 2700, loss[ctc_loss=0.1684, att_loss=0.2897, loss=0.2655, over 13956.00 frames. utt_duration=383.9 frames, utt_pad_proportion=0.3324, over 146.00 utterances.], tot_loss[ctc_loss=0.1007, att_loss=0.249, loss=0.2193, over 3255077.31 frames. utt_duration=1175 frames, utt_pad_proportion=0.07835, over 11091.37 utterances.], batch size: 146, lr: 9.00e-03, grad_scale: 8.0 2023-03-08 07:00:19,662 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 2.444e+02 2.933e+02 3.610e+02 6.469e+02, threshold=5.867e+02, percent-clipped=1.0 2023-03-08 07:01:24,154 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:01:37,869 INFO [train2.py:809] (3/4) Epoch 12, batch 2750, loss[ctc_loss=0.1151, att_loss=0.2722, loss=0.2408, over 17042.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009274, over 52.00 utterances.], tot_loss[ctc_loss=0.1008, att_loss=0.2492, loss=0.2195, over 3258138.28 frames. utt_duration=1180 frames, utt_pad_proportion=0.07635, over 11055.69 utterances.], batch size: 52, lr: 9.00e-03, grad_scale: 8.0 2023-03-08 07:02:15,254 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:02:26,702 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:02:57,840 INFO [train2.py:809] (3/4) Epoch 12, batch 2800, loss[ctc_loss=0.07868, att_loss=0.2137, loss=0.1867, over 15383.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01045, over 35.00 utterances.], tot_loss[ctc_loss=0.1008, att_loss=0.249, loss=0.2193, over 3261331.84 frames. utt_duration=1183 frames, utt_pad_proportion=0.07417, over 11038.66 utterances.], batch size: 35, lr: 8.99e-03, grad_scale: 8.0 2023-03-08 07:02:59,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 2.286e+02 2.805e+02 3.463e+02 9.248e+02, threshold=5.611e+02, percent-clipped=3.0 2023-03-08 07:03:14,335 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:03:52,974 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 07:04:04,705 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:04:15,563 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7005, 4.9610, 5.2202, 5.1081, 5.1240, 5.5940, 5.0447, 5.7460], device='cuda:3'), covar=tensor([0.0802, 0.0758, 0.0815, 0.1316, 0.2131, 0.1191, 0.0779, 0.0801], device='cuda:3'), in_proj_covar=tensor([0.0739, 0.0435, 0.0512, 0.0571, 0.0748, 0.0521, 0.0415, 0.0504], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 07:04:18,471 INFO [train2.py:809] (3/4) Epoch 12, batch 2850, loss[ctc_loss=0.1241, att_loss=0.2799, loss=0.2487, over 17043.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009361, over 52.00 utterances.], tot_loss[ctc_loss=0.09992, att_loss=0.2477, loss=0.2181, over 3253843.37 frames. utt_duration=1204 frames, utt_pad_proportion=0.07052, over 10819.59 utterances.], batch size: 52, lr: 8.99e-03, grad_scale: 8.0 2023-03-08 07:04:40,960 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3660, 3.5486, 3.4101, 2.9542, 3.5272, 3.5965, 3.3856, 2.4716], device='cuda:3'), covar=tensor([0.1392, 0.1773, 0.2844, 0.7442, 0.3737, 0.3256, 0.1342, 0.7868], device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0126, 0.0139, 0.0209, 0.0109, 0.0191, 0.0116, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:05:14,227 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0891, 5.3103, 5.5994, 5.4524, 5.5742, 6.0431, 5.2130, 6.1688], device='cuda:3'), covar=tensor([0.0598, 0.0602, 0.0723, 0.1026, 0.1628, 0.0785, 0.0553, 0.0598], device='cuda:3'), in_proj_covar=tensor([0.0733, 0.0431, 0.0509, 0.0566, 0.0743, 0.0514, 0.0414, 0.0499], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 07:05:39,964 INFO [train2.py:809] (3/4) Epoch 12, batch 2900, loss[ctc_loss=0.09189, att_loss=0.2328, loss=0.2046, over 16177.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.007052, over 41.00 utterances.], tot_loss[ctc_loss=0.09985, att_loss=0.248, loss=0.2183, over 3266314.75 frames. utt_duration=1224 frames, utt_pad_proportion=0.06305, over 10688.36 utterances.], batch size: 41, lr: 8.98e-03, grad_scale: 8.0 2023-03-08 07:05:41,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.133e+02 2.612e+02 3.383e+02 7.389e+02, threshold=5.223e+02, percent-clipped=3.0 2023-03-08 07:05:48,892 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1428, 2.0580, 2.7171, 2.3470, 2.4674, 2.7845, 1.9931, 1.9485], device='cuda:3'), covar=tensor([0.1030, 0.4128, 0.2585, 0.1856, 0.1919, 0.1163, 0.3675, 0.1642], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0085, 0.0089, 0.0073, 0.0077, 0.0068, 0.0088, 0.0061], device='cuda:3'), out_proj_covar=tensor([5.2523e-05, 5.9978e-05, 6.2645e-05, 5.1468e-05, 5.1916e-05, 5.0205e-05, 6.1050e-05, 4.6075e-05], device='cuda:3') 2023-03-08 07:06:05,677 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-08 07:06:37,871 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1954, 3.9674, 3.2044, 3.6286, 4.0558, 3.7228, 2.7432, 4.4513], device='cuda:3'), covar=tensor([0.1009, 0.0415, 0.1212, 0.0615, 0.0616, 0.0691, 0.1000, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0190, 0.0210, 0.0181, 0.0248, 0.0219, 0.0186, 0.0260], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:07:00,900 INFO [train2.py:809] (3/4) Epoch 12, batch 2950, loss[ctc_loss=0.07847, att_loss=0.2271, loss=0.1974, over 16270.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007996, over 43.00 utterances.], tot_loss[ctc_loss=0.09886, att_loss=0.2473, loss=0.2176, over 3266057.32 frames. utt_duration=1261 frames, utt_pad_proportion=0.05434, over 10376.06 utterances.], batch size: 43, lr: 8.98e-03, grad_scale: 8.0 2023-03-08 07:07:15,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 07:07:35,320 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3279, 5.0964, 4.9890, 4.8762, 5.5413, 5.3213, 4.9340, 2.5406], device='cuda:3'), covar=tensor([0.0070, 0.0124, 0.0103, 0.0151, 0.0691, 0.0065, 0.0132, 0.1878], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0136, 0.0136, 0.0149, 0.0336, 0.0122, 0.0125, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:07:50,565 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1523, 5.1794, 4.9900, 2.2972, 1.9517, 2.9730, 3.0662, 3.9053], device='cuda:3'), covar=tensor([0.0584, 0.0253, 0.0234, 0.4612, 0.6003, 0.2492, 0.2210, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0223, 0.0238, 0.0215, 0.0343, 0.0333, 0.0231, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.4942e-04, 8.3383e-05, 1.0344e-04, 9.4819e-05, 1.4930e-04, 1.3403e-04, 9.1004e-05, 1.4874e-04], device='cuda:3') 2023-03-08 07:08:03,485 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:08:12,852 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6511, 2.7551, 5.1016, 3.9016, 3.1362, 4.4023, 4.9407, 4.6769], device='cuda:3'), covar=tensor([0.0210, 0.1567, 0.0145, 0.0984, 0.1793, 0.0234, 0.0093, 0.0215], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0246, 0.0139, 0.0306, 0.0268, 0.0185, 0.0126, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:08:23,020 INFO [train2.py:809] (3/4) Epoch 12, batch 3000, loss[ctc_loss=0.08907, att_loss=0.233, loss=0.2042, over 16006.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.006975, over 40.00 utterances.], tot_loss[ctc_loss=0.09962, att_loss=0.2484, loss=0.2187, over 3268198.28 frames. utt_duration=1225 frames, utt_pad_proportion=0.06265, over 10680.57 utterances.], batch size: 40, lr: 8.97e-03, grad_scale: 8.0 2023-03-08 07:08:23,020 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 07:08:31,380 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3512, 5.5324, 5.3814, 5.4276, 5.6131, 5.4267, 4.7366, 5.3279], device='cuda:3'), covar=tensor([0.0075, 0.0065, 0.0095, 0.0064, 0.0063, 0.0082, 0.0428, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0073, 0.0091, 0.0055, 0.0062, 0.0073, 0.0095, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 07:08:37,752 INFO [train2.py:843] (3/4) Epoch 12, validation: ctc_loss=0.04782, att_loss=0.2378, loss=0.1998, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 07:08:37,753 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 07:08:39,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.287e+02 2.757e+02 3.651e+02 8.301e+02, threshold=5.514e+02, percent-clipped=2.0 2023-03-08 07:08:41,276 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2768, 3.4821, 3.3600, 2.9602, 3.5282, 3.5213, 3.4525, 2.4565], device='cuda:3'), covar=tensor([0.1607, 0.1982, 0.2784, 0.7454, 0.2707, 0.2980, 0.1345, 0.7402], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0127, 0.0140, 0.0211, 0.0108, 0.0194, 0.0116, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:09:56,620 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:09:57,798 INFO [train2.py:809] (3/4) Epoch 12, batch 3050, loss[ctc_loss=0.0901, att_loss=0.2509, loss=0.2187, over 16876.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007008, over 49.00 utterances.], tot_loss[ctc_loss=0.09874, att_loss=0.2478, loss=0.218, over 3261698.44 frames. utt_duration=1259 frames, utt_pad_proportion=0.05497, over 10379.14 utterances.], batch size: 49, lr: 8.97e-03, grad_scale: 8.0 2023-03-08 07:10:28,702 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-08 07:11:17,290 INFO [train2.py:809] (3/4) Epoch 12, batch 3100, loss[ctc_loss=0.0991, att_loss=0.2554, loss=0.2242, over 17201.00 frames. utt_duration=872.4 frames, utt_pad_proportion=0.0865, over 79.00 utterances.], tot_loss[ctc_loss=0.09875, att_loss=0.248, loss=0.2181, over 3263528.01 frames. utt_duration=1252 frames, utt_pad_proportion=0.05589, over 10437.80 utterances.], batch size: 79, lr: 8.96e-03, grad_scale: 8.0 2023-03-08 07:11:17,530 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3773, 4.5003, 4.5759, 4.5924, 4.6408, 4.6354, 4.3015, 4.1908], device='cuda:3'), covar=tensor([0.0894, 0.0617, 0.0325, 0.0413, 0.0310, 0.0308, 0.0348, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0300, 0.0261, 0.0292, 0.0351, 0.0367, 0.0300, 0.0333], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:11:18,788 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.293e+02 2.705e+02 3.311e+02 9.468e+02, threshold=5.410e+02, percent-clipped=5.0 2023-03-08 07:11:33,002 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:12:02,255 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 07:12:13,044 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:12:36,182 INFO [train2.py:809] (3/4) Epoch 12, batch 3150, loss[ctc_loss=0.09469, att_loss=0.2574, loss=0.2249, over 17057.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.008737, over 53.00 utterances.], tot_loss[ctc_loss=0.0991, att_loss=0.248, loss=0.2182, over 3259091.66 frames. utt_duration=1239 frames, utt_pad_proportion=0.05884, over 10534.19 utterances.], batch size: 53, lr: 8.96e-03, grad_scale: 8.0 2023-03-08 07:12:48,217 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:13:05,336 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.22 vs. limit=5.0 2023-03-08 07:13:55,729 INFO [train2.py:809] (3/4) Epoch 12, batch 3200, loss[ctc_loss=0.09748, att_loss=0.2286, loss=0.2024, over 16029.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006517, over 40.00 utterances.], tot_loss[ctc_loss=0.09879, att_loss=0.2478, loss=0.218, over 3254981.78 frames. utt_duration=1230 frames, utt_pad_proportion=0.06355, over 10600.06 utterances.], batch size: 40, lr: 8.95e-03, grad_scale: 8.0 2023-03-08 07:13:57,256 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.280e+02 2.660e+02 3.739e+02 9.855e+02, threshold=5.321e+02, percent-clipped=6.0 2023-03-08 07:14:05,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9777, 5.0996, 4.9880, 2.2404, 1.9245, 2.7854, 2.7139, 3.8776], device='cuda:3'), covar=tensor([0.0739, 0.0197, 0.0196, 0.4737, 0.5939, 0.2509, 0.2669, 0.1632], device='cuda:3'), in_proj_covar=tensor([0.0330, 0.0221, 0.0237, 0.0213, 0.0342, 0.0330, 0.0229, 0.0352], device='cuda:3'), out_proj_covar=tensor([1.4718e-04, 8.2536e-05, 1.0284e-04, 9.3832e-05, 1.4834e-04, 1.3295e-04, 9.0327e-05, 1.4766e-04], device='cuda:3') 2023-03-08 07:14:34,606 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5629, 2.8101, 3.4293, 4.4742, 3.9628, 4.1191, 2.9557, 2.1578], device='cuda:3'), covar=tensor([0.0648, 0.2190, 0.0967, 0.0521, 0.0681, 0.0349, 0.1582, 0.2503], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0214, 0.0192, 0.0197, 0.0194, 0.0158, 0.0200, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 07:15:00,381 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 07:15:16,019 INFO [train2.py:809] (3/4) Epoch 12, batch 3250, loss[ctc_loss=0.08403, att_loss=0.2316, loss=0.2021, over 16175.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007343, over 41.00 utterances.], tot_loss[ctc_loss=0.0981, att_loss=0.2482, loss=0.2182, over 3265400.02 frames. utt_duration=1245 frames, utt_pad_proportion=0.05774, over 10503.92 utterances.], batch size: 41, lr: 8.95e-03, grad_scale: 8.0 2023-03-08 07:16:34,791 INFO [train2.py:809] (3/4) Epoch 12, batch 3300, loss[ctc_loss=0.1148, att_loss=0.2608, loss=0.2316, over 17110.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01543, over 56.00 utterances.], tot_loss[ctc_loss=0.09718, att_loss=0.2474, loss=0.2174, over 3264743.54 frames. utt_duration=1269 frames, utt_pad_proportion=0.05251, over 10301.98 utterances.], batch size: 56, lr: 8.94e-03, grad_scale: 8.0 2023-03-08 07:16:36,329 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.270e+02 2.752e+02 3.318e+02 7.003e+02, threshold=5.505e+02, percent-clipped=5.0 2023-03-08 07:17:43,751 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:17:53,906 INFO [train2.py:809] (3/4) Epoch 12, batch 3350, loss[ctc_loss=0.09484, att_loss=0.2465, loss=0.2162, over 16472.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006297, over 46.00 utterances.], tot_loss[ctc_loss=0.09675, att_loss=0.2469, loss=0.2169, over 3269216.99 frames. utt_duration=1290 frames, utt_pad_proportion=0.04601, over 10151.49 utterances.], batch size: 46, lr: 8.94e-03, grad_scale: 8.0 2023-03-08 07:19:14,296 INFO [train2.py:809] (3/4) Epoch 12, batch 3400, loss[ctc_loss=0.07559, att_loss=0.2358, loss=0.2037, over 15963.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.006498, over 41.00 utterances.], tot_loss[ctc_loss=0.09702, att_loss=0.2475, loss=0.2174, over 3278127.11 frames. utt_duration=1303 frames, utt_pad_proportion=0.04042, over 10077.48 utterances.], batch size: 41, lr: 8.93e-03, grad_scale: 8.0 2023-03-08 07:19:15,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.335e+02 2.724e+02 3.467e+02 7.189e+02, threshold=5.448e+02, percent-clipped=4.0 2023-03-08 07:19:44,397 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:19:59,827 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:20:10,651 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:20:33,848 INFO [train2.py:809] (3/4) Epoch 12, batch 3450, loss[ctc_loss=0.07533, att_loss=0.2246, loss=0.1947, over 15780.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008181, over 38.00 utterances.], tot_loss[ctc_loss=0.09713, att_loss=0.2475, loss=0.2175, over 3270735.60 frames. utt_duration=1281 frames, utt_pad_proportion=0.0465, over 10228.47 utterances.], batch size: 38, lr: 8.93e-03, grad_scale: 8.0 2023-03-08 07:20:37,190 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0401, 4.5050, 4.1009, 4.7489, 2.3538, 4.4919, 2.1256, 1.8527], device='cuda:3'), covar=tensor([0.0361, 0.0151, 0.0867, 0.0135, 0.2183, 0.0167, 0.2161, 0.2009], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0118, 0.0258, 0.0117, 0.0220, 0.0110, 0.0227, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 07:21:15,940 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:21:20,804 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:21:26,663 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:21:38,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7840, 6.0503, 5.3734, 5.7626, 5.6881, 5.2839, 5.4625, 5.2208], device='cuda:3'), covar=tensor([0.1242, 0.0844, 0.0917, 0.0772, 0.0820, 0.1520, 0.2228, 0.2404], device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0518, 0.0390, 0.0389, 0.0377, 0.0431, 0.0536, 0.0464], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 07:21:53,257 INFO [train2.py:809] (3/4) Epoch 12, batch 3500, loss[ctc_loss=0.07835, att_loss=0.2328, loss=0.2019, over 16176.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.007159, over 41.00 utterances.], tot_loss[ctc_loss=0.09743, att_loss=0.2478, loss=0.2177, over 3269075.83 frames. utt_duration=1254 frames, utt_pad_proportion=0.05366, over 10436.89 utterances.], batch size: 41, lr: 8.92e-03, grad_scale: 8.0 2023-03-08 07:21:54,808 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.392e+02 2.957e+02 3.491e+02 6.570e+02, threshold=5.913e+02, percent-clipped=5.0 2023-03-08 07:22:26,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1833, 5.1365, 5.0305, 2.9201, 4.8691, 4.6901, 4.3298, 2.6320], device='cuda:3'), covar=tensor([0.0091, 0.0073, 0.0212, 0.1002, 0.0089, 0.0184, 0.0303, 0.1462], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0088, 0.0082, 0.0107, 0.0074, 0.0097, 0.0095, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 07:22:44,052 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3124, 3.8665, 3.2789, 3.6800, 4.1565, 3.6333, 3.0356, 4.5163], device='cuda:3'), covar=tensor([0.0892, 0.0477, 0.0978, 0.0596, 0.0632, 0.0722, 0.0883, 0.0426], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0189, 0.0207, 0.0178, 0.0244, 0.0216, 0.0184, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:23:13,804 INFO [train2.py:809] (3/4) Epoch 12, batch 3550, loss[ctc_loss=0.09641, att_loss=0.2479, loss=0.2176, over 16480.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006413, over 46.00 utterances.], tot_loss[ctc_loss=0.09707, att_loss=0.2471, loss=0.2171, over 3262042.30 frames. utt_duration=1260 frames, utt_pad_proportion=0.05529, over 10370.24 utterances.], batch size: 46, lr: 8.92e-03, grad_scale: 8.0 2023-03-08 07:23:25,974 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:24:24,433 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8685, 5.1357, 5.1132, 5.0383, 5.2082, 5.0884, 4.8838, 4.5273], device='cuda:3'), covar=tensor([0.0971, 0.0543, 0.0261, 0.0470, 0.0268, 0.0313, 0.0332, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0301, 0.0261, 0.0294, 0.0353, 0.0371, 0.0303, 0.0334], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:24:34,830 INFO [train2.py:809] (3/4) Epoch 12, batch 3600, loss[ctc_loss=0.1299, att_loss=0.2624, loss=0.2359, over 17073.00 frames. utt_duration=698.4 frames, utt_pad_proportion=0.1248, over 98.00 utterances.], tot_loss[ctc_loss=0.09674, att_loss=0.2469, loss=0.2169, over 3269267.45 frames. utt_duration=1267 frames, utt_pad_proportion=0.05157, over 10333.72 utterances.], batch size: 98, lr: 8.92e-03, grad_scale: 8.0 2023-03-08 07:24:36,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.371e+02 2.874e+02 3.552e+02 5.141e+02, threshold=5.747e+02, percent-clipped=0.0 2023-03-08 07:24:40,312 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8539, 6.1145, 5.5969, 5.9157, 5.8191, 5.3171, 5.4981, 5.3791], device='cuda:3'), covar=tensor([0.1174, 0.0824, 0.0735, 0.0745, 0.0860, 0.1548, 0.2003, 0.2315], device='cuda:3'), in_proj_covar=tensor([0.0448, 0.0519, 0.0389, 0.0385, 0.0376, 0.0429, 0.0535, 0.0467], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 07:25:04,436 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:25:25,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-08 07:25:47,363 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:25:51,008 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 07:25:56,420 INFO [train2.py:809] (3/4) Epoch 12, batch 3650, loss[ctc_loss=0.1102, att_loss=0.2501, loss=0.2221, over 16252.00 frames. utt_duration=1513 frames, utt_pad_proportion=0.007701, over 43.00 utterances.], tot_loss[ctc_loss=0.09701, att_loss=0.2473, loss=0.2173, over 3267725.87 frames. utt_duration=1254 frames, utt_pad_proportion=0.05466, over 10433.05 utterances.], batch size: 43, lr: 8.91e-03, grad_scale: 8.0 2023-03-08 07:27:03,210 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:27:10,059 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5684, 2.4543, 5.0218, 3.8477, 2.9727, 4.3263, 4.7186, 4.5502], device='cuda:3'), covar=tensor([0.0209, 0.1697, 0.0109, 0.1024, 0.1790, 0.0216, 0.0107, 0.0249], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0242, 0.0138, 0.0304, 0.0266, 0.0181, 0.0124, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:27:16,060 INFO [train2.py:809] (3/4) Epoch 12, batch 3700, loss[ctc_loss=0.107, att_loss=0.2612, loss=0.2304, over 16645.00 frames. utt_duration=673.9 frames, utt_pad_proportion=0.1513, over 99.00 utterances.], tot_loss[ctc_loss=0.09844, att_loss=0.2487, loss=0.2186, over 3275591.23 frames. utt_duration=1254 frames, utt_pad_proportion=0.05183, over 10457.92 utterances.], batch size: 99, lr: 8.91e-03, grad_scale: 8.0 2023-03-08 07:27:17,595 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.382e+02 2.831e+02 3.423e+02 6.375e+02, threshold=5.662e+02, percent-clipped=1.0 2023-03-08 07:28:38,161 INFO [train2.py:809] (3/4) Epoch 12, batch 3750, loss[ctc_loss=0.1216, att_loss=0.2679, loss=0.2386, over 17146.00 frames. utt_duration=869.7 frames, utt_pad_proportion=0.08641, over 79.00 utterances.], tot_loss[ctc_loss=0.09932, att_loss=0.2492, loss=0.2192, over 3276738.81 frames. utt_duration=1233 frames, utt_pad_proportion=0.05813, over 10644.68 utterances.], batch size: 79, lr: 8.90e-03, grad_scale: 8.0 2023-03-08 07:28:58,822 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9322, 3.5708, 2.9503, 3.3549, 3.7587, 3.4396, 2.7828, 4.0891], device='cuda:3'), covar=tensor([0.1029, 0.0488, 0.1105, 0.0701, 0.0749, 0.0680, 0.0867, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0187, 0.0205, 0.0177, 0.0241, 0.0213, 0.0181, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:29:17,206 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:29:58,052 INFO [train2.py:809] (3/4) Epoch 12, batch 3800, loss[ctc_loss=0.0892, att_loss=0.2486, loss=0.2167, over 16492.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005947, over 46.00 utterances.], tot_loss[ctc_loss=0.09916, att_loss=0.2494, loss=0.2193, over 3280832.77 frames. utt_duration=1228 frames, utt_pad_proportion=0.05856, over 10697.10 utterances.], batch size: 46, lr: 8.90e-03, grad_scale: 8.0 2023-03-08 07:29:59,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.386e+02 2.816e+02 3.593e+02 7.143e+02, threshold=5.632e+02, percent-clipped=4.0 2023-03-08 07:31:17,521 INFO [train2.py:809] (3/4) Epoch 12, batch 3850, loss[ctc_loss=0.0843, att_loss=0.2282, loss=0.1994, over 12363.00 frames. utt_duration=1833 frames, utt_pad_proportion=0.149, over 27.00 utterances.], tot_loss[ctc_loss=0.09939, att_loss=0.2497, loss=0.2197, over 3279056.96 frames. utt_duration=1210 frames, utt_pad_proportion=0.06181, over 10856.16 utterances.], batch size: 27, lr: 8.89e-03, grad_scale: 16.0 2023-03-08 07:32:34,727 INFO [train2.py:809] (3/4) Epoch 12, batch 3900, loss[ctc_loss=0.1018, att_loss=0.2577, loss=0.2265, over 16877.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00695, over 49.00 utterances.], tot_loss[ctc_loss=0.09861, att_loss=0.2486, loss=0.2186, over 3267593.57 frames. utt_duration=1239 frames, utt_pad_proportion=0.05702, over 10562.52 utterances.], batch size: 49, lr: 8.89e-03, grad_scale: 16.0 2023-03-08 07:32:36,203 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.278e+02 2.685e+02 3.452e+02 6.474e+02, threshold=5.369e+02, percent-clipped=2.0 2023-03-08 07:32:55,452 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:33:00,347 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:33:03,393 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9588, 5.2315, 5.1265, 5.1076, 5.2950, 5.2476, 4.9591, 4.6863], device='cuda:3'), covar=tensor([0.1089, 0.0553, 0.0268, 0.0526, 0.0263, 0.0279, 0.0366, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0302, 0.0263, 0.0294, 0.0353, 0.0371, 0.0301, 0.0335], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 07:33:52,999 INFO [train2.py:809] (3/4) Epoch 12, batch 3950, loss[ctc_loss=0.1108, att_loss=0.2646, loss=0.2338, over 17361.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02033, over 59.00 utterances.], tot_loss[ctc_loss=0.0981, att_loss=0.2485, loss=0.2184, over 3260390.26 frames. utt_duration=1211 frames, utt_pad_proportion=0.06526, over 10780.49 utterances.], batch size: 59, lr: 8.88e-03, grad_scale: 8.0 2023-03-08 07:34:17,745 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:34:35,126 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:35:12,421 INFO [train2.py:809] (3/4) Epoch 13, batch 0, loss[ctc_loss=0.1041, att_loss=0.2528, loss=0.2231, over 16683.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006632, over 46.00 utterances.], tot_loss[ctc_loss=0.1041, att_loss=0.2528, loss=0.2231, over 16683.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006632, over 46.00 utterances.], batch size: 46, lr: 8.53e-03, grad_scale: 8.0 2023-03-08 07:35:12,421 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 07:35:24,621 INFO [train2.py:843] (3/4) Epoch 13, validation: ctc_loss=0.04799, att_loss=0.2379, loss=0.2, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 07:35:24,622 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 07:35:29,641 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 07:35:54,071 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 2.314e+02 2.717e+02 3.534e+02 1.025e+03, threshold=5.435e+02, percent-clipped=1.0 2023-03-08 07:36:00,508 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-08 07:36:34,373 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:36:45,285 INFO [train2.py:809] (3/4) Epoch 13, batch 50, loss[ctc_loss=0.1165, att_loss=0.2613, loss=0.2324, over 16887.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007047, over 49.00 utterances.], tot_loss[ctc_loss=0.1014, att_loss=0.2515, loss=0.2215, over 743155.97 frames. utt_duration=1212 frames, utt_pad_proportion=0.06036, over 2455.16 utterances.], batch size: 49, lr: 8.53e-03, grad_scale: 8.0 2023-03-08 07:37:08,561 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 07:37:51,325 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:38:05,791 INFO [train2.py:809] (3/4) Epoch 13, batch 100, loss[ctc_loss=0.1046, att_loss=0.2594, loss=0.2284, over 17007.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009353, over 51.00 utterances.], tot_loss[ctc_loss=0.09978, att_loss=0.2501, loss=0.2201, over 1303377.67 frames. utt_duration=1251 frames, utt_pad_proportion=0.05502, over 4173.89 utterances.], batch size: 51, lr: 8.52e-03, grad_scale: 8.0 2023-03-08 07:38:35,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.449e+02 2.869e+02 3.545e+02 8.286e+02, threshold=5.739e+02, percent-clipped=3.0 2023-03-08 07:39:08,790 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:39:12,750 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-03-08 07:39:26,427 INFO [train2.py:809] (3/4) Epoch 13, batch 150, loss[ctc_loss=0.07195, att_loss=0.2139, loss=0.1855, over 15669.00 frames. utt_duration=1696 frames, utt_pad_proportion=0.006682, over 37.00 utterances.], tot_loss[ctc_loss=0.09875, att_loss=0.2484, loss=0.2185, over 1732592.36 frames. utt_duration=1223 frames, utt_pad_proportion=0.06375, over 5672.34 utterances.], batch size: 37, lr: 8.52e-03, grad_scale: 8.0 2023-03-08 07:40:24,987 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5618, 5.1761, 4.8812, 5.0236, 5.2055, 4.7751, 3.7601, 4.9762], device='cuda:3'), covar=tensor([0.0118, 0.0085, 0.0116, 0.0073, 0.0068, 0.0103, 0.0625, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0074, 0.0092, 0.0057, 0.0062, 0.0074, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 07:40:51,747 INFO [train2.py:809] (3/4) Epoch 13, batch 200, loss[ctc_loss=0.1171, att_loss=0.2736, loss=0.2423, over 17116.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.0151, over 56.00 utterances.], tot_loss[ctc_loss=0.09691, att_loss=0.2486, loss=0.2183, over 2082577.14 frames. utt_duration=1226 frames, utt_pad_proportion=0.05937, over 6805.49 utterances.], batch size: 56, lr: 8.52e-03, grad_scale: 8.0 2023-03-08 07:41:20,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.209e+02 2.594e+02 3.149e+02 6.379e+02, threshold=5.187e+02, percent-clipped=1.0 2023-03-08 07:41:35,524 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9851, 6.1417, 5.5338, 5.9144, 5.8213, 5.3212, 5.6072, 5.3942], device='cuda:3'), covar=tensor([0.1155, 0.0882, 0.0883, 0.0644, 0.0745, 0.1419, 0.2095, 0.2111], device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0521, 0.0388, 0.0384, 0.0371, 0.0424, 0.0530, 0.0466], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 07:41:38,779 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:42:12,097 INFO [train2.py:809] (3/4) Epoch 13, batch 250, loss[ctc_loss=0.08743, att_loss=0.2374, loss=0.2074, over 16174.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007251, over 41.00 utterances.], tot_loss[ctc_loss=0.09713, att_loss=0.2487, loss=0.2184, over 2353106.47 frames. utt_duration=1210 frames, utt_pad_proportion=0.06049, over 7786.14 utterances.], batch size: 41, lr: 8.51e-03, grad_scale: 8.0 2023-03-08 07:42:17,028 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5952, 4.8891, 5.3298, 4.8676, 4.6638, 5.4001, 4.9321, 5.4547], device='cuda:3'), covar=tensor([0.1451, 0.1448, 0.1090, 0.2208, 0.4084, 0.1800, 0.1172, 0.1273], device='cuda:3'), in_proj_covar=tensor([0.0745, 0.0438, 0.0513, 0.0577, 0.0756, 0.0522, 0.0417, 0.0515], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 07:42:35,943 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-03-08 07:42:38,650 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:42:55,999 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:43:13,010 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:43:31,448 INFO [train2.py:809] (3/4) Epoch 13, batch 300, loss[ctc_loss=0.09002, att_loss=0.2514, loss=0.2192, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.006579, over 50.00 utterances.], tot_loss[ctc_loss=0.09721, att_loss=0.2492, loss=0.2188, over 2565337.15 frames. utt_duration=1212 frames, utt_pad_proportion=0.0586, over 8477.24 utterances.], batch size: 50, lr: 8.51e-03, grad_scale: 8.0 2023-03-08 07:43:32,478 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8837, 4.3159, 4.3034, 4.6037, 2.2721, 4.4573, 2.4924, 1.6247], device='cuda:3'), covar=tensor([0.0431, 0.0125, 0.0732, 0.0117, 0.2096, 0.0134, 0.1747, 0.1968], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0122, 0.0260, 0.0115, 0.0223, 0.0111, 0.0229, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 07:43:59,888 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3940, 3.5943, 3.5037, 3.0331, 3.6080, 3.5760, 3.3779, 2.4889], device='cuda:3'), covar=tensor([0.1240, 0.1391, 0.2819, 0.5282, 0.1562, 0.3958, 0.1290, 0.6853], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0128, 0.0137, 0.0209, 0.0110, 0.0194, 0.0117, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:44:01,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.171e+02 2.716e+02 3.489e+02 9.070e+02, threshold=5.431e+02, percent-clipped=6.0 2023-03-08 07:44:16,024 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:44:16,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.42 vs. limit=5.0 2023-03-08 07:44:32,578 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:44:44,061 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-03-08 07:44:51,566 INFO [train2.py:809] (3/4) Epoch 13, batch 350, loss[ctc_loss=0.08535, att_loss=0.2469, loss=0.2146, over 16616.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005054, over 47.00 utterances.], tot_loss[ctc_loss=0.09597, att_loss=0.2476, loss=0.2173, over 2727017.38 frames. utt_duration=1255 frames, utt_pad_proportion=0.04834, over 8703.04 utterances.], batch size: 47, lr: 8.50e-03, grad_scale: 8.0 2023-03-08 07:45:06,311 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 07:46:10,722 INFO [train2.py:809] (3/4) Epoch 13, batch 400, loss[ctc_loss=0.09184, att_loss=0.2542, loss=0.2217, over 17399.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.04688, over 69.00 utterances.], tot_loss[ctc_loss=0.09593, att_loss=0.2479, loss=0.2175, over 2852690.64 frames. utt_duration=1269 frames, utt_pad_proportion=0.04498, over 9002.60 utterances.], batch size: 69, lr: 8.50e-03, grad_scale: 8.0 2023-03-08 07:46:21,644 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 07:46:40,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.275e+02 2.732e+02 3.494e+02 6.607e+02, threshold=5.464e+02, percent-clipped=2.0 2023-03-08 07:47:29,598 INFO [train2.py:809] (3/4) Epoch 13, batch 450, loss[ctc_loss=0.09755, att_loss=0.2538, loss=0.2225, over 16879.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006484, over 49.00 utterances.], tot_loss[ctc_loss=0.09618, att_loss=0.2476, loss=0.2173, over 2948044.16 frames. utt_duration=1266 frames, utt_pad_proportion=0.04723, over 9327.50 utterances.], batch size: 49, lr: 8.49e-03, grad_scale: 8.0 2023-03-08 07:47:56,931 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 07:48:19,595 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2023-03-08 07:48:48,219 INFO [train2.py:809] (3/4) Epoch 13, batch 500, loss[ctc_loss=0.1182, att_loss=0.2525, loss=0.2257, over 16761.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006278, over 48.00 utterances.], tot_loss[ctc_loss=0.09528, att_loss=0.2463, loss=0.2161, over 3011601.13 frames. utt_duration=1280 frames, utt_pad_proportion=0.04636, over 9421.94 utterances.], batch size: 48, lr: 8.49e-03, grad_scale: 8.0 2023-03-08 07:48:57,478 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2517, 4.6382, 4.6732, 4.8554, 2.5320, 4.7962, 2.6534, 1.4972], device='cuda:3'), covar=tensor([0.0273, 0.0173, 0.0673, 0.0111, 0.2029, 0.0115, 0.1678, 0.2058], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0121, 0.0256, 0.0115, 0.0219, 0.0109, 0.0224, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 07:49:01,815 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9861, 6.2016, 5.6619, 5.9579, 5.8710, 5.5002, 5.6895, 5.5236], device='cuda:3'), covar=tensor([0.1259, 0.0851, 0.0866, 0.0666, 0.0704, 0.1536, 0.2428, 0.2308], device='cuda:3'), in_proj_covar=tensor([0.0446, 0.0521, 0.0387, 0.0382, 0.0368, 0.0428, 0.0532, 0.0464], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 07:49:16,719 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.184e+02 2.674e+02 3.467e+02 7.635e+02, threshold=5.348e+02, percent-clipped=2.0 2023-03-08 07:49:17,122 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:49:27,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 07:50:07,390 INFO [train2.py:809] (3/4) Epoch 13, batch 550, loss[ctc_loss=0.1019, att_loss=0.2533, loss=0.2231, over 17052.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009615, over 53.00 utterances.], tot_loss[ctc_loss=0.09517, att_loss=0.2461, loss=0.2159, over 3067263.98 frames. utt_duration=1293 frames, utt_pad_proportion=0.04202, over 9496.93 utterances.], batch size: 53, lr: 8.49e-03, grad_scale: 8.0 2023-03-08 07:50:53,306 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5431, 4.5308, 4.5197, 4.5980, 5.0635, 4.7027, 4.6260, 2.4622], device='cuda:3'), covar=tensor([0.0208, 0.0246, 0.0224, 0.0177, 0.0719, 0.0171, 0.0215, 0.1956], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0137, 0.0140, 0.0152, 0.0336, 0.0122, 0.0125, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:50:54,958 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:51:08,883 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:51:13,504 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:51:27,705 INFO [train2.py:809] (3/4) Epoch 13, batch 600, loss[ctc_loss=0.1425, att_loss=0.2718, loss=0.2459, over 13671.00 frames. utt_duration=376 frames, utt_pad_proportion=0.3438, over 146.00 utterances.], tot_loss[ctc_loss=0.09521, att_loss=0.2464, loss=0.2162, over 3114514.86 frames. utt_duration=1284 frames, utt_pad_proportion=0.04496, over 9711.61 utterances.], batch size: 146, lr: 8.48e-03, grad_scale: 8.0 2023-03-08 07:51:38,884 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8804, 3.7597, 3.7302, 3.2683, 3.7274, 3.6837, 3.4362, 2.7518], device='cuda:3'), covar=tensor([0.1112, 0.1519, 0.2440, 0.4638, 0.1178, 0.1992, 0.1293, 0.6357], device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0131, 0.0138, 0.0212, 0.0111, 0.0196, 0.0118, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:51:56,286 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:51:57,442 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.243e+02 2.721e+02 3.200e+02 7.268e+02, threshold=5.442e+02, percent-clipped=2.0 2023-03-08 07:52:04,401 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:52:25,659 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:52:28,964 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:52:47,830 INFO [train2.py:809] (3/4) Epoch 13, batch 650, loss[ctc_loss=0.06942, att_loss=0.2078, loss=0.1801, over 15376.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.009572, over 35.00 utterances.], tot_loss[ctc_loss=0.09521, att_loss=0.2458, loss=0.2157, over 3151974.02 frames. utt_duration=1279 frames, utt_pad_proportion=0.04528, over 9872.43 utterances.], batch size: 35, lr: 8.48e-03, grad_scale: 8.0 2023-03-08 07:52:49,268 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-03-08 07:52:51,836 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:53:03,151 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 07:53:33,516 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:53:45,420 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:54:00,199 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6268, 2.3721, 2.2277, 2.3576, 3.1114, 2.4259, 1.9787, 3.0111], device='cuda:3'), covar=tensor([0.2067, 0.4015, 0.3617, 0.1703, 0.1187, 0.1442, 0.3445, 0.0969], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0088, 0.0091, 0.0074, 0.0079, 0.0070, 0.0091, 0.0062], device='cuda:3'), out_proj_covar=tensor([5.4678e-05, 6.2551e-05, 6.4740e-05, 5.3175e-05, 5.3992e-05, 5.2619e-05, 6.3686e-05, 4.7285e-05], device='cuda:3') 2023-03-08 07:54:07,436 INFO [train2.py:809] (3/4) Epoch 13, batch 700, loss[ctc_loss=0.1013, att_loss=0.2552, loss=0.2244, over 16463.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007467, over 46.00 utterances.], tot_loss[ctc_loss=0.0938, att_loss=0.2451, loss=0.2149, over 3175967.50 frames. utt_duration=1301 frames, utt_pad_proportion=0.04047, over 9774.73 utterances.], batch size: 46, lr: 8.47e-03, grad_scale: 8.0 2023-03-08 07:54:19,503 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 07:54:36,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.306e+02 2.831e+02 3.598e+02 7.957e+02, threshold=5.663e+02, percent-clipped=6.0 2023-03-08 07:55:26,624 INFO [train2.py:809] (3/4) Epoch 13, batch 750, loss[ctc_loss=0.135, att_loss=0.2778, loss=0.2493, over 17479.00 frames. utt_duration=1015 frames, utt_pad_proportion=0.04352, over 69.00 utterances.], tot_loss[ctc_loss=0.0947, att_loss=0.2468, loss=0.2164, over 3211117.28 frames. utt_duration=1282 frames, utt_pad_proportion=0.04102, over 10030.01 utterances.], batch size: 69, lr: 8.47e-03, grad_scale: 8.0 2023-03-08 07:55:29,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.70 vs. limit=5.0 2023-03-08 07:55:44,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 07:55:46,633 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 07:56:47,921 INFO [train2.py:809] (3/4) Epoch 13, batch 800, loss[ctc_loss=0.1177, att_loss=0.279, loss=0.2468, over 17058.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008665, over 52.00 utterances.], tot_loss[ctc_loss=0.09556, att_loss=0.2467, loss=0.2165, over 3214795.48 frames. utt_duration=1248 frames, utt_pad_proportion=0.05333, over 10315.13 utterances.], batch size: 52, lr: 8.46e-03, grad_scale: 8.0 2023-03-08 07:57:16,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.319e+02 2.822e+02 3.507e+02 7.468e+02, threshold=5.644e+02, percent-clipped=2.0 2023-03-08 07:58:07,624 INFO [train2.py:809] (3/4) Epoch 13, batch 850, loss[ctc_loss=0.06585, att_loss=0.2162, loss=0.1862, over 15768.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008652, over 38.00 utterances.], tot_loss[ctc_loss=0.09473, att_loss=0.2461, loss=0.2159, over 3227813.84 frames. utt_duration=1235 frames, utt_pad_proportion=0.05622, over 10463.78 utterances.], batch size: 38, lr: 8.46e-03, grad_scale: 8.0 2023-03-08 07:58:11,887 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7611, 4.5664, 4.6282, 4.6622, 5.1525, 4.7730, 4.7284, 2.4085], device='cuda:3'), covar=tensor([0.0172, 0.0310, 0.0250, 0.0225, 0.0970, 0.0161, 0.0234, 0.1943], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0138, 0.0142, 0.0153, 0.0339, 0.0123, 0.0126, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 07:58:46,041 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 07:58:54,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 07:59:11,413 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7574, 3.0198, 3.7219, 4.6170, 4.1270, 4.1790, 3.0725, 2.4942], device='cuda:3'), covar=tensor([0.0524, 0.1854, 0.0831, 0.0481, 0.0740, 0.0365, 0.1493, 0.2106], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0206, 0.0184, 0.0193, 0.0192, 0.0157, 0.0197, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 07:59:28,738 INFO [train2.py:809] (3/4) Epoch 13, batch 900, loss[ctc_loss=0.08877, att_loss=0.2557, loss=0.2223, over 16481.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005665, over 46.00 utterances.], tot_loss[ctc_loss=0.09421, att_loss=0.2458, loss=0.2155, over 3241386.37 frames. utt_duration=1244 frames, utt_pad_proportion=0.05315, over 10432.13 utterances.], batch size: 46, lr: 8.45e-03, grad_scale: 8.0 2023-03-08 07:59:48,793 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-08 07:59:56,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.412e+02 2.153e+02 2.503e+02 3.048e+02 6.271e+02, threshold=5.007e+02, percent-clipped=1.0 2023-03-08 08:00:04,005 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:00:44,221 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:00:49,767 INFO [train2.py:809] (3/4) Epoch 13, batch 950, loss[ctc_loss=0.08126, att_loss=0.2522, loss=0.218, over 16966.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007436, over 50.00 utterances.], tot_loss[ctc_loss=0.09469, att_loss=0.246, loss=0.2157, over 3251412.23 frames. utt_duration=1232 frames, utt_pad_proportion=0.05569, over 10569.52 utterances.], batch size: 50, lr: 8.45e-03, grad_scale: 8.0 2023-03-08 08:01:21,803 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:01:26,530 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:02:10,410 INFO [train2.py:809] (3/4) Epoch 13, batch 1000, loss[ctc_loss=0.1804, att_loss=0.2906, loss=0.2686, over 13992.00 frames. utt_duration=387.5 frames, utt_pad_proportion=0.326, over 145.00 utterances.], tot_loss[ctc_loss=0.09481, att_loss=0.2459, loss=0.2157, over 3251644.21 frames. utt_duration=1237 frames, utt_pad_proportion=0.05622, over 10523.37 utterances.], batch size: 145, lr: 8.45e-03, grad_scale: 8.0 2023-03-08 08:02:33,389 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:02:37,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.228e+02 2.748e+02 3.245e+02 7.781e+02, threshold=5.496e+02, percent-clipped=5.0 2023-03-08 08:03:00,451 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:03:11,992 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2163, 5.5025, 5.7572, 5.6054, 5.6620, 6.1033, 5.3845, 6.2445], device='cuda:3'), covar=tensor([0.0589, 0.0554, 0.0651, 0.1030, 0.1531, 0.0884, 0.0526, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0746, 0.0438, 0.0518, 0.0579, 0.0768, 0.0526, 0.0419, 0.0514], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:03:30,181 INFO [train2.py:809] (3/4) Epoch 13, batch 1050, loss[ctc_loss=0.09575, att_loss=0.2373, loss=0.209, over 16116.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006895, over 42.00 utterances.], tot_loss[ctc_loss=0.09397, att_loss=0.245, loss=0.2148, over 3249381.37 frames. utt_duration=1261 frames, utt_pad_proportion=0.05144, over 10315.91 utterances.], batch size: 42, lr: 8.44e-03, grad_scale: 8.0 2023-03-08 08:03:49,030 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 08:04:11,830 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:04:39,686 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:04:50,180 INFO [train2.py:809] (3/4) Epoch 13, batch 1100, loss[ctc_loss=0.06681, att_loss=0.2309, loss=0.1981, over 15879.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009783, over 39.00 utterances.], tot_loss[ctc_loss=0.0957, att_loss=0.2461, loss=0.216, over 3256577.20 frames. utt_duration=1224 frames, utt_pad_proportion=0.05922, over 10652.62 utterances.], batch size: 39, lr: 8.44e-03, grad_scale: 8.0 2023-03-08 08:05:05,420 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:05:18,176 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.462e+02 3.131e+02 3.917e+02 9.784e+02, threshold=6.261e+02, percent-clipped=6.0 2023-03-08 08:06:09,392 INFO [train2.py:809] (3/4) Epoch 13, batch 1150, loss[ctc_loss=0.1105, att_loss=0.2699, loss=0.238, over 17473.00 frames. utt_duration=1111 frames, utt_pad_proportion=0.02818, over 63.00 utterances.], tot_loss[ctc_loss=0.09583, att_loss=0.2465, loss=0.2163, over 3260956.51 frames. utt_duration=1231 frames, utt_pad_proportion=0.05668, over 10613.17 utterances.], batch size: 63, lr: 8.43e-03, grad_scale: 8.0 2023-03-08 08:06:47,198 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:07:28,989 INFO [train2.py:809] (3/4) Epoch 13, batch 1200, loss[ctc_loss=0.09494, att_loss=0.2224, loss=0.1969, over 15383.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.009856, over 35.00 utterances.], tot_loss[ctc_loss=0.0963, att_loss=0.2467, loss=0.2166, over 3259729.27 frames. utt_duration=1212 frames, utt_pad_proportion=0.06185, over 10768.69 utterances.], batch size: 35, lr: 8.43e-03, grad_scale: 8.0 2023-03-08 08:07:47,951 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3895, 5.0073, 4.7474, 4.9440, 5.1255, 4.6280, 3.8886, 4.8944], device='cuda:3'), covar=tensor([0.0125, 0.0113, 0.0145, 0.0100, 0.0099, 0.0124, 0.0541, 0.0193], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0072, 0.0091, 0.0056, 0.0061, 0.0071, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:07:57,674 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.494e+02 2.872e+02 3.566e+02 1.058e+03, threshold=5.743e+02, percent-clipped=4.0 2023-03-08 08:08:04,020 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:08:44,308 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:08:48,628 INFO [train2.py:809] (3/4) Epoch 13, batch 1250, loss[ctc_loss=0.08638, att_loss=0.2316, loss=0.2025, over 16172.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.00675, over 41.00 utterances.], tot_loss[ctc_loss=0.09652, att_loss=0.2467, loss=0.2166, over 3250121.33 frames. utt_duration=1186 frames, utt_pad_proportion=0.07318, over 10970.96 utterances.], batch size: 41, lr: 8.42e-03, grad_scale: 8.0 2023-03-08 08:08:58,134 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4858, 5.0664, 4.8872, 4.9806, 5.1722, 4.6757, 3.7848, 5.0420], device='cuda:3'), covar=tensor([0.0110, 0.0106, 0.0130, 0.0088, 0.0097, 0.0112, 0.0589, 0.0181], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0072, 0.0091, 0.0056, 0.0061, 0.0072, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:09:25,506 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:10:01,133 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:10:08,582 INFO [train2.py:809] (3/4) Epoch 13, batch 1300, loss[ctc_loss=0.08105, att_loss=0.2354, loss=0.2045, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006658, over 48.00 utterances.], tot_loss[ctc_loss=0.09608, att_loss=0.2465, loss=0.2164, over 3253275.56 frames. utt_duration=1213 frames, utt_pad_proportion=0.06612, over 10740.03 utterances.], batch size: 48, lr: 8.42e-03, grad_scale: 8.0 2023-03-08 08:10:36,808 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.317e+02 2.735e+02 3.523e+02 8.419e+02, threshold=5.469e+02, percent-clipped=3.0 2023-03-08 08:10:41,571 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:10:53,148 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-08 08:11:28,000 INFO [train2.py:809] (3/4) Epoch 13, batch 1350, loss[ctc_loss=0.1265, att_loss=0.2729, loss=0.2436, over 17375.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03277, over 63.00 utterances.], tot_loss[ctc_loss=0.09608, att_loss=0.2466, loss=0.2165, over 3257086.23 frames. utt_duration=1229 frames, utt_pad_proportion=0.06156, over 10616.11 utterances.], batch size: 63, lr: 8.42e-03, grad_scale: 8.0 2023-03-08 08:12:01,199 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:12:29,056 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:12:47,675 INFO [train2.py:809] (3/4) Epoch 13, batch 1400, loss[ctc_loss=0.1203, att_loss=0.2472, loss=0.2219, over 15781.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007835, over 38.00 utterances.], tot_loss[ctc_loss=0.09557, att_loss=0.2459, loss=0.2158, over 3252050.61 frames. utt_duration=1228 frames, utt_pad_proportion=0.0622, over 10604.31 utterances.], batch size: 38, lr: 8.41e-03, grad_scale: 8.0 2023-03-08 08:13:14,896 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0283, 3.7006, 3.1495, 3.4889, 4.0366, 3.6153, 2.7986, 4.3435], device='cuda:3'), covar=tensor([0.1118, 0.0527, 0.1138, 0.0711, 0.0675, 0.0727, 0.1010, 0.0523], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0193, 0.0210, 0.0181, 0.0245, 0.0219, 0.0187, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 08:13:15,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.165e+02 2.664e+02 3.367e+02 6.809e+02, threshold=5.329e+02, percent-clipped=2.0 2023-03-08 08:13:35,706 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:14:07,334 INFO [train2.py:809] (3/4) Epoch 13, batch 1450, loss[ctc_loss=0.09344, att_loss=0.2428, loss=0.2129, over 16478.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005921, over 46.00 utterances.], tot_loss[ctc_loss=0.09615, att_loss=0.2463, loss=0.2163, over 3242134.32 frames. utt_duration=1186 frames, utt_pad_proportion=0.07568, over 10950.08 utterances.], batch size: 46, lr: 8.41e-03, grad_scale: 8.0 2023-03-08 08:15:13,302 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:15:26,866 INFO [train2.py:809] (3/4) Epoch 13, batch 1500, loss[ctc_loss=0.1037, att_loss=0.2646, loss=0.2324, over 16471.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006478, over 46.00 utterances.], tot_loss[ctc_loss=0.09524, att_loss=0.2459, loss=0.2158, over 3250938.10 frames. utt_duration=1230 frames, utt_pad_proportion=0.06373, over 10584.20 utterances.], batch size: 46, lr: 8.40e-03, grad_scale: 8.0 2023-03-08 08:15:55,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.233e+02 2.652e+02 3.112e+02 8.079e+02, threshold=5.303e+02, percent-clipped=1.0 2023-03-08 08:16:24,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-03-08 08:16:44,585 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6461, 5.8465, 5.1667, 5.6322, 5.4421, 5.0802, 5.2682, 5.1229], device='cuda:3'), covar=tensor([0.1413, 0.1080, 0.0899, 0.0887, 0.0920, 0.1549, 0.2575, 0.2452], device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0531, 0.0391, 0.0392, 0.0372, 0.0424, 0.0538, 0.0474], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 08:16:47,514 INFO [train2.py:809] (3/4) Epoch 13, batch 1550, loss[ctc_loss=0.0802, att_loss=0.2329, loss=0.2023, over 16184.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006653, over 41.00 utterances.], tot_loss[ctc_loss=0.0962, att_loss=0.2473, loss=0.2171, over 3266952.45 frames. utt_duration=1222 frames, utt_pad_proportion=0.06095, over 10704.98 utterances.], batch size: 41, lr: 8.40e-03, grad_scale: 8.0 2023-03-08 08:17:29,400 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8062, 4.7720, 4.7537, 4.5975, 5.2921, 4.7211, 4.8252, 2.5865], device='cuda:3'), covar=tensor([0.0188, 0.0242, 0.0218, 0.0311, 0.0810, 0.0219, 0.0219, 0.1959], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0138, 0.0142, 0.0155, 0.0342, 0.0125, 0.0125, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:17:50,201 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4480, 2.3573, 5.0001, 3.6337, 2.9406, 4.1549, 4.7076, 4.5284], device='cuda:3'), covar=tensor([0.0221, 0.1763, 0.0110, 0.1310, 0.1812, 0.0226, 0.0117, 0.0221], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0240, 0.0142, 0.0303, 0.0265, 0.0183, 0.0128, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:18:08,485 INFO [train2.py:809] (3/4) Epoch 13, batch 1600, loss[ctc_loss=0.1613, att_loss=0.2828, loss=0.2585, over 13587.00 frames. utt_duration=373.7 frames, utt_pad_proportion=0.3467, over 146.00 utterances.], tot_loss[ctc_loss=0.09619, att_loss=0.2472, loss=0.217, over 3260707.17 frames. utt_duration=1207 frames, utt_pad_proportion=0.06796, over 10816.41 utterances.], batch size: 146, lr: 8.40e-03, grad_scale: 8.0 2023-03-08 08:18:29,608 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0584, 5.1521, 5.0093, 2.2490, 1.9156, 2.9285, 2.8522, 3.7731], device='cuda:3'), covar=tensor([0.0646, 0.0223, 0.0203, 0.4579, 0.5877, 0.2352, 0.2467, 0.1812], device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0227, 0.0238, 0.0216, 0.0343, 0.0329, 0.0228, 0.0352], device='cuda:3'), out_proj_covar=tensor([1.4926e-04, 8.3677e-05, 1.0270e-04, 9.4966e-05, 1.4749e-04, 1.3169e-04, 8.9925e-05, 1.4681e-04], device='cuda:3') 2023-03-08 08:18:34,216 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:18:36,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.327e+02 2.714e+02 3.320e+02 9.456e+02, threshold=5.428e+02, percent-clipped=3.0 2023-03-08 08:18:50,262 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:19:28,467 INFO [train2.py:809] (3/4) Epoch 13, batch 1650, loss[ctc_loss=0.1116, att_loss=0.2654, loss=0.2347, over 17440.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04491, over 69.00 utterances.], tot_loss[ctc_loss=0.0956, att_loss=0.2469, loss=0.2167, over 3268780.38 frames. utt_duration=1222 frames, utt_pad_proportion=0.06188, over 10708.98 utterances.], batch size: 69, lr: 8.39e-03, grad_scale: 8.0 2023-03-08 08:19:36,881 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2197, 2.1494, 3.3741, 2.4882, 3.1417, 4.5492, 4.4142, 2.7462], device='cuda:3'), covar=tensor([0.0558, 0.2792, 0.1159, 0.1820, 0.1243, 0.0784, 0.0519, 0.2066], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0233, 0.0253, 0.0203, 0.0248, 0.0319, 0.0229, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:20:03,231 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:20:13,184 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:20:29,293 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:20:30,730 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:20:37,543 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-03-08 08:20:48,665 INFO [train2.py:809] (3/4) Epoch 13, batch 1700, loss[ctc_loss=0.1024, att_loss=0.2355, loss=0.2089, over 16197.00 frames. utt_duration=1582 frames, utt_pad_proportion=0.005933, over 41.00 utterances.], tot_loss[ctc_loss=0.09583, att_loss=0.2476, loss=0.2172, over 3274618.52 frames. utt_duration=1240 frames, utt_pad_proportion=0.05647, over 10576.79 utterances.], batch size: 41, lr: 8.39e-03, grad_scale: 8.0 2023-03-08 08:21:02,237 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1612, 5.1657, 4.9765, 2.3171, 2.0514, 3.2222, 3.2472, 3.9430], device='cuda:3'), covar=tensor([0.0657, 0.0286, 0.0268, 0.5336, 0.5847, 0.2138, 0.2299, 0.1773], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0235, 0.0244, 0.0223, 0.0351, 0.0339, 0.0235, 0.0362], device='cuda:3'), out_proj_covar=tensor([1.5280e-04, 8.6592e-05, 1.0544e-04, 9.8281e-05, 1.5120e-04, 1.3563e-04, 9.2815e-05, 1.5103e-04], device='cuda:3') 2023-03-08 08:21:03,577 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0581, 5.1392, 4.8796, 2.7533, 4.8905, 4.7350, 4.1922, 2.7568], device='cuda:3'), covar=tensor([0.0169, 0.0088, 0.0230, 0.1246, 0.0088, 0.0178, 0.0402, 0.1539], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0087, 0.0083, 0.0105, 0.0074, 0.0095, 0.0094, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 08:21:16,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.324e+02 2.788e+02 3.752e+02 8.319e+02, threshold=5.576e+02, percent-clipped=5.0 2023-03-08 08:21:17,513 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1219, 4.6700, 4.5443, 4.7451, 2.8577, 4.7068, 2.8396, 1.8117], device='cuda:3'), covar=tensor([0.0361, 0.0181, 0.0673, 0.0152, 0.1597, 0.0159, 0.1439, 0.1802], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0123, 0.0257, 0.0118, 0.0218, 0.0112, 0.0224, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 08:21:18,730 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:21:46,039 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:22:07,061 INFO [train2.py:809] (3/4) Epoch 13, batch 1750, loss[ctc_loss=0.08139, att_loss=0.2364, loss=0.2054, over 16298.00 frames. utt_duration=1518 frames, utt_pad_proportion=0.006122, over 43.00 utterances.], tot_loss[ctc_loss=0.09604, att_loss=0.2471, loss=0.2169, over 3267575.67 frames. utt_duration=1265 frames, utt_pad_proportion=0.0526, over 10345.74 utterances.], batch size: 43, lr: 8.38e-03, grad_scale: 8.0 2023-03-08 08:23:03,825 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:23:25,900 INFO [train2.py:809] (3/4) Epoch 13, batch 1800, loss[ctc_loss=0.1018, att_loss=0.2606, loss=0.2288, over 16887.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007106, over 49.00 utterances.], tot_loss[ctc_loss=0.0961, att_loss=0.2471, loss=0.2169, over 3273976.72 frames. utt_duration=1261 frames, utt_pad_proportion=0.05155, over 10397.13 utterances.], batch size: 49, lr: 8.38e-03, grad_scale: 8.0 2023-03-08 08:23:52,110 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 08:23:54,567 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.431e+02 2.885e+02 3.492e+02 8.603e+02, threshold=5.770e+02, percent-clipped=6.0 2023-03-08 08:24:29,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-08 08:24:40,888 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2030, 2.5793, 3.1708, 4.0071, 3.6151, 3.7012, 2.6924, 2.0597], device='cuda:3'), covar=tensor([0.0747, 0.2117, 0.0939, 0.0619, 0.0863, 0.0420, 0.1623, 0.2395], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0207, 0.0187, 0.0196, 0.0195, 0.0161, 0.0197, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 08:24:45,028 INFO [train2.py:809] (3/4) Epoch 13, batch 1850, loss[ctc_loss=0.1237, att_loss=0.2607, loss=0.2333, over 17052.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008003, over 52.00 utterances.], tot_loss[ctc_loss=0.09593, att_loss=0.2476, loss=0.2173, over 3290156.73 frames. utt_duration=1270 frames, utt_pad_proportion=0.045, over 10372.04 utterances.], batch size: 52, lr: 8.37e-03, grad_scale: 8.0 2023-03-08 08:26:04,419 INFO [train2.py:809] (3/4) Epoch 13, batch 1900, loss[ctc_loss=0.1071, att_loss=0.2387, loss=0.2124, over 10604.00 frames. utt_duration=1846 frames, utt_pad_proportion=0.2299, over 23.00 utterances.], tot_loss[ctc_loss=0.09572, att_loss=0.2472, loss=0.2169, over 3278809.62 frames. utt_duration=1277 frames, utt_pad_proportion=0.04581, over 10280.78 utterances.], batch size: 23, lr: 8.37e-03, grad_scale: 8.0 2023-03-08 08:26:33,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.157e+02 2.753e+02 3.358e+02 9.197e+02, threshold=5.506e+02, percent-clipped=6.0 2023-03-08 08:26:55,295 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9885, 6.1462, 5.6150, 5.8430, 5.7845, 5.4429, 5.6882, 5.3469], device='cuda:3'), covar=tensor([0.1042, 0.0898, 0.0784, 0.0830, 0.0873, 0.1355, 0.2184, 0.2523], device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0538, 0.0397, 0.0400, 0.0381, 0.0431, 0.0549, 0.0482], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 08:27:24,890 INFO [train2.py:809] (3/4) Epoch 13, batch 1950, loss[ctc_loss=0.05874, att_loss=0.2077, loss=0.1779, over 14547.00 frames. utt_duration=1820 frames, utt_pad_proportion=0.03956, over 32.00 utterances.], tot_loss[ctc_loss=0.09446, att_loss=0.2461, loss=0.2157, over 3276449.10 frames. utt_duration=1288 frames, utt_pad_proportion=0.04409, over 10185.74 utterances.], batch size: 32, lr: 8.37e-03, grad_scale: 16.0 2023-03-08 08:27:33,263 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4791, 4.8458, 4.9095, 5.0072, 4.9461, 4.7307, 2.9433, 4.7268], device='cuda:3'), covar=tensor([0.0115, 0.0131, 0.0122, 0.0084, 0.0124, 0.0109, 0.1003, 0.0303], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0074, 0.0092, 0.0057, 0.0063, 0.0073, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:27:41,107 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:27:59,739 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:28:17,537 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:28:45,894 INFO [train2.py:809] (3/4) Epoch 13, batch 2000, loss[ctc_loss=0.09147, att_loss=0.2183, loss=0.1929, over 15478.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.01006, over 36.00 utterances.], tot_loss[ctc_loss=0.09456, att_loss=0.2459, loss=0.2156, over 3270154.28 frames. utt_duration=1267 frames, utt_pad_proportion=0.05162, over 10338.87 utterances.], batch size: 36, lr: 8.36e-03, grad_scale: 16.0 2023-03-08 08:29:14,827 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 2.246e+02 2.552e+02 3.168e+02 7.222e+02, threshold=5.105e+02, percent-clipped=2.0 2023-03-08 08:29:19,846 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:29:40,017 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0201, 4.1214, 3.7966, 4.1859, 3.8365, 3.7021, 4.1608, 4.0906], device='cuda:3'), covar=tensor([0.0509, 0.0308, 0.0769, 0.0328, 0.0408, 0.0864, 0.0289, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0271, 0.0327, 0.0276, 0.0278, 0.0211, 0.0257, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 08:29:40,260 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7280, 2.5503, 3.9588, 3.4457, 2.8154, 3.6995, 3.6881, 3.7296], device='cuda:3'), covar=tensor([0.0234, 0.1266, 0.0119, 0.0902, 0.1442, 0.0261, 0.0132, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0239, 0.0143, 0.0304, 0.0266, 0.0183, 0.0127, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:30:05,636 INFO [train2.py:809] (3/4) Epoch 13, batch 2050, loss[ctc_loss=0.08151, att_loss=0.2238, loss=0.1953, over 15620.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.01052, over 37.00 utterances.], tot_loss[ctc_loss=0.09557, att_loss=0.2463, loss=0.2162, over 3257377.63 frames. utt_duration=1234 frames, utt_pad_proportion=0.06341, over 10570.14 utterances.], batch size: 37, lr: 8.36e-03, grad_scale: 16.0 2023-03-08 08:30:34,984 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-03-08 08:31:03,767 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:31:05,432 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2552, 5.2200, 5.0220, 2.8748, 5.0520, 4.8454, 4.4927, 2.8636], device='cuda:3'), covar=tensor([0.0095, 0.0060, 0.0231, 0.1032, 0.0066, 0.0143, 0.0268, 0.1265], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0088, 0.0085, 0.0105, 0.0074, 0.0096, 0.0093, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 08:31:25,310 INFO [train2.py:809] (3/4) Epoch 13, batch 2100, loss[ctc_loss=0.08387, att_loss=0.2263, loss=0.1978, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007094, over 40.00 utterances.], tot_loss[ctc_loss=0.09524, att_loss=0.246, loss=0.2159, over 3251252.07 frames. utt_duration=1250 frames, utt_pad_proportion=0.05855, over 10415.26 utterances.], batch size: 40, lr: 8.35e-03, grad_scale: 16.0 2023-03-08 08:31:36,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4917, 3.5880, 3.4025, 3.0565, 3.4418, 3.4826, 3.4288, 2.2610], device='cuda:3'), covar=tensor([0.1119, 0.1498, 0.4230, 0.5675, 0.1980, 0.4321, 0.1209, 0.7659], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0131, 0.0141, 0.0210, 0.0109, 0.0195, 0.0118, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:31:53,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 2.361e+02 2.979e+02 3.749e+02 1.127e+03, threshold=5.957e+02, percent-clipped=6.0 2023-03-08 08:32:19,979 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:32:39,738 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-03-08 08:32:44,497 INFO [train2.py:809] (3/4) Epoch 13, batch 2150, loss[ctc_loss=0.1011, att_loss=0.2658, loss=0.2328, over 17064.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008099, over 52.00 utterances.], tot_loss[ctc_loss=0.0953, att_loss=0.2459, loss=0.2157, over 3250980.60 frames. utt_duration=1238 frames, utt_pad_proportion=0.06222, over 10512.64 utterances.], batch size: 52, lr: 8.35e-03, grad_scale: 16.0 2023-03-08 08:33:32,790 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7910, 5.9969, 5.4448, 5.8140, 5.5907, 5.3404, 5.4059, 5.2935], device='cuda:3'), covar=tensor([0.1114, 0.0887, 0.0857, 0.0802, 0.1084, 0.1393, 0.2485, 0.2395], device='cuda:3'), in_proj_covar=tensor([0.0452, 0.0531, 0.0396, 0.0399, 0.0382, 0.0431, 0.0551, 0.0478], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 08:34:08,271 INFO [train2.py:809] (3/4) Epoch 13, batch 2200, loss[ctc_loss=0.09097, att_loss=0.2352, loss=0.2063, over 16173.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.005956, over 41.00 utterances.], tot_loss[ctc_loss=0.09567, att_loss=0.2459, loss=0.2159, over 3247450.98 frames. utt_duration=1239 frames, utt_pad_proportion=0.06164, over 10495.01 utterances.], batch size: 41, lr: 8.35e-03, grad_scale: 16.0 2023-03-08 08:34:35,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.313e+02 2.828e+02 3.809e+02 8.151e+02, threshold=5.657e+02, percent-clipped=5.0 2023-03-08 08:34:39,638 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7441, 4.0064, 3.9290, 3.9820, 4.0090, 3.7782, 3.0371, 3.9340], device='cuda:3'), covar=tensor([0.0122, 0.0112, 0.0127, 0.0085, 0.0095, 0.0127, 0.0603, 0.0219], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0073, 0.0091, 0.0057, 0.0062, 0.0073, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:35:10,779 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8321, 5.0833, 5.3657, 5.2267, 5.3162, 5.7892, 5.0743, 5.9028], device='cuda:3'), covar=tensor([0.0653, 0.0711, 0.0775, 0.1193, 0.1552, 0.0848, 0.0713, 0.0636], device='cuda:3'), in_proj_covar=tensor([0.0753, 0.0445, 0.0527, 0.0583, 0.0766, 0.0527, 0.0425, 0.0523], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 08:35:25,943 INFO [train2.py:809] (3/4) Epoch 13, batch 2250, loss[ctc_loss=0.09556, att_loss=0.2651, loss=0.2312, over 16868.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.00764, over 49.00 utterances.], tot_loss[ctc_loss=0.09597, att_loss=0.2463, loss=0.2162, over 3256056.47 frames. utt_duration=1245 frames, utt_pad_proportion=0.0592, over 10470.44 utterances.], batch size: 49, lr: 8.34e-03, grad_scale: 16.0 2023-03-08 08:35:53,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-03-08 08:36:00,949 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:36:17,737 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 08:36:45,462 INFO [train2.py:809] (3/4) Epoch 13, batch 2300, loss[ctc_loss=0.09751, att_loss=0.2526, loss=0.2216, over 17301.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01201, over 55.00 utterances.], tot_loss[ctc_loss=0.09498, att_loss=0.2461, loss=0.2159, over 3257343.58 frames. utt_duration=1268 frames, utt_pad_proportion=0.05384, over 10286.24 utterances.], batch size: 55, lr: 8.34e-03, grad_scale: 8.0 2023-03-08 08:37:10,664 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:37:15,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.441e+02 2.731e+02 3.266e+02 6.421e+02, threshold=5.462e+02, percent-clipped=3.0 2023-03-08 08:37:17,333 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:37:21,433 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:37:34,153 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:38:04,752 INFO [train2.py:809] (3/4) Epoch 13, batch 2350, loss[ctc_loss=0.09637, att_loss=0.2561, loss=0.2242, over 17030.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007928, over 51.00 utterances.], tot_loss[ctc_loss=0.09652, att_loss=0.2475, loss=0.2173, over 3267537.81 frames. utt_duration=1226 frames, utt_pad_proportion=0.06033, over 10672.24 utterances.], batch size: 51, lr: 8.33e-03, grad_scale: 8.0 2023-03-08 08:38:58,512 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:39:10,484 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 08:39:24,034 INFO [train2.py:809] (3/4) Epoch 13, batch 2400, loss[ctc_loss=0.1045, att_loss=0.2586, loss=0.2278, over 17410.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04728, over 69.00 utterances.], tot_loss[ctc_loss=0.09629, att_loss=0.2471, loss=0.2169, over 3271832.81 frames. utt_duration=1230 frames, utt_pad_proportion=0.05802, over 10656.61 utterances.], batch size: 69, lr: 8.33e-03, grad_scale: 8.0 2023-03-08 08:39:54,445 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.261e+02 2.728e+02 3.440e+02 5.730e+02, threshold=5.455e+02, percent-clipped=1.0 2023-03-08 08:40:42,744 INFO [train2.py:809] (3/4) Epoch 13, batch 2450, loss[ctc_loss=0.1186, att_loss=0.2691, loss=0.239, over 17266.00 frames. utt_duration=1257 frames, utt_pad_proportion=0.01395, over 55.00 utterances.], tot_loss[ctc_loss=0.09544, att_loss=0.2466, loss=0.2163, over 3278668.70 frames. utt_duration=1252 frames, utt_pad_proportion=0.05104, over 10485.91 utterances.], batch size: 55, lr: 8.32e-03, grad_scale: 8.0 2023-03-08 08:40:46,210 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:42:02,226 INFO [train2.py:809] (3/4) Epoch 13, batch 2500, loss[ctc_loss=0.09721, att_loss=0.2633, loss=0.23, over 17071.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.00847, over 53.00 utterances.], tot_loss[ctc_loss=0.09518, att_loss=0.2464, loss=0.2161, over 3274490.22 frames. utt_duration=1245 frames, utt_pad_proportion=0.05482, over 10530.72 utterances.], batch size: 53, lr: 8.32e-03, grad_scale: 8.0 2023-03-08 08:42:31,318 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-03-08 08:42:33,323 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.276e+02 2.773e+02 3.463e+02 7.694e+02, threshold=5.547e+02, percent-clipped=7.0 2023-03-08 08:43:21,289 INFO [train2.py:809] (3/4) Epoch 13, batch 2550, loss[ctc_loss=0.1052, att_loss=0.2572, loss=0.2268, over 17336.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02272, over 59.00 utterances.], tot_loss[ctc_loss=0.09544, att_loss=0.2465, loss=0.2163, over 3282516.99 frames. utt_duration=1244 frames, utt_pad_proportion=0.05263, over 10567.09 utterances.], batch size: 59, lr: 8.32e-03, grad_scale: 8.0 2023-03-08 08:44:26,707 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0715, 5.0455, 4.9114, 2.3951, 1.9740, 2.7473, 2.5250, 3.6447], device='cuda:3'), covar=tensor([0.0708, 0.0242, 0.0269, 0.4435, 0.6012, 0.2712, 0.2869, 0.2050], device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0233, 0.0241, 0.0220, 0.0349, 0.0333, 0.0235, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.5083e-04, 8.6037e-05, 1.0445e-04, 9.6653e-05, 1.4985e-04, 1.3374e-04, 9.3070e-05, 1.4883e-04], device='cuda:3') 2023-03-08 08:44:40,231 INFO [train2.py:809] (3/4) Epoch 13, batch 2600, loss[ctc_loss=0.08828, att_loss=0.2577, loss=0.2238, over 17310.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02256, over 59.00 utterances.], tot_loss[ctc_loss=0.09428, att_loss=0.2455, loss=0.2153, over 3270299.01 frames. utt_duration=1254 frames, utt_pad_proportion=0.05226, over 10447.07 utterances.], batch size: 59, lr: 8.31e-03, grad_scale: 8.0 2023-03-08 08:45:06,234 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:45:10,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.288e+02 2.850e+02 3.496e+02 8.572e+02, threshold=5.699e+02, percent-clipped=4.0 2023-03-08 08:45:16,582 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-08 08:45:39,275 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4772, 2.9436, 3.7035, 4.5259, 4.0078, 3.9947, 2.9407, 2.3499], device='cuda:3'), covar=tensor([0.0721, 0.2188, 0.0863, 0.0691, 0.0738, 0.0446, 0.1529, 0.2339], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0214, 0.0192, 0.0201, 0.0199, 0.0163, 0.0201, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 08:45:59,192 INFO [train2.py:809] (3/4) Epoch 13, batch 2650, loss[ctc_loss=0.093, att_loss=0.2304, loss=0.2029, over 15358.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.009137, over 35.00 utterances.], tot_loss[ctc_loss=0.09343, att_loss=0.2447, loss=0.2144, over 3271310.46 frames. utt_duration=1261 frames, utt_pad_proportion=0.05021, over 10389.49 utterances.], batch size: 35, lr: 8.31e-03, grad_scale: 8.0 2023-03-08 08:46:02,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2023-03-08 08:46:22,077 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:46:44,861 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:47:17,951 INFO [train2.py:809] (3/4) Epoch 13, batch 2700, loss[ctc_loss=0.1014, att_loss=0.2635, loss=0.2311, over 17348.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02123, over 59.00 utterances.], tot_loss[ctc_loss=0.09494, att_loss=0.2456, loss=0.2155, over 3278457.89 frames. utt_duration=1232 frames, utt_pad_proportion=0.05596, over 10657.74 utterances.], batch size: 59, lr: 8.30e-03, grad_scale: 8.0 2023-03-08 08:47:49,717 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.365e+02 2.788e+02 3.367e+02 6.677e+02, threshold=5.576e+02, percent-clipped=4.0 2023-03-08 08:47:55,028 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5511, 2.4909, 5.0321, 3.6577, 2.8659, 4.2931, 4.7665, 4.5353], device='cuda:3'), covar=tensor([0.0231, 0.1650, 0.0139, 0.1332, 0.1926, 0.0249, 0.0119, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0243, 0.0145, 0.0309, 0.0269, 0.0188, 0.0129, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:48:23,036 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:48:33,791 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 08:48:35,509 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1290, 3.7422, 3.0562, 3.4286, 3.9074, 3.5933, 2.9114, 4.2988], device='cuda:3'), covar=tensor([0.0841, 0.0402, 0.0956, 0.0650, 0.0616, 0.0616, 0.0819, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0190, 0.0209, 0.0180, 0.0243, 0.0218, 0.0186, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 08:48:38,309 INFO [train2.py:809] (3/4) Epoch 13, batch 2750, loss[ctc_loss=0.07512, att_loss=0.2237, loss=0.194, over 15383.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.00917, over 35.00 utterances.], tot_loss[ctc_loss=0.09504, att_loss=0.2455, loss=0.2154, over 3273102.82 frames. utt_duration=1236 frames, utt_pad_proportion=0.05789, over 10608.43 utterances.], batch size: 35, lr: 8.30e-03, grad_scale: 8.0 2023-03-08 08:49:45,336 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9777, 3.7773, 3.6105, 3.1383, 3.8007, 3.7858, 3.7717, 2.6917], device='cuda:3'), covar=tensor([0.0910, 0.1542, 0.2625, 0.6525, 0.1568, 0.2681, 0.0822, 0.6258], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0133, 0.0143, 0.0213, 0.0109, 0.0196, 0.0120, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 08:49:50,317 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1180, 3.7115, 3.0157, 3.4776, 3.8792, 3.5780, 2.7592, 4.3227], device='cuda:3'), covar=tensor([0.0939, 0.0518, 0.1136, 0.0689, 0.0791, 0.0698, 0.0954, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0191, 0.0210, 0.0180, 0.0244, 0.0219, 0.0186, 0.0260], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 08:49:57,334 INFO [train2.py:809] (3/4) Epoch 13, batch 2800, loss[ctc_loss=0.08022, att_loss=0.2417, loss=0.2094, over 16766.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005715, over 48.00 utterances.], tot_loss[ctc_loss=0.09556, att_loss=0.2465, loss=0.2163, over 3275483.28 frames. utt_duration=1237 frames, utt_pad_proportion=0.0572, over 10600.59 utterances.], batch size: 48, lr: 8.30e-03, grad_scale: 8.0 2023-03-08 08:49:59,322 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:50:02,376 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4860, 2.7267, 3.6539, 4.4276, 3.9377, 4.0099, 2.8479, 2.2818], device='cuda:3'), covar=tensor([0.0709, 0.2365, 0.0830, 0.0570, 0.0859, 0.0426, 0.1592, 0.2287], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0215, 0.0192, 0.0201, 0.0200, 0.0164, 0.0201, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 08:50:27,129 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0064, 3.9900, 3.9169, 2.6570, 3.8397, 3.8242, 3.5714, 2.6742], device='cuda:3'), covar=tensor([0.0119, 0.0124, 0.0205, 0.0992, 0.0122, 0.0307, 0.0289, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0088, 0.0085, 0.0105, 0.0075, 0.0096, 0.0092, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 08:50:29,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.488e+02 2.143e+02 2.586e+02 3.162e+02 8.446e+02, threshold=5.172e+02, percent-clipped=2.0 2023-03-08 08:51:17,763 INFO [train2.py:809] (3/4) Epoch 13, batch 2850, loss[ctc_loss=0.07381, att_loss=0.2365, loss=0.204, over 17032.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007002, over 51.00 utterances.], tot_loss[ctc_loss=0.09538, att_loss=0.2464, loss=0.2162, over 3271432.82 frames. utt_duration=1224 frames, utt_pad_proportion=0.06085, over 10703.24 utterances.], batch size: 51, lr: 8.29e-03, grad_scale: 8.0 2023-03-08 08:52:37,441 INFO [train2.py:809] (3/4) Epoch 13, batch 2900, loss[ctc_loss=0.1212, att_loss=0.2678, loss=0.2385, over 17038.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.00971, over 53.00 utterances.], tot_loss[ctc_loss=0.09479, att_loss=0.2467, loss=0.2163, over 3280446.24 frames. utt_duration=1223 frames, utt_pad_proportion=0.05838, over 10742.85 utterances.], batch size: 53, lr: 8.29e-03, grad_scale: 8.0 2023-03-08 08:53:08,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.231e+02 2.674e+02 3.196e+02 5.262e+02, threshold=5.348e+02, percent-clipped=1.0 2023-03-08 08:53:46,462 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-03-08 08:53:56,557 INFO [train2.py:809] (3/4) Epoch 13, batch 2950, loss[ctc_loss=0.09646, att_loss=0.2595, loss=0.2269, over 17283.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02561, over 59.00 utterances.], tot_loss[ctc_loss=0.09504, att_loss=0.247, loss=0.2166, over 3284654.72 frames. utt_duration=1243 frames, utt_pad_proportion=0.05241, over 10580.93 utterances.], batch size: 59, lr: 8.28e-03, grad_scale: 8.0 2023-03-08 08:54:39,286 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:54:42,252 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:55:16,332 INFO [train2.py:809] (3/4) Epoch 13, batch 3000, loss[ctc_loss=0.06682, att_loss=0.2234, loss=0.192, over 16175.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006703, over 41.00 utterances.], tot_loss[ctc_loss=0.09613, att_loss=0.2478, loss=0.2174, over 3289965.05 frames. utt_duration=1227 frames, utt_pad_proportion=0.05467, over 10740.09 utterances.], batch size: 41, lr: 8.28e-03, grad_scale: 8.0 2023-03-08 08:55:16,333 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 08:55:30,013 INFO [train2.py:843] (3/4) Epoch 13, validation: ctc_loss=0.04571, att_loss=0.2368, loss=0.1986, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 08:55:30,014 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 08:55:33,436 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 08:55:56,832 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1768, 5.1656, 4.9734, 2.5723, 2.0604, 2.7611, 3.2790, 3.8637], device='cuda:3'), covar=tensor([0.0625, 0.0288, 0.0266, 0.4568, 0.5975, 0.2610, 0.2280, 0.1836], device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0234, 0.0240, 0.0218, 0.0346, 0.0331, 0.0234, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4975e-04, 8.6546e-05, 1.0372e-04, 9.5790e-05, 1.4861e-04, 1.3304e-04, 9.2756e-05, 1.4746e-04], device='cuda:3') 2023-03-08 08:56:00,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.441e+02 3.101e+02 4.045e+02 6.345e+02, threshold=6.202e+02, percent-clipped=8.0 2023-03-08 08:56:12,101 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:56:29,165 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:56:44,472 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:56:48,707 INFO [train2.py:809] (3/4) Epoch 13, batch 3050, loss[ctc_loss=0.1049, att_loss=0.2648, loss=0.2328, over 17043.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009245, over 52.00 utterances.], tot_loss[ctc_loss=0.09586, att_loss=0.2471, loss=0.2169, over 3278494.29 frames. utt_duration=1238 frames, utt_pad_proportion=0.05453, over 10606.31 utterances.], batch size: 52, lr: 8.28e-03, grad_scale: 8.0 2023-03-08 08:57:10,214 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:57:28,023 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-03-08 08:57:29,965 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-08 08:57:39,721 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0816, 3.8435, 3.2185, 3.6784, 4.0064, 3.7973, 2.8098, 4.3224], device='cuda:3'), covar=tensor([0.0856, 0.0470, 0.0958, 0.0554, 0.0648, 0.0577, 0.0900, 0.0460], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0192, 0.0207, 0.0179, 0.0243, 0.0218, 0.0186, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 08:57:49,291 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-08 08:58:00,755 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 08:58:02,422 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:58:08,261 INFO [train2.py:809] (3/4) Epoch 13, batch 3100, loss[ctc_loss=0.08712, att_loss=0.2467, loss=0.2148, over 16003.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007528, over 40.00 utterances.], tot_loss[ctc_loss=0.09573, att_loss=0.2473, loss=0.217, over 3282620.97 frames. utt_duration=1254 frames, utt_pad_proportion=0.05044, over 10481.14 utterances.], batch size: 40, lr: 8.27e-03, grad_scale: 8.0 2023-03-08 08:58:39,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.262e+02 2.755e+02 3.486e+02 1.234e+03, threshold=5.511e+02, percent-clipped=2.0 2023-03-08 08:59:11,647 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 08:59:26,546 INFO [train2.py:809] (3/4) Epoch 13, batch 3150, loss[ctc_loss=0.08163, att_loss=0.2323, loss=0.2022, over 16167.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006995, over 41.00 utterances.], tot_loss[ctc_loss=0.09564, att_loss=0.2472, loss=0.2169, over 3280862.42 frames. utt_duration=1272 frames, utt_pad_proportion=0.04727, over 10332.82 utterances.], batch size: 41, lr: 8.27e-03, grad_scale: 8.0 2023-03-08 09:00:46,290 INFO [train2.py:809] (3/4) Epoch 13, batch 3200, loss[ctc_loss=0.0907, att_loss=0.2508, loss=0.2188, over 17298.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01209, over 55.00 utterances.], tot_loss[ctc_loss=0.09527, att_loss=0.2466, loss=0.2163, over 3268104.81 frames. utt_duration=1264 frames, utt_pad_proportion=0.05121, over 10350.90 utterances.], batch size: 55, lr: 8.26e-03, grad_scale: 8.0 2023-03-08 09:00:48,236 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:01:14,587 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-08 09:01:17,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.247e+02 2.628e+02 3.429e+02 8.157e+02, threshold=5.255e+02, percent-clipped=5.0 2023-03-08 09:01:37,866 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1971, 2.7905, 3.2883, 4.4127, 3.8082, 3.8924, 2.8159, 2.0975], device='cuda:3'), covar=tensor([0.0780, 0.2122, 0.0961, 0.0563, 0.0838, 0.0448, 0.1605, 0.2332], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0211, 0.0188, 0.0199, 0.0199, 0.0159, 0.0196, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:02:01,021 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5163, 3.1468, 3.6364, 2.8613, 3.5289, 4.6765, 4.4583, 3.3011], device='cuda:3'), covar=tensor([0.0391, 0.1515, 0.1112, 0.1459, 0.1076, 0.0727, 0.0484, 0.1292], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0237, 0.0256, 0.0209, 0.0254, 0.0324, 0.0234, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:02:05,327 INFO [train2.py:809] (3/4) Epoch 13, batch 3250, loss[ctc_loss=0.09328, att_loss=0.2514, loss=0.2198, over 17355.00 frames. utt_duration=880.2 frames, utt_pad_proportion=0.07542, over 79.00 utterances.], tot_loss[ctc_loss=0.09437, att_loss=0.2463, loss=0.2159, over 3271953.15 frames. utt_duration=1275 frames, utt_pad_proportion=0.04763, over 10275.29 utterances.], batch size: 79, lr: 8.26e-03, grad_scale: 8.0 2023-03-08 09:02:14,819 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:02:21,178 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8405, 6.0596, 5.5747, 5.8195, 5.7232, 5.3169, 5.4540, 5.2830], device='cuda:3'), covar=tensor([0.1112, 0.0821, 0.0825, 0.0729, 0.0916, 0.1412, 0.2257, 0.2099], device='cuda:3'), in_proj_covar=tensor([0.0456, 0.0538, 0.0396, 0.0398, 0.0384, 0.0433, 0.0546, 0.0483], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 09:03:23,231 INFO [train2.py:809] (3/4) Epoch 13, batch 3300, loss[ctc_loss=0.1055, att_loss=0.2587, loss=0.2281, over 17360.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.05009, over 69.00 utterances.], tot_loss[ctc_loss=0.09515, att_loss=0.2469, loss=0.2166, over 3275175.19 frames. utt_duration=1263 frames, utt_pad_proportion=0.04961, over 10384.49 utterances.], batch size: 69, lr: 8.26e-03, grad_scale: 8.0 2023-03-08 09:03:48,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-03-08 09:03:51,941 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:03:54,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.303e+02 2.804e+02 3.427e+02 6.539e+02, threshold=5.609e+02, percent-clipped=6.0 2023-03-08 09:04:14,240 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:04:41,664 INFO [train2.py:809] (3/4) Epoch 13, batch 3350, loss[ctc_loss=0.112, att_loss=0.2677, loss=0.2366, over 17302.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02373, over 59.00 utterances.], tot_loss[ctc_loss=0.09522, att_loss=0.2472, loss=0.2168, over 3286701.45 frames. utt_duration=1260 frames, utt_pad_proportion=0.04621, over 10449.05 utterances.], batch size: 59, lr: 8.25e-03, grad_scale: 8.0 2023-03-08 09:04:55,616 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 09:05:07,036 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5058, 3.5417, 3.5086, 3.1040, 3.6238, 3.6200, 3.5038, 2.6960], device='cuda:3'), covar=tensor([0.1083, 0.2023, 0.3452, 0.5036, 0.1088, 0.4971, 0.1058, 0.5911], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0137, 0.0147, 0.0217, 0.0111, 0.0201, 0.0124, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:05:08,582 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:05:27,342 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6795, 2.3562, 5.0446, 4.1889, 3.0884, 4.5690, 5.0356, 4.7150], device='cuda:3'), covar=tensor([0.0151, 0.1558, 0.0193, 0.0701, 0.1615, 0.0160, 0.0083, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0242, 0.0146, 0.0306, 0.0271, 0.0187, 0.0130, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:05:55,379 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:06:01,140 INFO [train2.py:809] (3/4) Epoch 13, batch 3400, loss[ctc_loss=0.1026, att_loss=0.2199, loss=0.1965, over 15504.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008466, over 36.00 utterances.], tot_loss[ctc_loss=0.09449, att_loss=0.2457, loss=0.2154, over 3274964.26 frames. utt_duration=1273 frames, utt_pad_proportion=0.04544, over 10301.69 utterances.], batch size: 36, lr: 8.25e-03, grad_scale: 8.0 2023-03-08 09:06:33,020 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.487e+02 2.909e+02 3.549e+02 7.479e+02, threshold=5.817e+02, percent-clipped=4.0 2023-03-08 09:06:45,934 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 09:07:11,555 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:07:20,300 INFO [train2.py:809] (3/4) Epoch 13, batch 3450, loss[ctc_loss=0.09169, att_loss=0.2246, loss=0.198, over 15870.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009031, over 39.00 utterances.], tot_loss[ctc_loss=0.09318, att_loss=0.2454, loss=0.215, over 3274935.07 frames. utt_duration=1287 frames, utt_pad_proportion=0.04278, over 10188.64 utterances.], batch size: 39, lr: 8.24e-03, grad_scale: 8.0 2023-03-08 09:07:32,539 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6303, 5.8874, 5.3195, 5.6848, 5.4645, 5.1598, 5.2416, 5.1205], device='cuda:3'), covar=tensor([0.1376, 0.0956, 0.0870, 0.0789, 0.0910, 0.1677, 0.2639, 0.2448], device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0553, 0.0407, 0.0407, 0.0394, 0.0443, 0.0557, 0.0494], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:08:33,572 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:08:35,247 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1896, 5.1893, 4.9043, 2.8635, 4.9489, 4.4690, 4.3573, 2.9934], device='cuda:3'), covar=tensor([0.0104, 0.0067, 0.0245, 0.0983, 0.0077, 0.0191, 0.0263, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0086, 0.0084, 0.0103, 0.0073, 0.0096, 0.0091, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 09:08:39,553 INFO [train2.py:809] (3/4) Epoch 13, batch 3500, loss[ctc_loss=0.07467, att_loss=0.2316, loss=0.2003, over 16272.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.006557, over 43.00 utterances.], tot_loss[ctc_loss=0.0919, att_loss=0.2442, loss=0.2137, over 3270272.22 frames. utt_duration=1320 frames, utt_pad_proportion=0.03701, over 9921.98 utterances.], batch size: 43, lr: 8.24e-03, grad_scale: 8.0 2023-03-08 09:08:56,916 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8501, 5.2051, 4.5929, 5.2248, 4.6454, 4.8918, 5.2477, 5.0388], device='cuda:3'), covar=tensor([0.0553, 0.0243, 0.1015, 0.0271, 0.0456, 0.0285, 0.0313, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0278, 0.0333, 0.0282, 0.0286, 0.0216, 0.0265, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 09:09:10,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.305e+02 2.663e+02 3.285e+02 9.490e+02, threshold=5.326e+02, percent-clipped=4.0 2023-03-08 09:09:57,629 INFO [train2.py:809] (3/4) Epoch 13, batch 3550, loss[ctc_loss=0.1578, att_loss=0.2836, loss=0.2584, over 14616.00 frames. utt_duration=401.9 frames, utt_pad_proportion=0.3011, over 146.00 utterances.], tot_loss[ctc_loss=0.09246, att_loss=0.2447, loss=0.2142, over 3277229.12 frames. utt_duration=1316 frames, utt_pad_proportion=0.03712, over 9974.48 utterances.], batch size: 146, lr: 8.24e-03, grad_scale: 8.0 2023-03-08 09:10:15,520 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6140, 2.6384, 4.9824, 4.1095, 3.0008, 4.5611, 5.0636, 4.7286], device='cuda:3'), covar=tensor([0.0256, 0.1606, 0.0294, 0.0950, 0.1921, 0.0203, 0.0100, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0241, 0.0146, 0.0306, 0.0270, 0.0188, 0.0131, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:10:39,743 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:10:41,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3302, 2.5959, 3.3209, 4.3861, 3.8545, 3.9737, 2.6921, 1.9530], device='cuda:3'), covar=tensor([0.0748, 0.2457, 0.0998, 0.0569, 0.0792, 0.0411, 0.1843, 0.2680], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0213, 0.0190, 0.0202, 0.0202, 0.0161, 0.0198, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:10:54,965 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:11:16,649 INFO [train2.py:809] (3/4) Epoch 13, batch 3600, loss[ctc_loss=0.1034, att_loss=0.2613, loss=0.2298, over 17061.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.008287, over 53.00 utterances.], tot_loss[ctc_loss=0.0927, att_loss=0.2447, loss=0.2143, over 3276253.96 frames. utt_duration=1318 frames, utt_pad_proportion=0.03599, over 9954.50 utterances.], batch size: 53, lr: 8.23e-03, grad_scale: 8.0 2023-03-08 09:11:36,218 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:11:46,270 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.412e+02 2.884e+02 3.551e+02 7.030e+02, threshold=5.767e+02, percent-clipped=4.0 2023-03-08 09:12:05,903 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:12:13,335 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:12:19,092 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0314, 5.3488, 5.5986, 5.3750, 5.5307, 5.9759, 5.2540, 6.1359], device='cuda:3'), covar=tensor([0.0652, 0.0647, 0.0697, 0.1076, 0.1642, 0.0964, 0.0561, 0.0559], device='cuda:3'), in_proj_covar=tensor([0.0769, 0.0455, 0.0536, 0.0590, 0.0786, 0.0540, 0.0436, 0.0524], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:12:29,033 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:12:33,509 INFO [train2.py:809] (3/4) Epoch 13, batch 3650, loss[ctc_loss=0.06655, att_loss=0.2398, loss=0.2052, over 16466.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006599, over 46.00 utterances.], tot_loss[ctc_loss=0.09391, att_loss=0.2452, loss=0.2149, over 3274549.45 frames. utt_duration=1301 frames, utt_pad_proportion=0.03949, over 10081.79 utterances.], batch size: 46, lr: 8.23e-03, grad_scale: 8.0 2023-03-08 09:12:47,050 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 09:12:58,846 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-08 09:13:20,392 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:13:34,852 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:13:39,561 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8669, 5.1413, 5.4320, 5.3502, 5.3485, 5.8838, 5.0906, 5.9514], device='cuda:3'), covar=tensor([0.0654, 0.0699, 0.0762, 0.0995, 0.1718, 0.0853, 0.0621, 0.0602], device='cuda:3'), in_proj_covar=tensor([0.0766, 0.0452, 0.0535, 0.0587, 0.0780, 0.0539, 0.0434, 0.0522], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:13:52,934 INFO [train2.py:809] (3/4) Epoch 13, batch 3700, loss[ctc_loss=0.09563, att_loss=0.2579, loss=0.2254, over 17024.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007438, over 51.00 utterances.], tot_loss[ctc_loss=0.0952, att_loss=0.2462, loss=0.216, over 3273742.69 frames. utt_duration=1262 frames, utt_pad_proportion=0.05044, over 10388.72 utterances.], batch size: 51, lr: 8.22e-03, grad_scale: 8.0 2023-03-08 09:14:03,480 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:14:14,283 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1061, 4.5041, 4.5125, 4.9198, 2.9087, 4.5172, 2.5303, 1.9064], device='cuda:3'), covar=tensor([0.0400, 0.0192, 0.0724, 0.0118, 0.1522, 0.0146, 0.1639, 0.1758], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0127, 0.0258, 0.0120, 0.0223, 0.0114, 0.0227, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:14:23,030 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.328e+02 2.806e+02 3.830e+02 9.896e+02, threshold=5.611e+02, percent-clipped=3.0 2023-03-08 09:14:28,083 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 09:15:13,024 INFO [train2.py:809] (3/4) Epoch 13, batch 3750, loss[ctc_loss=0.07419, att_loss=0.2218, loss=0.1923, over 15897.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.00849, over 39.00 utterances.], tot_loss[ctc_loss=0.09384, att_loss=0.2455, loss=0.2152, over 3279133.39 frames. utt_duration=1283 frames, utt_pad_proportion=0.04426, over 10236.69 utterances.], batch size: 39, lr: 8.22e-03, grad_scale: 8.0 2023-03-08 09:15:13,462 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:15:38,973 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5040, 4.8960, 5.1667, 4.9537, 5.0099, 5.4942, 4.8678, 5.5400], device='cuda:3'), covar=tensor([0.0712, 0.0710, 0.0681, 0.1052, 0.1735, 0.0867, 0.0876, 0.0673], device='cuda:3'), in_proj_covar=tensor([0.0764, 0.0448, 0.0533, 0.0584, 0.0776, 0.0533, 0.0434, 0.0522], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:16:26,961 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:16:32,780 INFO [train2.py:809] (3/4) Epoch 13, batch 3800, loss[ctc_loss=0.1019, att_loss=0.2575, loss=0.2264, over 17063.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009035, over 53.00 utterances.], tot_loss[ctc_loss=0.09428, att_loss=0.246, loss=0.2156, over 3276936.81 frames. utt_duration=1263 frames, utt_pad_proportion=0.05003, over 10389.89 utterances.], batch size: 53, lr: 8.22e-03, grad_scale: 8.0 2023-03-08 09:16:34,488 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0997, 5.3541, 5.6564, 5.5181, 5.5750, 6.0744, 5.2177, 6.1742], device='cuda:3'), covar=tensor([0.0636, 0.0731, 0.0657, 0.1081, 0.1693, 0.0829, 0.0590, 0.0573], device='cuda:3'), in_proj_covar=tensor([0.0760, 0.0448, 0.0532, 0.0583, 0.0775, 0.0534, 0.0434, 0.0520], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:17:02,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.194e+02 2.674e+02 3.110e+02 7.500e+02, threshold=5.349e+02, percent-clipped=1.0 2023-03-08 09:17:41,369 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:17:51,575 INFO [train2.py:809] (3/4) Epoch 13, batch 3850, loss[ctc_loss=0.1009, att_loss=0.2523, loss=0.222, over 16868.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.008149, over 49.00 utterances.], tot_loss[ctc_loss=0.09501, att_loss=0.2467, loss=0.2164, over 3279228.78 frames. utt_duration=1254 frames, utt_pad_proportion=0.05308, over 10475.02 utterances.], batch size: 49, lr: 8.21e-03, grad_scale: 8.0 2023-03-08 09:17:56,556 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6327, 3.5769, 3.5066, 3.0970, 3.6357, 3.4800, 3.5473, 2.4252], device='cuda:3'), covar=tensor([0.0930, 0.1618, 0.2400, 0.4729, 0.0855, 0.2954, 0.0953, 0.6226], device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0135, 0.0146, 0.0215, 0.0111, 0.0200, 0.0123, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:18:12,636 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 2023-03-08 09:19:07,770 INFO [train2.py:809] (3/4) Epoch 13, batch 3900, loss[ctc_loss=0.1197, att_loss=0.2714, loss=0.2411, over 17414.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04537, over 69.00 utterances.], tot_loss[ctc_loss=0.09472, att_loss=0.2463, loss=0.216, over 3273072.58 frames. utt_duration=1238 frames, utt_pad_proportion=0.05736, over 10587.32 utterances.], batch size: 69, lr: 8.21e-03, grad_scale: 8.0 2023-03-08 09:19:26,177 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:19:36,508 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.485e+02 2.131e+02 2.698e+02 3.249e+02 8.076e+02, threshold=5.395e+02, percent-clipped=6.0 2023-03-08 09:19:45,767 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9576, 4.4198, 4.3873, 4.8863, 2.4955, 4.4214, 2.4105, 1.5996], device='cuda:3'), covar=tensor([0.0437, 0.0210, 0.0804, 0.0137, 0.1967, 0.0188, 0.1851, 0.1905], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0128, 0.0255, 0.0120, 0.0224, 0.0114, 0.0226, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:19:56,080 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:20:10,180 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:20:11,527 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:20:23,592 INFO [train2.py:809] (3/4) Epoch 13, batch 3950, loss[ctc_loss=0.1067, att_loss=0.2648, loss=0.2332, over 17369.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03559, over 63.00 utterances.], tot_loss[ctc_loss=0.09457, att_loss=0.2466, loss=0.2162, over 3277899.24 frames. utt_duration=1265 frames, utt_pad_proportion=0.04972, over 10375.75 utterances.], batch size: 63, lr: 8.20e-03, grad_scale: 8.0 2023-03-08 09:20:38,818 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:20:40,784 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0936, 4.3600, 4.4757, 4.8704, 2.6794, 4.2821, 2.3223, 1.4913], device='cuda:3'), covar=tensor([0.0385, 0.0184, 0.0596, 0.0169, 0.1758, 0.0173, 0.1850, 0.1967], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0127, 0.0255, 0.0120, 0.0222, 0.0114, 0.0226, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:20:48,317 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2259, 4.5225, 4.8396, 5.0660, 2.8271, 4.4087, 2.6053, 1.6525], device='cuda:3'), covar=tensor([0.0337, 0.0177, 0.0516, 0.0105, 0.1567, 0.0154, 0.1663, 0.1831], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0127, 0.0255, 0.0120, 0.0223, 0.0114, 0.0227, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:20:57,145 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3031, 5.2608, 5.0147, 2.9384, 5.0437, 4.8288, 4.5807, 2.7438], device='cuda:3'), covar=tensor([0.0084, 0.0086, 0.0295, 0.1050, 0.0092, 0.0160, 0.0261, 0.1351], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0088, 0.0085, 0.0105, 0.0074, 0.0097, 0.0092, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:21:41,016 INFO [train2.py:809] (3/4) Epoch 14, batch 0, loss[ctc_loss=0.08693, att_loss=0.2337, loss=0.2043, over 13277.00 frames. utt_duration=1833 frames, utt_pad_proportion=0.1016, over 29.00 utterances.], tot_loss[ctc_loss=0.08693, att_loss=0.2337, loss=0.2043, over 13277.00 frames. utt_duration=1833 frames, utt_pad_proportion=0.1016, over 29.00 utterances.], batch size: 29, lr: 7.90e-03, grad_scale: 8.0 2023-03-08 09:21:41,016 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 09:21:52,750 INFO [train2.py:843] (3/4) Epoch 14, validation: ctc_loss=0.04501, att_loss=0.2367, loss=0.1984, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 09:21:52,750 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 09:22:20,798 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:22:46,826 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 2.322e+02 2.890e+02 3.740e+02 8.433e+02, threshold=5.780e+02, percent-clipped=5.0 2023-03-08 09:22:52,384 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:22:56,837 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5717, 5.0896, 3.6807, 5.3062, 4.6658, 4.9738, 4.7913, 4.8880], device='cuda:3'), covar=tensor([0.0676, 0.0367, 0.1666, 0.0251, 0.0408, 0.0319, 0.0791, 0.0346], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0271, 0.0323, 0.0278, 0.0279, 0.0211, 0.0258, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-08 09:23:11,218 INFO [train2.py:809] (3/4) Epoch 14, batch 50, loss[ctc_loss=0.09088, att_loss=0.2504, loss=0.2185, over 16277.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007386, over 43.00 utterances.], tot_loss[ctc_loss=0.09363, att_loss=0.2444, loss=0.2143, over 732820.44 frames. utt_duration=1249 frames, utt_pad_proportion=0.06186, over 2348.78 utterances.], batch size: 43, lr: 7.90e-03, grad_scale: 8.0 2023-03-08 09:23:16,825 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6657, 2.2058, 2.0576, 2.2713, 2.9106, 2.3179, 2.1280, 2.8851], device='cuda:3'), covar=tensor([0.2852, 0.4497, 0.4032, 0.2305, 0.2115, 0.2070, 0.4135, 0.1373], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0091, 0.0095, 0.0079, 0.0084, 0.0073, 0.0096, 0.0064], device='cuda:3'), out_proj_covar=tensor([5.8600e-05, 6.5840e-05, 6.8748e-05, 5.7789e-05, 5.8509e-05, 5.5992e-05, 6.7412e-05, 5.0012e-05], device='cuda:3') 2023-03-08 09:23:29,062 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:23:44,948 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1191, 4.4098, 4.3862, 4.9249, 2.6902, 4.5656, 2.5367, 1.6112], device='cuda:3'), covar=tensor([0.0356, 0.0181, 0.0723, 0.0110, 0.1796, 0.0130, 0.1712, 0.1881], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0128, 0.0257, 0.0120, 0.0225, 0.0114, 0.0228, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:24:08,121 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:24:29,806 INFO [train2.py:809] (3/4) Epoch 14, batch 100, loss[ctc_loss=0.07228, att_loss=0.2147, loss=0.1862, over 15361.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01063, over 35.00 utterances.], tot_loss[ctc_loss=0.09189, att_loss=0.2449, loss=0.2143, over 1304925.82 frames. utt_duration=1251 frames, utt_pad_proportion=0.04732, over 4177.86 utterances.], batch size: 35, lr: 7.90e-03, grad_scale: 8.0 2023-03-08 09:25:07,672 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0950, 5.3624, 5.6147, 5.5104, 5.5554, 6.0517, 5.2133, 6.1298], device='cuda:3'), covar=tensor([0.0576, 0.0655, 0.0724, 0.1035, 0.1748, 0.0768, 0.0616, 0.0589], device='cuda:3'), in_proj_covar=tensor([0.0760, 0.0450, 0.0536, 0.0585, 0.0780, 0.0539, 0.0434, 0.0523], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:25:10,988 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2062, 4.4101, 4.2885, 4.8181, 2.6633, 4.4620, 2.3526, 1.7244], device='cuda:3'), covar=tensor([0.0285, 0.0133, 0.0684, 0.0111, 0.1820, 0.0142, 0.1859, 0.1887], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0129, 0.0259, 0.0121, 0.0227, 0.0115, 0.0230, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 09:25:24,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.393e+02 2.937e+02 3.541e+02 1.195e+03, threshold=5.873e+02, percent-clipped=2.0 2023-03-08 09:25:48,792 INFO [train2.py:809] (3/4) Epoch 14, batch 150, loss[ctc_loss=0.1298, att_loss=0.2638, loss=0.237, over 17117.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01504, over 56.00 utterances.], tot_loss[ctc_loss=0.09293, att_loss=0.2454, loss=0.2149, over 1735539.82 frames. utt_duration=1241 frames, utt_pad_proportion=0.05618, over 5602.13 utterances.], batch size: 56, lr: 7.89e-03, grad_scale: 8.0 2023-03-08 09:27:06,080 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3759, 5.0311, 4.7765, 4.8935, 4.9271, 4.6270, 3.1625, 4.8173], device='cuda:3'), covar=tensor([0.0139, 0.0122, 0.0130, 0.0094, 0.0111, 0.0129, 0.0862, 0.0275], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0075, 0.0093, 0.0058, 0.0063, 0.0074, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:27:06,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-08 09:27:07,214 INFO [train2.py:809] (3/4) Epoch 14, batch 200, loss[ctc_loss=0.1402, att_loss=0.2775, loss=0.25, over 16468.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006584, over 46.00 utterances.], tot_loss[ctc_loss=0.09197, att_loss=0.2452, loss=0.2145, over 2086384.29 frames. utt_duration=1268 frames, utt_pad_proportion=0.04485, over 6591.03 utterances.], batch size: 46, lr: 7.89e-03, grad_scale: 8.0 2023-03-08 09:27:07,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.03 vs. limit=5.0 2023-03-08 09:28:06,027 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.279e+02 2.700e+02 3.291e+02 7.221e+02, threshold=5.400e+02, percent-clipped=2.0 2023-03-08 09:28:27,689 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:28:30,295 INFO [train2.py:809] (3/4) Epoch 14, batch 250, loss[ctc_loss=0.205, att_loss=0.296, loss=0.2778, over 14187.00 frames. utt_duration=390.1 frames, utt_pad_proportion=0.3216, over 146.00 utterances.], tot_loss[ctc_loss=0.09204, att_loss=0.2446, loss=0.2141, over 2350858.45 frames. utt_duration=1280 frames, utt_pad_proportion=0.04354, over 7357.39 utterances.], batch size: 146, lr: 7.88e-03, grad_scale: 8.0 2023-03-08 09:28:30,677 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:28:32,168 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9447, 4.9592, 4.8564, 2.7032, 4.8218, 4.6567, 4.1763, 2.6939], device='cuda:3'), covar=tensor([0.0160, 0.0103, 0.0260, 0.1227, 0.0095, 0.0172, 0.0371, 0.1484], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0089, 0.0086, 0.0106, 0.0075, 0.0099, 0.0094, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:28:42,590 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:29:30,458 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2023-03-08 09:29:42,264 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:29:48,099 INFO [train2.py:809] (3/4) Epoch 14, batch 300, loss[ctc_loss=0.1394, att_loss=0.2797, loss=0.2516, over 16887.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007326, over 49.00 utterances.], tot_loss[ctc_loss=0.09321, att_loss=0.2453, loss=0.2149, over 2562204.51 frames. utt_duration=1304 frames, utt_pad_proportion=0.03635, over 7871.48 utterances.], batch size: 49, lr: 7.88e-03, grad_scale: 16.0 2023-03-08 09:29:57,178 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:30:05,853 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:30:08,641 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:30:20,603 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4887, 3.5385, 3.3058, 2.9717, 3.4025, 3.4146, 3.4739, 2.3678], device='cuda:3'), covar=tensor([0.1103, 0.1475, 0.2941, 0.5738, 0.2949, 0.4031, 0.1016, 0.6693], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0137, 0.0152, 0.0220, 0.0113, 0.0203, 0.0124, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:30:42,752 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.191e+02 2.821e+02 3.545e+02 1.097e+03, threshold=5.643e+02, percent-clipped=10.0 2023-03-08 09:31:06,327 INFO [train2.py:809] (3/4) Epoch 14, batch 350, loss[ctc_loss=0.1066, att_loss=0.262, loss=0.2309, over 17083.00 frames. utt_duration=1291 frames, utt_pad_proportion=0.008049, over 53.00 utterances.], tot_loss[ctc_loss=0.09369, att_loss=0.2455, loss=0.2151, over 2705693.51 frames. utt_duration=1281 frames, utt_pad_proportion=0.04552, over 8456.49 utterances.], batch size: 53, lr: 7.88e-03, grad_scale: 16.0 2023-03-08 09:31:09,873 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9784, 2.1941, 2.2312, 2.3683, 2.6982, 2.5997, 2.3039, 2.9904], device='cuda:3'), covar=tensor([0.1103, 0.3902, 0.3142, 0.1476, 0.1428, 0.0991, 0.2694, 0.0709], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0094, 0.0080, 0.0084, 0.0073, 0.0097, 0.0065], device='cuda:3'), out_proj_covar=tensor([5.8631e-05, 6.6272e-05, 6.8980e-05, 5.8119e-05, 5.8730e-05, 5.5897e-05, 6.7581e-05, 5.0399e-05], device='cuda:3') 2023-03-08 09:31:18,321 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-08 09:31:24,114 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:32:25,893 INFO [train2.py:809] (3/4) Epoch 14, batch 400, loss[ctc_loss=0.1226, att_loss=0.2626, loss=0.2346, over 16335.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.00605, over 45.00 utterances.], tot_loss[ctc_loss=0.0938, att_loss=0.2464, loss=0.2158, over 2839796.92 frames. utt_duration=1253 frames, utt_pad_proportion=0.04838, over 9077.89 utterances.], batch size: 45, lr: 7.87e-03, grad_scale: 16.0 2023-03-08 09:32:31,164 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:32:40,748 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:33:18,288 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 09:33:21,767 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.286e+02 2.626e+02 3.493e+02 1.383e+03, threshold=5.251e+02, percent-clipped=5.0 2023-03-08 09:33:37,136 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3418, 4.7792, 4.7337, 4.7228, 4.8241, 4.5762, 2.8387, 4.6046], device='cuda:3'), covar=tensor([0.0132, 0.0150, 0.0138, 0.0106, 0.0119, 0.0149, 0.1085, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0075, 0.0093, 0.0058, 0.0064, 0.0074, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:33:46,082 INFO [train2.py:809] (3/4) Epoch 14, batch 450, loss[ctc_loss=0.06671, att_loss=0.2101, loss=0.1814, over 15512.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.007448, over 36.00 utterances.], tot_loss[ctc_loss=0.09344, att_loss=0.2465, loss=0.2159, over 2939203.63 frames. utt_duration=1247 frames, utt_pad_proportion=0.05033, over 9440.02 utterances.], batch size: 36, lr: 7.87e-03, grad_scale: 16.0 2023-03-08 09:34:02,800 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8944, 5.1847, 5.2101, 5.0847, 5.2086, 5.1454, 4.8329, 4.6592], device='cuda:3'), covar=tensor([0.1029, 0.0508, 0.0255, 0.0524, 0.0319, 0.0364, 0.0344, 0.0396], device='cuda:3'), in_proj_covar=tensor([0.0484, 0.0321, 0.0284, 0.0314, 0.0370, 0.0390, 0.0318, 0.0356], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 09:34:09,163 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:34:12,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 09:34:24,454 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0438, 5.0620, 4.9565, 2.9676, 4.8792, 4.7508, 4.3791, 2.6423], device='cuda:3'), covar=tensor([0.0138, 0.0084, 0.0247, 0.1065, 0.0088, 0.0168, 0.0316, 0.1507], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0090, 0.0087, 0.0107, 0.0075, 0.0099, 0.0095, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:35:04,914 INFO [train2.py:809] (3/4) Epoch 14, batch 500, loss[ctc_loss=0.1203, att_loss=0.268, loss=0.2385, over 17371.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02072, over 59.00 utterances.], tot_loss[ctc_loss=0.09397, att_loss=0.2469, loss=0.2163, over 3016125.20 frames. utt_duration=1231 frames, utt_pad_proportion=0.05395, over 9810.39 utterances.], batch size: 59, lr: 7.87e-03, grad_scale: 16.0 2023-03-08 09:35:39,583 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:36:01,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.299e+02 2.275e+02 2.805e+02 3.649e+02 6.948e+02, threshold=5.610e+02, percent-clipped=5.0 2023-03-08 09:36:23,208 INFO [train2.py:809] (3/4) Epoch 14, batch 550, loss[ctc_loss=0.084, att_loss=0.2429, loss=0.2111, over 16759.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006851, over 48.00 utterances.], tot_loss[ctc_loss=0.09405, att_loss=0.2463, loss=0.2158, over 3073794.92 frames. utt_duration=1235 frames, utt_pad_proportion=0.05428, over 9966.88 utterances.], batch size: 48, lr: 7.86e-03, grad_scale: 8.0 2023-03-08 09:37:15,151 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:37:42,123 INFO [train2.py:809] (3/4) Epoch 14, batch 600, loss[ctc_loss=0.08569, att_loss=0.2533, loss=0.2198, over 16881.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007356, over 49.00 utterances.], tot_loss[ctc_loss=0.09408, att_loss=0.2461, loss=0.2157, over 3113743.40 frames. utt_duration=1210 frames, utt_pad_proportion=0.06236, over 10306.33 utterances.], batch size: 49, lr: 7.86e-03, grad_scale: 8.0 2023-03-08 09:37:51,417 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:38:03,147 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:38:12,679 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-08 09:38:39,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.147e+02 2.714e+02 3.419e+02 5.544e+02, threshold=5.429e+02, percent-clipped=0.0 2023-03-08 09:38:49,503 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8009, 4.8169, 4.6969, 2.9345, 4.5879, 4.5393, 4.0181, 2.7435], device='cuda:3'), covar=tensor([0.0097, 0.0092, 0.0221, 0.0977, 0.0112, 0.0165, 0.0338, 0.1323], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0089, 0.0086, 0.0106, 0.0075, 0.0099, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:39:01,291 INFO [train2.py:809] (3/4) Epoch 14, batch 650, loss[ctc_loss=0.09617, att_loss=0.2585, loss=0.226, over 16955.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007471, over 50.00 utterances.], tot_loss[ctc_loss=0.09253, att_loss=0.2443, loss=0.2139, over 3143050.14 frames. utt_duration=1223 frames, utt_pad_proportion=0.06114, over 10292.66 utterances.], batch size: 50, lr: 7.85e-03, grad_scale: 8.0 2023-03-08 09:39:18,703 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:40:19,853 INFO [train2.py:809] (3/4) Epoch 14, batch 700, loss[ctc_loss=0.06956, att_loss=0.2132, loss=0.1845, over 15512.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008019, over 36.00 utterances.], tot_loss[ctc_loss=0.09205, att_loss=0.2445, loss=0.214, over 3172957.64 frames. utt_duration=1221 frames, utt_pad_proportion=0.06253, over 10411.54 utterances.], batch size: 36, lr: 7.85e-03, grad_scale: 8.0 2023-03-08 09:40:30,763 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9045, 5.1682, 4.6884, 5.2555, 4.6344, 4.9249, 5.3014, 5.0898], device='cuda:3'), covar=tensor([0.0540, 0.0289, 0.0858, 0.0250, 0.0419, 0.0225, 0.0262, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0274, 0.0326, 0.0283, 0.0283, 0.0212, 0.0260, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-08 09:41:15,851 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 1.961e+02 2.440e+02 3.056e+02 5.134e+02, threshold=4.881e+02, percent-clipped=0.0 2023-03-08 09:41:37,732 INFO [train2.py:809] (3/4) Epoch 14, batch 750, loss[ctc_loss=0.07496, att_loss=0.2123, loss=0.1848, over 15766.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.00884, over 38.00 utterances.], tot_loss[ctc_loss=0.09185, att_loss=0.2439, loss=0.2135, over 3193465.83 frames. utt_duration=1224 frames, utt_pad_proportion=0.06164, over 10451.97 utterances.], batch size: 38, lr: 7.85e-03, grad_scale: 8.0 2023-03-08 09:41:52,177 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:42:02,160 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:42:09,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 09:42:25,767 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2840, 5.2134, 5.1173, 2.7690, 5.0281, 4.8874, 4.5191, 3.0259], device='cuda:3'), covar=tensor([0.0108, 0.0069, 0.0244, 0.1074, 0.0085, 0.0142, 0.0272, 0.1186], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0089, 0.0086, 0.0106, 0.0075, 0.0098, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:42:56,737 INFO [train2.py:809] (3/4) Epoch 14, batch 800, loss[ctc_loss=0.1165, att_loss=0.2694, loss=0.2388, over 16965.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007597, over 50.00 utterances.], tot_loss[ctc_loss=0.09161, att_loss=0.2441, loss=0.2136, over 3210488.11 frames. utt_duration=1223 frames, utt_pad_proportion=0.06138, over 10511.17 utterances.], batch size: 50, lr: 7.84e-03, grad_scale: 8.0 2023-03-08 09:43:38,090 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:43:51,583 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-08 09:43:54,265 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.071e+02 2.560e+02 3.132e+02 4.899e+02, threshold=5.119e+02, percent-clipped=1.0 2023-03-08 09:44:09,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-03-08 09:44:16,109 INFO [train2.py:809] (3/4) Epoch 14, batch 850, loss[ctc_loss=0.08831, att_loss=0.2489, loss=0.2168, over 17008.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.008574, over 51.00 utterances.], tot_loss[ctc_loss=0.0913, att_loss=0.2441, loss=0.2135, over 3233710.56 frames. utt_duration=1246 frames, utt_pad_proportion=0.05294, over 10393.48 utterances.], batch size: 51, lr: 7.84e-03, grad_scale: 8.0 2023-03-08 09:45:00,839 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:45:35,372 INFO [train2.py:809] (3/4) Epoch 14, batch 900, loss[ctc_loss=0.08669, att_loss=0.2326, loss=0.2034, over 14586.00 frames. utt_duration=1825 frames, utt_pad_proportion=0.04268, over 32.00 utterances.], tot_loss[ctc_loss=0.09127, att_loss=0.2441, loss=0.2135, over 3236751.30 frames. utt_duration=1239 frames, utt_pad_proportion=0.05555, over 10462.31 utterances.], batch size: 32, lr: 7.84e-03, grad_scale: 8.0 2023-03-08 09:45:44,747 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:46:32,364 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.184e+02 2.605e+02 3.188e+02 5.544e+02, threshold=5.210e+02, percent-clipped=2.0 2023-03-08 09:46:44,362 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2023-03-08 09:46:53,847 INFO [train2.py:809] (3/4) Epoch 14, batch 950, loss[ctc_loss=0.0876, att_loss=0.2517, loss=0.2189, over 17310.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02313, over 59.00 utterances.], tot_loss[ctc_loss=0.09105, att_loss=0.2439, loss=0.2133, over 3236844.80 frames. utt_duration=1251 frames, utt_pad_proportion=0.05413, over 10359.59 utterances.], batch size: 59, lr: 7.83e-03, grad_scale: 8.0 2023-03-08 09:47:00,138 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:47:13,093 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:48:04,810 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0126, 5.2984, 5.3100, 5.2127, 5.3436, 5.2989, 5.0328, 4.8184], device='cuda:3'), covar=tensor([0.1067, 0.0539, 0.0254, 0.0479, 0.0290, 0.0277, 0.0301, 0.0300], device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0310, 0.0275, 0.0303, 0.0357, 0.0373, 0.0305, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 09:48:12,277 INFO [train2.py:809] (3/4) Epoch 14, batch 1000, loss[ctc_loss=0.06615, att_loss=0.208, loss=0.1796, over 15779.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008228, over 38.00 utterances.], tot_loss[ctc_loss=0.09121, att_loss=0.244, loss=0.2134, over 3238889.07 frames. utt_duration=1220 frames, utt_pad_proportion=0.06361, over 10633.67 utterances.], batch size: 38, lr: 7.83e-03, grad_scale: 8.0 2023-03-08 09:48:17,259 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1890, 3.8942, 3.3665, 3.6322, 4.0751, 3.8402, 3.3301, 4.4986], device='cuda:3'), covar=tensor([0.0942, 0.0544, 0.1028, 0.0657, 0.0647, 0.0629, 0.0715, 0.0396], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0198, 0.0213, 0.0185, 0.0252, 0.0223, 0.0189, 0.0267], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 09:48:49,227 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:48:58,207 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-08 09:49:06,998 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0998, 5.1429, 4.9557, 2.9020, 4.9692, 4.6972, 4.4758, 2.6590], device='cuda:3'), covar=tensor([0.0164, 0.0078, 0.0254, 0.0992, 0.0086, 0.0170, 0.0277, 0.1377], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0088, 0.0085, 0.0104, 0.0074, 0.0098, 0.0094, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 09:49:09,713 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.243e+02 2.663e+02 3.401e+02 6.618e+02, threshold=5.327e+02, percent-clipped=3.0 2023-03-08 09:49:31,218 INFO [train2.py:809] (3/4) Epoch 14, batch 1050, loss[ctc_loss=0.08299, att_loss=0.245, loss=0.2126, over 16884.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007312, over 49.00 utterances.], tot_loss[ctc_loss=0.09156, att_loss=0.2442, loss=0.2137, over 3240531.32 frames. utt_duration=1188 frames, utt_pad_proportion=0.07343, over 10927.16 utterances.], batch size: 49, lr: 7.82e-03, grad_scale: 4.0 2023-03-08 09:49:42,350 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1256, 3.7807, 3.7053, 2.9495, 3.8023, 3.9298, 3.7297, 2.4615], device='cuda:3'), covar=tensor([0.1049, 0.1858, 0.3304, 0.8537, 0.2430, 0.3337, 0.1148, 1.1024], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0139, 0.0150, 0.0222, 0.0116, 0.0206, 0.0128, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:49:45,266 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:50:49,487 INFO [train2.py:809] (3/4) Epoch 14, batch 1100, loss[ctc_loss=0.09243, att_loss=0.2487, loss=0.2174, over 16493.00 frames. utt_duration=1436 frames, utt_pad_proportion=0.004941, over 46.00 utterances.], tot_loss[ctc_loss=0.09137, att_loss=0.244, loss=0.2135, over 3242927.98 frames. utt_duration=1202 frames, utt_pad_proportion=0.07031, over 10801.16 utterances.], batch size: 46, lr: 7.82e-03, grad_scale: 4.0 2023-03-08 09:51:00,146 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:51:23,342 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 09:51:48,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.277e+02 2.760e+02 3.306e+02 6.438e+02, threshold=5.521e+02, percent-clipped=4.0 2023-03-08 09:52:07,940 INFO [train2.py:809] (3/4) Epoch 14, batch 1150, loss[ctc_loss=0.1121, att_loss=0.2521, loss=0.2241, over 16102.00 frames. utt_duration=1535 frames, utt_pad_proportion=0.006977, over 42.00 utterances.], tot_loss[ctc_loss=0.0914, att_loss=0.2439, loss=0.2134, over 3246970.19 frames. utt_duration=1207 frames, utt_pad_proportion=0.06878, over 10776.99 utterances.], batch size: 42, lr: 7.82e-03, grad_scale: 4.0 2023-03-08 09:52:52,760 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:53:26,828 INFO [train2.py:809] (3/4) Epoch 14, batch 1200, loss[ctc_loss=0.1006, att_loss=0.2458, loss=0.2168, over 16396.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.008175, over 44.00 utterances.], tot_loss[ctc_loss=0.09026, att_loss=0.2431, loss=0.2125, over 3254194.77 frames. utt_duration=1232 frames, utt_pad_proportion=0.06228, over 10580.37 utterances.], batch size: 44, lr: 7.81e-03, grad_scale: 8.0 2023-03-08 09:54:08,682 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:54:25,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 2.184e+02 2.648e+02 3.563e+02 1.260e+03, threshold=5.296e+02, percent-clipped=9.0 2023-03-08 09:54:33,634 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9307, 5.1720, 5.4900, 5.2747, 5.3635, 5.8820, 5.1574, 5.9176], device='cuda:3'), covar=tensor([0.0671, 0.0721, 0.0755, 0.1302, 0.1681, 0.0900, 0.0722, 0.0719], device='cuda:3'), in_proj_covar=tensor([0.0773, 0.0458, 0.0539, 0.0596, 0.0783, 0.0545, 0.0437, 0.0536], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 09:54:39,904 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7680, 4.9944, 4.8770, 4.7940, 5.4446, 4.9742, 4.9582, 2.6710], device='cuda:3'), covar=tensor([0.0177, 0.0195, 0.0228, 0.0277, 0.0653, 0.0146, 0.0226, 0.1882], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0141, 0.0147, 0.0158, 0.0343, 0.0127, 0.0131, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 09:54:45,537 INFO [train2.py:809] (3/4) Epoch 14, batch 1250, loss[ctc_loss=0.1007, att_loss=0.2729, loss=0.2385, over 17050.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009086, over 53.00 utterances.], tot_loss[ctc_loss=0.08998, att_loss=0.2433, loss=0.2126, over 3261708.13 frames. utt_duration=1235 frames, utt_pad_proportion=0.05915, over 10579.28 utterances.], batch size: 53, lr: 7.81e-03, grad_scale: 8.0 2023-03-08 09:55:10,577 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:55:54,496 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-03-08 09:56:03,885 INFO [train2.py:809] (3/4) Epoch 14, batch 1300, loss[ctc_loss=0.07742, att_loss=0.2432, loss=0.21, over 16335.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005943, over 45.00 utterances.], tot_loss[ctc_loss=0.08918, att_loss=0.2426, loss=0.2119, over 3260242.63 frames. utt_duration=1269 frames, utt_pad_proportion=0.05105, over 10291.10 utterances.], batch size: 45, lr: 7.81e-03, grad_scale: 8.0 2023-03-08 09:56:23,370 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-03-08 09:56:32,058 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:56:47,510 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 09:57:03,199 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.168e+02 2.572e+02 3.596e+02 7.210e+02, threshold=5.144e+02, percent-clipped=5.0 2023-03-08 09:57:23,234 INFO [train2.py:809] (3/4) Epoch 14, batch 1350, loss[ctc_loss=0.05882, att_loss=0.2084, loss=0.1785, over 15498.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008439, over 36.00 utterances.], tot_loss[ctc_loss=0.08962, att_loss=0.2433, loss=0.2126, over 3258389.73 frames. utt_duration=1253 frames, utt_pad_proportion=0.05617, over 10416.32 utterances.], batch size: 36, lr: 7.80e-03, grad_scale: 8.0 2023-03-08 09:58:41,979 INFO [train2.py:809] (3/4) Epoch 14, batch 1400, loss[ctc_loss=0.07756, att_loss=0.2405, loss=0.2079, over 16464.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006825, over 46.00 utterances.], tot_loss[ctc_loss=0.09003, att_loss=0.2438, loss=0.213, over 3262784.10 frames. utt_duration=1249 frames, utt_pad_proportion=0.05645, over 10462.52 utterances.], batch size: 46, lr: 7.80e-03, grad_scale: 8.0 2023-03-08 09:59:16,380 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 09:59:41,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.260e+02 2.643e+02 3.236e+02 5.232e+02, threshold=5.286e+02, percent-clipped=1.0 2023-03-08 10:00:01,541 INFO [train2.py:809] (3/4) Epoch 14, batch 1450, loss[ctc_loss=0.08197, att_loss=0.2471, loss=0.214, over 16890.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006333, over 49.00 utterances.], tot_loss[ctc_loss=0.09001, att_loss=0.244, loss=0.2132, over 3260042.21 frames. utt_duration=1220 frames, utt_pad_proportion=0.06485, over 10699.21 utterances.], batch size: 49, lr: 7.80e-03, grad_scale: 8.0 2023-03-08 10:00:03,450 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:00:26,138 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2554, 4.6379, 4.8173, 4.7361, 2.6379, 4.9496, 2.7587, 1.7229], device='cuda:3'), covar=tensor([0.0321, 0.0237, 0.0584, 0.0177, 0.1842, 0.0127, 0.1609, 0.1789], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0128, 0.0250, 0.0120, 0.0219, 0.0113, 0.0223, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 10:00:31,932 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 10:00:35,790 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9035, 4.9530, 4.7108, 2.7514, 4.7448, 4.5275, 4.1567, 2.5553], device='cuda:3'), covar=tensor([0.0144, 0.0079, 0.0266, 0.1043, 0.0088, 0.0209, 0.0334, 0.1435], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0089, 0.0085, 0.0106, 0.0075, 0.0099, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:01:21,618 INFO [train2.py:809] (3/4) Epoch 14, batch 1500, loss[ctc_loss=0.1031, att_loss=0.2513, loss=0.2216, over 16630.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005095, over 47.00 utterances.], tot_loss[ctc_loss=0.09122, att_loss=0.245, loss=0.2142, over 3271538.76 frames. utt_duration=1211 frames, utt_pad_proportion=0.06404, over 10822.12 utterances.], batch size: 47, lr: 7.79e-03, grad_scale: 8.0 2023-03-08 10:01:41,567 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:01:44,589 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:02:02,608 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:02:20,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.404e+02 2.202e+02 2.545e+02 3.206e+02 8.434e+02, threshold=5.090e+02, percent-clipped=3.0 2023-03-08 10:02:41,148 INFO [train2.py:809] (3/4) Epoch 14, batch 1550, loss[ctc_loss=0.07665, att_loss=0.2393, loss=0.2068, over 16677.00 frames. utt_duration=1451 frames, utt_pad_proportion=0.006636, over 46.00 utterances.], tot_loss[ctc_loss=0.09109, att_loss=0.2451, loss=0.2143, over 3273416.71 frames. utt_duration=1214 frames, utt_pad_proportion=0.06328, over 10798.00 utterances.], batch size: 46, lr: 7.79e-03, grad_scale: 8.0 2023-03-08 10:02:47,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-08 10:03:15,045 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7253, 5.1275, 5.0008, 5.0610, 5.2248, 4.8465, 3.6354, 5.1532], device='cuda:3'), covar=tensor([0.0112, 0.0102, 0.0112, 0.0097, 0.0081, 0.0095, 0.0653, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0076, 0.0094, 0.0059, 0.0064, 0.0075, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:03:21,770 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:03:24,998 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:03:26,314 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1371, 5.4508, 4.8659, 5.3121, 5.0592, 4.7649, 4.8904, 4.7542], device='cuda:3'), covar=tensor([0.1458, 0.0948, 0.0990, 0.0779, 0.0958, 0.1501, 0.2522, 0.2263], device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0548, 0.0411, 0.0403, 0.0389, 0.0442, 0.0562, 0.0498], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:03:38,803 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:03:59,970 INFO [train2.py:809] (3/4) Epoch 14, batch 1600, loss[ctc_loss=0.08987, att_loss=0.2452, loss=0.2141, over 16397.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.006461, over 44.00 utterances.], tot_loss[ctc_loss=0.09148, att_loss=0.2453, loss=0.2145, over 3271401.60 frames. utt_duration=1212 frames, utt_pad_proportion=0.06492, over 10810.60 utterances.], batch size: 44, lr: 7.78e-03, grad_scale: 8.0 2023-03-08 10:04:29,546 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:04:36,029 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:04:58,718 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.287e+02 2.681e+02 3.372e+02 7.467e+02, threshold=5.361e+02, percent-clipped=4.0 2023-03-08 10:05:00,623 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 10:05:19,366 INFO [train2.py:809] (3/4) Epoch 14, batch 1650, loss[ctc_loss=0.08496, att_loss=0.2226, loss=0.1951, over 15349.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01247, over 35.00 utterances.], tot_loss[ctc_loss=0.09116, att_loss=0.2441, loss=0.2135, over 3258297.67 frames. utt_duration=1240 frames, utt_pad_proportion=0.06014, over 10525.15 utterances.], batch size: 35, lr: 7.78e-03, grad_scale: 8.0 2023-03-08 10:05:44,649 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:06:19,333 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-03-08 10:06:38,748 INFO [train2.py:809] (3/4) Epoch 14, batch 1700, loss[ctc_loss=0.1097, att_loss=0.2564, loss=0.227, over 17264.00 frames. utt_duration=1257 frames, utt_pad_proportion=0.0133, over 55.00 utterances.], tot_loss[ctc_loss=0.09162, att_loss=0.2449, loss=0.2142, over 3267625.07 frames. utt_duration=1211 frames, utt_pad_proportion=0.06457, over 10807.22 utterances.], batch size: 55, lr: 7.78e-03, grad_scale: 8.0 2023-03-08 10:06:46,851 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1818, 2.0013, 2.2572, 2.6717, 2.6684, 2.1570, 2.3278, 3.2082], device='cuda:3'), covar=tensor([0.1457, 0.5106, 0.3702, 0.1800, 0.1709, 0.2635, 0.3795, 0.1070], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0094, 0.0095, 0.0080, 0.0085, 0.0076, 0.0098, 0.0068], device='cuda:3'), out_proj_covar=tensor([6.0334e-05, 6.7791e-05, 7.0111e-05, 5.9183e-05, 5.9600e-05, 5.8404e-05, 6.9002e-05, 5.2524e-05], device='cuda:3') 2023-03-08 10:07:31,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-08 10:07:37,131 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.130e+02 2.657e+02 3.171e+02 6.492e+02, threshold=5.314e+02, percent-clipped=1.0 2023-03-08 10:07:39,037 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9006, 3.6872, 4.0085, 3.7250, 3.9284, 4.9436, 4.7272, 3.9288], device='cuda:3'), covar=tensor([0.0292, 0.1248, 0.0973, 0.1021, 0.0936, 0.0721, 0.0622, 0.0916], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0234, 0.0256, 0.0207, 0.0249, 0.0329, 0.0235, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:07:57,988 INFO [train2.py:809] (3/4) Epoch 14, batch 1750, loss[ctc_loss=0.1137, att_loss=0.2504, loss=0.223, over 16413.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.0062, over 44.00 utterances.], tot_loss[ctc_loss=0.09031, att_loss=0.2443, loss=0.2135, over 3263037.39 frames. utt_duration=1226 frames, utt_pad_proportion=0.06137, over 10656.27 utterances.], batch size: 44, lr: 7.77e-03, grad_scale: 8.0 2023-03-08 10:08:18,945 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9260, 3.6770, 3.1648, 3.4235, 3.8275, 3.5951, 2.9089, 4.2472], device='cuda:3'), covar=tensor([0.1158, 0.0548, 0.1105, 0.0782, 0.0866, 0.0742, 0.0906, 0.0496], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0197, 0.0213, 0.0186, 0.0253, 0.0223, 0.0191, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 10:09:17,126 INFO [train2.py:809] (3/4) Epoch 14, batch 1800, loss[ctc_loss=0.09419, att_loss=0.2454, loss=0.2151, over 16958.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008006, over 50.00 utterances.], tot_loss[ctc_loss=0.09111, att_loss=0.2445, loss=0.2138, over 3267279.62 frames. utt_duration=1216 frames, utt_pad_proportion=0.06326, over 10759.76 utterances.], batch size: 50, lr: 7.77e-03, grad_scale: 8.0 2023-03-08 10:09:18,864 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6969, 5.0219, 4.9507, 4.8800, 5.0693, 4.7864, 3.3614, 4.9596], device='cuda:3'), covar=tensor([0.0090, 0.0110, 0.0097, 0.0086, 0.0087, 0.0098, 0.0702, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0076, 0.0094, 0.0058, 0.0064, 0.0075, 0.0095, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:09:28,927 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:09:34,116 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5992, 2.4560, 5.0059, 3.9988, 2.9818, 4.1823, 4.9061, 4.7077], device='cuda:3'), covar=tensor([0.0234, 0.1734, 0.0179, 0.0944, 0.1889, 0.0249, 0.0117, 0.0208], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0239, 0.0148, 0.0298, 0.0261, 0.0186, 0.0132, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:10:17,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.192e+02 2.789e+02 3.495e+02 6.708e+02, threshold=5.578e+02, percent-clipped=5.0 2023-03-08 10:10:20,923 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-03-08 10:10:39,127 INFO [train2.py:809] (3/4) Epoch 14, batch 1850, loss[ctc_loss=0.07908, att_loss=0.2453, loss=0.212, over 17131.00 frames. utt_duration=868.8 frames, utt_pad_proportion=0.09024, over 79.00 utterances.], tot_loss[ctc_loss=0.09093, att_loss=0.2445, loss=0.2138, over 3273836.01 frames. utt_duration=1217 frames, utt_pad_proportion=0.06047, over 10777.79 utterances.], batch size: 79, lr: 7.77e-03, grad_scale: 8.0 2023-03-08 10:11:12,783 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:11:14,262 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0679, 5.3487, 5.2955, 5.2603, 5.4058, 5.3423, 5.0774, 4.8211], device='cuda:3'), covar=tensor([0.0932, 0.0404, 0.0249, 0.0425, 0.0237, 0.0274, 0.0303, 0.0313], device='cuda:3'), in_proj_covar=tensor([0.0469, 0.0311, 0.0277, 0.0309, 0.0359, 0.0374, 0.0306, 0.0347], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 10:11:29,359 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:11:59,261 INFO [train2.py:809] (3/4) Epoch 14, batch 1900, loss[ctc_loss=0.09547, att_loss=0.2663, loss=0.2321, over 17293.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01245, over 55.00 utterances.], tot_loss[ctc_loss=0.09113, att_loss=0.2443, loss=0.2136, over 3270828.75 frames. utt_duration=1220 frames, utt_pad_proportion=0.06088, over 10738.12 utterances.], batch size: 55, lr: 7.76e-03, grad_scale: 8.0 2023-03-08 10:12:26,464 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 10:12:35,625 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:12:45,535 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 10:12:52,354 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 10:12:58,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 2.093e+02 2.485e+02 3.000e+02 6.734e+02, threshold=4.970e+02, percent-clipped=1.0 2023-03-08 10:13:18,791 INFO [train2.py:809] (3/4) Epoch 14, batch 1950, loss[ctc_loss=0.09621, att_loss=0.258, loss=0.2256, over 17449.00 frames. utt_duration=1013 frames, utt_pad_proportion=0.04523, over 69.00 utterances.], tot_loss[ctc_loss=0.09044, att_loss=0.2442, loss=0.2134, over 3269663.06 frames. utt_duration=1215 frames, utt_pad_proportion=0.06274, over 10775.44 utterances.], batch size: 69, lr: 7.76e-03, grad_scale: 8.0 2023-03-08 10:13:40,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-08 10:13:50,783 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:14:37,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2023-03-08 10:14:38,395 INFO [train2.py:809] (3/4) Epoch 14, batch 2000, loss[ctc_loss=0.1291, att_loss=0.2695, loss=0.2414, over 14142.00 frames. utt_duration=388.9 frames, utt_pad_proportion=0.3213, over 146.00 utterances.], tot_loss[ctc_loss=0.08982, att_loss=0.2428, loss=0.2122, over 3260233.93 frames. utt_duration=1225 frames, utt_pad_proportion=0.06335, over 10659.79 utterances.], batch size: 146, lr: 7.76e-03, grad_scale: 8.0 2023-03-08 10:15:06,197 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5754, 2.9624, 3.6542, 2.9728, 3.6104, 4.7561, 4.5719, 3.1695], device='cuda:3'), covar=tensor([0.0389, 0.1917, 0.1149, 0.1478, 0.1097, 0.0900, 0.0487, 0.1550], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0232, 0.0253, 0.0205, 0.0246, 0.0328, 0.0233, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:15:36,777 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.201e+02 2.632e+02 3.371e+02 1.571e+03, threshold=5.264e+02, percent-clipped=3.0 2023-03-08 10:15:50,096 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0778, 6.2075, 5.7481, 5.9656, 5.9150, 5.5056, 5.6786, 5.4441], device='cuda:3'), covar=tensor([0.0973, 0.0861, 0.0737, 0.0711, 0.0709, 0.1221, 0.2188, 0.2250], device='cuda:3'), in_proj_covar=tensor([0.0469, 0.0540, 0.0407, 0.0399, 0.0384, 0.0437, 0.0554, 0.0488], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 10:15:57,729 INFO [train2.py:809] (3/4) Epoch 14, batch 2050, loss[ctc_loss=0.07988, att_loss=0.2453, loss=0.2123, over 16858.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008228, over 49.00 utterances.], tot_loss[ctc_loss=0.09012, att_loss=0.243, loss=0.2124, over 3258226.25 frames. utt_duration=1227 frames, utt_pad_proportion=0.06285, over 10630.72 utterances.], batch size: 49, lr: 7.75e-03, grad_scale: 8.0 2023-03-08 10:16:09,514 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5975, 5.0440, 4.8113, 4.9777, 5.0849, 4.7179, 3.6306, 4.9033], device='cuda:3'), covar=tensor([0.0112, 0.0110, 0.0128, 0.0074, 0.0085, 0.0130, 0.0666, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0076, 0.0095, 0.0059, 0.0065, 0.0076, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:16:17,492 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4220, 3.4523, 3.2932, 2.8725, 3.3752, 3.3741, 3.3999, 2.2079], device='cuda:3'), covar=tensor([0.1221, 0.1391, 0.2612, 0.5491, 0.2540, 0.5517, 0.1233, 0.6908], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0139, 0.0147, 0.0220, 0.0116, 0.0206, 0.0126, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:16:53,340 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0870, 5.1542, 4.9475, 2.9836, 4.9244, 4.8359, 4.3383, 2.5992], device='cuda:3'), covar=tensor([0.0123, 0.0085, 0.0269, 0.1094, 0.0091, 0.0165, 0.0345, 0.1523], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0090, 0.0086, 0.0106, 0.0075, 0.0100, 0.0095, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:17:18,540 INFO [train2.py:809] (3/4) Epoch 14, batch 2100, loss[ctc_loss=0.0765, att_loss=0.236, loss=0.2041, over 16132.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005205, over 42.00 utterances.], tot_loss[ctc_loss=0.08981, att_loss=0.2429, loss=0.2123, over 3263131.82 frames. utt_duration=1255 frames, utt_pad_proportion=0.05455, over 10415.34 utterances.], batch size: 42, lr: 7.75e-03, grad_scale: 8.0 2023-03-08 10:17:30,233 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:17:57,755 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1703, 4.5760, 4.7322, 4.6980, 2.7595, 4.7741, 2.8903, 1.9705], device='cuda:3'), covar=tensor([0.0371, 0.0208, 0.0563, 0.0148, 0.1799, 0.0138, 0.1356, 0.1646], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0132, 0.0256, 0.0121, 0.0224, 0.0117, 0.0227, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 10:18:17,164 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.088e+02 2.471e+02 3.126e+02 7.099e+02, threshold=4.942e+02, percent-clipped=2.0 2023-03-08 10:18:37,841 INFO [train2.py:809] (3/4) Epoch 14, batch 2150, loss[ctc_loss=0.09827, att_loss=0.2594, loss=0.2272, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008137, over 51.00 utterances.], tot_loss[ctc_loss=0.0903, att_loss=0.2437, loss=0.213, over 3268670.62 frames. utt_duration=1239 frames, utt_pad_proportion=0.05685, over 10565.15 utterances.], batch size: 51, lr: 7.75e-03, grad_scale: 8.0 2023-03-08 10:18:46,264 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:19:10,086 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:19:27,884 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:19:57,017 INFO [train2.py:809] (3/4) Epoch 14, batch 2200, loss[ctc_loss=0.08564, att_loss=0.2264, loss=0.1982, over 15894.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.00877, over 39.00 utterances.], tot_loss[ctc_loss=0.09077, att_loss=0.2439, loss=0.2133, over 3257383.77 frames. utt_duration=1240 frames, utt_pad_proportion=0.05812, over 10517.03 utterances.], batch size: 39, lr: 7.74e-03, grad_scale: 8.0 2023-03-08 10:20:30,048 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:20:47,496 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:20:53,813 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 10:20:59,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.188e+02 2.816e+02 3.562e+02 5.650e+02, threshold=5.631e+02, percent-clipped=6.0 2023-03-08 10:21:05,633 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0538, 5.3456, 5.2758, 5.2342, 5.4104, 5.3478, 5.0861, 4.7829], device='cuda:3'), covar=tensor([0.1013, 0.0470, 0.0268, 0.0484, 0.0250, 0.0290, 0.0275, 0.0326], device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0312, 0.0279, 0.0309, 0.0363, 0.0379, 0.0309, 0.0349], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 10:21:07,854 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7848, 5.9933, 5.4639, 5.7908, 5.6197, 5.3105, 5.4334, 5.2211], device='cuda:3'), covar=tensor([0.1149, 0.0920, 0.0872, 0.0769, 0.0847, 0.1450, 0.2349, 0.2401], device='cuda:3'), in_proj_covar=tensor([0.0470, 0.0539, 0.0406, 0.0402, 0.0385, 0.0438, 0.0554, 0.0490], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 10:21:20,209 INFO [train2.py:809] (3/4) Epoch 14, batch 2250, loss[ctc_loss=0.0752, att_loss=0.2448, loss=0.2109, over 16472.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006955, over 46.00 utterances.], tot_loss[ctc_loss=0.09092, att_loss=0.2442, loss=0.2135, over 3257001.67 frames. utt_duration=1243 frames, utt_pad_proportion=0.05824, over 10489.51 utterances.], batch size: 46, lr: 7.74e-03, grad_scale: 8.0 2023-03-08 10:22:09,527 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:22:39,619 INFO [train2.py:809] (3/4) Epoch 14, batch 2300, loss[ctc_loss=0.08057, att_loss=0.2344, loss=0.2036, over 16008.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.00697, over 40.00 utterances.], tot_loss[ctc_loss=0.09082, att_loss=0.2442, loss=0.2135, over 3263744.55 frames. utt_duration=1252 frames, utt_pad_proportion=0.05486, over 10438.71 utterances.], batch size: 40, lr: 7.73e-03, grad_scale: 8.0 2023-03-08 10:23:36,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.251e+02 2.657e+02 3.358e+02 7.808e+02, threshold=5.315e+02, percent-clipped=1.0 2023-03-08 10:23:57,483 INFO [train2.py:809] (3/4) Epoch 14, batch 2350, loss[ctc_loss=0.103, att_loss=0.2613, loss=0.2296, over 16470.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005789, over 46.00 utterances.], tot_loss[ctc_loss=0.09136, att_loss=0.2454, loss=0.2146, over 3275285.31 frames. utt_duration=1257 frames, utt_pad_proportion=0.04972, over 10430.99 utterances.], batch size: 46, lr: 7.73e-03, grad_scale: 8.0 2023-03-08 10:24:13,721 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0173, 5.0244, 4.9858, 2.1638, 1.9429, 2.6938, 2.7735, 3.6663], device='cuda:3'), covar=tensor([0.0689, 0.0203, 0.0195, 0.4589, 0.5942, 0.2723, 0.2494, 0.1836], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0238, 0.0245, 0.0221, 0.0345, 0.0335, 0.0232, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.4880e-04, 8.8407e-05, 1.0499e-04, 9.6636e-05, 1.4711e-04, 1.3339e-04, 9.2405e-05, 1.4712e-04], device='cuda:3') 2023-03-08 10:25:16,438 INFO [train2.py:809] (3/4) Epoch 14, batch 2400, loss[ctc_loss=0.07958, att_loss=0.2553, loss=0.2201, over 16748.00 frames. utt_duration=678.1 frames, utt_pad_proportion=0.1491, over 99.00 utterances.], tot_loss[ctc_loss=0.0912, att_loss=0.2456, loss=0.2147, over 3282245.79 frames. utt_duration=1241 frames, utt_pad_proportion=0.05233, over 10590.23 utterances.], batch size: 99, lr: 7.73e-03, grad_scale: 8.0 2023-03-08 10:25:16,803 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3489, 3.4187, 3.3120, 2.8532, 3.3375, 3.4264, 3.3546, 2.2071], device='cuda:3'), covar=tensor([0.1470, 0.1616, 0.2680, 0.6364, 0.1931, 0.2945, 0.1471, 0.7097], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0140, 0.0149, 0.0221, 0.0117, 0.0207, 0.0128, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:26:05,755 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-03-08 10:26:15,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.443e+02 2.183e+02 2.531e+02 3.194e+02 7.343e+02, threshold=5.062e+02, percent-clipped=2.0 2023-03-08 10:26:15,722 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4338, 2.7667, 3.5700, 2.6379, 3.3165, 4.5041, 4.3348, 2.9124], device='cuda:3'), covar=tensor([0.0376, 0.2033, 0.1134, 0.1716, 0.1128, 0.0908, 0.0577, 0.1583], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0230, 0.0251, 0.0204, 0.0242, 0.0326, 0.0232, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:26:15,763 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4470, 3.4121, 3.3962, 2.8920, 3.4390, 3.4198, 3.4015, 2.2214], device='cuda:3'), covar=tensor([0.1127, 0.1863, 0.3193, 0.4814, 0.1740, 0.3708, 0.1172, 0.6533], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0140, 0.0148, 0.0219, 0.0116, 0.0205, 0.0127, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:26:35,973 INFO [train2.py:809] (3/4) Epoch 14, batch 2450, loss[ctc_loss=0.0732, att_loss=0.2284, loss=0.1974, over 15998.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.0079, over 40.00 utterances.], tot_loss[ctc_loss=0.09084, att_loss=0.2446, loss=0.2138, over 3274493.97 frames. utt_duration=1262 frames, utt_pad_proportion=0.04948, over 10388.65 utterances.], batch size: 40, lr: 7.72e-03, grad_scale: 8.0 2023-03-08 10:27:10,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 2023-03-08 10:27:27,621 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-08 10:27:54,973 INFO [train2.py:809] (3/4) Epoch 14, batch 2500, loss[ctc_loss=0.0835, att_loss=0.2456, loss=0.2132, over 16955.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008196, over 50.00 utterances.], tot_loss[ctc_loss=0.09156, att_loss=0.2453, loss=0.2146, over 3277074.25 frames. utt_duration=1265 frames, utt_pad_proportion=0.04899, over 10371.87 utterances.], batch size: 50, lr: 7.72e-03, grad_scale: 8.0 2023-03-08 10:28:24,403 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5659, 2.9970, 3.6928, 2.9765, 3.4709, 4.6421, 4.3499, 3.1210], device='cuda:3'), covar=tensor([0.0349, 0.1579, 0.0990, 0.1329, 0.1033, 0.0596, 0.0567, 0.1288], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0231, 0.0252, 0.0204, 0.0243, 0.0327, 0.0233, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:28:53,991 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.435e+02 2.776e+02 3.474e+02 9.434e+02, threshold=5.553e+02, percent-clipped=7.0 2023-03-08 10:29:14,826 INFO [train2.py:809] (3/4) Epoch 14, batch 2550, loss[ctc_loss=0.0827, att_loss=0.2352, loss=0.2047, over 16555.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005387, over 45.00 utterances.], tot_loss[ctc_loss=0.09119, att_loss=0.2448, loss=0.2141, over 3283483.23 frames. utt_duration=1285 frames, utt_pad_proportion=0.04298, over 10233.70 utterances.], batch size: 45, lr: 7.72e-03, grad_scale: 8.0 2023-03-08 10:30:37,348 INFO [train2.py:809] (3/4) Epoch 14, batch 2600, loss[ctc_loss=0.07008, att_loss=0.2359, loss=0.2028, over 17018.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007904, over 51.00 utterances.], tot_loss[ctc_loss=0.09022, att_loss=0.2444, loss=0.2136, over 3284020.42 frames. utt_duration=1278 frames, utt_pad_proportion=0.0455, over 10291.50 utterances.], batch size: 51, lr: 7.71e-03, grad_scale: 8.0 2023-03-08 10:30:53,706 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.41 vs. limit=5.0 2023-03-08 10:31:39,668 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.230e+02 2.611e+02 3.212e+02 5.306e+02, threshold=5.223e+02, percent-clipped=0.0 2023-03-08 10:31:46,329 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-08 10:32:01,196 INFO [train2.py:809] (3/4) Epoch 14, batch 2650, loss[ctc_loss=0.1116, att_loss=0.2455, loss=0.2187, over 16675.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.007078, over 46.00 utterances.], tot_loss[ctc_loss=0.09009, att_loss=0.2438, loss=0.2131, over 3278437.32 frames. utt_duration=1264 frames, utt_pad_proportion=0.04772, over 10383.96 utterances.], batch size: 46, lr: 7.71e-03, grad_scale: 8.0 2023-03-08 10:33:23,996 INFO [train2.py:809] (3/4) Epoch 14, batch 2700, loss[ctc_loss=0.07052, att_loss=0.215, loss=0.1861, over 15769.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008604, over 38.00 utterances.], tot_loss[ctc_loss=0.08988, att_loss=0.2438, loss=0.213, over 3282067.19 frames. utt_duration=1259 frames, utt_pad_proportion=0.04892, over 10441.60 utterances.], batch size: 38, lr: 7.71e-03, grad_scale: 8.0 2023-03-08 10:34:25,552 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.188e+02 2.730e+02 3.256e+02 6.934e+02, threshold=5.460e+02, percent-clipped=5.0 2023-03-08 10:34:47,225 INFO [train2.py:809] (3/4) Epoch 14, batch 2750, loss[ctc_loss=0.08795, att_loss=0.2398, loss=0.2095, over 16123.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.005775, over 42.00 utterances.], tot_loss[ctc_loss=0.08881, att_loss=0.2428, loss=0.212, over 3281935.90 frames. utt_duration=1279 frames, utt_pad_proportion=0.04395, over 10276.81 utterances.], batch size: 42, lr: 7.70e-03, grad_scale: 8.0 2023-03-08 10:34:55,712 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4622, 4.8899, 4.7207, 4.7893, 4.9910, 4.6159, 3.4513, 4.8022], device='cuda:3'), covar=tensor([0.0107, 0.0104, 0.0119, 0.0082, 0.0070, 0.0105, 0.0667, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0076, 0.0093, 0.0058, 0.0063, 0.0073, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:35:14,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-08 10:36:10,904 INFO [train2.py:809] (3/4) Epoch 14, batch 2800, loss[ctc_loss=0.06609, att_loss=0.2353, loss=0.2015, over 16472.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006418, over 46.00 utterances.], tot_loss[ctc_loss=0.08914, att_loss=0.2432, loss=0.2124, over 3278166.40 frames. utt_duration=1250 frames, utt_pad_proportion=0.05177, over 10502.49 utterances.], batch size: 46, lr: 7.70e-03, grad_scale: 8.0 2023-03-08 10:37:12,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.211e+02 2.653e+02 3.453e+02 1.073e+03, threshold=5.307e+02, percent-clipped=4.0 2023-03-08 10:37:34,794 INFO [train2.py:809] (3/4) Epoch 14, batch 2850, loss[ctc_loss=0.0734, att_loss=0.2352, loss=0.2028, over 16541.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005631, over 45.00 utterances.], tot_loss[ctc_loss=0.08909, att_loss=0.2434, loss=0.2125, over 3269776.40 frames. utt_duration=1234 frames, utt_pad_proportion=0.05888, over 10615.81 utterances.], batch size: 45, lr: 7.70e-03, grad_scale: 8.0 2023-03-08 10:37:45,330 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 10:38:12,934 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1941, 5.4673, 5.3888, 5.3792, 5.5232, 5.5190, 5.2100, 4.9745], device='cuda:3'), covar=tensor([0.0864, 0.0466, 0.0258, 0.0479, 0.0234, 0.0242, 0.0295, 0.0298], device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0319, 0.0285, 0.0315, 0.0370, 0.0391, 0.0316, 0.0353], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 10:38:14,671 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 10:38:57,457 INFO [train2.py:809] (3/4) Epoch 14, batch 2900, loss[ctc_loss=0.08333, att_loss=0.2283, loss=0.1993, over 16386.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007803, over 44.00 utterances.], tot_loss[ctc_loss=0.09009, att_loss=0.2439, loss=0.2131, over 3262470.09 frames. utt_duration=1228 frames, utt_pad_proportion=0.06213, over 10644.15 utterances.], batch size: 44, lr: 7.69e-03, grad_scale: 8.0 2023-03-08 10:39:40,749 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4541, 2.6740, 3.5593, 4.3758, 3.9362, 3.9530, 2.9569, 2.0572], device='cuda:3'), covar=tensor([0.0695, 0.2356, 0.0874, 0.0563, 0.0682, 0.0514, 0.1583, 0.2425], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0210, 0.0188, 0.0199, 0.0200, 0.0164, 0.0198, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 10:39:55,661 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 10:39:58,424 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.074e+02 2.453e+02 3.116e+02 5.349e+02, threshold=4.905e+02, percent-clipped=1.0 2023-03-08 10:40:02,130 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4048, 2.2783, 2.3469, 2.4155, 2.4969, 2.4092, 2.3766, 2.8284], device='cuda:3'), covar=tensor([0.2189, 0.3489, 0.2507, 0.1292, 0.1689, 0.1299, 0.2444, 0.1048], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0093, 0.0095, 0.0080, 0.0085, 0.0077, 0.0099, 0.0069], device='cuda:3'), out_proj_covar=tensor([6.1814e-05, 6.7860e-05, 7.0672e-05, 5.9258e-05, 6.0307e-05, 5.8889e-05, 6.9972e-05, 5.3526e-05], device='cuda:3') 2023-03-08 10:40:19,747 INFO [train2.py:809] (3/4) Epoch 14, batch 2950, loss[ctc_loss=0.1139, att_loss=0.2662, loss=0.2358, over 17347.00 frames. utt_duration=879.8 frames, utt_pad_proportion=0.07873, over 79.00 utterances.], tot_loss[ctc_loss=0.09115, att_loss=0.2448, loss=0.214, over 3271366.77 frames. utt_duration=1217 frames, utt_pad_proportion=0.06194, over 10763.05 utterances.], batch size: 79, lr: 7.69e-03, grad_scale: 8.0 2023-03-08 10:41:11,149 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2457, 5.2319, 5.1362, 2.4708, 2.1045, 3.1476, 2.6924, 3.9178], device='cuda:3'), covar=tensor([0.0644, 0.0275, 0.0198, 0.4891, 0.6281, 0.2378, 0.2929, 0.1742], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0235, 0.0243, 0.0222, 0.0344, 0.0334, 0.0233, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4879e-04, 8.7100e-05, 1.0474e-04, 9.6950e-05, 1.4690e-04, 1.3258e-04, 9.2725e-05, 1.4679e-04], device='cuda:3') 2023-03-08 10:41:19,147 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0915, 4.3194, 4.1967, 4.4780, 2.8024, 4.6436, 2.6247, 1.8009], device='cuda:3'), covar=tensor([0.0356, 0.0195, 0.0707, 0.0154, 0.1592, 0.0121, 0.1486, 0.1694], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0131, 0.0250, 0.0120, 0.0218, 0.0114, 0.0222, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 10:41:41,654 INFO [train2.py:809] (3/4) Epoch 14, batch 3000, loss[ctc_loss=0.1036, att_loss=0.2547, loss=0.2244, over 16480.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005966, over 46.00 utterances.], tot_loss[ctc_loss=0.09098, att_loss=0.2449, loss=0.2141, over 3280834.12 frames. utt_duration=1225 frames, utt_pad_proportion=0.05793, over 10724.74 utterances.], batch size: 46, lr: 7.69e-03, grad_scale: 8.0 2023-03-08 10:41:41,654 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 10:41:56,406 INFO [train2.py:843] (3/4) Epoch 14, validation: ctc_loss=0.0441, att_loss=0.2368, loss=0.1983, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 10:41:56,407 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 10:42:43,724 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7190, 4.5188, 4.4673, 4.4959, 5.0847, 4.6820, 4.6072, 2.2667], device='cuda:3'), covar=tensor([0.0166, 0.0304, 0.0347, 0.0234, 0.0903, 0.0180, 0.0305, 0.2216], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0147, 0.0153, 0.0164, 0.0353, 0.0132, 0.0139, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:42:55,796 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 2.238e+02 2.720e+02 3.334e+02 7.498e+02, threshold=5.441e+02, percent-clipped=5.0 2023-03-08 10:43:16,812 INFO [train2.py:809] (3/4) Epoch 14, batch 3050, loss[ctc_loss=0.09864, att_loss=0.2385, loss=0.2105, over 16546.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005792, over 45.00 utterances.], tot_loss[ctc_loss=0.09078, att_loss=0.2441, loss=0.2134, over 3274654.54 frames. utt_duration=1253 frames, utt_pad_proportion=0.05184, over 10463.73 utterances.], batch size: 45, lr: 7.68e-03, grad_scale: 16.0 2023-03-08 10:43:21,263 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7331, 2.5448, 4.1053, 3.5434, 2.9475, 3.6881, 3.7034, 3.8285], device='cuda:3'), covar=tensor([0.0257, 0.1321, 0.0128, 0.0788, 0.1342, 0.0294, 0.0157, 0.0252], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0244, 0.0151, 0.0306, 0.0266, 0.0191, 0.0137, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:43:52,924 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:44:37,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8899, 6.1725, 5.6466, 5.9253, 5.8288, 5.4550, 5.6216, 5.3425], device='cuda:3'), covar=tensor([0.1317, 0.0945, 0.0872, 0.0829, 0.0802, 0.1367, 0.2300, 0.2603], device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0548, 0.0411, 0.0408, 0.0390, 0.0441, 0.0557, 0.0495], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 10:44:38,675 INFO [train2.py:809] (3/4) Epoch 14, batch 3100, loss[ctc_loss=0.07135, att_loss=0.2232, loss=0.1929, over 15772.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008416, over 38.00 utterances.], tot_loss[ctc_loss=0.09131, att_loss=0.2445, loss=0.2139, over 3276924.26 frames. utt_duration=1238 frames, utt_pad_proportion=0.05615, over 10604.91 utterances.], batch size: 38, lr: 7.68e-03, grad_scale: 16.0 2023-03-08 10:45:32,223 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 10:45:34,591 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 10:45:38,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.291e+02 2.720e+02 3.507e+02 7.867e+02, threshold=5.439e+02, percent-clipped=2.0 2023-03-08 10:45:41,992 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1098, 5.4122, 4.8356, 5.2445, 4.9525, 4.6351, 4.8655, 4.5962], device='cuda:3'), covar=tensor([0.1336, 0.1035, 0.0938, 0.0891, 0.1078, 0.1549, 0.2372, 0.2636], device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0539, 0.0406, 0.0403, 0.0385, 0.0437, 0.0551, 0.0488], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 10:45:59,420 INFO [train2.py:809] (3/4) Epoch 14, batch 3150, loss[ctc_loss=0.08594, att_loss=0.2164, loss=0.1903, over 14473.00 frames. utt_duration=1810 frames, utt_pad_proportion=0.04521, over 32.00 utterances.], tot_loss[ctc_loss=0.09151, att_loss=0.2451, loss=0.2144, over 3281463.83 frames. utt_duration=1228 frames, utt_pad_proportion=0.05709, over 10704.79 utterances.], batch size: 32, lr: 7.68e-03, grad_scale: 16.0 2023-03-08 10:46:31,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-03-08 10:46:34,653 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6598, 3.8885, 3.6506, 3.6386, 3.9459, 3.7545, 3.6239, 2.5682], device='cuda:3'), covar=tensor([0.0249, 0.0266, 0.0333, 0.0358, 0.0682, 0.0234, 0.0342, 0.1578], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0144, 0.0151, 0.0162, 0.0347, 0.0130, 0.0137, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:46:42,440 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7939, 3.6222, 2.9109, 3.2606, 3.8177, 3.4201, 2.6502, 4.0549], device='cuda:3'), covar=tensor([0.1230, 0.0556, 0.1266, 0.0800, 0.0779, 0.0797, 0.1072, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0195, 0.0210, 0.0183, 0.0249, 0.0220, 0.0188, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 10:47:15,321 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1181, 4.4913, 4.5618, 4.6783, 2.7493, 5.0676, 3.0755, 1.6826], device='cuda:3'), covar=tensor([0.0374, 0.0235, 0.0718, 0.0186, 0.1875, 0.0111, 0.1399, 0.1933], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0134, 0.0254, 0.0123, 0.0223, 0.0117, 0.0227, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 10:47:20,537 INFO [train2.py:809] (3/4) Epoch 14, batch 3200, loss[ctc_loss=0.0868, att_loss=0.2464, loss=0.2145, over 17406.00 frames. utt_duration=882.9 frames, utt_pad_proportion=0.07651, over 79.00 utterances.], tot_loss[ctc_loss=0.09029, att_loss=0.244, loss=0.2133, over 3278181.57 frames. utt_duration=1237 frames, utt_pad_proportion=0.05569, over 10617.55 utterances.], batch size: 79, lr: 7.67e-03, grad_scale: 16.0 2023-03-08 10:48:12,474 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 10:48:23,969 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 2.213e+02 2.608e+02 3.360e+02 8.145e+02, threshold=5.216e+02, percent-clipped=2.0 2023-03-08 10:48:46,134 INFO [train2.py:809] (3/4) Epoch 14, batch 3250, loss[ctc_loss=0.08874, att_loss=0.2267, loss=0.1991, over 16121.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006541, over 42.00 utterances.], tot_loss[ctc_loss=0.08959, att_loss=0.2432, loss=0.2125, over 3281135.53 frames. utt_duration=1254 frames, utt_pad_proportion=0.0502, over 10477.82 utterances.], batch size: 42, lr: 7.67e-03, grad_scale: 16.0 2023-03-08 10:48:56,999 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5151, 4.9130, 4.7432, 4.8467, 5.0180, 4.6049, 3.5755, 4.8953], device='cuda:3'), covar=tensor([0.0115, 0.0113, 0.0136, 0.0081, 0.0088, 0.0116, 0.0651, 0.0164], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0079, 0.0097, 0.0059, 0.0066, 0.0076, 0.0096, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:49:14,432 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7490, 2.5896, 4.0677, 3.5190, 3.0009, 3.7002, 3.7967, 3.8240], device='cuda:3'), covar=tensor([0.0269, 0.1403, 0.0138, 0.0765, 0.1422, 0.0308, 0.0183, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0250, 0.0154, 0.0312, 0.0271, 0.0194, 0.0141, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 10:49:34,328 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7401, 5.1578, 4.9846, 5.1255, 5.2618, 4.8419, 3.5499, 5.1590], device='cuda:3'), covar=tensor([0.0100, 0.0118, 0.0128, 0.0085, 0.0100, 0.0106, 0.0681, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0078, 0.0096, 0.0059, 0.0066, 0.0076, 0.0096, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:50:12,434 INFO [train2.py:809] (3/4) Epoch 14, batch 3300, loss[ctc_loss=0.09102, att_loss=0.2253, loss=0.1985, over 15487.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.008065, over 36.00 utterances.], tot_loss[ctc_loss=0.08932, att_loss=0.2423, loss=0.2117, over 3267375.77 frames. utt_duration=1250 frames, utt_pad_proportion=0.05539, over 10467.92 utterances.], batch size: 36, lr: 7.66e-03, grad_scale: 16.0 2023-03-08 10:51:15,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.260e+02 2.683e+02 3.437e+02 9.607e+02, threshold=5.365e+02, percent-clipped=9.0 2023-03-08 10:51:37,897 INFO [train2.py:809] (3/4) Epoch 14, batch 3350, loss[ctc_loss=0.06674, att_loss=0.2243, loss=0.1928, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007919, over 43.00 utterances.], tot_loss[ctc_loss=0.08904, att_loss=0.2422, loss=0.2116, over 3265491.62 frames. utt_duration=1261 frames, utt_pad_proportion=0.0532, over 10372.79 utterances.], batch size: 43, lr: 7.66e-03, grad_scale: 16.0 2023-03-08 10:51:51,544 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:52:16,342 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.83 vs. limit=5.0 2023-03-08 10:52:50,768 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-03-08 10:53:01,321 INFO [train2.py:809] (3/4) Epoch 14, batch 3400, loss[ctc_loss=0.06647, att_loss=0.2411, loss=0.2062, over 17018.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008713, over 51.00 utterances.], tot_loss[ctc_loss=0.08943, att_loss=0.2428, loss=0.2122, over 3269678.34 frames. utt_duration=1274 frames, utt_pad_proportion=0.0494, over 10275.72 utterances.], batch size: 51, lr: 7.66e-03, grad_scale: 16.0 2023-03-08 10:53:34,790 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:53:38,111 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:53:50,361 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 10:54:05,576 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.249e+02 2.663e+02 3.402e+02 7.529e+02, threshold=5.327e+02, percent-clipped=2.0 2023-03-08 10:54:27,351 INFO [train2.py:809] (3/4) Epoch 14, batch 3450, loss[ctc_loss=0.09116, att_loss=0.2268, loss=0.1997, over 15615.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.01005, over 37.00 utterances.], tot_loss[ctc_loss=0.09075, att_loss=0.2439, loss=0.2133, over 3268779.29 frames. utt_duration=1243 frames, utt_pad_proportion=0.05862, over 10531.71 utterances.], batch size: 37, lr: 7.65e-03, grad_scale: 16.0 2023-03-08 10:54:38,254 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-08 10:55:22,046 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 10:55:35,618 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1291, 5.4381, 4.9818, 5.4965, 4.9270, 5.0582, 5.6272, 5.3739], device='cuda:3'), covar=tensor([0.0575, 0.0264, 0.0727, 0.0279, 0.0347, 0.0197, 0.0209, 0.0173], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0282, 0.0334, 0.0293, 0.0285, 0.0219, 0.0275, 0.0253], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 10:55:52,480 INFO [train2.py:809] (3/4) Epoch 14, batch 3500, loss[ctc_loss=0.0825, att_loss=0.2327, loss=0.2026, over 16257.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007198, over 43.00 utterances.], tot_loss[ctc_loss=0.0908, att_loss=0.2436, loss=0.2131, over 3256087.36 frames. utt_duration=1225 frames, utt_pad_proportion=0.06753, over 10647.08 utterances.], batch size: 43, lr: 7.65e-03, grad_scale: 8.0 2023-03-08 10:56:37,821 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6167, 5.2015, 4.9849, 5.2205, 5.3047, 4.8223, 3.8261, 5.1507], device='cuda:3'), covar=tensor([0.0119, 0.0108, 0.0131, 0.0064, 0.0103, 0.0100, 0.0592, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0077, 0.0095, 0.0059, 0.0065, 0.0076, 0.0094, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 10:56:43,464 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 10:56:56,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.024e+02 2.386e+02 2.873e+02 9.037e+02, threshold=4.771e+02, percent-clipped=2.0 2023-03-08 10:57:17,394 INFO [train2.py:809] (3/4) Epoch 14, batch 3550, loss[ctc_loss=0.08006, att_loss=0.2363, loss=0.2051, over 16164.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007072, over 41.00 utterances.], tot_loss[ctc_loss=0.09066, att_loss=0.2436, loss=0.213, over 3257241.89 frames. utt_duration=1226 frames, utt_pad_proportion=0.06446, over 10639.42 utterances.], batch size: 41, lr: 7.65e-03, grad_scale: 8.0 2023-03-08 10:57:22,653 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1901, 5.1044, 4.9311, 2.8715, 4.9252, 4.8208, 4.2669, 2.8822], device='cuda:3'), covar=tensor([0.0087, 0.0090, 0.0218, 0.1064, 0.0086, 0.0181, 0.0347, 0.1292], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0090, 0.0088, 0.0108, 0.0076, 0.0102, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:58:05,153 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 10:58:41,910 INFO [train2.py:809] (3/4) Epoch 14, batch 3600, loss[ctc_loss=0.09241, att_loss=0.255, loss=0.2225, over 17307.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02344, over 59.00 utterances.], tot_loss[ctc_loss=0.08976, att_loss=0.2433, loss=0.2126, over 3263166.49 frames. utt_duration=1242 frames, utt_pad_proportion=0.05963, over 10524.01 utterances.], batch size: 59, lr: 7.64e-03, grad_scale: 8.0 2023-03-08 10:58:43,908 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1891, 5.0814, 4.9798, 2.9727, 4.9740, 4.8201, 4.2424, 2.8957], device='cuda:3'), covar=tensor([0.0093, 0.0098, 0.0222, 0.0952, 0.0078, 0.0152, 0.0329, 0.1164], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0091, 0.0088, 0.0107, 0.0076, 0.0102, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:59:41,918 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8668, 6.0789, 5.5776, 5.8256, 5.7258, 5.2634, 5.3925, 5.2925], device='cuda:3'), covar=tensor([0.1164, 0.0858, 0.0882, 0.0805, 0.0864, 0.1663, 0.2323, 0.2452], device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0549, 0.0418, 0.0409, 0.0395, 0.0445, 0.0567, 0.0499], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 10:59:46,533 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.040e+02 2.412e+02 2.968e+02 1.250e+03, threshold=4.823e+02, percent-clipped=3.0 2023-03-08 11:00:07,018 INFO [train2.py:809] (3/4) Epoch 14, batch 3650, loss[ctc_loss=0.1002, att_loss=0.2618, loss=0.2295, over 16898.00 frames. utt_duration=684.4 frames, utt_pad_proportion=0.1392, over 99.00 utterances.], tot_loss[ctc_loss=0.08944, att_loss=0.2435, loss=0.2127, over 3270943.98 frames. utt_duration=1241 frames, utt_pad_proportion=0.05671, over 10551.55 utterances.], batch size: 99, lr: 7.64e-03, grad_scale: 8.0 2023-03-08 11:00:09,075 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7026, 3.6096, 3.4306, 3.2290, 3.6381, 3.6802, 3.6797, 2.8294], device='cuda:3'), covar=tensor([0.1119, 0.2088, 0.3234, 0.4304, 0.1535, 0.3615, 0.0982, 0.5303], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0143, 0.0154, 0.0222, 0.0116, 0.0208, 0.0130, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:01:29,822 INFO [train2.py:809] (3/4) Epoch 14, batch 3700, loss[ctc_loss=0.07729, att_loss=0.2396, loss=0.2071, over 16617.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005872, over 47.00 utterances.], tot_loss[ctc_loss=0.08953, att_loss=0.2436, loss=0.2128, over 3268413.76 frames. utt_duration=1277 frames, utt_pad_proportion=0.04895, over 10252.05 utterances.], batch size: 47, lr: 7.64e-03, grad_scale: 8.0 2023-03-08 11:01:53,194 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:02:08,184 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4681, 5.0843, 5.1816, 2.9757, 2.7203, 3.2325, 2.9438, 4.0560], device='cuda:3'), covar=tensor([0.0731, 0.0560, 0.0282, 0.3523, 0.5113, 0.2755, 0.2862, 0.1943], device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0237, 0.0242, 0.0219, 0.0344, 0.0336, 0.0232, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4930e-04, 8.8695e-05, 1.0459e-04, 9.5595e-05, 1.4668e-04, 1.3330e-04, 9.2539e-05, 1.4691e-04], device='cuda:3') 2023-03-08 11:02:16,953 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 11:02:33,340 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.162e+02 2.668e+02 3.280e+02 7.175e+02, threshold=5.335e+02, percent-clipped=8.0 2023-03-08 11:02:43,171 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1782, 5.4341, 5.7290, 5.7008, 5.6419, 6.1175, 5.2829, 6.2603], device='cuda:3'), covar=tensor([0.0608, 0.0612, 0.0656, 0.1050, 0.1676, 0.0875, 0.0627, 0.0581], device='cuda:3'), in_proj_covar=tensor([0.0775, 0.0462, 0.0546, 0.0601, 0.0798, 0.0553, 0.0451, 0.0532], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 11:02:53,080 INFO [train2.py:809] (3/4) Epoch 14, batch 3750, loss[ctc_loss=0.06128, att_loss=0.2337, loss=0.1992, over 16863.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008501, over 49.00 utterances.], tot_loss[ctc_loss=0.08974, att_loss=0.2431, loss=0.2124, over 3257880.94 frames. utt_duration=1275 frames, utt_pad_proportion=0.05166, over 10233.65 utterances.], batch size: 49, lr: 7.63e-03, grad_scale: 8.0 2023-03-08 11:03:16,964 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3339, 3.7962, 3.1543, 3.6450, 4.0639, 3.6970, 2.9371, 4.4149], device='cuda:3'), covar=tensor([0.0828, 0.0440, 0.1035, 0.0602, 0.0578, 0.0681, 0.0882, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0199, 0.0212, 0.0189, 0.0255, 0.0226, 0.0192, 0.0272], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 11:03:36,240 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:03:37,998 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:03:38,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 2023-03-08 11:03:38,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-03-08 11:04:16,306 INFO [train2.py:809] (3/4) Epoch 14, batch 3800, loss[ctc_loss=0.09754, att_loss=0.2547, loss=0.2233, over 17339.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02154, over 59.00 utterances.], tot_loss[ctc_loss=0.09109, att_loss=0.2445, loss=0.2138, over 3264918.51 frames. utt_duration=1215 frames, utt_pad_proportion=0.06346, over 10761.99 utterances.], batch size: 59, lr: 7.63e-03, grad_scale: 8.0 2023-03-08 11:04:30,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-03-08 11:05:15,641 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9990, 3.8375, 3.2679, 3.6171, 4.0708, 3.7423, 2.9346, 4.4619], device='cuda:3'), covar=tensor([0.1000, 0.0438, 0.0924, 0.0590, 0.0594, 0.0578, 0.0840, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0197, 0.0210, 0.0186, 0.0252, 0.0224, 0.0190, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 11:05:20,070 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 2.212e+02 2.620e+02 3.323e+02 8.677e+02, threshold=5.239e+02, percent-clipped=3.0 2023-03-08 11:05:40,105 INFO [train2.py:809] (3/4) Epoch 14, batch 3850, loss[ctc_loss=0.06467, att_loss=0.2277, loss=0.1951, over 16275.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007584, over 43.00 utterances.], tot_loss[ctc_loss=0.09076, att_loss=0.2445, loss=0.2137, over 3264199.34 frames. utt_duration=1220 frames, utt_pad_proportion=0.06235, over 10718.21 utterances.], batch size: 43, lr: 7.63e-03, grad_scale: 8.0 2023-03-08 11:06:11,709 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-08 11:07:00,907 INFO [train2.py:809] (3/4) Epoch 14, batch 3900, loss[ctc_loss=0.08691, att_loss=0.2551, loss=0.2214, over 17331.00 frames. utt_duration=1006 frames, utt_pad_proportion=0.05167, over 69.00 utterances.], tot_loss[ctc_loss=0.09015, att_loss=0.2442, loss=0.2134, over 3259502.89 frames. utt_duration=1201 frames, utt_pad_proportion=0.06843, over 10867.15 utterances.], batch size: 69, lr: 7.62e-03, grad_scale: 8.0 2023-03-08 11:07:30,562 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4181, 2.7408, 3.6082, 3.0115, 3.5004, 4.5374, 4.3264, 3.3205], device='cuda:3'), covar=tensor([0.0346, 0.1896, 0.1122, 0.1219, 0.1000, 0.0852, 0.0603, 0.1148], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0237, 0.0260, 0.0209, 0.0250, 0.0338, 0.0239, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 11:07:31,246 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-08 11:08:01,631 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.280e+02 2.167e+02 2.705e+02 3.104e+02 1.128e+03, threshold=5.409e+02, percent-clipped=4.0 2023-03-08 11:08:20,151 INFO [train2.py:809] (3/4) Epoch 14, batch 3950, loss[ctc_loss=0.08831, att_loss=0.2463, loss=0.2147, over 16386.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.008493, over 44.00 utterances.], tot_loss[ctc_loss=0.08959, att_loss=0.2434, loss=0.2126, over 3257569.85 frames. utt_duration=1226 frames, utt_pad_proportion=0.06321, over 10644.28 utterances.], batch size: 44, lr: 7.62e-03, grad_scale: 8.0 2023-03-08 11:08:32,855 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:09:39,021 INFO [train2.py:809] (3/4) Epoch 15, batch 0, loss[ctc_loss=0.1087, att_loss=0.2316, loss=0.207, over 15894.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008708, over 39.00 utterances.], tot_loss[ctc_loss=0.1087, att_loss=0.2316, loss=0.207, over 15894.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008708, over 39.00 utterances.], batch size: 39, lr: 7.36e-03, grad_scale: 8.0 2023-03-08 11:09:39,021 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 11:09:51,761 INFO [train2.py:843] (3/4) Epoch 15, validation: ctc_loss=0.04404, att_loss=0.2365, loss=0.198, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 11:09:51,762 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 11:10:43,448 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:10:51,507 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:11:13,891 INFO [train2.py:809] (3/4) Epoch 15, batch 50, loss[ctc_loss=0.08388, att_loss=0.2544, loss=0.2203, over 16538.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006227, over 45.00 utterances.], tot_loss[ctc_loss=0.09164, att_loss=0.2461, loss=0.2152, over 742116.08 frames. utt_duration=1221 frames, utt_pad_proportion=0.05599, over 2434.17 utterances.], batch size: 45, lr: 7.35e-03, grad_scale: 8.0 2023-03-08 11:11:21,609 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 2.261e+02 2.719e+02 3.409e+02 7.287e+02, threshold=5.439e+02, percent-clipped=2.0 2023-03-08 11:11:55,455 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:12:01,410 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:12:25,428 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:12:29,338 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7692, 2.8286, 3.5397, 4.6214, 3.9793, 4.1507, 2.9616, 2.1102], device='cuda:3'), covar=tensor([0.0483, 0.2108, 0.0943, 0.0519, 0.0786, 0.0352, 0.1515, 0.2258], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0216, 0.0193, 0.0201, 0.0207, 0.0164, 0.0199, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 11:12:35,391 INFO [train2.py:809] (3/4) Epoch 15, batch 100, loss[ctc_loss=0.1048, att_loss=0.2656, loss=0.2335, over 17406.00 frames. utt_duration=882.7 frames, utt_pad_proportion=0.07566, over 79.00 utterances.], tot_loss[ctc_loss=0.09033, att_loss=0.2428, loss=0.2123, over 1283580.83 frames. utt_duration=1219 frames, utt_pad_proportion=0.06984, over 4215.73 utterances.], batch size: 79, lr: 7.35e-03, grad_scale: 8.0 2023-03-08 11:13:35,207 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:13:35,315 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0749, 5.0087, 4.9121, 2.0768, 1.8880, 2.7091, 2.4619, 3.8846], device='cuda:3'), covar=tensor([0.0751, 0.0205, 0.0238, 0.5369, 0.6156, 0.2844, 0.3325, 0.1628], device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0236, 0.0243, 0.0221, 0.0343, 0.0335, 0.0230, 0.0351], device='cuda:3'), out_proj_covar=tensor([1.4927e-04, 8.8498e-05, 1.0491e-04, 9.6678e-05, 1.4605e-04, 1.3301e-04, 9.2053e-05, 1.4582e-04], device='cuda:3') 2023-03-08 11:13:44,373 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:13:57,610 INFO [train2.py:809] (3/4) Epoch 15, batch 150, loss[ctc_loss=0.04577, att_loss=0.2218, loss=0.1866, over 15954.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006346, over 41.00 utterances.], tot_loss[ctc_loss=0.0896, att_loss=0.2429, loss=0.2122, over 1725898.90 frames. utt_duration=1252 frames, utt_pad_proportion=0.05779, over 5522.81 utterances.], batch size: 41, lr: 7.35e-03, grad_scale: 8.0 2023-03-08 11:14:05,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.292e+02 2.863e+02 3.453e+02 8.582e+02, threshold=5.726e+02, percent-clipped=4.0 2023-03-08 11:14:48,902 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5072, 5.0767, 4.8718, 4.9826, 5.1014, 4.7130, 3.8564, 5.0355], device='cuda:3'), covar=tensor([0.0128, 0.0104, 0.0140, 0.0084, 0.0082, 0.0125, 0.0572, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0078, 0.0096, 0.0060, 0.0066, 0.0077, 0.0095, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 11:15:19,360 INFO [train2.py:809] (3/4) Epoch 15, batch 200, loss[ctc_loss=0.07723, att_loss=0.2287, loss=0.1984, over 16010.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007848, over 40.00 utterances.], tot_loss[ctc_loss=0.08922, att_loss=0.2433, loss=0.2125, over 2063426.34 frames. utt_duration=1217 frames, utt_pad_proportion=0.06852, over 6787.81 utterances.], batch size: 40, lr: 7.34e-03, grad_scale: 8.0 2023-03-08 11:16:45,494 INFO [train2.py:809] (3/4) Epoch 15, batch 250, loss[ctc_loss=0.06882, att_loss=0.2421, loss=0.2074, over 16859.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008052, over 49.00 utterances.], tot_loss[ctc_loss=0.08884, att_loss=0.244, loss=0.213, over 2340966.30 frames. utt_duration=1229 frames, utt_pad_proportion=0.0601, over 7630.34 utterances.], batch size: 49, lr: 7.34e-03, grad_scale: 8.0 2023-03-08 11:16:53,155 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.127e+02 2.516e+02 2.865e+02 8.492e+02, threshold=5.032e+02, percent-clipped=2.0 2023-03-08 11:18:06,088 INFO [train2.py:809] (3/4) Epoch 15, batch 300, loss[ctc_loss=0.1209, att_loss=0.2628, loss=0.2344, over 14144.00 frames. utt_duration=389.2 frames, utt_pad_proportion=0.3232, over 146.00 utterances.], tot_loss[ctc_loss=0.08912, att_loss=0.244, loss=0.213, over 2543750.27 frames. utt_duration=1231 frames, utt_pad_proportion=0.06042, over 8278.64 utterances.], batch size: 146, lr: 7.34e-03, grad_scale: 8.0 2023-03-08 11:18:55,691 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:19:25,567 INFO [train2.py:809] (3/4) Epoch 15, batch 350, loss[ctc_loss=0.09122, att_loss=0.2481, loss=0.2168, over 17417.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04692, over 69.00 utterances.], tot_loss[ctc_loss=0.0904, att_loss=0.2446, loss=0.2138, over 2712762.89 frames. utt_duration=1205 frames, utt_pad_proportion=0.06405, over 9012.90 utterances.], batch size: 69, lr: 7.34e-03, grad_scale: 8.0 2023-03-08 11:19:25,796 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4777, 4.8753, 3.6163, 5.1208, 4.5375, 4.8564, 4.7008, 4.7298], device='cuda:3'), covar=tensor([0.0691, 0.0491, 0.1866, 0.0294, 0.0435, 0.0301, 0.0663, 0.0376], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0282, 0.0339, 0.0298, 0.0287, 0.0221, 0.0274, 0.0251], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 11:19:34,067 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.148e+02 2.622e+02 3.077e+02 6.565e+02, threshold=5.243e+02, percent-clipped=3.0 2023-03-08 11:20:01,864 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:20:45,055 INFO [train2.py:809] (3/4) Epoch 15, batch 400, loss[ctc_loss=0.1089, att_loss=0.253, loss=0.2242, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01087, over 55.00 utterances.], tot_loss[ctc_loss=0.09104, att_loss=0.2438, loss=0.2132, over 2832271.58 frames. utt_duration=1207 frames, utt_pad_proportion=0.06523, over 9397.87 utterances.], batch size: 55, lr: 7.33e-03, grad_scale: 8.0 2023-03-08 11:21:02,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2023-03-08 11:21:22,530 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1005, 4.5532, 4.1283, 4.5829, 2.6373, 4.5997, 2.3422, 2.0143], device='cuda:3'), covar=tensor([0.0330, 0.0161, 0.0834, 0.0161, 0.1933, 0.0137, 0.1883, 0.1817], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0134, 0.0254, 0.0124, 0.0220, 0.0114, 0.0226, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 11:21:34,936 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:21:35,502 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-08 11:21:38,228 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:22:04,493 INFO [train2.py:809] (3/4) Epoch 15, batch 450, loss[ctc_loss=0.09678, att_loss=0.2615, loss=0.2286, over 16764.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006614, over 48.00 utterances.], tot_loss[ctc_loss=0.09044, att_loss=0.2437, loss=0.213, over 2933250.09 frames. utt_duration=1217 frames, utt_pad_proportion=0.06094, over 9655.00 utterances.], batch size: 48, lr: 7.33e-03, grad_scale: 8.0 2023-03-08 11:22:12,114 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.121e+02 2.729e+02 3.274e+02 6.212e+02, threshold=5.457e+02, percent-clipped=2.0 2023-03-08 11:23:23,984 INFO [train2.py:809] (3/4) Epoch 15, batch 500, loss[ctc_loss=0.1225, att_loss=0.2652, loss=0.2367, over 17315.00 frames. utt_duration=1005 frames, utt_pad_proportion=0.05251, over 69.00 utterances.], tot_loss[ctc_loss=0.09043, att_loss=0.2435, loss=0.2129, over 3004176.33 frames. utt_duration=1190 frames, utt_pad_proportion=0.07057, over 10113.68 utterances.], batch size: 69, lr: 7.33e-03, grad_scale: 8.0 2023-03-08 11:24:43,548 INFO [train2.py:809] (3/4) Epoch 15, batch 550, loss[ctc_loss=0.0877, att_loss=0.2435, loss=0.2124, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008341, over 51.00 utterances.], tot_loss[ctc_loss=0.09052, att_loss=0.2434, loss=0.2129, over 3057192.33 frames. utt_duration=1188 frames, utt_pad_proportion=0.07235, over 10302.78 utterances.], batch size: 51, lr: 7.32e-03, grad_scale: 8.0 2023-03-08 11:24:51,298 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.199e+02 2.714e+02 3.543e+02 1.087e+03, threshold=5.428e+02, percent-clipped=4.0 2023-03-08 11:26:03,951 INFO [train2.py:809] (3/4) Epoch 15, batch 600, loss[ctc_loss=0.1036, att_loss=0.245, loss=0.2167, over 17060.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009398, over 53.00 utterances.], tot_loss[ctc_loss=0.09167, att_loss=0.2448, loss=0.2142, over 3108453.63 frames. utt_duration=1168 frames, utt_pad_proportion=0.07586, over 10660.30 utterances.], batch size: 53, lr: 7.32e-03, grad_scale: 8.0 2023-03-08 11:26:54,539 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:27:14,889 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1566, 5.4565, 5.7232, 5.6239, 5.6209, 6.1385, 5.3577, 6.2001], device='cuda:3'), covar=tensor([0.0568, 0.0684, 0.0578, 0.1081, 0.1626, 0.0688, 0.0556, 0.0535], device='cuda:3'), in_proj_covar=tensor([0.0776, 0.0462, 0.0539, 0.0601, 0.0792, 0.0548, 0.0446, 0.0536], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 11:27:23,614 INFO [train2.py:809] (3/4) Epoch 15, batch 650, loss[ctc_loss=0.08978, att_loss=0.2462, loss=0.2149, over 16276.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.006888, over 43.00 utterances.], tot_loss[ctc_loss=0.09104, att_loss=0.2439, loss=0.2134, over 3144655.47 frames. utt_duration=1220 frames, utt_pad_proportion=0.06356, over 10326.10 utterances.], batch size: 43, lr: 7.32e-03, grad_scale: 8.0 2023-03-08 11:27:31,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.098e+02 2.434e+02 3.232e+02 5.405e+02, threshold=4.867e+02, percent-clipped=0.0 2023-03-08 11:28:09,902 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:28:42,742 INFO [train2.py:809] (3/4) Epoch 15, batch 700, loss[ctc_loss=0.1095, att_loss=0.2625, loss=0.2319, over 17300.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.0112, over 55.00 utterances.], tot_loss[ctc_loss=0.09062, att_loss=0.2436, loss=0.213, over 3177528.58 frames. utt_duration=1255 frames, utt_pad_proportion=0.054, over 10138.39 utterances.], batch size: 55, lr: 7.31e-03, grad_scale: 8.0 2023-03-08 11:29:28,164 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:29:33,161 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:30:02,715 INFO [train2.py:809] (3/4) Epoch 15, batch 750, loss[ctc_loss=0.0638, att_loss=0.2165, loss=0.186, over 15513.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.007987, over 36.00 utterances.], tot_loss[ctc_loss=0.08945, att_loss=0.2426, loss=0.2119, over 3195976.21 frames. utt_duration=1281 frames, utt_pad_proportion=0.04777, over 9987.98 utterances.], batch size: 36, lr: 7.31e-03, grad_scale: 8.0 2023-03-08 11:30:03,067 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 11:30:07,844 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6155, 4.6272, 4.5429, 4.4939, 5.1133, 4.6948, 4.4980, 2.3680], device='cuda:3'), covar=tensor([0.0219, 0.0252, 0.0274, 0.0277, 0.0877, 0.0211, 0.0320, 0.1932], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0145, 0.0152, 0.0164, 0.0347, 0.0132, 0.0138, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:30:11,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.381e+02 2.975e+02 3.740e+02 8.505e+02, threshold=5.949e+02, percent-clipped=7.0 2023-03-08 11:30:49,605 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:31:03,225 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-08 11:31:22,459 INFO [train2.py:809] (3/4) Epoch 15, batch 800, loss[ctc_loss=0.1106, att_loss=0.2593, loss=0.2295, over 16999.00 frames. utt_duration=688.4 frames, utt_pad_proportion=0.1351, over 99.00 utterances.], tot_loss[ctc_loss=0.08922, att_loss=0.2429, loss=0.2121, over 3214922.94 frames. utt_duration=1284 frames, utt_pad_proportion=0.04616, over 10026.82 utterances.], batch size: 99, lr: 7.31e-03, grad_scale: 8.0 2023-03-08 11:31:42,523 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 11:32:44,276 INFO [train2.py:809] (3/4) Epoch 15, batch 850, loss[ctc_loss=0.06694, att_loss=0.2254, loss=0.1937, over 15753.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009515, over 38.00 utterances.], tot_loss[ctc_loss=0.08845, att_loss=0.2428, loss=0.212, over 3228987.66 frames. utt_duration=1290 frames, utt_pad_proportion=0.04459, over 10023.95 utterances.], batch size: 38, lr: 7.30e-03, grad_scale: 8.0 2023-03-08 11:32:52,930 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.144e+02 2.509e+02 3.065e+02 7.548e+02, threshold=5.017e+02, percent-clipped=1.0 2023-03-08 11:32:57,993 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7569, 4.7621, 4.6338, 4.5966, 5.1377, 4.8983, 4.6109, 2.4031], device='cuda:3'), covar=tensor([0.0201, 0.0198, 0.0234, 0.0248, 0.0904, 0.0178, 0.0260, 0.2012], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0147, 0.0153, 0.0165, 0.0352, 0.0134, 0.0139, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:34:04,334 INFO [train2.py:809] (3/4) Epoch 15, batch 900, loss[ctc_loss=0.1048, att_loss=0.2474, loss=0.2189, over 16945.00 frames. utt_duration=686.2 frames, utt_pad_proportion=0.139, over 99.00 utterances.], tot_loss[ctc_loss=0.08844, att_loss=0.2428, loss=0.2119, over 3238835.38 frames. utt_duration=1276 frames, utt_pad_proportion=0.04792, over 10161.70 utterances.], batch size: 99, lr: 7.30e-03, grad_scale: 8.0 2023-03-08 11:35:24,398 INFO [train2.py:809] (3/4) Epoch 15, batch 950, loss[ctc_loss=0.08915, att_loss=0.2583, loss=0.2245, over 17290.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02082, over 59.00 utterances.], tot_loss[ctc_loss=0.08939, att_loss=0.2438, loss=0.2129, over 3250072.07 frames. utt_duration=1265 frames, utt_pad_proportion=0.05008, over 10291.83 utterances.], batch size: 59, lr: 7.30e-03, grad_scale: 8.0 2023-03-08 11:35:25,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-08 11:35:32,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.300e+02 2.400e+02 2.758e+02 3.423e+02 1.509e+03, threshold=5.516e+02, percent-clipped=3.0 2023-03-08 11:36:44,605 INFO [train2.py:809] (3/4) Epoch 15, batch 1000, loss[ctc_loss=0.05971, att_loss=0.2156, loss=0.1844, over 15490.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.009456, over 36.00 utterances.], tot_loss[ctc_loss=0.08878, att_loss=0.2435, loss=0.2125, over 3255578.67 frames. utt_duration=1279 frames, utt_pad_proportion=0.04649, over 10197.55 utterances.], batch size: 36, lr: 7.29e-03, grad_scale: 8.0 2023-03-08 11:37:30,279 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:37:33,404 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:38:05,061 INFO [train2.py:809] (3/4) Epoch 15, batch 1050, loss[ctc_loss=0.0866, att_loss=0.2414, loss=0.2104, over 16332.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006126, over 45.00 utterances.], tot_loss[ctc_loss=0.08808, att_loss=0.2432, loss=0.2122, over 3262294.00 frames. utt_duration=1293 frames, utt_pad_proportion=0.04157, over 10104.03 utterances.], batch size: 45, lr: 7.29e-03, grad_scale: 8.0 2023-03-08 11:38:12,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.111e+02 2.438e+02 3.282e+02 6.926e+02, threshold=4.877e+02, percent-clipped=2.0 2023-03-08 11:38:45,003 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6818, 4.7030, 4.5748, 4.6470, 5.1328, 4.8279, 4.4800, 2.2756], device='cuda:3'), covar=tensor([0.0215, 0.0219, 0.0273, 0.0198, 0.0878, 0.0201, 0.0353, 0.1879], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0147, 0.0152, 0.0165, 0.0350, 0.0135, 0.0139, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:38:46,238 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:38:51,153 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2971, 1.7416, 1.9545, 2.0265, 2.4839, 1.8516, 2.1641, 2.4838], device='cuda:3'), covar=tensor([0.2082, 0.4405, 0.3408, 0.1866, 0.2284, 0.1808, 0.2396, 0.0961], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0098, 0.0086, 0.0088, 0.0080, 0.0104, 0.0070], device='cuda:3'), out_proj_covar=tensor([6.4501e-05, 7.1312e-05, 7.3705e-05, 6.3364e-05, 6.3175e-05, 6.2148e-05, 7.3455e-05, 5.4978e-05], device='cuda:3') 2023-03-08 11:39:10,458 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:39:13,555 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2492, 4.5366, 4.5278, 4.8270, 2.8132, 4.6511, 2.8348, 1.7872], device='cuda:3'), covar=tensor([0.0306, 0.0209, 0.0608, 0.0164, 0.1546, 0.0144, 0.1305, 0.1661], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0135, 0.0252, 0.0126, 0.0218, 0.0115, 0.0223, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 11:39:24,722 INFO [train2.py:809] (3/4) Epoch 15, batch 1100, loss[ctc_loss=0.09442, att_loss=0.2511, loss=0.2198, over 17623.00 frames. utt_duration=1009 frames, utt_pad_proportion=0.04487, over 70.00 utterances.], tot_loss[ctc_loss=0.08785, att_loss=0.2425, loss=0.2116, over 3264427.67 frames. utt_duration=1305 frames, utt_pad_proportion=0.03835, over 10018.80 utterances.], batch size: 70, lr: 7.29e-03, grad_scale: 8.0 2023-03-08 11:39:34,154 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 11:40:19,049 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8152, 4.7384, 4.5967, 4.6268, 5.1466, 4.8260, 4.5752, 2.2706], device='cuda:3'), covar=tensor([0.0216, 0.0288, 0.0339, 0.0265, 0.0933, 0.0223, 0.0348, 0.2067], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0147, 0.0152, 0.0165, 0.0349, 0.0135, 0.0138, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:40:20,520 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:40:44,821 INFO [train2.py:809] (3/4) Epoch 15, batch 1150, loss[ctc_loss=0.09599, att_loss=0.2629, loss=0.2295, over 17294.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02412, over 59.00 utterances.], tot_loss[ctc_loss=0.08765, att_loss=0.2424, loss=0.2115, over 3269569.10 frames. utt_duration=1305 frames, utt_pad_proportion=0.0387, over 10035.43 utterances.], batch size: 59, lr: 7.28e-03, grad_scale: 8.0 2023-03-08 11:40:48,922 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-08 11:40:52,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.173e+02 2.768e+02 3.173e+02 4.889e+02, threshold=5.536e+02, percent-clipped=1.0 2023-03-08 11:41:20,313 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6176, 5.1480, 4.8778, 5.0023, 5.1910, 4.6864, 3.5361, 5.0192], device='cuda:3'), covar=tensor([0.0115, 0.0108, 0.0139, 0.0082, 0.0116, 0.0127, 0.0697, 0.0202], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0078, 0.0097, 0.0059, 0.0066, 0.0077, 0.0095, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 11:41:37,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9412, 4.7518, 4.7171, 2.2297, 2.1145, 2.7903, 2.5124, 3.7289], device='cuda:3'), covar=tensor([0.0719, 0.0183, 0.0207, 0.4018, 0.5062, 0.2480, 0.2812, 0.1535], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0237, 0.0243, 0.0223, 0.0342, 0.0333, 0.0231, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4960e-04, 8.8905e-05, 1.0493e-04, 9.7327e-05, 1.4558e-04, 1.3198e-04, 9.2655e-05, 1.4665e-04], device='cuda:3') 2023-03-08 11:41:58,536 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:42:04,424 INFO [train2.py:809] (3/4) Epoch 15, batch 1200, loss[ctc_loss=0.07434, att_loss=0.2171, loss=0.1886, over 15353.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01236, over 35.00 utterances.], tot_loss[ctc_loss=0.08813, att_loss=0.2426, loss=0.2117, over 3268905.06 frames. utt_duration=1269 frames, utt_pad_proportion=0.04702, over 10316.59 utterances.], batch size: 35, lr: 7.28e-03, grad_scale: 8.0 2023-03-08 11:42:36,826 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:42:39,829 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:43:23,350 INFO [train2.py:809] (3/4) Epoch 15, batch 1250, loss[ctc_loss=0.07741, att_loss=0.2481, loss=0.2139, over 17033.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.00862, over 52.00 utterances.], tot_loss[ctc_loss=0.08761, att_loss=0.2425, loss=0.2115, over 3266782.28 frames. utt_duration=1289 frames, utt_pad_proportion=0.04225, over 10152.16 utterances.], batch size: 52, lr: 7.28e-03, grad_scale: 8.0 2023-03-08 11:43:31,118 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.571e+02 2.221e+02 2.683e+02 3.397e+02 1.061e+03, threshold=5.366e+02, percent-clipped=4.0 2023-03-08 11:44:13,696 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:44:17,080 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:44:43,937 INFO [train2.py:809] (3/4) Epoch 15, batch 1300, loss[ctc_loss=0.08658, att_loss=0.234, loss=0.2045, over 16167.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007056, over 41.00 utterances.], tot_loss[ctc_loss=0.08796, att_loss=0.243, loss=0.212, over 3277161.72 frames. utt_duration=1267 frames, utt_pad_proportion=0.0449, over 10362.35 utterances.], batch size: 41, lr: 7.27e-03, grad_scale: 8.0 2023-03-08 11:46:03,453 INFO [train2.py:809] (3/4) Epoch 15, batch 1350, loss[ctc_loss=0.08523, att_loss=0.2144, loss=0.1886, over 15767.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.007467, over 38.00 utterances.], tot_loss[ctc_loss=0.08746, att_loss=0.2425, loss=0.2115, over 3277511.69 frames. utt_duration=1281 frames, utt_pad_proportion=0.04173, over 10242.56 utterances.], batch size: 38, lr: 7.27e-03, grad_scale: 8.0 2023-03-08 11:46:11,170 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.073e+02 2.593e+02 3.161e+02 1.501e+03, threshold=5.185e+02, percent-clipped=3.0 2023-03-08 11:47:00,489 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:47:22,526 INFO [train2.py:809] (3/4) Epoch 15, batch 1400, loss[ctc_loss=0.06641, att_loss=0.2213, loss=0.1904, over 15749.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.01005, over 38.00 utterances.], tot_loss[ctc_loss=0.08879, att_loss=0.2435, loss=0.2126, over 3285666.36 frames. utt_duration=1276 frames, utt_pad_proportion=0.04206, over 10315.69 utterances.], batch size: 38, lr: 7.27e-03, grad_scale: 8.0 2023-03-08 11:47:32,893 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 11:47:44,297 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:47:45,069 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-08 11:48:00,882 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:48:15,963 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0050, 3.8110, 3.7622, 3.2726, 3.8758, 3.8485, 3.8934, 2.7575], device='cuda:3'), covar=tensor([0.0974, 0.1842, 0.2421, 0.5186, 0.0788, 0.3120, 0.0776, 0.6535], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0148, 0.0157, 0.0226, 0.0119, 0.0213, 0.0135, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:48:22,137 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9306, 3.9652, 3.7322, 2.5311, 3.8145, 3.8338, 3.4941, 2.4660], device='cuda:3'), covar=tensor([0.0181, 0.0136, 0.0329, 0.1290, 0.0143, 0.0464, 0.0435, 0.1828], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0092, 0.0088, 0.0107, 0.0076, 0.0101, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 11:48:33,746 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0444, 4.4126, 4.1865, 4.6937, 2.6971, 4.5248, 2.8217, 1.9865], device='cuda:3'), covar=tensor([0.0371, 0.0237, 0.0842, 0.0185, 0.1751, 0.0186, 0.1391, 0.1662], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0135, 0.0252, 0.0126, 0.0219, 0.0116, 0.0223, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 11:48:42,464 INFO [train2.py:809] (3/4) Epoch 15, batch 1450, loss[ctc_loss=0.08016, att_loss=0.2295, loss=0.1996, over 16406.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006623, over 44.00 utterances.], tot_loss[ctc_loss=0.0884, att_loss=0.243, loss=0.2121, over 3280650.89 frames. utt_duration=1257 frames, utt_pad_proportion=0.04793, over 10453.01 utterances.], batch size: 44, lr: 7.26e-03, grad_scale: 8.0 2023-03-08 11:48:48,506 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 11:48:49,835 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.071e+02 2.451e+02 3.320e+02 6.721e+02, threshold=4.903e+02, percent-clipped=4.0 2023-03-08 11:49:04,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-08 11:49:15,909 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:49:20,460 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:49:38,331 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 11:49:48,071 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:50:02,007 INFO [train2.py:809] (3/4) Epoch 15, batch 1500, loss[ctc_loss=0.07906, att_loss=0.2297, loss=0.1995, over 16279.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007326, over 43.00 utterances.], tot_loss[ctc_loss=0.08827, att_loss=0.2429, loss=0.212, over 3277437.37 frames. utt_duration=1247 frames, utt_pad_proportion=0.05283, over 10527.67 utterances.], batch size: 43, lr: 7.26e-03, grad_scale: 16.0 2023-03-08 11:50:53,399 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:51:20,964 INFO [train2.py:809] (3/4) Epoch 15, batch 1550, loss[ctc_loss=0.128, att_loss=0.2747, loss=0.2454, over 17107.00 frames. utt_duration=867.8 frames, utt_pad_proportion=0.08944, over 79.00 utterances.], tot_loss[ctc_loss=0.08869, att_loss=0.2426, loss=0.2118, over 3271031.48 frames. utt_duration=1258 frames, utt_pad_proportion=0.05154, over 10415.91 utterances.], batch size: 79, lr: 7.26e-03, grad_scale: 16.0 2023-03-08 11:51:29,316 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.286e+02 2.083e+02 2.433e+02 3.048e+02 8.459e+02, threshold=4.866e+02, percent-clipped=4.0 2023-03-08 11:52:02,044 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:52:05,152 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:52:39,936 INFO [train2.py:809] (3/4) Epoch 15, batch 1600, loss[ctc_loss=0.09123, att_loss=0.2587, loss=0.2252, over 17138.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01389, over 56.00 utterances.], tot_loss[ctc_loss=0.08872, att_loss=0.2429, loss=0.2121, over 3278517.09 frames. utt_duration=1271 frames, utt_pad_proportion=0.04614, over 10330.92 utterances.], batch size: 56, lr: 7.26e-03, grad_scale: 16.0 2023-03-08 11:53:20,400 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:54:00,796 INFO [train2.py:809] (3/4) Epoch 15, batch 1650, loss[ctc_loss=0.1204, att_loss=0.275, loss=0.2441, over 17298.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02385, over 59.00 utterances.], tot_loss[ctc_loss=0.08788, att_loss=0.2429, loss=0.2119, over 3275902.88 frames. utt_duration=1274 frames, utt_pad_proportion=0.04733, over 10299.74 utterances.], batch size: 59, lr: 7.25e-03, grad_scale: 16.0 2023-03-08 11:54:09,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.363e+02 2.043e+02 2.377e+02 2.961e+02 5.254e+02, threshold=4.753e+02, percent-clipped=2.0 2023-03-08 11:54:31,988 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0021, 3.7432, 3.0971, 3.2919, 3.8625, 3.5423, 2.7179, 4.2040], device='cuda:3'), covar=tensor([0.1045, 0.0529, 0.1105, 0.0735, 0.0745, 0.0752, 0.1019, 0.0512], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0200, 0.0214, 0.0187, 0.0258, 0.0227, 0.0192, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 11:54:59,150 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:54:59,246 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:55:03,831 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:55:20,182 INFO [train2.py:809] (3/4) Epoch 15, batch 1700, loss[ctc_loss=0.1086, att_loss=0.2594, loss=0.2293, over 17333.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02276, over 59.00 utterances.], tot_loss[ctc_loss=0.08753, att_loss=0.2425, loss=0.2115, over 3267721.06 frames. utt_duration=1265 frames, utt_pad_proportion=0.05099, over 10344.51 utterances.], batch size: 59, lr: 7.25e-03, grad_scale: 16.0 2023-03-08 11:55:27,151 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1202, 5.3940, 5.4150, 5.3459, 5.4705, 5.4307, 5.1595, 4.8870], device='cuda:3'), covar=tensor([0.1026, 0.0460, 0.0229, 0.0403, 0.0241, 0.0281, 0.0320, 0.0313], device='cuda:3'), in_proj_covar=tensor([0.0488, 0.0325, 0.0291, 0.0319, 0.0376, 0.0398, 0.0321, 0.0358], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 11:55:54,979 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:56:06,479 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2809, 5.2271, 5.1709, 2.4342, 2.0178, 2.9151, 3.2597, 3.9754], device='cuda:3'), covar=tensor([0.0614, 0.0292, 0.0204, 0.3858, 0.5420, 0.2574, 0.2146, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0235, 0.0242, 0.0221, 0.0340, 0.0331, 0.0230, 0.0350], device='cuda:3'), out_proj_covar=tensor([1.4829e-04, 8.8467e-05, 1.0448e-04, 9.6150e-05, 1.4457e-04, 1.3126e-04, 9.2199e-05, 1.4523e-04], device='cuda:3') 2023-03-08 11:56:14,335 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:56:38,716 INFO [train2.py:809] (3/4) Epoch 15, batch 1750, loss[ctc_loss=0.06915, att_loss=0.2193, loss=0.1893, over 16008.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007234, over 40.00 utterances.], tot_loss[ctc_loss=0.08706, att_loss=0.2422, loss=0.2111, over 3269045.59 frames. utt_duration=1263 frames, utt_pad_proportion=0.05174, over 10363.57 utterances.], batch size: 40, lr: 7.25e-03, grad_scale: 16.0 2023-03-08 11:56:39,758 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:56:42,623 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6653, 4.8841, 4.3654, 4.7892, 4.5211, 4.1425, 4.4336, 4.1847], device='cuda:3'), covar=tensor([0.1225, 0.1074, 0.1020, 0.0862, 0.1067, 0.1498, 0.1991, 0.2507], device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0537, 0.0409, 0.0405, 0.0387, 0.0427, 0.0553, 0.0493], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 11:56:46,885 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.541e+02 2.236e+02 2.703e+02 3.413e+02 7.176e+02, threshold=5.406e+02, percent-clipped=5.0 2023-03-08 11:57:08,498 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:57:26,202 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 11:57:26,375 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7657, 1.8141, 2.1442, 1.9397, 2.5873, 2.4957, 1.9117, 2.9299], device='cuda:3'), covar=tensor([0.1890, 0.4601, 0.5574, 0.3555, 0.3968, 0.2025, 0.4359, 0.1334], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0097, 0.0100, 0.0087, 0.0090, 0.0080, 0.0104, 0.0071], device='cuda:3'), out_proj_covar=tensor([6.5051e-05, 7.1561e-05, 7.4652e-05, 6.4554e-05, 6.4352e-05, 6.2172e-05, 7.4085e-05, 5.5856e-05], device='cuda:3') 2023-03-08 11:57:31,863 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:57:44,167 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:57:58,284 INFO [train2.py:809] (3/4) Epoch 15, batch 1800, loss[ctc_loss=0.09819, att_loss=0.256, loss=0.2244, over 17302.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.0112, over 55.00 utterances.], tot_loss[ctc_loss=0.08659, att_loss=0.2416, loss=0.2106, over 3263167.57 frames. utt_duration=1265 frames, utt_pad_proportion=0.05449, over 10327.07 utterances.], batch size: 55, lr: 7.24e-03, grad_scale: 16.0 2023-03-08 11:58:29,087 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-03-08 11:58:41,635 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:58:56,309 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6459, 4.7242, 4.6289, 4.7680, 5.2450, 4.9319, 4.6662, 2.4717], device='cuda:3'), covar=tensor([0.0214, 0.0229, 0.0238, 0.0208, 0.0715, 0.0168, 0.0263, 0.1871], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0148, 0.0153, 0.0164, 0.0350, 0.0134, 0.0140, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 11:59:00,593 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 11:59:19,513 INFO [train2.py:809] (3/4) Epoch 15, batch 1850, loss[ctc_loss=0.08938, att_loss=0.2395, loss=0.2095, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008766, over 37.00 utterances.], tot_loss[ctc_loss=0.08634, att_loss=0.2415, loss=0.2104, over 3264539.81 frames. utt_duration=1283 frames, utt_pad_proportion=0.05044, over 10189.90 utterances.], batch size: 37, lr: 7.24e-03, grad_scale: 16.0 2023-03-08 11:59:27,210 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.149e+02 2.578e+02 3.271e+02 8.979e+02, threshold=5.155e+02, percent-clipped=4.0 2023-03-08 11:59:33,905 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8331, 5.1203, 4.6346, 5.2329, 4.5474, 4.8678, 5.2532, 5.0309], device='cuda:3'), covar=tensor([0.0558, 0.0272, 0.0906, 0.0271, 0.0472, 0.0280, 0.0243, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0284, 0.0342, 0.0299, 0.0294, 0.0223, 0.0277, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 12:00:00,778 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:00:03,809 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:00:38,111 INFO [train2.py:809] (3/4) Epoch 15, batch 1900, loss[ctc_loss=0.08629, att_loss=0.2496, loss=0.2169, over 17033.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.006381, over 51.00 utterances.], tot_loss[ctc_loss=0.08602, att_loss=0.2411, loss=0.2101, over 3259949.22 frames. utt_duration=1284 frames, utt_pad_proportion=0.04971, over 10164.11 utterances.], batch size: 51, lr: 7.24e-03, grad_scale: 16.0 2023-03-08 12:01:08,087 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1581, 5.4312, 4.9576, 5.5080, 4.8913, 5.0660, 5.6104, 5.3736], device='cuda:3'), covar=tensor([0.0460, 0.0233, 0.0737, 0.0214, 0.0367, 0.0193, 0.0186, 0.0164], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0285, 0.0343, 0.0301, 0.0295, 0.0225, 0.0278, 0.0256], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 12:01:12,790 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6243, 3.0350, 3.6833, 3.0348, 3.5561, 4.7184, 4.4006, 3.4958], device='cuda:3'), covar=tensor([0.0322, 0.1715, 0.1162, 0.1458, 0.1067, 0.0623, 0.0578, 0.1146], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0236, 0.0264, 0.0209, 0.0255, 0.0335, 0.0239, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:01:15,661 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:01:18,924 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:01:19,265 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5826, 4.6167, 4.5265, 4.4895, 5.0518, 4.8174, 4.4752, 2.2348], device='cuda:3'), covar=tensor([0.0203, 0.0246, 0.0270, 0.0237, 0.0788, 0.0160, 0.0311, 0.2026], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0148, 0.0155, 0.0166, 0.0353, 0.0135, 0.0141, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 12:01:56,431 INFO [train2.py:809] (3/4) Epoch 15, batch 1950, loss[ctc_loss=0.1, att_loss=0.2606, loss=0.2284, over 16466.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006674, over 46.00 utterances.], tot_loss[ctc_loss=0.08673, att_loss=0.2416, loss=0.2107, over 3262152.40 frames. utt_duration=1281 frames, utt_pad_proportion=0.04911, over 10195.14 utterances.], batch size: 46, lr: 7.23e-03, grad_scale: 16.0 2023-03-08 12:02:05,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.187e+02 2.629e+02 3.507e+02 9.887e+02, threshold=5.258e+02, percent-clipped=4.0 2023-03-08 12:02:13,658 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:02:20,545 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 12:02:47,510 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:03:17,166 INFO [train2.py:809] (3/4) Epoch 15, batch 2000, loss[ctc_loss=0.0707, att_loss=0.2235, loss=0.1929, over 15965.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.005724, over 41.00 utterances.], tot_loss[ctc_loss=0.0862, att_loss=0.2414, loss=0.2104, over 3257254.75 frames. utt_duration=1276 frames, utt_pad_proportion=0.05127, over 10226.02 utterances.], batch size: 41, lr: 7.23e-03, grad_scale: 16.0 2023-03-08 12:03:31,266 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 2023-03-08 12:03:43,296 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 12:03:50,220 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:04:28,945 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:04:30,815 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5614, 2.6682, 3.6477, 2.7329, 3.4840, 4.6814, 4.5894, 2.9702], device='cuda:3'), covar=tensor([0.0414, 0.1960, 0.0991, 0.1621, 0.0987, 0.0710, 0.0435, 0.1639], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0237, 0.0265, 0.0209, 0.0255, 0.0333, 0.0237, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:04:37,300 INFO [train2.py:809] (3/4) Epoch 15, batch 2050, loss[ctc_loss=0.08035, att_loss=0.2426, loss=0.2102, over 17021.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007948, over 51.00 utterances.], tot_loss[ctc_loss=0.08715, att_loss=0.2419, loss=0.2109, over 3263233.77 frames. utt_duration=1279 frames, utt_pad_proportion=0.04897, over 10219.66 utterances.], batch size: 51, lr: 7.23e-03, grad_scale: 16.0 2023-03-08 12:04:45,712 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.062e+02 2.536e+02 3.073e+02 8.062e+02, threshold=5.072e+02, percent-clipped=2.0 2023-03-08 12:05:07,139 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:05:22,343 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:05:25,450 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 12:05:42,477 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8333, 6.1235, 5.5175, 5.8790, 5.7344, 5.2795, 5.5267, 5.3551], device='cuda:3'), covar=tensor([0.1312, 0.0783, 0.0975, 0.0665, 0.0916, 0.1468, 0.2398, 0.2147], device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0546, 0.0419, 0.0416, 0.0398, 0.0435, 0.0564, 0.0500], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 12:05:56,962 INFO [train2.py:809] (3/4) Epoch 15, batch 2100, loss[ctc_loss=0.06527, att_loss=0.2206, loss=0.1895, over 15878.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009689, over 39.00 utterances.], tot_loss[ctc_loss=0.08734, att_loss=0.2423, loss=0.2113, over 3261711.68 frames. utt_duration=1271 frames, utt_pad_proportion=0.05081, over 10280.00 utterances.], batch size: 39, lr: 7.22e-03, grad_scale: 16.0 2023-03-08 12:06:23,276 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:06:28,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-03-08 12:06:40,442 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:06:41,923 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:07:17,263 INFO [train2.py:809] (3/4) Epoch 15, batch 2150, loss[ctc_loss=0.1096, att_loss=0.2633, loss=0.2326, over 17057.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008018, over 52.00 utterances.], tot_loss[ctc_loss=0.08703, att_loss=0.2422, loss=0.2112, over 3264587.17 frames. utt_duration=1271 frames, utt_pad_proportion=0.05098, over 10286.55 utterances.], batch size: 52, lr: 7.22e-03, grad_scale: 16.0 2023-03-08 12:07:20,082 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-08 12:07:25,215 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.164e+02 2.681e+02 3.169e+02 6.485e+02, threshold=5.361e+02, percent-clipped=3.0 2023-03-08 12:07:32,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 12:07:43,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-08 12:07:57,493 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:08:30,474 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-03-08 12:08:36,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-03-08 12:08:37,645 INFO [train2.py:809] (3/4) Epoch 15, batch 2200, loss[ctc_loss=0.08715, att_loss=0.2557, loss=0.222, over 17063.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009035, over 53.00 utterances.], tot_loss[ctc_loss=0.08734, att_loss=0.2427, loss=0.2116, over 3264635.75 frames. utt_duration=1291 frames, utt_pad_proportion=0.04671, over 10129.62 utterances.], batch size: 53, lr: 7.22e-03, grad_scale: 16.0 2023-03-08 12:08:43,298 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-08 12:10:01,087 INFO [train2.py:809] (3/4) Epoch 15, batch 2250, loss[ctc_loss=0.09139, att_loss=0.2473, loss=0.2161, over 17048.00 frames. utt_duration=1339 frames, utt_pad_proportion=0.006982, over 51.00 utterances.], tot_loss[ctc_loss=0.08741, att_loss=0.2428, loss=0.2117, over 3263308.53 frames. utt_duration=1285 frames, utt_pad_proportion=0.04723, over 10166.43 utterances.], batch size: 51, lr: 7.22e-03, grad_scale: 16.0 2023-03-08 12:10:08,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.417e+02 2.180e+02 2.771e+02 3.383e+02 6.703e+02, threshold=5.542e+02, percent-clipped=1.0 2023-03-08 12:10:49,953 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:10:56,256 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2410, 4.6663, 4.6438, 4.8247, 3.2154, 4.9881, 2.8746, 2.1809], device='cuda:3'), covar=tensor([0.0301, 0.0193, 0.0598, 0.0175, 0.1433, 0.0131, 0.1321, 0.1538], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0136, 0.0253, 0.0127, 0.0223, 0.0117, 0.0224, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:11:07,993 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1753, 5.1736, 4.9900, 2.1428, 2.0223, 2.9605, 2.5627, 3.9314], device='cuda:3'), covar=tensor([0.0682, 0.0218, 0.0227, 0.4846, 0.5808, 0.2488, 0.3058, 0.1573], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0241, 0.0245, 0.0226, 0.0344, 0.0337, 0.0234, 0.0355], device='cuda:3'), out_proj_covar=tensor([1.5076e-04, 9.0774e-05, 1.0629e-04, 9.8408e-05, 1.4675e-04, 1.3373e-04, 9.3765e-05, 1.4717e-04], device='cuda:3') 2023-03-08 12:11:19,144 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4688, 2.7758, 3.5513, 2.9274, 3.4420, 4.5820, 4.3689, 3.1252], device='cuda:3'), covar=tensor([0.0361, 0.1912, 0.1145, 0.1384, 0.1144, 0.0698, 0.0633, 0.1462], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0236, 0.0262, 0.0208, 0.0253, 0.0332, 0.0237, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:11:20,345 INFO [train2.py:809] (3/4) Epoch 15, batch 2300, loss[ctc_loss=0.08602, att_loss=0.2534, loss=0.22, over 17019.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008582, over 51.00 utterances.], tot_loss[ctc_loss=0.08625, att_loss=0.242, loss=0.2108, over 3269595.25 frames. utt_duration=1303 frames, utt_pad_proportion=0.0417, over 10049.25 utterances.], batch size: 51, lr: 7.21e-03, grad_scale: 8.0 2023-03-08 12:11:27,801 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-03-08 12:11:45,869 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:12:06,461 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:12:31,598 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2519, 1.8827, 2.0152, 2.4245, 2.6019, 2.1978, 2.2298, 2.9679], device='cuda:3'), covar=tensor([0.1629, 0.3531, 0.2763, 0.1452, 0.1710, 0.1636, 0.2699, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0097, 0.0085, 0.0087, 0.0078, 0.0101, 0.0069], device='cuda:3'), out_proj_covar=tensor([6.3912e-05, 6.9866e-05, 7.3158e-05, 6.3259e-05, 6.2784e-05, 6.0970e-05, 7.2739e-05, 5.4543e-05], device='cuda:3') 2023-03-08 12:12:33,107 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:12:37,054 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.20 vs. limit=5.0 2023-03-08 12:12:40,615 INFO [train2.py:809] (3/4) Epoch 15, batch 2350, loss[ctc_loss=0.06714, att_loss=0.2483, loss=0.2121, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01093, over 55.00 utterances.], tot_loss[ctc_loss=0.08603, att_loss=0.242, loss=0.2108, over 3270670.27 frames. utt_duration=1297 frames, utt_pad_proportion=0.04349, over 10099.33 utterances.], batch size: 55, lr: 7.21e-03, grad_scale: 8.0 2023-03-08 12:12:44,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-03-08 12:12:45,691 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1689, 5.0647, 4.9086, 2.8613, 4.8578, 4.7575, 4.4821, 2.7565], device='cuda:3'), covar=tensor([0.0115, 0.0092, 0.0308, 0.1061, 0.0100, 0.0190, 0.0268, 0.1345], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0093, 0.0090, 0.0109, 0.0078, 0.0103, 0.0097, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 12:12:49,822 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.243e+02 2.592e+02 3.349e+02 5.777e+02, threshold=5.184e+02, percent-clipped=2.0 2023-03-08 12:13:08,889 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:13:14,195 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:13:24,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:13:26,378 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3967, 4.3835, 4.4196, 4.3730, 4.9184, 4.6160, 4.3565, 2.2157], device='cuda:3'), covar=tensor([0.0253, 0.0305, 0.0298, 0.0253, 0.0975, 0.0214, 0.0307, 0.2159], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0150, 0.0155, 0.0168, 0.0353, 0.0135, 0.0143, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 12:13:48,260 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:13:58,917 INFO [train2.py:809] (3/4) Epoch 15, batch 2400, loss[ctc_loss=0.09023, att_loss=0.2551, loss=0.2221, over 17122.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01485, over 56.00 utterances.], tot_loss[ctc_loss=0.08668, att_loss=0.2422, loss=0.2111, over 3265850.88 frames. utt_duration=1278 frames, utt_pad_proportion=0.05005, over 10236.95 utterances.], batch size: 56, lr: 7.21e-03, grad_scale: 8.0 2023-03-08 12:14:39,205 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0494, 4.1661, 3.8698, 4.2294, 3.9068, 3.8273, 4.2268, 4.1313], device='cuda:3'), covar=tensor([0.0564, 0.0357, 0.0747, 0.0398, 0.0410, 0.0790, 0.0310, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0286, 0.0344, 0.0302, 0.0296, 0.0222, 0.0278, 0.0256], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 12:14:40,625 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:14:45,854 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:14:50,482 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:15:18,771 INFO [train2.py:809] (3/4) Epoch 15, batch 2450, loss[ctc_loss=0.08285, att_loss=0.2278, loss=0.1988, over 16286.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.005795, over 43.00 utterances.], tot_loss[ctc_loss=0.08691, att_loss=0.2421, loss=0.2111, over 3268861.96 frames. utt_duration=1246 frames, utt_pad_proportion=0.05586, over 10510.79 utterances.], batch size: 43, lr: 7.20e-03, grad_scale: 8.0 2023-03-08 12:15:27,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.141e+02 2.652e+02 3.416e+02 6.108e+02, threshold=5.305e+02, percent-clipped=4.0 2023-03-08 12:16:33,130 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-08 12:16:38,079 INFO [train2.py:809] (3/4) Epoch 15, batch 2500, loss[ctc_loss=0.1469, att_loss=0.2785, loss=0.2522, over 14641.00 frames. utt_duration=402.7 frames, utt_pad_proportion=0.2997, over 146.00 utterances.], tot_loss[ctc_loss=0.08706, att_loss=0.2422, loss=0.2112, over 3272740.35 frames. utt_duration=1233 frames, utt_pad_proportion=0.05758, over 10627.92 utterances.], batch size: 146, lr: 7.20e-03, grad_scale: 8.0 2023-03-08 12:17:57,544 INFO [train2.py:809] (3/4) Epoch 15, batch 2550, loss[ctc_loss=0.08623, att_loss=0.2335, loss=0.204, over 16264.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007635, over 43.00 utterances.], tot_loss[ctc_loss=0.08726, att_loss=0.2415, loss=0.2107, over 3261069.36 frames. utt_duration=1236 frames, utt_pad_proportion=0.05796, over 10565.45 utterances.], batch size: 43, lr: 7.20e-03, grad_scale: 8.0 2023-03-08 12:18:06,732 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.062e+02 2.631e+02 3.242e+02 5.829e+02, threshold=5.263e+02, percent-clipped=2.0 2023-03-08 12:18:10,233 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3483, 2.9244, 2.8904, 2.7057, 3.0223, 2.8906, 2.9975, 2.2840], device='cuda:3'), covar=tensor([0.0956, 0.2017, 0.3118, 0.4206, 0.1116, 0.4354, 0.1302, 0.5164], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0149, 0.0159, 0.0225, 0.0120, 0.0215, 0.0136, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 12:18:43,138 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-08 12:19:17,178 INFO [train2.py:809] (3/4) Epoch 15, batch 2600, loss[ctc_loss=0.066, att_loss=0.2213, loss=0.1903, over 16159.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.00642, over 41.00 utterances.], tot_loss[ctc_loss=0.08696, att_loss=0.242, loss=0.211, over 3270561.77 frames. utt_duration=1255 frames, utt_pad_proportion=0.05102, over 10437.23 utterances.], batch size: 41, lr: 7.19e-03, grad_scale: 8.0 2023-03-08 12:19:43,497 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:20:37,376 INFO [train2.py:809] (3/4) Epoch 15, batch 2650, loss[ctc_loss=0.0737, att_loss=0.2435, loss=0.2095, over 16553.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005552, over 45.00 utterances.], tot_loss[ctc_loss=0.08757, att_loss=0.2432, loss=0.2121, over 3279062.56 frames. utt_duration=1231 frames, utt_pad_proportion=0.0555, over 10669.47 utterances.], batch size: 45, lr: 7.19e-03, grad_scale: 8.0 2023-03-08 12:20:42,313 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9388, 5.2716, 5.4610, 5.3349, 5.3691, 5.9117, 5.1972, 5.9828], device='cuda:3'), covar=tensor([0.0623, 0.0611, 0.0720, 0.1074, 0.1791, 0.0812, 0.0664, 0.0636], device='cuda:3'), in_proj_covar=tensor([0.0779, 0.0461, 0.0538, 0.0599, 0.0790, 0.0552, 0.0442, 0.0525], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:20:46,673 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.218e+02 2.453e+02 3.355e+02 6.382e+02, threshold=4.906e+02, percent-clipped=2.0 2023-03-08 12:21:00,186 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:21:01,921 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1258, 5.4435, 5.3964, 5.3110, 5.4784, 5.4378, 5.1405, 4.9284], device='cuda:3'), covar=tensor([0.1018, 0.0463, 0.0267, 0.0464, 0.0255, 0.0264, 0.0323, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0483, 0.0325, 0.0294, 0.0322, 0.0377, 0.0397, 0.0319, 0.0360], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:21:18,079 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9943, 5.2164, 5.1408, 5.1462, 5.2986, 5.2878, 4.9362, 4.8020], device='cuda:3'), covar=tensor([0.1029, 0.0530, 0.0357, 0.0505, 0.0285, 0.0294, 0.0379, 0.0334], device='cuda:3'), in_proj_covar=tensor([0.0483, 0.0325, 0.0294, 0.0322, 0.0377, 0.0397, 0.0320, 0.0360], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:21:22,724 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1097, 5.3514, 5.5901, 5.3980, 5.5827, 6.0203, 5.2786, 6.1153], device='cuda:3'), covar=tensor([0.0578, 0.0604, 0.0737, 0.1170, 0.1671, 0.0846, 0.0613, 0.0615], device='cuda:3'), in_proj_covar=tensor([0.0779, 0.0461, 0.0538, 0.0600, 0.0790, 0.0553, 0.0441, 0.0525], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:21:57,758 INFO [train2.py:809] (3/4) Epoch 15, batch 2700, loss[ctc_loss=0.1021, att_loss=0.261, loss=0.2293, over 17039.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.01033, over 53.00 utterances.], tot_loss[ctc_loss=0.0876, att_loss=0.243, loss=0.212, over 3277794.15 frames. utt_duration=1231 frames, utt_pad_proportion=0.05503, over 10664.48 utterances.], batch size: 53, lr: 7.19e-03, grad_scale: 8.0 2023-03-08 12:22:32,827 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 12:22:34,256 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 12:22:36,306 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:22:41,393 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:23:17,024 INFO [train2.py:809] (3/4) Epoch 15, batch 2750, loss[ctc_loss=0.09058, att_loss=0.253, loss=0.2205, over 16954.00 frames. utt_duration=686.4 frames, utt_pad_proportion=0.1366, over 99.00 utterances.], tot_loss[ctc_loss=0.08737, att_loss=0.2428, loss=0.2117, over 3278476.83 frames. utt_duration=1245 frames, utt_pad_proportion=0.05159, over 10548.80 utterances.], batch size: 99, lr: 7.18e-03, grad_scale: 8.0 2023-03-08 12:23:26,923 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.307e+02 2.848e+02 3.393e+02 6.929e+02, threshold=5.697e+02, percent-clipped=6.0 2023-03-08 12:24:35,857 INFO [train2.py:809] (3/4) Epoch 15, batch 2800, loss[ctc_loss=0.1202, att_loss=0.2635, loss=0.2348, over 17349.00 frames. utt_duration=879.9 frames, utt_pad_proportion=0.07478, over 79.00 utterances.], tot_loss[ctc_loss=0.0879, att_loss=0.2433, loss=0.2122, over 3285450.86 frames. utt_duration=1250 frames, utt_pad_proportion=0.0492, over 10530.27 utterances.], batch size: 79, lr: 7.18e-03, grad_scale: 8.0 2023-03-08 12:24:51,040 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9376, 4.4783, 4.1652, 4.5881, 2.6136, 4.4328, 2.3075, 1.9697], device='cuda:3'), covar=tensor([0.0420, 0.0190, 0.0918, 0.0157, 0.2137, 0.0192, 0.2019, 0.1931], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0133, 0.0247, 0.0124, 0.0218, 0.0117, 0.0219, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:25:53,745 INFO [train2.py:809] (3/4) Epoch 15, batch 2850, loss[ctc_loss=0.1135, att_loss=0.2361, loss=0.2116, over 14531.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.0395, over 32.00 utterances.], tot_loss[ctc_loss=0.08722, att_loss=0.2425, loss=0.2115, over 3279557.36 frames. utt_duration=1271 frames, utt_pad_proportion=0.04652, over 10334.79 utterances.], batch size: 32, lr: 7.18e-03, grad_scale: 8.0 2023-03-08 12:26:02,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.157e+02 2.674e+02 3.074e+02 6.924e+02, threshold=5.347e+02, percent-clipped=2.0 2023-03-08 12:26:07,518 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3216, 2.6274, 3.0332, 4.3254, 3.7634, 3.8322, 2.7631, 2.0844], device='cuda:3'), covar=tensor([0.0710, 0.2109, 0.1126, 0.0523, 0.0803, 0.0500, 0.1658, 0.2317], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0211, 0.0189, 0.0199, 0.0206, 0.0167, 0.0198, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:27:12,213 INFO [train2.py:809] (3/4) Epoch 15, batch 2900, loss[ctc_loss=0.08855, att_loss=0.2292, loss=0.2011, over 15655.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.007473, over 37.00 utterances.], tot_loss[ctc_loss=0.08671, att_loss=0.2425, loss=0.2113, over 3275771.91 frames. utt_duration=1298 frames, utt_pad_proportion=0.0408, over 10103.22 utterances.], batch size: 37, lr: 7.18e-03, grad_scale: 8.0 2023-03-08 12:27:18,811 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8797, 4.0376, 3.7891, 4.2009, 2.5308, 4.0826, 2.5387, 2.0519], device='cuda:3'), covar=tensor([0.0369, 0.0190, 0.0773, 0.0192, 0.1855, 0.0215, 0.1508, 0.1558], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0133, 0.0248, 0.0124, 0.0220, 0.0118, 0.0219, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:27:23,869 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9500, 5.2736, 5.1847, 5.1089, 5.2683, 5.2570, 4.9700, 4.7394], device='cuda:3'), covar=tensor([0.1049, 0.0485, 0.0296, 0.0558, 0.0287, 0.0315, 0.0405, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0323, 0.0293, 0.0318, 0.0377, 0.0395, 0.0319, 0.0360], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:27:26,489 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:28:19,455 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7519, 3.1929, 3.7898, 3.3583, 3.6954, 4.7919, 4.5713, 3.6767], device='cuda:3'), covar=tensor([0.0317, 0.1500, 0.1105, 0.1237, 0.0946, 0.0749, 0.0474, 0.1023], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0234, 0.0262, 0.0209, 0.0252, 0.0330, 0.0236, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:28:31,200 INFO [train2.py:809] (3/4) Epoch 15, batch 2950, loss[ctc_loss=0.06321, att_loss=0.2137, loss=0.1836, over 15892.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008137, over 39.00 utterances.], tot_loss[ctc_loss=0.08755, att_loss=0.2427, loss=0.2116, over 3268637.62 frames. utt_duration=1247 frames, utt_pad_proportion=0.05465, over 10495.18 utterances.], batch size: 39, lr: 7.17e-03, grad_scale: 8.0 2023-03-08 12:28:41,001 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.063e+02 2.403e+02 2.955e+02 5.605e+02, threshold=4.806e+02, percent-clipped=1.0 2023-03-08 12:29:01,901 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:29:49,919 INFO [train2.py:809] (3/4) Epoch 15, batch 3000, loss[ctc_loss=0.09955, att_loss=0.2458, loss=0.2165, over 15615.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.0101, over 37.00 utterances.], tot_loss[ctc_loss=0.0867, att_loss=0.2418, loss=0.2107, over 3264627.86 frames. utt_duration=1261 frames, utt_pad_proportion=0.05246, over 10370.94 utterances.], batch size: 37, lr: 7.17e-03, grad_scale: 8.0 2023-03-08 12:29:49,919 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 12:30:03,569 INFO [train2.py:843] (3/4) Epoch 15, validation: ctc_loss=0.04336, att_loss=0.2362, loss=0.1976, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 12:30:03,570 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 12:30:07,197 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:30:39,522 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6221, 3.0135, 3.6922, 3.1916, 3.5574, 4.6791, 4.4801, 3.5436], device='cuda:3'), covar=tensor([0.0383, 0.1730, 0.1231, 0.1344, 0.1079, 0.0871, 0.0532, 0.1144], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0233, 0.0262, 0.0209, 0.0252, 0.0330, 0.0236, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:30:42,551 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:30:47,647 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:31:24,434 INFO [train2.py:809] (3/4) Epoch 15, batch 3050, loss[ctc_loss=0.08456, att_loss=0.2451, loss=0.213, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005513, over 47.00 utterances.], tot_loss[ctc_loss=0.08632, att_loss=0.2414, loss=0.2104, over 3257186.24 frames. utt_duration=1276 frames, utt_pad_proportion=0.05031, over 10220.36 utterances.], batch size: 47, lr: 7.17e-03, grad_scale: 8.0 2023-03-08 12:31:34,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.376e+02 2.164e+02 2.584e+02 3.141e+02 6.750e+02, threshold=5.168e+02, percent-clipped=6.0 2023-03-08 12:31:45,928 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:31:59,461 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:32:04,249 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:32:43,306 INFO [train2.py:809] (3/4) Epoch 15, batch 3100, loss[ctc_loss=0.07933, att_loss=0.2497, loss=0.2156, over 16468.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.007166, over 46.00 utterances.], tot_loss[ctc_loss=0.08688, att_loss=0.2422, loss=0.2111, over 3269582.43 frames. utt_duration=1286 frames, utt_pad_proportion=0.04446, over 10182.50 utterances.], batch size: 46, lr: 7.16e-03, grad_scale: 8.0 2023-03-08 12:33:10,736 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-08 12:33:34,873 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:34:00,655 INFO [train2.py:809] (3/4) Epoch 15, batch 3150, loss[ctc_loss=0.1005, att_loss=0.2599, loss=0.228, over 17071.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008455, over 53.00 utterances.], tot_loss[ctc_loss=0.0871, att_loss=0.2423, loss=0.2112, over 3265658.48 frames. utt_duration=1279 frames, utt_pad_proportion=0.04887, over 10222.64 utterances.], batch size: 53, lr: 7.16e-03, grad_scale: 8.0 2023-03-08 12:34:02,703 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4004, 2.3826, 4.7832, 3.8317, 2.7390, 4.1064, 4.4974, 4.4387], device='cuda:3'), covar=tensor([0.0193, 0.1770, 0.0117, 0.0908, 0.1885, 0.0243, 0.0139, 0.0222], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0243, 0.0157, 0.0308, 0.0264, 0.0191, 0.0141, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 12:34:09,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 2.265e+02 2.592e+02 3.060e+02 7.077e+02, threshold=5.184e+02, percent-clipped=3.0 2023-03-08 12:35:10,524 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:35:19,407 INFO [train2.py:809] (3/4) Epoch 15, batch 3200, loss[ctc_loss=0.08954, att_loss=0.2524, loss=0.2199, over 17420.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04586, over 69.00 utterances.], tot_loss[ctc_loss=0.0866, att_loss=0.2421, loss=0.211, over 3271493.98 frames. utt_duration=1290 frames, utt_pad_proportion=0.04407, over 10154.31 utterances.], batch size: 69, lr: 7.16e-03, grad_scale: 8.0 2023-03-08 12:35:20,970 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9778, 6.2438, 5.6965, 5.9821, 5.8689, 5.4586, 5.6391, 5.4490], device='cuda:3'), covar=tensor([0.1422, 0.0957, 0.0914, 0.0823, 0.0889, 0.1584, 0.2710, 0.2420], device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0557, 0.0423, 0.0420, 0.0400, 0.0437, 0.0566, 0.0506], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 12:35:53,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-03-08 12:36:38,221 INFO [train2.py:809] (3/4) Epoch 15, batch 3250, loss[ctc_loss=0.0859, att_loss=0.23, loss=0.2012, over 15894.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008864, over 39.00 utterances.], tot_loss[ctc_loss=0.08672, att_loss=0.2423, loss=0.2112, over 3275344.79 frames. utt_duration=1272 frames, utt_pad_proportion=0.04644, over 10315.53 utterances.], batch size: 39, lr: 7.15e-03, grad_scale: 8.0 2023-03-08 12:36:48,043 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.216e+02 2.757e+02 3.407e+02 8.183e+02, threshold=5.513e+02, percent-clipped=5.0 2023-03-08 12:37:00,988 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:37:36,104 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-08 12:37:57,613 INFO [train2.py:809] (3/4) Epoch 15, batch 3300, loss[ctc_loss=0.1091, att_loss=0.2618, loss=0.2313, over 17266.00 frames. utt_duration=875.6 frames, utt_pad_proportion=0.08317, over 79.00 utterances.], tot_loss[ctc_loss=0.08679, att_loss=0.2421, loss=0.211, over 3272699.63 frames. utt_duration=1255 frames, utt_pad_proportion=0.05225, over 10442.64 utterances.], batch size: 79, lr: 7.15e-03, grad_scale: 8.0 2023-03-08 12:39:16,674 INFO [train2.py:809] (3/4) Epoch 15, batch 3350, loss[ctc_loss=0.1529, att_loss=0.2827, loss=0.2568, over 16846.00 frames. utt_duration=682 frames, utt_pad_proportion=0.1432, over 99.00 utterances.], tot_loss[ctc_loss=0.08762, att_loss=0.2426, loss=0.2116, over 3269106.00 frames. utt_duration=1218 frames, utt_pad_proportion=0.06162, over 10746.56 utterances.], batch size: 99, lr: 7.15e-03, grad_scale: 8.0 2023-03-08 12:39:27,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 1.917e+02 2.406e+02 2.955e+02 7.524e+02, threshold=4.811e+02, percent-clipped=3.0 2023-03-08 12:39:30,504 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:40:22,587 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:40:36,061 INFO [train2.py:809] (3/4) Epoch 15, batch 3400, loss[ctc_loss=0.09328, att_loss=0.2451, loss=0.2147, over 16484.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.00568, over 46.00 utterances.], tot_loss[ctc_loss=0.08818, att_loss=0.2435, loss=0.2124, over 3278138.96 frames. utt_duration=1224 frames, utt_pad_proportion=0.05792, over 10725.32 utterances.], batch size: 46, lr: 7.15e-03, grad_scale: 8.0 2023-03-08 12:41:48,551 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6222, 5.0973, 5.1393, 5.0053, 5.0927, 5.0592, 4.8164, 4.6497], device='cuda:3'), covar=tensor([0.1395, 0.0661, 0.0313, 0.0618, 0.0495, 0.0432, 0.0444, 0.0444], device='cuda:3'), in_proj_covar=tensor([0.0484, 0.0326, 0.0292, 0.0322, 0.0379, 0.0401, 0.0323, 0.0361], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:41:56,072 INFO [train2.py:809] (3/4) Epoch 15, batch 3450, loss[ctc_loss=0.07914, att_loss=0.2173, loss=0.1897, over 15379.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01062, over 35.00 utterances.], tot_loss[ctc_loss=0.08797, att_loss=0.2431, loss=0.212, over 3272309.29 frames. utt_duration=1200 frames, utt_pad_proportion=0.06622, over 10918.35 utterances.], batch size: 35, lr: 7.14e-03, grad_scale: 8.0 2023-03-08 12:42:00,132 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:42:06,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.401e+02 2.877e+02 3.602e+02 6.308e+02, threshold=5.755e+02, percent-clipped=6.0 2023-03-08 12:42:11,794 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1648, 3.8077, 3.2290, 3.6426, 4.1430, 3.6569, 2.8689, 4.3730], device='cuda:3'), covar=tensor([0.0880, 0.0424, 0.0923, 0.0575, 0.0558, 0.0639, 0.0883, 0.0496], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0216, 0.0189, 0.0259, 0.0227, 0.0192, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:42:58,753 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:43:16,319 INFO [train2.py:809] (3/4) Epoch 15, batch 3500, loss[ctc_loss=0.07245, att_loss=0.2249, loss=0.1944, over 16476.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.00676, over 46.00 utterances.], tot_loss[ctc_loss=0.08756, att_loss=0.2422, loss=0.2113, over 3264751.22 frames. utt_duration=1226 frames, utt_pad_proportion=0.06141, over 10665.68 utterances.], batch size: 46, lr: 7.14e-03, grad_scale: 8.0 2023-03-08 12:44:35,597 INFO [train2.py:809] (3/4) Epoch 15, batch 3550, loss[ctc_loss=0.0789, att_loss=0.2401, loss=0.2078, over 16870.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00593, over 49.00 utterances.], tot_loss[ctc_loss=0.08713, att_loss=0.2422, loss=0.2112, over 3274356.48 frames. utt_duration=1229 frames, utt_pad_proportion=0.05782, over 10667.68 utterances.], batch size: 49, lr: 7.14e-03, grad_scale: 8.0 2023-03-08 12:44:45,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 2.222e+02 2.629e+02 3.391e+02 5.578e+02, threshold=5.259e+02, percent-clipped=0.0 2023-03-08 12:44:57,812 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:45:31,140 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6122, 3.6332, 3.5709, 3.8159, 2.6439, 3.8054, 2.7000, 2.1465], device='cuda:3'), covar=tensor([0.0422, 0.0271, 0.0650, 0.0208, 0.1465, 0.0201, 0.1252, 0.1401], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0135, 0.0247, 0.0125, 0.0219, 0.0118, 0.0222, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:45:54,581 INFO [train2.py:809] (3/4) Epoch 15, batch 3600, loss[ctc_loss=0.08099, att_loss=0.2368, loss=0.2057, over 16318.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006962, over 45.00 utterances.], tot_loss[ctc_loss=0.08678, att_loss=0.2422, loss=0.2111, over 3269130.95 frames. utt_duration=1228 frames, utt_pad_proportion=0.05906, over 10659.91 utterances.], batch size: 45, lr: 7.13e-03, grad_scale: 8.0 2023-03-08 12:46:06,902 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 12:46:13,893 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:47:15,176 INFO [train2.py:809] (3/4) Epoch 15, batch 3650, loss[ctc_loss=0.08054, att_loss=0.2209, loss=0.1928, over 16012.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007265, over 40.00 utterances.], tot_loss[ctc_loss=0.08659, att_loss=0.2425, loss=0.2113, over 3277065.90 frames. utt_duration=1232 frames, utt_pad_proportion=0.05754, over 10652.80 utterances.], batch size: 40, lr: 7.13e-03, grad_scale: 8.0 2023-03-08 12:47:24,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.155e+02 2.711e+02 3.466e+02 9.955e+02, threshold=5.422e+02, percent-clipped=4.0 2023-03-08 12:47:27,730 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:48:07,492 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1756, 5.0725, 5.1008, 2.1263, 1.8850, 2.7435, 2.5376, 3.8322], device='cuda:3'), covar=tensor([0.0640, 0.0271, 0.0189, 0.4940, 0.6301, 0.2778, 0.2875, 0.1724], device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0241, 0.0247, 0.0226, 0.0343, 0.0335, 0.0234, 0.0357], device='cuda:3'), out_proj_covar=tensor([1.4941e-04, 9.0431e-05, 1.0668e-04, 9.8322e-05, 1.4581e-04, 1.3288e-04, 9.3707e-05, 1.4758e-04], device='cuda:3') 2023-03-08 12:48:34,105 INFO [train2.py:809] (3/4) Epoch 15, batch 3700, loss[ctc_loss=0.1129, att_loss=0.2551, loss=0.2267, over 17003.00 frames. utt_duration=688.5 frames, utt_pad_proportion=0.1361, over 99.00 utterances.], tot_loss[ctc_loss=0.08629, att_loss=0.2417, loss=0.2106, over 3274250.28 frames. utt_duration=1240 frames, utt_pad_proportion=0.05695, over 10576.87 utterances.], batch size: 99, lr: 7.13e-03, grad_scale: 8.0 2023-03-08 12:48:43,279 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:49:20,223 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:49:49,033 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:49:53,481 INFO [train2.py:809] (3/4) Epoch 15, batch 3750, loss[ctc_loss=0.08133, att_loss=0.2305, loss=0.2007, over 16279.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.00731, over 43.00 utterances.], tot_loss[ctc_loss=0.08716, att_loss=0.2426, loss=0.2115, over 3276245.93 frames. utt_duration=1229 frames, utt_pad_proportion=0.05961, over 10680.17 utterances.], batch size: 43, lr: 7.12e-03, grad_scale: 8.0 2023-03-08 12:50:02,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 2.043e+02 2.443e+02 3.044e+02 5.040e+02, threshold=4.885e+02, percent-clipped=0.0 2023-03-08 12:50:54,597 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:50:54,744 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7008, 3.1590, 5.0983, 4.4698, 3.4102, 4.5245, 4.9259, 4.7232], device='cuda:3'), covar=tensor([0.0269, 0.1293, 0.0258, 0.0696, 0.1549, 0.0232, 0.0132, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0239, 0.0156, 0.0308, 0.0263, 0.0191, 0.0141, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 12:50:56,184 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:51:07,928 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2706, 5.4680, 5.8707, 5.6351, 5.7812, 6.1805, 5.3814, 6.2466], device='cuda:3'), covar=tensor([0.0632, 0.0717, 0.0640, 0.1099, 0.1463, 0.0940, 0.0562, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0803, 0.0471, 0.0550, 0.0612, 0.0803, 0.0564, 0.0456, 0.0544], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:51:12,185 INFO [train2.py:809] (3/4) Epoch 15, batch 3800, loss[ctc_loss=0.08078, att_loss=0.2301, loss=0.2002, over 16180.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006899, over 41.00 utterances.], tot_loss[ctc_loss=0.0872, att_loss=0.2423, loss=0.2113, over 3264017.02 frames. utt_duration=1218 frames, utt_pad_proportion=0.06438, over 10729.33 utterances.], batch size: 41, lr: 7.12e-03, grad_scale: 8.0 2023-03-08 12:51:15,854 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.26 vs. limit=5.0 2023-03-08 12:51:20,049 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8707, 3.6389, 3.0667, 3.2722, 3.9317, 3.5520, 2.6891, 4.2830], device='cuda:3'), covar=tensor([0.1121, 0.0578, 0.1190, 0.0810, 0.0698, 0.0736, 0.1045, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0199, 0.0216, 0.0190, 0.0256, 0.0228, 0.0191, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 12:51:22,361 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 12:51:51,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7179, 3.0333, 3.8523, 3.2118, 3.6774, 4.7082, 4.5578, 3.3016], device='cuda:3'), covar=tensor([0.0328, 0.1650, 0.1028, 0.1329, 0.1018, 0.0835, 0.0526, 0.1432], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0233, 0.0262, 0.0207, 0.0252, 0.0331, 0.0235, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:52:09,976 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:52:31,310 INFO [train2.py:809] (3/4) Epoch 15, batch 3850, loss[ctc_loss=0.07743, att_loss=0.2373, loss=0.2054, over 17528.00 frames. utt_duration=1018 frames, utt_pad_proportion=0.03994, over 69.00 utterances.], tot_loss[ctc_loss=0.08656, att_loss=0.2423, loss=0.2112, over 3266967.89 frames. utt_duration=1217 frames, utt_pad_proportion=0.06504, over 10752.49 utterances.], batch size: 69, lr: 7.12e-03, grad_scale: 8.0 2023-03-08 12:52:40,310 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.221e+02 2.684e+02 3.228e+02 7.729e+02, threshold=5.369e+02, percent-clipped=7.0 2023-03-08 12:52:59,636 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 12:53:12,592 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9884, 5.0044, 4.9704, 2.1835, 1.8903, 2.6923, 2.5470, 3.8309], device='cuda:3'), covar=tensor([0.0722, 0.0220, 0.0209, 0.4907, 0.6016, 0.2733, 0.3162, 0.1604], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0245, 0.0250, 0.0228, 0.0345, 0.0337, 0.0237, 0.0358], device='cuda:3'), out_proj_covar=tensor([1.4994e-04, 9.1441e-05, 1.0823e-04, 9.9400e-05, 1.4661e-04, 1.3384e-04, 9.4831e-05, 1.4798e-04], device='cuda:3') 2023-03-08 12:53:46,842 INFO [train2.py:809] (3/4) Epoch 15, batch 3900, loss[ctc_loss=0.05398, att_loss=0.2055, loss=0.1752, over 15766.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.008918, over 38.00 utterances.], tot_loss[ctc_loss=0.08684, att_loss=0.2423, loss=0.2112, over 3259868.22 frames. utt_duration=1218 frames, utt_pad_proportion=0.06626, over 10715.98 utterances.], batch size: 38, lr: 7.12e-03, grad_scale: 8.0 2023-03-08 12:54:10,327 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6024, 2.8175, 3.5142, 4.5018, 3.9717, 3.9225, 2.9377, 2.1343], device='cuda:3'), covar=tensor([0.0588, 0.2007, 0.0929, 0.0473, 0.0857, 0.0409, 0.1486, 0.2317], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0213, 0.0188, 0.0202, 0.0207, 0.0165, 0.0199, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 12:54:57,126 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 2023-03-08 12:55:03,847 INFO [train2.py:809] (3/4) Epoch 15, batch 3950, loss[ctc_loss=0.07502, att_loss=0.243, loss=0.2094, over 17033.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007869, over 51.00 utterances.], tot_loss[ctc_loss=0.08618, att_loss=0.2418, loss=0.2106, over 3252335.08 frames. utt_duration=1214 frames, utt_pad_proportion=0.07026, over 10729.07 utterances.], batch size: 51, lr: 7.11e-03, grad_scale: 8.0 2023-03-08 12:55:12,750 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.994e+02 2.418e+02 2.996e+02 5.766e+02, threshold=4.836e+02, percent-clipped=1.0 2023-03-08 12:56:20,784 INFO [train2.py:809] (3/4) Epoch 16, batch 0, loss[ctc_loss=0.07673, att_loss=0.2302, loss=0.1995, over 16007.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007218, over 40.00 utterances.], tot_loss[ctc_loss=0.07673, att_loss=0.2302, loss=0.1995, over 16007.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007218, over 40.00 utterances.], batch size: 40, lr: 6.88e-03, grad_scale: 8.0 2023-03-08 12:56:20,784 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 12:56:32,578 INFO [train2.py:843] (3/4) Epoch 16, validation: ctc_loss=0.04399, att_loss=0.2363, loss=0.1978, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 12:56:32,579 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 12:57:38,738 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1006, 4.4521, 4.3240, 4.4675, 4.4493, 4.1732, 3.1644, 4.2954], device='cuda:3'), covar=tensor([0.0126, 0.0119, 0.0127, 0.0079, 0.0099, 0.0120, 0.0673, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0079, 0.0098, 0.0062, 0.0066, 0.0078, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 12:57:52,428 INFO [train2.py:809] (3/4) Epoch 16, batch 50, loss[ctc_loss=0.0756, att_loss=0.2133, loss=0.1858, over 14098.00 frames. utt_duration=1821 frames, utt_pad_proportion=0.04874, over 31.00 utterances.], tot_loss[ctc_loss=0.08642, att_loss=0.2426, loss=0.2113, over 742099.48 frames. utt_duration=1191 frames, utt_pad_proportion=0.05872, over 2495.48 utterances.], batch size: 31, lr: 6.88e-03, grad_scale: 8.0 2023-03-08 12:58:13,656 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:58:27,021 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 2023-03-08 12:58:27,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.394e+02 2.112e+02 2.420e+02 3.068e+02 7.828e+02, threshold=4.840e+02, percent-clipped=6.0 2023-03-08 12:58:59,296 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:59:12,663 INFO [train2.py:809] (3/4) Epoch 16, batch 100, loss[ctc_loss=0.09146, att_loss=0.237, loss=0.2079, over 16627.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005274, over 47.00 utterances.], tot_loss[ctc_loss=0.0847, att_loss=0.2416, loss=0.2103, over 1308153.05 frames. utt_duration=1288 frames, utt_pad_proportion=0.03727, over 4068.22 utterances.], batch size: 47, lr: 6.88e-03, grad_scale: 8.0 2023-03-08 12:59:14,411 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:59:30,343 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 12:59:30,527 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9227, 4.9688, 4.5495, 2.3801, 4.6229, 4.5976, 4.0494, 2.1718], device='cuda:3'), covar=tensor([0.0143, 0.0112, 0.0365, 0.1707, 0.0132, 0.0238, 0.0509, 0.2496], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0093, 0.0090, 0.0107, 0.0076, 0.0101, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 13:00:03,271 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:00:14,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-08 13:00:26,553 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-08 13:00:31,897 INFO [train2.py:809] (3/4) Epoch 16, batch 150, loss[ctc_loss=0.08363, att_loss=0.2436, loss=0.2116, over 17277.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01325, over 55.00 utterances.], tot_loss[ctc_loss=0.08539, att_loss=0.2426, loss=0.2111, over 1748692.19 frames. utt_duration=1263 frames, utt_pad_proportion=0.04223, over 5543.45 utterances.], batch size: 55, lr: 6.87e-03, grad_scale: 8.0 2023-03-08 13:00:32,876 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.98 vs. limit=5.0 2023-03-08 13:00:35,479 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:01:06,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.211e+02 2.747e+02 3.339e+02 6.489e+02, threshold=5.494e+02, percent-clipped=4.0 2023-03-08 13:01:40,850 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:01:51,217 INFO [train2.py:809] (3/4) Epoch 16, batch 200, loss[ctc_loss=0.09948, att_loss=0.2551, loss=0.224, over 17334.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03583, over 63.00 utterances.], tot_loss[ctc_loss=0.08673, att_loss=0.2435, loss=0.2122, over 2085927.54 frames. utt_duration=1224 frames, utt_pad_proportion=0.05512, over 6827.14 utterances.], batch size: 63, lr: 6.87e-03, grad_scale: 8.0 2023-03-08 13:02:25,095 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9208, 5.1640, 4.7435, 5.2894, 4.6145, 4.8958, 5.3118, 5.0925], device='cuda:3'), covar=tensor([0.0566, 0.0276, 0.0710, 0.0274, 0.0447, 0.0255, 0.0223, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0289, 0.0339, 0.0301, 0.0297, 0.0223, 0.0277, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 13:03:15,168 INFO [train2.py:809] (3/4) Epoch 16, batch 250, loss[ctc_loss=0.1097, att_loss=0.2581, loss=0.2284, over 16483.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005544, over 46.00 utterances.], tot_loss[ctc_loss=0.08631, att_loss=0.2427, loss=0.2114, over 2345088.85 frames. utt_duration=1234 frames, utt_pad_proportion=0.05545, over 7611.91 utterances.], batch size: 46, lr: 6.87e-03, grad_scale: 8.0 2023-03-08 13:03:24,247 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0251, 3.6080, 3.6690, 3.1674, 3.6542, 3.7620, 3.6862, 2.8983], device='cuda:3'), covar=tensor([0.0855, 0.1682, 0.2526, 0.4354, 0.1492, 0.1808, 0.1136, 0.4512], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0153, 0.0164, 0.0229, 0.0126, 0.0218, 0.0139, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:03:50,523 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.138e+02 2.534e+02 3.199e+02 6.195e+02, threshold=5.067e+02, percent-clipped=2.0 2023-03-08 13:03:55,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-08 13:04:09,646 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:04:28,387 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:04:35,589 INFO [train2.py:809] (3/4) Epoch 16, batch 300, loss[ctc_loss=0.07955, att_loss=0.2443, loss=0.2114, over 16617.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005648, over 47.00 utterances.], tot_loss[ctc_loss=0.08601, att_loss=0.2419, loss=0.2108, over 2547005.58 frames. utt_duration=1242 frames, utt_pad_proportion=0.05574, over 8214.61 utterances.], batch size: 47, lr: 6.87e-03, grad_scale: 16.0 2023-03-08 13:04:49,180 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:05:09,403 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:05:27,671 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0320, 5.3352, 5.5392, 5.4466, 5.4335, 5.9313, 5.2426, 6.0253], device='cuda:3'), covar=tensor([0.0678, 0.0612, 0.0689, 0.1114, 0.1671, 0.0892, 0.0690, 0.0706], device='cuda:3'), in_proj_covar=tensor([0.0789, 0.0468, 0.0548, 0.0607, 0.0808, 0.0557, 0.0452, 0.0542], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:05:46,272 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 13:05:55,888 INFO [train2.py:809] (3/4) Epoch 16, batch 350, loss[ctc_loss=0.08168, att_loss=0.2491, loss=0.2156, over 17064.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009151, over 53.00 utterances.], tot_loss[ctc_loss=0.08623, att_loss=0.2419, loss=0.2108, over 2706541.85 frames. utt_duration=1246 frames, utt_pad_proportion=0.05525, over 8699.32 utterances.], batch size: 53, lr: 6.86e-03, grad_scale: 16.0 2023-03-08 13:06:05,770 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 13:06:26,356 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:06:30,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 2.169e+02 2.724e+02 3.771e+02 9.216e+02, threshold=5.447e+02, percent-clipped=7.0 2023-03-08 13:06:47,900 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:07:14,372 INFO [train2.py:809] (3/4) Epoch 16, batch 400, loss[ctc_loss=0.09957, att_loss=0.2636, loss=0.2308, over 16680.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006383, over 46.00 utterances.], tot_loss[ctc_loss=0.08631, att_loss=0.2418, loss=0.2107, over 2833324.44 frames. utt_duration=1271 frames, utt_pad_proportion=0.04818, over 8924.68 utterances.], batch size: 46, lr: 6.86e-03, grad_scale: 8.0 2023-03-08 13:07:16,900 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:08:16,878 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4661, 2.5607, 4.9037, 3.6709, 2.9440, 4.2204, 4.6513, 4.5388], device='cuda:3'), covar=tensor([0.0176, 0.1607, 0.0139, 0.1257, 0.1869, 0.0251, 0.0128, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0237, 0.0157, 0.0304, 0.0259, 0.0191, 0.0139, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:08:19,861 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:08:29,143 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:08:29,283 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5423, 2.9383, 3.7323, 2.9071, 3.6535, 4.6798, 4.4447, 3.1495], device='cuda:3'), covar=tensor([0.0379, 0.1827, 0.1034, 0.1523, 0.1102, 0.0884, 0.0577, 0.1405], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0237, 0.0264, 0.0209, 0.0254, 0.0335, 0.0240, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:08:32,213 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:08:33,621 INFO [train2.py:809] (3/4) Epoch 16, batch 450, loss[ctc_loss=0.08119, att_loss=0.2412, loss=0.2092, over 16959.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007816, over 50.00 utterances.], tot_loss[ctc_loss=0.08571, att_loss=0.2416, loss=0.2104, over 2934800.81 frames. utt_duration=1271 frames, utt_pad_proportion=0.04769, over 9250.28 utterances.], batch size: 50, lr: 6.86e-03, grad_scale: 8.0 2023-03-08 13:08:45,925 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1489, 5.3016, 5.0989, 2.3961, 2.1216, 3.3132, 2.7209, 3.8241], device='cuda:3'), covar=tensor([0.0697, 0.0262, 0.0242, 0.4971, 0.5631, 0.2086, 0.3034, 0.1819], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0241, 0.0246, 0.0224, 0.0340, 0.0330, 0.0231, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4803e-04, 9.0050e-05, 1.0578e-04, 9.7424e-05, 1.4427e-04, 1.3089e-04, 9.2666e-05, 1.4558e-04], device='cuda:3') 2023-03-08 13:09:10,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.196e+02 2.695e+02 3.219e+02 5.768e+02, threshold=5.391e+02, percent-clipped=1.0 2023-03-08 13:09:34,763 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:09:54,248 INFO [train2.py:809] (3/4) Epoch 16, batch 500, loss[ctc_loss=0.09107, att_loss=0.2545, loss=0.2218, over 17030.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007278, over 51.00 utterances.], tot_loss[ctc_loss=0.08465, att_loss=0.2413, loss=0.21, over 3016035.85 frames. utt_duration=1269 frames, utt_pad_proportion=0.04663, over 9520.31 utterances.], batch size: 51, lr: 6.85e-03, grad_scale: 8.0 2023-03-08 13:09:57,679 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:10:09,686 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7004, 2.8818, 5.0550, 4.2187, 3.1242, 4.5040, 4.8368, 4.6733], device='cuda:3'), covar=tensor([0.0305, 0.1492, 0.0260, 0.0840, 0.1736, 0.0218, 0.0172, 0.0294], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0239, 0.0159, 0.0306, 0.0261, 0.0192, 0.0140, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:10:24,272 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.13 vs. limit=5.0 2023-03-08 13:10:30,023 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5413, 2.8485, 3.7306, 2.8946, 3.6508, 4.6534, 4.3883, 3.2014], device='cuda:3'), covar=tensor([0.0348, 0.1785, 0.0915, 0.1406, 0.0892, 0.0927, 0.0568, 0.1290], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0236, 0.0263, 0.0208, 0.0253, 0.0335, 0.0240, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:11:01,379 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7140, 2.5826, 5.1146, 4.1808, 3.0417, 4.3714, 4.8681, 4.7127], device='cuda:3'), covar=tensor([0.0222, 0.1609, 0.0168, 0.0822, 0.1689, 0.0215, 0.0126, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0237, 0.0157, 0.0304, 0.0259, 0.0191, 0.0139, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:11:14,953 INFO [train2.py:809] (3/4) Epoch 16, batch 550, loss[ctc_loss=0.09498, att_loss=0.2574, loss=0.2249, over 17137.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01405, over 56.00 utterances.], tot_loss[ctc_loss=0.08479, att_loss=0.2413, loss=0.21, over 3073788.36 frames. utt_duration=1277 frames, utt_pad_proportion=0.04488, over 9636.74 utterances.], batch size: 56, lr: 6.85e-03, grad_scale: 8.0 2023-03-08 13:11:20,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-08 13:11:30,067 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7227, 2.6197, 5.0817, 4.1091, 3.0997, 4.3409, 4.8875, 4.7133], device='cuda:3'), covar=tensor([0.0245, 0.1587, 0.0216, 0.0912, 0.1713, 0.0226, 0.0136, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0238, 0.0158, 0.0305, 0.0260, 0.0192, 0.0140, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:11:51,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 1.986e+02 2.421e+02 3.099e+02 6.630e+02, threshold=4.842e+02, percent-clipped=3.0 2023-03-08 13:12:31,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 13:12:35,538 INFO [train2.py:809] (3/4) Epoch 16, batch 600, loss[ctc_loss=0.08856, att_loss=0.2522, loss=0.2195, over 17063.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.007655, over 52.00 utterances.], tot_loss[ctc_loss=0.08454, att_loss=0.2401, loss=0.209, over 3103237.86 frames. utt_duration=1274 frames, utt_pad_proportion=0.05129, over 9752.44 utterances.], batch size: 52, lr: 6.85e-03, grad_scale: 8.0 2023-03-08 13:13:39,238 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 13:13:57,394 INFO [train2.py:809] (3/4) Epoch 16, batch 650, loss[ctc_loss=0.08757, att_loss=0.2266, loss=0.1988, over 14506.00 frames. utt_duration=1815 frames, utt_pad_proportion=0.04535, over 32.00 utterances.], tot_loss[ctc_loss=0.0851, att_loss=0.2407, loss=0.2096, over 3144462.46 frames. utt_duration=1276 frames, utt_pad_proportion=0.04998, over 9871.23 utterances.], batch size: 32, lr: 6.85e-03, grad_scale: 8.0 2023-03-08 13:13:59,132 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 13:14:20,229 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:14:23,393 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0490, 5.3366, 4.8978, 5.3597, 4.7375, 5.0112, 5.5326, 5.2789], device='cuda:3'), covar=tensor([0.0591, 0.0334, 0.0776, 0.0278, 0.0425, 0.0237, 0.0203, 0.0189], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0292, 0.0345, 0.0303, 0.0299, 0.0227, 0.0280, 0.0263], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 13:14:33,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.189e+02 2.670e+02 3.387e+02 7.027e+02, threshold=5.339e+02, percent-clipped=7.0 2023-03-08 13:14:40,861 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:15:17,212 INFO [train2.py:809] (3/4) Epoch 16, batch 700, loss[ctc_loss=0.06431, att_loss=0.2129, loss=0.1832, over 15792.00 frames. utt_duration=1664 frames, utt_pad_proportion=0.007144, over 38.00 utterances.], tot_loss[ctc_loss=0.08508, att_loss=0.2407, loss=0.2096, over 3177680.74 frames. utt_duration=1270 frames, utt_pad_proportion=0.05009, over 10024.02 utterances.], batch size: 38, lr: 6.84e-03, grad_scale: 8.0 2023-03-08 13:15:46,270 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-08 13:15:54,361 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7866, 5.1783, 5.0540, 5.2167, 5.2964, 4.9029, 3.9825, 5.1615], device='cuda:3'), covar=tensor([0.0087, 0.0083, 0.0101, 0.0062, 0.0066, 0.0109, 0.0564, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0080, 0.0098, 0.0062, 0.0067, 0.0079, 0.0097, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:16:33,166 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:16:37,556 INFO [train2.py:809] (3/4) Epoch 16, batch 750, loss[ctc_loss=0.109, att_loss=0.261, loss=0.2306, over 17341.00 frames. utt_duration=1007 frames, utt_pad_proportion=0.04859, over 69.00 utterances.], tot_loss[ctc_loss=0.08524, att_loss=0.2416, loss=0.2103, over 3202273.52 frames. utt_duration=1261 frames, utt_pad_proportion=0.05102, over 10172.11 utterances.], batch size: 69, lr: 6.84e-03, grad_scale: 8.0 2023-03-08 13:17:08,905 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0365, 5.0357, 4.9952, 2.0791, 1.9752, 3.0265, 2.2940, 3.7477], device='cuda:3'), covar=tensor([0.0678, 0.0236, 0.0204, 0.4866, 0.5706, 0.2246, 0.3170, 0.1782], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0241, 0.0246, 0.0223, 0.0341, 0.0332, 0.0233, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.4788e-04, 8.9975e-05, 1.0562e-04, 9.6911e-05, 1.4473e-04, 1.3144e-04, 9.3209e-05, 1.4569e-04], device='cuda:3') 2023-03-08 13:17:14,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 2.152e+02 2.643e+02 3.369e+02 1.101e+03, threshold=5.285e+02, percent-clipped=4.0 2023-03-08 13:17:40,023 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:17:40,195 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3298, 2.3756, 4.7109, 3.7088, 2.8973, 4.1232, 4.3863, 4.3861], device='cuda:3'), covar=tensor([0.0203, 0.1793, 0.0129, 0.0892, 0.1752, 0.0250, 0.0142, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0239, 0.0159, 0.0306, 0.0261, 0.0193, 0.0141, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:17:50,565 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:17:54,415 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:17:58,933 INFO [train2.py:809] (3/4) Epoch 16, batch 800, loss[ctc_loss=0.09958, att_loss=0.2549, loss=0.2239, over 16900.00 frames. utt_duration=691.4 frames, utt_pad_proportion=0.1336, over 98.00 utterances.], tot_loss[ctc_loss=0.0852, att_loss=0.2417, loss=0.2104, over 3226210.34 frames. utt_duration=1269 frames, utt_pad_proportion=0.04617, over 10182.48 utterances.], batch size: 98, lr: 6.84e-03, grad_scale: 8.0 2023-03-08 13:18:38,569 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5413, 3.7011, 2.9171, 3.1089, 3.8018, 3.3931, 2.3407, 4.0711], device='cuda:3'), covar=tensor([0.1334, 0.0532, 0.1147, 0.0912, 0.0800, 0.0776, 0.1300, 0.0599], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0199, 0.0214, 0.0187, 0.0257, 0.0227, 0.0190, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 13:18:54,146 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-08 13:18:56,199 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:19:18,473 INFO [train2.py:809] (3/4) Epoch 16, batch 850, loss[ctc_loss=0.0812, att_loss=0.2408, loss=0.2089, over 17405.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04576, over 69.00 utterances.], tot_loss[ctc_loss=0.08477, att_loss=0.241, loss=0.2098, over 3233319.07 frames. utt_duration=1274 frames, utt_pad_proportion=0.04559, over 10160.98 utterances.], batch size: 69, lr: 6.83e-03, grad_scale: 8.0 2023-03-08 13:19:49,046 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3229, 5.2469, 5.1672, 3.1161, 5.0468, 4.9360, 4.5833, 2.9002], device='cuda:3'), covar=tensor([0.0128, 0.0084, 0.0204, 0.0943, 0.0081, 0.0161, 0.0277, 0.1339], device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0094, 0.0091, 0.0108, 0.0077, 0.0104, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 13:19:55,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.256e+02 2.159e+02 2.689e+02 3.192e+02 6.800e+02, threshold=5.378e+02, percent-clipped=2.0 2023-03-08 13:20:38,473 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5650, 2.9100, 2.9574, 2.6027, 3.0021, 2.9820, 2.9416, 2.1035], device='cuda:3'), covar=tensor([0.0984, 0.2315, 0.2394, 0.4890, 0.1354, 0.2820, 0.1402, 0.5372], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0155, 0.0165, 0.0229, 0.0130, 0.0221, 0.0140, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 13:20:39,634 INFO [train2.py:809] (3/4) Epoch 16, batch 900, loss[ctc_loss=0.09285, att_loss=0.2592, loss=0.226, over 16783.00 frames. utt_duration=679.5 frames, utt_pad_proportion=0.1442, over 99.00 utterances.], tot_loss[ctc_loss=0.08495, att_loss=0.2408, loss=0.2096, over 3234696.54 frames. utt_duration=1261 frames, utt_pad_proportion=0.05189, over 10272.00 utterances.], batch size: 99, lr: 6.83e-03, grad_scale: 8.0 2023-03-08 13:21:37,353 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3190, 4.7657, 4.6828, 4.8121, 4.9013, 4.5345, 3.5737, 4.7658], device='cuda:3'), covar=tensor([0.0125, 0.0108, 0.0115, 0.0081, 0.0085, 0.0119, 0.0611, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0079, 0.0098, 0.0062, 0.0067, 0.0079, 0.0097, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:21:42,277 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:22:00,167 INFO [train2.py:809] (3/4) Epoch 16, batch 950, loss[ctc_loss=0.08632, att_loss=0.2558, loss=0.2219, over 16781.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005697, over 48.00 utterances.], tot_loss[ctc_loss=0.08515, att_loss=0.241, loss=0.2098, over 3246591.12 frames. utt_duration=1270 frames, utt_pad_proportion=0.04875, over 10240.58 utterances.], batch size: 48, lr: 6.83e-03, grad_scale: 8.0 2023-03-08 13:22:02,072 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:22:18,044 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:22:22,536 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:22:36,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.437e+02 2.156e+02 2.670e+02 3.136e+02 6.539e+02, threshold=5.340e+02, percent-clipped=2.0 2023-03-08 13:22:43,835 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:22:53,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 13:22:59,110 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:23:04,023 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6792, 5.0428, 5.2280, 5.0649, 5.1008, 5.5552, 5.0094, 5.6680], device='cuda:3'), covar=tensor([0.0696, 0.0762, 0.0826, 0.1173, 0.1915, 0.1018, 0.0773, 0.0764], device='cuda:3'), in_proj_covar=tensor([0.0793, 0.0472, 0.0551, 0.0608, 0.0811, 0.0568, 0.0451, 0.0540], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:23:14,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-03-08 13:23:19,144 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:23:20,644 INFO [train2.py:809] (3/4) Epoch 16, batch 1000, loss[ctc_loss=0.07799, att_loss=0.2492, loss=0.215, over 16633.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004961, over 47.00 utterances.], tot_loss[ctc_loss=0.08569, att_loss=0.2418, loss=0.2106, over 3261688.89 frames. utt_duration=1238 frames, utt_pad_proportion=0.05317, over 10547.65 utterances.], batch size: 47, lr: 6.83e-03, grad_scale: 8.0 2023-03-08 13:23:37,885 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3509, 2.2282, 3.5075, 2.3244, 3.3417, 4.5986, 4.5185, 2.7677], device='cuda:3'), covar=tensor([0.0440, 0.2372, 0.0957, 0.1842, 0.0962, 0.0641, 0.0482, 0.1696], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0237, 0.0265, 0.0209, 0.0254, 0.0339, 0.0242, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:23:39,233 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:23:55,709 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:24:00,514 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:24:40,429 INFO [train2.py:809] (3/4) Epoch 16, batch 1050, loss[ctc_loss=0.0881, att_loss=0.2199, loss=0.1935, over 15402.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.008237, over 35.00 utterances.], tot_loss[ctc_loss=0.08658, att_loss=0.2422, loss=0.2111, over 3258764.40 frames. utt_duration=1238 frames, utt_pad_proportion=0.05489, over 10539.94 utterances.], batch size: 35, lr: 6.82e-03, grad_scale: 8.0 2023-03-08 13:25:17,107 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.141e+02 2.597e+02 3.263e+02 1.129e+03, threshold=5.194e+02, percent-clipped=2.0 2023-03-08 13:25:56,056 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:26:00,450 INFO [train2.py:809] (3/4) Epoch 16, batch 1100, loss[ctc_loss=0.07533, att_loss=0.2384, loss=0.2058, over 17052.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008032, over 52.00 utterances.], tot_loss[ctc_loss=0.08665, att_loss=0.2423, loss=0.2112, over 3265598.26 frames. utt_duration=1220 frames, utt_pad_proportion=0.05893, over 10716.97 utterances.], batch size: 52, lr: 6.82e-03, grad_scale: 8.0 2023-03-08 13:27:12,936 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:27:21,209 INFO [train2.py:809] (3/4) Epoch 16, batch 1150, loss[ctc_loss=0.08956, att_loss=0.2527, loss=0.2201, over 16872.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007141, over 49.00 utterances.], tot_loss[ctc_loss=0.08623, att_loss=0.2415, loss=0.2104, over 3252198.48 frames. utt_duration=1229 frames, utt_pad_proportion=0.06121, over 10594.35 utterances.], batch size: 49, lr: 6.82e-03, grad_scale: 8.0 2023-03-08 13:27:37,513 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9823, 4.6969, 4.8352, 2.2939, 1.9593, 2.7780, 2.2257, 3.7439], device='cuda:3'), covar=tensor([0.0709, 0.0250, 0.0196, 0.4843, 0.6058, 0.2663, 0.3530, 0.1596], device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0242, 0.0246, 0.0226, 0.0341, 0.0333, 0.0234, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.4855e-04, 9.0281e-05, 1.0539e-04, 9.7986e-05, 1.4460e-04, 1.3188e-04, 9.4045e-05, 1.4581e-04], device='cuda:3') 2023-03-08 13:27:59,214 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 2.021e+02 2.572e+02 3.109e+02 6.628e+02, threshold=5.143e+02, percent-clipped=1.0 2023-03-08 13:28:42,079 INFO [train2.py:809] (3/4) Epoch 16, batch 1200, loss[ctc_loss=0.09103, att_loss=0.2508, loss=0.2189, over 17524.00 frames. utt_duration=888.7 frames, utt_pad_proportion=0.07039, over 79.00 utterances.], tot_loss[ctc_loss=0.08597, att_loss=0.2415, loss=0.2104, over 3265245.77 frames. utt_duration=1226 frames, utt_pad_proportion=0.05858, over 10665.40 utterances.], batch size: 79, lr: 6.81e-03, grad_scale: 8.0 2023-03-08 13:28:49,194 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-08 13:29:59,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.24 vs. limit=5.0 2023-03-08 13:30:02,897 INFO [train2.py:809] (3/4) Epoch 16, batch 1250, loss[ctc_loss=0.09035, att_loss=0.2294, loss=0.2016, over 16015.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006784, over 40.00 utterances.], tot_loss[ctc_loss=0.08678, att_loss=0.2422, loss=0.2111, over 3264530.81 frames. utt_duration=1197 frames, utt_pad_proportion=0.06798, over 10920.11 utterances.], batch size: 40, lr: 6.81e-03, grad_scale: 8.0 2023-03-08 13:30:23,530 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0767, 2.5802, 3.5036, 2.5870, 3.3216, 4.3108, 4.1111, 3.0799], device='cuda:3'), covar=tensor([0.0509, 0.2041, 0.1261, 0.1721, 0.1183, 0.1033, 0.0705, 0.1355], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0236, 0.0265, 0.0208, 0.0253, 0.0336, 0.0242, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:30:39,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.300e+02 1.974e+02 2.401e+02 2.960e+02 6.292e+02, threshold=4.803e+02, percent-clipped=4.0 2023-03-08 13:30:47,910 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7349, 6.0838, 5.4349, 5.8364, 5.7333, 5.2725, 5.4084, 5.2525], device='cuda:3'), covar=tensor([0.1350, 0.0851, 0.0874, 0.0779, 0.0803, 0.1270, 0.2209, 0.2448], device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0550, 0.0418, 0.0416, 0.0395, 0.0435, 0.0563, 0.0499], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 13:30:59,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-08 13:31:22,899 INFO [train2.py:809] (3/4) Epoch 16, batch 1300, loss[ctc_loss=0.06285, att_loss=0.211, loss=0.1813, over 15361.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01124, over 35.00 utterances.], tot_loss[ctc_loss=0.08568, att_loss=0.2412, loss=0.2101, over 3259459.01 frames. utt_duration=1232 frames, utt_pad_proportion=0.06134, over 10596.00 utterances.], batch size: 35, lr: 6.81e-03, grad_scale: 8.0 2023-03-08 13:31:29,591 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:31:50,217 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:31:51,891 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2543, 4.6359, 4.6184, 4.6095, 4.7157, 4.4780, 3.1600, 4.4224], device='cuda:3'), covar=tensor([0.0135, 0.0137, 0.0138, 0.0106, 0.0104, 0.0117, 0.0842, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0080, 0.0099, 0.0062, 0.0067, 0.0079, 0.0098, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:32:42,648 INFO [train2.py:809] (3/4) Epoch 16, batch 1350, loss[ctc_loss=0.09401, att_loss=0.2482, loss=0.2174, over 16877.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007723, over 49.00 utterances.], tot_loss[ctc_loss=0.08514, att_loss=0.2407, loss=0.2096, over 3261469.42 frames. utt_duration=1257 frames, utt_pad_proportion=0.05423, over 10393.02 utterances.], batch size: 49, lr: 6.81e-03, grad_scale: 8.0 2023-03-08 13:33:06,266 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:33:18,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.149e+02 2.520e+02 3.082e+02 5.998e+02, threshold=5.039e+02, percent-clipped=2.0 2023-03-08 13:34:02,072 INFO [train2.py:809] (3/4) Epoch 16, batch 1400, loss[ctc_loss=0.0775, att_loss=0.2437, loss=0.2105, over 16866.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.007567, over 49.00 utterances.], tot_loss[ctc_loss=0.08524, att_loss=0.2412, loss=0.21, over 3266691.21 frames. utt_duration=1257 frames, utt_pad_proportion=0.05395, over 10409.81 utterances.], batch size: 49, lr: 6.80e-03, grad_scale: 8.0 2023-03-08 13:34:11,336 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7434, 3.9754, 3.9597, 3.9848, 4.0611, 3.8258, 3.0337, 3.8578], device='cuda:3'), covar=tensor([0.0132, 0.0113, 0.0129, 0.0090, 0.0083, 0.0134, 0.0594, 0.0228], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0080, 0.0099, 0.0063, 0.0067, 0.0079, 0.0098, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:35:21,832 INFO [train2.py:809] (3/4) Epoch 16, batch 1450, loss[ctc_loss=0.1474, att_loss=0.2744, loss=0.249, over 13715.00 frames. utt_duration=377.3 frames, utt_pad_proportion=0.3415, over 146.00 utterances.], tot_loss[ctc_loss=0.08569, att_loss=0.2415, loss=0.2103, over 3265788.70 frames. utt_duration=1246 frames, utt_pad_proportion=0.05753, over 10493.32 utterances.], batch size: 146, lr: 6.80e-03, grad_scale: 8.0 2023-03-08 13:35:58,473 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.227e+02 2.591e+02 3.129e+02 8.266e+02, threshold=5.182e+02, percent-clipped=3.0 2023-03-08 13:36:40,849 INFO [train2.py:809] (3/4) Epoch 16, batch 1500, loss[ctc_loss=0.08695, att_loss=0.233, loss=0.2038, over 16287.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007021, over 43.00 utterances.], tot_loss[ctc_loss=0.08595, att_loss=0.2414, loss=0.2103, over 3269598.32 frames. utt_duration=1274 frames, utt_pad_proportion=0.04993, over 10278.98 utterances.], batch size: 43, lr: 6.80e-03, grad_scale: 8.0 2023-03-08 13:37:59,314 INFO [train2.py:809] (3/4) Epoch 16, batch 1550, loss[ctc_loss=0.0787, att_loss=0.2473, loss=0.2136, over 17303.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02417, over 59.00 utterances.], tot_loss[ctc_loss=0.08603, att_loss=0.2414, loss=0.2103, over 3267155.29 frames. utt_duration=1253 frames, utt_pad_proportion=0.0554, over 10438.46 utterances.], batch size: 59, lr: 6.80e-03, grad_scale: 8.0 2023-03-08 13:38:35,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 2.094e+02 2.601e+02 3.579e+02 7.093e+02, threshold=5.201e+02, percent-clipped=2.0 2023-03-08 13:39:03,028 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9131, 5.2937, 4.7982, 5.3242, 4.7392, 4.9225, 5.4114, 5.1462], device='cuda:3'), covar=tensor([0.0665, 0.0235, 0.0818, 0.0255, 0.0412, 0.0262, 0.0205, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0293, 0.0348, 0.0306, 0.0301, 0.0226, 0.0280, 0.0265], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 13:39:18,917 INFO [train2.py:809] (3/4) Epoch 16, batch 1600, loss[ctc_loss=0.05567, att_loss=0.2086, loss=0.178, over 15858.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.009623, over 39.00 utterances.], tot_loss[ctc_loss=0.08622, att_loss=0.2414, loss=0.2103, over 3262222.63 frames. utt_duration=1228 frames, utt_pad_proportion=0.06351, over 10641.08 utterances.], batch size: 39, lr: 6.79e-03, grad_scale: 8.0 2023-03-08 13:39:46,413 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:40:01,863 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0138, 5.0974, 4.9667, 2.4238, 2.1007, 3.0013, 2.6544, 3.7689], device='cuda:3'), covar=tensor([0.0768, 0.0284, 0.0274, 0.4707, 0.5771, 0.2440, 0.3201, 0.1899], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0244, 0.0249, 0.0227, 0.0343, 0.0336, 0.0236, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.5031e-04, 9.1250e-05, 1.0717e-04, 9.8629e-05, 1.4562e-04, 1.3295e-04, 9.4692e-05, 1.4689e-04], device='cuda:3') 2023-03-08 13:40:38,374 INFO [train2.py:809] (3/4) Epoch 16, batch 1650, loss[ctc_loss=0.09291, att_loss=0.2365, loss=0.2078, over 15941.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007249, over 41.00 utterances.], tot_loss[ctc_loss=0.0859, att_loss=0.2417, loss=0.2106, over 3269964.82 frames. utt_duration=1225 frames, utt_pad_proportion=0.06236, over 10692.65 utterances.], batch size: 41, lr: 6.79e-03, grad_scale: 8.0 2023-03-08 13:40:54,211 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:41:02,526 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:41:05,735 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0293, 5.4000, 4.8125, 5.4961, 4.8550, 5.1190, 5.5309, 5.2295], device='cuda:3'), covar=tensor([0.0598, 0.0239, 0.0903, 0.0235, 0.0364, 0.0181, 0.0215, 0.0196], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0292, 0.0345, 0.0305, 0.0300, 0.0225, 0.0278, 0.0264], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 13:41:14,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 2.105e+02 2.549e+02 2.944e+02 7.207e+02, threshold=5.098e+02, percent-clipped=3.0 2023-03-08 13:41:56,933 INFO [train2.py:809] (3/4) Epoch 16, batch 1700, loss[ctc_loss=0.09383, att_loss=0.2336, loss=0.2056, over 16181.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006274, over 41.00 utterances.], tot_loss[ctc_loss=0.08603, att_loss=0.2422, loss=0.211, over 3266467.40 frames. utt_duration=1231 frames, utt_pad_proportion=0.06258, over 10630.41 utterances.], batch size: 41, lr: 6.79e-03, grad_scale: 8.0 2023-03-08 13:42:53,815 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6750, 4.6390, 4.4166, 2.7747, 4.4163, 4.3406, 4.0303, 2.6631], device='cuda:3'), covar=tensor([0.0111, 0.0107, 0.0277, 0.1064, 0.0106, 0.0239, 0.0312, 0.1392], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0094, 0.0093, 0.0109, 0.0079, 0.0106, 0.0097, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 13:43:16,832 INFO [train2.py:809] (3/4) Epoch 16, batch 1750, loss[ctc_loss=0.07226, att_loss=0.2306, loss=0.199, over 16277.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007341, over 43.00 utterances.], tot_loss[ctc_loss=0.08586, att_loss=0.2416, loss=0.2104, over 3267199.96 frames. utt_duration=1235 frames, utt_pad_proportion=0.06095, over 10594.76 utterances.], batch size: 43, lr: 6.78e-03, grad_scale: 8.0 2023-03-08 13:43:53,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 1.998e+02 2.470e+02 3.033e+02 6.251e+02, threshold=4.940e+02, percent-clipped=1.0 2023-03-08 13:44:15,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-03-08 13:44:36,284 INFO [train2.py:809] (3/4) Epoch 16, batch 1800, loss[ctc_loss=0.1061, att_loss=0.2327, loss=0.2074, over 15657.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008165, over 37.00 utterances.], tot_loss[ctc_loss=0.08638, att_loss=0.242, loss=0.2109, over 3272192.74 frames. utt_duration=1204 frames, utt_pad_proportion=0.06714, over 10881.01 utterances.], batch size: 37, lr: 6.78e-03, grad_scale: 8.0 2023-03-08 13:45:38,798 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9412, 6.1403, 5.5920, 5.8913, 5.8022, 5.3180, 5.6374, 5.2871], device='cuda:3'), covar=tensor([0.1161, 0.1080, 0.0816, 0.0907, 0.0967, 0.1689, 0.2230, 0.2642], device='cuda:3'), in_proj_covar=tensor([0.0484, 0.0550, 0.0419, 0.0423, 0.0403, 0.0440, 0.0569, 0.0498], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 13:45:55,243 INFO [train2.py:809] (3/4) Epoch 16, batch 1850, loss[ctc_loss=0.06809, att_loss=0.2233, loss=0.1923, over 16006.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007203, over 40.00 utterances.], tot_loss[ctc_loss=0.08645, att_loss=0.242, loss=0.2109, over 3264159.72 frames. utt_duration=1211 frames, utt_pad_proportion=0.0676, over 10792.10 utterances.], batch size: 40, lr: 6.78e-03, grad_scale: 8.0 2023-03-08 13:46:12,164 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-08 13:46:31,941 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.233e+02 2.629e+02 3.133e+02 5.185e+02, threshold=5.258e+02, percent-clipped=1.0 2023-03-08 13:47:14,835 INFO [train2.py:809] (3/4) Epoch 16, batch 1900, loss[ctc_loss=0.07973, att_loss=0.2257, loss=0.1965, over 16009.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007833, over 40.00 utterances.], tot_loss[ctc_loss=0.08748, att_loss=0.243, loss=0.2119, over 3268405.68 frames. utt_duration=1224 frames, utt_pad_proportion=0.06293, over 10694.21 utterances.], batch size: 40, lr: 6.78e-03, grad_scale: 8.0 2023-03-08 13:48:33,458 INFO [train2.py:809] (3/4) Epoch 16, batch 1950, loss[ctc_loss=0.08075, att_loss=0.2446, loss=0.2118, over 17108.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.0158, over 56.00 utterances.], tot_loss[ctc_loss=0.08572, att_loss=0.2418, loss=0.2106, over 3271061.02 frames. utt_duration=1256 frames, utt_pad_proportion=0.05469, over 10431.33 utterances.], batch size: 56, lr: 6.77e-03, grad_scale: 8.0 2023-03-08 13:48:49,376 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:49:09,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.102e+02 2.537e+02 2.891e+02 6.099e+02, threshold=5.073e+02, percent-clipped=2.0 2023-03-08 13:49:13,300 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:49:17,996 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9655, 5.1391, 4.9773, 2.3110, 2.0601, 2.9185, 2.5315, 3.9744], device='cuda:3'), covar=tensor([0.0788, 0.0243, 0.0236, 0.4691, 0.5789, 0.2437, 0.3252, 0.1484], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0243, 0.0247, 0.0225, 0.0340, 0.0333, 0.0235, 0.0353], device='cuda:3'), out_proj_covar=tensor([1.4877e-04, 9.0657e-05, 1.0627e-04, 9.7702e-05, 1.4422e-04, 1.3168e-04, 9.4527e-05, 1.4588e-04], device='cuda:3') 2023-03-08 13:49:28,766 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6065, 4.8012, 4.7829, 4.8274, 4.8691, 4.8363, 4.5375, 4.4000], device='cuda:3'), covar=tensor([0.0929, 0.0555, 0.0308, 0.0472, 0.0287, 0.0330, 0.0363, 0.0328], device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0331, 0.0301, 0.0325, 0.0385, 0.0405, 0.0327, 0.0364], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 13:49:30,481 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:49:53,041 INFO [train2.py:809] (3/4) Epoch 16, batch 2000, loss[ctc_loss=0.06473, att_loss=0.233, loss=0.1993, over 16166.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007558, over 41.00 utterances.], tot_loss[ctc_loss=0.08424, att_loss=0.2411, loss=0.2098, over 3280001.55 frames. utt_duration=1268 frames, utt_pad_proportion=0.04859, over 10361.25 utterances.], batch size: 41, lr: 6.77e-03, grad_scale: 8.0 2023-03-08 13:50:05,442 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:50:37,620 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8995, 5.0784, 5.0959, 5.0613, 5.1716, 5.1456, 4.7685, 4.6475], device='cuda:3'), covar=tensor([0.0997, 0.0611, 0.0287, 0.0535, 0.0334, 0.0318, 0.0384, 0.0347], device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0331, 0.0301, 0.0326, 0.0385, 0.0404, 0.0327, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 13:50:49,954 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:51:08,277 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:51:12,561 INFO [train2.py:809] (3/4) Epoch 16, batch 2050, loss[ctc_loss=0.1059, att_loss=0.2488, loss=0.2203, over 16128.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.006141, over 42.00 utterances.], tot_loss[ctc_loss=0.0852, att_loss=0.2412, loss=0.21, over 3262400.14 frames. utt_duration=1242 frames, utt_pad_proportion=0.06051, over 10523.69 utterances.], batch size: 42, lr: 6.77e-03, grad_scale: 8.0 2023-03-08 13:51:48,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.229e+02 2.633e+02 3.391e+02 8.027e+02, threshold=5.266e+02, percent-clipped=7.0 2023-03-08 13:52:32,059 INFO [train2.py:809] (3/4) Epoch 16, batch 2100, loss[ctc_loss=0.08534, att_loss=0.2261, loss=0.198, over 15511.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008035, over 36.00 utterances.], tot_loss[ctc_loss=0.08594, att_loss=0.2421, loss=0.2109, over 3265200.07 frames. utt_duration=1199 frames, utt_pad_proportion=0.06937, over 10909.33 utterances.], batch size: 36, lr: 6.77e-03, grad_scale: 8.0 2023-03-08 13:53:12,287 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-08 13:53:27,054 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 13:53:47,305 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.53 vs. limit=5.0 2023-03-08 13:53:51,105 INFO [train2.py:809] (3/4) Epoch 16, batch 2150, loss[ctc_loss=0.08903, att_loss=0.2304, loss=0.2022, over 16110.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.007311, over 42.00 utterances.], tot_loss[ctc_loss=0.08664, att_loss=0.2423, loss=0.2112, over 3260450.89 frames. utt_duration=1199 frames, utt_pad_proportion=0.07083, over 10888.10 utterances.], batch size: 42, lr: 6.76e-03, grad_scale: 8.0 2023-03-08 13:54:03,322 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:54:27,519 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.191e+02 2.731e+02 3.267e+02 7.309e+02, threshold=5.461e+02, percent-clipped=1.0 2023-03-08 13:54:48,290 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3777, 4.6434, 4.2075, 4.6968, 4.1377, 4.3554, 4.7263, 4.5475], device='cuda:3'), covar=tensor([0.0585, 0.0310, 0.0823, 0.0313, 0.0483, 0.0424, 0.0235, 0.0204], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0289, 0.0340, 0.0303, 0.0297, 0.0221, 0.0275, 0.0258], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 13:55:11,211 INFO [train2.py:809] (3/4) Epoch 16, batch 2200, loss[ctc_loss=0.07843, att_loss=0.2177, loss=0.1898, over 15500.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008721, over 36.00 utterances.], tot_loss[ctc_loss=0.08544, att_loss=0.2416, loss=0.2104, over 3262865.76 frames. utt_duration=1207 frames, utt_pad_proportion=0.06812, over 10823.15 utterances.], batch size: 36, lr: 6.76e-03, grad_scale: 8.0 2023-03-08 13:55:40,185 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:56:00,987 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9655, 5.0450, 4.8620, 1.9721, 1.9839, 2.8632, 2.5392, 3.8710], device='cuda:3'), covar=tensor([0.0760, 0.0243, 0.0235, 0.5556, 0.5955, 0.2537, 0.3210, 0.1671], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0246, 0.0250, 0.0227, 0.0343, 0.0336, 0.0236, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.4987e-04, 9.1741e-05, 1.0767e-04, 9.8817e-05, 1.4543e-04, 1.3294e-04, 9.4603e-05, 1.4706e-04], device='cuda:3') 2023-03-08 13:56:17,135 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3505, 4.8264, 4.6791, 4.7736, 4.8298, 4.5029, 3.3454, 4.7472], device='cuda:3'), covar=tensor([0.0121, 0.0114, 0.0132, 0.0089, 0.0116, 0.0118, 0.0708, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0080, 0.0099, 0.0062, 0.0067, 0.0079, 0.0097, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 13:56:35,307 INFO [train2.py:809] (3/4) Epoch 16, batch 2250, loss[ctc_loss=0.06722, att_loss=0.2226, loss=0.1915, over 15386.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.009234, over 35.00 utterances.], tot_loss[ctc_loss=0.0855, att_loss=0.2417, loss=0.2105, over 3267698.91 frames. utt_duration=1202 frames, utt_pad_proportion=0.06849, over 10885.32 utterances.], batch size: 35, lr: 6.76e-03, grad_scale: 8.0 2023-03-08 13:57:11,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.098e+02 2.625e+02 3.264e+02 7.065e+02, threshold=5.250e+02, percent-clipped=5.0 2023-03-08 13:57:54,973 INFO [train2.py:809] (3/4) Epoch 16, batch 2300, loss[ctc_loss=0.08918, att_loss=0.2377, loss=0.208, over 16274.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007117, over 43.00 utterances.], tot_loss[ctc_loss=0.08531, att_loss=0.2417, loss=0.2104, over 3270130.76 frames. utt_duration=1220 frames, utt_pad_proportion=0.06279, over 10736.03 utterances.], batch size: 43, lr: 6.75e-03, grad_scale: 8.0 2023-03-08 13:58:44,420 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:59:02,063 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:59:02,182 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2401, 2.8140, 3.2618, 4.3669, 3.8463, 3.8588, 2.8232, 2.1592], device='cuda:3'), covar=tensor([0.0719, 0.1972, 0.0980, 0.0618, 0.0753, 0.0477, 0.1479, 0.2334], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0214, 0.0191, 0.0203, 0.0215, 0.0170, 0.0198, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 13:59:13,559 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5327, 3.8038, 2.9368, 3.1845, 3.8335, 3.5892, 2.4390, 4.1553], device='cuda:3'), covar=tensor([0.1395, 0.0543, 0.1165, 0.0849, 0.0800, 0.0668, 0.1291, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0204, 0.0217, 0.0191, 0.0263, 0.0229, 0.0196, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 13:59:14,699 INFO [train2.py:809] (3/4) Epoch 16, batch 2350, loss[ctc_loss=0.08489, att_loss=0.2456, loss=0.2135, over 16971.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.006637, over 50.00 utterances.], tot_loss[ctc_loss=0.08555, att_loss=0.2423, loss=0.211, over 3269225.22 frames. utt_duration=1205 frames, utt_pad_proportion=0.06534, over 10864.90 utterances.], batch size: 50, lr: 6.75e-03, grad_scale: 8.0 2023-03-08 13:59:17,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2023-03-08 13:59:38,240 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 13:59:49,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.132e+02 2.401e+02 3.401e+02 9.519e+02, threshold=4.802e+02, percent-clipped=3.0 2023-03-08 14:00:11,783 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8812, 3.6060, 3.5516, 3.0711, 3.6477, 3.6802, 3.5758, 2.5281], device='cuda:3'), covar=tensor([0.1121, 0.1422, 0.3027, 0.5041, 0.1168, 0.2406, 0.1229, 0.6252], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0159, 0.0169, 0.0232, 0.0132, 0.0225, 0.0146, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:00:30,821 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6004, 4.7125, 4.5946, 4.5746, 5.3243, 4.4920, 4.4854, 2.5231], device='cuda:3'), covar=tensor([0.0214, 0.0280, 0.0318, 0.0331, 0.0899, 0.0222, 0.0362, 0.2042], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0154, 0.0158, 0.0169, 0.0350, 0.0135, 0.0146, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:00:34,131 INFO [train2.py:809] (3/4) Epoch 16, batch 2400, loss[ctc_loss=0.08079, att_loss=0.251, loss=0.217, over 17069.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008658, over 53.00 utterances.], tot_loss[ctc_loss=0.08459, att_loss=0.241, loss=0.2097, over 3262974.89 frames. utt_duration=1244 frames, utt_pad_proportion=0.05869, over 10506.86 utterances.], batch size: 53, lr: 6.75e-03, grad_scale: 16.0 2023-03-08 14:01:15,286 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:01:17,990 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8912, 6.0807, 5.5613, 5.8967, 5.7721, 5.3416, 5.6177, 5.3670], device='cuda:3'), covar=tensor([0.1117, 0.0959, 0.0860, 0.0791, 0.0796, 0.1389, 0.2061, 0.2441], device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0559, 0.0431, 0.0435, 0.0409, 0.0446, 0.0586, 0.0509], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 14:01:32,343 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1817, 5.4303, 5.4072, 5.3899, 5.4968, 5.4507, 5.1657, 4.9493], device='cuda:3'), covar=tensor([0.0894, 0.0423, 0.0231, 0.0413, 0.0239, 0.0238, 0.0275, 0.0303], device='cuda:3'), in_proj_covar=tensor([0.0482, 0.0323, 0.0295, 0.0317, 0.0377, 0.0393, 0.0320, 0.0359], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:01:53,485 INFO [train2.py:809] (3/4) Epoch 16, batch 2450, loss[ctc_loss=0.069, att_loss=0.2419, loss=0.2074, over 16470.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.007226, over 46.00 utterances.], tot_loss[ctc_loss=0.0851, att_loss=0.2413, loss=0.21, over 3267741.09 frames. utt_duration=1260 frames, utt_pad_proportion=0.05416, over 10389.97 utterances.], batch size: 46, lr: 6.75e-03, grad_scale: 16.0 2023-03-08 14:02:20,424 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-08 14:02:29,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 2.234e+02 2.691e+02 3.084e+02 9.437e+02, threshold=5.381e+02, percent-clipped=5.0 2023-03-08 14:03:00,415 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5648, 4.6124, 4.6748, 4.5551, 5.2397, 4.6172, 4.5438, 2.4443], device='cuda:3'), covar=tensor([0.0214, 0.0272, 0.0242, 0.0295, 0.0644, 0.0198, 0.0288, 0.1904], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0155, 0.0157, 0.0169, 0.0349, 0.0135, 0.0146, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:03:13,139 INFO [train2.py:809] (3/4) Epoch 16, batch 2500, loss[ctc_loss=0.0781, att_loss=0.2504, loss=0.216, over 16742.00 frames. utt_duration=1396 frames, utt_pad_proportion=0.006757, over 48.00 utterances.], tot_loss[ctc_loss=0.08508, att_loss=0.2416, loss=0.2103, over 3273321.17 frames. utt_duration=1257 frames, utt_pad_proportion=0.05436, over 10431.66 utterances.], batch size: 48, lr: 6.74e-03, grad_scale: 16.0 2023-03-08 14:03:33,439 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:03:38,167 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:04:32,516 INFO [train2.py:809] (3/4) Epoch 16, batch 2550, loss[ctc_loss=0.08909, att_loss=0.2394, loss=0.2094, over 16522.00 frames. utt_duration=1470 frames, utt_pad_proportion=0.0058, over 45.00 utterances.], tot_loss[ctc_loss=0.08454, att_loss=0.2407, loss=0.2095, over 3271284.03 frames. utt_duration=1257 frames, utt_pad_proportion=0.05283, over 10421.79 utterances.], batch size: 45, lr: 6.74e-03, grad_scale: 8.0 2023-03-08 14:05:09,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 2.191e+02 2.595e+02 3.145e+02 7.901e+02, threshold=5.189e+02, percent-clipped=4.0 2023-03-08 14:05:13,621 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7624, 3.6570, 3.0434, 3.3073, 3.9021, 3.5515, 2.9697, 4.1276], device='cuda:3'), covar=tensor([0.1086, 0.0486, 0.1105, 0.0742, 0.0620, 0.0709, 0.0875, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0205, 0.0218, 0.0191, 0.0262, 0.0229, 0.0194, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:05:15,223 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7197, 2.2619, 2.5326, 2.4595, 3.1031, 2.7050, 2.2224, 2.4914], device='cuda:3'), covar=tensor([0.2274, 0.3798, 0.2930, 0.2114, 0.1576, 0.1604, 0.3781, 0.1403], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0098, 0.0101, 0.0090, 0.0093, 0.0085, 0.0105, 0.0074], device='cuda:3'), out_proj_covar=tensor([6.7786e-05, 7.4082e-05, 7.7109e-05, 6.7389e-05, 6.7974e-05, 6.6474e-05, 7.6658e-05, 5.8971e-05], device='cuda:3') 2023-03-08 14:05:15,229 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:05:52,043 INFO [train2.py:809] (3/4) Epoch 16, batch 2600, loss[ctc_loss=0.07383, att_loss=0.2221, loss=0.1924, over 15893.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008464, over 39.00 utterances.], tot_loss[ctc_loss=0.08385, att_loss=0.2405, loss=0.2092, over 3272578.20 frames. utt_duration=1257 frames, utt_pad_proportion=0.05295, over 10427.84 utterances.], batch size: 39, lr: 6.74e-03, grad_scale: 8.0 2023-03-08 14:06:41,234 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:06:45,809 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:06:58,811 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:07:11,495 INFO [train2.py:809] (3/4) Epoch 16, batch 2650, loss[ctc_loss=0.07973, att_loss=0.2092, loss=0.1833, over 15499.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008897, over 36.00 utterances.], tot_loss[ctc_loss=0.08399, att_loss=0.2402, loss=0.2089, over 3269061.13 frames. utt_duration=1264 frames, utt_pad_proportion=0.05295, over 10359.93 utterances.], batch size: 36, lr: 6.74e-03, grad_scale: 8.0 2023-03-08 14:07:49,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.130e+02 2.553e+02 3.253e+02 7.281e+02, threshold=5.105e+02, percent-clipped=6.0 2023-03-08 14:07:57,925 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:08:15,339 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:08:24,079 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:08:31,464 INFO [train2.py:809] (3/4) Epoch 16, batch 2700, loss[ctc_loss=0.08659, att_loss=0.2455, loss=0.2137, over 17141.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01301, over 56.00 utterances.], tot_loss[ctc_loss=0.08432, att_loss=0.2405, loss=0.2093, over 3273467.84 frames. utt_duration=1252 frames, utt_pad_proportion=0.05353, over 10472.34 utterances.], batch size: 56, lr: 6.73e-03, grad_scale: 8.0 2023-03-08 14:09:05,230 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:09:51,787 INFO [train2.py:809] (3/4) Epoch 16, batch 2750, loss[ctc_loss=0.07025, att_loss=0.2276, loss=0.1961, over 15942.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007695, over 41.00 utterances.], tot_loss[ctc_loss=0.08488, att_loss=0.2409, loss=0.2097, over 3267542.33 frames. utt_duration=1246 frames, utt_pad_proportion=0.05716, over 10503.56 utterances.], batch size: 41, lr: 6.73e-03, grad_scale: 8.0 2023-03-08 14:10:29,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.080e+02 2.636e+02 3.250e+02 8.951e+02, threshold=5.273e+02, percent-clipped=4.0 2023-03-08 14:10:29,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5261, 2.2175, 2.0553, 2.3801, 2.6698, 2.4117, 2.2366, 2.5969], device='cuda:3'), covar=tensor([0.1657, 0.3330, 0.2846, 0.1887, 0.1981, 0.1553, 0.2925, 0.1096], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0097, 0.0100, 0.0090, 0.0094, 0.0084, 0.0103, 0.0073], device='cuda:3'), out_proj_covar=tensor([6.7229e-05, 7.3482e-05, 7.6404e-05, 6.7179e-05, 6.8331e-05, 6.5921e-05, 7.5308e-05, 5.8523e-05], device='cuda:3') 2023-03-08 14:10:52,936 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:11:10,991 INFO [train2.py:809] (3/4) Epoch 16, batch 2800, loss[ctc_loss=0.06087, att_loss=0.2228, loss=0.1904, over 16004.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007652, over 40.00 utterances.], tot_loss[ctc_loss=0.08524, att_loss=0.2411, loss=0.2099, over 3268377.26 frames. utt_duration=1228 frames, utt_pad_proportion=0.06143, over 10660.55 utterances.], batch size: 40, lr: 6.73e-03, grad_scale: 8.0 2023-03-08 14:11:31,257 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:11:52,728 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9838, 5.0137, 4.8721, 2.3005, 1.9624, 2.7213, 2.9570, 3.7872], device='cuda:3'), covar=tensor([0.0792, 0.0245, 0.0267, 0.3995, 0.5721, 0.2758, 0.2517, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0247, 0.0249, 0.0230, 0.0343, 0.0336, 0.0237, 0.0360], device='cuda:3'), out_proj_covar=tensor([1.4994e-04, 9.2593e-05, 1.0763e-04, 1.0035e-04, 1.4570e-04, 1.3289e-04, 9.5117e-05, 1.4830e-04], device='cuda:3') 2023-03-08 14:11:55,450 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:12:29,672 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:12:30,822 INFO [train2.py:809] (3/4) Epoch 16, batch 2850, loss[ctc_loss=0.09021, att_loss=0.2526, loss=0.2201, over 17090.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01662, over 56.00 utterances.], tot_loss[ctc_loss=0.08513, att_loss=0.241, loss=0.2098, over 3272834.85 frames. utt_duration=1248 frames, utt_pad_proportion=0.05432, over 10501.64 utterances.], batch size: 56, lr: 6.72e-03, grad_scale: 8.0 2023-03-08 14:12:48,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:13:06,017 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:13:09,362 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 1.902e+02 2.514e+02 2.955e+02 4.757e+02, threshold=5.028e+02, percent-clipped=0.0 2023-03-08 14:13:34,645 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:13:51,158 INFO [train2.py:809] (3/4) Epoch 16, batch 2900, loss[ctc_loss=0.1101, att_loss=0.2545, loss=0.2257, over 16875.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007884, over 49.00 utterances.], tot_loss[ctc_loss=0.08515, att_loss=0.2411, loss=0.2099, over 3266329.21 frames. utt_duration=1215 frames, utt_pad_proportion=0.06473, over 10765.22 utterances.], batch size: 49, lr: 6.72e-03, grad_scale: 8.0 2023-03-08 14:14:06,746 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:15:00,093 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6399, 3.1031, 3.7824, 3.0666, 3.6700, 4.6586, 4.5345, 3.3899], device='cuda:3'), covar=tensor([0.0344, 0.1651, 0.1065, 0.1359, 0.0987, 0.0881, 0.0534, 0.1202], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0232, 0.0261, 0.0205, 0.0245, 0.0333, 0.0236, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:15:12,440 INFO [train2.py:809] (3/4) Epoch 16, batch 2950, loss[ctc_loss=0.1193, att_loss=0.2645, loss=0.2355, over 17322.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.03648, over 63.00 utterances.], tot_loss[ctc_loss=0.08516, att_loss=0.2412, loss=0.21, over 3267114.54 frames. utt_duration=1209 frames, utt_pad_proportion=0.06527, over 10819.48 utterances.], batch size: 63, lr: 6.72e-03, grad_scale: 8.0 2023-03-08 14:15:44,561 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:15:50,983 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 2.017e+02 2.387e+02 2.901e+02 6.735e+02, threshold=4.774e+02, percent-clipped=4.0 2023-03-08 14:16:17,627 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:16:32,914 INFO [train2.py:809] (3/4) Epoch 16, batch 3000, loss[ctc_loss=0.05789, att_loss=0.214, loss=0.1828, over 14471.00 frames. utt_duration=1811 frames, utt_pad_proportion=0.03849, over 32.00 utterances.], tot_loss[ctc_loss=0.08437, att_loss=0.2404, loss=0.2092, over 3251681.32 frames. utt_duration=1219 frames, utt_pad_proportion=0.06534, over 10681.77 utterances.], batch size: 32, lr: 6.72e-03, grad_scale: 8.0 2023-03-08 14:16:32,914 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 14:16:46,718 INFO [train2.py:843] (3/4) Epoch 16, validation: ctc_loss=0.0433, att_loss=0.235, loss=0.1967, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 14:16:46,719 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 14:16:47,059 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8713, 3.7118, 3.1078, 3.2871, 3.9489, 3.5540, 2.8307, 4.2559], device='cuda:3'), covar=tensor([0.1053, 0.0474, 0.1098, 0.0736, 0.0666, 0.0682, 0.0909, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0205, 0.0217, 0.0189, 0.0263, 0.0229, 0.0193, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:17:11,775 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:17:19,154 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:18:03,761 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-08 14:18:06,028 INFO [train2.py:809] (3/4) Epoch 16, batch 3050, loss[ctc_loss=0.07601, att_loss=0.2466, loss=0.2125, over 17412.00 frames. utt_duration=883.2 frames, utt_pad_proportion=0.07517, over 79.00 utterances.], tot_loss[ctc_loss=0.08536, att_loss=0.2416, loss=0.2103, over 3255177.41 frames. utt_duration=1204 frames, utt_pad_proportion=0.06803, over 10824.44 utterances.], batch size: 79, lr: 6.71e-03, grad_scale: 8.0 2023-03-08 14:18:36,373 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:18:43,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 2.145e+02 2.502e+02 3.037e+02 6.032e+02, threshold=5.003e+02, percent-clipped=3.0 2023-03-08 14:18:48,950 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:19:20,448 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1546, 5.3303, 5.7019, 5.5680, 5.6163, 6.1655, 5.2338, 6.2414], device='cuda:3'), covar=tensor([0.0753, 0.0721, 0.0738, 0.1145, 0.1729, 0.0770, 0.0597, 0.0544], device='cuda:3'), in_proj_covar=tensor([0.0807, 0.0474, 0.0558, 0.0618, 0.0815, 0.0567, 0.0449, 0.0547], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:19:28,202 INFO [train2.py:809] (3/4) Epoch 16, batch 3100, loss[ctc_loss=0.0757, att_loss=0.2446, loss=0.2108, over 17309.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01157, over 55.00 utterances.], tot_loss[ctc_loss=0.08617, att_loss=0.2425, loss=0.2112, over 3257655.36 frames. utt_duration=1188 frames, utt_pad_proportion=0.07245, over 10978.11 utterances.], batch size: 55, lr: 6.71e-03, grad_scale: 8.0 2023-03-08 14:20:42,405 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:20:51,888 INFO [train2.py:809] (3/4) Epoch 16, batch 3150, loss[ctc_loss=0.08497, att_loss=0.2381, loss=0.2075, over 16541.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006137, over 45.00 utterances.], tot_loss[ctc_loss=0.08606, att_loss=0.2424, loss=0.2111, over 3258761.90 frames. utt_duration=1187 frames, utt_pad_proportion=0.07249, over 10994.84 utterances.], batch size: 45, lr: 6.71e-03, grad_scale: 8.0 2023-03-08 14:21:28,213 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:21:31,247 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.078e+02 2.532e+02 3.325e+02 7.691e+02, threshold=5.063e+02, percent-clipped=6.0 2023-03-08 14:21:34,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-08 14:21:49,293 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:22:05,883 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5966, 2.4712, 3.6975, 2.9141, 3.4506, 4.7145, 4.6215, 2.9039], device='cuda:3'), covar=tensor([0.0495, 0.2562, 0.1132, 0.1711, 0.1182, 0.0862, 0.0526, 0.1989], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0238, 0.0268, 0.0211, 0.0252, 0.0341, 0.0241, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:22:15,346 INFO [train2.py:809] (3/4) Epoch 16, batch 3200, loss[ctc_loss=0.06814, att_loss=0.2215, loss=0.1908, over 16171.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006765, over 41.00 utterances.], tot_loss[ctc_loss=0.08577, att_loss=0.2423, loss=0.211, over 3264247.67 frames. utt_duration=1197 frames, utt_pad_proportion=0.06809, over 10926.31 utterances.], batch size: 41, lr: 6.71e-03, grad_scale: 8.0 2023-03-08 14:22:47,679 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:23:37,497 INFO [train2.py:809] (3/4) Epoch 16, batch 3250, loss[ctc_loss=0.1016, att_loss=0.2513, loss=0.2214, over 16877.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007782, over 49.00 utterances.], tot_loss[ctc_loss=0.08558, att_loss=0.2423, loss=0.2109, over 3272355.91 frames. utt_duration=1225 frames, utt_pad_proportion=0.06008, over 10695.86 utterances.], batch size: 49, lr: 6.70e-03, grad_scale: 8.0 2023-03-08 14:23:58,628 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1395, 5.1577, 4.9908, 2.3777, 2.0462, 2.8839, 2.6190, 3.8933], device='cuda:3'), covar=tensor([0.0673, 0.0263, 0.0246, 0.4718, 0.5429, 0.2527, 0.3068, 0.1619], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0246, 0.0249, 0.0227, 0.0340, 0.0331, 0.0239, 0.0355], device='cuda:3'), out_proj_covar=tensor([1.4976e-04, 9.2514e-05, 1.0775e-04, 9.8780e-05, 1.4468e-04, 1.3122e-04, 9.5634e-05, 1.4667e-04], device='cuda:3') 2023-03-08 14:24:03,052 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:24:17,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.062e+02 2.466e+02 2.972e+02 8.137e+02, threshold=4.932e+02, percent-clipped=1.0 2023-03-08 14:24:45,832 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:25:01,610 INFO [train2.py:809] (3/4) Epoch 16, batch 3300, loss[ctc_loss=0.09216, att_loss=0.2567, loss=0.2238, over 17062.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008447, over 52.00 utterances.], tot_loss[ctc_loss=0.0849, att_loss=0.2421, loss=0.2106, over 3276371.05 frames. utt_duration=1252 frames, utt_pad_proportion=0.05258, over 10484.45 utterances.], batch size: 52, lr: 6.70e-03, grad_scale: 8.0 2023-03-08 14:26:04,531 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:26:24,103 INFO [train2.py:809] (3/4) Epoch 16, batch 3350, loss[ctc_loss=0.1157, att_loss=0.2503, loss=0.2234, over 16458.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007678, over 46.00 utterances.], tot_loss[ctc_loss=0.08528, att_loss=0.2416, loss=0.2104, over 3273944.63 frames. utt_duration=1238 frames, utt_pad_proportion=0.05707, over 10592.48 utterances.], batch size: 46, lr: 6.70e-03, grad_scale: 8.0 2023-03-08 14:26:33,006 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8671, 5.1590, 5.4360, 5.2945, 5.3359, 5.8238, 5.0898, 5.9080], device='cuda:3'), covar=tensor([0.0752, 0.0810, 0.0787, 0.1270, 0.1920, 0.0896, 0.0665, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0808, 0.0473, 0.0557, 0.0617, 0.0816, 0.0569, 0.0451, 0.0548], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:27:00,483 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:27:03,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 2.090e+02 2.534e+02 3.183e+02 5.519e+02, threshold=5.068e+02, percent-clipped=6.0 2023-03-08 14:27:47,073 INFO [train2.py:809] (3/4) Epoch 16, batch 3400, loss[ctc_loss=0.08573, att_loss=0.236, loss=0.206, over 15884.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009253, over 39.00 utterances.], tot_loss[ctc_loss=0.08379, att_loss=0.2403, loss=0.209, over 3265089.81 frames. utt_duration=1260 frames, utt_pad_proportion=0.05228, over 10379.54 utterances.], batch size: 39, lr: 6.70e-03, grad_scale: 8.0 2023-03-08 14:27:47,315 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9212, 5.1559, 5.0747, 5.0395, 5.1975, 5.1470, 4.8966, 4.5761], device='cuda:3'), covar=tensor([0.1034, 0.0531, 0.0327, 0.0551, 0.0283, 0.0371, 0.0348, 0.0388], device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0329, 0.0304, 0.0322, 0.0379, 0.0403, 0.0326, 0.0362], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:29:01,699 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:29:11,061 INFO [train2.py:809] (3/4) Epoch 16, batch 3450, loss[ctc_loss=0.08933, att_loss=0.2369, loss=0.2074, over 15936.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.008177, over 41.00 utterances.], tot_loss[ctc_loss=0.08362, att_loss=0.2402, loss=0.2089, over 3262604.41 frames. utt_duration=1273 frames, utt_pad_proportion=0.04908, over 10259.92 utterances.], batch size: 41, lr: 6.69e-03, grad_scale: 8.0 2023-03-08 14:29:34,424 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:29:49,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 2.022e+02 2.458e+02 2.979e+02 7.741e+02, threshold=4.916e+02, percent-clipped=5.0 2023-03-08 14:30:06,827 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:30:20,199 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:30:29,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 14:30:32,953 INFO [train2.py:809] (3/4) Epoch 16, batch 3500, loss[ctc_loss=0.1378, att_loss=0.2681, loss=0.242, over 14446.00 frames. utt_duration=397.4 frames, utt_pad_proportion=0.3053, over 146.00 utterances.], tot_loss[ctc_loss=0.0854, att_loss=0.2416, loss=0.2104, over 3267191.18 frames. utt_duration=1226 frames, utt_pad_proportion=0.06081, over 10668.83 utterances.], batch size: 146, lr: 6.69e-03, grad_scale: 8.0 2023-03-08 14:31:14,077 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:31:25,096 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:31:51,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 14:31:52,842 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8136, 5.0856, 4.6690, 5.1701, 4.5550, 4.8173, 5.2541, 5.0165], device='cuda:3'), covar=tensor([0.0554, 0.0292, 0.0778, 0.0306, 0.0446, 0.0250, 0.0239, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0297, 0.0347, 0.0314, 0.0303, 0.0225, 0.0284, 0.0262], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 14:31:55,972 INFO [train2.py:809] (3/4) Epoch 16, batch 3550, loss[ctc_loss=0.1362, att_loss=0.2603, loss=0.2355, over 13605.00 frames. utt_duration=374.3 frames, utt_pad_proportion=0.3467, over 146.00 utterances.], tot_loss[ctc_loss=0.08546, att_loss=0.2419, loss=0.2106, over 3267552.57 frames. utt_duration=1211 frames, utt_pad_proportion=0.06609, over 10805.42 utterances.], batch size: 146, lr: 6.69e-03, grad_scale: 8.0 2023-03-08 14:32:20,753 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:32:36,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 2.051e+02 2.467e+02 3.018e+02 7.226e+02, threshold=4.933e+02, percent-clipped=6.0 2023-03-08 14:33:06,286 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9186, 3.7422, 3.1230, 3.3668, 3.8736, 3.5852, 2.7558, 4.1964], device='cuda:3'), covar=tensor([0.1032, 0.0508, 0.1010, 0.0678, 0.0697, 0.0706, 0.0952, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0206, 0.0219, 0.0190, 0.0264, 0.0231, 0.0195, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:33:19,239 INFO [train2.py:809] (3/4) Epoch 16, batch 3600, loss[ctc_loss=0.08538, att_loss=0.2383, loss=0.2077, over 16119.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006757, over 42.00 utterances.], tot_loss[ctc_loss=0.08534, att_loss=0.2419, loss=0.2106, over 3262508.64 frames. utt_duration=1183 frames, utt_pad_proportion=0.07329, over 11043.69 utterances.], batch size: 42, lr: 6.69e-03, grad_scale: 8.0 2023-03-08 14:33:39,695 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:34:42,618 INFO [train2.py:809] (3/4) Epoch 16, batch 3650, loss[ctc_loss=0.05796, att_loss=0.2349, loss=0.1995, over 16978.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.00595, over 50.00 utterances.], tot_loss[ctc_loss=0.08448, att_loss=0.2414, loss=0.21, over 3266958.33 frames. utt_duration=1215 frames, utt_pad_proportion=0.06491, over 10772.42 utterances.], batch size: 50, lr: 6.68e-03, grad_scale: 8.0 2023-03-08 14:34:51,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.06 vs. limit=5.0 2023-03-08 14:34:52,491 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6088, 3.7381, 3.6082, 3.6002, 3.9192, 3.6783, 3.4938, 2.4472], device='cuda:3'), covar=tensor([0.0335, 0.0443, 0.0450, 0.0440, 0.0906, 0.0275, 0.0445, 0.1716], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0155, 0.0162, 0.0173, 0.0349, 0.0137, 0.0146, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:34:53,846 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9135, 4.0024, 4.0206, 4.0594, 4.1273, 4.1308, 3.9493, 3.8265], device='cuda:3'), covar=tensor([0.0938, 0.0676, 0.0546, 0.0499, 0.0333, 0.0369, 0.0385, 0.0354], device='cuda:3'), in_proj_covar=tensor([0.0489, 0.0329, 0.0304, 0.0322, 0.0381, 0.0406, 0.0328, 0.0362], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:35:03,590 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:35:07,472 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:35:19,569 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:35:22,397 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.142e+02 2.479e+02 3.214e+02 4.969e+02, threshold=4.958e+02, percent-clipped=1.0 2023-03-08 14:35:57,355 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 14:36:06,126 INFO [train2.py:809] (3/4) Epoch 16, batch 3700, loss[ctc_loss=0.07443, att_loss=0.2269, loss=0.1964, over 15888.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008495, over 39.00 utterances.], tot_loss[ctc_loss=0.08456, att_loss=0.241, loss=0.2098, over 3264121.40 frames. utt_duration=1209 frames, utt_pad_proportion=0.06844, over 10815.52 utterances.], batch size: 39, lr: 6.68e-03, grad_scale: 8.0 2023-03-08 14:36:33,844 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8782, 4.7206, 4.7431, 2.2995, 2.0563, 2.8108, 2.2048, 3.6089], device='cuda:3'), covar=tensor([0.0780, 0.0234, 0.0235, 0.4382, 0.5809, 0.2494, 0.3336, 0.1683], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0244, 0.0247, 0.0223, 0.0335, 0.0325, 0.0237, 0.0350], device='cuda:3'), out_proj_covar=tensor([1.4749e-04, 9.1149e-05, 1.0684e-04, 9.6582e-05, 1.4233e-04, 1.2873e-04, 9.4681e-05, 1.4447e-04], device='cuda:3') 2023-03-08 14:36:40,562 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:36:47,218 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:36:50,265 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:37:27,472 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2014, 5.1074, 4.8698, 3.2886, 4.8924, 4.6339, 4.4332, 2.6448], device='cuda:3'), covar=tensor([0.0111, 0.0102, 0.0291, 0.0886, 0.0092, 0.0205, 0.0302, 0.1525], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0097, 0.0096, 0.0110, 0.0080, 0.0107, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 14:37:30,410 INFO [train2.py:809] (3/4) Epoch 16, batch 3750, loss[ctc_loss=0.08808, att_loss=0.2584, loss=0.2244, over 17320.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.03833, over 63.00 utterances.], tot_loss[ctc_loss=0.08429, att_loss=0.2409, loss=0.2096, over 3264285.25 frames. utt_duration=1216 frames, utt_pad_proportion=0.06483, over 10753.69 utterances.], batch size: 63, lr: 6.68e-03, grad_scale: 8.0 2023-03-08 14:38:09,923 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.179e+02 2.509e+02 3.051e+02 5.574e+02, threshold=5.018e+02, percent-clipped=2.0 2023-03-08 14:38:52,444 INFO [train2.py:809] (3/4) Epoch 16, batch 3800, loss[ctc_loss=0.07498, att_loss=0.2275, loss=0.197, over 15504.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008498, over 36.00 utterances.], tot_loss[ctc_loss=0.08431, att_loss=0.2403, loss=0.2091, over 3260530.85 frames. utt_duration=1220 frames, utt_pad_proportion=0.06413, over 10702.08 utterances.], batch size: 36, lr: 6.67e-03, grad_scale: 8.0 2023-03-08 14:39:11,907 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:39:25,936 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:39:41,239 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:40:15,130 INFO [train2.py:809] (3/4) Epoch 16, batch 3850, loss[ctc_loss=0.08707, att_loss=0.2296, loss=0.2011, over 15771.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.00873, over 38.00 utterances.], tot_loss[ctc_loss=0.0844, att_loss=0.2405, loss=0.2093, over 3262716.37 frames. utt_duration=1214 frames, utt_pad_proportion=0.065, over 10766.01 utterances.], batch size: 38, lr: 6.67e-03, grad_scale: 8.0 2023-03-08 14:40:17,008 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0155, 5.2888, 5.1743, 5.2003, 5.3041, 5.2637, 4.9965, 4.6645], device='cuda:3'), covar=tensor([0.0971, 0.0501, 0.0279, 0.0530, 0.0285, 0.0292, 0.0343, 0.0366], device='cuda:3'), in_proj_covar=tensor([0.0490, 0.0329, 0.0306, 0.0324, 0.0384, 0.0407, 0.0329, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 14:40:34,497 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8430, 6.1251, 5.5860, 5.9234, 5.8432, 5.3729, 5.5401, 5.4152], device='cuda:3'), covar=tensor([0.1355, 0.0833, 0.0856, 0.0794, 0.0736, 0.1332, 0.2147, 0.2079], device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0566, 0.0431, 0.0434, 0.0407, 0.0452, 0.0582, 0.0508], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 14:40:39,618 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1257, 5.0382, 4.8636, 3.2134, 4.8791, 4.7054, 4.2721, 2.7267], device='cuda:3'), covar=tensor([0.0093, 0.0097, 0.0236, 0.0885, 0.0085, 0.0190, 0.0314, 0.1384], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0097, 0.0095, 0.0111, 0.0080, 0.0108, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 14:40:52,470 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:40:53,568 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 2.062e+02 2.555e+02 3.342e+02 5.623e+02, threshold=5.109e+02, percent-clipped=2.0 2023-03-08 14:41:15,027 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7232, 5.2082, 5.0264, 5.0028, 5.2360, 4.8027, 3.4752, 5.1355], device='cuda:3'), covar=tensor([0.0100, 0.0085, 0.0112, 0.0085, 0.0098, 0.0116, 0.0718, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0080, 0.0100, 0.0063, 0.0068, 0.0079, 0.0098, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:41:19,857 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:41:35,252 INFO [train2.py:809] (3/4) Epoch 16, batch 3900, loss[ctc_loss=0.0915, att_loss=0.25, loss=0.2183, over 16901.00 frames. utt_duration=1381 frames, utt_pad_proportion=0.005583, over 49.00 utterances.], tot_loss[ctc_loss=0.08388, att_loss=0.2404, loss=0.2091, over 3261758.50 frames. utt_duration=1212 frames, utt_pad_proportion=0.06638, over 10779.81 utterances.], batch size: 49, lr: 6.67e-03, grad_scale: 8.0 2023-03-08 14:42:07,208 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 14:42:54,344 INFO [train2.py:809] (3/4) Epoch 16, batch 3950, loss[ctc_loss=0.09384, att_loss=0.261, loss=0.2276, over 17117.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01452, over 56.00 utterances.], tot_loss[ctc_loss=0.08464, att_loss=0.2414, loss=0.21, over 3263836.20 frames. utt_duration=1210 frames, utt_pad_proportion=0.06521, over 10800.79 utterances.], batch size: 56, lr: 6.67e-03, grad_scale: 8.0 2023-03-08 14:43:31,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.047e+02 2.429e+02 3.327e+02 6.424e+02, threshold=4.857e+02, percent-clipped=3.0 2023-03-08 14:44:08,715 INFO [train2.py:809] (3/4) Epoch 17, batch 0, loss[ctc_loss=0.09258, att_loss=0.251, loss=0.2193, over 17063.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.007422, over 52.00 utterances.], tot_loss[ctc_loss=0.09258, att_loss=0.251, loss=0.2193, over 17063.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.007422, over 52.00 utterances.], batch size: 52, lr: 6.46e-03, grad_scale: 8.0 2023-03-08 14:44:08,715 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 14:44:15,013 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6947, 4.1756, 4.1680, 2.2314, 1.8680, 2.6471, 1.8708, 3.4964], device='cuda:3'), covar=tensor([0.0718, 0.0342, 0.0349, 0.4864, 0.6124, 0.2690, 0.4127, 0.1443], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0244, 0.0249, 0.0225, 0.0338, 0.0329, 0.0239, 0.0352], device='cuda:3'), out_proj_covar=tensor([1.4724e-04, 9.0898e-05, 1.0734e-04, 9.7254e-05, 1.4338e-04, 1.2993e-04, 9.5658e-05, 1.4504e-04], device='cuda:3') 2023-03-08 14:44:15,462 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.5993, 6.2234, 6.2233, 6.0793, 6.1165, 6.5733, 5.6053, 6.6147], device='cuda:3'), covar=tensor([0.0446, 0.0522, 0.0540, 0.0779, 0.1418, 0.0601, 0.0354, 0.0382], device='cuda:3'), in_proj_covar=tensor([0.0799, 0.0469, 0.0555, 0.0615, 0.0814, 0.0568, 0.0449, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:44:19,595 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8347, 4.1889, 4.1830, 3.9350, 4.2777, 4.2341, 4.1726, 3.3876], device='cuda:3'), covar=tensor([0.0532, 0.1030, 0.1819, 0.1686, 0.1009, 0.2336, 0.0680, 0.3252], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0158, 0.0170, 0.0232, 0.0131, 0.0227, 0.0145, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:44:21,759 INFO [train2.py:843] (3/4) Epoch 17, validation: ctc_loss=0.04327, att_loss=0.2362, loss=0.1976, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 14:44:21,760 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 14:44:54,151 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7741, 2.7903, 5.1502, 4.0725, 3.2765, 4.4829, 5.1276, 4.8443], device='cuda:3'), covar=tensor([0.0215, 0.1395, 0.0187, 0.1019, 0.1654, 0.0197, 0.0089, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0240, 0.0163, 0.0308, 0.0263, 0.0196, 0.0143, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:45:20,140 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:45:23,459 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:45:44,563 INFO [train2.py:809] (3/4) Epoch 17, batch 50, loss[ctc_loss=0.07718, att_loss=0.244, loss=0.2106, over 17433.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.0305, over 63.00 utterances.], tot_loss[ctc_loss=0.08625, att_loss=0.2409, loss=0.2099, over 732122.28 frames. utt_duration=1150 frames, utt_pad_proportion=0.08669, over 2549.40 utterances.], batch size: 63, lr: 6.46e-03, grad_scale: 8.0 2023-03-08 14:46:08,966 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6228, 3.4618, 3.3642, 2.9645, 3.4244, 3.4251, 3.5523, 2.3460], device='cuda:3'), covar=tensor([0.1031, 0.1375, 0.2535, 0.4657, 0.2182, 0.3119, 0.0907, 0.5651], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0158, 0.0170, 0.0231, 0.0131, 0.0227, 0.0145, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:46:17,006 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:46:50,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 2.041e+02 2.465e+02 3.095e+02 5.834e+02, threshold=4.929e+02, percent-clipped=2.0 2023-03-08 14:46:52,914 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1516, 2.5349, 3.0267, 4.2024, 3.8418, 3.8256, 2.8084, 2.0589], device='cuda:3'), covar=tensor([0.0871, 0.2287, 0.1129, 0.0571, 0.0733, 0.0444, 0.1491, 0.2441], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0213, 0.0189, 0.0202, 0.0210, 0.0169, 0.0195, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 14:47:08,061 INFO [train2.py:809] (3/4) Epoch 17, batch 100, loss[ctc_loss=0.05927, att_loss=0.2198, loss=0.1877, over 15754.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009672, over 38.00 utterances.], tot_loss[ctc_loss=0.08368, att_loss=0.2403, loss=0.209, over 1304273.39 frames. utt_duration=1236 frames, utt_pad_proportion=0.05487, over 4227.43 utterances.], batch size: 38, lr: 6.46e-03, grad_scale: 8.0 2023-03-08 14:47:16,178 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:47:58,809 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:48:08,249 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:48:31,340 INFO [train2.py:809] (3/4) Epoch 17, batch 150, loss[ctc_loss=0.09115, att_loss=0.256, loss=0.2231, over 17377.00 frames. utt_duration=1221 frames, utt_pad_proportion=0.01545, over 57.00 utterances.], tot_loss[ctc_loss=0.0821, att_loss=0.2393, loss=0.2078, over 1729708.29 frames. utt_duration=1261 frames, utt_pad_proportion=0.05441, over 5495.08 utterances.], batch size: 57, lr: 6.46e-03, grad_scale: 4.0 2023-03-08 14:48:56,289 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:49:12,688 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:49:26,811 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:49:26,917 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:49:37,547 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 1.983e+02 2.552e+02 2.961e+02 5.419e+02, threshold=5.104e+02, percent-clipped=2.0 2023-03-08 14:49:52,620 INFO [train2.py:809] (3/4) Epoch 17, batch 200, loss[ctc_loss=0.06389, att_loss=0.2163, loss=0.1858, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008297, over 37.00 utterances.], tot_loss[ctc_loss=0.08139, att_loss=0.24, loss=0.2083, over 2075874.10 frames. utt_duration=1257 frames, utt_pad_proportion=0.05054, over 6611.19 utterances.], batch size: 37, lr: 6.45e-03, grad_scale: 4.0 2023-03-08 14:49:55,662 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:50:48,298 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-03-08 14:50:52,531 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:51:15,250 INFO [train2.py:809] (3/4) Epoch 17, batch 250, loss[ctc_loss=0.07223, att_loss=0.2178, loss=0.1887, over 15366.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01125, over 35.00 utterances.], tot_loss[ctc_loss=0.08197, att_loss=0.2404, loss=0.2087, over 2350303.98 frames. utt_duration=1269 frames, utt_pad_proportion=0.04382, over 7415.55 utterances.], batch size: 35, lr: 6.45e-03, grad_scale: 4.0 2023-03-08 14:51:39,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-08 14:51:57,430 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 14:52:14,293 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:52:28,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 2.064e+02 2.460e+02 3.052e+02 5.985e+02, threshold=4.919e+02, percent-clipped=1.0 2023-03-08 14:52:30,258 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1540, 4.1205, 4.1725, 4.0679, 4.7339, 4.2535, 4.1024, 2.4157], device='cuda:3'), covar=tensor([0.0282, 0.0433, 0.0386, 0.0288, 0.0693, 0.0245, 0.0334, 0.1911], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0157, 0.0164, 0.0175, 0.0353, 0.0137, 0.0147, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-08 14:52:43,729 INFO [train2.py:809] (3/4) Epoch 17, batch 300, loss[ctc_loss=0.06056, att_loss=0.2141, loss=0.1834, over 16112.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.007049, over 42.00 utterances.], tot_loss[ctc_loss=0.0823, att_loss=0.2405, loss=0.2089, over 2556337.90 frames. utt_duration=1252 frames, utt_pad_proportion=0.04926, over 8175.07 utterances.], batch size: 42, lr: 6.45e-03, grad_scale: 4.0 2023-03-08 14:53:25,543 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2967, 5.2492, 5.0034, 2.9138, 5.0466, 4.7295, 4.4867, 2.8964], device='cuda:3'), covar=tensor([0.0092, 0.0071, 0.0238, 0.1008, 0.0069, 0.0169, 0.0286, 0.1200], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0096, 0.0095, 0.0110, 0.0080, 0.0107, 0.0098, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 14:53:39,790 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:53:42,903 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:53:53,067 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:54:04,131 INFO [train2.py:809] (3/4) Epoch 17, batch 350, loss[ctc_loss=0.1009, att_loss=0.2371, loss=0.2098, over 15950.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.006782, over 41.00 utterances.], tot_loss[ctc_loss=0.08288, att_loss=0.2407, loss=0.2091, over 2712118.65 frames. utt_duration=1258 frames, utt_pad_proportion=0.05096, over 8631.44 utterances.], batch size: 41, lr: 6.45e-03, grad_scale: 4.0 2023-03-08 14:54:25,585 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-03-08 14:54:50,657 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9176, 6.1052, 5.5490, 5.8562, 5.7340, 5.2977, 5.5534, 5.3513], device='cuda:3'), covar=tensor([0.1183, 0.0838, 0.0971, 0.0777, 0.0985, 0.1610, 0.2211, 0.2078], device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0561, 0.0430, 0.0431, 0.0413, 0.0455, 0.0586, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 14:54:56,997 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:55:00,053 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:55:09,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.060e+02 2.368e+02 3.330e+02 9.339e+02, threshold=4.736e+02, percent-clipped=6.0 2023-03-08 14:55:18,159 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2019, 5.4248, 4.7992, 5.2345, 5.0239, 4.5772, 4.8522, 4.5518], device='cuda:3'), covar=tensor([0.1150, 0.0849, 0.1034, 0.0814, 0.1045, 0.1631, 0.2136, 0.2276], device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0560, 0.0430, 0.0430, 0.0412, 0.0455, 0.0585, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 14:55:21,613 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1872, 4.5979, 4.5315, 4.6278, 2.7894, 4.7547, 2.8290, 1.9614], device='cuda:3'), covar=tensor([0.0355, 0.0215, 0.0656, 0.0182, 0.1596, 0.0127, 0.1416, 0.1696], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0142, 0.0259, 0.0136, 0.0221, 0.0125, 0.0230, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 14:55:25,125 INFO [train2.py:809] (3/4) Epoch 17, batch 400, loss[ctc_loss=0.1077, att_loss=0.2641, loss=0.2328, over 17321.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02327, over 59.00 utterances.], tot_loss[ctc_loss=0.08325, att_loss=0.241, loss=0.2094, over 2840822.85 frames. utt_duration=1261 frames, utt_pad_proportion=0.04877, over 9018.68 utterances.], batch size: 59, lr: 6.44e-03, grad_scale: 8.0 2023-03-08 14:56:07,208 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:56:36,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2586, 4.6270, 4.6942, 4.6868, 2.9079, 4.7693, 3.0891, 1.9285], device='cuda:3'), covar=tensor([0.0385, 0.0219, 0.0607, 0.0374, 0.1531, 0.0139, 0.1258, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0141, 0.0256, 0.0135, 0.0218, 0.0123, 0.0228, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 14:56:47,843 INFO [train2.py:809] (3/4) Epoch 17, batch 450, loss[ctc_loss=0.06578, att_loss=0.2252, loss=0.1933, over 16008.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.00594, over 40.00 utterances.], tot_loss[ctc_loss=0.08357, att_loss=0.2409, loss=0.2095, over 2934454.34 frames. utt_duration=1242 frames, utt_pad_proportion=0.05532, over 9461.86 utterances.], batch size: 40, lr: 6.44e-03, grad_scale: 8.0 2023-03-08 14:57:02,675 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2958, 5.5160, 5.8078, 5.6219, 5.7550, 6.1116, 5.2486, 6.3091], device='cuda:3'), covar=tensor([0.0580, 0.0746, 0.0702, 0.1107, 0.1583, 0.0906, 0.0598, 0.0534], device='cuda:3'), in_proj_covar=tensor([0.0784, 0.0465, 0.0549, 0.0603, 0.0801, 0.0561, 0.0440, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 14:57:04,228 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:57:42,983 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:57:53,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.502e+02 2.140e+02 2.597e+02 3.193e+02 5.967e+02, threshold=5.195e+02, percent-clipped=4.0 2023-03-08 14:57:54,033 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:58:09,211 INFO [train2.py:809] (3/4) Epoch 17, batch 500, loss[ctc_loss=0.06207, att_loss=0.2255, loss=0.1928, over 16396.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.008115, over 44.00 utterances.], tot_loss[ctc_loss=0.08186, att_loss=0.2396, loss=0.2081, over 3006803.47 frames. utt_duration=1270 frames, utt_pad_proportion=0.04893, over 9482.72 utterances.], batch size: 44, lr: 6.44e-03, grad_scale: 8.0 2023-03-08 14:58:11,144 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:58:58,197 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:58:59,748 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:59:27,934 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:59:29,441 INFO [train2.py:809] (3/4) Epoch 17, batch 550, loss[ctc_loss=0.07484, att_loss=0.2212, loss=0.1919, over 16025.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006382, over 40.00 utterances.], tot_loss[ctc_loss=0.08156, att_loss=0.2387, loss=0.2073, over 3056889.55 frames. utt_duration=1276 frames, utt_pad_proportion=0.05191, over 9596.92 utterances.], batch size: 40, lr: 6.44e-03, grad_scale: 8.0 2023-03-08 14:59:33,716 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 14:59:53,209 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1934, 5.1815, 4.9482, 3.2318, 4.9799, 4.8011, 4.4764, 2.8909], device='cuda:3'), covar=tensor([0.0148, 0.0088, 0.0250, 0.0890, 0.0085, 0.0180, 0.0275, 0.1253], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0097, 0.0096, 0.0111, 0.0081, 0.0108, 0.0099, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:00:34,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 1.879e+02 2.342e+02 2.961e+02 5.931e+02, threshold=4.683e+02, percent-clipped=3.0 2023-03-08 15:00:50,376 INFO [train2.py:809] (3/4) Epoch 17, batch 600, loss[ctc_loss=0.07516, att_loss=0.246, loss=0.2118, over 16615.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005932, over 47.00 utterances.], tot_loss[ctc_loss=0.08186, att_loss=0.2389, loss=0.2075, over 3108027.09 frames. utt_duration=1278 frames, utt_pad_proportion=0.0478, over 9738.75 utterances.], batch size: 47, lr: 6.43e-03, grad_scale: 8.0 2023-03-08 15:00:57,638 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:01:42,128 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8859, 6.0548, 5.6098, 5.8233, 5.7274, 5.3237, 5.6221, 5.3662], device='cuda:3'), covar=tensor([0.1183, 0.0879, 0.0902, 0.0760, 0.0908, 0.1423, 0.2050, 0.2019], device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0568, 0.0432, 0.0431, 0.0412, 0.0456, 0.0586, 0.0508], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:01:51,636 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:02:12,275 INFO [train2.py:809] (3/4) Epoch 17, batch 650, loss[ctc_loss=0.1076, att_loss=0.2472, loss=0.2192, over 16286.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006808, over 43.00 utterances.], tot_loss[ctc_loss=0.08159, att_loss=0.2389, loss=0.2074, over 3149475.77 frames. utt_duration=1286 frames, utt_pad_proportion=0.04546, over 9806.75 utterances.], batch size: 43, lr: 6.43e-03, grad_scale: 8.0 2023-03-08 15:02:36,765 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:03:17,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.002e+02 2.471e+02 3.095e+02 7.929e+02, threshold=4.943e+02, percent-clipped=8.0 2023-03-08 15:03:32,635 INFO [train2.py:809] (3/4) Epoch 17, batch 700, loss[ctc_loss=0.07502, att_loss=0.2228, loss=0.1932, over 16022.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006428, over 40.00 utterances.], tot_loss[ctc_loss=0.08291, att_loss=0.2394, loss=0.2081, over 3178851.21 frames. utt_duration=1260 frames, utt_pad_proportion=0.05033, over 10105.96 utterances.], batch size: 40, lr: 6.43e-03, grad_scale: 8.0 2023-03-08 15:04:09,393 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2168, 2.7685, 3.0140, 4.2613, 3.7805, 3.9037, 2.7057, 1.9673], device='cuda:3'), covar=tensor([0.0724, 0.1931, 0.1125, 0.0422, 0.0821, 0.0394, 0.1565, 0.2411], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0218, 0.0190, 0.0206, 0.0215, 0.0172, 0.0200, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:04:12,685 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:04:14,131 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:04:52,370 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3818, 4.7258, 5.1228, 4.4735, 4.5733, 5.2505, 4.8280, 5.2885], device='cuda:3'), covar=tensor([0.1338, 0.1557, 0.1101, 0.2579, 0.3349, 0.1551, 0.1296, 0.1320], device='cuda:3'), in_proj_covar=tensor([0.0780, 0.0463, 0.0548, 0.0603, 0.0800, 0.0560, 0.0440, 0.0544], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:04:55,327 INFO [train2.py:809] (3/4) Epoch 17, batch 750, loss[ctc_loss=0.103, att_loss=0.2641, loss=0.2319, over 17289.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01255, over 55.00 utterances.], tot_loss[ctc_loss=0.08386, att_loss=0.2406, loss=0.2092, over 3205892.45 frames. utt_duration=1241 frames, utt_pad_proportion=0.05339, over 10342.61 utterances.], batch size: 55, lr: 6.43e-03, grad_scale: 8.0 2023-03-08 15:05:07,165 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.30 vs. limit=5.0 2023-03-08 15:05:11,193 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:05:18,266 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8683, 2.3246, 5.2279, 4.3708, 3.3768, 4.4919, 5.0574, 4.8796], device='cuda:3'), covar=tensor([0.0159, 0.1584, 0.0133, 0.0674, 0.1460, 0.0165, 0.0080, 0.0158], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0237, 0.0163, 0.0303, 0.0259, 0.0194, 0.0144, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 15:05:31,760 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:05:51,422 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:06:01,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.298e+02 2.805e+02 3.347e+02 6.423e+02, threshold=5.609e+02, percent-clipped=6.0 2023-03-08 15:06:16,290 INFO [train2.py:809] (3/4) Epoch 17, batch 800, loss[ctc_loss=0.08444, att_loss=0.222, loss=0.1945, over 15481.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009856, over 36.00 utterances.], tot_loss[ctc_loss=0.08337, att_loss=0.2405, loss=0.2091, over 3225399.45 frames. utt_duration=1246 frames, utt_pad_proportion=0.05174, over 10366.54 utterances.], batch size: 36, lr: 6.42e-03, grad_scale: 8.0 2023-03-08 15:06:28,798 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:07:05,772 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:07:32,189 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:07:33,841 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:07:36,532 INFO [train2.py:809] (3/4) Epoch 17, batch 850, loss[ctc_loss=0.08425, att_loss=0.2398, loss=0.2087, over 17056.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008621, over 52.00 utterances.], tot_loss[ctc_loss=0.0832, att_loss=0.2401, loss=0.2087, over 3240484.50 frames. utt_duration=1266 frames, utt_pad_proportion=0.04637, over 10248.39 utterances.], batch size: 52, lr: 6.42e-03, grad_scale: 8.0 2023-03-08 15:08:24,224 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:08:43,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.027e+02 2.349e+02 3.075e+02 7.936e+02, threshold=4.698e+02, percent-clipped=3.0 2023-03-08 15:08:58,453 INFO [train2.py:809] (3/4) Epoch 17, batch 900, loss[ctc_loss=0.09567, att_loss=0.2638, loss=0.2302, over 17054.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.008955, over 53.00 utterances.], tot_loss[ctc_loss=0.08352, att_loss=0.2406, loss=0.2092, over 3257567.21 frames. utt_duration=1250 frames, utt_pad_proportion=0.04771, over 10436.55 utterances.], batch size: 53, lr: 6.42e-03, grad_scale: 8.0 2023-03-08 15:09:01,958 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-08 15:09:10,780 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:09:14,136 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:09:27,553 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2200, 4.5165, 4.6604, 4.5899, 3.1358, 4.5970, 3.1637, 2.3611], device='cuda:3'), covar=tensor([0.0392, 0.0227, 0.0640, 0.0172, 0.1387, 0.0161, 0.1283, 0.1613], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0143, 0.0258, 0.0137, 0.0220, 0.0125, 0.0231, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:10:02,309 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:10:21,902 INFO [train2.py:809] (3/4) Epoch 17, batch 950, loss[ctc_loss=0.07056, att_loss=0.2404, loss=0.2065, over 16481.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005921, over 46.00 utterances.], tot_loss[ctc_loss=0.08315, att_loss=0.2403, loss=0.2088, over 3262215.83 frames. utt_duration=1262 frames, utt_pad_proportion=0.04545, over 10349.91 utterances.], batch size: 46, lr: 6.42e-03, grad_scale: 8.0 2023-03-08 15:10:38,883 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:10:42,354 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:10:51,764 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:11:19,764 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:11:28,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 2.001e+02 2.338e+02 2.858e+02 5.487e+02, threshold=4.676e+02, percent-clipped=1.0 2023-03-08 15:11:28,454 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2603, 2.0859, 2.6158, 2.6291, 2.6965, 2.6739, 2.3456, 3.2105], device='cuda:3'), covar=tensor([0.1413, 0.4531, 0.2887, 0.2415, 0.2358, 0.1924, 0.4302, 0.1197], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0105, 0.0109, 0.0095, 0.0101, 0.0089, 0.0110, 0.0079], device='cuda:3'), out_proj_covar=tensor([7.1851e-05, 7.9657e-05, 8.3065e-05, 7.1684e-05, 7.3731e-05, 7.0509e-05, 8.0949e-05, 6.3359e-05], device='cuda:3') 2023-03-08 15:11:42,851 INFO [train2.py:809] (3/4) Epoch 17, batch 1000, loss[ctc_loss=0.06685, att_loss=0.2228, loss=0.1916, over 15645.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008392, over 37.00 utterances.], tot_loss[ctc_loss=0.08355, att_loss=0.2407, loss=0.2093, over 3259486.82 frames. utt_duration=1253 frames, utt_pad_proportion=0.04959, over 10418.48 utterances.], batch size: 37, lr: 6.41e-03, grad_scale: 8.0 2023-03-08 15:11:49,908 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:12:01,525 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:12:20,117 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:13:02,950 INFO [train2.py:809] (3/4) Epoch 17, batch 1050, loss[ctc_loss=0.09164, att_loss=0.2505, loss=0.2187, over 17056.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009441, over 53.00 utterances.], tot_loss[ctc_loss=0.08385, att_loss=0.2409, loss=0.2095, over 3273620.16 frames. utt_duration=1263 frames, utt_pad_proportion=0.04449, over 10380.04 utterances.], batch size: 53, lr: 6.41e-03, grad_scale: 8.0 2023-03-08 15:13:27,826 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:13:38,378 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:13:50,513 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:14:08,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.188e+02 2.644e+02 3.168e+02 6.507e+02, threshold=5.288e+02, percent-clipped=8.0 2023-03-08 15:14:22,626 INFO [train2.py:809] (3/4) Epoch 17, batch 1100, loss[ctc_loss=0.05431, att_loss=0.2233, loss=0.1895, over 16466.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007362, over 46.00 utterances.], tot_loss[ctc_loss=0.08244, att_loss=0.24, loss=0.2085, over 3279742.17 frames. utt_duration=1282 frames, utt_pad_proportion=0.03956, over 10242.93 utterances.], batch size: 46, lr: 6.41e-03, grad_scale: 8.0 2023-03-08 15:14:44,781 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5349, 2.5323, 2.5404, 2.1628, 2.5113, 2.3234, 2.5342, 1.7930], device='cuda:3'), covar=tensor([0.1184, 0.2291, 0.2645, 0.6841, 0.1558, 0.2720, 0.1788, 0.8326], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0163, 0.0177, 0.0239, 0.0139, 0.0237, 0.0151, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 15:15:19,148 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-08 15:15:37,956 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:15:42,107 INFO [train2.py:809] (3/4) Epoch 17, batch 1150, loss[ctc_loss=0.06975, att_loss=0.2316, loss=0.1993, over 16249.00 frames. utt_duration=1513 frames, utt_pad_proportion=0.008489, over 43.00 utterances.], tot_loss[ctc_loss=0.08297, att_loss=0.2407, loss=0.2091, over 3276883.90 frames. utt_duration=1253 frames, utt_pad_proportion=0.04778, over 10471.61 utterances.], batch size: 43, lr: 6.41e-03, grad_scale: 8.0 2023-03-08 15:15:49,586 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9688, 4.3001, 4.1088, 4.5766, 2.6198, 4.5011, 2.6753, 2.0286], device='cuda:3'), covar=tensor([0.0447, 0.0219, 0.0833, 0.0169, 0.1771, 0.0162, 0.1532, 0.1794], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0142, 0.0257, 0.0135, 0.0217, 0.0124, 0.0230, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:16:24,117 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1347, 5.1077, 4.9219, 2.0548, 1.9381, 2.8353, 2.1929, 3.7754], device='cuda:3'), covar=tensor([0.0613, 0.0222, 0.0231, 0.5054, 0.5695, 0.2454, 0.3169, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0250, 0.0250, 0.0229, 0.0339, 0.0331, 0.0237, 0.0357], device='cuda:3'), out_proj_covar=tensor([1.4800e-04, 9.1936e-05, 1.0752e-04, 9.9374e-05, 1.4357e-04, 1.3076e-04, 9.5073e-05, 1.4650e-04], device='cuda:3') 2023-03-08 15:16:47,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 2.059e+02 2.511e+02 2.874e+02 5.628e+02, threshold=5.021e+02, percent-clipped=1.0 2023-03-08 15:16:54,658 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:17:02,368 INFO [train2.py:809] (3/4) Epoch 17, batch 1200, loss[ctc_loss=0.05766, att_loss=0.2046, loss=0.1752, over 15752.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009097, over 38.00 utterances.], tot_loss[ctc_loss=0.0833, att_loss=0.2408, loss=0.2093, over 3279877.59 frames. utt_duration=1232 frames, utt_pad_proportion=0.05358, over 10661.75 utterances.], batch size: 38, lr: 6.40e-03, grad_scale: 8.0 2023-03-08 15:17:09,401 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:17:10,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 15:17:58,163 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:18:22,338 INFO [train2.py:809] (3/4) Epoch 17, batch 1250, loss[ctc_loss=0.07331, att_loss=0.2237, loss=0.1936, over 16019.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006873, over 40.00 utterances.], tot_loss[ctc_loss=0.08353, att_loss=0.2409, loss=0.2095, over 3269673.36 frames. utt_duration=1220 frames, utt_pad_proportion=0.05899, over 10731.23 utterances.], batch size: 40, lr: 6.40e-03, grad_scale: 8.0 2023-03-08 15:18:31,751 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 15:18:34,265 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6899, 2.1214, 2.1455, 2.5073, 2.6312, 2.3634, 2.2537, 2.6764], device='cuda:3'), covar=tensor([0.1475, 0.3420, 0.2730, 0.1254, 0.1624, 0.1321, 0.2608, 0.1354], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0107, 0.0109, 0.0096, 0.0102, 0.0091, 0.0110, 0.0080], device='cuda:3'), out_proj_covar=tensor([7.2460e-05, 8.0748e-05, 8.3368e-05, 7.2258e-05, 7.4272e-05, 7.1288e-05, 8.1166e-05, 6.4050e-05], device='cuda:3') 2023-03-08 15:18:39,123 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:18:44,354 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:19:27,888 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.065e+02 2.436e+02 3.098e+02 5.553e+02, threshold=4.871e+02, percent-clipped=2.0 2023-03-08 15:19:36,267 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 15:19:41,908 INFO [train2.py:809] (3/4) Epoch 17, batch 1300, loss[ctc_loss=0.09587, att_loss=0.2516, loss=0.2205, over 16979.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.006301, over 50.00 utterances.], tot_loss[ctc_loss=0.08303, att_loss=0.2407, loss=0.2092, over 3280957.71 frames. utt_duration=1220 frames, utt_pad_proportion=0.05626, over 10768.87 utterances.], batch size: 50, lr: 6.40e-03, grad_scale: 8.0 2023-03-08 15:19:55,569 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:20:05,142 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:20:11,146 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:20:45,112 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-08 15:21:01,328 INFO [train2.py:809] (3/4) Epoch 17, batch 1350, loss[ctc_loss=0.06009, att_loss=0.2189, loss=0.1872, over 10166.00 frames. utt_duration=1850 frames, utt_pad_proportion=0.2529, over 22.00 utterances.], tot_loss[ctc_loss=0.0832, att_loss=0.2411, loss=0.2095, over 3275387.36 frames. utt_duration=1228 frames, utt_pad_proportion=0.05571, over 10683.47 utterances.], batch size: 22, lr: 6.40e-03, grad_scale: 8.0 2023-03-08 15:21:12,328 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0077, 5.2148, 5.2020, 5.1091, 5.2467, 5.2290, 4.9579, 4.6968], device='cuda:3'), covar=tensor([0.0865, 0.0478, 0.0237, 0.0523, 0.0280, 0.0297, 0.0319, 0.0354], device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0339, 0.0312, 0.0334, 0.0392, 0.0411, 0.0333, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 15:21:18,505 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:21:29,263 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:21:41,908 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:21:49,460 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:22:07,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.057e+02 2.538e+02 3.006e+02 4.389e+02, threshold=5.076e+02, percent-clipped=0.0 2023-03-08 15:22:22,269 INFO [train2.py:809] (3/4) Epoch 17, batch 1400, loss[ctc_loss=0.08253, att_loss=0.2542, loss=0.2199, over 17026.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.008029, over 51.00 utterances.], tot_loss[ctc_loss=0.08328, att_loss=0.2408, loss=0.2093, over 3271233.63 frames. utt_duration=1226 frames, utt_pad_proportion=0.05879, over 10683.52 utterances.], batch size: 51, lr: 6.39e-03, grad_scale: 8.0 2023-03-08 15:23:03,342 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2170, 5.1769, 4.9170, 3.0173, 4.9472, 4.7704, 4.1966, 2.8863], device='cuda:3'), covar=tensor([0.0115, 0.0090, 0.0279, 0.0934, 0.0086, 0.0179, 0.0342, 0.1270], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0096, 0.0095, 0.0109, 0.0080, 0.0107, 0.0098, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:23:06,280 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:23:35,254 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2345, 5.2007, 4.9719, 3.1079, 4.9338, 4.7535, 4.2437, 2.7386], device='cuda:3'), covar=tensor([0.0092, 0.0083, 0.0256, 0.0918, 0.0091, 0.0187, 0.0349, 0.1335], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0096, 0.0095, 0.0108, 0.0080, 0.0106, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:23:41,178 INFO [train2.py:809] (3/4) Epoch 17, batch 1450, loss[ctc_loss=0.08669, att_loss=0.2527, loss=0.2195, over 17331.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02219, over 59.00 utterances.], tot_loss[ctc_loss=0.08395, att_loss=0.2413, loss=0.2098, over 3276028.73 frames. utt_duration=1226 frames, utt_pad_proportion=0.05721, over 10698.42 utterances.], batch size: 59, lr: 6.39e-03, grad_scale: 8.0 2023-03-08 15:24:04,392 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3703, 5.4147, 5.2009, 3.3610, 5.1773, 4.9440, 4.4895, 2.9633], device='cuda:3'), covar=tensor([0.0117, 0.0085, 0.0199, 0.0849, 0.0078, 0.0172, 0.0320, 0.1286], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0095, 0.0094, 0.0108, 0.0079, 0.0106, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:24:13,604 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4699, 2.7118, 5.0276, 3.8691, 2.9136, 4.2466, 4.8015, 4.6442], device='cuda:3'), covar=tensor([0.0290, 0.1652, 0.0198, 0.1021, 0.1904, 0.0252, 0.0132, 0.0248], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0246, 0.0171, 0.0318, 0.0269, 0.0202, 0.0150, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 15:24:37,679 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0783, 6.2396, 5.7476, 5.9642, 5.9741, 5.4364, 5.7859, 5.4752], device='cuda:3'), covar=tensor([0.1159, 0.0822, 0.0843, 0.0818, 0.0777, 0.1373, 0.1969, 0.2381], device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0571, 0.0435, 0.0434, 0.0416, 0.0459, 0.0589, 0.0513], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:24:46,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.067e+02 2.518e+02 3.214e+02 7.164e+02, threshold=5.035e+02, percent-clipped=4.0 2023-03-08 15:25:01,181 INFO [train2.py:809] (3/4) Epoch 17, batch 1500, loss[ctc_loss=0.07512, att_loss=0.2462, loss=0.2119, over 16613.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006096, over 47.00 utterances.], tot_loss[ctc_loss=0.08317, att_loss=0.2406, loss=0.2091, over 3280265.65 frames. utt_duration=1239 frames, utt_pad_proportion=0.05244, over 10599.62 utterances.], batch size: 47, lr: 6.39e-03, grad_scale: 8.0 2023-03-08 15:25:08,281 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:26:10,146 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2983, 3.0815, 3.6614, 3.1044, 3.4826, 4.5608, 4.3414, 3.1152], device='cuda:3'), covar=tensor([0.0413, 0.1428, 0.1051, 0.1358, 0.1166, 0.0800, 0.0596, 0.1415], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0239, 0.0268, 0.0212, 0.0255, 0.0348, 0.0247, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:26:11,719 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:26:21,192 INFO [train2.py:809] (3/4) Epoch 17, batch 1550, loss[ctc_loss=0.08197, att_loss=0.2443, loss=0.2119, over 17286.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.0122, over 55.00 utterances.], tot_loss[ctc_loss=0.0829, att_loss=0.2403, loss=0.2088, over 3273403.45 frames. utt_duration=1238 frames, utt_pad_proportion=0.05522, over 10585.79 utterances.], batch size: 55, lr: 6.39e-03, grad_scale: 8.0 2023-03-08 15:26:24,411 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:26:42,549 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:27:11,608 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7880, 6.0107, 5.5067, 5.7819, 5.7026, 5.1877, 5.4003, 5.1499], device='cuda:3'), covar=tensor([0.1276, 0.0922, 0.0961, 0.0700, 0.0927, 0.1583, 0.2412, 0.2320], device='cuda:3'), in_proj_covar=tensor([0.0500, 0.0567, 0.0434, 0.0430, 0.0414, 0.0457, 0.0590, 0.0512], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:27:27,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.050e+02 2.369e+02 2.748e+02 8.749e+02, threshold=4.738e+02, percent-clipped=3.0 2023-03-08 15:27:27,836 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:27:41,718 INFO [train2.py:809] (3/4) Epoch 17, batch 1600, loss[ctc_loss=0.07174, att_loss=0.231, loss=0.1992, over 16390.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007051, over 44.00 utterances.], tot_loss[ctc_loss=0.08277, att_loss=0.2403, loss=0.2088, over 3275163.22 frames. utt_duration=1244 frames, utt_pad_proportion=0.05415, over 10543.32 utterances.], batch size: 44, lr: 6.38e-03, grad_scale: 8.0 2023-03-08 15:27:51,077 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:28:00,084 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:28:10,917 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:28:30,366 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5173, 4.9369, 5.1367, 4.9515, 4.9901, 5.5005, 4.9585, 5.5614], device='cuda:3'), covar=tensor([0.0767, 0.0693, 0.0818, 0.1106, 0.1870, 0.0789, 0.0922, 0.0727], device='cuda:3'), in_proj_covar=tensor([0.0807, 0.0474, 0.0564, 0.0622, 0.0827, 0.0572, 0.0459, 0.0558], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:29:02,427 INFO [train2.py:809] (3/4) Epoch 17, batch 1650, loss[ctc_loss=0.06206, att_loss=0.2083, loss=0.179, over 15640.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008629, over 37.00 utterances.], tot_loss[ctc_loss=0.08258, att_loss=0.2396, loss=0.2082, over 3267117.38 frames. utt_duration=1249 frames, utt_pad_proportion=0.05571, over 10475.87 utterances.], batch size: 37, lr: 6.38e-03, grad_scale: 8.0 2023-03-08 15:29:10,826 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6584, 3.3716, 3.7250, 3.2696, 3.6197, 4.7756, 4.5112, 3.2570], device='cuda:3'), covar=tensor([0.0331, 0.1325, 0.1215, 0.1230, 0.1046, 0.0810, 0.0491, 0.1494], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0241, 0.0269, 0.0214, 0.0256, 0.0350, 0.0249, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:29:20,649 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:29:29,995 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:29:31,733 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:29:36,217 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:30:09,497 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.074e+02 2.531e+02 3.179e+02 8.424e+02, threshold=5.062e+02, percent-clipped=4.0 2023-03-08 15:30:23,559 INFO [train2.py:809] (3/4) Epoch 17, batch 1700, loss[ctc_loss=0.09614, att_loss=0.2576, loss=0.2253, over 17012.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008195, over 51.00 utterances.], tot_loss[ctc_loss=0.08231, att_loss=0.2393, loss=0.2079, over 3264193.01 frames. utt_duration=1274 frames, utt_pad_proportion=0.04949, over 10263.53 utterances.], batch size: 51, lr: 6.38e-03, grad_scale: 8.0 2023-03-08 15:30:29,894 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3407, 2.8790, 3.6236, 3.0692, 3.4720, 4.4909, 4.2288, 2.9907], device='cuda:3'), covar=tensor([0.0401, 0.1750, 0.1210, 0.1253, 0.1043, 0.0857, 0.0622, 0.1437], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0238, 0.0266, 0.0211, 0.0252, 0.0347, 0.0246, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:30:37,508 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:30:48,517 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:30:53,722 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2378, 4.2592, 4.2957, 4.0861, 4.8513, 4.3263, 4.2910, 2.2193], device='cuda:3'), covar=tensor([0.0258, 0.0359, 0.0332, 0.0318, 0.0701, 0.0232, 0.0292, 0.2145], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0159, 0.0166, 0.0180, 0.0360, 0.0140, 0.0151, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:31:05,808 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1258, 5.2391, 4.7292, 2.6467, 4.9044, 4.8120, 4.3097, 2.4866], device='cuda:3'), covar=tensor([0.0177, 0.0086, 0.0302, 0.1334, 0.0116, 0.0182, 0.0404, 0.1977], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0095, 0.0094, 0.0107, 0.0079, 0.0105, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:31:44,395 INFO [train2.py:809] (3/4) Epoch 17, batch 1750, loss[ctc_loss=0.07593, att_loss=0.243, loss=0.2096, over 16900.00 frames. utt_duration=684.1 frames, utt_pad_proportion=0.1427, over 99.00 utterances.], tot_loss[ctc_loss=0.08232, att_loss=0.2395, loss=0.2081, over 3269647.09 frames. utt_duration=1280 frames, utt_pad_proportion=0.04681, over 10226.00 utterances.], batch size: 99, lr: 6.38e-03, grad_scale: 8.0 2023-03-08 15:32:49,895 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 2.005e+02 2.493e+02 3.105e+02 6.516e+02, threshold=4.986e+02, percent-clipped=3.0 2023-03-08 15:32:50,249 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:32:51,778 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4541, 2.0651, 2.1843, 2.5921, 2.6415, 2.4627, 2.2054, 2.4999], device='cuda:3'), covar=tensor([0.1361, 0.2982, 0.2799, 0.1266, 0.1667, 0.1347, 0.2218, 0.1074], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0105, 0.0109, 0.0094, 0.0101, 0.0088, 0.0107, 0.0078], device='cuda:3'), out_proj_covar=tensor([7.1307e-05, 7.9328e-05, 8.2688e-05, 7.0721e-05, 7.3479e-05, 6.9750e-05, 7.9105e-05, 6.2547e-05], device='cuda:3') 2023-03-08 15:33:05,041 INFO [train2.py:809] (3/4) Epoch 17, batch 1800, loss[ctc_loss=0.05816, att_loss=0.2204, loss=0.188, over 15993.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008731, over 40.00 utterances.], tot_loss[ctc_loss=0.08358, att_loss=0.2407, loss=0.2092, over 3262914.23 frames. utt_duration=1212 frames, utt_pad_proportion=0.06512, over 10785.32 utterances.], batch size: 40, lr: 6.37e-03, grad_scale: 8.0 2023-03-08 15:34:19,289 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0722, 5.1201, 4.9735, 2.5052, 2.0025, 2.9669, 2.3039, 3.8396], device='cuda:3'), covar=tensor([0.0691, 0.0255, 0.0252, 0.4736, 0.5650, 0.2382, 0.3472, 0.1699], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0249, 0.0251, 0.0228, 0.0339, 0.0331, 0.0238, 0.0356], device='cuda:3'), out_proj_covar=tensor([1.4757e-04, 9.1661e-05, 1.0781e-04, 9.8479e-05, 1.4326e-04, 1.3052e-04, 9.5255e-05, 1.4617e-04], device='cuda:3') 2023-03-08 15:34:26,299 INFO [train2.py:809] (3/4) Epoch 17, batch 1850, loss[ctc_loss=0.09363, att_loss=0.2624, loss=0.2287, over 17291.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01165, over 55.00 utterances.], tot_loss[ctc_loss=0.08276, att_loss=0.2401, loss=0.2086, over 3266596.67 frames. utt_duration=1238 frames, utt_pad_proportion=0.05693, over 10563.84 utterances.], batch size: 55, lr: 6.37e-03, grad_scale: 8.0 2023-03-08 15:34:29,978 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 15:35:23,990 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8503, 6.1066, 5.6419, 5.9119, 5.8082, 5.3433, 5.5718, 5.3205], device='cuda:3'), covar=tensor([0.1206, 0.0810, 0.0833, 0.0727, 0.0839, 0.1579, 0.2204, 0.2409], device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0571, 0.0434, 0.0429, 0.0413, 0.0458, 0.0586, 0.0508], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:35:31,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 1.987e+02 2.376e+02 2.824e+02 5.455e+02, threshold=4.753e+02, percent-clipped=1.0 2023-03-08 15:35:31,973 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:35:46,843 INFO [train2.py:809] (3/4) Epoch 17, batch 1900, loss[ctc_loss=0.06024, att_loss=0.2178, loss=0.1863, over 16268.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007178, over 43.00 utterances.], tot_loss[ctc_loss=0.08248, att_loss=0.2405, loss=0.2089, over 3274854.66 frames. utt_duration=1244 frames, utt_pad_proportion=0.05382, over 10541.91 utterances.], batch size: 43, lr: 6.37e-03, grad_scale: 8.0 2023-03-08 15:35:47,052 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:36:48,028 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:37:05,993 INFO [train2.py:809] (3/4) Epoch 17, batch 1950, loss[ctc_loss=0.07118, att_loss=0.2442, loss=0.2096, over 16877.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007782, over 49.00 utterances.], tot_loss[ctc_loss=0.08235, att_loss=0.2404, loss=0.2088, over 3278802.43 frames. utt_duration=1249 frames, utt_pad_proportion=0.05172, over 10509.02 utterances.], batch size: 49, lr: 6.37e-03, grad_scale: 8.0 2023-03-08 15:37:36,757 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:38:10,052 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.136e+02 2.531e+02 3.248e+02 6.819e+02, threshold=5.061e+02, percent-clipped=5.0 2023-03-08 15:38:25,203 INFO [train2.py:809] (3/4) Epoch 17, batch 2000, loss[ctc_loss=0.09375, att_loss=0.2545, loss=0.2223, over 17032.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007089, over 51.00 utterances.], tot_loss[ctc_loss=0.08217, att_loss=0.2402, loss=0.2086, over 3275040.91 frames. utt_duration=1260 frames, utt_pad_proportion=0.0499, over 10409.49 utterances.], batch size: 51, lr: 6.37e-03, grad_scale: 8.0 2023-03-08 15:38:30,702 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-08 15:38:53,013 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:39:44,926 INFO [train2.py:809] (3/4) Epoch 17, batch 2050, loss[ctc_loss=0.07712, att_loss=0.2525, loss=0.2174, over 16274.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007208, over 43.00 utterances.], tot_loss[ctc_loss=0.08283, att_loss=0.2411, loss=0.2094, over 3272865.87 frames. utt_duration=1248 frames, utt_pad_proportion=0.05383, over 10506.79 utterances.], batch size: 43, lr: 6.36e-03, grad_scale: 8.0 2023-03-08 15:40:33,394 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2343, 5.2410, 4.9545, 3.1769, 4.9925, 4.7907, 4.4236, 2.5854], device='cuda:3'), covar=tensor([0.0120, 0.0066, 0.0253, 0.0924, 0.0088, 0.0187, 0.0298, 0.1452], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0096, 0.0095, 0.0108, 0.0080, 0.0106, 0.0096, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:40:51,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.921e+02 2.500e+02 3.054e+02 5.987e+02, threshold=5.000e+02, percent-clipped=2.0 2023-03-08 15:41:00,853 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-03-08 15:41:01,845 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:41:06,586 INFO [train2.py:809] (3/4) Epoch 17, batch 2100, loss[ctc_loss=0.0945, att_loss=0.2535, loss=0.2217, over 17454.00 frames. utt_duration=1013 frames, utt_pad_proportion=0.04407, over 69.00 utterances.], tot_loss[ctc_loss=0.08339, att_loss=0.2413, loss=0.2097, over 3278723.49 frames. utt_duration=1256 frames, utt_pad_proportion=0.05064, over 10455.81 utterances.], batch size: 69, lr: 6.36e-03, grad_scale: 8.0 2023-03-08 15:41:53,439 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0121, 4.8512, 5.0499, 2.1399, 1.8295, 2.3097, 2.4507, 3.6165], device='cuda:3'), covar=tensor([0.0840, 0.0538, 0.0252, 0.4622, 0.6733, 0.3640, 0.3271, 0.2002], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0253, 0.0253, 0.0231, 0.0343, 0.0333, 0.0240, 0.0360], device='cuda:3'), out_proj_covar=tensor([1.4905e-04, 9.3160e-05, 1.0859e-04, 9.9867e-05, 1.4483e-04, 1.3126e-04, 9.6005e-05, 1.4770e-04], device='cuda:3') 2023-03-08 15:42:21,798 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 15:42:26,212 INFO [train2.py:809] (3/4) Epoch 17, batch 2150, loss[ctc_loss=0.09771, att_loss=0.2514, loss=0.2207, over 17077.00 frames. utt_duration=1221 frames, utt_pad_proportion=0.01664, over 56.00 utterances.], tot_loss[ctc_loss=0.08357, att_loss=0.2412, loss=0.2096, over 3279940.87 frames. utt_duration=1236 frames, utt_pad_proportion=0.05572, over 10624.82 utterances.], batch size: 56, lr: 6.36e-03, grad_scale: 16.0 2023-03-08 15:42:38,842 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 15:43:30,062 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5481, 3.1264, 3.5004, 2.9814, 3.4424, 4.6079, 4.3467, 3.2893], device='cuda:3'), covar=tensor([0.0370, 0.1689, 0.1320, 0.1446, 0.1217, 0.0782, 0.0679, 0.1343], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0239, 0.0268, 0.0211, 0.0256, 0.0350, 0.0247, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 15:43:31,159 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 2.187e+02 2.638e+02 3.352e+02 9.129e+02, threshold=5.276e+02, percent-clipped=7.0 2023-03-08 15:43:45,842 INFO [train2.py:809] (3/4) Epoch 17, batch 2200, loss[ctc_loss=0.06159, att_loss=0.2253, loss=0.1926, over 16124.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.00631, over 42.00 utterances.], tot_loss[ctc_loss=0.08261, att_loss=0.2404, loss=0.2088, over 3279628.17 frames. utt_duration=1255 frames, utt_pad_proportion=0.05171, over 10465.15 utterances.], batch size: 42, lr: 6.36e-03, grad_scale: 16.0 2023-03-08 15:43:46,152 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:45:01,931 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:45:04,882 INFO [train2.py:809] (3/4) Epoch 17, batch 2250, loss[ctc_loss=0.06355, att_loss=0.2262, loss=0.1937, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006662, over 45.00 utterances.], tot_loss[ctc_loss=0.08199, att_loss=0.2397, loss=0.2082, over 3276227.59 frames. utt_duration=1243 frames, utt_pad_proportion=0.05486, over 10551.45 utterances.], batch size: 45, lr: 6.35e-03, grad_scale: 16.0 2023-03-08 15:45:15,523 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0219, 4.2264, 4.0306, 4.5034, 2.6633, 4.2856, 2.5965, 1.9398], device='cuda:3'), covar=tensor([0.0367, 0.0197, 0.0738, 0.0167, 0.1721, 0.0171, 0.1576, 0.1738], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0145, 0.0262, 0.0139, 0.0223, 0.0126, 0.0232, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:45:54,157 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2465, 3.7686, 3.4137, 3.5875, 4.0774, 3.6814, 3.2120, 4.3635], device='cuda:3'), covar=tensor([0.0863, 0.0550, 0.0919, 0.0592, 0.0613, 0.0713, 0.0747, 0.0394], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0209, 0.0221, 0.0193, 0.0265, 0.0234, 0.0196, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 15:46:12,740 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.093e+02 2.407e+02 3.008e+02 8.516e+02, threshold=4.813e+02, percent-clipped=4.0 2023-03-08 15:46:27,302 INFO [train2.py:809] (3/4) Epoch 17, batch 2300, loss[ctc_loss=0.08628, att_loss=0.2433, loss=0.2119, over 16760.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.005423, over 48.00 utterances.], tot_loss[ctc_loss=0.08343, att_loss=0.2407, loss=0.2093, over 3274886.69 frames. utt_duration=1216 frames, utt_pad_proportion=0.06238, over 10784.70 utterances.], batch size: 48, lr: 6.35e-03, grad_scale: 8.0 2023-03-08 15:47:34,923 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8693, 6.0916, 5.5563, 5.8928, 5.7072, 5.2714, 5.5819, 5.2004], device='cuda:3'), covar=tensor([0.1354, 0.0943, 0.0946, 0.0817, 0.0895, 0.1719, 0.2375, 0.2439], device='cuda:3'), in_proj_covar=tensor([0.0499, 0.0573, 0.0438, 0.0435, 0.0415, 0.0455, 0.0592, 0.0510], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:47:48,113 INFO [train2.py:809] (3/4) Epoch 17, batch 2350, loss[ctc_loss=0.08836, att_loss=0.2437, loss=0.2126, over 17038.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01047, over 53.00 utterances.], tot_loss[ctc_loss=0.08303, att_loss=0.2402, loss=0.2088, over 3270774.45 frames. utt_duration=1237 frames, utt_pad_proportion=0.05815, over 10585.23 utterances.], batch size: 53, lr: 6.35e-03, grad_scale: 8.0 2023-03-08 15:48:02,658 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0139, 4.3012, 3.9274, 4.4895, 2.7026, 4.2316, 2.5661, 2.5752], device='cuda:3'), covar=tensor([0.0370, 0.0196, 0.0815, 0.0172, 0.1597, 0.0199, 0.1568, 0.1297], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0144, 0.0258, 0.0138, 0.0220, 0.0125, 0.0229, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:48:55,986 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.214e+02 2.683e+02 3.352e+02 8.000e+02, threshold=5.366e+02, percent-clipped=7.0 2023-03-08 15:49:09,126 INFO [train2.py:809] (3/4) Epoch 17, batch 2400, loss[ctc_loss=0.07007, att_loss=0.2243, loss=0.1935, over 15865.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009296, over 39.00 utterances.], tot_loss[ctc_loss=0.08161, att_loss=0.2391, loss=0.2076, over 3269700.24 frames. utt_duration=1264 frames, utt_pad_proportion=0.0525, over 10355.52 utterances.], batch size: 39, lr: 6.35e-03, grad_scale: 8.0 2023-03-08 15:49:11,812 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-08 15:49:12,632 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6371, 4.8823, 4.8970, 4.8614, 4.9263, 4.9185, 4.6087, 4.4245], device='cuda:3'), covar=tensor([0.1115, 0.0658, 0.0324, 0.0549, 0.0332, 0.0358, 0.0438, 0.0387], device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0336, 0.0313, 0.0328, 0.0396, 0.0405, 0.0331, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 15:49:14,400 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:49:39,859 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:49:51,473 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 15:50:19,980 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7464, 6.0377, 5.4561, 5.7565, 5.6689, 5.2912, 5.4094, 5.1728], device='cuda:3'), covar=tensor([0.1329, 0.0828, 0.0848, 0.0817, 0.0847, 0.1390, 0.2368, 0.2271], device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0562, 0.0430, 0.0430, 0.0409, 0.0447, 0.0579, 0.0501], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 15:50:24,741 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 15:50:29,108 INFO [train2.py:809] (3/4) Epoch 17, batch 2450, loss[ctc_loss=0.08225, att_loss=0.2212, loss=0.1934, over 15648.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008043, over 37.00 utterances.], tot_loss[ctc_loss=0.08204, att_loss=0.2398, loss=0.2083, over 3271249.32 frames. utt_duration=1272 frames, utt_pad_proportion=0.04955, over 10295.72 utterances.], batch size: 37, lr: 6.34e-03, grad_scale: 8.0 2023-03-08 15:50:33,906 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 15:50:51,807 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:51:16,846 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 15:51:29,753 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6079, 2.8374, 5.0696, 3.9094, 3.0682, 4.4211, 4.9001, 4.6463], device='cuda:3'), covar=tensor([0.0275, 0.1463, 0.0227, 0.1002, 0.1784, 0.0253, 0.0130, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0245, 0.0170, 0.0314, 0.0269, 0.0201, 0.0152, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 15:51:35,450 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 1.994e+02 2.398e+02 2.990e+02 4.880e+02, threshold=4.796e+02, percent-clipped=0.0 2023-03-08 15:51:40,271 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 15:51:48,522 INFO [train2.py:809] (3/4) Epoch 17, batch 2500, loss[ctc_loss=0.07683, att_loss=0.2226, loss=0.1934, over 16287.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006869, over 43.00 utterances.], tot_loss[ctc_loss=0.08258, att_loss=0.2398, loss=0.2084, over 3269841.36 frames. utt_duration=1268 frames, utt_pad_proportion=0.05052, over 10328.30 utterances.], batch size: 43, lr: 6.34e-03, grad_scale: 8.0 2023-03-08 15:52:16,108 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:53:08,205 INFO [train2.py:809] (3/4) Epoch 17, batch 2550, loss[ctc_loss=0.08004, att_loss=0.2386, loss=0.2069, over 16318.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006886, over 45.00 utterances.], tot_loss[ctc_loss=0.08221, att_loss=0.2396, loss=0.2081, over 3266071.86 frames. utt_duration=1266 frames, utt_pad_proportion=0.05258, over 10334.20 utterances.], batch size: 45, lr: 6.34e-03, grad_scale: 8.0 2023-03-08 15:53:22,189 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1656, 5.3987, 4.8736, 5.2418, 5.0028, 4.6011, 4.8950, 4.6202], device='cuda:3'), covar=tensor([0.1261, 0.1011, 0.0959, 0.0872, 0.1016, 0.1587, 0.2177, 0.2367], device='cuda:3'), in_proj_covar=tensor([0.0495, 0.0570, 0.0435, 0.0432, 0.0415, 0.0453, 0.0587, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 15:53:52,746 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 15:54:14,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 2.284e+02 2.590e+02 3.284e+02 7.249e+02, threshold=5.180e+02, percent-clipped=6.0 2023-03-08 15:54:27,436 INFO [train2.py:809] (3/4) Epoch 17, batch 2600, loss[ctc_loss=0.06408, att_loss=0.2346, loss=0.2005, over 16540.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006497, over 45.00 utterances.], tot_loss[ctc_loss=0.08322, att_loss=0.2402, loss=0.2088, over 3266969.60 frames. utt_duration=1245 frames, utt_pad_proportion=0.05786, over 10510.41 utterances.], batch size: 45, lr: 6.34e-03, grad_scale: 8.0 2023-03-08 15:55:06,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7578, 3.6459, 3.5512, 3.1228, 3.6545, 3.7591, 3.6150, 2.7305], device='cuda:3'), covar=tensor([0.0935, 0.1121, 0.1877, 0.3843, 0.1396, 0.1651, 0.0831, 0.4188], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0162, 0.0175, 0.0235, 0.0138, 0.0231, 0.0152, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 15:55:36,733 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.28 vs. limit=5.0 2023-03-08 15:55:47,247 INFO [train2.py:809] (3/4) Epoch 17, batch 2650, loss[ctc_loss=0.0723, att_loss=0.231, loss=0.1993, over 16136.00 frames. utt_duration=1539 frames, utt_pad_proportion=0.005479, over 42.00 utterances.], tot_loss[ctc_loss=0.08328, att_loss=0.2402, loss=0.2088, over 3257573.30 frames. utt_duration=1237 frames, utt_pad_proportion=0.06212, over 10548.58 utterances.], batch size: 42, lr: 6.33e-03, grad_scale: 8.0 2023-03-08 15:56:19,788 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5475, 1.6943, 2.2505, 2.3456, 2.1207, 2.1196, 1.9160, 2.2664], device='cuda:3'), covar=tensor([0.1202, 0.2936, 0.2273, 0.1477, 0.1697, 0.1112, 0.1688, 0.0905], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0104, 0.0108, 0.0092, 0.0101, 0.0088, 0.0107, 0.0079], device='cuda:3'), out_proj_covar=tensor([7.1141e-05, 7.8826e-05, 8.2416e-05, 7.0041e-05, 7.3408e-05, 6.9678e-05, 7.9129e-05, 6.2829e-05], device='cuda:3') 2023-03-08 15:56:28,550 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0016, 6.2082, 5.7196, 5.9639, 5.8693, 5.3845, 5.7326, 5.4715], device='cuda:3'), covar=tensor([0.1203, 0.0780, 0.0808, 0.0742, 0.0810, 0.1515, 0.1999, 0.2317], device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0573, 0.0438, 0.0439, 0.0418, 0.0456, 0.0587, 0.0511], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 15:56:53,521 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 2.144e+02 2.549e+02 3.292e+02 5.806e+02, threshold=5.098e+02, percent-clipped=4.0 2023-03-08 15:57:06,281 INFO [train2.py:809] (3/4) Epoch 17, batch 2700, loss[ctc_loss=0.09605, att_loss=0.2509, loss=0.22, over 16939.00 frames. utt_duration=685.9 frames, utt_pad_proportion=0.1394, over 99.00 utterances.], tot_loss[ctc_loss=0.08491, att_loss=0.2411, loss=0.2099, over 3259644.44 frames. utt_duration=1203 frames, utt_pad_proportion=0.0707, over 10849.09 utterances.], batch size: 99, lr: 6.33e-03, grad_scale: 8.0 2023-03-08 15:57:25,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 15:58:06,562 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2484, 2.7015, 3.2948, 4.3822, 3.8174, 3.8761, 2.8952, 2.2975], device='cuda:3'), covar=tensor([0.0735, 0.2151, 0.0946, 0.0522, 0.0834, 0.0458, 0.1499, 0.2064], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0214, 0.0187, 0.0206, 0.0211, 0.0170, 0.0197, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 15:58:12,536 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-08 15:58:25,425 INFO [train2.py:809] (3/4) Epoch 17, batch 2750, loss[ctc_loss=0.08158, att_loss=0.2131, loss=0.1868, over 14541.00 frames. utt_duration=1819 frames, utt_pad_proportion=0.0365, over 32.00 utterances.], tot_loss[ctc_loss=0.08429, att_loss=0.2405, loss=0.2093, over 3244263.38 frames. utt_duration=1206 frames, utt_pad_proportion=0.07218, over 10772.39 utterances.], batch size: 32, lr: 6.33e-03, grad_scale: 8.0 2023-03-08 15:58:30,326 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 15:58:40,087 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 15:58:43,799 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 15:59:05,253 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 15:59:12,735 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 15:59:31,461 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.407e+02 2.169e+02 2.491e+02 3.348e+02 1.363e+03, threshold=4.982e+02, percent-clipped=6.0 2023-03-08 15:59:43,576 INFO [train2.py:809] (3/4) Epoch 17, batch 2800, loss[ctc_loss=0.1199, att_loss=0.2682, loss=0.2385, over 16624.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005454, over 47.00 utterances.], tot_loss[ctc_loss=0.08416, att_loss=0.2409, loss=0.2095, over 3257616.02 frames. utt_duration=1216 frames, utt_pad_proportion=0.0658, over 10729.44 utterances.], batch size: 47, lr: 6.33e-03, grad_scale: 8.0 2023-03-08 15:59:45,180 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:00:48,786 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:01:02,316 INFO [train2.py:809] (3/4) Epoch 17, batch 2850, loss[ctc_loss=0.07974, att_loss=0.2405, loss=0.2083, over 16973.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007232, over 50.00 utterances.], tot_loss[ctc_loss=0.08381, att_loss=0.241, loss=0.2096, over 3261541.05 frames. utt_duration=1222 frames, utt_pad_proportion=0.06363, over 10691.65 utterances.], batch size: 50, lr: 6.32e-03, grad_scale: 8.0 2023-03-08 16:01:39,565 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:02:09,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.066e+02 2.507e+02 3.083e+02 5.054e+02, threshold=5.014e+02, percent-clipped=1.0 2023-03-08 16:02:21,164 INFO [train2.py:809] (3/4) Epoch 17, batch 2900, loss[ctc_loss=0.1234, att_loss=0.2638, loss=0.2357, over 17449.00 frames. utt_duration=1109 frames, utt_pad_proportion=0.03034, over 63.00 utterances.], tot_loss[ctc_loss=0.0842, att_loss=0.2412, loss=0.2098, over 3262234.83 frames. utt_duration=1189 frames, utt_pad_proportion=0.07265, over 10990.37 utterances.], batch size: 63, lr: 6.32e-03, grad_scale: 8.0 2023-03-08 16:03:41,156 INFO [train2.py:809] (3/4) Epoch 17, batch 2950, loss[ctc_loss=0.07511, att_loss=0.2282, loss=0.1976, over 15891.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008926, over 39.00 utterances.], tot_loss[ctc_loss=0.08462, att_loss=0.2419, loss=0.2104, over 3265788.93 frames. utt_duration=1184 frames, utt_pad_proportion=0.07337, over 11043.19 utterances.], batch size: 39, lr: 6.32e-03, grad_scale: 8.0 2023-03-08 16:04:44,859 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:04:47,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.044e+02 2.450e+02 3.054e+02 6.191e+02, threshold=4.900e+02, percent-clipped=3.0 2023-03-08 16:04:52,559 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9989, 5.3242, 4.9246, 5.4280, 4.8026, 5.0633, 5.5019, 5.2581], device='cuda:3'), covar=tensor([0.0597, 0.0289, 0.0703, 0.0257, 0.0397, 0.0230, 0.0210, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0297, 0.0347, 0.0312, 0.0302, 0.0223, 0.0281, 0.0264], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 16:05:00,733 INFO [train2.py:809] (3/4) Epoch 17, batch 3000, loss[ctc_loss=0.05737, att_loss=0.2204, loss=0.1878, over 15884.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009393, over 39.00 utterances.], tot_loss[ctc_loss=0.08338, att_loss=0.2416, loss=0.2099, over 3275104.97 frames. utt_duration=1208 frames, utt_pad_proportion=0.06542, over 10860.78 utterances.], batch size: 39, lr: 6.32e-03, grad_scale: 8.0 2023-03-08 16:05:00,733 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 16:05:14,967 INFO [train2.py:843] (3/4) Epoch 17, validation: ctc_loss=0.04199, att_loss=0.2349, loss=0.1964, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 16:05:14,968 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 16:06:16,607 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7134, 6.0034, 5.4697, 5.7087, 5.6310, 5.2648, 5.4121, 5.2658], device='cuda:3'), covar=tensor([0.1313, 0.0959, 0.0983, 0.0876, 0.0949, 0.1678, 0.2630, 0.2411], device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0574, 0.0442, 0.0441, 0.0417, 0.0456, 0.0594, 0.0511], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 16:06:34,021 INFO [train2.py:809] (3/4) Epoch 17, batch 3050, loss[ctc_loss=0.09463, att_loss=0.2548, loss=0.2228, over 17281.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01164, over 55.00 utterances.], tot_loss[ctc_loss=0.08281, att_loss=0.2416, loss=0.2098, over 3283795.08 frames. utt_duration=1226 frames, utt_pad_proportion=0.05773, over 10728.71 utterances.], batch size: 55, lr: 6.32e-03, grad_scale: 4.0 2023-03-08 16:06:35,946 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:06:49,683 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:07:14,792 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 16:07:42,430 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 2.123e+02 2.532e+02 3.472e+02 1.749e+03, threshold=5.065e+02, percent-clipped=12.0 2023-03-08 16:07:53,810 INFO [train2.py:809] (3/4) Epoch 17, batch 3100, loss[ctc_loss=0.07683, att_loss=0.2271, loss=0.1971, over 15516.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.007859, over 36.00 utterances.], tot_loss[ctc_loss=0.08257, att_loss=0.2413, loss=0.2096, over 3268686.91 frames. utt_duration=1238 frames, utt_pad_proportion=0.05594, over 10576.71 utterances.], batch size: 36, lr: 6.31e-03, grad_scale: 4.0 2023-03-08 16:08:05,853 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:08:30,883 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:08:39,373 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-03-08 16:08:50,689 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:09:12,469 INFO [train2.py:809] (3/4) Epoch 17, batch 3150, loss[ctc_loss=0.07715, att_loss=0.2402, loss=0.2076, over 16773.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006052, over 48.00 utterances.], tot_loss[ctc_loss=0.08216, att_loss=0.2405, loss=0.2088, over 3261490.53 frames. utt_duration=1242 frames, utt_pad_proportion=0.05675, over 10519.57 utterances.], batch size: 48, lr: 6.31e-03, grad_scale: 4.0 2023-03-08 16:09:50,618 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:10:21,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 2.012e+02 2.406e+02 2.883e+02 7.010e+02, threshold=4.812e+02, percent-clipped=2.0 2023-03-08 16:10:32,788 INFO [train2.py:809] (3/4) Epoch 17, batch 3200, loss[ctc_loss=0.07892, att_loss=0.2439, loss=0.2109, over 17188.00 frames. utt_duration=695.9 frames, utt_pad_proportion=0.1268, over 99.00 utterances.], tot_loss[ctc_loss=0.081, att_loss=0.2394, loss=0.2077, over 3263303.19 frames. utt_duration=1272 frames, utt_pad_proportion=0.04952, over 10273.66 utterances.], batch size: 99, lr: 6.31e-03, grad_scale: 8.0 2023-03-08 16:10:58,927 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:11:06,328 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:11:15,760 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2695, 5.2400, 5.0229, 3.2115, 5.0112, 4.8342, 4.4962, 2.9180], device='cuda:3'), covar=tensor([0.0106, 0.0087, 0.0237, 0.0870, 0.0088, 0.0176, 0.0265, 0.1264], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0096, 0.0096, 0.0109, 0.0081, 0.0106, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 16:11:51,565 INFO [train2.py:809] (3/4) Epoch 17, batch 3250, loss[ctc_loss=0.05128, att_loss=0.2129, loss=0.1805, over 15775.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008353, over 38.00 utterances.], tot_loss[ctc_loss=0.08139, att_loss=0.2401, loss=0.2084, over 3269480.02 frames. utt_duration=1246 frames, utt_pad_proportion=0.05459, over 10511.18 utterances.], batch size: 38, lr: 6.31e-03, grad_scale: 8.0 2023-03-08 16:12:19,458 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7997, 4.2572, 4.4878, 4.3279, 4.3751, 4.7331, 4.4199, 4.7852], device='cuda:3'), covar=tensor([0.0829, 0.0924, 0.0842, 0.1223, 0.1843, 0.0964, 0.2036, 0.0808], device='cuda:3'), in_proj_covar=tensor([0.0818, 0.0480, 0.0571, 0.0634, 0.0828, 0.0587, 0.0463, 0.0572], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:12:34,788 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 16:12:58,913 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.115e+02 2.445e+02 2.942e+02 7.309e+02, threshold=4.890e+02, percent-clipped=3.0 2023-03-08 16:13:11,488 INFO [train2.py:809] (3/4) Epoch 17, batch 3300, loss[ctc_loss=0.1386, att_loss=0.269, loss=0.2429, over 17281.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02261, over 59.00 utterances.], tot_loss[ctc_loss=0.08222, att_loss=0.2407, loss=0.209, over 3263630.31 frames. utt_duration=1211 frames, utt_pad_proportion=0.0649, over 10792.53 utterances.], batch size: 59, lr: 6.30e-03, grad_scale: 8.0 2023-03-08 16:13:33,234 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-03-08 16:14:17,399 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:14:24,017 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 16:14:31,293 INFO [train2.py:809] (3/4) Epoch 17, batch 3350, loss[ctc_loss=0.06322, att_loss=0.2499, loss=0.2126, over 16796.00 frames. utt_duration=1401 frames, utt_pad_proportion=0.004838, over 48.00 utterances.], tot_loss[ctc_loss=0.08153, att_loss=0.24, loss=0.2083, over 3264607.67 frames. utt_duration=1242 frames, utt_pad_proportion=0.05799, over 10526.20 utterances.], batch size: 48, lr: 6.30e-03, grad_scale: 8.0 2023-03-08 16:15:37,321 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.015e+02 2.419e+02 3.041e+02 5.956e+02, threshold=4.838e+02, percent-clipped=4.0 2023-03-08 16:15:49,840 INFO [train2.py:809] (3/4) Epoch 17, batch 3400, loss[ctc_loss=0.09647, att_loss=0.26, loss=0.2273, over 17393.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03423, over 63.00 utterances.], tot_loss[ctc_loss=0.08231, att_loss=0.2405, loss=0.2089, over 3264250.29 frames. utt_duration=1223 frames, utt_pad_proportion=0.06173, over 10692.21 utterances.], batch size: 63, lr: 6.30e-03, grad_scale: 8.0 2023-03-08 16:15:54,782 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:16:45,568 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:17:08,455 INFO [train2.py:809] (3/4) Epoch 17, batch 3450, loss[ctc_loss=0.08328, att_loss=0.2517, loss=0.218, over 16953.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.00824, over 50.00 utterances.], tot_loss[ctc_loss=0.08166, att_loss=0.2401, loss=0.2084, over 3271231.54 frames. utt_duration=1243 frames, utt_pad_proportion=0.05455, over 10541.25 utterances.], batch size: 50, lr: 6.30e-03, grad_scale: 8.0 2023-03-08 16:18:01,727 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:18:15,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.178e+02 2.510e+02 3.172e+02 6.489e+02, threshold=5.020e+02, percent-clipped=3.0 2023-03-08 16:18:27,085 INFO [train2.py:809] (3/4) Epoch 17, batch 3500, loss[ctc_loss=0.1001, att_loss=0.2515, loss=0.2212, over 17134.00 frames. utt_duration=868.9 frames, utt_pad_proportion=0.09108, over 79.00 utterances.], tot_loss[ctc_loss=0.08174, att_loss=0.2402, loss=0.2085, over 3268332.79 frames. utt_duration=1264 frames, utt_pad_proportion=0.05079, over 10354.98 utterances.], batch size: 79, lr: 6.29e-03, grad_scale: 8.0 2023-03-08 16:19:25,195 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6137, 3.1963, 3.7381, 2.9852, 3.7202, 4.7602, 4.5314, 3.4264], device='cuda:3'), covar=tensor([0.0383, 0.1587, 0.1150, 0.1514, 0.1039, 0.0755, 0.0555, 0.1264], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0238, 0.0270, 0.0213, 0.0258, 0.0350, 0.0248, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:19:46,689 INFO [train2.py:809] (3/4) Epoch 17, batch 3550, loss[ctc_loss=0.07514, att_loss=0.2556, loss=0.2195, over 17298.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02395, over 59.00 utterances.], tot_loss[ctc_loss=0.08191, att_loss=0.2405, loss=0.2088, over 3277014.27 frames. utt_duration=1270 frames, utt_pad_proportion=0.04607, over 10331.84 utterances.], batch size: 59, lr: 6.29e-03, grad_scale: 8.0 2023-03-08 16:20:05,748 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6117, 5.9666, 5.5266, 5.7112, 5.6572, 5.1749, 5.3688, 5.1485], device='cuda:3'), covar=tensor([0.1597, 0.0891, 0.0866, 0.0870, 0.0913, 0.1689, 0.2294, 0.2272], device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0582, 0.0440, 0.0439, 0.0418, 0.0457, 0.0596, 0.0519], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 16:20:21,502 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:20:54,011 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.300e+02 1.995e+02 2.311e+02 2.839e+02 6.877e+02, threshold=4.621e+02, percent-clipped=1.0 2023-03-08 16:21:06,790 INFO [train2.py:809] (3/4) Epoch 17, batch 3600, loss[ctc_loss=0.0995, att_loss=0.2589, loss=0.227, over 17411.00 frames. utt_duration=883.2 frames, utt_pad_proportion=0.07422, over 79.00 utterances.], tot_loss[ctc_loss=0.08165, att_loss=0.2399, loss=0.2083, over 3269251.08 frames. utt_duration=1264 frames, utt_pad_proportion=0.05065, over 10361.35 utterances.], batch size: 79, lr: 6.29e-03, grad_scale: 8.0 2023-03-08 16:22:19,259 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 16:22:26,316 INFO [train2.py:809] (3/4) Epoch 17, batch 3650, loss[ctc_loss=0.1174, att_loss=0.2591, loss=0.2308, over 17143.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01358, over 56.00 utterances.], tot_loss[ctc_loss=0.08178, att_loss=0.2402, loss=0.2085, over 3275730.57 frames. utt_duration=1269 frames, utt_pad_proportion=0.0473, over 10340.85 utterances.], batch size: 56, lr: 6.29e-03, grad_scale: 8.0 2023-03-08 16:23:15,735 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9345, 3.5312, 3.0636, 3.2402, 3.6325, 3.3703, 2.7386, 3.7719], device='cuda:3'), covar=tensor([0.0992, 0.0461, 0.1127, 0.0707, 0.0755, 0.0725, 0.0950, 0.0532], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0207, 0.0218, 0.0191, 0.0264, 0.0231, 0.0194, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 16:23:30,783 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-03-08 16:23:34,282 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.053e+02 2.468e+02 2.850e+02 9.204e+02, threshold=4.935e+02, percent-clipped=2.0 2023-03-08 16:23:37,290 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:23:44,140 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:23:47,245 INFO [train2.py:809] (3/4) Epoch 17, batch 3700, loss[ctc_loss=0.07588, att_loss=0.2501, loss=0.2152, over 17361.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02056, over 59.00 utterances.], tot_loss[ctc_loss=0.08212, att_loss=0.2408, loss=0.2091, over 3285666.82 frames. utt_duration=1242 frames, utt_pad_proportion=0.05117, over 10596.08 utterances.], batch size: 59, lr: 6.28e-03, grad_scale: 8.0 2023-03-08 16:25:02,516 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8714, 6.0778, 5.5481, 5.7762, 5.7451, 5.1876, 5.5329, 5.2704], device='cuda:3'), covar=tensor([0.1134, 0.0857, 0.0810, 0.0837, 0.0824, 0.1497, 0.1970, 0.2434], device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0585, 0.0439, 0.0443, 0.0416, 0.0455, 0.0593, 0.0518], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 16:25:07,050 INFO [train2.py:809] (3/4) Epoch 17, batch 3750, loss[ctc_loss=0.06149, att_loss=0.2257, loss=0.1929, over 15496.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008551, over 36.00 utterances.], tot_loss[ctc_loss=0.08214, att_loss=0.24, loss=0.2084, over 3270596.04 frames. utt_duration=1240 frames, utt_pad_proportion=0.05677, over 10564.52 utterances.], batch size: 36, lr: 6.28e-03, grad_scale: 8.0 2023-03-08 16:25:42,197 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:26:14,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.055e+02 2.519e+02 3.069e+02 5.925e+02, threshold=5.038e+02, percent-clipped=2.0 2023-03-08 16:26:25,910 INFO [train2.py:809] (3/4) Epoch 17, batch 3800, loss[ctc_loss=0.1389, att_loss=0.2645, loss=0.2394, over 13924.00 frames. utt_duration=380.5 frames, utt_pad_proportion=0.3337, over 147.00 utterances.], tot_loss[ctc_loss=0.08261, att_loss=0.2409, loss=0.2093, over 3280232.68 frames. utt_duration=1231 frames, utt_pad_proportion=0.05718, over 10672.64 utterances.], batch size: 147, lr: 6.28e-03, grad_scale: 8.0 2023-03-08 16:27:18,187 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:27:46,049 INFO [train2.py:809] (3/4) Epoch 17, batch 3850, loss[ctc_loss=0.06105, att_loss=0.235, loss=0.2002, over 16131.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.00611, over 42.00 utterances.], tot_loss[ctc_loss=0.08205, att_loss=0.2406, loss=0.2089, over 3281835.13 frames. utt_duration=1233 frames, utt_pad_proportion=0.05659, over 10661.93 utterances.], batch size: 42, lr: 6.28e-03, grad_scale: 8.0 2023-03-08 16:28:20,443 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:28:53,213 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 2.153e+02 2.695e+02 3.326e+02 5.215e+02, threshold=5.391e+02, percent-clipped=2.0 2023-03-08 16:28:59,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6182, 3.1693, 3.7464, 3.0508, 3.6472, 4.6912, 4.5173, 3.3395], device='cuda:3'), covar=tensor([0.0360, 0.1595, 0.1175, 0.1266, 0.0996, 0.0803, 0.0566, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0241, 0.0270, 0.0212, 0.0257, 0.0352, 0.0249, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:29:04,166 INFO [train2.py:809] (3/4) Epoch 17, batch 3900, loss[ctc_loss=0.07768, att_loss=0.2405, loss=0.2079, over 16556.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005252, over 45.00 utterances.], tot_loss[ctc_loss=0.08207, att_loss=0.2403, loss=0.2086, over 3278806.16 frames. utt_duration=1218 frames, utt_pad_proportion=0.06121, over 10782.04 utterances.], batch size: 45, lr: 6.28e-03, grad_scale: 8.0 2023-03-08 16:29:34,771 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:30:04,353 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:30:21,235 INFO [train2.py:809] (3/4) Epoch 17, batch 3950, loss[ctc_loss=0.07832, att_loss=0.2381, loss=0.2062, over 16401.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007571, over 44.00 utterances.], tot_loss[ctc_loss=0.08217, att_loss=0.2403, loss=0.2087, over 3261224.98 frames. utt_duration=1211 frames, utt_pad_proportion=0.06563, over 10786.77 utterances.], batch size: 44, lr: 6.27e-03, grad_scale: 8.0 2023-03-08 16:30:23,194 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:30:50,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1100, 5.4352, 4.9750, 5.5213, 4.8940, 5.1137, 5.5899, 5.3352], device='cuda:3'), covar=tensor([0.0584, 0.0276, 0.0786, 0.0284, 0.0371, 0.0198, 0.0197, 0.0168], device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0298, 0.0349, 0.0316, 0.0301, 0.0225, 0.0285, 0.0268], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 16:31:06,489 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6797, 2.6153, 3.9689, 3.4884, 2.9407, 3.7037, 3.6318, 3.6979], device='cuda:3'), covar=tensor([0.0299, 0.1389, 0.0153, 0.0810, 0.1381, 0.0278, 0.0197, 0.0291], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0239, 0.0168, 0.0305, 0.0263, 0.0199, 0.0151, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:31:39,751 INFO [train2.py:809] (3/4) Epoch 18, batch 0, loss[ctc_loss=0.1199, att_loss=0.2679, loss=0.2383, over 17055.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.007239, over 52.00 utterances.], tot_loss[ctc_loss=0.1199, att_loss=0.2679, loss=0.2383, over 17055.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.007239, over 52.00 utterances.], batch size: 52, lr: 6.09e-03, grad_scale: 8.0 2023-03-08 16:31:39,752 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 16:31:52,765 INFO [train2.py:843] (3/4) Epoch 18, validation: ctc_loss=0.04224, att_loss=0.2352, loss=0.1966, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 16:31:52,766 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 16:32:06,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 2.121e+02 2.560e+02 3.250e+02 7.102e+02, threshold=5.120e+02, percent-clipped=3.0 2023-03-08 16:32:14,452 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:32:17,603 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:32:31,008 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0859, 2.2908, 3.4808, 2.2015, 3.2810, 4.3295, 4.3042, 2.4461], device='cuda:3'), covar=tensor([0.0545, 0.2324, 0.0982, 0.2054, 0.0953, 0.0669, 0.0522, 0.2101], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0241, 0.0271, 0.0214, 0.0259, 0.0353, 0.0249, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:32:33,006 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-08 16:32:35,642 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:32:40,105 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1813, 5.4633, 4.9078, 5.2671, 5.0861, 4.6976, 4.9457, 4.7119], device='cuda:3'), covar=tensor([0.1326, 0.0920, 0.0944, 0.0875, 0.1035, 0.1576, 0.2203, 0.2392], device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0570, 0.0432, 0.0434, 0.0412, 0.0446, 0.0584, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 16:33:03,899 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-08 16:33:11,762 INFO [train2.py:809] (3/4) Epoch 18, batch 50, loss[ctc_loss=0.06245, att_loss=0.2214, loss=0.1896, over 15659.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.007928, over 37.00 utterances.], tot_loss[ctc_loss=0.08291, att_loss=0.2393, loss=0.2081, over 734947.88 frames. utt_duration=1282 frames, utt_pad_proportion=0.04645, over 2295.45 utterances.], batch size: 37, lr: 6.09e-03, grad_scale: 8.0 2023-03-08 16:33:21,480 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4903, 2.2505, 2.0219, 1.9545, 2.9741, 2.4623, 2.1643, 2.7025], device='cuda:3'), covar=tensor([0.1772, 0.3615, 0.3149, 0.2208, 0.1553, 0.2475, 0.2706, 0.1117], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0111, 0.0115, 0.0098, 0.0106, 0.0092, 0.0114, 0.0084], device='cuda:3'), out_proj_covar=tensor([7.5619e-05, 8.4314e-05, 8.8094e-05, 7.4526e-05, 7.7994e-05, 7.3580e-05, 8.4376e-05, 6.7383e-05], device='cuda:3') 2023-03-08 16:33:30,478 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:34:31,359 INFO [train2.py:809] (3/4) Epoch 18, batch 100, loss[ctc_loss=0.0732, att_loss=0.2508, loss=0.2153, over 17313.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01066, over 55.00 utterances.], tot_loss[ctc_loss=0.08278, att_loss=0.2393, loss=0.208, over 1296959.32 frames. utt_duration=1287 frames, utt_pad_proportion=0.0485, over 4034.22 utterances.], batch size: 55, lr: 6.09e-03, grad_scale: 8.0 2023-03-08 16:34:38,089 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8700, 3.5740, 3.5826, 3.1525, 3.5746, 3.7052, 3.6455, 2.7974], device='cuda:3'), covar=tensor([0.0911, 0.1742, 0.2510, 0.4038, 0.3005, 0.2100, 0.1111, 0.3985], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0163, 0.0172, 0.0234, 0.0140, 0.0230, 0.0151, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:34:45,115 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.310e+02 2.796e+02 3.498e+02 6.865e+02, threshold=5.593e+02, percent-clipped=9.0 2023-03-08 16:35:40,468 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 16:35:49,982 INFO [train2.py:809] (3/4) Epoch 18, batch 150, loss[ctc_loss=0.08912, att_loss=0.2594, loss=0.2253, over 16778.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005963, over 48.00 utterances.], tot_loss[ctc_loss=0.08109, att_loss=0.2394, loss=0.2077, over 1735629.47 frames. utt_duration=1296 frames, utt_pad_proportion=0.04489, over 5362.16 utterances.], batch size: 48, lr: 6.09e-03, grad_scale: 8.0 2023-03-08 16:36:13,769 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1653, 4.4606, 4.2625, 4.6023, 2.8066, 4.5324, 2.5805, 1.5407], device='cuda:3'), covar=tensor([0.0378, 0.0198, 0.0686, 0.0179, 0.1621, 0.0165, 0.1566, 0.1830], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0146, 0.0254, 0.0141, 0.0219, 0.0126, 0.0228, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 16:37:08,982 INFO [train2.py:809] (3/4) Epoch 18, batch 200, loss[ctc_loss=0.06474, att_loss=0.2247, loss=0.1927, over 16401.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007722, over 44.00 utterances.], tot_loss[ctc_loss=0.07949, att_loss=0.2381, loss=0.2064, over 2074865.04 frames. utt_duration=1306 frames, utt_pad_proportion=0.04263, over 6363.15 utterances.], batch size: 44, lr: 6.08e-03, grad_scale: 8.0 2023-03-08 16:37:22,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.000e+02 2.339e+02 2.884e+02 4.176e+02, threshold=4.678e+02, percent-clipped=0.0 2023-03-08 16:37:30,812 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6378, 3.4902, 3.5000, 3.0125, 3.5086, 3.5395, 3.5613, 2.5500], device='cuda:3'), covar=tensor([0.1203, 0.2081, 0.2028, 0.4069, 0.1570, 0.3005, 0.1078, 0.4933], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0165, 0.0174, 0.0235, 0.0141, 0.0233, 0.0153, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:37:36,820 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8329, 3.6129, 3.6425, 3.1289, 3.6563, 3.6680, 3.7010, 2.6723], device='cuda:3'), covar=tensor([0.0939, 0.1189, 0.1316, 0.3911, 0.1639, 0.2564, 0.0854, 0.4237], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0164, 0.0174, 0.0235, 0.0141, 0.0233, 0.0153, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:38:27,433 INFO [train2.py:809] (3/4) Epoch 18, batch 250, loss[ctc_loss=0.08095, att_loss=0.2441, loss=0.2115, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005427, over 46.00 utterances.], tot_loss[ctc_loss=0.08011, att_loss=0.2386, loss=0.2069, over 2342846.79 frames. utt_duration=1279 frames, utt_pad_proportion=0.04619, over 7336.99 utterances.], batch size: 46, lr: 6.08e-03, grad_scale: 8.0 2023-03-08 16:38:36,836 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:39:50,974 INFO [train2.py:809] (3/4) Epoch 18, batch 300, loss[ctc_loss=0.0786, att_loss=0.2456, loss=0.2122, over 17031.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01017, over 53.00 utterances.], tot_loss[ctc_loss=0.07982, att_loss=0.2383, loss=0.2066, over 2547789.04 frames. utt_duration=1269 frames, utt_pad_proportion=0.04963, over 8037.76 utterances.], batch size: 53, lr: 6.08e-03, grad_scale: 8.0 2023-03-08 16:40:04,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 2.007e+02 2.345e+02 2.847e+02 6.033e+02, threshold=4.689e+02, percent-clipped=3.0 2023-03-08 16:40:08,020 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:40:17,192 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:40:26,343 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 16:41:09,397 INFO [train2.py:809] (3/4) Epoch 18, batch 350, loss[ctc_loss=0.1308, att_loss=0.2749, loss=0.2461, over 14635.00 frames. utt_duration=399.7 frames, utt_pad_proportion=0.2999, over 147.00 utterances.], tot_loss[ctc_loss=0.0815, att_loss=0.2393, loss=0.2078, over 2700252.91 frames. utt_duration=1236 frames, utt_pad_proportion=0.05982, over 8752.62 utterances.], batch size: 147, lr: 6.08e-03, grad_scale: 8.0 2023-03-08 16:41:13,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 16:41:27,805 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:42:27,981 INFO [train2.py:809] (3/4) Epoch 18, batch 400, loss[ctc_loss=0.06538, att_loss=0.2443, loss=0.2086, over 16623.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005558, over 47.00 utterances.], tot_loss[ctc_loss=0.08056, att_loss=0.2386, loss=0.207, over 2829046.96 frames. utt_duration=1269 frames, utt_pad_proportion=0.05053, over 8927.15 utterances.], batch size: 47, lr: 6.07e-03, grad_scale: 8.0 2023-03-08 16:42:41,447 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.125e+02 2.519e+02 3.295e+02 8.158e+02, threshold=5.039e+02, percent-clipped=7.0 2023-03-08 16:43:03,789 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 16:43:37,557 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:43:46,626 INFO [train2.py:809] (3/4) Epoch 18, batch 450, loss[ctc_loss=0.06302, att_loss=0.2076, loss=0.1787, over 15374.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.0104, over 35.00 utterances.], tot_loss[ctc_loss=0.08048, att_loss=0.2391, loss=0.2074, over 2923113.13 frames. utt_duration=1269 frames, utt_pad_proportion=0.05154, over 9221.17 utterances.], batch size: 35, lr: 6.07e-03, grad_scale: 8.0 2023-03-08 16:44:53,068 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:45:05,408 INFO [train2.py:809] (3/4) Epoch 18, batch 500, loss[ctc_loss=0.06436, att_loss=0.2284, loss=0.1956, over 16121.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006587, over 42.00 utterances.], tot_loss[ctc_loss=0.08056, att_loss=0.2394, loss=0.2076, over 3000491.02 frames. utt_duration=1255 frames, utt_pad_proportion=0.0539, over 9570.94 utterances.], batch size: 42, lr: 6.07e-03, grad_scale: 8.0 2023-03-08 16:45:12,026 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7743, 5.2391, 4.9877, 5.0391, 5.2879, 4.7990, 3.7622, 5.1363], device='cuda:3'), covar=tensor([0.0111, 0.0103, 0.0119, 0.0096, 0.0080, 0.0092, 0.0624, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0085, 0.0106, 0.0066, 0.0070, 0.0082, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:45:19,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.432e+02 1.924e+02 2.355e+02 3.288e+02 6.167e+02, threshold=4.710e+02, percent-clipped=3.0 2023-03-08 16:46:23,446 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1251, 5.4156, 5.7351, 5.6086, 5.6012, 6.0846, 5.2639, 6.1494], device='cuda:3'), covar=tensor([0.0688, 0.0634, 0.0740, 0.1168, 0.1803, 0.0841, 0.0623, 0.0709], device='cuda:3'), in_proj_covar=tensor([0.0824, 0.0478, 0.0566, 0.0629, 0.0826, 0.0580, 0.0463, 0.0574], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:46:24,806 INFO [train2.py:809] (3/4) Epoch 18, batch 550, loss[ctc_loss=0.07471, att_loss=0.2248, loss=0.1948, over 15881.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009425, over 39.00 utterances.], tot_loss[ctc_loss=0.08117, att_loss=0.2395, loss=0.2079, over 3061225.77 frames. utt_duration=1242 frames, utt_pad_proportion=0.05786, over 9868.00 utterances.], batch size: 39, lr: 6.07e-03, grad_scale: 8.0 2023-03-08 16:47:44,062 INFO [train2.py:809] (3/4) Epoch 18, batch 600, loss[ctc_loss=0.1498, att_loss=0.2837, loss=0.2569, over 14188.00 frames. utt_duration=390.3 frames, utt_pad_proportion=0.3212, over 146.00 utterances.], tot_loss[ctc_loss=0.08088, att_loss=0.2386, loss=0.2071, over 3097885.60 frames. utt_duration=1232 frames, utt_pad_proportion=0.06466, over 10071.55 utterances.], batch size: 146, lr: 6.07e-03, grad_scale: 8.0 2023-03-08 16:47:58,122 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.193e+02 2.599e+02 3.203e+02 7.782e+02, threshold=5.199e+02, percent-clipped=2.0 2023-03-08 16:48:01,687 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:48:03,112 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:48:19,825 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:49:03,208 INFO [train2.py:809] (3/4) Epoch 18, batch 650, loss[ctc_loss=0.07531, att_loss=0.2323, loss=0.2009, over 16546.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005702, over 45.00 utterances.], tot_loss[ctc_loss=0.08146, att_loss=0.2385, loss=0.2071, over 3126411.38 frames. utt_duration=1211 frames, utt_pad_proportion=0.07029, over 10343.57 utterances.], batch size: 45, lr: 6.06e-03, grad_scale: 8.0 2023-03-08 16:49:17,266 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:49:35,682 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:49:40,485 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:50:22,686 INFO [train2.py:809] (3/4) Epoch 18, batch 700, loss[ctc_loss=0.08857, att_loss=0.2499, loss=0.2176, over 16944.00 frames. utt_duration=686.1 frames, utt_pad_proportion=0.1392, over 99.00 utterances.], tot_loss[ctc_loss=0.08159, att_loss=0.2389, loss=0.2074, over 3163743.55 frames. utt_duration=1209 frames, utt_pad_proportion=0.06801, over 10480.57 utterances.], batch size: 99, lr: 6.06e-03, grad_scale: 8.0 2023-03-08 16:50:36,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.020e+02 2.578e+02 3.082e+02 5.307e+02, threshold=5.155e+02, percent-clipped=1.0 2023-03-08 16:50:50,327 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 16:51:18,388 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:51:41,632 INFO [train2.py:809] (3/4) Epoch 18, batch 750, loss[ctc_loss=0.06072, att_loss=0.2273, loss=0.194, over 16182.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006182, over 41.00 utterances.], tot_loss[ctc_loss=0.08113, att_loss=0.2392, loss=0.2076, over 3193070.53 frames. utt_duration=1228 frames, utt_pad_proportion=0.05979, over 10417.09 utterances.], batch size: 41, lr: 6.06e-03, grad_scale: 8.0 2023-03-08 16:51:45,646 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-08 16:53:00,275 INFO [train2.py:809] (3/4) Epoch 18, batch 800, loss[ctc_loss=0.06511, att_loss=0.2346, loss=0.2007, over 16329.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006323, over 45.00 utterances.], tot_loss[ctc_loss=0.08085, att_loss=0.2384, loss=0.2069, over 3209346.41 frames. utt_duration=1239 frames, utt_pad_proportion=0.05829, over 10373.01 utterances.], batch size: 45, lr: 6.06e-03, grad_scale: 8.0 2023-03-08 16:53:13,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.218e+02 2.657e+02 2.960e+02 6.407e+02, threshold=5.314e+02, percent-clipped=4.0 2023-03-08 16:54:08,983 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2272, 4.6130, 4.4243, 4.5922, 4.7053, 4.2921, 3.2703, 4.5359], device='cuda:3'), covar=tensor([0.0128, 0.0110, 0.0148, 0.0094, 0.0084, 0.0132, 0.0690, 0.0212], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0085, 0.0105, 0.0067, 0.0071, 0.0083, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 16:54:19,206 INFO [train2.py:809] (3/4) Epoch 18, batch 850, loss[ctc_loss=0.1122, att_loss=0.2686, loss=0.2373, over 16952.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.0075, over 50.00 utterances.], tot_loss[ctc_loss=0.08106, att_loss=0.239, loss=0.2074, over 3232075.60 frames. utt_duration=1239 frames, utt_pad_proportion=0.05535, over 10450.50 utterances.], batch size: 50, lr: 6.05e-03, grad_scale: 8.0 2023-03-08 16:54:24,182 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:54:47,253 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4612, 2.4626, 4.9354, 3.7043, 3.0166, 4.2109, 4.5295, 4.6130], device='cuda:3'), covar=tensor([0.0229, 0.1793, 0.0136, 0.1032, 0.1776, 0.0255, 0.0155, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0240, 0.0169, 0.0306, 0.0263, 0.0201, 0.0153, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:54:59,474 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6203, 3.1606, 3.2010, 2.7144, 3.3215, 3.3539, 3.2757, 2.2835], device='cuda:3'), covar=tensor([0.1172, 0.2413, 0.2604, 0.5954, 0.2749, 0.2491, 0.1331, 0.5897], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0167, 0.0176, 0.0242, 0.0144, 0.0239, 0.0156, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 16:55:23,429 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1739, 5.1279, 5.0531, 2.4611, 1.8587, 2.5766, 2.3668, 3.8591], device='cuda:3'), covar=tensor([0.0640, 0.0302, 0.0215, 0.4101, 0.6189, 0.3004, 0.3221, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0257, 0.0255, 0.0234, 0.0343, 0.0333, 0.0242, 0.0358], device='cuda:3'), out_proj_covar=tensor([1.4794e-04, 9.4828e-05, 1.0931e-04, 1.0125e-04, 1.4421e-04, 1.3086e-04, 9.6613e-05, 1.4669e-04], device='cuda:3') 2023-03-08 16:55:38,617 INFO [train2.py:809] (3/4) Epoch 18, batch 900, loss[ctc_loss=0.06825, att_loss=0.2399, loss=0.2056, over 17357.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03357, over 63.00 utterances.], tot_loss[ctc_loss=0.08132, att_loss=0.2393, loss=0.2077, over 3241986.61 frames. utt_duration=1241 frames, utt_pad_proportion=0.05535, over 10461.82 utterances.], batch size: 63, lr: 6.05e-03, grad_scale: 8.0 2023-03-08 16:55:52,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 1.982e+02 2.432e+02 3.091e+02 4.816e+02, threshold=4.864e+02, percent-clipped=0.0 2023-03-08 16:55:57,203 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:56:00,289 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:56:06,355 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2377, 4.7379, 4.7287, 4.8043, 3.3216, 4.5425, 3.0225, 2.2566], device='cuda:3'), covar=tensor([0.0341, 0.0220, 0.0593, 0.0179, 0.1240, 0.0192, 0.1288, 0.1506], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0145, 0.0251, 0.0141, 0.0218, 0.0127, 0.0227, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 16:56:57,070 INFO [train2.py:809] (3/4) Epoch 18, batch 950, loss[ctc_loss=0.07272, att_loss=0.2306, loss=0.199, over 16397.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007137, over 44.00 utterances.], tot_loss[ctc_loss=0.08033, att_loss=0.2388, loss=0.2071, over 3254317.19 frames. utt_duration=1258 frames, utt_pad_proportion=0.04992, over 10362.24 utterances.], batch size: 44, lr: 6.05e-03, grad_scale: 8.0 2023-03-08 16:57:12,201 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:57:12,449 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:58:15,811 INFO [train2.py:809] (3/4) Epoch 18, batch 1000, loss[ctc_loss=0.08765, att_loss=0.2377, loss=0.2077, over 16555.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005402, over 45.00 utterances.], tot_loss[ctc_loss=0.08017, att_loss=0.2384, loss=0.2067, over 3252609.67 frames. utt_duration=1248 frames, utt_pad_proportion=0.05509, over 10438.30 utterances.], batch size: 45, lr: 6.05e-03, grad_scale: 8.0 2023-03-08 16:58:29,418 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.078e+02 2.479e+02 2.964e+02 7.208e+02, threshold=4.957e+02, percent-clipped=4.0 2023-03-08 16:58:42,898 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-08 16:58:43,622 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 16:58:48,381 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:59:02,251 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 16:59:02,417 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1720, 3.8973, 3.2472, 3.5974, 4.0269, 3.7557, 3.2807, 4.3492], device='cuda:3'), covar=tensor([0.0956, 0.0479, 0.1123, 0.0691, 0.0705, 0.0665, 0.0748, 0.0485], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0206, 0.0218, 0.0191, 0.0259, 0.0229, 0.0195, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 16:59:18,394 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0704, 5.1089, 4.8514, 2.7545, 4.9296, 4.6863, 4.2497, 2.6704], device='cuda:3'), covar=tensor([0.0099, 0.0090, 0.0244, 0.1052, 0.0086, 0.0202, 0.0331, 0.1335], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0096, 0.0095, 0.0109, 0.0080, 0.0106, 0.0096, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 16:59:33,513 INFO [train2.py:809] (3/4) Epoch 18, batch 1050, loss[ctc_loss=0.06901, att_loss=0.224, loss=0.193, over 15888.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009035, over 39.00 utterances.], tot_loss[ctc_loss=0.07995, att_loss=0.2386, loss=0.2069, over 3260689.93 frames. utt_duration=1259 frames, utt_pad_proportion=0.05168, over 10371.79 utterances.], batch size: 39, lr: 6.05e-03, grad_scale: 16.0 2023-03-08 16:59:58,355 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:00:21,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2023-03-08 17:00:53,126 INFO [train2.py:809] (3/4) Epoch 18, batch 1100, loss[ctc_loss=0.08916, att_loss=0.2493, loss=0.2172, over 16608.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006305, over 47.00 utterances.], tot_loss[ctc_loss=0.0798, att_loss=0.2384, loss=0.2067, over 3261134.30 frames. utt_duration=1273 frames, utt_pad_proportion=0.04855, over 10262.98 utterances.], batch size: 47, lr: 6.04e-03, grad_scale: 16.0 2023-03-08 17:00:53,476 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6260, 3.2157, 3.6522, 3.0988, 3.6851, 4.7205, 4.4457, 3.3244], device='cuda:3'), covar=tensor([0.0368, 0.1611, 0.1307, 0.1311, 0.1053, 0.0707, 0.0615, 0.1359], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0239, 0.0267, 0.0210, 0.0256, 0.0348, 0.0247, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:01:06,987 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.026e+02 2.542e+02 3.113e+02 8.116e+02, threshold=5.085e+02, percent-clipped=3.0 2023-03-08 17:01:58,590 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2769, 2.9477, 3.1900, 4.4010, 3.9483, 3.8719, 2.9075, 2.1784], device='cuda:3'), covar=tensor([0.0705, 0.1847, 0.0978, 0.0498, 0.0718, 0.0446, 0.1413, 0.2131], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0214, 0.0189, 0.0208, 0.0212, 0.0170, 0.0198, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:02:01,879 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9871, 4.0275, 3.9469, 4.2973, 2.8841, 3.9616, 2.6489, 1.8829], device='cuda:3'), covar=tensor([0.0383, 0.0229, 0.0729, 0.0200, 0.1402, 0.0242, 0.1414, 0.1700], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0145, 0.0249, 0.0140, 0.0215, 0.0127, 0.0224, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:02:12,252 INFO [train2.py:809] (3/4) Epoch 18, batch 1150, loss[ctc_loss=0.0751, att_loss=0.2172, loss=0.1888, over 15633.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008396, over 37.00 utterances.], tot_loss[ctc_loss=0.07967, att_loss=0.2384, loss=0.2067, over 3265535.00 frames. utt_duration=1277 frames, utt_pad_proportion=0.04688, over 10243.62 utterances.], batch size: 37, lr: 6.04e-03, grad_scale: 16.0 2023-03-08 17:02:53,981 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4974, 4.9601, 4.7208, 4.8633, 4.9885, 4.5817, 3.5906, 4.8611], device='cuda:3'), covar=tensor([0.0116, 0.0114, 0.0136, 0.0107, 0.0091, 0.0122, 0.0621, 0.0227], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0084, 0.0105, 0.0066, 0.0071, 0.0082, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:03:31,043 INFO [train2.py:809] (3/4) Epoch 18, batch 1200, loss[ctc_loss=0.07256, att_loss=0.2398, loss=0.2063, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006107, over 45.00 utterances.], tot_loss[ctc_loss=0.07961, att_loss=0.2388, loss=0.207, over 3278195.87 frames. utt_duration=1284 frames, utt_pad_proportion=0.04222, over 10226.44 utterances.], batch size: 45, lr: 6.04e-03, grad_scale: 16.0 2023-03-08 17:03:44,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.483e+02 2.054e+02 2.571e+02 3.261e+02 6.642e+02, threshold=5.143e+02, percent-clipped=4.0 2023-03-08 17:03:45,176 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:04:25,363 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 17:04:50,135 INFO [train2.py:809] (3/4) Epoch 18, batch 1250, loss[ctc_loss=0.09659, att_loss=0.2553, loss=0.2236, over 16480.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006519, over 46.00 utterances.], tot_loss[ctc_loss=0.07948, att_loss=0.2382, loss=0.2064, over 3265558.34 frames. utt_duration=1290 frames, utt_pad_proportion=0.04295, over 10136.91 utterances.], batch size: 46, lr: 6.04e-03, grad_scale: 16.0 2023-03-08 17:05:10,275 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5598, 4.9102, 4.7410, 4.9054, 4.9703, 4.5936, 3.3991, 4.8561], device='cuda:3'), covar=tensor([0.0114, 0.0120, 0.0128, 0.0097, 0.0116, 0.0123, 0.0718, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0084, 0.0104, 0.0066, 0.0070, 0.0082, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:06:08,502 INFO [train2.py:809] (3/4) Epoch 18, batch 1300, loss[ctc_loss=0.0662, att_loss=0.2332, loss=0.1998, over 16322.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006597, over 45.00 utterances.], tot_loss[ctc_loss=0.07945, att_loss=0.2384, loss=0.2066, over 3272660.28 frames. utt_duration=1297 frames, utt_pad_proportion=0.03957, over 10105.43 utterances.], batch size: 45, lr: 6.04e-03, grad_scale: 16.0 2023-03-08 17:06:22,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.084e+02 2.462e+02 2.947e+02 6.955e+02, threshold=4.923e+02, percent-clipped=4.0 2023-03-08 17:06:30,563 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1289, 5.1484, 4.9066, 2.5367, 2.0243, 3.0239, 2.6414, 3.9607], device='cuda:3'), covar=tensor([0.0654, 0.0282, 0.0300, 0.4190, 0.5526, 0.2350, 0.3152, 0.1527], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0254, 0.0253, 0.0234, 0.0337, 0.0329, 0.0241, 0.0355], device='cuda:3'), out_proj_covar=tensor([1.4622e-04, 9.3953e-05, 1.0855e-04, 1.0120e-04, 1.4207e-04, 1.2927e-04, 9.6090e-05, 1.4551e-04], device='cuda:3') 2023-03-08 17:06:33,207 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:06:36,473 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:06:55,494 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:07:27,320 INFO [train2.py:809] (3/4) Epoch 18, batch 1350, loss[ctc_loss=0.0896, att_loss=0.218, loss=0.1923, over 15877.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009705, over 39.00 utterances.], tot_loss[ctc_loss=0.08014, att_loss=0.2384, loss=0.2068, over 3271696.61 frames. utt_duration=1287 frames, utt_pad_proportion=0.04126, over 10177.07 utterances.], batch size: 39, lr: 6.03e-03, grad_scale: 16.0 2023-03-08 17:07:51,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-08 17:08:10,440 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:08:12,189 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:08:46,057 INFO [train2.py:809] (3/4) Epoch 18, batch 1400, loss[ctc_loss=0.07128, att_loss=0.2302, loss=0.1984, over 16630.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.00514, over 47.00 utterances.], tot_loss[ctc_loss=0.07984, att_loss=0.2381, loss=0.2065, over 3260018.75 frames. utt_duration=1281 frames, utt_pad_proportion=0.04568, over 10189.49 utterances.], batch size: 47, lr: 6.03e-03, grad_scale: 16.0 2023-03-08 17:08:59,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.123e+02 2.542e+02 3.068e+02 7.934e+02, threshold=5.085e+02, percent-clipped=3.0 2023-03-08 17:09:44,849 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:09:54,762 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:10:05,181 INFO [train2.py:809] (3/4) Epoch 18, batch 1450, loss[ctc_loss=0.1559, att_loss=0.282, loss=0.2568, over 14208.00 frames. utt_duration=388.3 frames, utt_pad_proportion=0.3176, over 147.00 utterances.], tot_loss[ctc_loss=0.08027, att_loss=0.2388, loss=0.2071, over 3267052.05 frames. utt_duration=1266 frames, utt_pad_proportion=0.04872, over 10333.89 utterances.], batch size: 147, lr: 6.03e-03, grad_scale: 16.0 2023-03-08 17:10:50,669 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 17:11:16,445 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1046, 5.0993, 4.9188, 2.4762, 2.0527, 2.8608, 2.8209, 3.7695], device='cuda:3'), covar=tensor([0.0619, 0.0260, 0.0244, 0.4392, 0.5323, 0.2450, 0.2922, 0.1750], device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0251, 0.0250, 0.0231, 0.0333, 0.0324, 0.0239, 0.0352], device='cuda:3'), out_proj_covar=tensor([1.4423e-04, 9.2703e-05, 1.0731e-04, 9.9674e-05, 1.4017e-04, 1.2749e-04, 9.5206e-05, 1.4432e-04], device='cuda:3') 2023-03-08 17:11:20,879 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:11:23,582 INFO [train2.py:809] (3/4) Epoch 18, batch 1500, loss[ctc_loss=0.098, att_loss=0.2592, loss=0.227, over 17100.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01551, over 56.00 utterances.], tot_loss[ctc_loss=0.08088, att_loss=0.2402, loss=0.2084, over 3277707.19 frames. utt_duration=1243 frames, utt_pad_proportion=0.05305, over 10564.00 utterances.], batch size: 56, lr: 6.03e-03, grad_scale: 16.0 2023-03-08 17:11:23,983 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:11:29,981 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:11:37,071 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.061e+02 2.532e+02 3.025e+02 5.458e+02, threshold=5.063e+02, percent-clipped=1.0 2023-03-08 17:11:37,467 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:11:54,407 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 17:12:13,379 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3592, 4.5018, 4.7057, 4.6606, 2.8438, 4.5114, 3.0229, 2.1243], device='cuda:3'), covar=tensor([0.0307, 0.0289, 0.0579, 0.0220, 0.1592, 0.0206, 0.1350, 0.1563], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0147, 0.0254, 0.0142, 0.0219, 0.0128, 0.0228, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:12:39,675 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:12:42,421 INFO [train2.py:809] (3/4) Epoch 18, batch 1550, loss[ctc_loss=0.07628, att_loss=0.2398, loss=0.2071, over 16534.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006602, over 45.00 utterances.], tot_loss[ctc_loss=0.07984, att_loss=0.2396, loss=0.2076, over 3282121.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.04768, over 10426.14 utterances.], batch size: 45, lr: 6.02e-03, grad_scale: 16.0 2023-03-08 17:12:51,658 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9723, 3.7569, 3.1988, 3.5220, 3.8877, 3.5572, 2.9445, 4.2004], device='cuda:3'), covar=tensor([0.0991, 0.0508, 0.1007, 0.0619, 0.0642, 0.0708, 0.0869, 0.0473], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0208, 0.0221, 0.0193, 0.0262, 0.0233, 0.0197, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:12:52,880 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:12:59,178 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:13:31,398 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 17:13:53,292 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:14:00,357 INFO [train2.py:809] (3/4) Epoch 18, batch 1600, loss[ctc_loss=0.06092, att_loss=0.2193, loss=0.1876, over 15756.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009484, over 38.00 utterances.], tot_loss[ctc_loss=0.07979, att_loss=0.2394, loss=0.2075, over 3281879.67 frames. utt_duration=1276 frames, utt_pad_proportion=0.04336, over 10301.76 utterances.], batch size: 38, lr: 6.02e-03, grad_scale: 16.0 2023-03-08 17:14:14,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 1.944e+02 2.286e+02 2.991e+02 6.797e+02, threshold=4.572e+02, percent-clipped=3.0 2023-03-08 17:14:14,900 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:14:25,754 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:14:35,242 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1517, 5.1551, 4.9327, 2.5523, 1.9019, 2.8598, 2.9173, 3.8472], device='cuda:3'), covar=tensor([0.0673, 0.0309, 0.0269, 0.4668, 0.6048, 0.2538, 0.2889, 0.1737], device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0251, 0.0251, 0.0231, 0.0336, 0.0326, 0.0240, 0.0354], device='cuda:3'), out_proj_covar=tensor([1.4492e-04, 9.2969e-05, 1.0767e-04, 9.9693e-05, 1.4119e-04, 1.2804e-04, 9.5750e-05, 1.4510e-04], device='cuda:3') 2023-03-08 17:15:01,552 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4772, 2.5518, 2.5037, 2.2493, 2.5154, 2.4755, 2.5841, 1.8884], device='cuda:3'), covar=tensor([0.1217, 0.1811, 0.2404, 0.4073, 0.1269, 0.2710, 0.1275, 0.4936], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0166, 0.0176, 0.0237, 0.0142, 0.0239, 0.0156, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 17:15:10,865 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:15:21,709 INFO [train2.py:809] (3/4) Epoch 18, batch 1650, loss[ctc_loss=0.1314, att_loss=0.2691, loss=0.2416, over 13944.00 frames. utt_duration=381 frames, utt_pad_proportion=0.3316, over 147.00 utterances.], tot_loss[ctc_loss=0.08127, att_loss=0.2406, loss=0.2087, over 3273278.92 frames. utt_duration=1216 frames, utt_pad_proportion=0.06282, over 10783.64 utterances.], batch size: 147, lr: 6.02e-03, grad_scale: 16.0 2023-03-08 17:15:31,705 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:15:44,523 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:15:52,872 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:16:02,225 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:16:04,273 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4455, 3.0108, 2.6338, 2.8740, 3.1251, 3.0062, 2.4348, 3.0201], device='cuda:3'), covar=tensor([0.0893, 0.0419, 0.0856, 0.0587, 0.0656, 0.0615, 0.0876, 0.0478], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0207, 0.0219, 0.0192, 0.0261, 0.0231, 0.0196, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:16:44,778 INFO [train2.py:809] (3/4) Epoch 18, batch 1700, loss[ctc_loss=0.08553, att_loss=0.2412, loss=0.2101, over 16423.00 frames. utt_duration=1495 frames, utt_pad_proportion=0.006301, over 44.00 utterances.], tot_loss[ctc_loss=0.08162, att_loss=0.2403, loss=0.2086, over 3267012.06 frames. utt_duration=1212 frames, utt_pad_proportion=0.06516, over 10794.88 utterances.], batch size: 44, lr: 6.02e-03, grad_scale: 16.0 2023-03-08 17:16:51,590 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:16:58,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.149e+02 2.524e+02 3.202e+02 9.381e+02, threshold=5.048e+02, percent-clipped=6.0 2023-03-08 17:17:34,272 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:18:03,413 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4643, 2.8756, 3.5530, 2.9727, 3.4150, 4.5948, 4.4099, 3.3550], device='cuda:3'), covar=tensor([0.0340, 0.1768, 0.1245, 0.1350, 0.1120, 0.0684, 0.0510, 0.1268], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0238, 0.0269, 0.0212, 0.0259, 0.0349, 0.0249, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:18:06,233 INFO [train2.py:809] (3/4) Epoch 18, batch 1750, loss[ctc_loss=0.1062, att_loss=0.255, loss=0.2252, over 16546.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006032, over 45.00 utterances.], tot_loss[ctc_loss=0.08094, att_loss=0.2396, loss=0.2079, over 3269479.53 frames. utt_duration=1244 frames, utt_pad_proportion=0.05653, over 10528.10 utterances.], batch size: 45, lr: 6.02e-03, grad_scale: 16.0 2023-03-08 17:18:45,742 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1994, 2.7421, 3.0244, 4.3054, 3.8350, 3.7418, 2.8264, 2.0784], device='cuda:3'), covar=tensor([0.0762, 0.2159, 0.1099, 0.0535, 0.0805, 0.0534, 0.1557, 0.2438], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0216, 0.0190, 0.0210, 0.0213, 0.0173, 0.0201, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:19:17,109 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:19:26,214 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:19:27,559 INFO [train2.py:809] (3/4) Epoch 18, batch 1800, loss[ctc_loss=0.06662, att_loss=0.2224, loss=0.1912, over 15948.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.006704, over 41.00 utterances.], tot_loss[ctc_loss=0.08113, att_loss=0.2395, loss=0.2078, over 3268419.29 frames. utt_duration=1242 frames, utt_pad_proportion=0.057, over 10538.30 utterances.], batch size: 41, lr: 6.01e-03, grad_scale: 16.0 2023-03-08 17:19:41,873 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.055e+02 2.270e+02 2.783e+02 4.907e+02, threshold=4.541e+02, percent-clipped=0.0 2023-03-08 17:19:46,952 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2400, 2.6554, 3.0848, 4.2864, 3.7944, 3.7459, 2.8627, 2.1179], device='cuda:3'), covar=tensor([0.0764, 0.2200, 0.0998, 0.0467, 0.0861, 0.0490, 0.1446, 0.2347], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0213, 0.0188, 0.0208, 0.0211, 0.0171, 0.0198, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:20:03,262 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3559, 2.8283, 3.5866, 2.9240, 3.4339, 4.4702, 4.2866, 3.3204], device='cuda:3'), covar=tensor([0.0416, 0.1786, 0.1233, 0.1361, 0.1177, 0.1028, 0.0706, 0.1239], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0240, 0.0269, 0.0213, 0.0260, 0.0350, 0.0251, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:20:31,966 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7289, 4.9615, 4.9188, 4.8915, 5.0283, 5.0030, 4.7073, 4.4768], device='cuda:3'), covar=tensor([0.0961, 0.0562, 0.0328, 0.0503, 0.0273, 0.0305, 0.0333, 0.0343], device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0334, 0.0315, 0.0332, 0.0393, 0.0403, 0.0332, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:20:49,348 INFO [train2.py:809] (3/4) Epoch 18, batch 1850, loss[ctc_loss=0.05728, att_loss=0.2116, loss=0.1807, over 15504.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008594, over 36.00 utterances.], tot_loss[ctc_loss=0.08059, att_loss=0.2396, loss=0.2078, over 3272496.04 frames. utt_duration=1215 frames, utt_pad_proportion=0.06215, over 10788.96 utterances.], batch size: 36, lr: 6.01e-03, grad_scale: 16.0 2023-03-08 17:20:58,918 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:21:31,783 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 17:21:39,823 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4527, 2.9576, 3.6817, 3.0727, 3.4827, 4.6122, 4.4250, 3.4120], device='cuda:3'), covar=tensor([0.0395, 0.1715, 0.1098, 0.1252, 0.1076, 0.0820, 0.0557, 0.1182], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0239, 0.0268, 0.0212, 0.0258, 0.0349, 0.0250, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:22:09,409 INFO [train2.py:809] (3/4) Epoch 18, batch 1900, loss[ctc_loss=0.09165, att_loss=0.2325, loss=0.2043, over 15633.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.008539, over 37.00 utterances.], tot_loss[ctc_loss=0.08216, att_loss=0.2412, loss=0.2094, over 3282236.11 frames. utt_duration=1191 frames, utt_pad_proportion=0.06549, over 11032.78 utterances.], batch size: 37, lr: 6.01e-03, grad_scale: 16.0 2023-03-08 17:22:12,202 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-08 17:22:15,690 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 17:22:23,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 2.048e+02 2.360e+02 2.878e+02 5.569e+02, threshold=4.720e+02, percent-clipped=3.0 2023-03-08 17:23:02,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-03-08 17:23:28,938 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 17:23:30,119 INFO [train2.py:809] (3/4) Epoch 18, batch 1950, loss[ctc_loss=0.1416, att_loss=0.2772, loss=0.2501, over 14463.00 frames. utt_duration=397.8 frames, utt_pad_proportion=0.307, over 146.00 utterances.], tot_loss[ctc_loss=0.08182, att_loss=0.2404, loss=0.2087, over 3272156.27 frames. utt_duration=1202 frames, utt_pad_proportion=0.06452, over 10898.91 utterances.], batch size: 146, lr: 6.01e-03, grad_scale: 16.0 2023-03-08 17:23:31,819 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:23:52,233 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8362, 4.8203, 4.7648, 4.5597, 5.3676, 4.6349, 4.8003, 2.4746], device='cuda:3'), covar=tensor([0.0217, 0.0310, 0.0297, 0.0437, 0.0954, 0.0229, 0.0272, 0.1898], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0165, 0.0167, 0.0184, 0.0357, 0.0141, 0.0155, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:24:09,261 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:24:20,251 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8384, 4.9012, 4.6284, 2.5687, 4.6085, 4.5040, 4.1330, 2.6130], device='cuda:3'), covar=tensor([0.0154, 0.0105, 0.0307, 0.1183, 0.0115, 0.0224, 0.0341, 0.1434], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0096, 0.0096, 0.0107, 0.0080, 0.0105, 0.0096, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 17:24:24,675 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-08 17:24:38,034 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6233, 2.2174, 2.3526, 2.6902, 2.8434, 2.6735, 2.3215, 3.0392], device='cuda:3'), covar=tensor([0.1401, 0.3034, 0.2379, 0.1421, 0.1361, 0.0952, 0.2175, 0.0867], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0113, 0.0112, 0.0098, 0.0108, 0.0096, 0.0116, 0.0084], device='cuda:3'), out_proj_covar=tensor([7.6810e-05, 8.5876e-05, 8.7089e-05, 7.4998e-05, 7.9068e-05, 7.5873e-05, 8.5752e-05, 6.7919e-05], device='cuda:3') 2023-03-08 17:24:48,848 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:24:50,207 INFO [train2.py:809] (3/4) Epoch 18, batch 2000, loss[ctc_loss=0.07639, att_loss=0.2272, loss=0.1971, over 16001.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007605, over 40.00 utterances.], tot_loss[ctc_loss=0.08116, att_loss=0.2391, loss=0.2075, over 3267298.76 frames. utt_duration=1232 frames, utt_pad_proportion=0.05765, over 10618.73 utterances.], batch size: 40, lr: 6.00e-03, grad_scale: 8.0 2023-03-08 17:25:01,358 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1268, 5.4432, 5.0028, 5.5195, 4.9036, 5.1218, 5.5891, 5.3388], device='cuda:3'), covar=tensor([0.0498, 0.0250, 0.0715, 0.0281, 0.0392, 0.0192, 0.0201, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0308, 0.0353, 0.0324, 0.0309, 0.0234, 0.0291, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 17:25:05,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.144e+02 2.656e+02 3.349e+02 1.249e+03, threshold=5.313e+02, percent-clipped=13.0 2023-03-08 17:25:06,181 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 17:25:26,866 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:25:31,592 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:26:10,680 INFO [train2.py:809] (3/4) Epoch 18, batch 2050, loss[ctc_loss=0.05768, att_loss=0.2162, loss=0.1845, over 15775.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008228, over 38.00 utterances.], tot_loss[ctc_loss=0.08117, att_loss=0.2389, loss=0.2074, over 3258803.70 frames. utt_duration=1236 frames, utt_pad_proportion=0.05883, over 10560.42 utterances.], batch size: 38, lr: 6.00e-03, grad_scale: 8.0 2023-03-08 17:26:17,024 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7666, 4.7976, 4.7191, 4.4979, 5.3501, 4.5403, 4.7084, 2.6173], device='cuda:3'), covar=tensor([0.0234, 0.0276, 0.0290, 0.0422, 0.0800, 0.0245, 0.0312, 0.1760], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0166, 0.0169, 0.0186, 0.0358, 0.0142, 0.0157, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:26:56,240 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4560, 4.7844, 4.6441, 4.6623, 4.8149, 4.5883, 3.5164, 4.7646], device='cuda:3'), covar=tensor([0.0121, 0.0116, 0.0134, 0.0101, 0.0099, 0.0103, 0.0656, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0084, 0.0105, 0.0066, 0.0071, 0.0083, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:27:07,545 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4536, 2.4598, 4.7420, 3.7184, 2.8435, 4.1443, 4.4340, 4.4508], device='cuda:3'), covar=tensor([0.0208, 0.1822, 0.0154, 0.0986, 0.1879, 0.0282, 0.0160, 0.0248], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0242, 0.0171, 0.0311, 0.0268, 0.0204, 0.0155, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 17:27:10,819 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-03-08 17:27:19,513 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:27:28,574 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:27:29,782 INFO [train2.py:809] (3/4) Epoch 18, batch 2100, loss[ctc_loss=0.0808, att_loss=0.2432, loss=0.2107, over 17031.00 frames. utt_duration=689.5 frames, utt_pad_proportion=0.1305, over 99.00 utterances.], tot_loss[ctc_loss=0.08008, att_loss=0.238, loss=0.2064, over 3259344.00 frames. utt_duration=1270 frames, utt_pad_proportion=0.05106, over 10275.67 utterances.], batch size: 99, lr: 6.00e-03, grad_scale: 8.0 2023-03-08 17:27:45,296 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 1.987e+02 2.392e+02 3.047e+02 6.775e+02, threshold=4.784e+02, percent-clipped=2.0 2023-03-08 17:27:54,767 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3972, 2.8294, 3.3958, 4.5320, 3.9851, 3.9189, 2.9138, 2.0627], device='cuda:3'), covar=tensor([0.0731, 0.2274, 0.0953, 0.0487, 0.0803, 0.0487, 0.1629, 0.2524], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0211, 0.0185, 0.0207, 0.0211, 0.0169, 0.0198, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:28:36,238 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:28:45,330 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:28:49,865 INFO [train2.py:809] (3/4) Epoch 18, batch 2150, loss[ctc_loss=0.06566, att_loss=0.2255, loss=0.1936, over 15939.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.008021, over 41.00 utterances.], tot_loss[ctc_loss=0.07969, att_loss=0.2382, loss=0.2065, over 3261382.14 frames. utt_duration=1277 frames, utt_pad_proportion=0.04875, over 10229.34 utterances.], batch size: 41, lr: 6.00e-03, grad_scale: 8.0 2023-03-08 17:28:59,314 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:29:31,237 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 17:29:39,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0733, 3.8524, 3.1996, 3.4741, 3.9523, 3.5716, 3.1282, 4.2749], device='cuda:3'), covar=tensor([0.1067, 0.0568, 0.1176, 0.0709, 0.0783, 0.0799, 0.0899, 0.0527], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0211, 0.0223, 0.0193, 0.0266, 0.0233, 0.0199, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:29:51,983 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2438, 5.5615, 5.8357, 5.6394, 5.7218, 6.1801, 5.2937, 6.3329], device='cuda:3'), covar=tensor([0.0637, 0.0666, 0.0772, 0.1309, 0.1825, 0.0843, 0.0647, 0.0503], device='cuda:3'), in_proj_covar=tensor([0.0835, 0.0486, 0.0572, 0.0637, 0.0842, 0.0589, 0.0475, 0.0578], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:30:08,951 INFO [train2.py:809] (3/4) Epoch 18, batch 2200, loss[ctc_loss=0.07721, att_loss=0.2503, loss=0.2157, over 16963.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007772, over 50.00 utterances.], tot_loss[ctc_loss=0.08048, att_loss=0.239, loss=0.2073, over 3269450.03 frames. utt_duration=1271 frames, utt_pad_proportion=0.0482, over 10305.07 utterances.], batch size: 50, lr: 6.00e-03, grad_scale: 8.0 2023-03-08 17:30:15,156 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:30:15,376 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 17:30:24,845 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.283e+02 2.018e+02 2.397e+02 3.065e+02 5.583e+02, threshold=4.795e+02, percent-clipped=3.0 2023-03-08 17:30:47,805 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:31:27,199 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4664, 4.8017, 4.7610, 4.8707, 4.9254, 4.6345, 3.3253, 4.8323], device='cuda:3'), covar=tensor([0.0171, 0.0178, 0.0196, 0.0139, 0.0162, 0.0170, 0.0933, 0.0455], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0084, 0.0104, 0.0066, 0.0071, 0.0083, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:31:28,416 INFO [train2.py:809] (3/4) Epoch 18, batch 2250, loss[ctc_loss=0.08867, att_loss=0.2304, loss=0.202, over 15987.00 frames. utt_duration=1600 frames, utt_pad_proportion=0.007967, over 40.00 utterances.], tot_loss[ctc_loss=0.08057, att_loss=0.2398, loss=0.2079, over 3276203.93 frames. utt_duration=1266 frames, utt_pad_proportion=0.04894, over 10362.78 utterances.], batch size: 40, lr: 5.99e-03, grad_scale: 8.0 2023-03-08 17:31:30,252 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:31:31,512 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:32:51,056 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:32:51,256 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:32:52,443 INFO [train2.py:809] (3/4) Epoch 18, batch 2300, loss[ctc_loss=0.06883, att_loss=0.2238, loss=0.1928, over 15622.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.01048, over 37.00 utterances.], tot_loss[ctc_loss=0.07987, att_loss=0.2386, loss=0.2069, over 3274158.57 frames. utt_duration=1258 frames, utt_pad_proportion=0.05126, over 10424.42 utterances.], batch size: 37, lr: 5.99e-03, grad_scale: 8.0 2023-03-08 17:33:01,038 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:33:09,216 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.121e+02 2.733e+02 3.440e+02 9.117e+02, threshold=5.466e+02, percent-clipped=8.0 2023-03-08 17:33:34,449 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:34:07,957 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:34:12,452 INFO [train2.py:809] (3/4) Epoch 18, batch 2350, loss[ctc_loss=0.08935, att_loss=0.2617, loss=0.2272, over 17278.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01329, over 55.00 utterances.], tot_loss[ctc_loss=0.07974, att_loss=0.239, loss=0.2071, over 3280891.78 frames. utt_duration=1264 frames, utt_pad_proportion=0.04842, over 10392.80 utterances.], batch size: 55, lr: 5.99e-03, grad_scale: 8.0 2023-03-08 17:34:50,257 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:35:08,702 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:35:31,917 INFO [train2.py:809] (3/4) Epoch 18, batch 2400, loss[ctc_loss=0.07903, att_loss=0.2415, loss=0.209, over 16467.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006734, over 46.00 utterances.], tot_loss[ctc_loss=0.07974, att_loss=0.239, loss=0.2071, over 3283806.53 frames. utt_duration=1269 frames, utt_pad_proportion=0.04673, over 10363.68 utterances.], batch size: 46, lr: 5.99e-03, grad_scale: 8.0 2023-03-08 17:35:42,536 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9957, 3.6075, 3.6854, 3.1946, 3.8258, 3.7802, 3.7351, 2.7248], device='cuda:3'), covar=tensor([0.0972, 0.2034, 0.3908, 0.3607, 0.0954, 0.3230, 0.1016, 0.4775], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0168, 0.0178, 0.0237, 0.0143, 0.0239, 0.0158, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 17:35:48,286 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.485e+02 2.097e+02 2.428e+02 2.770e+02 6.273e+02, threshold=4.856e+02, percent-clipped=1.0 2023-03-08 17:36:20,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1907, 3.8431, 3.2521, 3.5010, 3.9988, 3.6488, 3.1192, 4.2371], device='cuda:3'), covar=tensor([0.0889, 0.0522, 0.1052, 0.0652, 0.0712, 0.0699, 0.0828, 0.0521], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0211, 0.0222, 0.0193, 0.0266, 0.0232, 0.0198, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:36:28,207 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1609, 2.8003, 3.0521, 4.3754, 3.9801, 3.9077, 2.7450, 2.0734], device='cuda:3'), covar=tensor([0.0757, 0.1953, 0.0992, 0.0420, 0.0655, 0.0424, 0.1582, 0.2323], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0215, 0.0188, 0.0210, 0.0215, 0.0172, 0.0202, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:36:45,656 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:36:52,022 INFO [train2.py:809] (3/4) Epoch 18, batch 2450, loss[ctc_loss=0.1049, att_loss=0.2667, loss=0.2343, over 17133.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01432, over 56.00 utterances.], tot_loss[ctc_loss=0.07955, att_loss=0.2391, loss=0.2072, over 3287983.14 frames. utt_duration=1279 frames, utt_pad_proportion=0.04244, over 10295.97 utterances.], batch size: 56, lr: 5.99e-03, grad_scale: 8.0 2023-03-08 17:37:48,688 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9677, 5.2317, 4.8177, 5.2669, 4.7193, 4.8963, 5.3547, 5.1540], device='cuda:3'), covar=tensor([0.0496, 0.0304, 0.0710, 0.0288, 0.0364, 0.0265, 0.0232, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0304, 0.0352, 0.0321, 0.0307, 0.0232, 0.0288, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 17:38:11,071 INFO [train2.py:809] (3/4) Epoch 18, batch 2500, loss[ctc_loss=0.09686, att_loss=0.2367, loss=0.2087, over 16403.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007767, over 44.00 utterances.], tot_loss[ctc_loss=0.07927, att_loss=0.2384, loss=0.2066, over 3282861.63 frames. utt_duration=1290 frames, utt_pad_proportion=0.04154, over 10191.05 utterances.], batch size: 44, lr: 5.98e-03, grad_scale: 8.0 2023-03-08 17:38:26,726 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.106e+02 2.442e+02 2.921e+02 5.704e+02, threshold=4.884e+02, percent-clipped=1.0 2023-03-08 17:39:06,608 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:39:16,825 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1904, 5.1794, 4.9704, 2.8893, 4.9949, 4.7846, 4.5287, 2.9102], device='cuda:3'), covar=tensor([0.0096, 0.0118, 0.0262, 0.1087, 0.0091, 0.0168, 0.0271, 0.1284], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0097, 0.0097, 0.0109, 0.0081, 0.0106, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 17:39:31,633 INFO [train2.py:809] (3/4) Epoch 18, batch 2550, loss[ctc_loss=0.06907, att_loss=0.2197, loss=0.1896, over 16176.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.007067, over 41.00 utterances.], tot_loss[ctc_loss=0.07967, att_loss=0.238, loss=0.2063, over 3281897.31 frames. utt_duration=1294 frames, utt_pad_proportion=0.04066, over 10156.14 utterances.], batch size: 41, lr: 5.98e-03, grad_scale: 8.0 2023-03-08 17:40:43,978 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:40:50,214 INFO [train2.py:809] (3/4) Epoch 18, batch 2600, loss[ctc_loss=0.07839, att_loss=0.2206, loss=0.1922, over 16155.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.007777, over 41.00 utterances.], tot_loss[ctc_loss=0.08004, att_loss=0.2377, loss=0.2062, over 3271871.93 frames. utt_duration=1265 frames, utt_pad_proportion=0.05152, over 10361.41 utterances.], batch size: 41, lr: 5.98e-03, grad_scale: 8.0 2023-03-08 17:40:58,274 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:41:05,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 1.967e+02 2.510e+02 3.094e+02 6.487e+02, threshold=5.020e+02, percent-clipped=1.0 2023-03-08 17:42:08,697 INFO [train2.py:809] (3/4) Epoch 18, batch 2650, loss[ctc_loss=0.1002, att_loss=0.2529, loss=0.2224, over 17048.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009144, over 52.00 utterances.], tot_loss[ctc_loss=0.08057, att_loss=0.2384, loss=0.2069, over 3278372.81 frames. utt_duration=1257 frames, utt_pad_proportion=0.05218, over 10446.86 utterances.], batch size: 52, lr: 5.98e-03, grad_scale: 8.0 2023-03-08 17:42:13,358 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 17:42:14,723 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8694, 6.1113, 5.5441, 5.8841, 5.7677, 5.3229, 5.5032, 5.3545], device='cuda:3'), covar=tensor([0.1193, 0.0761, 0.0767, 0.0744, 0.0818, 0.1353, 0.2054, 0.1936], device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0574, 0.0435, 0.0431, 0.0408, 0.0445, 0.0585, 0.0500], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 17:43:27,231 INFO [train2.py:809] (3/4) Epoch 18, batch 2700, loss[ctc_loss=0.1002, att_loss=0.2536, loss=0.223, over 16899.00 frames. utt_duration=684.3 frames, utt_pad_proportion=0.1393, over 99.00 utterances.], tot_loss[ctc_loss=0.08089, att_loss=0.2389, loss=0.2073, over 3280545.24 frames. utt_duration=1234 frames, utt_pad_proportion=0.05626, over 10649.78 utterances.], batch size: 99, lr: 5.98e-03, grad_scale: 8.0 2023-03-08 17:43:30,664 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5766, 2.8895, 3.6055, 3.1276, 3.5342, 4.5676, 4.4678, 3.1586], device='cuda:3'), covar=tensor([0.0307, 0.1737, 0.1243, 0.1217, 0.1048, 0.0746, 0.0464, 0.1436], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0240, 0.0270, 0.0213, 0.0261, 0.0352, 0.0249, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 17:43:43,124 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 1.957e+02 2.578e+02 3.179e+02 5.678e+02, threshold=5.155e+02, percent-clipped=6.0 2023-03-08 17:43:54,198 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0302, 4.3334, 4.1150, 4.5173, 2.4417, 4.5952, 2.5164, 2.0506], device='cuda:3'), covar=tensor([0.0500, 0.0242, 0.1085, 0.0195, 0.2441, 0.0167, 0.2137, 0.2007], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0150, 0.0256, 0.0145, 0.0221, 0.0129, 0.0231, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:44:15,439 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6609, 5.9148, 5.3400, 5.6503, 5.5680, 5.0214, 5.2784, 5.1246], device='cuda:3'), covar=tensor([0.1296, 0.0824, 0.0941, 0.0814, 0.0944, 0.1545, 0.2274, 0.2089], device='cuda:3'), in_proj_covar=tensor([0.0496, 0.0577, 0.0439, 0.0435, 0.0412, 0.0450, 0.0591, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 17:44:31,283 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:44:33,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.63 vs. limit=5.0 2023-03-08 17:44:46,110 INFO [train2.py:809] (3/4) Epoch 18, batch 2750, loss[ctc_loss=0.05745, att_loss=0.2075, loss=0.1775, over 15769.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.008667, over 38.00 utterances.], tot_loss[ctc_loss=0.08145, att_loss=0.2395, loss=0.2079, over 3279818.00 frames. utt_duration=1205 frames, utt_pad_proportion=0.06355, over 10897.29 utterances.], batch size: 38, lr: 5.97e-03, grad_scale: 8.0 2023-03-08 17:46:05,053 INFO [train2.py:809] (3/4) Epoch 18, batch 2800, loss[ctc_loss=0.07014, att_loss=0.2231, loss=0.1925, over 16402.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.00751, over 44.00 utterances.], tot_loss[ctc_loss=0.08049, att_loss=0.2385, loss=0.2069, over 3268240.72 frames. utt_duration=1207 frames, utt_pad_proportion=0.06502, over 10841.16 utterances.], batch size: 44, lr: 5.97e-03, grad_scale: 8.0 2023-03-08 17:46:20,626 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.948e+02 2.363e+02 2.957e+02 5.460e+02, threshold=4.725e+02, percent-clipped=2.0 2023-03-08 17:47:17,801 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9324, 6.1307, 5.5614, 5.8961, 5.7889, 5.2867, 5.5516, 5.3365], device='cuda:3'), covar=tensor([0.1070, 0.0790, 0.0716, 0.0724, 0.0743, 0.1296, 0.2104, 0.2234], device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0576, 0.0437, 0.0430, 0.0410, 0.0446, 0.0586, 0.0505], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 17:47:24,026 INFO [train2.py:809] (3/4) Epoch 18, batch 2850, loss[ctc_loss=0.07514, att_loss=0.2358, loss=0.2037, over 16329.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.00611, over 45.00 utterances.], tot_loss[ctc_loss=0.07974, att_loss=0.2383, loss=0.2066, over 3272377.74 frames. utt_duration=1236 frames, utt_pad_proportion=0.057, over 10601.84 utterances.], batch size: 45, lr: 5.97e-03, grad_scale: 8.0 2023-03-08 17:47:54,987 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6546, 2.9361, 3.5180, 4.5139, 3.9872, 4.0041, 3.0986, 2.3566], device='cuda:3'), covar=tensor([0.0573, 0.1875, 0.0831, 0.0455, 0.0756, 0.0429, 0.1308, 0.2079], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0213, 0.0185, 0.0210, 0.0211, 0.0170, 0.0198, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 17:48:29,687 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:48:43,294 INFO [train2.py:809] (3/4) Epoch 18, batch 2900, loss[ctc_loss=0.09138, att_loss=0.2536, loss=0.2212, over 17419.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.03113, over 63.00 utterances.], tot_loss[ctc_loss=0.07945, att_loss=0.2378, loss=0.2061, over 3269850.62 frames. utt_duration=1235 frames, utt_pad_proportion=0.05642, over 10601.68 utterances.], batch size: 63, lr: 5.97e-03, grad_scale: 8.0 2023-03-08 17:48:59,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.930e+02 2.447e+02 2.876e+02 5.952e+02, threshold=4.894e+02, percent-clipped=4.0 2023-03-08 17:48:59,787 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8898, 5.1551, 5.0980, 4.9894, 5.1955, 5.1616, 4.8625, 4.6160], device='cuda:3'), covar=tensor([0.1029, 0.0491, 0.0259, 0.0590, 0.0304, 0.0316, 0.0346, 0.0328], device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0335, 0.0316, 0.0335, 0.0397, 0.0408, 0.0338, 0.0374], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:50:04,771 INFO [train2.py:809] (3/4) Epoch 18, batch 2950, loss[ctc_loss=0.07843, att_loss=0.2543, loss=0.2191, over 17042.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009695, over 52.00 utterances.], tot_loss[ctc_loss=0.07974, att_loss=0.2378, loss=0.2062, over 3275572.25 frames. utt_duration=1239 frames, utt_pad_proportion=0.05396, over 10583.92 utterances.], batch size: 52, lr: 5.96e-03, grad_scale: 8.0 2023-03-08 17:51:26,431 INFO [train2.py:809] (3/4) Epoch 18, batch 3000, loss[ctc_loss=0.07466, att_loss=0.2394, loss=0.2065, over 16331.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006034, over 45.00 utterances.], tot_loss[ctc_loss=0.0805, att_loss=0.2388, loss=0.2071, over 3280561.19 frames. utt_duration=1248 frames, utt_pad_proportion=0.05114, over 10529.45 utterances.], batch size: 45, lr: 5.96e-03, grad_scale: 8.0 2023-03-08 17:51:26,432 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 17:51:43,588 INFO [train2.py:843] (3/4) Epoch 18, validation: ctc_loss=0.04147, att_loss=0.2347, loss=0.196, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 17:51:43,589 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 17:51:59,785 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 1.987e+02 2.490e+02 3.021e+02 7.545e+02, threshold=4.980e+02, percent-clipped=3.0 2023-03-08 17:52:49,460 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:53:03,426 INFO [train2.py:809] (3/4) Epoch 18, batch 3050, loss[ctc_loss=0.09057, att_loss=0.2464, loss=0.2152, over 16966.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007524, over 50.00 utterances.], tot_loss[ctc_loss=0.07988, att_loss=0.2382, loss=0.2065, over 3267158.24 frames. utt_duration=1270 frames, utt_pad_proportion=0.05052, over 10305.56 utterances.], batch size: 50, lr: 5.96e-03, grad_scale: 8.0 2023-03-08 17:53:13,214 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8431, 5.0267, 4.7640, 2.1406, 2.0206, 2.6971, 2.4027, 3.7181], device='cuda:3'), covar=tensor([0.0913, 0.0246, 0.0265, 0.5085, 0.5857, 0.2605, 0.3347, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0263, 0.0260, 0.0239, 0.0347, 0.0335, 0.0249, 0.0364], device='cuda:3'), out_proj_covar=tensor([1.5042e-04, 9.8031e-05, 1.1182e-04, 1.0338e-04, 1.4586e-04, 1.3183e-04, 9.9490e-05, 1.4929e-04], device='cuda:3') 2023-03-08 17:54:07,375 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:54:11,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-03-08 17:54:26,018 INFO [train2.py:809] (3/4) Epoch 18, batch 3100, loss[ctc_loss=0.07582, att_loss=0.2509, loss=0.2159, over 17017.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008079, over 51.00 utterances.], tot_loss[ctc_loss=0.07973, att_loss=0.2382, loss=0.2065, over 3262820.81 frames. utt_duration=1260 frames, utt_pad_proportion=0.05354, over 10368.69 utterances.], batch size: 51, lr: 5.96e-03, grad_scale: 8.0 2023-03-08 17:54:41,879 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 2.010e+02 2.337e+02 3.007e+02 9.236e+02, threshold=4.674e+02, percent-clipped=2.0 2023-03-08 17:55:30,089 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1210, 5.4096, 5.3622, 5.2804, 5.4386, 5.4292, 5.1022, 4.8567], device='cuda:3'), covar=tensor([0.1086, 0.0497, 0.0237, 0.0474, 0.0325, 0.0306, 0.0361, 0.0342], device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0337, 0.0318, 0.0334, 0.0397, 0.0411, 0.0339, 0.0377], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 17:55:46,928 INFO [train2.py:809] (3/4) Epoch 18, batch 3150, loss[ctc_loss=0.05846, att_loss=0.2048, loss=0.1755, over 15352.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01253, over 35.00 utterances.], tot_loss[ctc_loss=0.07895, att_loss=0.2379, loss=0.2061, over 3261648.21 frames. utt_duration=1275 frames, utt_pad_proportion=0.05048, over 10242.14 utterances.], batch size: 35, lr: 5.96e-03, grad_scale: 8.0 2023-03-08 17:56:52,228 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:57:07,455 INFO [train2.py:809] (3/4) Epoch 18, batch 3200, loss[ctc_loss=0.0891, att_loss=0.2582, loss=0.2244, over 17562.00 frames. utt_duration=1005 frames, utt_pad_proportion=0.0473, over 70.00 utterances.], tot_loss[ctc_loss=0.07863, att_loss=0.2379, loss=0.2061, over 3267725.77 frames. utt_duration=1272 frames, utt_pad_proportion=0.04931, over 10288.90 utterances.], batch size: 70, lr: 5.95e-03, grad_scale: 8.0 2023-03-08 17:57:23,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 2.043e+02 2.547e+02 2.997e+02 7.270e+02, threshold=5.093e+02, percent-clipped=4.0 2023-03-08 17:58:10,132 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:58:28,694 INFO [train2.py:809] (3/4) Epoch 18, batch 3250, loss[ctc_loss=0.1029, att_loss=0.252, loss=0.2222, over 16341.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.00573, over 45.00 utterances.], tot_loss[ctc_loss=0.07892, att_loss=0.2375, loss=0.2058, over 3273228.08 frames. utt_duration=1284 frames, utt_pad_proportion=0.04569, over 10212.44 utterances.], batch size: 45, lr: 5.95e-03, grad_scale: 8.0 2023-03-08 17:58:56,357 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 17:59:47,177 INFO [train2.py:809] (3/4) Epoch 18, batch 3300, loss[ctc_loss=0.08904, att_loss=0.2608, loss=0.2265, over 17032.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01091, over 53.00 utterances.], tot_loss[ctc_loss=0.08, att_loss=0.2386, loss=0.2069, over 3277634.41 frames. utt_duration=1269 frames, utt_pad_proportion=0.04744, over 10342.47 utterances.], batch size: 53, lr: 5.95e-03, grad_scale: 8.0 2023-03-08 17:59:55,991 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:00:02,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.989e+02 2.342e+02 2.870e+02 6.957e+02, threshold=4.685e+02, percent-clipped=3.0 2023-03-08 18:00:31,335 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:00:51,190 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2037, 5.4401, 5.0137, 5.5161, 4.9330, 5.1281, 5.6086, 5.3953], device='cuda:3'), covar=tensor([0.0533, 0.0272, 0.0833, 0.0266, 0.0419, 0.0199, 0.0206, 0.0178], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0305, 0.0357, 0.0324, 0.0308, 0.0233, 0.0290, 0.0272], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:00:58,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7786, 3.4820, 3.5394, 3.0521, 3.6287, 3.5285, 3.6430, 2.4954], device='cuda:3'), covar=tensor([0.1019, 0.1183, 0.1829, 0.3687, 0.0831, 0.2270, 0.0718, 0.4556], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0167, 0.0180, 0.0240, 0.0144, 0.0238, 0.0160, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:01:05,451 INFO [train2.py:809] (3/4) Epoch 18, batch 3350, loss[ctc_loss=0.07746, att_loss=0.2419, loss=0.209, over 16407.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006759, over 44.00 utterances.], tot_loss[ctc_loss=0.08041, att_loss=0.2385, loss=0.2069, over 3266736.07 frames. utt_duration=1261 frames, utt_pad_proportion=0.05303, over 10371.51 utterances.], batch size: 44, lr: 5.95e-03, grad_scale: 8.0 2023-03-08 18:01:31,210 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:01:43,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-08 18:02:18,415 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9637, 5.2761, 4.8651, 5.3852, 4.7453, 5.0170, 5.4303, 5.2432], device='cuda:3'), covar=tensor([0.0510, 0.0250, 0.0728, 0.0264, 0.0394, 0.0215, 0.0213, 0.0172], device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0304, 0.0356, 0.0323, 0.0307, 0.0232, 0.0289, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:02:25,937 INFO [train2.py:809] (3/4) Epoch 18, batch 3400, loss[ctc_loss=0.07545, att_loss=0.2563, loss=0.2201, over 16328.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006202, over 45.00 utterances.], tot_loss[ctc_loss=0.08046, att_loss=0.239, loss=0.2073, over 3268206.10 frames. utt_duration=1241 frames, utt_pad_proportion=0.05756, over 10548.15 utterances.], batch size: 45, lr: 5.95e-03, grad_scale: 8.0 2023-03-08 18:02:42,142 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.089e+02 2.385e+02 3.204e+02 8.386e+02, threshold=4.770e+02, percent-clipped=4.0 2023-03-08 18:03:46,797 INFO [train2.py:809] (3/4) Epoch 18, batch 3450, loss[ctc_loss=0.06418, att_loss=0.2122, loss=0.1826, over 15627.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.008195, over 37.00 utterances.], tot_loss[ctc_loss=0.08038, att_loss=0.2382, loss=0.2067, over 3260349.77 frames. utt_duration=1238 frames, utt_pad_proportion=0.06072, over 10545.21 utterances.], batch size: 37, lr: 5.94e-03, grad_scale: 8.0 2023-03-08 18:03:47,771 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 18:04:01,727 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7848, 5.0898, 4.6374, 5.1670, 4.5580, 4.7100, 5.2220, 5.0174], device='cuda:3'), covar=tensor([0.0558, 0.0265, 0.0837, 0.0338, 0.0434, 0.0354, 0.0245, 0.0206], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0306, 0.0359, 0.0326, 0.0308, 0.0234, 0.0291, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:05:06,670 INFO [train2.py:809] (3/4) Epoch 18, batch 3500, loss[ctc_loss=0.06786, att_loss=0.2438, loss=0.2086, over 17250.00 frames. utt_duration=874.9 frames, utt_pad_proportion=0.07904, over 79.00 utterances.], tot_loss[ctc_loss=0.07913, att_loss=0.238, loss=0.2062, over 3274825.21 frames. utt_duration=1235 frames, utt_pad_proportion=0.05599, over 10619.19 utterances.], batch size: 79, lr: 5.94e-03, grad_scale: 8.0 2023-03-08 18:05:22,701 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.989e+02 2.438e+02 2.819e+02 5.372e+02, threshold=4.876e+02, percent-clipped=3.0 2023-03-08 18:05:29,656 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2023-03-08 18:05:30,884 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5365, 2.9009, 3.7025, 3.1162, 3.4182, 4.6453, 4.4213, 3.1855], device='cuda:3'), covar=tensor([0.0333, 0.1738, 0.1095, 0.1248, 0.1146, 0.0598, 0.0497, 0.1312], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0239, 0.0271, 0.0214, 0.0258, 0.0352, 0.0250, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:05:42,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 18:06:26,583 INFO [train2.py:809] (3/4) Epoch 18, batch 3550, loss[ctc_loss=0.08015, att_loss=0.2429, loss=0.2104, over 17299.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02452, over 59.00 utterances.], tot_loss[ctc_loss=0.07964, att_loss=0.2387, loss=0.2069, over 3277585.06 frames. utt_duration=1218 frames, utt_pad_proportion=0.05972, over 10779.87 utterances.], batch size: 59, lr: 5.94e-03, grad_scale: 8.0 2023-03-08 18:06:38,917 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-03-08 18:07:29,660 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3558, 2.6627, 3.0929, 4.3408, 3.7579, 3.8302, 2.7442, 1.8884], device='cuda:3'), covar=tensor([0.0673, 0.2055, 0.1016, 0.0625, 0.0851, 0.0440, 0.1519, 0.2567], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0211, 0.0183, 0.0208, 0.0212, 0.0169, 0.0196, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:07:46,260 INFO [train2.py:809] (3/4) Epoch 18, batch 3600, loss[ctc_loss=0.06417, att_loss=0.2185, loss=0.1876, over 16014.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007492, over 40.00 utterances.], tot_loss[ctc_loss=0.07906, att_loss=0.2381, loss=0.2063, over 3273588.04 frames. utt_duration=1232 frames, utt_pad_proportion=0.05686, over 10638.95 utterances.], batch size: 40, lr: 5.94e-03, grad_scale: 8.0 2023-03-08 18:08:02,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 1.997e+02 2.305e+02 2.777e+02 7.963e+02, threshold=4.609e+02, percent-clipped=3.0 2023-03-08 18:08:10,395 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7185, 4.5075, 4.6911, 4.5496, 5.2193, 4.5605, 4.5590, 2.3834], device='cuda:3'), covar=tensor([0.0184, 0.0291, 0.0259, 0.0291, 0.0919, 0.0186, 0.0284, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0167, 0.0171, 0.0187, 0.0358, 0.0141, 0.0159, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:08:23,286 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:09:09,353 INFO [train2.py:809] (3/4) Epoch 18, batch 3650, loss[ctc_loss=0.106, att_loss=0.2553, loss=0.2254, over 17335.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02175, over 59.00 utterances.], tot_loss[ctc_loss=0.07951, att_loss=0.2384, loss=0.2066, over 3269323.08 frames. utt_duration=1231 frames, utt_pad_proportion=0.05792, over 10638.10 utterances.], batch size: 59, lr: 5.94e-03, grad_scale: 8.0 2023-03-08 18:09:28,407 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:10:04,978 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 18:10:06,310 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3977, 2.5991, 4.9090, 3.8023, 2.9643, 4.1688, 4.5795, 4.5337], device='cuda:3'), covar=tensor([0.0241, 0.1652, 0.0159, 0.0922, 0.1725, 0.0273, 0.0155, 0.0244], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0238, 0.0168, 0.0307, 0.0263, 0.0204, 0.0155, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:10:33,065 INFO [train2.py:809] (3/4) Epoch 18, batch 3700, loss[ctc_loss=0.04938, att_loss=0.2259, loss=0.1906, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.006667, over 50.00 utterances.], tot_loss[ctc_loss=0.07951, att_loss=0.2388, loss=0.207, over 3264525.71 frames. utt_duration=1235 frames, utt_pad_proportion=0.056, over 10586.15 utterances.], batch size: 50, lr: 5.93e-03, grad_scale: 8.0 2023-03-08 18:10:49,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.052e+02 2.398e+02 2.900e+02 5.315e+02, threshold=4.797e+02, percent-clipped=3.0 2023-03-08 18:11:56,308 INFO [train2.py:809] (3/4) Epoch 18, batch 3750, loss[ctc_loss=0.09594, att_loss=0.2591, loss=0.2265, over 17288.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01187, over 55.00 utterances.], tot_loss[ctc_loss=0.0788, att_loss=0.2376, loss=0.2058, over 3261548.99 frames. utt_duration=1266 frames, utt_pad_proportion=0.05084, over 10319.85 utterances.], batch size: 55, lr: 5.93e-03, grad_scale: 8.0 2023-03-08 18:13:18,688 INFO [train2.py:809] (3/4) Epoch 18, batch 3800, loss[ctc_loss=0.0624, att_loss=0.2392, loss=0.2038, over 16870.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006543, over 49.00 utterances.], tot_loss[ctc_loss=0.07871, att_loss=0.2381, loss=0.2062, over 3272709.18 frames. utt_duration=1262 frames, utt_pad_proportion=0.04923, over 10384.34 utterances.], batch size: 49, lr: 5.93e-03, grad_scale: 8.0 2023-03-08 18:13:34,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.121e+02 2.472e+02 3.058e+02 5.629e+02, threshold=4.945e+02, percent-clipped=2.0 2023-03-08 18:14:18,290 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7990, 3.3947, 3.8437, 3.4186, 3.8272, 4.8395, 4.6220, 3.7549], device='cuda:3'), covar=tensor([0.0292, 0.1288, 0.1053, 0.1226, 0.0971, 0.0804, 0.0595, 0.1000], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0239, 0.0271, 0.0213, 0.0258, 0.0352, 0.0251, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:14:37,809 INFO [train2.py:809] (3/4) Epoch 18, batch 3850, loss[ctc_loss=0.05778, att_loss=0.2282, loss=0.1941, over 16400.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007676, over 44.00 utterances.], tot_loss[ctc_loss=0.07914, att_loss=0.2379, loss=0.2061, over 3260093.16 frames. utt_duration=1254 frames, utt_pad_proportion=0.05497, over 10408.98 utterances.], batch size: 44, lr: 5.93e-03, grad_scale: 8.0 2023-03-08 18:15:31,002 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-03-08 18:15:41,539 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:15:54,895 INFO [train2.py:809] (3/4) Epoch 18, batch 3900, loss[ctc_loss=0.0795, att_loss=0.2499, loss=0.2158, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005782, over 47.00 utterances.], tot_loss[ctc_loss=0.07943, att_loss=0.2385, loss=0.2067, over 3259887.29 frames. utt_duration=1231 frames, utt_pad_proportion=0.06099, over 10606.90 utterances.], batch size: 47, lr: 5.93e-03, grad_scale: 8.0 2023-03-08 18:16:10,964 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.131e+02 2.535e+02 3.133e+02 8.645e+02, threshold=5.071e+02, percent-clipped=4.0 2023-03-08 18:16:31,502 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:17:07,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.15 vs. limit=5.0 2023-03-08 18:17:13,024 INFO [train2.py:809] (3/4) Epoch 18, batch 3950, loss[ctc_loss=0.07175, att_loss=0.2262, loss=0.1953, over 15960.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.006607, over 41.00 utterances.], tot_loss[ctc_loss=0.07889, att_loss=0.2379, loss=0.2061, over 3261020.39 frames. utt_duration=1252 frames, utt_pad_proportion=0.05453, over 10431.84 utterances.], batch size: 41, lr: 5.92e-03, grad_scale: 8.0 2023-03-08 18:17:16,292 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:17:29,975 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:17:45,661 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:18:26,975 INFO [train2.py:809] (3/4) Epoch 19, batch 0, loss[ctc_loss=0.07776, att_loss=0.2513, loss=0.2166, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005229, over 47.00 utterances.], tot_loss[ctc_loss=0.07776, att_loss=0.2513, loss=0.2166, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005229, over 47.00 utterances.], batch size: 47, lr: 5.76e-03, grad_scale: 16.0 2023-03-08 18:18:26,976 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 18:18:38,962 INFO [train2.py:843] (3/4) Epoch 19, validation: ctc_loss=0.04291, att_loss=0.2348, loss=0.1964, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 18:18:38,963 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 18:19:14,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-08 18:19:18,775 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6912, 5.0612, 4.8793, 4.8937, 5.0293, 4.7053, 3.7195, 4.9345], device='cuda:3'), covar=tensor([0.0109, 0.0132, 0.0128, 0.0108, 0.0098, 0.0115, 0.0653, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0085, 0.0105, 0.0065, 0.0071, 0.0082, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:19:20,155 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:19:21,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.522e+02 2.028e+02 2.615e+02 3.173e+02 7.413e+02, threshold=5.229e+02, percent-clipped=5.0 2023-03-08 18:19:57,608 INFO [train2.py:809] (3/4) Epoch 19, batch 50, loss[ctc_loss=0.06535, att_loss=0.2135, loss=0.1839, over 15767.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.008824, over 38.00 utterances.], tot_loss[ctc_loss=0.07785, att_loss=0.237, loss=0.2052, over 734875.70 frames. utt_duration=1390 frames, utt_pad_proportion=0.0293, over 2117.18 utterances.], batch size: 38, lr: 5.76e-03, grad_scale: 16.0 2023-03-08 18:19:57,882 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1135, 5.4460, 4.9913, 5.5360, 4.8933, 5.1273, 5.5865, 5.3867], device='cuda:3'), covar=tensor([0.0584, 0.0310, 0.0797, 0.0336, 0.0415, 0.0229, 0.0231, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0309, 0.0358, 0.0327, 0.0311, 0.0234, 0.0293, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:20:43,419 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8149, 5.3151, 5.3326, 5.2591, 5.3544, 5.2991, 5.0498, 4.8382], device='cuda:3'), covar=tensor([0.1401, 0.0577, 0.0343, 0.0510, 0.0376, 0.0407, 0.0418, 0.0386], device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0337, 0.0319, 0.0334, 0.0396, 0.0408, 0.0338, 0.0377], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 18:21:17,852 INFO [train2.py:809] (3/4) Epoch 19, batch 100, loss[ctc_loss=0.07562, att_loss=0.2238, loss=0.1941, over 14118.00 frames. utt_duration=1823 frames, utt_pad_proportion=0.0465, over 31.00 utterances.], tot_loss[ctc_loss=0.0786, att_loss=0.2382, loss=0.2063, over 1300029.32 frames. utt_duration=1324 frames, utt_pad_proportion=0.03766, over 3932.28 utterances.], batch size: 31, lr: 5.76e-03, grad_scale: 16.0 2023-03-08 18:22:00,751 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.106e+02 2.518e+02 3.036e+02 5.583e+02, threshold=5.037e+02, percent-clipped=2.0 2023-03-08 18:22:36,463 INFO [train2.py:809] (3/4) Epoch 19, batch 150, loss[ctc_loss=0.09358, att_loss=0.2656, loss=0.2312, over 17274.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01183, over 55.00 utterances.], tot_loss[ctc_loss=0.07839, att_loss=0.2384, loss=0.2064, over 1738883.92 frames. utt_duration=1304 frames, utt_pad_proportion=0.04362, over 5338.12 utterances.], batch size: 55, lr: 5.76e-03, grad_scale: 16.0 2023-03-08 18:23:05,235 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7584, 5.0732, 4.9749, 5.0765, 5.2379, 4.7421, 3.5647, 5.0799], device='cuda:3'), covar=tensor([0.0098, 0.0117, 0.0111, 0.0083, 0.0077, 0.0106, 0.0687, 0.0173], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0085, 0.0105, 0.0066, 0.0071, 0.0082, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:23:56,062 INFO [train2.py:809] (3/4) Epoch 19, batch 200, loss[ctc_loss=0.08001, att_loss=0.2577, loss=0.2222, over 17154.00 frames. utt_duration=1227 frames, utt_pad_proportion=0.01292, over 56.00 utterances.], tot_loss[ctc_loss=0.07854, att_loss=0.2391, loss=0.207, over 2074600.93 frames. utt_duration=1281 frames, utt_pad_proportion=0.04642, over 6485.49 utterances.], batch size: 56, lr: 5.75e-03, grad_scale: 16.0 2023-03-08 18:24:14,384 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:24:36,185 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9149, 5.2875, 4.8186, 5.3465, 4.7045, 4.9076, 5.3854, 5.2389], device='cuda:3'), covar=tensor([0.0582, 0.0308, 0.0864, 0.0315, 0.0407, 0.0271, 0.0232, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0307, 0.0358, 0.0326, 0.0310, 0.0234, 0.0291, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:24:39,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.975e+02 2.308e+02 2.986e+02 5.349e+02, threshold=4.617e+02, percent-clipped=2.0 2023-03-08 18:25:14,793 INFO [train2.py:809] (3/4) Epoch 19, batch 250, loss[ctc_loss=0.05391, att_loss=0.2051, loss=0.1748, over 15626.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.008465, over 37.00 utterances.], tot_loss[ctc_loss=0.07744, att_loss=0.2378, loss=0.2057, over 2331229.93 frames. utt_duration=1293 frames, utt_pad_proportion=0.04591, over 7222.57 utterances.], batch size: 37, lr: 5.75e-03, grad_scale: 16.0 2023-03-08 18:25:21,532 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 18:25:37,800 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:25:51,077 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:26:40,276 INFO [train2.py:809] (3/4) Epoch 19, batch 300, loss[ctc_loss=0.07997, att_loss=0.2433, loss=0.2106, over 17298.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01135, over 55.00 utterances.], tot_loss[ctc_loss=0.07922, att_loss=0.2391, loss=0.2072, over 2533545.93 frames. utt_duration=1191 frames, utt_pad_proportion=0.07281, over 8523.05 utterances.], batch size: 55, lr: 5.75e-03, grad_scale: 16.0 2023-03-08 18:27:22,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.009e+02 2.404e+02 3.052e+02 6.487e+02, threshold=4.807e+02, percent-clipped=6.0 2023-03-08 18:27:59,505 INFO [train2.py:809] (3/4) Epoch 19, batch 350, loss[ctc_loss=0.1172, att_loss=0.2554, loss=0.2277, over 16379.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.008211, over 44.00 utterances.], tot_loss[ctc_loss=0.07914, att_loss=0.2395, loss=0.2074, over 2704670.37 frames. utt_duration=1213 frames, utt_pad_proportion=0.06431, over 8929.42 utterances.], batch size: 44, lr: 5.75e-03, grad_scale: 16.0 2023-03-08 18:28:01,390 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2841, 2.3154, 2.8153, 4.3176, 3.8399, 3.8563, 2.7432, 2.0086], device='cuda:3'), covar=tensor([0.0714, 0.2406, 0.1327, 0.0492, 0.0764, 0.0441, 0.1632, 0.2507], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0212, 0.0184, 0.0209, 0.0215, 0.0171, 0.0199, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:28:13,591 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9521, 5.2842, 5.5144, 5.3108, 5.3881, 5.9080, 5.1783, 6.0356], device='cuda:3'), covar=tensor([0.0672, 0.0812, 0.0794, 0.1264, 0.1817, 0.0951, 0.0653, 0.0640], device='cuda:3'), in_proj_covar=tensor([0.0823, 0.0484, 0.0567, 0.0626, 0.0835, 0.0589, 0.0464, 0.0575], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:28:33,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 18:28:36,946 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7901, 5.1363, 4.6332, 5.1762, 4.5743, 4.8063, 5.2329, 5.0702], device='cuda:3'), covar=tensor([0.0597, 0.0251, 0.0820, 0.0265, 0.0409, 0.0353, 0.0214, 0.0173], device='cuda:3'), in_proj_covar=tensor([0.0380, 0.0309, 0.0359, 0.0327, 0.0312, 0.0236, 0.0292, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 18:29:05,015 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2482, 5.4168, 4.9849, 3.2009, 5.2597, 5.0677, 4.6374, 2.8003], device='cuda:3'), covar=tensor([0.0169, 0.0097, 0.0317, 0.1181, 0.0088, 0.0174, 0.0358, 0.1972], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0098, 0.0098, 0.0108, 0.0081, 0.0108, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:29:17,685 INFO [train2.py:809] (3/4) Epoch 19, batch 400, loss[ctc_loss=0.09547, att_loss=0.2601, loss=0.2272, over 17245.00 frames. utt_duration=1171 frames, utt_pad_proportion=0.02261, over 59.00 utterances.], tot_loss[ctc_loss=0.07888, att_loss=0.2393, loss=0.2072, over 2831828.00 frames. utt_duration=1234 frames, utt_pad_proportion=0.05715, over 9187.91 utterances.], batch size: 59, lr: 5.75e-03, grad_scale: 8.0 2023-03-08 18:29:51,763 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:30:02,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.245e+02 2.032e+02 2.343e+02 2.887e+02 1.413e+03, threshold=4.687e+02, percent-clipped=3.0 2023-03-08 18:30:06,195 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4750, 4.6600, 4.6125, 4.5485, 5.2029, 4.5167, 4.6436, 2.6006], device='cuda:3'), covar=tensor([0.0282, 0.0317, 0.0305, 0.0348, 0.1142, 0.0261, 0.0308, 0.1868], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0170, 0.0172, 0.0188, 0.0364, 0.0144, 0.0161, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:30:37,548 INFO [train2.py:809] (3/4) Epoch 19, batch 450, loss[ctc_loss=0.08, att_loss=0.2416, loss=0.2093, over 17000.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009644, over 51.00 utterances.], tot_loss[ctc_loss=0.07938, att_loss=0.2397, loss=0.2076, over 2940154.74 frames. utt_duration=1217 frames, utt_pad_proportion=0.05694, over 9678.10 utterances.], batch size: 51, lr: 5.74e-03, grad_scale: 8.0 2023-03-08 18:31:05,935 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0680, 5.0210, 4.8286, 3.1280, 4.8782, 4.7632, 4.1840, 2.5633], device='cuda:3'), covar=tensor([0.0107, 0.0097, 0.0264, 0.0881, 0.0091, 0.0175, 0.0332, 0.1430], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0098, 0.0098, 0.0108, 0.0081, 0.0107, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:31:28,156 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:31:29,577 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9750, 5.2850, 5.1899, 5.1235, 5.2898, 5.2511, 4.9631, 4.7263], device='cuda:3'), covar=tensor([0.1073, 0.0538, 0.0276, 0.0575, 0.0307, 0.0313, 0.0406, 0.0379], device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0343, 0.0324, 0.0340, 0.0404, 0.0417, 0.0347, 0.0382], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 18:31:53,726 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-08 18:31:56,580 INFO [train2.py:809] (3/4) Epoch 19, batch 500, loss[ctc_loss=0.07648, att_loss=0.2534, loss=0.218, over 17278.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02351, over 59.00 utterances.], tot_loss[ctc_loss=0.07933, att_loss=0.2392, loss=0.2072, over 3007721.79 frames. utt_duration=1190 frames, utt_pad_proportion=0.06496, over 10126.72 utterances.], batch size: 59, lr: 5.74e-03, grad_scale: 8.0 2023-03-08 18:32:18,525 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8448, 5.2232, 5.0233, 5.1323, 5.3038, 4.8610, 3.6101, 5.2035], device='cuda:3'), covar=tensor([0.0098, 0.0108, 0.0118, 0.0078, 0.0094, 0.0111, 0.0676, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0085, 0.0106, 0.0065, 0.0071, 0.0082, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:32:32,218 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0328, 5.0523, 4.8309, 2.9422, 4.8338, 4.5844, 4.2130, 2.6997], device='cuda:3'), covar=tensor([0.0110, 0.0096, 0.0259, 0.0994, 0.0096, 0.0215, 0.0336, 0.1391], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0098, 0.0099, 0.0109, 0.0082, 0.0108, 0.0098, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:32:35,314 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7516, 2.7676, 5.1596, 4.2301, 3.2166, 4.3391, 4.8836, 4.7258], device='cuda:3'), covar=tensor([0.0234, 0.1422, 0.0159, 0.0789, 0.1598, 0.0238, 0.0149, 0.0237], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0244, 0.0174, 0.0314, 0.0271, 0.0207, 0.0160, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:32:39,385 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.024e+02 2.496e+02 3.106e+02 1.224e+03, threshold=4.991e+02, percent-clipped=5.0 2023-03-08 18:33:14,207 INFO [train2.py:809] (3/4) Epoch 19, batch 550, loss[ctc_loss=0.07075, att_loss=0.2263, loss=0.1952, over 15960.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.005958, over 41.00 utterances.], tot_loss[ctc_loss=0.07878, att_loss=0.2383, loss=0.2064, over 3058541.74 frames. utt_duration=1216 frames, utt_pad_proportion=0.06248, over 10069.51 utterances.], batch size: 41, lr: 5.74e-03, grad_scale: 8.0 2023-03-08 18:33:28,926 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7723, 6.0307, 5.3929, 5.8029, 5.6866, 5.1842, 5.4146, 5.2134], device='cuda:3'), covar=tensor([0.1202, 0.0868, 0.0918, 0.0702, 0.0866, 0.1386, 0.2226, 0.2063], device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0590, 0.0443, 0.0442, 0.0416, 0.0456, 0.0599, 0.0516], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:33:37,411 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:33:41,792 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:34:08,464 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0410, 5.0570, 4.8237, 2.8348, 4.8702, 4.6334, 4.2796, 2.6832], device='cuda:3'), covar=tensor([0.0114, 0.0111, 0.0289, 0.1085, 0.0104, 0.0214, 0.0340, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0099, 0.0099, 0.0109, 0.0083, 0.0109, 0.0099, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:34:32,967 INFO [train2.py:809] (3/4) Epoch 19, batch 600, loss[ctc_loss=0.07234, att_loss=0.2245, loss=0.1941, over 16139.00 frames. utt_duration=1539 frames, utt_pad_proportion=0.005433, over 42.00 utterances.], tot_loss[ctc_loss=0.07936, att_loss=0.2388, loss=0.2069, over 3106434.10 frames. utt_duration=1188 frames, utt_pad_proportion=0.07002, over 10470.55 utterances.], batch size: 42, lr: 5.74e-03, grad_scale: 8.0 2023-03-08 18:34:52,096 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:34:54,368 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9605, 5.2549, 5.5118, 5.3636, 5.4645, 5.9365, 5.2419, 6.0414], device='cuda:3'), covar=tensor([0.0663, 0.0713, 0.0732, 0.1166, 0.1646, 0.0824, 0.0607, 0.0593], device='cuda:3'), in_proj_covar=tensor([0.0832, 0.0489, 0.0572, 0.0631, 0.0843, 0.0593, 0.0465, 0.0580], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:35:02,583 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:35:18,142 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 1.966e+02 2.354e+02 3.021e+02 6.403e+02, threshold=4.709e+02, percent-clipped=3.0 2023-03-08 18:35:53,227 INFO [train2.py:809] (3/4) Epoch 19, batch 650, loss[ctc_loss=0.0784, att_loss=0.2359, loss=0.2044, over 16283.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.00696, over 43.00 utterances.], tot_loss[ctc_loss=0.07882, att_loss=0.2385, loss=0.2066, over 3140887.55 frames. utt_duration=1202 frames, utt_pad_proportion=0.06764, over 10463.61 utterances.], batch size: 43, lr: 5.74e-03, grad_scale: 8.0 2023-03-08 18:35:57,558 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2549, 3.8738, 3.4224, 3.5714, 4.0589, 3.7133, 3.1197, 4.4243], device='cuda:3'), covar=tensor([0.0947, 0.0521, 0.0919, 0.0653, 0.0707, 0.0709, 0.0822, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0215, 0.0223, 0.0193, 0.0269, 0.0234, 0.0197, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 18:36:40,905 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:37:13,708 INFO [train2.py:809] (3/4) Epoch 19, batch 700, loss[ctc_loss=0.09048, att_loss=0.2532, loss=0.2207, over 17143.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01304, over 56.00 utterances.], tot_loss[ctc_loss=0.07923, att_loss=0.2386, loss=0.2068, over 3173658.91 frames. utt_duration=1213 frames, utt_pad_proportion=0.06387, over 10481.34 utterances.], batch size: 56, lr: 5.73e-03, grad_scale: 8.0 2023-03-08 18:37:28,539 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5834, 2.9995, 3.8314, 3.2379, 3.7304, 4.6966, 4.4494, 3.3381], device='cuda:3'), covar=tensor([0.0399, 0.1722, 0.1034, 0.1304, 0.0980, 0.0837, 0.0690, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0242, 0.0273, 0.0216, 0.0260, 0.0353, 0.0252, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:37:57,757 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 2.010e+02 2.298e+02 2.722e+02 6.112e+02, threshold=4.595e+02, percent-clipped=4.0 2023-03-08 18:38:32,586 INFO [train2.py:809] (3/4) Epoch 19, batch 750, loss[ctc_loss=0.08769, att_loss=0.2573, loss=0.2234, over 17039.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007637, over 51.00 utterances.], tot_loss[ctc_loss=0.07933, att_loss=0.239, loss=0.2071, over 3197468.41 frames. utt_duration=1207 frames, utt_pad_proportion=0.06528, over 10606.62 utterances.], batch size: 51, lr: 5.73e-03, grad_scale: 8.0 2023-03-08 18:39:16,289 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:39:51,884 INFO [train2.py:809] (3/4) Epoch 19, batch 800, loss[ctc_loss=0.09082, att_loss=0.2304, loss=0.2025, over 15881.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009518, over 39.00 utterances.], tot_loss[ctc_loss=0.0798, att_loss=0.2395, loss=0.2076, over 3213646.69 frames. utt_duration=1218 frames, utt_pad_proportion=0.06311, over 10566.25 utterances.], batch size: 39, lr: 5.73e-03, grad_scale: 8.0 2023-03-08 18:40:37,047 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 2.045e+02 2.501e+02 3.023e+02 6.959e+02, threshold=5.001e+02, percent-clipped=5.0 2023-03-08 18:41:12,445 INFO [train2.py:809] (3/4) Epoch 19, batch 850, loss[ctc_loss=0.07302, att_loss=0.2352, loss=0.2028, over 16984.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.006457, over 50.00 utterances.], tot_loss[ctc_loss=0.07967, att_loss=0.2397, loss=0.2077, over 3227528.06 frames. utt_duration=1208 frames, utt_pad_proportion=0.06506, over 10700.96 utterances.], batch size: 50, lr: 5.73e-03, grad_scale: 8.0 2023-03-08 18:41:18,916 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2651, 2.7276, 3.1388, 4.2434, 3.8812, 3.8464, 2.8396, 1.9622], device='cuda:3'), covar=tensor([0.0726, 0.1993, 0.0937, 0.0546, 0.0744, 0.0425, 0.1438, 0.2503], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0210, 0.0184, 0.0209, 0.0213, 0.0169, 0.0197, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:41:36,246 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7784, 2.2769, 2.8461, 2.2909, 2.7194, 2.6530, 2.4716, 2.9899], device='cuda:3'), covar=tensor([0.1987, 0.3798, 0.2613, 0.3405, 0.2160, 0.1744, 0.3339, 0.1281], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0116, 0.0113, 0.0104, 0.0114, 0.0100, 0.0122, 0.0089], device='cuda:3'), out_proj_covar=tensor([8.1380e-05, 8.9066e-05, 8.8982e-05, 7.9734e-05, 8.3857e-05, 7.9257e-05, 9.0294e-05, 7.1637e-05], device='cuda:3') 2023-03-08 18:41:39,175 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:41:59,565 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2327, 2.6143, 2.9346, 3.9872, 3.6990, 3.6544, 2.6772, 1.9824], device='cuda:3'), covar=tensor([0.0658, 0.1904, 0.0960, 0.0583, 0.0765, 0.0465, 0.1424, 0.2336], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0209, 0.0183, 0.0208, 0.0213, 0.0169, 0.0196, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:42:31,643 INFO [train2.py:809] (3/4) Epoch 19, batch 900, loss[ctc_loss=0.08529, att_loss=0.2611, loss=0.2259, over 17113.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01461, over 56.00 utterances.], tot_loss[ctc_loss=0.07933, att_loss=0.2397, loss=0.2076, over 3242685.09 frames. utt_duration=1219 frames, utt_pad_proportion=0.06007, over 10651.95 utterances.], batch size: 56, lr: 5.73e-03, grad_scale: 8.0 2023-03-08 18:42:55,554 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:43:16,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.045e+02 2.502e+02 3.069e+02 5.173e+02, threshold=5.005e+02, percent-clipped=1.0 2023-03-08 18:43:29,475 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8551, 4.8783, 4.4698, 2.5285, 4.7612, 4.5654, 3.6537, 2.5980], device='cuda:3'), covar=tensor([0.0172, 0.0123, 0.0429, 0.1298, 0.0112, 0.0258, 0.0546, 0.1611], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0098, 0.0098, 0.0108, 0.0082, 0.0108, 0.0098, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:43:41,530 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-08 18:43:51,342 INFO [train2.py:809] (3/4) Epoch 19, batch 950, loss[ctc_loss=0.08029, att_loss=0.2449, loss=0.212, over 16610.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.00647, over 47.00 utterances.], tot_loss[ctc_loss=0.07905, att_loss=0.2394, loss=0.2073, over 3242889.02 frames. utt_duration=1206 frames, utt_pad_proportion=0.06644, over 10772.41 utterances.], batch size: 47, lr: 5.72e-03, grad_scale: 8.0 2023-03-08 18:44:30,134 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:44:40,234 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4946, 2.7176, 5.0167, 3.8911, 3.0905, 4.2432, 4.7897, 4.5464], device='cuda:3'), covar=tensor([0.0284, 0.1576, 0.0173, 0.0871, 0.1698, 0.0242, 0.0139, 0.0288], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0240, 0.0173, 0.0308, 0.0265, 0.0204, 0.0159, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:45:11,228 INFO [train2.py:809] (3/4) Epoch 19, batch 1000, loss[ctc_loss=0.06483, att_loss=0.2241, loss=0.1922, over 16128.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.005637, over 42.00 utterances.], tot_loss[ctc_loss=0.07883, att_loss=0.2395, loss=0.2074, over 3251686.88 frames. utt_duration=1192 frames, utt_pad_proportion=0.06978, over 10927.91 utterances.], batch size: 42, lr: 5.72e-03, grad_scale: 8.0 2023-03-08 18:45:56,178 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.016e+02 2.281e+02 2.809e+02 6.026e+02, threshold=4.562e+02, percent-clipped=5.0 2023-03-08 18:46:14,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 18:46:31,114 INFO [train2.py:809] (3/4) Epoch 19, batch 1050, loss[ctc_loss=0.09243, att_loss=0.2459, loss=0.2152, over 16943.00 frames. utt_duration=693 frames, utt_pad_proportion=0.1316, over 98.00 utterances.], tot_loss[ctc_loss=0.07746, att_loss=0.2375, loss=0.2055, over 3237071.68 frames. utt_duration=1223 frames, utt_pad_proportion=0.06571, over 10596.67 utterances.], batch size: 98, lr: 5.72e-03, grad_scale: 8.0 2023-03-08 18:47:13,811 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:47:19,347 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4212, 2.3664, 4.8015, 3.7859, 3.0082, 4.1045, 4.3429, 4.4361], device='cuda:3'), covar=tensor([0.0206, 0.1700, 0.0169, 0.0925, 0.1641, 0.0260, 0.0212, 0.0260], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0240, 0.0174, 0.0308, 0.0265, 0.0204, 0.0159, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:47:50,615 INFO [train2.py:809] (3/4) Epoch 19, batch 1100, loss[ctc_loss=0.06511, att_loss=0.232, loss=0.1987, over 16455.00 frames. utt_duration=1432 frames, utt_pad_proportion=0.007382, over 46.00 utterances.], tot_loss[ctc_loss=0.07818, att_loss=0.2381, loss=0.2061, over 3245773.29 frames. utt_duration=1234 frames, utt_pad_proportion=0.0613, over 10534.43 utterances.], batch size: 46, lr: 5.72e-03, grad_scale: 8.0 2023-03-08 18:48:30,794 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:48:35,940 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 1.909e+02 2.229e+02 2.585e+02 5.838e+02, threshold=4.459e+02, percent-clipped=1.0 2023-03-08 18:48:49,171 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-03-08 18:48:49,908 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6225, 5.8627, 5.2475, 5.6450, 5.4691, 5.0840, 5.3108, 4.9585], device='cuda:3'), covar=tensor([0.1236, 0.0896, 0.0929, 0.0767, 0.0939, 0.1554, 0.2304, 0.2437], device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0588, 0.0438, 0.0440, 0.0413, 0.0450, 0.0595, 0.0514], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 18:48:51,660 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6926, 2.3167, 2.7149, 2.3763, 3.0610, 2.7453, 2.6153, 3.1161], device='cuda:3'), covar=tensor([0.1484, 0.2981, 0.2020, 0.1970, 0.1156, 0.1087, 0.2411, 0.0924], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0116, 0.0114, 0.0103, 0.0113, 0.0099, 0.0123, 0.0090], device='cuda:3'), out_proj_covar=tensor([8.1081e-05, 8.9047e-05, 8.9277e-05, 7.9509e-05, 8.3293e-05, 7.8883e-05, 9.0764e-05, 7.2083e-05], device='cuda:3') 2023-03-08 18:49:10,573 INFO [train2.py:809] (3/4) Epoch 19, batch 1150, loss[ctc_loss=0.07318, att_loss=0.2376, loss=0.2047, over 16128.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005941, over 42.00 utterances.], tot_loss[ctc_loss=0.07839, att_loss=0.2382, loss=0.2063, over 3252569.80 frames. utt_duration=1230 frames, utt_pad_proportion=0.06193, over 10593.36 utterances.], batch size: 42, lr: 5.72e-03, grad_scale: 8.0 2023-03-08 18:49:18,637 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-08 18:50:08,410 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:50:30,470 INFO [train2.py:809] (3/4) Epoch 19, batch 1200, loss[ctc_loss=0.1138, att_loss=0.2566, loss=0.228, over 13983.00 frames. utt_duration=384.6 frames, utt_pad_proportion=0.3276, over 146.00 utterances.], tot_loss[ctc_loss=0.07837, att_loss=0.2375, loss=0.2057, over 3237774.24 frames. utt_duration=1221 frames, utt_pad_proportion=0.06791, over 10617.09 utterances.], batch size: 146, lr: 5.71e-03, grad_scale: 8.0 2023-03-08 18:50:49,209 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6693, 3.0891, 3.7454, 3.1346, 3.6240, 4.6963, 4.5627, 3.5236], device='cuda:3'), covar=tensor([0.0338, 0.1590, 0.1127, 0.1378, 0.1034, 0.0917, 0.0509, 0.1146], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0241, 0.0274, 0.0215, 0.0259, 0.0355, 0.0252, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:51:15,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 1.921e+02 2.301e+02 2.976e+02 5.020e+02, threshold=4.601e+02, percent-clipped=3.0 2023-03-08 18:51:41,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2023-03-08 18:51:45,240 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:51:50,079 INFO [train2.py:809] (3/4) Epoch 19, batch 1250, loss[ctc_loss=0.06885, att_loss=0.2409, loss=0.2065, over 16769.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006525, over 48.00 utterances.], tot_loss[ctc_loss=0.07869, att_loss=0.2377, loss=0.2059, over 3237630.11 frames. utt_duration=1241 frames, utt_pad_proportion=0.06255, over 10452.01 utterances.], batch size: 48, lr: 5.71e-03, grad_scale: 8.0 2023-03-08 18:51:59,973 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 18:52:28,333 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:52:48,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-08 18:52:51,472 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0207, 5.0360, 4.8637, 2.3317, 1.9283, 2.7520, 2.3107, 3.6983], device='cuda:3'), covar=tensor([0.0744, 0.0231, 0.0250, 0.4748, 0.6005, 0.2686, 0.3772, 0.1836], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0269, 0.0266, 0.0243, 0.0348, 0.0338, 0.0254, 0.0365], device='cuda:3'), out_proj_covar=tensor([1.5133e-04, 9.9706e-05, 1.1312e-04, 1.0542e-04, 1.4631e-04, 1.3252e-04, 1.0187e-04, 1.4963e-04], device='cuda:3') 2023-03-08 18:52:56,970 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1631, 4.4407, 4.3628, 4.4557, 4.5145, 4.2918, 3.2989, 4.3536], device='cuda:3'), covar=tensor([0.0118, 0.0124, 0.0124, 0.0077, 0.0083, 0.0099, 0.0636, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0086, 0.0108, 0.0066, 0.0072, 0.0083, 0.0103, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 18:53:08,948 INFO [train2.py:809] (3/4) Epoch 19, batch 1300, loss[ctc_loss=0.07583, att_loss=0.2167, loss=0.1885, over 15509.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008226, over 36.00 utterances.], tot_loss[ctc_loss=0.07824, att_loss=0.2368, loss=0.2051, over 3241223.54 frames. utt_duration=1268 frames, utt_pad_proportion=0.05546, over 10237.62 utterances.], batch size: 36, lr: 5.71e-03, grad_scale: 8.0 2023-03-08 18:53:37,545 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 18:53:45,285 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:53:50,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 18:53:55,193 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 2.017e+02 2.532e+02 3.108e+02 5.795e+02, threshold=5.064e+02, percent-clipped=4.0 2023-03-08 18:54:30,469 INFO [train2.py:809] (3/4) Epoch 19, batch 1350, loss[ctc_loss=0.06182, att_loss=0.2254, loss=0.1927, over 14517.00 frames. utt_duration=1816 frames, utt_pad_proportion=0.03547, over 32.00 utterances.], tot_loss[ctc_loss=0.07717, att_loss=0.2363, loss=0.2045, over 3249213.47 frames. utt_duration=1286 frames, utt_pad_proportion=0.04929, over 10117.59 utterances.], batch size: 32, lr: 5.71e-03, grad_scale: 8.0 2023-03-08 18:55:50,794 INFO [train2.py:809] (3/4) Epoch 19, batch 1400, loss[ctc_loss=0.07916, att_loss=0.2426, loss=0.2099, over 16956.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008123, over 50.00 utterances.], tot_loss[ctc_loss=0.07618, att_loss=0.2355, loss=0.2036, over 3252238.42 frames. utt_duration=1311 frames, utt_pad_proportion=0.04321, over 9931.68 utterances.], batch size: 50, lr: 5.71e-03, grad_scale: 8.0 2023-03-08 18:55:53,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.21 vs. limit=5.0 2023-03-08 18:56:32,803 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5798, 4.5814, 4.7148, 4.6220, 5.2394, 4.5163, 4.5532, 2.6648], device='cuda:3'), covar=tensor([0.0213, 0.0365, 0.0229, 0.0283, 0.0917, 0.0222, 0.0317, 0.1806], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0168, 0.0170, 0.0186, 0.0356, 0.0143, 0.0158, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 18:56:35,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 2.120e+02 2.397e+02 2.926e+02 6.281e+02, threshold=4.794e+02, percent-clipped=3.0 2023-03-08 18:56:46,470 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9180, 3.5932, 3.6055, 3.1501, 3.7474, 3.7724, 3.7410, 2.8286], device='cuda:3'), covar=tensor([0.0767, 0.1430, 0.2391, 0.3379, 0.0944, 0.2308, 0.0810, 0.3939], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0170, 0.0185, 0.0244, 0.0147, 0.0242, 0.0162, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 18:57:10,510 INFO [train2.py:809] (3/4) Epoch 19, batch 1450, loss[ctc_loss=0.1015, att_loss=0.2602, loss=0.2284, over 17103.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.0159, over 56.00 utterances.], tot_loss[ctc_loss=0.07661, att_loss=0.2362, loss=0.2042, over 3260836.85 frames. utt_duration=1318 frames, utt_pad_proportion=0.03911, over 9906.85 utterances.], batch size: 56, lr: 5.70e-03, grad_scale: 8.0 2023-03-08 18:57:14,578 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-08 18:58:06,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.20 vs. limit=5.0 2023-03-08 18:58:30,640 INFO [train2.py:809] (3/4) Epoch 19, batch 1500, loss[ctc_loss=0.06543, att_loss=0.2205, loss=0.1895, over 16185.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.005464, over 41.00 utterances.], tot_loss[ctc_loss=0.07717, att_loss=0.2371, loss=0.2051, over 3267388.49 frames. utt_duration=1309 frames, utt_pad_proportion=0.04073, over 9998.96 utterances.], batch size: 41, lr: 5.70e-03, grad_scale: 8.0 2023-03-08 18:58:51,321 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1206, 5.4522, 5.3705, 5.2818, 5.4389, 5.3685, 5.1074, 4.8230], device='cuda:3'), covar=tensor([0.0975, 0.0470, 0.0315, 0.0568, 0.0280, 0.0320, 0.0392, 0.0343], device='cuda:3'), in_proj_covar=tensor([0.0508, 0.0342, 0.0327, 0.0343, 0.0402, 0.0414, 0.0343, 0.0378], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 18:59:15,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.048e+02 2.505e+02 3.116e+02 1.439e+03, threshold=5.009e+02, percent-clipped=5.0 2023-03-08 18:59:36,933 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 18:59:48,611 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1636, 5.4325, 5.4007, 5.2972, 5.4488, 5.3976, 5.1181, 4.8472], device='cuda:3'), covar=tensor([0.1012, 0.0473, 0.0266, 0.0536, 0.0270, 0.0292, 0.0383, 0.0309], device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0343, 0.0328, 0.0344, 0.0403, 0.0415, 0.0344, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 18:59:49,905 INFO [train2.py:809] (3/4) Epoch 19, batch 1550, loss[ctc_loss=0.05079, att_loss=0.2196, loss=0.1859, over 16009.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007234, over 40.00 utterances.], tot_loss[ctc_loss=0.07808, att_loss=0.2372, loss=0.2054, over 3256512.16 frames. utt_duration=1283 frames, utt_pad_proportion=0.05003, over 10161.58 utterances.], batch size: 40, lr: 5.70e-03, grad_scale: 8.0 2023-03-08 19:00:45,892 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:01:09,931 INFO [train2.py:809] (3/4) Epoch 19, batch 1600, loss[ctc_loss=0.05753, att_loss=0.2189, loss=0.1866, over 16170.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007435, over 41.00 utterances.], tot_loss[ctc_loss=0.07763, att_loss=0.2375, loss=0.2055, over 3265223.68 frames. utt_duration=1276 frames, utt_pad_proportion=0.04935, over 10248.03 utterances.], batch size: 41, lr: 5.70e-03, grad_scale: 8.0 2023-03-08 19:01:29,354 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 19:01:36,416 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-08 19:01:54,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.054e+02 2.356e+02 2.874e+02 5.257e+02, threshold=4.713e+02, percent-clipped=1.0 2023-03-08 19:02:23,073 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:02:29,413 INFO [train2.py:809] (3/4) Epoch 19, batch 1650, loss[ctc_loss=0.05266, att_loss=0.2038, loss=0.1736, over 15353.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01123, over 35.00 utterances.], tot_loss[ctc_loss=0.07776, att_loss=0.2381, loss=0.206, over 3263493.83 frames. utt_duration=1267 frames, utt_pad_proportion=0.05183, over 10314.65 utterances.], batch size: 35, lr: 5.70e-03, grad_scale: 8.0 2023-03-08 19:03:48,205 INFO [train2.py:809] (3/4) Epoch 19, batch 1700, loss[ctc_loss=0.06536, att_loss=0.2356, loss=0.2016, over 16395.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007812, over 44.00 utterances.], tot_loss[ctc_loss=0.07759, att_loss=0.2375, loss=0.2055, over 3263991.71 frames. utt_duration=1283 frames, utt_pad_proportion=0.0482, over 10190.58 utterances.], batch size: 44, lr: 5.69e-03, grad_scale: 8.0 2023-03-08 19:04:32,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.058e+02 2.463e+02 2.943e+02 5.957e+02, threshold=4.927e+02, percent-clipped=3.0 2023-03-08 19:05:06,958 INFO [train2.py:809] (3/4) Epoch 19, batch 1750, loss[ctc_loss=0.08812, att_loss=0.2568, loss=0.2231, over 17473.00 frames. utt_duration=1014 frames, utt_pad_proportion=0.04307, over 69.00 utterances.], tot_loss[ctc_loss=0.07795, att_loss=0.2379, loss=0.2059, over 3272208.86 frames. utt_duration=1270 frames, utt_pad_proportion=0.04877, over 10317.70 utterances.], batch size: 69, lr: 5.69e-03, grad_scale: 8.0 2023-03-08 19:05:25,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-08 19:06:02,108 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0481, 5.0664, 4.8090, 2.8457, 4.9026, 4.6773, 4.3297, 2.7009], device='cuda:3'), covar=tensor([0.0122, 0.0097, 0.0275, 0.1042, 0.0093, 0.0211, 0.0299, 0.1444], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0099, 0.0099, 0.0109, 0.0082, 0.0109, 0.0097, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:06:25,923 INFO [train2.py:809] (3/4) Epoch 19, batch 1800, loss[ctc_loss=0.05872, att_loss=0.2101, loss=0.1798, over 13260.00 frames. utt_duration=1831 frames, utt_pad_proportion=0.1056, over 29.00 utterances.], tot_loss[ctc_loss=0.07722, att_loss=0.2367, loss=0.2048, over 3266179.19 frames. utt_duration=1299 frames, utt_pad_proportion=0.04332, over 10069.58 utterances.], batch size: 29, lr: 5.69e-03, grad_scale: 8.0 2023-03-08 19:06:52,204 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7002, 5.0545, 4.6023, 5.1188, 4.5151, 4.6293, 5.2044, 4.9778], device='cuda:3'), covar=tensor([0.0689, 0.0327, 0.0904, 0.0337, 0.0492, 0.0363, 0.0224, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0309, 0.0361, 0.0330, 0.0313, 0.0232, 0.0293, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 19:07:10,632 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.911e+02 2.242e+02 2.883e+02 9.834e+02, threshold=4.485e+02, percent-clipped=3.0 2023-03-08 19:07:33,092 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:07:45,874 INFO [train2.py:809] (3/4) Epoch 19, batch 1850, loss[ctc_loss=0.06732, att_loss=0.2296, loss=0.1972, over 16383.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.008659, over 44.00 utterances.], tot_loss[ctc_loss=0.07744, att_loss=0.2375, loss=0.2055, over 3267155.75 frames. utt_duration=1246 frames, utt_pad_proportion=0.05614, over 10504.91 utterances.], batch size: 44, lr: 5.69e-03, grad_scale: 8.0 2023-03-08 19:08:36,626 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:08:49,393 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:08:56,797 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-08 19:08:57,784 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6854, 2.3464, 5.0318, 4.0204, 3.1162, 4.4430, 4.8706, 4.7344], device='cuda:3'), covar=tensor([0.0165, 0.1597, 0.0143, 0.0784, 0.1545, 0.0159, 0.0090, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0241, 0.0175, 0.0310, 0.0266, 0.0205, 0.0161, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 19:09:04,979 INFO [train2.py:809] (3/4) Epoch 19, batch 1900, loss[ctc_loss=0.06207, att_loss=0.207, loss=0.178, over 16010.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007156, over 40.00 utterances.], tot_loss[ctc_loss=0.07805, att_loss=0.2378, loss=0.2059, over 3270087.09 frames. utt_duration=1223 frames, utt_pad_proportion=0.06109, over 10705.43 utterances.], batch size: 40, lr: 5.69e-03, grad_scale: 8.0 2023-03-08 19:09:23,770 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 19:09:48,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 1.956e+02 2.391e+02 2.954e+02 5.478e+02, threshold=4.781e+02, percent-clipped=4.0 2023-03-08 19:10:10,685 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:10:14,010 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:10:24,369 INFO [train2.py:809] (3/4) Epoch 19, batch 1950, loss[ctc_loss=0.0659, att_loss=0.2195, loss=0.1888, over 15863.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009913, over 39.00 utterances.], tot_loss[ctc_loss=0.07824, att_loss=0.2378, loss=0.2059, over 3272171.17 frames. utt_duration=1233 frames, utt_pad_proportion=0.05939, over 10631.52 utterances.], batch size: 39, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:10:40,462 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 19:11:31,537 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0503, 5.3059, 5.1952, 5.2032, 5.3241, 5.3087, 5.0271, 4.7874], device='cuda:3'), covar=tensor([0.0958, 0.0496, 0.0299, 0.0464, 0.0296, 0.0305, 0.0341, 0.0303], device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0339, 0.0322, 0.0337, 0.0396, 0.0407, 0.0337, 0.0374], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 19:11:43,431 INFO [train2.py:809] (3/4) Epoch 19, batch 2000, loss[ctc_loss=0.09405, att_loss=0.2488, loss=0.2178, over 17121.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01418, over 56.00 utterances.], tot_loss[ctc_loss=0.07853, att_loss=0.2383, loss=0.2064, over 3279826.79 frames. utt_duration=1247 frames, utt_pad_proportion=0.05426, over 10531.45 utterances.], batch size: 56, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:11:48,920 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1871, 5.1496, 4.9395, 2.8622, 4.9587, 4.8676, 4.4032, 2.8468], device='cuda:3'), covar=tensor([0.0097, 0.0091, 0.0274, 0.0986, 0.0090, 0.0167, 0.0284, 0.1252], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0099, 0.0099, 0.0109, 0.0082, 0.0109, 0.0097, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:12:22,535 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:12:28,300 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 2.150e+02 2.565e+02 3.312e+02 1.867e+03, threshold=5.130e+02, percent-clipped=8.0 2023-03-08 19:13:03,766 INFO [train2.py:809] (3/4) Epoch 19, batch 2050, loss[ctc_loss=0.07826, att_loss=0.2341, loss=0.2029, over 16122.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006618, over 42.00 utterances.], tot_loss[ctc_loss=0.07891, att_loss=0.2386, loss=0.2067, over 3279833.04 frames. utt_duration=1215 frames, utt_pad_proportion=0.06035, over 10810.87 utterances.], batch size: 42, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:13:10,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.39 vs. limit=5.0 2023-03-08 19:13:42,165 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2023-03-08 19:13:59,679 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:14:24,526 INFO [train2.py:809] (3/4) Epoch 19, batch 2100, loss[ctc_loss=0.09054, att_loss=0.2543, loss=0.2215, over 16333.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005913, over 45.00 utterances.], tot_loss[ctc_loss=0.07986, att_loss=0.2391, loss=0.2072, over 3269019.96 frames. utt_duration=1179 frames, utt_pad_proportion=0.07278, over 11103.46 utterances.], batch size: 45, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:14:52,287 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0520, 4.3676, 4.2893, 4.5304, 2.7746, 4.3839, 2.7314, 1.9173], device='cuda:3'), covar=tensor([0.0402, 0.0264, 0.0784, 0.0260, 0.1775, 0.0211, 0.1554, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0152, 0.0257, 0.0149, 0.0221, 0.0132, 0.0229, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:15:08,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 2.020e+02 2.356e+02 2.940e+02 5.527e+02, threshold=4.712e+02, percent-clipped=4.0 2023-03-08 19:15:43,845 INFO [train2.py:809] (3/4) Epoch 19, batch 2150, loss[ctc_loss=0.07668, att_loss=0.239, loss=0.2065, over 16628.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005289, over 47.00 utterances.], tot_loss[ctc_loss=0.08005, att_loss=0.2388, loss=0.2071, over 3263732.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.0747, over 11082.85 utterances.], batch size: 47, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:17:03,552 INFO [train2.py:809] (3/4) Epoch 19, batch 2200, loss[ctc_loss=0.07562, att_loss=0.2521, loss=0.2168, over 16758.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006426, over 48.00 utterances.], tot_loss[ctc_loss=0.08008, att_loss=0.2394, loss=0.2075, over 3274485.80 frames. utt_duration=1187 frames, utt_pad_proportion=0.06957, over 11044.86 utterances.], batch size: 48, lr: 5.68e-03, grad_scale: 8.0 2023-03-08 19:17:43,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-08 19:17:47,492 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.057e+02 2.535e+02 3.122e+02 7.194e+02, threshold=5.070e+02, percent-clipped=8.0 2023-03-08 19:18:03,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2023-03-08 19:18:04,656 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:18:09,408 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:18:10,185 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2023-03-08 19:18:23,578 INFO [train2.py:809] (3/4) Epoch 19, batch 2250, loss[ctc_loss=0.08494, att_loss=0.2423, loss=0.2109, over 16553.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005552, over 45.00 utterances.], tot_loss[ctc_loss=0.07923, att_loss=0.2389, loss=0.207, over 3275031.78 frames. utt_duration=1224 frames, utt_pad_proportion=0.06022, over 10716.63 utterances.], batch size: 45, lr: 5.67e-03, grad_scale: 8.0 2023-03-08 19:18:41,628 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1102, 4.4495, 4.2450, 4.7097, 2.7068, 4.4794, 2.5442, 1.6280], device='cuda:3'), covar=tensor([0.0343, 0.0217, 0.0831, 0.0186, 0.1885, 0.0193, 0.1790, 0.2047], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0154, 0.0260, 0.0150, 0.0222, 0.0133, 0.0232, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:18:45,220 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.09 vs. limit=5.0 2023-03-08 19:19:15,899 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0368, 3.8990, 3.2412, 3.3281, 4.0399, 3.6868, 2.9042, 4.2087], device='cuda:3'), covar=tensor([0.1074, 0.0521, 0.1073, 0.0744, 0.0729, 0.0657, 0.0930, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0211, 0.0222, 0.0193, 0.0267, 0.0233, 0.0199, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 19:19:25,470 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:19:47,858 INFO [train2.py:809] (3/4) Epoch 19, batch 2300, loss[ctc_loss=0.07883, att_loss=0.2523, loss=0.2176, over 17135.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01406, over 56.00 utterances.], tot_loss[ctc_loss=0.07873, att_loss=0.2382, loss=0.2063, over 3267177.56 frames. utt_duration=1237 frames, utt_pad_proportion=0.05758, over 10579.09 utterances.], batch size: 56, lr: 5.67e-03, grad_scale: 8.0 2023-03-08 19:20:30,626 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 2.198e+02 2.569e+02 3.135e+02 6.426e+02, threshold=5.138e+02, percent-clipped=5.0 2023-03-08 19:21:05,899 INFO [train2.py:809] (3/4) Epoch 19, batch 2350, loss[ctc_loss=0.08129, att_loss=0.256, loss=0.2211, over 17046.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008642, over 52.00 utterances.], tot_loss[ctc_loss=0.07948, att_loss=0.2393, loss=0.2074, over 3277593.51 frames. utt_duration=1239 frames, utt_pad_proportion=0.05394, over 10591.13 utterances.], batch size: 52, lr: 5.67e-03, grad_scale: 8.0 2023-03-08 19:21:15,275 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:21:52,241 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:22:24,338 INFO [train2.py:809] (3/4) Epoch 19, batch 2400, loss[ctc_loss=0.09266, att_loss=0.2562, loss=0.2235, over 17251.00 frames. utt_duration=1171 frames, utt_pad_proportion=0.0263, over 59.00 utterances.], tot_loss[ctc_loss=0.07866, att_loss=0.2388, loss=0.2068, over 3280948.05 frames. utt_duration=1247 frames, utt_pad_proportion=0.0516, over 10539.93 utterances.], batch size: 59, lr: 5.67e-03, grad_scale: 16.0 2023-03-08 19:22:51,379 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:23:07,412 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.383e+02 2.133e+02 2.520e+02 3.216e+02 6.079e+02, threshold=5.039e+02, percent-clipped=1.0 2023-03-08 19:23:43,616 INFO [train2.py:809] (3/4) Epoch 19, batch 2450, loss[ctc_loss=0.0722, att_loss=0.238, loss=0.2049, over 17027.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007569, over 51.00 utterances.], tot_loss[ctc_loss=0.07855, att_loss=0.2389, loss=0.2068, over 3285934.51 frames. utt_duration=1250 frames, utt_pad_proportion=0.04954, over 10529.33 utterances.], batch size: 51, lr: 5.67e-03, grad_scale: 16.0 2023-03-08 19:23:48,510 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4643, 2.5997, 4.9055, 3.8900, 3.0718, 4.2714, 4.6663, 4.6187], device='cuda:3'), covar=tensor([0.0228, 0.1672, 0.0176, 0.0928, 0.1762, 0.0243, 0.0136, 0.0239], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0239, 0.0174, 0.0306, 0.0264, 0.0203, 0.0160, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:3') 2023-03-08 19:24:25,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-08 19:25:02,847 INFO [train2.py:809] (3/4) Epoch 19, batch 2500, loss[ctc_loss=0.06726, att_loss=0.2294, loss=0.1969, over 17047.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009992, over 53.00 utterances.], tot_loss[ctc_loss=0.07806, att_loss=0.2386, loss=0.2065, over 3284856.91 frames. utt_duration=1232 frames, utt_pad_proportion=0.05432, over 10675.35 utterances.], batch size: 53, lr: 5.66e-03, grad_scale: 16.0 2023-03-08 19:25:36,332 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1978, 5.4916, 5.7569, 5.5659, 5.6880, 6.1433, 5.3906, 6.2189], device='cuda:3'), covar=tensor([0.0713, 0.0687, 0.0830, 0.1302, 0.1756, 0.0870, 0.0627, 0.0647], device='cuda:3'), in_proj_covar=tensor([0.0836, 0.0490, 0.0578, 0.0639, 0.0851, 0.0601, 0.0477, 0.0585], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 19:25:40,233 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-08 19:25:46,746 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-08 19:25:47,411 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.096e+02 2.486e+02 2.922e+02 5.921e+02, threshold=4.971e+02, percent-clipped=2.0 2023-03-08 19:25:54,827 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:26:03,997 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:26:23,587 INFO [train2.py:809] (3/4) Epoch 19, batch 2550, loss[ctc_loss=0.06771, att_loss=0.2137, loss=0.1845, over 15753.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.009861, over 38.00 utterances.], tot_loss[ctc_loss=0.07769, att_loss=0.2382, loss=0.2061, over 3284672.88 frames. utt_duration=1245 frames, utt_pad_proportion=0.05176, over 10566.23 utterances.], batch size: 38, lr: 5.66e-03, grad_scale: 16.0 2023-03-08 19:26:36,010 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8441, 5.1177, 4.9899, 5.0884, 5.1275, 5.1234, 4.8241, 4.6503], device='cuda:3'), covar=tensor([0.1071, 0.0586, 0.0327, 0.0448, 0.0309, 0.0334, 0.0387, 0.0342], device='cuda:3'), in_proj_covar=tensor([0.0518, 0.0349, 0.0333, 0.0345, 0.0407, 0.0419, 0.0349, 0.0386], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 19:27:20,419 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:27:23,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9673, 4.0292, 3.7740, 2.7271, 3.8453, 3.8076, 3.6106, 2.8106], device='cuda:3'), covar=tensor([0.0114, 0.0122, 0.0264, 0.0932, 0.0115, 0.0383, 0.0308, 0.1168], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0100, 0.0101, 0.0111, 0.0084, 0.0111, 0.0098, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:27:32,406 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:27:43,385 INFO [train2.py:809] (3/4) Epoch 19, batch 2600, loss[ctc_loss=0.08651, att_loss=0.2593, loss=0.2247, over 17353.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03563, over 63.00 utterances.], tot_loss[ctc_loss=0.0773, att_loss=0.2379, loss=0.2057, over 3273098.82 frames. utt_duration=1229 frames, utt_pad_proportion=0.05897, over 10663.48 utterances.], batch size: 63, lr: 5.66e-03, grad_scale: 8.0 2023-03-08 19:28:10,367 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6183, 5.0145, 4.8720, 4.9726, 5.1490, 4.6827, 3.5915, 5.0073], device='cuda:3'), covar=tensor([0.0123, 0.0118, 0.0124, 0.0083, 0.0084, 0.0110, 0.0679, 0.0177], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0085, 0.0107, 0.0066, 0.0071, 0.0083, 0.0102, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 19:28:29,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.314e+02 1.890e+02 2.336e+02 2.791e+02 4.929e+02, threshold=4.673e+02, percent-clipped=0.0 2023-03-08 19:28:31,914 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-03-08 19:29:04,097 INFO [train2.py:809] (3/4) Epoch 19, batch 2650, loss[ctc_loss=0.05905, att_loss=0.2051, loss=0.1759, over 14499.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.03661, over 32.00 utterances.], tot_loss[ctc_loss=0.07684, att_loss=0.2376, loss=0.2054, over 3273445.52 frames. utt_duration=1237 frames, utt_pad_proportion=0.05596, over 10596.58 utterances.], batch size: 32, lr: 5.66e-03, grad_scale: 8.0 2023-03-08 19:29:04,260 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9262, 5.1371, 5.7089, 5.1348, 5.0261, 5.6148, 5.2874, 5.7674], device='cuda:3'), covar=tensor([0.1142, 0.1615, 0.1020, 0.2299, 0.3411, 0.1808, 0.0919, 0.1165], device='cuda:3'), in_proj_covar=tensor([0.0834, 0.0488, 0.0576, 0.0637, 0.0848, 0.0600, 0.0471, 0.0581], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 19:29:52,097 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:30:23,754 INFO [train2.py:809] (3/4) Epoch 19, batch 2700, loss[ctc_loss=0.08433, att_loss=0.2295, loss=0.2005, over 16276.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.006965, over 43.00 utterances.], tot_loss[ctc_loss=0.07677, att_loss=0.2365, loss=0.2045, over 3269598.44 frames. utt_duration=1257 frames, utt_pad_proportion=0.05303, over 10418.76 utterances.], batch size: 43, lr: 5.66e-03, grad_scale: 8.0 2023-03-08 19:30:42,821 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:31:09,481 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:31:10,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.123e+02 2.465e+02 2.927e+02 6.214e+02, threshold=4.930e+02, percent-clipped=3.0 2023-03-08 19:31:46,100 INFO [train2.py:809] (3/4) Epoch 19, batch 2750, loss[ctc_loss=0.06629, att_loss=0.2421, loss=0.2069, over 16935.00 frames. utt_duration=1356 frames, utt_pad_proportion=0.008816, over 50.00 utterances.], tot_loss[ctc_loss=0.07672, att_loss=0.2365, loss=0.2045, over 3276879.50 frames. utt_duration=1276 frames, utt_pad_proportion=0.0465, over 10283.80 utterances.], batch size: 50, lr: 5.65e-03, grad_scale: 8.0 2023-03-08 19:33:07,061 INFO [train2.py:809] (3/4) Epoch 19, batch 2800, loss[ctc_loss=0.1257, att_loss=0.2689, loss=0.2402, over 14227.00 frames. utt_duration=388.6 frames, utt_pad_proportion=0.3195, over 147.00 utterances.], tot_loss[ctc_loss=0.07762, att_loss=0.2374, loss=0.2055, over 3280643.69 frames. utt_duration=1252 frames, utt_pad_proportion=0.05096, over 10493.28 utterances.], batch size: 147, lr: 5.65e-03, grad_scale: 8.0 2023-03-08 19:33:53,761 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.053e+02 2.431e+02 2.775e+02 7.062e+02, threshold=4.862e+02, percent-clipped=1.0 2023-03-08 19:34:27,617 INFO [train2.py:809] (3/4) Epoch 19, batch 2850, loss[ctc_loss=0.07686, att_loss=0.2531, loss=0.2178, over 17406.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.03101, over 63.00 utterances.], tot_loss[ctc_loss=0.07732, att_loss=0.2369, loss=0.205, over 3279645.77 frames. utt_duration=1248 frames, utt_pad_proportion=0.05254, over 10526.50 utterances.], batch size: 63, lr: 5.65e-03, grad_scale: 8.0 2023-03-08 19:34:59,786 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0777, 5.4156, 4.9622, 5.4692, 4.8830, 5.0283, 5.5591, 5.3382], device='cuda:3'), covar=tensor([0.0648, 0.0261, 0.0853, 0.0316, 0.0401, 0.0236, 0.0220, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0306, 0.0356, 0.0328, 0.0311, 0.0232, 0.0292, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 19:35:27,841 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:35:29,850 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:35:48,361 INFO [train2.py:809] (3/4) Epoch 19, batch 2900, loss[ctc_loss=0.06174, att_loss=0.2239, loss=0.1915, over 15962.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.005973, over 41.00 utterances.], tot_loss[ctc_loss=0.07726, att_loss=0.2372, loss=0.2052, over 3274838.89 frames. utt_duration=1231 frames, utt_pad_proportion=0.05773, over 10654.88 utterances.], batch size: 41, lr: 5.65e-03, grad_scale: 8.0 2023-03-08 19:36:34,200 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 1.992e+02 2.354e+02 2.922e+02 1.423e+03, threshold=4.708e+02, percent-clipped=1.0 2023-03-08 19:36:42,302 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6079, 5.0083, 4.9542, 4.9627, 5.0454, 4.7808, 3.3966, 4.9889], device='cuda:3'), covar=tensor([0.0122, 0.0159, 0.0121, 0.0088, 0.0108, 0.0107, 0.0797, 0.0219], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0085, 0.0107, 0.0066, 0.0071, 0.0083, 0.0102, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 19:37:06,223 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:37:08,934 INFO [train2.py:809] (3/4) Epoch 19, batch 2950, loss[ctc_loss=0.08878, att_loss=0.2323, loss=0.2036, over 15757.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009342, over 38.00 utterances.], tot_loss[ctc_loss=0.07673, att_loss=0.2365, loss=0.2045, over 3277979.15 frames. utt_duration=1261 frames, utt_pad_proportion=0.05001, over 10411.15 utterances.], batch size: 38, lr: 5.65e-03, grad_scale: 8.0 2023-03-08 19:37:51,938 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4747, 2.8409, 3.1928, 4.5129, 4.1659, 4.0695, 3.3049, 2.2172], device='cuda:3'), covar=tensor([0.0692, 0.2125, 0.1335, 0.0629, 0.0700, 0.0511, 0.1116, 0.2327], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0213, 0.0188, 0.0214, 0.0219, 0.0175, 0.0199, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:37:56,665 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8488, 6.1040, 5.6085, 5.7797, 5.7863, 5.3467, 5.5551, 5.2679], device='cuda:3'), covar=tensor([0.1305, 0.0827, 0.0854, 0.0855, 0.0930, 0.1380, 0.2168, 0.2494], device='cuda:3'), in_proj_covar=tensor([0.0514, 0.0598, 0.0451, 0.0446, 0.0420, 0.0459, 0.0606, 0.0520], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:38:28,665 INFO [train2.py:809] (3/4) Epoch 19, batch 3000, loss[ctc_loss=0.07667, att_loss=0.2372, loss=0.2051, over 16001.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007745, over 40.00 utterances.], tot_loss[ctc_loss=0.07723, att_loss=0.2368, loss=0.2049, over 3274804.72 frames. utt_duration=1262 frames, utt_pad_proportion=0.05093, over 10394.48 utterances.], batch size: 40, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:38:28,666 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 19:38:42,983 INFO [train2.py:843] (3/4) Epoch 19, validation: ctc_loss=0.04253, att_loss=0.235, loss=0.1965, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 19:38:42,984 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 19:38:53,017 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2674, 4.4572, 4.3795, 4.7135, 2.6066, 4.4045, 2.4087, 1.8493], device='cuda:3'), covar=tensor([0.0332, 0.0212, 0.0752, 0.0185, 0.1839, 0.0184, 0.1874, 0.1754], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0153, 0.0253, 0.0148, 0.0216, 0.0132, 0.0226, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:39:00,509 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7681, 6.0207, 5.4527, 5.7381, 5.6635, 5.2905, 5.4966, 5.2555], device='cuda:3'), covar=tensor([0.1103, 0.0847, 0.0820, 0.0834, 0.1011, 0.1344, 0.2206, 0.2345], device='cuda:3'), in_proj_covar=tensor([0.0508, 0.0590, 0.0447, 0.0443, 0.0416, 0.0455, 0.0601, 0.0514], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:39:02,217 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:39:29,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.054e+02 2.397e+02 2.986e+02 9.234e+02, threshold=4.795e+02, percent-clipped=4.0 2023-03-08 19:40:05,303 INFO [train2.py:809] (3/4) Epoch 19, batch 3050, loss[ctc_loss=0.09078, att_loss=0.255, loss=0.2221, over 17131.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.0146, over 56.00 utterances.], tot_loss[ctc_loss=0.07742, att_loss=0.2374, loss=0.2054, over 3282907.58 frames. utt_duration=1258 frames, utt_pad_proportion=0.0493, over 10448.37 utterances.], batch size: 56, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:40:20,715 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:41:26,172 INFO [train2.py:809] (3/4) Epoch 19, batch 3100, loss[ctc_loss=0.07569, att_loss=0.2322, loss=0.2009, over 16008.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007141, over 40.00 utterances.], tot_loss[ctc_loss=0.0773, att_loss=0.2376, loss=0.2055, over 3286138.52 frames. utt_duration=1245 frames, utt_pad_proportion=0.05111, over 10573.53 utterances.], batch size: 40, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:42:11,542 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 1.971e+02 2.423e+02 3.077e+02 6.495e+02, threshold=4.846e+02, percent-clipped=7.0 2023-03-08 19:42:12,759 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-08 19:42:46,968 INFO [train2.py:809] (3/4) Epoch 19, batch 3150, loss[ctc_loss=0.0846, att_loss=0.258, loss=0.2233, over 17273.00 frames. utt_duration=876.1 frames, utt_pad_proportion=0.08264, over 79.00 utterances.], tot_loss[ctc_loss=0.07711, att_loss=0.2375, loss=0.2054, over 3294780.90 frames. utt_duration=1260 frames, utt_pad_proportion=0.04483, over 10468.51 utterances.], batch size: 79, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:42:47,905 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.03 vs. limit=5.0 2023-03-08 19:43:02,714 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9407, 3.9739, 3.3427, 3.5865, 4.0966, 3.7537, 3.1582, 4.3985], device='cuda:3'), covar=tensor([0.1029, 0.0468, 0.0936, 0.0575, 0.0634, 0.0609, 0.0802, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0211, 0.0223, 0.0193, 0.0267, 0.0233, 0.0198, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-08 19:43:47,035 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:44:06,758 INFO [train2.py:809] (3/4) Epoch 19, batch 3200, loss[ctc_loss=0.06441, att_loss=0.2231, loss=0.1914, over 16195.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006025, over 41.00 utterances.], tot_loss[ctc_loss=0.07648, att_loss=0.2364, loss=0.2044, over 3284548.59 frames. utt_duration=1290 frames, utt_pad_proportion=0.04023, over 10194.56 utterances.], batch size: 41, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:44:08,705 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:44:51,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 1.907e+02 2.285e+02 3.012e+02 5.502e+02, threshold=4.570e+02, percent-clipped=3.0 2023-03-08 19:45:03,893 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:45:15,413 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:45:26,715 INFO [train2.py:809] (3/4) Epoch 19, batch 3250, loss[ctc_loss=0.1001, att_loss=0.2545, loss=0.2236, over 17227.00 frames. utt_duration=873.7 frames, utt_pad_proportion=0.08605, over 79.00 utterances.], tot_loss[ctc_loss=0.0777, att_loss=0.2372, loss=0.2053, over 3284629.25 frames. utt_duration=1260 frames, utt_pad_proportion=0.04725, over 10441.65 utterances.], batch size: 79, lr: 5.64e-03, grad_scale: 8.0 2023-03-08 19:45:36,420 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:45:45,819 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:46:46,971 INFO [train2.py:809] (3/4) Epoch 19, batch 3300, loss[ctc_loss=0.09649, att_loss=0.2605, loss=0.2277, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006732, over 48.00 utterances.], tot_loss[ctc_loss=0.07732, att_loss=0.2369, loss=0.205, over 3276407.03 frames. utt_duration=1252 frames, utt_pad_proportion=0.05209, over 10479.11 utterances.], batch size: 48, lr: 5.63e-03, grad_scale: 8.0 2023-03-08 19:46:57,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-08 19:47:13,539 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:47:31,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.356e+02 2.051e+02 2.410e+02 2.967e+02 4.688e+02, threshold=4.820e+02, percent-clipped=1.0 2023-03-08 19:48:03,131 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-08 19:48:06,390 INFO [train2.py:809] (3/4) Epoch 19, batch 3350, loss[ctc_loss=0.0775, att_loss=0.2454, loss=0.2118, over 17475.00 frames. utt_duration=1015 frames, utt_pad_proportion=0.04284, over 69.00 utterances.], tot_loss[ctc_loss=0.07755, att_loss=0.2374, loss=0.2055, over 3281783.40 frames. utt_duration=1274 frames, utt_pad_proportion=0.04511, over 10312.77 utterances.], batch size: 69, lr: 5.63e-03, grad_scale: 8.0 2023-03-08 19:48:28,644 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6780, 3.4786, 3.4950, 3.0136, 3.5001, 3.4879, 3.5177, 2.4049], device='cuda:3'), covar=tensor([0.1101, 0.1631, 0.1905, 0.3569, 0.2169, 0.2681, 0.1295, 0.4680], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0172, 0.0187, 0.0246, 0.0148, 0.0248, 0.0164, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:49:26,189 INFO [train2.py:809] (3/4) Epoch 19, batch 3400, loss[ctc_loss=0.07818, att_loss=0.2509, loss=0.2164, over 17122.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01419, over 56.00 utterances.], tot_loss[ctc_loss=0.07694, att_loss=0.2371, loss=0.205, over 3279722.32 frames. utt_duration=1274 frames, utt_pad_proportion=0.04557, over 10311.51 utterances.], batch size: 56, lr: 5.63e-03, grad_scale: 8.0 2023-03-08 19:49:54,566 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9390, 5.2250, 5.4630, 5.2898, 5.4352, 5.8393, 5.1890, 5.9937], device='cuda:3'), covar=tensor([0.0728, 0.0660, 0.0815, 0.1373, 0.1678, 0.0993, 0.0729, 0.0642], device='cuda:3'), in_proj_covar=tensor([0.0851, 0.0501, 0.0589, 0.0651, 0.0866, 0.0612, 0.0485, 0.0592], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 19:50:11,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 2.118e+02 2.457e+02 2.938e+02 6.241e+02, threshold=4.915e+02, percent-clipped=4.0 2023-03-08 19:50:46,057 INFO [train2.py:809] (3/4) Epoch 19, batch 3450, loss[ctc_loss=0.07408, att_loss=0.2119, loss=0.1843, over 15631.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009167, over 37.00 utterances.], tot_loss[ctc_loss=0.07736, att_loss=0.237, loss=0.2051, over 3276338.87 frames. utt_duration=1269 frames, utt_pad_proportion=0.04795, over 10338.28 utterances.], batch size: 37, lr: 5.63e-03, grad_scale: 8.0 2023-03-08 19:52:06,158 INFO [train2.py:809] (3/4) Epoch 19, batch 3500, loss[ctc_loss=0.09155, att_loss=0.2406, loss=0.2108, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005893, over 48.00 utterances.], tot_loss[ctc_loss=0.07742, att_loss=0.2372, loss=0.2053, over 3276687.62 frames. utt_duration=1258 frames, utt_pad_proportion=0.0512, over 10429.67 utterances.], batch size: 48, lr: 5.63e-03, grad_scale: 8.0 2023-03-08 19:52:15,831 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4493, 4.5171, 4.5760, 4.6226, 5.2091, 4.5266, 4.4870, 2.3385], device='cuda:3'), covar=tensor([0.0229, 0.0343, 0.0353, 0.0314, 0.0902, 0.0220, 0.0346, 0.1964], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0175, 0.0181, 0.0195, 0.0364, 0.0149, 0.0167, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:52:19,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-08 19:52:38,294 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:52:52,420 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.026e+02 2.447e+02 3.089e+02 8.845e+02, threshold=4.893e+02, percent-clipped=4.0 2023-03-08 19:53:15,796 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:53:26,630 INFO [train2.py:809] (3/4) Epoch 19, batch 3550, loss[ctc_loss=0.07984, att_loss=0.2332, loss=0.2025, over 16543.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006272, over 45.00 utterances.], tot_loss[ctc_loss=0.0768, att_loss=0.2368, loss=0.2048, over 3275778.82 frames. utt_duration=1251 frames, utt_pad_proportion=0.05459, over 10490.02 utterances.], batch size: 45, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 19:53:37,486 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:53:45,219 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:54:15,816 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:54:23,215 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5306, 2.9583, 3.4419, 4.6044, 4.0659, 4.0787, 3.1132, 2.3207], device='cuda:3'), covar=tensor([0.0664, 0.2008, 0.0882, 0.0459, 0.0739, 0.0420, 0.1442, 0.2092], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0212, 0.0187, 0.0212, 0.0218, 0.0174, 0.0201, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 19:54:32,166 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:54:45,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.13 vs. limit=5.0 2023-03-08 19:54:46,285 INFO [train2.py:809] (3/4) Epoch 19, batch 3600, loss[ctc_loss=0.1007, att_loss=0.2591, loss=0.2274, over 17290.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02268, over 59.00 utterances.], tot_loss[ctc_loss=0.07807, att_loss=0.2375, loss=0.2056, over 3272161.46 frames. utt_duration=1236 frames, utt_pad_proportion=0.05981, over 10602.58 utterances.], batch size: 59, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 19:55:04,971 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 19:55:04,998 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9455, 5.2021, 5.1070, 5.0649, 5.1902, 5.1444, 4.8919, 4.6553], device='cuda:3'), covar=tensor([0.1060, 0.0481, 0.0323, 0.0532, 0.0301, 0.0361, 0.0394, 0.0353], device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0348, 0.0330, 0.0345, 0.0401, 0.0417, 0.0345, 0.0382], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 19:55:14,940 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5975, 5.8956, 5.3374, 5.6455, 5.5617, 5.1673, 5.3077, 5.0492], device='cuda:3'), covar=tensor([0.1168, 0.0829, 0.0911, 0.0809, 0.0864, 0.1475, 0.2332, 0.2593], device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0594, 0.0447, 0.0445, 0.0419, 0.0461, 0.0601, 0.0519], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:55:23,542 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 19:55:32,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.965e+02 2.324e+02 2.872e+02 5.380e+02, threshold=4.648e+02, percent-clipped=3.0 2023-03-08 19:56:05,134 INFO [train2.py:809] (3/4) Epoch 19, batch 3650, loss[ctc_loss=0.08582, att_loss=0.2455, loss=0.2135, over 16769.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.00654, over 48.00 utterances.], tot_loss[ctc_loss=0.07809, att_loss=0.2371, loss=0.2053, over 3266800.05 frames. utt_duration=1238 frames, utt_pad_proportion=0.06125, over 10569.68 utterances.], batch size: 48, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 19:57:25,712 INFO [train2.py:809] (3/4) Epoch 19, batch 3700, loss[ctc_loss=0.06686, att_loss=0.2379, loss=0.2037, over 16770.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006362, over 48.00 utterances.], tot_loss[ctc_loss=0.07824, att_loss=0.2377, loss=0.2058, over 3263518.54 frames. utt_duration=1225 frames, utt_pad_proportion=0.06252, over 10667.06 utterances.], batch size: 48, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 19:57:38,045 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9093, 6.1832, 5.6530, 5.8564, 5.8441, 5.3550, 5.5092, 5.3088], device='cuda:3'), covar=tensor([0.1059, 0.0814, 0.0810, 0.0929, 0.0716, 0.1551, 0.2269, 0.2297], device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0590, 0.0445, 0.0443, 0.0416, 0.0457, 0.0597, 0.0517], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 19:58:11,213 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 2.064e+02 2.535e+02 3.260e+02 9.270e+02, threshold=5.071e+02, percent-clipped=6.0 2023-03-08 19:58:44,489 INFO [train2.py:809] (3/4) Epoch 19, batch 3750, loss[ctc_loss=0.073, att_loss=0.2452, loss=0.2108, over 17349.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02182, over 59.00 utterances.], tot_loss[ctc_loss=0.07802, att_loss=0.2376, loss=0.2057, over 3257261.44 frames. utt_duration=1218 frames, utt_pad_proportion=0.06523, over 10714.51 utterances.], batch size: 59, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 19:59:01,006 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-08 19:59:48,633 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1017, 5.1505, 4.8613, 3.0511, 4.9310, 4.8020, 4.4180, 2.9850], device='cuda:3'), covar=tensor([0.0129, 0.0093, 0.0277, 0.0921, 0.0090, 0.0171, 0.0282, 0.1254], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0100, 0.0101, 0.0111, 0.0084, 0.0110, 0.0098, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:00:03,974 INFO [train2.py:809] (3/4) Epoch 19, batch 3800, loss[ctc_loss=0.08261, att_loss=0.2491, loss=0.2158, over 16476.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006041, over 46.00 utterances.], tot_loss[ctc_loss=0.0774, att_loss=0.2371, loss=0.2051, over 3256337.33 frames. utt_duration=1247 frames, utt_pad_proportion=0.05756, over 10461.57 utterances.], batch size: 46, lr: 5.62e-03, grad_scale: 8.0 2023-03-08 20:00:04,284 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3397, 5.3584, 5.1333, 3.0324, 5.1957, 5.0376, 4.7787, 3.2931], device='cuda:3'), covar=tensor([0.0135, 0.0078, 0.0242, 0.0942, 0.0068, 0.0144, 0.0210, 0.1072], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0100, 0.0101, 0.0111, 0.0084, 0.0110, 0.0098, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:00:19,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5710, 4.8876, 5.3749, 4.8146, 4.6448, 5.3738, 5.0108, 5.4288], device='cuda:3'), covar=tensor([0.1625, 0.1808, 0.1437, 0.2712, 0.4301, 0.1970, 0.1377, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0852, 0.0501, 0.0587, 0.0645, 0.0862, 0.0609, 0.0482, 0.0590], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:00:22,006 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 20:00:50,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.066e+02 2.537e+02 3.345e+02 7.460e+02, threshold=5.074e+02, percent-clipped=4.0 2023-03-08 20:01:23,608 INFO [train2.py:809] (3/4) Epoch 19, batch 3850, loss[ctc_loss=0.07004, att_loss=0.2474, loss=0.2119, over 16774.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006126, over 48.00 utterances.], tot_loss[ctc_loss=0.07728, att_loss=0.2365, loss=0.2046, over 3250657.45 frames. utt_duration=1239 frames, utt_pad_proportion=0.06176, over 10508.43 utterances.], batch size: 48, lr: 5.61e-03, grad_scale: 8.0 2023-03-08 20:01:34,614 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:02:03,510 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:02:40,370 INFO [train2.py:809] (3/4) Epoch 19, batch 3900, loss[ctc_loss=0.0722, att_loss=0.2517, loss=0.2158, over 16330.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006141, over 45.00 utterances.], tot_loss[ctc_loss=0.0766, att_loss=0.2367, loss=0.2047, over 3265768.65 frames. utt_duration=1254 frames, utt_pad_proportion=0.05422, over 10427.74 utterances.], batch size: 45, lr: 5.61e-03, grad_scale: 8.0 2023-03-08 20:02:47,828 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:02:58,574 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:03:07,702 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 20:03:24,259 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.998e+02 2.333e+02 2.845e+02 5.732e+02, threshold=4.667e+02, percent-clipped=1.0 2023-03-08 20:03:34,385 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3608, 5.6382, 5.0315, 5.3708, 5.2505, 4.8856, 5.1149, 4.7834], device='cuda:3'), covar=tensor([0.1223, 0.0951, 0.1017, 0.0923, 0.0922, 0.1388, 0.1986, 0.2532], device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0593, 0.0449, 0.0444, 0.0419, 0.0456, 0.0599, 0.0517], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:03:57,014 INFO [train2.py:809] (3/4) Epoch 19, batch 3950, loss[ctc_loss=0.06783, att_loss=0.2389, loss=0.2047, over 17022.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007846, over 51.00 utterances.], tot_loss[ctc_loss=0.07697, att_loss=0.2367, loss=0.2048, over 3263170.82 frames. utt_duration=1252 frames, utt_pad_proportion=0.05355, over 10437.84 utterances.], batch size: 51, lr: 5.61e-03, grad_scale: 8.0 2023-03-08 20:04:00,570 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5063, 4.5193, 4.5874, 4.6733, 5.2563, 4.5657, 4.5009, 2.4643], device='cuda:3'), covar=tensor([0.0210, 0.0388, 0.0316, 0.0257, 0.0670, 0.0198, 0.0342, 0.1827], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0179, 0.0194, 0.0360, 0.0148, 0.0166, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:04:12,257 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:04:14,697 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-08 20:05:12,144 INFO [train2.py:809] (3/4) Epoch 20, batch 0, loss[ctc_loss=0.06592, att_loss=0.2189, loss=0.1883, over 15350.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01198, over 35.00 utterances.], tot_loss[ctc_loss=0.06592, att_loss=0.2189, loss=0.1883, over 15350.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01198, over 35.00 utterances.], batch size: 35, lr: 5.46e-03, grad_scale: 8.0 2023-03-08 20:05:12,145 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 20:05:24,206 INFO [train2.py:843] (3/4) Epoch 20, validation: ctc_loss=0.04136, att_loss=0.235, loss=0.1963, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 20:05:24,207 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 20:06:35,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.109e+02 2.667e+02 3.273e+02 6.239e+02, threshold=5.334e+02, percent-clipped=6.0 2023-03-08 20:06:43,085 INFO [train2.py:809] (3/4) Epoch 20, batch 50, loss[ctc_loss=0.07526, att_loss=0.2376, loss=0.2051, over 16402.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.00631, over 44.00 utterances.], tot_loss[ctc_loss=0.07996, att_loss=0.24, loss=0.208, over 740207.51 frames. utt_duration=1239 frames, utt_pad_proportion=0.05378, over 2392.34 utterances.], batch size: 44, lr: 5.46e-03, grad_scale: 8.0 2023-03-08 20:08:03,711 INFO [train2.py:809] (3/4) Epoch 20, batch 100, loss[ctc_loss=0.08412, att_loss=0.2489, loss=0.216, over 17331.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.0368, over 63.00 utterances.], tot_loss[ctc_loss=0.07765, att_loss=0.2389, loss=0.2067, over 1310087.19 frames. utt_duration=1284 frames, utt_pad_proportion=0.03871, over 4085.59 utterances.], batch size: 63, lr: 5.46e-03, grad_scale: 8.0 2023-03-08 20:09:16,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.985e+02 2.361e+02 2.843e+02 6.853e+02, threshold=4.722e+02, percent-clipped=2.0 2023-03-08 20:09:24,379 INFO [train2.py:809] (3/4) Epoch 20, batch 150, loss[ctc_loss=0.08129, att_loss=0.2512, loss=0.2172, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006026, over 46.00 utterances.], tot_loss[ctc_loss=0.07822, att_loss=0.2381, loss=0.2061, over 1751237.77 frames. utt_duration=1291 frames, utt_pad_proportion=0.03733, over 5433.76 utterances.], batch size: 46, lr: 5.46e-03, grad_scale: 8.0 2023-03-08 20:10:33,725 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:10:46,196 INFO [train2.py:809] (3/4) Epoch 20, batch 200, loss[ctc_loss=0.08415, att_loss=0.2463, loss=0.2138, over 17482.00 frames. utt_duration=1111 frames, utt_pad_proportion=0.0268, over 63.00 utterances.], tot_loss[ctc_loss=0.07875, att_loss=0.2393, loss=0.2072, over 2093088.26 frames. utt_duration=1266 frames, utt_pad_proportion=0.04348, over 6618.69 utterances.], batch size: 63, lr: 5.46e-03, grad_scale: 8.0 2023-03-08 20:11:08,120 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8551, 5.2024, 5.4485, 5.3013, 5.3946, 5.7965, 5.1597, 5.8988], device='cuda:3'), covar=tensor([0.0717, 0.0789, 0.0759, 0.1372, 0.1657, 0.0867, 0.0765, 0.0730], device='cuda:3'), in_proj_covar=tensor([0.0846, 0.0495, 0.0581, 0.0639, 0.0850, 0.0599, 0.0474, 0.0586], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:11:42,069 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 20:11:51,156 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:11:59,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.084e+02 2.421e+02 2.926e+02 5.998e+02, threshold=4.842e+02, percent-clipped=2.0 2023-03-08 20:12:06,826 INFO [train2.py:809] (3/4) Epoch 20, batch 250, loss[ctc_loss=0.07106, att_loss=0.2465, loss=0.2114, over 17382.00 frames. utt_duration=1105 frames, utt_pad_proportion=0.03395, over 63.00 utterances.], tot_loss[ctc_loss=0.07793, att_loss=0.2385, loss=0.2064, over 2353935.32 frames. utt_duration=1239 frames, utt_pad_proportion=0.05167, over 7606.10 utterances.], batch size: 63, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:12:35,953 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8869, 5.1976, 4.7597, 5.3016, 4.5892, 4.9515, 5.3647, 5.1673], device='cuda:3'), covar=tensor([0.0639, 0.0307, 0.0808, 0.0301, 0.0470, 0.0212, 0.0222, 0.0216], device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0312, 0.0362, 0.0337, 0.0315, 0.0234, 0.0297, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 20:12:59,208 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:12:59,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-03-08 20:13:26,594 INFO [train2.py:809] (3/4) Epoch 20, batch 300, loss[ctc_loss=0.08025, att_loss=0.2303, loss=0.2003, over 16188.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006362, over 41.00 utterances.], tot_loss[ctc_loss=0.078, att_loss=0.2384, loss=0.2063, over 2556819.34 frames. utt_duration=1238 frames, utt_pad_proportion=0.0535, over 8273.42 utterances.], batch size: 41, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:14:32,076 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9968, 5.2686, 5.1749, 5.1565, 5.2505, 5.2060, 4.9222, 4.7414], device='cuda:3'), covar=tensor([0.0823, 0.0449, 0.0270, 0.0408, 0.0256, 0.0270, 0.0353, 0.0269], device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0354, 0.0335, 0.0349, 0.0414, 0.0425, 0.0347, 0.0389], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 20:14:42,639 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.144e+02 2.516e+02 3.059e+02 1.142e+03, threshold=5.032e+02, percent-clipped=7.0 2023-03-08 20:14:50,247 INFO [train2.py:809] (3/4) Epoch 20, batch 350, loss[ctc_loss=0.07652, att_loss=0.2267, loss=0.1967, over 12388.00 frames. utt_duration=1837 frames, utt_pad_proportion=0.1459, over 27.00 utterances.], tot_loss[ctc_loss=0.07731, att_loss=0.2382, loss=0.206, over 2711994.03 frames. utt_duration=1266 frames, utt_pad_proportion=0.04847, over 8580.43 utterances.], batch size: 27, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:15:14,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0322, 5.2465, 5.5540, 5.3553, 5.5617, 5.9602, 5.2588, 6.0873], device='cuda:3'), covar=tensor([0.0674, 0.0753, 0.0762, 0.1332, 0.1521, 0.0884, 0.0570, 0.0630], device='cuda:3'), in_proj_covar=tensor([0.0848, 0.0496, 0.0579, 0.0635, 0.0851, 0.0600, 0.0475, 0.0584], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:15:55,372 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1913, 5.4597, 4.9225, 5.2979, 5.1116, 4.7213, 4.9129, 4.7671], device='cuda:3'), covar=tensor([0.1346, 0.0936, 0.0988, 0.0763, 0.0947, 0.1500, 0.2390, 0.2363], device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0600, 0.0454, 0.0448, 0.0424, 0.0464, 0.0608, 0.0525], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:16:10,639 INFO [train2.py:809] (3/4) Epoch 20, batch 400, loss[ctc_loss=0.0923, att_loss=0.2544, loss=0.222, over 16473.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006865, over 46.00 utterances.], tot_loss[ctc_loss=0.07773, att_loss=0.2389, loss=0.2067, over 2839397.94 frames. utt_duration=1257 frames, utt_pad_proportion=0.05002, over 9047.66 utterances.], batch size: 46, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:16:59,800 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0433, 5.0711, 4.9212, 2.1775, 2.0828, 2.7560, 2.3763, 3.8399], device='cuda:3'), covar=tensor([0.0710, 0.0257, 0.0256, 0.5573, 0.5537, 0.2524, 0.3482, 0.1666], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0265, 0.0262, 0.0239, 0.0342, 0.0332, 0.0248, 0.0363], device='cuda:3'), out_proj_covar=tensor([1.4987e-04, 9.8214e-05, 1.1196e-04, 1.0285e-04, 1.4360e-04, 1.3037e-04, 9.9471e-05, 1.4779e-04], device='cuda:3') 2023-03-08 20:17:22,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 1.983e+02 2.448e+02 2.910e+02 4.561e+02, threshold=4.897e+02, percent-clipped=0.0 2023-03-08 20:17:29,975 INFO [train2.py:809] (3/4) Epoch 20, batch 450, loss[ctc_loss=0.06199, att_loss=0.2344, loss=0.1999, over 16130.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.00611, over 42.00 utterances.], tot_loss[ctc_loss=0.07715, att_loss=0.2376, loss=0.2055, over 2932982.55 frames. utt_duration=1283 frames, utt_pad_proportion=0.04559, over 9153.72 utterances.], batch size: 42, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:18:49,887 INFO [train2.py:809] (3/4) Epoch 20, batch 500, loss[ctc_loss=0.06985, att_loss=0.2245, loss=0.1936, over 16177.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.00632, over 41.00 utterances.], tot_loss[ctc_loss=0.07727, att_loss=0.2372, loss=0.2052, over 3004722.51 frames. utt_duration=1275 frames, utt_pad_proportion=0.04879, over 9439.80 utterances.], batch size: 41, lr: 5.45e-03, grad_scale: 8.0 2023-03-08 20:20:01,815 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 1.846e+02 2.334e+02 2.812e+02 5.634e+02, threshold=4.669e+02, percent-clipped=1.0 2023-03-08 20:20:09,799 INFO [train2.py:809] (3/4) Epoch 20, batch 550, loss[ctc_loss=0.08184, att_loss=0.2474, loss=0.2143, over 17339.00 frames. utt_duration=879.4 frames, utt_pad_proportion=0.07622, over 79.00 utterances.], tot_loss[ctc_loss=0.07749, att_loss=0.2375, loss=0.2055, over 3054909.91 frames. utt_duration=1234 frames, utt_pad_proportion=0.06143, over 9914.15 utterances.], batch size: 79, lr: 5.44e-03, grad_scale: 8.0 2023-03-08 20:20:10,150 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8200, 3.4871, 3.5489, 2.9758, 3.5092, 3.6623, 3.5654, 2.4510], device='cuda:3'), covar=tensor([0.1017, 0.1875, 0.2360, 0.4238, 0.1383, 0.2828, 0.0953, 0.4623], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0175, 0.0187, 0.0249, 0.0150, 0.0248, 0.0167, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:20:40,264 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1356, 4.5355, 4.8284, 4.9607, 3.1277, 4.4550, 3.2508, 2.3008], device='cuda:3'), covar=tensor([0.0413, 0.0237, 0.0513, 0.0165, 0.1371, 0.0191, 0.1178, 0.1461], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0156, 0.0257, 0.0152, 0.0218, 0.0135, 0.0229, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:20:54,262 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:21:29,926 INFO [train2.py:809] (3/4) Epoch 20, batch 600, loss[ctc_loss=0.1011, att_loss=0.2542, loss=0.2236, over 16839.00 frames. utt_duration=681.9 frames, utt_pad_proportion=0.1444, over 99.00 utterances.], tot_loss[ctc_loss=0.07694, att_loss=0.2373, loss=0.2052, over 3108631.94 frames. utt_duration=1248 frames, utt_pad_proportion=0.05461, over 9972.22 utterances.], batch size: 99, lr: 5.44e-03, grad_scale: 16.0 2023-03-08 20:21:33,839 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:21:44,498 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9436, 3.6633, 3.1519, 3.3654, 3.8206, 3.6033, 2.9446, 4.0647], device='cuda:3'), covar=tensor([0.1031, 0.0466, 0.1059, 0.0685, 0.0751, 0.0669, 0.0908, 0.0457], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0213, 0.0225, 0.0198, 0.0274, 0.0237, 0.0200, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 20:22:20,726 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:22:31,396 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 20:22:41,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 1.945e+02 2.306e+02 2.690e+02 7.016e+02, threshold=4.612e+02, percent-clipped=2.0 2023-03-08 20:22:49,616 INFO [train2.py:809] (3/4) Epoch 20, batch 650, loss[ctc_loss=0.07803, att_loss=0.248, loss=0.214, over 16755.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006352, over 48.00 utterances.], tot_loss[ctc_loss=0.07688, att_loss=0.2373, loss=0.2052, over 3142407.28 frames. utt_duration=1261 frames, utt_pad_proportion=0.05139, over 9983.31 utterances.], batch size: 48, lr: 5.44e-03, grad_scale: 16.0 2023-03-08 20:23:11,364 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:23:18,799 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8837, 4.9219, 4.6670, 2.9095, 4.6673, 4.5219, 4.1520, 2.7010], device='cuda:3'), covar=tensor([0.0104, 0.0088, 0.0255, 0.1006, 0.0093, 0.0201, 0.0352, 0.1382], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0100, 0.0101, 0.0110, 0.0083, 0.0110, 0.0097, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:23:56,998 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:24:09,249 INFO [train2.py:809] (3/4) Epoch 20, batch 700, loss[ctc_loss=0.08243, att_loss=0.2462, loss=0.2134, over 16624.00 frames. utt_duration=673.1 frames, utt_pad_proportion=0.1544, over 99.00 utterances.], tot_loss[ctc_loss=0.07617, att_loss=0.2369, loss=0.2048, over 3176885.42 frames. utt_duration=1262 frames, utt_pad_proportion=0.04948, over 10079.50 utterances.], batch size: 99, lr: 5.44e-03, grad_scale: 16.0 2023-03-08 20:25:20,471 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 1.882e+02 2.322e+02 2.702e+02 6.742e+02, threshold=4.645e+02, percent-clipped=2.0 2023-03-08 20:25:28,806 INFO [train2.py:809] (3/4) Epoch 20, batch 750, loss[ctc_loss=0.07249, att_loss=0.2463, loss=0.2115, over 17291.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01269, over 55.00 utterances.], tot_loss[ctc_loss=0.07652, att_loss=0.2372, loss=0.205, over 3184207.32 frames. utt_duration=1251 frames, utt_pad_proportion=0.05583, over 10191.50 utterances.], batch size: 55, lr: 5.44e-03, grad_scale: 16.0 2023-03-08 20:26:20,847 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 20:26:48,174 INFO [train2.py:809] (3/4) Epoch 20, batch 800, loss[ctc_loss=0.06064, att_loss=0.2217, loss=0.1895, over 16173.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006278, over 41.00 utterances.], tot_loss[ctc_loss=0.07656, att_loss=0.2369, loss=0.2048, over 3206876.63 frames. utt_duration=1246 frames, utt_pad_proportion=0.05518, over 10304.42 utterances.], batch size: 41, lr: 5.44e-03, grad_scale: 16.0 2023-03-08 20:28:00,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 1.827e+02 2.351e+02 2.962e+02 7.560e+02, threshold=4.702e+02, percent-clipped=5.0 2023-03-08 20:28:07,274 INFO [train2.py:809] (3/4) Epoch 20, batch 850, loss[ctc_loss=0.08649, att_loss=0.2343, loss=0.2047, over 14556.00 frames. utt_duration=1821 frames, utt_pad_proportion=0.04161, over 32.00 utterances.], tot_loss[ctc_loss=0.07594, att_loss=0.2363, loss=0.2042, over 3221239.48 frames. utt_duration=1270 frames, utt_pad_proportion=0.05045, over 10155.54 utterances.], batch size: 32, lr: 5.43e-03, grad_scale: 8.0 2023-03-08 20:28:45,964 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0848, 4.0738, 4.0180, 4.0458, 4.4386, 4.1187, 3.9262, 2.5574], device='cuda:3'), covar=tensor([0.0241, 0.0422, 0.0418, 0.0341, 0.0826, 0.0230, 0.0401, 0.1811], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0175, 0.0178, 0.0193, 0.0360, 0.0149, 0.0166, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:28:55,447 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-08 20:29:26,683 INFO [train2.py:809] (3/4) Epoch 20, batch 900, loss[ctc_loss=0.08298, att_loss=0.2402, loss=0.2088, over 17052.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009775, over 53.00 utterances.], tot_loss[ctc_loss=0.07609, att_loss=0.2365, loss=0.2045, over 3234155.68 frames. utt_duration=1244 frames, utt_pad_proportion=0.0539, over 10415.20 utterances.], batch size: 53, lr: 5.43e-03, grad_scale: 8.0 2023-03-08 20:30:21,638 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 20:30:41,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.428e+02 2.021e+02 2.452e+02 3.045e+02 6.796e+02, threshold=4.904e+02, percent-clipped=5.0 2023-03-08 20:30:47,771 INFO [train2.py:809] (3/4) Epoch 20, batch 950, loss[ctc_loss=0.07639, att_loss=0.245, loss=0.2113, over 17211.00 frames. utt_duration=872.9 frames, utt_pad_proportion=0.08594, over 79.00 utterances.], tot_loss[ctc_loss=0.07645, att_loss=0.2371, loss=0.2049, over 3240622.30 frames. utt_duration=1237 frames, utt_pad_proportion=0.05748, over 10491.89 utterances.], batch size: 79, lr: 5.43e-03, grad_scale: 8.0 2023-03-08 20:31:02,343 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:31:47,599 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 20:32:08,848 INFO [train2.py:809] (3/4) Epoch 20, batch 1000, loss[ctc_loss=0.05381, att_loss=0.2256, loss=0.1913, over 16137.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005695, over 42.00 utterances.], tot_loss[ctc_loss=0.07645, att_loss=0.2371, loss=0.205, over 3245392.01 frames. utt_duration=1240 frames, utt_pad_proportion=0.05562, over 10479.31 utterances.], batch size: 42, lr: 5.43e-03, grad_scale: 8.0 2023-03-08 20:33:21,296 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 1.956e+02 2.404e+02 2.887e+02 6.438e+02, threshold=4.808e+02, percent-clipped=2.0 2023-03-08 20:33:28,421 INFO [train2.py:809] (3/4) Epoch 20, batch 1050, loss[ctc_loss=0.05423, att_loss=0.2165, loss=0.184, over 16196.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.005507, over 41.00 utterances.], tot_loss[ctc_loss=0.07622, att_loss=0.2365, loss=0.2045, over 3242640.67 frames. utt_duration=1222 frames, utt_pad_proportion=0.06293, over 10627.54 utterances.], batch size: 41, lr: 5.43e-03, grad_scale: 8.0 2023-03-08 20:34:07,672 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-03-08 20:34:48,265 INFO [train2.py:809] (3/4) Epoch 20, batch 1100, loss[ctc_loss=0.07012, att_loss=0.2405, loss=0.2064, over 16774.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005978, over 48.00 utterances.], tot_loss[ctc_loss=0.0764, att_loss=0.2366, loss=0.2046, over 3250471.23 frames. utt_duration=1242 frames, utt_pad_proportion=0.05772, over 10481.82 utterances.], batch size: 48, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:34:53,888 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1800, 3.9799, 3.4415, 3.6858, 4.1934, 3.8364, 3.2631, 4.5142], device='cuda:3'), covar=tensor([0.1001, 0.0460, 0.0997, 0.0618, 0.0665, 0.0660, 0.0811, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0212, 0.0223, 0.0196, 0.0272, 0.0236, 0.0199, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 20:35:01,852 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2893, 5.5018, 5.4832, 5.4422, 5.5587, 5.4805, 5.2511, 5.0580], device='cuda:3'), covar=tensor([0.0964, 0.0527, 0.0245, 0.0451, 0.0297, 0.0294, 0.0314, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0358, 0.0339, 0.0354, 0.0414, 0.0429, 0.0350, 0.0390], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 20:36:02,083 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 2.008e+02 2.417e+02 3.031e+02 6.018e+02, threshold=4.833e+02, percent-clipped=3.0 2023-03-08 20:36:03,991 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:36:08,266 INFO [train2.py:809] (3/4) Epoch 20, batch 1150, loss[ctc_loss=0.07321, att_loss=0.2379, loss=0.2049, over 17070.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008658, over 53.00 utterances.], tot_loss[ctc_loss=0.07617, att_loss=0.2365, loss=0.2044, over 3256440.05 frames. utt_duration=1245 frames, utt_pad_proportion=0.05642, over 10473.05 utterances.], batch size: 53, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:36:31,196 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0142, 5.2578, 5.5325, 5.4188, 5.4767, 5.9283, 5.2923, 6.0440], device='cuda:3'), covar=tensor([0.0798, 0.0852, 0.0902, 0.1316, 0.2084, 0.0976, 0.0596, 0.0781], device='cuda:3'), in_proj_covar=tensor([0.0856, 0.0501, 0.0583, 0.0646, 0.0857, 0.0611, 0.0477, 0.0592], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:37:28,708 INFO [train2.py:809] (3/4) Epoch 20, batch 1200, loss[ctc_loss=0.06502, att_loss=0.2151, loss=0.1851, over 15924.00 frames. utt_duration=1635 frames, utt_pad_proportion=0.006901, over 39.00 utterances.], tot_loss[ctc_loss=0.07618, att_loss=0.2363, loss=0.2043, over 3251791.82 frames. utt_duration=1233 frames, utt_pad_proportion=0.0625, over 10562.18 utterances.], batch size: 39, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:37:42,000 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:38:21,860 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 20:38:42,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 1.866e+02 2.347e+02 2.914e+02 5.639e+02, threshold=4.693e+02, percent-clipped=1.0 2023-03-08 20:38:49,699 INFO [train2.py:809] (3/4) Epoch 20, batch 1250, loss[ctc_loss=0.05399, att_loss=0.207, loss=0.1764, over 15518.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007651, over 36.00 utterances.], tot_loss[ctc_loss=0.07551, att_loss=0.2358, loss=0.2038, over 3252670.86 frames. utt_duration=1258 frames, utt_pad_proportion=0.0554, over 10352.61 utterances.], batch size: 36, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:39:02,630 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:39:19,857 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9040, 3.5580, 3.6526, 2.8918, 3.5797, 3.7200, 3.7106, 2.3548], device='cuda:3'), covar=tensor([0.1287, 0.2006, 0.2060, 0.5851, 0.1425, 0.2828, 0.0828, 0.6831], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0175, 0.0187, 0.0246, 0.0150, 0.0246, 0.0166, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:39:38,664 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:39:48,062 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:40:10,276 INFO [train2.py:809] (3/4) Epoch 20, batch 1300, loss[ctc_loss=0.07298, att_loss=0.2475, loss=0.2126, over 17020.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007599, over 51.00 utterances.], tot_loss[ctc_loss=0.07545, att_loss=0.2357, loss=0.2037, over 3258976.79 frames. utt_duration=1262 frames, utt_pad_proportion=0.05287, over 10339.42 utterances.], batch size: 51, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:40:19,515 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:41:04,689 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:41:22,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.046e+02 2.509e+02 3.001e+02 6.074e+02, threshold=5.017e+02, percent-clipped=4.0 2023-03-08 20:41:29,842 INFO [train2.py:809] (3/4) Epoch 20, batch 1350, loss[ctc_loss=0.1397, att_loss=0.2722, loss=0.2457, over 13970.00 frames. utt_duration=386.8 frames, utt_pad_proportion=0.3284, over 145.00 utterances.], tot_loss[ctc_loss=0.07568, att_loss=0.2361, loss=0.204, over 3253484.16 frames. utt_duration=1239 frames, utt_pad_proportion=0.06018, over 10515.39 utterances.], batch size: 145, lr: 5.42e-03, grad_scale: 8.0 2023-03-08 20:42:48,969 INFO [train2.py:809] (3/4) Epoch 20, batch 1400, loss[ctc_loss=0.07156, att_loss=0.2448, loss=0.2102, over 16857.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008154, over 49.00 utterances.], tot_loss[ctc_loss=0.07664, att_loss=0.2371, loss=0.205, over 3262492.96 frames. utt_duration=1232 frames, utt_pad_proportion=0.06016, over 10604.26 utterances.], batch size: 49, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:42:54,079 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:43:15,375 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0738, 5.2104, 5.6332, 5.3694, 5.5192, 5.9828, 5.2600, 6.1238], device='cuda:3'), covar=tensor([0.0673, 0.0795, 0.0799, 0.1319, 0.1732, 0.0979, 0.0617, 0.0644], device='cuda:3'), in_proj_covar=tensor([0.0863, 0.0499, 0.0585, 0.0645, 0.0858, 0.0611, 0.0476, 0.0590], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:44:01,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 1.974e+02 2.358e+02 2.926e+02 6.381e+02, threshold=4.717e+02, percent-clipped=2.0 2023-03-08 20:44:05,921 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 20:44:08,007 INFO [train2.py:809] (3/4) Epoch 20, batch 1450, loss[ctc_loss=0.084, att_loss=0.2545, loss=0.2204, over 17401.00 frames. utt_duration=1181 frames, utt_pad_proportion=0.01816, over 59.00 utterances.], tot_loss[ctc_loss=0.07668, att_loss=0.237, loss=0.205, over 3271766.83 frames. utt_duration=1251 frames, utt_pad_proportion=0.053, over 10473.60 utterances.], batch size: 59, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:44:29,623 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={3} 2023-03-08 20:45:07,789 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5542, 2.2427, 2.4726, 2.5133, 2.8163, 2.5899, 2.6029, 3.1792], device='cuda:3'), covar=tensor([0.1331, 0.3155, 0.2164, 0.1175, 0.1275, 0.0941, 0.1740, 0.0921], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0122, 0.0116, 0.0105, 0.0117, 0.0101, 0.0125, 0.0093], device='cuda:3'), out_proj_covar=tensor([8.4490e-05, 9.3599e-05, 9.1947e-05, 8.1595e-05, 8.6381e-05, 8.1347e-05, 9.3374e-05, 7.4969e-05], device='cuda:3') 2023-03-08 20:45:27,451 INFO [train2.py:809] (3/4) Epoch 20, batch 1500, loss[ctc_loss=0.06463, att_loss=0.2295, loss=0.1965, over 16297.00 frames. utt_duration=1518 frames, utt_pad_proportion=0.006168, over 43.00 utterances.], tot_loss[ctc_loss=0.07627, att_loss=0.2366, loss=0.2045, over 3267565.92 frames. utt_duration=1265 frames, utt_pad_proportion=0.05203, over 10347.30 utterances.], batch size: 43, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:45:31,986 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:45:57,180 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:46:34,105 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-08 20:46:40,576 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 1.945e+02 2.357e+02 2.826e+02 4.354e+02, threshold=4.714e+02, percent-clipped=0.0 2023-03-08 20:46:46,769 INFO [train2.py:809] (3/4) Epoch 20, batch 1550, loss[ctc_loss=0.07426, att_loss=0.2482, loss=0.2134, over 17333.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02264, over 59.00 utterances.], tot_loss[ctc_loss=0.07574, att_loss=0.2365, loss=0.2044, over 3273779.59 frames. utt_duration=1278 frames, utt_pad_proportion=0.0469, over 10256.77 utterances.], batch size: 59, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:47:29,719 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6596, 2.2241, 2.4630, 2.5047, 2.7932, 2.6364, 2.6204, 3.2451], device='cuda:3'), covar=tensor([0.1778, 0.3494, 0.2386, 0.1677, 0.1599, 0.1278, 0.2423, 0.1131], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0124, 0.0119, 0.0107, 0.0119, 0.0104, 0.0127, 0.0095], device='cuda:3'), out_proj_covar=tensor([8.6777e-05, 9.4969e-05, 9.3537e-05, 8.3091e-05, 8.7929e-05, 8.3297e-05, 9.4958e-05, 7.6389e-05], device='cuda:3') 2023-03-08 20:47:34,337 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:47:35,866 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:47:42,453 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:48:05,778 INFO [train2.py:809] (3/4) Epoch 20, batch 1600, loss[ctc_loss=0.06918, att_loss=0.2242, loss=0.1932, over 15936.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.007544, over 41.00 utterances.], tot_loss[ctc_loss=0.07562, att_loss=0.2366, loss=0.2044, over 3273734.27 frames. utt_duration=1280 frames, utt_pad_proportion=0.0484, over 10243.48 utterances.], batch size: 41, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:48:58,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 20:49:11,780 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:49:18,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.314e+02 1.917e+02 2.281e+02 2.951e+02 6.682e+02, threshold=4.563e+02, percent-clipped=3.0 2023-03-08 20:49:18,736 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:49:25,152 INFO [train2.py:809] (3/4) Epoch 20, batch 1650, loss[ctc_loss=0.08869, att_loss=0.2385, loss=0.2086, over 16685.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.00613, over 46.00 utterances.], tot_loss[ctc_loss=0.07491, att_loss=0.2365, loss=0.2042, over 3276994.24 frames. utt_duration=1292 frames, utt_pad_proportion=0.04368, over 10153.72 utterances.], batch size: 46, lr: 5.41e-03, grad_scale: 8.0 2023-03-08 20:50:44,527 INFO [train2.py:809] (3/4) Epoch 20, batch 1700, loss[ctc_loss=0.1189, att_loss=0.2699, loss=0.2397, over 17364.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03484, over 63.00 utterances.], tot_loss[ctc_loss=0.07531, att_loss=0.2366, loss=0.2044, over 3279090.99 frames. utt_duration=1284 frames, utt_pad_proportion=0.04566, over 10229.10 utterances.], batch size: 63, lr: 5.40e-03, grad_scale: 8.0 2023-03-08 20:51:57,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.333e+02 1.985e+02 2.370e+02 2.830e+02 5.351e+02, threshold=4.740e+02, percent-clipped=1.0 2023-03-08 20:52:04,407 INFO [train2.py:809] (3/4) Epoch 20, batch 1750, loss[ctc_loss=0.07155, att_loss=0.2177, loss=0.1885, over 15878.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009783, over 39.00 utterances.], tot_loss[ctc_loss=0.07578, att_loss=0.2366, loss=0.2045, over 3273441.55 frames. utt_duration=1255 frames, utt_pad_proportion=0.05444, over 10442.35 utterances.], batch size: 39, lr: 5.40e-03, grad_scale: 8.0 2023-03-08 20:52:07,912 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6073, 3.1335, 3.5722, 3.0200, 3.5327, 4.6286, 4.4400, 3.2988], device='cuda:3'), covar=tensor([0.0296, 0.1491, 0.1290, 0.1255, 0.1094, 0.0786, 0.0538, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0241, 0.0274, 0.0213, 0.0263, 0.0356, 0.0256, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:52:18,582 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 20:52:58,688 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1368, 4.4721, 4.5849, 4.7233, 2.9081, 4.5586, 2.5087, 1.8798], device='cuda:3'), covar=tensor([0.0353, 0.0208, 0.0595, 0.0184, 0.1495, 0.0155, 0.1615, 0.1723], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0158, 0.0259, 0.0155, 0.0221, 0.0138, 0.0232, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:53:24,307 INFO [train2.py:809] (3/4) Epoch 20, batch 1800, loss[ctc_loss=0.05939, att_loss=0.2167, loss=0.1852, over 15868.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01042, over 39.00 utterances.], tot_loss[ctc_loss=0.07635, att_loss=0.2366, loss=0.2045, over 3266673.67 frames. utt_duration=1252 frames, utt_pad_proportion=0.05626, over 10446.50 utterances.], batch size: 39, lr: 5.40e-03, grad_scale: 8.0 2023-03-08 20:53:29,184 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:54:37,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.944e+02 2.387e+02 2.834e+02 5.542e+02, threshold=4.773e+02, percent-clipped=3.0 2023-03-08 20:54:44,323 INFO [train2.py:809] (3/4) Epoch 20, batch 1850, loss[ctc_loss=0.06254, att_loss=0.236, loss=0.2013, over 17455.00 frames. utt_duration=1110 frames, utt_pad_proportion=0.02986, over 63.00 utterances.], tot_loss[ctc_loss=0.07712, att_loss=0.2375, loss=0.2055, over 3266097.54 frames. utt_duration=1199 frames, utt_pad_proportion=0.06899, over 10912.04 utterances.], batch size: 63, lr: 5.40e-03, grad_scale: 8.0 2023-03-08 20:54:44,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8541, 6.1681, 5.5218, 5.8769, 5.7568, 5.3344, 5.6156, 5.4042], device='cuda:3'), covar=tensor([0.1233, 0.0897, 0.0958, 0.0738, 0.0891, 0.1417, 0.2094, 0.2120], device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0602, 0.0455, 0.0445, 0.0427, 0.0462, 0.0605, 0.0524], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 20:54:45,978 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:55:23,576 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:55:35,134 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4656, 3.2449, 3.8110, 4.5562, 4.0165, 3.9567, 3.1992, 2.5829], device='cuda:3'), covar=tensor([0.0723, 0.1782, 0.0689, 0.0553, 0.0769, 0.0503, 0.1377, 0.2040], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0217, 0.0189, 0.0216, 0.0219, 0.0176, 0.0203, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 20:56:03,931 INFO [train2.py:809] (3/4) Epoch 20, batch 1900, loss[ctc_loss=0.06294, att_loss=0.2372, loss=0.2023, over 16961.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.006959, over 50.00 utterances.], tot_loss[ctc_loss=0.07729, att_loss=0.2375, loss=0.2055, over 3270562.80 frames. utt_duration=1214 frames, utt_pad_proportion=0.06477, over 10785.45 utterances.], batch size: 50, lr: 5.40e-03, grad_scale: 8.0 2023-03-08 20:56:42,730 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:57:03,881 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:57:10,170 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:57:17,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.953e+02 2.465e+02 3.107e+02 2.003e+03, threshold=4.931e+02, percent-clipped=5.0 2023-03-08 20:57:24,504 INFO [train2.py:809] (3/4) Epoch 20, batch 1950, loss[ctc_loss=0.05691, att_loss=0.2174, loss=0.1853, over 15774.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008369, over 38.00 utterances.], tot_loss[ctc_loss=0.07616, att_loss=0.2367, loss=0.2046, over 3271364.44 frames. utt_duration=1230 frames, utt_pad_proportion=0.06043, over 10654.60 utterances.], batch size: 38, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 20:57:38,741 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8833, 5.2696, 5.0667, 5.2123, 5.3166, 4.9043, 3.6447, 5.2297], device='cuda:3'), covar=tensor([0.0092, 0.0090, 0.0109, 0.0063, 0.0070, 0.0089, 0.0580, 0.0142], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0084, 0.0107, 0.0066, 0.0072, 0.0083, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 20:57:39,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-08 20:58:19,950 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:58:19,983 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 20:58:44,129 INFO [train2.py:809] (3/4) Epoch 20, batch 2000, loss[ctc_loss=0.08439, att_loss=0.2532, loss=0.2194, over 17377.00 frames. utt_duration=881.2 frames, utt_pad_proportion=0.07636, over 79.00 utterances.], tot_loss[ctc_loss=0.07663, att_loss=0.2368, loss=0.2048, over 3272966.18 frames. utt_duration=1233 frames, utt_pad_proportion=0.05799, over 10628.41 utterances.], batch size: 79, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 20:59:57,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.073e+02 2.649e+02 3.244e+02 1.218e+03, threshold=5.298e+02, percent-clipped=5.0 2023-03-08 20:59:57,979 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:00:03,841 INFO [train2.py:809] (3/4) Epoch 20, batch 2050, loss[ctc_loss=0.0587, att_loss=0.2145, loss=0.1833, over 13705.00 frames. utt_duration=1829 frames, utt_pad_proportion=0.06874, over 30.00 utterances.], tot_loss[ctc_loss=0.07748, att_loss=0.2379, loss=0.2058, over 3270014.54 frames. utt_duration=1204 frames, utt_pad_proportion=0.06691, over 10881.74 utterances.], batch size: 30, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 21:00:17,889 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 21:00:28,788 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0553, 5.0228, 4.8469, 2.2513, 2.0043, 2.6228, 2.4381, 3.8201], device='cuda:3'), covar=tensor([0.0708, 0.0263, 0.0225, 0.4432, 0.5357, 0.2782, 0.3314, 0.1727], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0269, 0.0264, 0.0243, 0.0344, 0.0335, 0.0251, 0.0367], device='cuda:3'), out_proj_covar=tensor([1.4995e-04, 9.9500e-05, 1.1215e-04, 1.0489e-04, 1.4440e-04, 1.3119e-04, 1.0054e-04, 1.4891e-04], device='cuda:3') 2023-03-08 21:01:23,889 INFO [train2.py:809] (3/4) Epoch 20, batch 2100, loss[ctc_loss=0.1176, att_loss=0.2631, loss=0.234, over 17124.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.0147, over 56.00 utterances.], tot_loss[ctc_loss=0.07774, att_loss=0.2381, loss=0.2061, over 3278512.57 frames. utt_duration=1215 frames, utt_pad_proportion=0.06187, over 10808.54 utterances.], batch size: 56, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 21:01:34,655 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:02:12,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6712, 5.0028, 4.8398, 4.9979, 5.0467, 4.6987, 3.4405, 4.9808], device='cuda:3'), covar=tensor([0.0112, 0.0115, 0.0149, 0.0077, 0.0116, 0.0123, 0.0700, 0.0231], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0084, 0.0107, 0.0066, 0.0073, 0.0082, 0.0101, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:02:12,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 21:02:36,993 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 1.913e+02 2.227e+02 2.675e+02 5.227e+02, threshold=4.453e+02, percent-clipped=0.0 2023-03-08 21:02:43,922 INFO [train2.py:809] (3/4) Epoch 20, batch 2150, loss[ctc_loss=0.08017, att_loss=0.2441, loss=0.2113, over 17342.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.0222, over 59.00 utterances.], tot_loss[ctc_loss=0.07695, att_loss=0.2376, loss=0.2055, over 3281840.02 frames. utt_duration=1228 frames, utt_pad_proportion=0.05885, over 10704.40 utterances.], batch size: 59, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 21:03:24,180 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:04:04,240 INFO [train2.py:809] (3/4) Epoch 20, batch 2200, loss[ctc_loss=0.05569, att_loss=0.2077, loss=0.1773, over 16174.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006263, over 41.00 utterances.], tot_loss[ctc_loss=0.07698, att_loss=0.2375, loss=0.2054, over 3276272.79 frames. utt_duration=1218 frames, utt_pad_proportion=0.06113, over 10769.53 utterances.], batch size: 41, lr: 5.39e-03, grad_scale: 8.0 2023-03-08 21:04:41,057 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:04:49,641 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9695, 3.9695, 3.7902, 2.7771, 3.7425, 3.7950, 3.5742, 2.7000], device='cuda:3'), covar=tensor([0.0126, 0.0126, 0.0262, 0.0914, 0.0146, 0.0369, 0.0318, 0.1260], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0100, 0.0101, 0.0109, 0.0083, 0.0109, 0.0097, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 21:05:03,619 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:05:10,185 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:05:17,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.298e+02 1.939e+02 2.291e+02 2.587e+02 5.643e+02, threshold=4.582e+02, percent-clipped=3.0 2023-03-08 21:05:23,449 INFO [train2.py:809] (3/4) Epoch 20, batch 2250, loss[ctc_loss=0.063, att_loss=0.2217, loss=0.19, over 16195.00 frames. utt_duration=1582 frames, utt_pad_proportion=0.005902, over 41.00 utterances.], tot_loss[ctc_loss=0.0766, att_loss=0.2366, loss=0.2046, over 3269349.15 frames. utt_duration=1233 frames, utt_pad_proportion=0.05951, over 10615.52 utterances.], batch size: 41, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:06:07,310 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1133, 4.3277, 4.3354, 4.4146, 2.6797, 4.3839, 2.6146, 1.6358], device='cuda:3'), covar=tensor([0.0425, 0.0253, 0.0648, 0.0252, 0.1683, 0.0199, 0.1548, 0.1903], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0158, 0.0259, 0.0154, 0.0220, 0.0139, 0.0230, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:06:11,731 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:06:19,586 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:06:26,481 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:06:43,280 INFO [train2.py:809] (3/4) Epoch 20, batch 2300, loss[ctc_loss=0.07854, att_loss=0.2373, loss=0.2056, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006827, over 45.00 utterances.], tot_loss[ctc_loss=0.07649, att_loss=0.2367, loss=0.2047, over 3266165.36 frames. utt_duration=1219 frames, utt_pad_proportion=0.06455, over 10727.94 utterances.], batch size: 45, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:07:53,495 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:08:01,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.189e+02 2.492e+02 2.915e+02 6.479e+02, threshold=4.984e+02, percent-clipped=5.0 2023-03-08 21:08:07,070 INFO [train2.py:809] (3/4) Epoch 20, batch 2350, loss[ctc_loss=0.08334, att_loss=0.2605, loss=0.225, over 17308.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02426, over 59.00 utterances.], tot_loss[ctc_loss=0.07668, att_loss=0.2371, loss=0.205, over 3268875.11 frames. utt_duration=1229 frames, utt_pad_proportion=0.06102, over 10652.21 utterances.], batch size: 59, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:08:09,662 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-08 21:08:55,199 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6858, 5.1071, 4.9066, 4.9331, 5.1408, 4.7560, 3.4693, 5.0952], device='cuda:3'), covar=tensor([0.0109, 0.0110, 0.0120, 0.0090, 0.0084, 0.0108, 0.0728, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0084, 0.0106, 0.0066, 0.0072, 0.0082, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:09:26,275 INFO [train2.py:809] (3/4) Epoch 20, batch 2400, loss[ctc_loss=0.07612, att_loss=0.2416, loss=0.2085, over 16992.00 frames. utt_duration=688.2 frames, utt_pad_proportion=0.1344, over 99.00 utterances.], tot_loss[ctc_loss=0.07675, att_loss=0.237, loss=0.2049, over 3263395.76 frames. utt_duration=1204 frames, utt_pad_proportion=0.06682, over 10856.96 utterances.], batch size: 99, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:09:30,276 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5504, 2.3108, 2.4406, 2.2761, 2.6575, 2.5756, 2.5130, 3.1194], device='cuda:3'), covar=tensor([0.1702, 0.3170, 0.2406, 0.1845, 0.1619, 0.1220, 0.2380, 0.1267], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0124, 0.0117, 0.0107, 0.0119, 0.0104, 0.0128, 0.0096], device='cuda:3'), out_proj_covar=tensor([8.7213e-05, 9.5115e-05, 9.3168e-05, 8.3408e-05, 8.8123e-05, 8.3290e-05, 9.5687e-05, 7.7088e-05], device='cuda:3') 2023-03-08 21:10:11,807 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5335, 2.7011, 5.1091, 3.9329, 2.8419, 4.2620, 4.8444, 4.5841], device='cuda:3'), covar=tensor([0.0273, 0.1702, 0.0155, 0.0954, 0.1984, 0.0263, 0.0141, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0244, 0.0181, 0.0312, 0.0268, 0.0211, 0.0169, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:10:39,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.004e+02 2.481e+02 3.189e+02 5.932e+02, threshold=4.963e+02, percent-clipped=2.0 2023-03-08 21:10:46,375 INFO [train2.py:809] (3/4) Epoch 20, batch 2450, loss[ctc_loss=0.09038, att_loss=0.2545, loss=0.2217, over 17285.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01269, over 55.00 utterances.], tot_loss[ctc_loss=0.07681, att_loss=0.2369, loss=0.2049, over 3263643.84 frames. utt_duration=1193 frames, utt_pad_proportion=0.06897, over 10960.01 utterances.], batch size: 55, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:11:44,858 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0933, 2.5978, 3.3861, 2.7321, 3.3273, 4.1761, 4.0422, 2.9398], device='cuda:3'), covar=tensor([0.0406, 0.2026, 0.1517, 0.1415, 0.1144, 0.0947, 0.0690, 0.1403], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0244, 0.0281, 0.0217, 0.0268, 0.0362, 0.0261, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:12:06,745 INFO [train2.py:809] (3/4) Epoch 20, batch 2500, loss[ctc_loss=0.07671, att_loss=0.2335, loss=0.2022, over 16282.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007204, over 43.00 utterances.], tot_loss[ctc_loss=0.07648, att_loss=0.2373, loss=0.2052, over 3270597.89 frames. utt_duration=1214 frames, utt_pad_proportion=0.06236, over 10787.73 utterances.], batch size: 43, lr: 5.38e-03, grad_scale: 8.0 2023-03-08 21:13:12,198 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:13:21,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 1.956e+02 2.237e+02 2.783e+02 5.149e+02, threshold=4.474e+02, percent-clipped=1.0 2023-03-08 21:13:27,753 INFO [train2.py:809] (3/4) Epoch 20, batch 2550, loss[ctc_loss=0.07593, att_loss=0.2453, loss=0.2115, over 17045.00 frames. utt_duration=690.2 frames, utt_pad_proportion=0.134, over 99.00 utterances.], tot_loss[ctc_loss=0.07534, att_loss=0.236, loss=0.2039, over 3268277.29 frames. utt_duration=1239 frames, utt_pad_proportion=0.05594, over 10562.05 utterances.], batch size: 99, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:14:15,806 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:14:47,852 INFO [train2.py:809] (3/4) Epoch 20, batch 2600, loss[ctc_loss=0.08838, att_loss=0.2526, loss=0.2198, over 17067.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.007968, over 52.00 utterances.], tot_loss[ctc_loss=0.07526, att_loss=0.2368, loss=0.2045, over 3280208.48 frames. utt_duration=1251 frames, utt_pad_proportion=0.05126, over 10496.75 utterances.], batch size: 52, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:14:49,877 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:14:52,759 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7590, 6.0521, 5.4953, 5.8404, 5.7124, 5.1831, 5.4085, 5.2929], device='cuda:3'), covar=tensor([0.1397, 0.0870, 0.0966, 0.0744, 0.0881, 0.1607, 0.2418, 0.2338], device='cuda:3'), in_proj_covar=tensor([0.0509, 0.0593, 0.0449, 0.0444, 0.0419, 0.0454, 0.0599, 0.0511], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 21:15:32,448 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:15:53,591 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:16:00,909 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.986e+02 2.327e+02 2.880e+02 6.602e+02, threshold=4.653e+02, percent-clipped=4.0 2023-03-08 21:16:03,371 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8000, 5.1247, 4.6479, 5.1843, 4.5310, 4.7832, 5.2628, 5.0461], device='cuda:3'), covar=tensor([0.0612, 0.0346, 0.0851, 0.0346, 0.0462, 0.0310, 0.0231, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0315, 0.0360, 0.0336, 0.0313, 0.0235, 0.0294, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 21:16:07,836 INFO [train2.py:809] (3/4) Epoch 20, batch 2650, loss[ctc_loss=0.0634, att_loss=0.2358, loss=0.2013, over 16263.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007045, over 43.00 utterances.], tot_loss[ctc_loss=0.07503, att_loss=0.2359, loss=0.2038, over 3262803.65 frames. utt_duration=1261 frames, utt_pad_proportion=0.05228, over 10360.41 utterances.], batch size: 43, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:16:37,401 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8005, 5.0689, 4.5523, 5.1238, 4.4871, 4.7238, 5.2163, 4.9790], device='cuda:3'), covar=tensor([0.0577, 0.0337, 0.0921, 0.0322, 0.0465, 0.0347, 0.0230, 0.0217], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0316, 0.0360, 0.0336, 0.0314, 0.0235, 0.0294, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 21:17:10,073 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:17:27,556 INFO [train2.py:809] (3/4) Epoch 20, batch 2700, loss[ctc_loss=0.08068, att_loss=0.2361, loss=0.205, over 16277.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007478, over 43.00 utterances.], tot_loss[ctc_loss=0.07476, att_loss=0.2358, loss=0.2036, over 3265825.59 frames. utt_duration=1273 frames, utt_pad_proportion=0.04962, over 10271.76 utterances.], batch size: 43, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:17:47,275 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8730, 5.2363, 5.4543, 5.2849, 5.3769, 5.8602, 5.1804, 5.9325], device='cuda:3'), covar=tensor([0.0797, 0.0725, 0.0809, 0.1267, 0.1792, 0.0922, 0.0739, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0863, 0.0504, 0.0587, 0.0654, 0.0858, 0.0609, 0.0479, 0.0595], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:18:41,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 1.853e+02 2.270e+02 2.670e+02 9.199e+02, threshold=4.540e+02, percent-clipped=5.0 2023-03-08 21:18:47,884 INFO [train2.py:809] (3/4) Epoch 20, batch 2750, loss[ctc_loss=0.1092, att_loss=0.2642, loss=0.2332, over 13955.00 frames. utt_duration=378.7 frames, utt_pad_proportion=0.3333, over 148.00 utterances.], tot_loss[ctc_loss=0.07505, att_loss=0.2363, loss=0.2041, over 3274231.68 frames. utt_duration=1252 frames, utt_pad_proportion=0.05253, over 10469.87 utterances.], batch size: 148, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:19:17,756 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:20:08,383 INFO [train2.py:809] (3/4) Epoch 20, batch 2800, loss[ctc_loss=0.06463, att_loss=0.2127, loss=0.1831, over 15521.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007396, over 36.00 utterances.], tot_loss[ctc_loss=0.07617, att_loss=0.2374, loss=0.2051, over 3278865.75 frames. utt_duration=1212 frames, utt_pad_proportion=0.06002, over 10831.75 utterances.], batch size: 36, lr: 5.37e-03, grad_scale: 8.0 2023-03-08 21:20:29,534 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 21:20:33,925 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2023-03-08 21:20:55,814 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:21:21,798 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.103e+02 2.542e+02 3.045e+02 6.490e+02, threshold=5.083e+02, percent-clipped=4.0 2023-03-08 21:21:28,203 INFO [train2.py:809] (3/4) Epoch 20, batch 2850, loss[ctc_loss=0.06519, att_loss=0.2396, loss=0.2047, over 16793.00 frames. utt_duration=1401 frames, utt_pad_proportion=0.005268, over 48.00 utterances.], tot_loss[ctc_loss=0.07627, att_loss=0.2368, loss=0.2047, over 3273319.43 frames. utt_duration=1237 frames, utt_pad_proportion=0.0558, over 10595.97 utterances.], batch size: 48, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:21:51,255 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:22:42,491 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:22:48,504 INFO [train2.py:809] (3/4) Epoch 20, batch 2900, loss[ctc_loss=0.08603, att_loss=0.2529, loss=0.2195, over 17443.00 frames. utt_duration=1013 frames, utt_pad_proportion=0.04569, over 69.00 utterances.], tot_loss[ctc_loss=0.07624, att_loss=0.2366, loss=0.2045, over 3274902.96 frames. utt_duration=1246 frames, utt_pad_proportion=0.05343, over 10522.15 utterances.], batch size: 69, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:23:02,348 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3444, 2.3715, 3.5483, 2.8136, 3.3643, 4.5385, 4.4586, 2.9303], device='cuda:3'), covar=tensor([0.0492, 0.2289, 0.1068, 0.1637, 0.1018, 0.0693, 0.0506, 0.1657], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0242, 0.0276, 0.0214, 0.0262, 0.0357, 0.0256, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:23:30,362 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:24:02,855 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 1.965e+02 2.501e+02 3.006e+02 5.766e+02, threshold=5.002e+02, percent-clipped=1.0 2023-03-08 21:24:08,865 INFO [train2.py:809] (3/4) Epoch 20, batch 2950, loss[ctc_loss=0.07061, att_loss=0.2325, loss=0.2001, over 16478.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005981, over 46.00 utterances.], tot_loss[ctc_loss=0.0753, att_loss=0.2359, loss=0.2038, over 3273314.23 frames. utt_duration=1247 frames, utt_pad_proportion=0.05391, over 10510.28 utterances.], batch size: 46, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:24:42,851 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6991, 3.4489, 3.4360, 3.0679, 3.4567, 3.5272, 3.4913, 2.5333], device='cuda:3'), covar=tensor([0.1147, 0.1831, 0.3006, 0.3616, 0.1415, 0.1767, 0.1079, 0.4544], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0185, 0.0198, 0.0254, 0.0156, 0.0259, 0.0175, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:24:49,618 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1449, 2.7751, 3.2186, 4.3451, 3.8004, 3.8238, 2.8370, 2.0029], device='cuda:3'), covar=tensor([0.0804, 0.1903, 0.0974, 0.0515, 0.0803, 0.0470, 0.1433, 0.2310], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0214, 0.0187, 0.0213, 0.0215, 0.0174, 0.0200, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:24:58,871 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8735, 4.3390, 4.2518, 4.5438, 2.6325, 4.5268, 2.5752, 1.6471], device='cuda:3'), covar=tensor([0.0484, 0.0249, 0.0722, 0.0322, 0.1813, 0.0191, 0.1670, 0.1871], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0159, 0.0259, 0.0154, 0.0220, 0.0139, 0.0230, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:25:28,470 INFO [train2.py:809] (3/4) Epoch 20, batch 3000, loss[ctc_loss=0.06771, att_loss=0.2261, loss=0.1944, over 16022.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.00692, over 40.00 utterances.], tot_loss[ctc_loss=0.07582, att_loss=0.2365, loss=0.2044, over 3268554.13 frames. utt_duration=1214 frames, utt_pad_proportion=0.06282, over 10784.45 utterances.], batch size: 40, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:25:28,470 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 21:25:42,048 INFO [train2.py:843] (3/4) Epoch 20, validation: ctc_loss=0.04025, att_loss=0.2341, loss=0.1953, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 21:25:42,049 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 21:26:22,198 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8758, 5.2017, 5.4528, 5.2504, 5.3552, 5.8330, 5.1504, 5.9176], device='cuda:3'), covar=tensor([0.0670, 0.0691, 0.0786, 0.1258, 0.1688, 0.0846, 0.0708, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0869, 0.0508, 0.0592, 0.0659, 0.0870, 0.0611, 0.0482, 0.0603], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:26:55,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.875e+02 2.186e+02 2.641e+02 4.823e+02, threshold=4.373e+02, percent-clipped=0.0 2023-03-08 21:27:02,055 INFO [train2.py:809] (3/4) Epoch 20, batch 3050, loss[ctc_loss=0.1246, att_loss=0.268, loss=0.2393, over 14124.00 frames. utt_duration=388.4 frames, utt_pad_proportion=0.3245, over 146.00 utterances.], tot_loss[ctc_loss=0.07493, att_loss=0.2361, loss=0.2038, over 3272341.50 frames. utt_duration=1230 frames, utt_pad_proportion=0.05871, over 10658.85 utterances.], batch size: 146, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:28:14,749 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6080, 4.8677, 4.4483, 4.8941, 4.3926, 4.4976, 4.9576, 4.8043], device='cuda:3'), covar=tensor([0.0574, 0.0313, 0.0780, 0.0347, 0.0424, 0.0399, 0.0248, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0385, 0.0317, 0.0362, 0.0338, 0.0315, 0.0236, 0.0297, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 21:28:22,268 INFO [train2.py:809] (3/4) Epoch 20, batch 3100, loss[ctc_loss=0.0633, att_loss=0.2341, loss=0.1999, over 16416.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006981, over 44.00 utterances.], tot_loss[ctc_loss=0.0746, att_loss=0.2361, loss=0.2038, over 3275951.64 frames. utt_duration=1236 frames, utt_pad_proportion=0.0562, over 10616.38 utterances.], batch size: 44, lr: 5.36e-03, grad_scale: 16.0 2023-03-08 21:28:42,333 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4234, 2.3745, 4.9279, 3.6572, 2.9519, 4.0692, 4.5785, 4.4923], device='cuda:3'), covar=tensor([0.0245, 0.1780, 0.0131, 0.1030, 0.1702, 0.0277, 0.0161, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0241, 0.0179, 0.0309, 0.0263, 0.0211, 0.0169, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:28:55,814 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-08 21:29:01,282 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:29:03,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-08 21:29:32,037 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4440, 2.3830, 4.9374, 3.7270, 3.0645, 4.0690, 4.5957, 4.5690], device='cuda:3'), covar=tensor([0.0241, 0.1743, 0.0130, 0.0903, 0.1525, 0.0269, 0.0146, 0.0231], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0240, 0.0178, 0.0307, 0.0261, 0.0209, 0.0168, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:29:36,351 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.977e+02 2.476e+02 2.944e+02 6.319e+02, threshold=4.952e+02, percent-clipped=6.0 2023-03-08 21:29:42,606 INFO [train2.py:809] (3/4) Epoch 20, batch 3150, loss[ctc_loss=0.07512, att_loss=0.2464, loss=0.2121, over 16847.00 frames. utt_duration=682.3 frames, utt_pad_proportion=0.1428, over 99.00 utterances.], tot_loss[ctc_loss=0.07452, att_loss=0.2359, loss=0.2036, over 3277133.38 frames. utt_duration=1228 frames, utt_pad_proportion=0.05754, over 10691.76 utterances.], batch size: 99, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:30:55,900 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:31:01,819 INFO [train2.py:809] (3/4) Epoch 20, batch 3200, loss[ctc_loss=0.06433, att_loss=0.219, loss=0.1881, over 15865.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.0105, over 39.00 utterances.], tot_loss[ctc_loss=0.07441, att_loss=0.2362, loss=0.2038, over 3283724.12 frames. utt_duration=1250 frames, utt_pad_proportion=0.05046, over 10522.74 utterances.], batch size: 39, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:31:25,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 21:31:34,111 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:32:12,197 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:32:15,122 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.045e+02 2.242e+02 2.704e+02 5.175e+02, threshold=4.484e+02, percent-clipped=1.0 2023-03-08 21:32:21,325 INFO [train2.py:809] (3/4) Epoch 20, batch 3250, loss[ctc_loss=0.06004, att_loss=0.2205, loss=0.1884, over 15796.00 frames. utt_duration=1664 frames, utt_pad_proportion=0.007238, over 38.00 utterances.], tot_loss[ctc_loss=0.07424, att_loss=0.2368, loss=0.2043, over 3295292.16 frames. utt_duration=1269 frames, utt_pad_proportion=0.04331, over 10401.22 utterances.], batch size: 38, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:32:58,077 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:33:40,330 INFO [train2.py:809] (3/4) Epoch 20, batch 3300, loss[ctc_loss=0.08393, att_loss=0.2491, loss=0.2161, over 17281.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01168, over 55.00 utterances.], tot_loss[ctc_loss=0.07443, att_loss=0.2365, loss=0.2041, over 3282617.04 frames. utt_duration=1275 frames, utt_pad_proportion=0.04526, over 10311.84 utterances.], batch size: 55, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:33:43,782 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1012, 5.4030, 4.9963, 5.4579, 4.8533, 5.0788, 5.5423, 5.3375], device='cuda:3'), covar=tensor([0.0572, 0.0242, 0.0747, 0.0293, 0.0387, 0.0206, 0.0211, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0316, 0.0362, 0.0338, 0.0315, 0.0235, 0.0297, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 21:34:13,029 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1530, 5.0950, 4.8437, 2.4385, 2.0256, 3.0021, 2.8472, 3.7097], device='cuda:3'), covar=tensor([0.0673, 0.0272, 0.0277, 0.4771, 0.5811, 0.2389, 0.3029, 0.1933], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0270, 0.0265, 0.0241, 0.0345, 0.0335, 0.0252, 0.0367], device='cuda:3'), out_proj_covar=tensor([1.4972e-04, 9.9967e-05, 1.1275e-04, 1.0354e-04, 1.4434e-04, 1.3131e-04, 1.0093e-04, 1.4874e-04], device='cuda:3') 2023-03-08 21:34:35,278 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:34:54,087 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 1.921e+02 2.242e+02 2.810e+02 6.510e+02, threshold=4.483e+02, percent-clipped=2.0 2023-03-08 21:35:00,360 INFO [train2.py:809] (3/4) Epoch 20, batch 3350, loss[ctc_loss=0.05371, att_loss=0.2229, loss=0.189, over 16393.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007409, over 44.00 utterances.], tot_loss[ctc_loss=0.07487, att_loss=0.2362, loss=0.2039, over 3268471.98 frames. utt_duration=1251 frames, utt_pad_proportion=0.05477, over 10462.04 utterances.], batch size: 44, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:36:21,445 INFO [train2.py:809] (3/4) Epoch 20, batch 3400, loss[ctc_loss=0.0685, att_loss=0.2453, loss=0.2099, over 17218.00 frames. utt_duration=1169 frames, utt_pad_proportion=0.02856, over 59.00 utterances.], tot_loss[ctc_loss=0.07511, att_loss=0.2363, loss=0.2041, over 3265600.84 frames. utt_duration=1242 frames, utt_pad_proportion=0.05815, over 10529.09 utterances.], batch size: 59, lr: 5.35e-03, grad_scale: 16.0 2023-03-08 21:36:59,667 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:37:12,437 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-08 21:37:14,546 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8040, 4.0265, 3.7387, 4.1258, 2.7834, 4.0915, 2.7637, 1.9092], device='cuda:3'), covar=tensor([0.0493, 0.0274, 0.0924, 0.0312, 0.1591, 0.0261, 0.1517, 0.1807], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0163, 0.0263, 0.0157, 0.0223, 0.0142, 0.0233, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:37:34,908 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 1.976e+02 2.387e+02 2.813e+02 1.211e+03, threshold=4.774e+02, percent-clipped=8.0 2023-03-08 21:37:41,720 INFO [train2.py:809] (3/4) Epoch 20, batch 3450, loss[ctc_loss=0.09777, att_loss=0.2475, loss=0.2176, over 16954.00 frames. utt_duration=686.5 frames, utt_pad_proportion=0.1398, over 99.00 utterances.], tot_loss[ctc_loss=0.07484, att_loss=0.2362, loss=0.2039, over 3267119.30 frames. utt_duration=1251 frames, utt_pad_proportion=0.05506, over 10456.17 utterances.], batch size: 99, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:38:16,636 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:38:48,994 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-08 21:39:01,285 INFO [train2.py:809] (3/4) Epoch 20, batch 3500, loss[ctc_loss=0.05474, att_loss=0.2081, loss=0.1774, over 15638.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008143, over 37.00 utterances.], tot_loss[ctc_loss=0.07554, att_loss=0.2367, loss=0.2045, over 3276607.81 frames. utt_duration=1250 frames, utt_pad_proportion=0.05256, over 10496.64 utterances.], batch size: 37, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:39:06,382 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7414, 2.5580, 5.1121, 4.1102, 3.0337, 4.3398, 4.9571, 4.8265], device='cuda:3'), covar=tensor([0.0170, 0.1450, 0.0132, 0.0900, 0.1706, 0.0182, 0.0091, 0.0181], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0241, 0.0179, 0.0309, 0.0263, 0.0210, 0.0169, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:39:15,992 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7581, 2.3549, 2.6144, 2.8442, 2.8285, 2.9820, 2.5149, 3.2633], device='cuda:3'), covar=tensor([0.1569, 0.2793, 0.2057, 0.1489, 0.1751, 0.0977, 0.2318, 0.1484], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0122, 0.0116, 0.0106, 0.0119, 0.0102, 0.0124, 0.0096], device='cuda:3'), out_proj_covar=tensor([8.5328e-05, 9.3771e-05, 9.2043e-05, 8.2165e-05, 8.8193e-05, 8.1967e-05, 9.3098e-05, 7.6558e-05], device='cuda:3') 2023-03-08 21:39:33,908 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:40:06,328 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8199, 6.0954, 5.5086, 5.8110, 5.7281, 5.2827, 5.5115, 5.2333], device='cuda:3'), covar=tensor([0.1124, 0.0779, 0.0921, 0.0784, 0.0908, 0.1383, 0.2147, 0.2316], device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0595, 0.0451, 0.0444, 0.0420, 0.0453, 0.0603, 0.0517], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 21:40:15,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 1.933e+02 2.262e+02 2.788e+02 9.048e+02, threshold=4.523e+02, percent-clipped=2.0 2023-03-08 21:40:22,331 INFO [train2.py:809] (3/4) Epoch 20, batch 3550, loss[ctc_loss=0.07977, att_loss=0.2457, loss=0.2125, over 16682.00 frames. utt_duration=675.4 frames, utt_pad_proportion=0.1525, over 99.00 utterances.], tot_loss[ctc_loss=0.07527, att_loss=0.2364, loss=0.2041, over 3280217.23 frames. utt_duration=1267 frames, utt_pad_proportion=0.04727, over 10364.87 utterances.], batch size: 99, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:40:51,180 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:41:07,153 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5964, 2.2883, 2.3922, 2.5547, 2.7175, 2.7455, 2.4367, 3.1250], device='cuda:3'), covar=tensor([0.1743, 0.3347, 0.2479, 0.1615, 0.2221, 0.1113, 0.2231, 0.1253], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0123, 0.0117, 0.0107, 0.0121, 0.0104, 0.0125, 0.0097], device='cuda:3'), out_proj_covar=tensor([8.6741e-05, 9.5201e-05, 9.3339e-05, 8.3480e-05, 8.9688e-05, 8.3202e-05, 9.4389e-05, 7.7724e-05], device='cuda:3') 2023-03-08 21:41:25,836 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0798, 4.3507, 4.1016, 4.5541, 2.6887, 4.4684, 2.7204, 1.9419], device='cuda:3'), covar=tensor([0.0432, 0.0215, 0.0744, 0.0199, 0.1749, 0.0183, 0.1568, 0.1755], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0161, 0.0258, 0.0154, 0.0219, 0.0140, 0.0229, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:41:31,946 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:41:42,764 INFO [train2.py:809] (3/4) Epoch 20, batch 3600, loss[ctc_loss=0.09945, att_loss=0.2575, loss=0.2259, over 17304.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01093, over 55.00 utterances.], tot_loss[ctc_loss=0.07609, att_loss=0.2374, loss=0.2051, over 3284488.46 frames. utt_duration=1250 frames, utt_pad_proportion=0.0519, over 10522.32 utterances.], batch size: 55, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:42:19,623 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9862, 5.3585, 4.8854, 5.3566, 4.7376, 5.0353, 5.4404, 5.2453], device='cuda:3'), covar=tensor([0.0565, 0.0290, 0.0693, 0.0306, 0.0435, 0.0238, 0.0223, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0317, 0.0362, 0.0341, 0.0315, 0.0236, 0.0299, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 21:42:29,137 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:42:39,936 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-03-08 21:42:48,955 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:42:56,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.246e+02 1.889e+02 2.390e+02 3.056e+02 6.577e+02, threshold=4.779e+02, percent-clipped=3.0 2023-03-08 21:43:03,008 INFO [train2.py:809] (3/4) Epoch 20, batch 3650, loss[ctc_loss=0.07337, att_loss=0.2268, loss=0.1961, over 16177.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.005771, over 41.00 utterances.], tot_loss[ctc_loss=0.07561, att_loss=0.237, loss=0.2047, over 3276886.17 frames. utt_duration=1263 frames, utt_pad_proportion=0.04966, over 10386.97 utterances.], batch size: 41, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:43:09,359 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:43:37,010 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:43:37,873 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-08 21:43:54,667 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:44:21,502 INFO [train2.py:809] (3/4) Epoch 20, batch 3700, loss[ctc_loss=0.07104, att_loss=0.2197, loss=0.19, over 15521.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.00692, over 36.00 utterances.], tot_loss[ctc_loss=0.07616, att_loss=0.2373, loss=0.2051, over 3276520.94 frames. utt_duration=1268 frames, utt_pad_proportion=0.04812, over 10345.23 utterances.], batch size: 36, lr: 5.34e-03, grad_scale: 16.0 2023-03-08 21:44:25,012 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:44:53,631 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0847, 5.1058, 4.8333, 2.8365, 4.7458, 4.6663, 4.1531, 2.5255], device='cuda:3'), covar=tensor([0.0118, 0.0090, 0.0265, 0.1057, 0.0101, 0.0207, 0.0339, 0.1510], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0102, 0.0102, 0.0110, 0.0083, 0.0111, 0.0099, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 21:45:14,893 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:45:32,721 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:45:33,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-08 21:45:35,344 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 1.959e+02 2.392e+02 2.997e+02 5.823e+02, threshold=4.784e+02, percent-clipped=2.0 2023-03-08 21:45:41,556 INFO [train2.py:809] (3/4) Epoch 20, batch 3750, loss[ctc_loss=0.07502, att_loss=0.2417, loss=0.2083, over 17490.00 frames. utt_duration=1016 frames, utt_pad_proportion=0.04103, over 69.00 utterances.], tot_loss[ctc_loss=0.07606, att_loss=0.2376, loss=0.2053, over 3279987.69 frames. utt_duration=1256 frames, utt_pad_proportion=0.04993, over 10457.38 utterances.], batch size: 69, lr: 5.33e-03, grad_scale: 16.0 2023-03-08 21:45:48,093 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1169, 4.5514, 4.6362, 4.8371, 2.8841, 4.6842, 2.9923, 1.9103], device='cuda:3'), covar=tensor([0.0431, 0.0283, 0.0579, 0.0177, 0.1582, 0.0185, 0.1322, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0163, 0.0261, 0.0157, 0.0223, 0.0142, 0.0232, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:46:14,896 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6454, 3.7813, 3.1697, 3.1376, 3.8461, 3.6188, 2.5150, 4.1386], device='cuda:3'), covar=tensor([0.1265, 0.0516, 0.1013, 0.0870, 0.0822, 0.0674, 0.1179, 0.0564], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0214, 0.0223, 0.0196, 0.0273, 0.0236, 0.0199, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 21:47:01,108 INFO [train2.py:809] (3/4) Epoch 20, batch 3800, loss[ctc_loss=0.0854, att_loss=0.2424, loss=0.211, over 17379.00 frames. utt_duration=1009 frames, utt_pad_proportion=0.04909, over 69.00 utterances.], tot_loss[ctc_loss=0.07611, att_loss=0.2377, loss=0.2054, over 3286468.95 frames. utt_duration=1276 frames, utt_pad_proportion=0.04316, over 10310.98 utterances.], batch size: 69, lr: 5.33e-03, grad_scale: 16.0 2023-03-08 21:48:15,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 1.856e+02 2.318e+02 2.933e+02 8.070e+02, threshold=4.637e+02, percent-clipped=4.0 2023-03-08 21:48:20,697 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2961, 5.2925, 5.0064, 3.1701, 4.9831, 4.8922, 4.5803, 3.0207], device='cuda:3'), covar=tensor([0.0067, 0.0083, 0.0241, 0.0883, 0.0094, 0.0166, 0.0260, 0.1154], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0101, 0.0101, 0.0109, 0.0083, 0.0110, 0.0098, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 21:48:21,945 INFO [train2.py:809] (3/4) Epoch 20, batch 3850, loss[ctc_loss=0.05988, att_loss=0.2249, loss=0.1919, over 15873.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01006, over 39.00 utterances.], tot_loss[ctc_loss=0.07554, att_loss=0.2368, loss=0.2045, over 3275103.40 frames. utt_duration=1260 frames, utt_pad_proportion=0.05029, over 10413.01 utterances.], batch size: 39, lr: 5.33e-03, grad_scale: 16.0 2023-03-08 21:49:02,639 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9651, 6.1971, 5.6247, 5.9340, 5.7852, 5.2871, 5.5882, 5.4340], device='cuda:3'), covar=tensor([0.1187, 0.0851, 0.0922, 0.0735, 0.1073, 0.1552, 0.2133, 0.2282], device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0602, 0.0455, 0.0447, 0.0429, 0.0457, 0.0608, 0.0524], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 21:49:29,042 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:49:39,448 INFO [train2.py:809] (3/4) Epoch 20, batch 3900, loss[ctc_loss=0.1077, att_loss=0.2583, loss=0.2282, over 16477.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006132, over 46.00 utterances.], tot_loss[ctc_loss=0.07545, att_loss=0.2367, loss=0.2045, over 3273331.73 frames. utt_duration=1256 frames, utt_pad_proportion=0.05229, over 10439.37 utterances.], batch size: 46, lr: 5.33e-03, grad_scale: 16.0 2023-03-08 21:50:23,873 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:50:36,224 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0679, 4.4515, 4.3461, 4.7505, 2.9416, 4.4855, 2.6961, 1.6733], device='cuda:3'), covar=tensor([0.0430, 0.0217, 0.0656, 0.0174, 0.1537, 0.0188, 0.1537, 0.1750], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0162, 0.0259, 0.0155, 0.0221, 0.0140, 0.0229, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:50:50,109 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 1.965e+02 2.336e+02 2.819e+02 5.688e+02, threshold=4.673e+02, percent-clipped=3.0 2023-03-08 21:50:54,940 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 21:50:56,230 INFO [train2.py:809] (3/4) Epoch 20, batch 3950, loss[ctc_loss=0.07095, att_loss=0.2074, loss=0.1801, over 15486.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009222, over 36.00 utterances.], tot_loss[ctc_loss=0.07606, att_loss=0.2375, loss=0.2052, over 3279380.77 frames. utt_duration=1239 frames, utt_pad_proportion=0.05418, over 10597.05 utterances.], batch size: 36, lr: 5.33e-03, grad_scale: 16.0 2023-03-08 21:51:02,669 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:51:15,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-08 21:51:19,643 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:51:38,220 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:52:14,305 INFO [train2.py:809] (3/4) Epoch 21, batch 0, loss[ctc_loss=0.05788, att_loss=0.2017, loss=0.1729, over 15750.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009798, over 38.00 utterances.], tot_loss[ctc_loss=0.05788, att_loss=0.2017, loss=0.1729, over 15750.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009798, over 38.00 utterances.], batch size: 38, lr: 5.20e-03, grad_scale: 16.0 2023-03-08 21:52:14,305 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 21:52:26,402 INFO [train2.py:843] (3/4) Epoch 21, validation: ctc_loss=0.04229, att_loss=0.2351, loss=0.1965, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 21:52:26,403 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 21:52:47,110 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:53:33,517 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:53:36,524 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:53:46,011 INFO [train2.py:809] (3/4) Epoch 21, batch 50, loss[ctc_loss=0.1426, att_loss=0.266, loss=0.2413, over 13666.00 frames. utt_duration=375.8 frames, utt_pad_proportion=0.3453, over 146.00 utterances.], tot_loss[ctc_loss=0.08011, att_loss=0.2399, loss=0.2079, over 735956.20 frames. utt_duration=1066 frames, utt_pad_proportion=0.1037, over 2765.41 utterances.], batch size: 146, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 21:53:54,362 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 21:54:05,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 1.975e+02 2.497e+02 3.178e+02 8.121e+02, threshold=4.994e+02, percent-clipped=4.0 2023-03-08 21:54:11,402 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.41 vs. limit=5.0 2023-03-08 21:54:43,432 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:55:05,928 INFO [train2.py:809] (3/4) Epoch 21, batch 100, loss[ctc_loss=0.1594, att_loss=0.2797, loss=0.2556, over 14270.00 frames. utt_duration=389.8 frames, utt_pad_proportion=0.3173, over 147.00 utterances.], tot_loss[ctc_loss=0.07862, att_loss=0.2378, loss=0.206, over 1291389.34 frames. utt_duration=1137 frames, utt_pad_proportion=0.08725, over 4548.32 utterances.], batch size: 147, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 21:56:22,725 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:56:26,926 INFO [train2.py:809] (3/4) Epoch 21, batch 150, loss[ctc_loss=0.07305, att_loss=0.2482, loss=0.2132, over 16769.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.00651, over 48.00 utterances.], tot_loss[ctc_loss=0.0768, att_loss=0.2372, loss=0.2051, over 1729914.77 frames. utt_duration=1155 frames, utt_pad_proportion=0.08139, over 5996.48 utterances.], batch size: 48, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 21:56:46,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 1.968e+02 2.419e+02 2.959e+02 7.390e+02, threshold=4.838e+02, percent-clipped=3.0 2023-03-08 21:57:22,637 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6977, 2.5077, 3.9093, 3.5284, 2.9756, 3.6932, 3.6887, 3.8035], device='cuda:3'), covar=tensor([0.0307, 0.1317, 0.0178, 0.0727, 0.1261, 0.0285, 0.0250, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0238, 0.0179, 0.0307, 0.0261, 0.0209, 0.0169, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:57:47,567 INFO [train2.py:809] (3/4) Epoch 21, batch 200, loss[ctc_loss=0.0667, att_loss=0.2352, loss=0.2015, over 16136.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005633, over 42.00 utterances.], tot_loss[ctc_loss=0.07572, att_loss=0.2363, loss=0.2042, over 2064863.27 frames. utt_duration=1162 frames, utt_pad_proportion=0.08225, over 7119.93 utterances.], batch size: 42, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 21:58:01,344 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-08 21:58:06,756 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1252, 5.4326, 5.6590, 5.4799, 5.6870, 6.0615, 5.1972, 6.1738], device='cuda:3'), covar=tensor([0.0653, 0.0657, 0.0771, 0.1320, 0.1564, 0.0852, 0.0684, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0861, 0.0508, 0.0587, 0.0653, 0.0859, 0.0613, 0.0481, 0.0600], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 21:58:18,483 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9799, 4.3172, 4.3015, 4.7282, 2.6735, 4.4529, 2.7578, 1.7640], device='cuda:3'), covar=tensor([0.0485, 0.0263, 0.0701, 0.0185, 0.1705, 0.0201, 0.1445, 0.1784], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0163, 0.0260, 0.0156, 0.0223, 0.0141, 0.0230, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 21:58:35,174 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2741, 5.2255, 5.0169, 2.7198, 2.2033, 3.1578, 2.5946, 3.9952], device='cuda:3'), covar=tensor([0.0643, 0.0355, 0.0280, 0.4829, 0.5474, 0.2370, 0.3742, 0.1667], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0266, 0.0263, 0.0241, 0.0340, 0.0330, 0.0250, 0.0362], device='cuda:3'), out_proj_covar=tensor([1.4825e-04, 9.8679e-05, 1.1190e-04, 1.0294e-04, 1.4252e-04, 1.2926e-04, 1.0017e-04, 1.4718e-04], device='cuda:3') 2023-03-08 21:59:08,383 INFO [train2.py:809] (3/4) Epoch 21, batch 250, loss[ctc_loss=0.05867, att_loss=0.2348, loss=0.1996, over 16615.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005947, over 47.00 utterances.], tot_loss[ctc_loss=0.07513, att_loss=0.2364, loss=0.2041, over 2339039.17 frames. utt_duration=1185 frames, utt_pad_proportion=0.07199, over 7904.64 utterances.], batch size: 47, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 21:59:28,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.404e+02 2.056e+02 2.417e+02 2.822e+02 7.152e+02, threshold=4.834e+02, percent-clipped=3.0 2023-03-08 21:59:33,267 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 21:59:33,378 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 22:00:29,281 INFO [train2.py:809] (3/4) Epoch 21, batch 300, loss[ctc_loss=0.06749, att_loss=0.2125, loss=0.1835, over 16182.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006305, over 41.00 utterances.], tot_loss[ctc_loss=0.07487, att_loss=0.236, loss=0.2038, over 2541655.92 frames. utt_duration=1190 frames, utt_pad_proportion=0.07257, over 8552.33 utterances.], batch size: 41, lr: 5.19e-03, grad_scale: 16.0 2023-03-08 22:00:49,787 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 22:00:49,913 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:01:32,423 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:01:43,955 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:01:53,324 INFO [train2.py:809] (3/4) Epoch 21, batch 350, loss[ctc_loss=0.06972, att_loss=0.2189, loss=0.189, over 15866.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01044, over 39.00 utterances.], tot_loss[ctc_loss=0.0751, att_loss=0.2367, loss=0.2044, over 2710368.78 frames. utt_duration=1198 frames, utt_pad_proportion=0.06802, over 9061.60 utterances.], batch size: 39, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:01:59,665 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-08 22:02:02,044 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 22:02:11,092 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:02:12,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 1.969e+02 2.385e+02 2.918e+02 9.167e+02, threshold=4.769e+02, percent-clipped=4.0 2023-03-08 22:03:01,914 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:03:14,455 INFO [train2.py:809] (3/4) Epoch 21, batch 400, loss[ctc_loss=0.109, att_loss=0.265, loss=0.2338, over 13988.00 frames. utt_duration=384.8 frames, utt_pad_proportion=0.3284, over 146.00 utterances.], tot_loss[ctc_loss=0.07565, att_loss=0.2371, loss=0.2048, over 2826150.66 frames. utt_duration=1158 frames, utt_pad_proportion=0.08139, over 9774.69 utterances.], batch size: 146, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:03:19,129 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-08 22:03:19,728 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 22:03:26,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.00 vs. limit=5.0 2023-03-08 22:04:22,270 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:04:24,727 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-08 22:04:34,493 INFO [train2.py:809] (3/4) Epoch 21, batch 450, loss[ctc_loss=0.0842, att_loss=0.2522, loss=0.2186, over 17121.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01508, over 56.00 utterances.], tot_loss[ctc_loss=0.07513, att_loss=0.2367, loss=0.2044, over 2928109.34 frames. utt_duration=1189 frames, utt_pad_proportion=0.07104, over 9862.80 utterances.], batch size: 56, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:04:53,682 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 2.042e+02 2.548e+02 3.452e+02 1.189e+03, threshold=5.096e+02, percent-clipped=8.0 2023-03-08 22:05:01,281 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2102, 5.1559, 5.0989, 2.4131, 2.0213, 2.9507, 2.6929, 3.9427], device='cuda:3'), covar=tensor([0.0682, 0.0332, 0.0209, 0.4645, 0.5931, 0.2554, 0.3021, 0.1487], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0269, 0.0264, 0.0242, 0.0342, 0.0331, 0.0250, 0.0363], device='cuda:3'), out_proj_covar=tensor([1.4895e-04, 9.9660e-05, 1.1244e-04, 1.0351e-04, 1.4308e-04, 1.2975e-04, 1.0004e-04, 1.4766e-04], device='cuda:3') 2023-03-08 22:05:17,516 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7143, 3.2313, 3.7528, 3.0820, 3.7091, 4.7264, 4.6007, 3.4846], device='cuda:3'), covar=tensor([0.0319, 0.1494, 0.1162, 0.1487, 0.1019, 0.0805, 0.0539, 0.1161], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0246, 0.0280, 0.0220, 0.0268, 0.0365, 0.0262, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:05:55,605 INFO [train2.py:809] (3/4) Epoch 21, batch 500, loss[ctc_loss=0.06075, att_loss=0.2367, loss=0.2015, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006496, over 48.00 utterances.], tot_loss[ctc_loss=0.07471, att_loss=0.2362, loss=0.2039, over 3007895.09 frames. utt_duration=1213 frames, utt_pad_proportion=0.06462, over 9933.57 utterances.], batch size: 48, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:07:16,769 INFO [train2.py:809] (3/4) Epoch 21, batch 550, loss[ctc_loss=0.05137, att_loss=0.2137, loss=0.1812, over 15939.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007233, over 41.00 utterances.], tot_loss[ctc_loss=0.07392, att_loss=0.2363, loss=0.2038, over 3074263.59 frames. utt_duration=1208 frames, utt_pad_proportion=0.0622, over 10194.27 utterances.], batch size: 41, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:07:35,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.287e+01 1.925e+02 2.164e+02 2.477e+02 4.088e+02, threshold=4.327e+02, percent-clipped=0.0 2023-03-08 22:07:41,490 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:08:04,058 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5863, 1.8716, 2.2058, 2.4411, 2.5382, 2.6056, 2.4071, 3.1286], device='cuda:3'), covar=tensor([0.2473, 0.3958, 0.2739, 0.1611, 0.2156, 0.1363, 0.2683, 0.1116], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0123, 0.0117, 0.0108, 0.0121, 0.0105, 0.0127, 0.0098], device='cuda:3'), out_proj_covar=tensor([8.7704e-05, 9.5275e-05, 9.3749e-05, 8.4051e-05, 8.9873e-05, 8.4357e-05, 9.5492e-05, 7.7885e-05], device='cuda:3') 2023-03-08 22:08:37,660 INFO [train2.py:809] (3/4) Epoch 21, batch 600, loss[ctc_loss=0.07022, att_loss=0.2329, loss=0.2004, over 16120.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.00648, over 42.00 utterances.], tot_loss[ctc_loss=0.07361, att_loss=0.2361, loss=0.2036, over 3121013.87 frames. utt_duration=1233 frames, utt_pad_proportion=0.05612, over 10135.75 utterances.], batch size: 42, lr: 5.18e-03, grad_scale: 16.0 2023-03-08 22:08:49,141 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:08:59,370 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:08:59,426 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7249, 6.0434, 5.5225, 5.7918, 5.6210, 5.2601, 5.3573, 5.1636], device='cuda:3'), covar=tensor([0.1181, 0.0841, 0.0885, 0.0790, 0.0909, 0.1542, 0.2602, 0.2314], device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0607, 0.0457, 0.0449, 0.0429, 0.0466, 0.0610, 0.0524], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:09:38,461 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:09:58,602 INFO [train2.py:809] (3/4) Epoch 21, batch 650, loss[ctc_loss=0.0668, att_loss=0.2312, loss=0.1983, over 16779.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.00583, over 48.00 utterances.], tot_loss[ctc_loss=0.07453, att_loss=0.2369, loss=0.2044, over 3160374.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.05788, over 10335.81 utterances.], batch size: 48, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:10:18,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.849e+02 2.223e+02 2.876e+02 6.376e+02, threshold=4.446e+02, percent-clipped=5.0 2023-03-08 22:10:27,859 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:10:55,597 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:11:19,498 INFO [train2.py:809] (3/4) Epoch 21, batch 700, loss[ctc_loss=0.0681, att_loss=0.2424, loss=0.2075, over 17329.00 frames. utt_duration=867.9 frames, utt_pad_proportion=0.08161, over 80.00 utterances.], tot_loss[ctc_loss=0.07422, att_loss=0.2362, loss=0.2038, over 3177014.85 frames. utt_duration=1220 frames, utt_pad_proportion=0.06235, over 10428.55 utterances.], batch size: 80, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:11:21,434 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8801, 2.6922, 3.4253, 2.6850, 3.3123, 4.0093, 3.8951, 2.8641], device='cuda:3'), covar=tensor([0.0395, 0.1750, 0.1056, 0.1393, 0.0970, 0.0930, 0.0606, 0.1347], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0243, 0.0276, 0.0216, 0.0265, 0.0361, 0.0259, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:11:46,959 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:12:08,349 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9716, 3.9953, 3.7969, 2.7752, 3.8286, 3.8027, 3.5146, 2.6653], device='cuda:3'), covar=tensor([0.0109, 0.0143, 0.0292, 0.0875, 0.0124, 0.0362, 0.0335, 0.1254], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0102, 0.0103, 0.0110, 0.0084, 0.0111, 0.0099, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:12:28,253 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:12:29,724 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3128, 2.4712, 3.4827, 2.6603, 3.3358, 4.5015, 4.4099, 2.7417], device='cuda:3'), covar=tensor([0.0522, 0.2433, 0.1215, 0.1799, 0.1123, 0.0789, 0.0592, 0.1868], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0245, 0.0278, 0.0217, 0.0265, 0.0363, 0.0260, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:12:37,509 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5122, 2.8870, 3.5869, 2.9356, 3.4955, 4.5553, 4.3674, 3.2153], device='cuda:3'), covar=tensor([0.0332, 0.1964, 0.1367, 0.1445, 0.1124, 0.0970, 0.0639, 0.1359], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0245, 0.0278, 0.0217, 0.0265, 0.0363, 0.0260, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:12:40,237 INFO [train2.py:809] (3/4) Epoch 21, batch 750, loss[ctc_loss=0.0678, att_loss=0.2294, loss=0.1971, over 16390.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.008462, over 44.00 utterances.], tot_loss[ctc_loss=0.07454, att_loss=0.2363, loss=0.204, over 3201028.22 frames. utt_duration=1220 frames, utt_pad_proportion=0.06149, over 10509.58 utterances.], batch size: 44, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:12:59,014 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 1.906e+02 2.332e+02 2.965e+02 6.816e+02, threshold=4.664e+02, percent-clipped=5.0 2023-03-08 22:13:25,154 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:13:39,568 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:13:44,272 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:13:59,495 INFO [train2.py:809] (3/4) Epoch 21, batch 800, loss[ctc_loss=0.1309, att_loss=0.2652, loss=0.2383, over 13989.00 frames. utt_duration=387.6 frames, utt_pad_proportion=0.3271, over 145.00 utterances.], tot_loss[ctc_loss=0.07514, att_loss=0.2368, loss=0.2044, over 3220300.29 frames. utt_duration=1226 frames, utt_pad_proportion=0.06011, over 10521.96 utterances.], batch size: 145, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:14:29,063 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:15:17,441 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:15:20,315 INFO [train2.py:809] (3/4) Epoch 21, batch 850, loss[ctc_loss=0.08637, att_loss=0.252, loss=0.2189, over 17004.00 frames. utt_duration=688.5 frames, utt_pad_proportion=0.1372, over 99.00 utterances.], tot_loss[ctc_loss=0.07521, att_loss=0.2372, loss=0.2048, over 3235514.76 frames. utt_duration=1224 frames, utt_pad_proportion=0.05931, over 10588.87 utterances.], batch size: 99, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:15:30,064 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4139, 2.5157, 4.9207, 3.8266, 2.9414, 4.1523, 4.7008, 4.5248], device='cuda:3'), covar=tensor([0.0254, 0.1667, 0.0169, 0.0917, 0.1839, 0.0266, 0.0153, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0240, 0.0182, 0.0310, 0.0261, 0.0211, 0.0172, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 22:15:40,971 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 2.073e+02 2.361e+02 2.912e+02 1.019e+03, threshold=4.723e+02, percent-clipped=4.0 2023-03-08 22:16:07,050 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:16:40,200 INFO [train2.py:809] (3/4) Epoch 21, batch 900, loss[ctc_loss=0.05974, att_loss=0.2205, loss=0.1883, over 14586.00 frames. utt_duration=1825 frames, utt_pad_proportion=0.0425, over 32.00 utterances.], tot_loss[ctc_loss=0.07519, att_loss=0.237, loss=0.2047, over 3246224.21 frames. utt_duration=1247 frames, utt_pad_proportion=0.05419, over 10427.80 utterances.], batch size: 32, lr: 5.17e-03, grad_scale: 16.0 2023-03-08 22:16:41,890 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9967, 6.2846, 5.7661, 6.0148, 5.9639, 5.4593, 5.6820, 5.4277], device='cuda:3'), covar=tensor([0.1441, 0.0828, 0.1003, 0.0677, 0.0814, 0.1452, 0.2396, 0.2197], device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0602, 0.0455, 0.0449, 0.0427, 0.0462, 0.0605, 0.0518], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:17:08,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-08 22:17:57,987 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6280, 4.8895, 4.4026, 4.7571, 4.5707, 4.1764, 4.4186, 4.1663], device='cuda:3'), covar=tensor([0.1334, 0.1161, 0.1097, 0.0968, 0.1138, 0.1574, 0.2177, 0.2408], device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0601, 0.0453, 0.0448, 0.0423, 0.0461, 0.0602, 0.0517], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 22:18:00,930 INFO [train2.py:809] (3/4) Epoch 21, batch 950, loss[ctc_loss=0.06891, att_loss=0.2441, loss=0.209, over 16491.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005258, over 46.00 utterances.], tot_loss[ctc_loss=0.07499, att_loss=0.2371, loss=0.2047, over 3253856.28 frames. utt_duration=1245 frames, utt_pad_proportion=0.05433, over 10470.14 utterances.], batch size: 46, lr: 5.16e-03, grad_scale: 16.0 2023-03-08 22:18:22,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 1.900e+02 2.361e+02 2.857e+02 6.538e+02, threshold=4.723e+02, percent-clipped=3.0 2023-03-08 22:18:22,787 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:19:22,689 INFO [train2.py:809] (3/4) Epoch 21, batch 1000, loss[ctc_loss=0.07749, att_loss=0.2381, loss=0.206, over 17497.00 frames. utt_duration=887.4 frames, utt_pad_proportion=0.06691, over 79.00 utterances.], tot_loss[ctc_loss=0.0747, att_loss=0.2371, loss=0.2046, over 3264156.42 frames. utt_duration=1252 frames, utt_pad_proportion=0.0511, over 10438.38 utterances.], batch size: 79, lr: 5.16e-03, grad_scale: 8.0 2023-03-08 22:20:30,967 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-08 22:20:44,011 INFO [train2.py:809] (3/4) Epoch 21, batch 1050, loss[ctc_loss=0.05636, att_loss=0.2165, loss=0.1844, over 15938.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.008099, over 41.00 utterances.], tot_loss[ctc_loss=0.07372, att_loss=0.2361, loss=0.2036, over 3266107.39 frames. utt_duration=1265 frames, utt_pad_proportion=0.04839, over 10338.96 utterances.], batch size: 41, lr: 5.16e-03, grad_scale: 4.0 2023-03-08 22:21:00,967 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:21:08,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.786e+02 2.169e+02 2.535e+02 5.351e+02, threshold=4.338e+02, percent-clipped=1.0 2023-03-08 22:21:21,079 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:21:30,522 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-08 22:22:00,966 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:22:05,372 INFO [train2.py:809] (3/4) Epoch 21, batch 1100, loss[ctc_loss=0.07236, att_loss=0.2358, loss=0.2031, over 17340.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03623, over 63.00 utterances.], tot_loss[ctc_loss=0.07453, att_loss=0.2362, loss=0.2039, over 3258514.08 frames. utt_duration=1243 frames, utt_pad_proportion=0.0578, over 10496.18 utterances.], batch size: 63, lr: 5.16e-03, grad_scale: 4.0 2023-03-08 22:22:39,455 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:22:40,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-08 22:23:16,403 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:23:27,022 INFO [train2.py:809] (3/4) Epoch 21, batch 1150, loss[ctc_loss=0.09085, att_loss=0.2465, loss=0.2154, over 17410.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04743, over 69.00 utterances.], tot_loss[ctc_loss=0.0753, att_loss=0.2365, loss=0.2042, over 3257703.32 frames. utt_duration=1224 frames, utt_pad_proportion=0.06336, over 10659.62 utterances.], batch size: 69, lr: 5.16e-03, grad_scale: 4.0 2023-03-08 22:23:40,088 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:23:51,463 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 2.024e+02 2.478e+02 3.117e+02 8.099e+02, threshold=4.955e+02, percent-clipped=10.0 2023-03-08 22:24:06,586 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:24:48,751 INFO [train2.py:809] (3/4) Epoch 21, batch 1200, loss[ctc_loss=0.07099, att_loss=0.2304, loss=0.1985, over 16406.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.00754, over 44.00 utterances.], tot_loss[ctc_loss=0.07544, att_loss=0.2366, loss=0.2044, over 3257464.90 frames. utt_duration=1220 frames, utt_pad_proportion=0.06493, over 10696.40 utterances.], batch size: 44, lr: 5.16e-03, grad_scale: 8.0 2023-03-08 22:25:41,327 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1880, 5.4511, 5.4559, 5.4057, 5.5127, 5.4178, 5.1763, 4.9046], device='cuda:3'), covar=tensor([0.0942, 0.0532, 0.0272, 0.0510, 0.0292, 0.0285, 0.0349, 0.0300], device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0361, 0.0344, 0.0357, 0.0419, 0.0428, 0.0354, 0.0394], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:26:04,139 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9818, 4.6610, 4.6699, 2.3300, 2.1567, 2.9196, 2.4596, 3.7593], device='cuda:3'), covar=tensor([0.0675, 0.0279, 0.0244, 0.5437, 0.5339, 0.2508, 0.3539, 0.1375], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0273, 0.0267, 0.0245, 0.0344, 0.0335, 0.0253, 0.0365], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-08 22:26:09,928 INFO [train2.py:809] (3/4) Epoch 21, batch 1250, loss[ctc_loss=0.07189, att_loss=0.2379, loss=0.2047, over 17277.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01346, over 55.00 utterances.], tot_loss[ctc_loss=0.07577, att_loss=0.2371, loss=0.2049, over 3265074.51 frames. utt_duration=1224 frames, utt_pad_proportion=0.06299, over 10683.95 utterances.], batch size: 55, lr: 5.16e-03, grad_scale: 8.0 2023-03-08 22:26:31,992 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:26:34,823 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.023e+02 2.469e+02 3.077e+02 5.945e+02, threshold=4.937e+02, percent-clipped=1.0 2023-03-08 22:26:51,358 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-08 22:27:33,044 INFO [train2.py:809] (3/4) Epoch 21, batch 1300, loss[ctc_loss=0.0703, att_loss=0.2504, loss=0.2144, over 17017.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008006, over 51.00 utterances.], tot_loss[ctc_loss=0.07537, att_loss=0.2368, loss=0.2045, over 3261171.83 frames. utt_duration=1247 frames, utt_pad_proportion=0.05772, over 10474.50 utterances.], batch size: 51, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:27:50,982 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80987.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:28:12,770 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9648, 5.2727, 4.8208, 5.3293, 4.7153, 4.9305, 5.4242, 5.1715], device='cuda:3'), covar=tensor([0.0629, 0.0286, 0.0821, 0.0324, 0.0414, 0.0258, 0.0217, 0.0186], device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0316, 0.0363, 0.0343, 0.0315, 0.0237, 0.0297, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 22:28:55,447 INFO [train2.py:809] (3/4) Epoch 21, batch 1350, loss[ctc_loss=0.0594, att_loss=0.2075, loss=0.1778, over 15624.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.01011, over 37.00 utterances.], tot_loss[ctc_loss=0.07484, att_loss=0.2363, loss=0.204, over 3264408.38 frames. utt_duration=1257 frames, utt_pad_proportion=0.05464, over 10404.18 utterances.], batch size: 37, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:29:19,939 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.963e+02 2.296e+02 2.713e+02 5.032e+02, threshold=4.592e+02, percent-clipped=2.0 2023-03-08 22:29:21,855 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9251, 5.2229, 4.7317, 5.2725, 4.5882, 4.8743, 5.3231, 5.1065], device='cuda:3'), covar=tensor([0.0580, 0.0289, 0.0850, 0.0294, 0.0487, 0.0265, 0.0241, 0.0202], device='cuda:3'), in_proj_covar=tensor([0.0380, 0.0315, 0.0362, 0.0342, 0.0314, 0.0236, 0.0297, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 22:29:33,235 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81049.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:29:53,276 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7265, 5.1026, 4.9326, 5.1038, 5.1348, 4.7788, 3.1709, 5.0608], device='cuda:3'), covar=tensor([0.0113, 0.0118, 0.0128, 0.0086, 0.0089, 0.0124, 0.0815, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0087, 0.0111, 0.0069, 0.0075, 0.0085, 0.0103, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:30:16,875 INFO [train2.py:809] (3/4) Epoch 21, batch 1400, loss[ctc_loss=0.05833, att_loss=0.2365, loss=0.2009, over 16459.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006483, over 46.00 utterances.], tot_loss[ctc_loss=0.07557, att_loss=0.2371, loss=0.2048, over 3271231.65 frames. utt_duration=1229 frames, utt_pad_proportion=0.05961, over 10659.17 utterances.], batch size: 46, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:30:42,796 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81092.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:30:45,983 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6511, 5.0662, 4.8257, 5.0905, 5.0542, 4.7386, 3.3221, 5.0169], device='cuda:3'), covar=tensor([0.0123, 0.0116, 0.0142, 0.0073, 0.0109, 0.0124, 0.0762, 0.0193], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0087, 0.0111, 0.0070, 0.0075, 0.0085, 0.0104, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:30:50,490 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81097.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:31:22,086 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81116.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:31:26,393 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81119.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:31:36,836 INFO [train2.py:809] (3/4) Epoch 21, batch 1450, loss[ctc_loss=0.09742, att_loss=0.2578, loss=0.2257, over 16892.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006156, over 49.00 utterances.], tot_loss[ctc_loss=0.07486, att_loss=0.2361, loss=0.2038, over 3263680.72 frames. utt_duration=1249 frames, utt_pad_proportion=0.05786, over 10461.37 utterances.], batch size: 49, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:31:41,570 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81129.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:32:00,573 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.791e+02 2.173e+02 2.721e+02 6.345e+02, threshold=4.347e+02, percent-clipped=3.0 2023-03-08 22:32:15,034 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81150.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:32:42,436 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7854, 2.2227, 2.7012, 2.6747, 2.8178, 2.8291, 2.4198, 3.3568], device='cuda:3'), covar=tensor([0.1257, 0.2653, 0.1649, 0.1358, 0.1463, 0.1260, 0.2609, 0.0927], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0124, 0.0117, 0.0109, 0.0122, 0.0106, 0.0130, 0.0098], device='cuda:3'), out_proj_covar=tensor([8.7753e-05, 9.6003e-05, 9.4000e-05, 8.4857e-05, 9.0718e-05, 8.5086e-05, 9.6947e-05, 7.8688e-05], device='cuda:3') 2023-03-08 22:32:43,703 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81167.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:32:57,634 INFO [train2.py:809] (3/4) Epoch 21, batch 1500, loss[ctc_loss=0.05512, att_loss=0.2136, loss=0.1819, over 15630.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009136, over 37.00 utterances.], tot_loss[ctc_loss=0.07367, att_loss=0.2349, loss=0.2027, over 3252495.47 frames. utt_duration=1272 frames, utt_pad_proportion=0.05467, over 10239.51 utterances.], batch size: 37, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:32:59,581 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81177.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:33:33,285 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81198.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:33:56,781 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0640, 5.3428, 5.2543, 5.2189, 5.3345, 5.2966, 5.0167, 4.7799], device='cuda:3'), covar=tensor([0.1112, 0.0539, 0.0340, 0.0555, 0.0343, 0.0335, 0.0402, 0.0352], device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0360, 0.0343, 0.0356, 0.0418, 0.0427, 0.0354, 0.0392], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 22:34:14,439 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0680, 5.3591, 4.9115, 5.4033, 4.7482, 5.0228, 5.4665, 5.2667], device='cuda:3'), covar=tensor([0.0569, 0.0296, 0.0704, 0.0297, 0.0400, 0.0239, 0.0201, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0313, 0.0358, 0.0342, 0.0313, 0.0235, 0.0294, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 22:34:19,136 INFO [train2.py:809] (3/4) Epoch 21, batch 1550, loss[ctc_loss=0.04391, att_loss=0.2095, loss=0.1764, over 15863.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01064, over 39.00 utterances.], tot_loss[ctc_loss=0.0732, att_loss=0.2347, loss=0.2024, over 3257244.03 frames. utt_duration=1291 frames, utt_pad_proportion=0.05017, over 10101.45 utterances.], batch size: 39, lr: 5.15e-03, grad_scale: 8.0 2023-03-08 22:34:43,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 1.828e+02 2.165e+02 2.736e+02 6.423e+02, threshold=4.329e+02, percent-clipped=4.0 2023-03-08 22:35:40,193 INFO [train2.py:809] (3/4) Epoch 21, batch 1600, loss[ctc_loss=0.06645, att_loss=0.2416, loss=0.2066, over 17355.00 frames. utt_duration=702.7 frames, utt_pad_proportion=0.1194, over 99.00 utterances.], tot_loss[ctc_loss=0.07341, att_loss=0.2357, loss=0.2033, over 3269646.14 frames. utt_duration=1285 frames, utt_pad_proportion=0.04668, over 10190.46 utterances.], batch size: 99, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:36:16,927 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81299.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:36:20,812 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81301.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:37:01,169 INFO [train2.py:809] (3/4) Epoch 21, batch 1650, loss[ctc_loss=0.04828, att_loss=0.2123, loss=0.1795, over 16190.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.00627, over 41.00 utterances.], tot_loss[ctc_loss=0.07251, att_loss=0.235, loss=0.2025, over 3275059.81 frames. utt_duration=1303 frames, utt_pad_proportion=0.04139, over 10068.90 utterances.], batch size: 41, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:37:25,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.957e+02 2.223e+02 2.631e+02 7.416e+02, threshold=4.446e+02, percent-clipped=3.0 2023-03-08 22:37:26,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-08 22:37:57,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-08 22:37:58,145 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81360.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:38:01,314 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81362.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:38:23,009 INFO [train2.py:809] (3/4) Epoch 21, batch 1700, loss[ctc_loss=0.06979, att_loss=0.2475, loss=0.212, over 16873.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007884, over 49.00 utterances.], tot_loss[ctc_loss=0.07257, att_loss=0.2357, loss=0.203, over 3281052.57 frames. utt_duration=1274 frames, utt_pad_proportion=0.04619, over 10316.22 utterances.], batch size: 49, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:38:49,091 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81392.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:39:15,021 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9805, 4.0804, 4.0299, 4.0855, 4.1492, 4.1367, 3.9008, 3.8323], device='cuda:3'), covar=tensor([0.0963, 0.0747, 0.0960, 0.0561, 0.0369, 0.0409, 0.0495, 0.0385], device='cuda:3'), in_proj_covar=tensor([0.0517, 0.0359, 0.0339, 0.0355, 0.0417, 0.0425, 0.0352, 0.0390], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 22:39:26,163 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9818, 3.7451, 3.5946, 3.1504, 3.6576, 3.6773, 3.7658, 2.7129], device='cuda:3'), covar=tensor([0.1171, 0.1227, 0.2016, 0.3776, 0.1109, 0.2377, 0.0935, 0.4126], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0183, 0.0194, 0.0247, 0.0153, 0.0254, 0.0173, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 22:39:44,356 INFO [train2.py:809] (3/4) Epoch 21, batch 1750, loss[ctc_loss=0.1103, att_loss=0.2626, loss=0.2321, over 14141.00 frames. utt_duration=388.8 frames, utt_pad_proportion=0.3227, over 146.00 utterances.], tot_loss[ctc_loss=0.07384, att_loss=0.2363, loss=0.2038, over 3279196.36 frames. utt_duration=1240 frames, utt_pad_proportion=0.05491, over 10587.74 utterances.], batch size: 146, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:39:49,373 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81429.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:40:06,988 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81440.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:40:08,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.066e+02 2.468e+02 3.056e+02 6.142e+02, threshold=4.936e+02, percent-clipped=4.0 2023-03-08 22:40:59,672 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81472.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:41:05,526 INFO [train2.py:809] (3/4) Epoch 21, batch 1800, loss[ctc_loss=0.06069, att_loss=0.2106, loss=0.1806, over 14491.00 frames. utt_duration=1813 frames, utt_pad_proportion=0.0443, over 32.00 utterances.], tot_loss[ctc_loss=0.07354, att_loss=0.2354, loss=0.203, over 3272877.44 frames. utt_duration=1254 frames, utt_pad_proportion=0.05303, over 10450.49 utterances.], batch size: 32, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:41:07,217 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81477.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:41:26,902 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7981, 6.0720, 5.5374, 5.7927, 5.7680, 5.2411, 5.5141, 5.2927], device='cuda:3'), covar=tensor([0.1173, 0.0815, 0.0773, 0.0789, 0.0766, 0.1380, 0.2175, 0.2175], device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0610, 0.0455, 0.0454, 0.0429, 0.0466, 0.0607, 0.0528], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:42:25,221 INFO [train2.py:809] (3/4) Epoch 21, batch 1850, loss[ctc_loss=0.1151, att_loss=0.2752, loss=0.2432, over 17289.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01268, over 55.00 utterances.], tot_loss[ctc_loss=0.07404, att_loss=0.236, loss=0.2036, over 3278100.32 frames. utt_duration=1241 frames, utt_pad_proportion=0.05222, over 10578.98 utterances.], batch size: 55, lr: 5.14e-03, grad_scale: 8.0 2023-03-08 22:42:49,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 1.850e+02 2.304e+02 2.710e+02 5.976e+02, threshold=4.608e+02, percent-clipped=1.0 2023-03-08 22:43:34,924 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1770, 5.1908, 4.8982, 3.0891, 4.9001, 4.6795, 4.3575, 2.7741], device='cuda:3'), covar=tensor([0.0111, 0.0084, 0.0266, 0.0910, 0.0105, 0.0206, 0.0307, 0.1362], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0101, 0.0104, 0.0109, 0.0084, 0.0111, 0.0099, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 22:43:45,869 INFO [train2.py:809] (3/4) Epoch 21, batch 1900, loss[ctc_loss=0.07613, att_loss=0.2287, loss=0.1982, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007859, over 43.00 utterances.], tot_loss[ctc_loss=0.07314, att_loss=0.2351, loss=0.2027, over 3273413.87 frames. utt_duration=1265 frames, utt_pad_proportion=0.04751, over 10362.18 utterances.], batch size: 43, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:45:07,136 INFO [train2.py:809] (3/4) Epoch 21, batch 1950, loss[ctc_loss=0.05443, att_loss=0.2413, loss=0.204, over 16485.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005499, over 46.00 utterances.], tot_loss[ctc_loss=0.0737, att_loss=0.2359, loss=0.2035, over 3277212.80 frames. utt_duration=1237 frames, utt_pad_proportion=0.05419, over 10612.43 utterances.], batch size: 46, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:45:31,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 1.984e+02 2.266e+02 2.808e+02 6.592e+02, threshold=4.533e+02, percent-clipped=3.0 2023-03-08 22:45:54,457 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81655.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:45:58,075 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81657.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:46:18,731 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81670.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:46:27,655 INFO [train2.py:809] (3/4) Epoch 21, batch 2000, loss[ctc_loss=0.06809, att_loss=0.2215, loss=0.1908, over 15624.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.008935, over 37.00 utterances.], tot_loss[ctc_loss=0.07344, att_loss=0.2359, loss=0.2034, over 3271450.31 frames. utt_duration=1239 frames, utt_pad_proportion=0.05477, over 10576.64 utterances.], batch size: 37, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:46:50,493 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7447, 2.1903, 2.5875, 2.4665, 2.6719, 2.5681, 2.4693, 3.4819], device='cuda:3'), covar=tensor([0.1737, 0.3430, 0.2451, 0.1974, 0.2080, 0.1870, 0.2950, 0.1224], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0126, 0.0120, 0.0110, 0.0126, 0.0109, 0.0131, 0.0100], device='cuda:3'), out_proj_covar=tensor([8.9365e-05, 9.7672e-05, 9.5847e-05, 8.6260e-05, 9.2872e-05, 8.6956e-05, 9.8433e-05, 8.0118e-05], device='cuda:3') 2023-03-08 22:47:21,265 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7724, 5.0468, 5.3451, 5.1607, 5.2565, 5.7336, 5.0972, 5.8583], device='cuda:3'), covar=tensor([0.0859, 0.0699, 0.0788, 0.1323, 0.1968, 0.0984, 0.0893, 0.0663], device='cuda:3'), in_proj_covar=tensor([0.0883, 0.0516, 0.0602, 0.0671, 0.0886, 0.0637, 0.0492, 0.0613], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:47:47,659 INFO [train2.py:809] (3/4) Epoch 21, batch 2050, loss[ctc_loss=0.06076, att_loss=0.2214, loss=0.1893, over 15956.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006902, over 41.00 utterances.], tot_loss[ctc_loss=0.07399, att_loss=0.2363, loss=0.2039, over 3273633.05 frames. utt_duration=1267 frames, utt_pad_proportion=0.04745, over 10344.91 utterances.], batch size: 41, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:47:55,809 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81731.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 22:48:11,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 1.913e+02 2.441e+02 2.878e+02 1.219e+03, threshold=4.882e+02, percent-clipped=4.0 2023-03-08 22:49:02,516 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81772.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:49:08,389 INFO [train2.py:809] (3/4) Epoch 21, batch 2100, loss[ctc_loss=0.09005, att_loss=0.256, loss=0.2228, over 17045.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008584, over 52.00 utterances.], tot_loss[ctc_loss=0.07482, att_loss=0.2368, loss=0.2044, over 3273557.47 frames. utt_duration=1237 frames, utt_pad_proportion=0.05552, over 10596.79 utterances.], batch size: 52, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:50:20,928 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81820.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:50:30,150 INFO [train2.py:809] (3/4) Epoch 21, batch 2150, loss[ctc_loss=0.08228, att_loss=0.2575, loss=0.2225, over 17279.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.0132, over 55.00 utterances.], tot_loss[ctc_loss=0.07474, att_loss=0.237, loss=0.2046, over 3282944.48 frames. utt_duration=1225 frames, utt_pad_proportion=0.05505, over 10729.43 utterances.], batch size: 55, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:50:39,734 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.31 vs. limit=5.0 2023-03-08 22:50:54,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 1.964e+02 2.354e+02 2.825e+02 8.417e+02, threshold=4.708e+02, percent-clipped=3.0 2023-03-08 22:51:38,811 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3029, 4.8741, 4.9649, 4.9802, 3.5513, 4.7981, 3.5050, 2.5892], device='cuda:3'), covar=tensor([0.0395, 0.0220, 0.0406, 0.0210, 0.1104, 0.0159, 0.1086, 0.1408], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0162, 0.0256, 0.0155, 0.0219, 0.0142, 0.0228, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 22:51:50,725 INFO [train2.py:809] (3/4) Epoch 21, batch 2200, loss[ctc_loss=0.07025, att_loss=0.238, loss=0.2044, over 16687.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006364, over 46.00 utterances.], tot_loss[ctc_loss=0.07479, att_loss=0.2373, loss=0.2048, over 3281132.03 frames. utt_duration=1223 frames, utt_pad_proportion=0.05497, over 10748.52 utterances.], batch size: 46, lr: 5.13e-03, grad_scale: 8.0 2023-03-08 22:53:11,514 INFO [train2.py:809] (3/4) Epoch 21, batch 2250, loss[ctc_loss=0.05475, att_loss=0.227, loss=0.1926, over 16172.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006826, over 41.00 utterances.], tot_loss[ctc_loss=0.07378, att_loss=0.2363, loss=0.2038, over 3285431.04 frames. utt_duration=1247 frames, utt_pad_proportion=0.04844, over 10554.69 utterances.], batch size: 41, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 22:53:16,599 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-08 22:53:35,602 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.106e+02 2.465e+02 3.018e+02 5.655e+02, threshold=4.929e+02, percent-clipped=3.0 2023-03-08 22:53:58,771 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81955.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:53:59,448 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-03-08 22:54:02,674 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81957.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:54:11,557 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-08 22:54:32,963 INFO [train2.py:809] (3/4) Epoch 21, batch 2300, loss[ctc_loss=0.05644, att_loss=0.2209, loss=0.188, over 15961.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006747, over 41.00 utterances.], tot_loss[ctc_loss=0.07344, att_loss=0.2363, loss=0.2038, over 3289524.41 frames. utt_duration=1238 frames, utt_pad_proportion=0.04936, over 10643.82 utterances.], batch size: 41, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 22:54:49,514 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 22:54:50,568 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81987.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:55:22,053 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82003.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:55:25,322 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:55:39,907 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0021, 5.2923, 4.8563, 5.3808, 4.7250, 4.9900, 5.4307, 5.2110], device='cuda:3'), covar=tensor([0.0629, 0.0343, 0.0827, 0.0392, 0.0463, 0.0295, 0.0259, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0384, 0.0318, 0.0361, 0.0346, 0.0318, 0.0237, 0.0297, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 22:55:59,321 INFO [train2.py:809] (3/4) Epoch 21, batch 2350, loss[ctc_loss=0.06877, att_loss=0.2375, loss=0.2038, over 17434.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04538, over 69.00 utterances.], tot_loss[ctc_loss=0.07366, att_loss=0.2361, loss=0.2036, over 3283110.55 frames. utt_duration=1249 frames, utt_pad_proportion=0.04968, over 10526.35 utterances.], batch size: 69, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 22:55:59,545 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82026.0, num_to_drop=1, layers_to_drop={1} 2023-03-08 22:56:22,871 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.390e+02 1.965e+02 2.374e+02 2.963e+02 4.469e+02, threshold=4.748e+02, percent-clipped=0.0 2023-03-08 22:56:35,684 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82048.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:56:47,297 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82055.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:57:20,438 INFO [train2.py:809] (3/4) Epoch 21, batch 2400, loss[ctc_loss=0.09162, att_loss=0.2568, loss=0.2238, over 17311.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01149, over 55.00 utterances.], tot_loss[ctc_loss=0.07317, att_loss=0.2357, loss=0.2032, over 3282690.94 frames. utt_duration=1255 frames, utt_pad_proportion=0.04844, over 10477.26 utterances.], batch size: 55, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 22:58:25,047 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82116.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 22:58:40,651 INFO [train2.py:809] (3/4) Epoch 21, batch 2450, loss[ctc_loss=0.0615, att_loss=0.2105, loss=0.1807, over 15864.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.008583, over 39.00 utterances.], tot_loss[ctc_loss=0.07435, att_loss=0.2361, loss=0.2038, over 3284764.13 frames. utt_duration=1227 frames, utt_pad_proportion=0.05464, over 10719.62 utterances.], batch size: 39, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 22:59:04,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.920e+01 1.856e+02 2.212e+02 2.703e+02 4.819e+02, threshold=4.424e+02, percent-clipped=1.0 2023-03-08 22:59:07,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6196, 4.9580, 4.8471, 4.9326, 5.0567, 4.7095, 3.6275, 4.9635], device='cuda:3'), covar=tensor([0.0132, 0.0116, 0.0133, 0.0083, 0.0098, 0.0127, 0.0652, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0088, 0.0113, 0.0071, 0.0076, 0.0087, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 22:59:46,932 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-08 23:00:00,559 INFO [train2.py:809] (3/4) Epoch 21, batch 2500, loss[ctc_loss=0.07571, att_loss=0.2438, loss=0.2101, over 17107.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01422, over 56.00 utterances.], tot_loss[ctc_loss=0.07483, att_loss=0.2362, loss=0.204, over 3285607.46 frames. utt_duration=1220 frames, utt_pad_proportion=0.05652, over 10788.19 utterances.], batch size: 56, lr: 5.12e-03, grad_scale: 8.0 2023-03-08 23:01:21,096 INFO [train2.py:809] (3/4) Epoch 21, batch 2550, loss[ctc_loss=0.1276, att_loss=0.2635, loss=0.2363, over 14549.00 frames. utt_duration=394.8 frames, utt_pad_proportion=0.3037, over 148.00 utterances.], tot_loss[ctc_loss=0.07471, att_loss=0.236, loss=0.2037, over 3273008.99 frames. utt_duration=1215 frames, utt_pad_proportion=0.06037, over 10791.75 utterances.], batch size: 148, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:01:45,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 1.926e+02 2.375e+02 3.006e+02 1.037e+03, threshold=4.749e+02, percent-clipped=4.0 2023-03-08 23:02:22,892 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7959, 4.2658, 4.5452, 4.3510, 4.4134, 4.7286, 4.3713, 4.7752], device='cuda:3'), covar=tensor([0.0874, 0.0813, 0.0777, 0.1247, 0.1681, 0.0933, 0.2118, 0.0791], device='cuda:3'), in_proj_covar=tensor([0.0880, 0.0518, 0.0605, 0.0677, 0.0886, 0.0640, 0.0493, 0.0614], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:02:42,046 INFO [train2.py:809] (3/4) Epoch 21, batch 2600, loss[ctc_loss=0.05661, att_loss=0.2347, loss=0.1991, over 16972.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007202, over 50.00 utterances.], tot_loss[ctc_loss=0.07445, att_loss=0.2359, loss=0.2036, over 3272597.40 frames. utt_duration=1220 frames, utt_pad_proportion=0.05997, over 10743.00 utterances.], batch size: 50, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:02:45,578 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5785, 2.1509, 2.4383, 2.5831, 2.7392, 2.2366, 2.3042, 3.0823], device='cuda:3'), covar=tensor([0.2978, 0.5932, 0.3565, 0.2756, 0.3156, 0.2769, 0.3748, 0.1930], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0124, 0.0120, 0.0108, 0.0123, 0.0108, 0.0129, 0.0098], device='cuda:3'), out_proj_covar=tensor([8.8027e-05, 9.6070e-05, 9.5416e-05, 8.4623e-05, 9.1412e-05, 8.6084e-05, 9.6901e-05, 7.8933e-05], device='cuda:3') 2023-03-08 23:03:29,410 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9695, 5.2604, 4.8482, 5.3791, 4.6929, 4.9707, 5.4255, 5.1635], device='cuda:3'), covar=tensor([0.0584, 0.0264, 0.0821, 0.0273, 0.0450, 0.0253, 0.0204, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0315, 0.0357, 0.0341, 0.0314, 0.0235, 0.0294, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 23:04:03,377 INFO [train2.py:809] (3/4) Epoch 21, batch 2650, loss[ctc_loss=0.07218, att_loss=0.2226, loss=0.1925, over 16258.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.008031, over 43.00 utterances.], tot_loss[ctc_loss=0.07466, att_loss=0.2361, loss=0.2038, over 3272878.95 frames. utt_duration=1216 frames, utt_pad_proportion=0.06198, over 10781.33 utterances.], batch size: 43, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:04:03,721 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82326.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:04:27,513 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.898e+02 2.205e+02 2.899e+02 5.130e+02, threshold=4.409e+02, percent-clipped=1.0 2023-03-08 23:04:31,613 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82343.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:04:36,573 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82346.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:05:15,692 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9861, 3.6232, 3.1139, 3.2607, 3.8630, 3.4583, 2.8994, 3.9865], device='cuda:3'), covar=tensor([0.0977, 0.0554, 0.1020, 0.0748, 0.0670, 0.0712, 0.0886, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0216, 0.0225, 0.0199, 0.0277, 0.0241, 0.0200, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 23:05:21,686 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82374.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:05:24,749 INFO [train2.py:809] (3/4) Epoch 21, batch 2700, loss[ctc_loss=0.07317, att_loss=0.2321, loss=0.2003, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006152, over 45.00 utterances.], tot_loss[ctc_loss=0.07427, att_loss=0.2357, loss=0.2034, over 3272439.72 frames. utt_duration=1220 frames, utt_pad_proportion=0.06135, over 10746.62 utterances.], batch size: 45, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:06:16,104 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82407.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:06:22,257 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82411.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:06:41,923 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4443, 4.2733, 4.2844, 4.2967, 4.7888, 4.4074, 4.3914, 2.2833], device='cuda:3'), covar=tensor([0.0228, 0.0376, 0.0352, 0.0317, 0.0928, 0.0228, 0.0308, 0.1958], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0186, 0.0185, 0.0203, 0.0365, 0.0157, 0.0174, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:06:46,070 INFO [train2.py:809] (3/4) Epoch 21, batch 2750, loss[ctc_loss=0.06345, att_loss=0.2129, loss=0.183, over 16193.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006163, over 41.00 utterances.], tot_loss[ctc_loss=0.07461, att_loss=0.2358, loss=0.2036, over 3269646.62 frames. utt_duration=1205 frames, utt_pad_proportion=0.06673, over 10867.74 utterances.], batch size: 41, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:07:10,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 1.903e+02 2.360e+02 2.836e+02 7.135e+02, threshold=4.721e+02, percent-clipped=4.0 2023-03-08 23:07:35,116 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5504, 3.0481, 3.4376, 4.5827, 3.9709, 3.9800, 3.0020, 2.2899], device='cuda:3'), covar=tensor([0.0673, 0.1827, 0.0933, 0.0483, 0.0824, 0.0419, 0.1536, 0.2293], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0217, 0.0193, 0.0222, 0.0226, 0.0180, 0.0205, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:08:06,987 INFO [train2.py:809] (3/4) Epoch 21, batch 2800, loss[ctc_loss=0.127, att_loss=0.2597, loss=0.2332, over 14956.00 frames. utt_duration=411.5 frames, utt_pad_proportion=0.2819, over 146.00 utterances.], tot_loss[ctc_loss=0.07479, att_loss=0.2367, loss=0.2043, over 3265207.77 frames. utt_duration=1197 frames, utt_pad_proportion=0.06897, over 10928.42 utterances.], batch size: 146, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:08:34,709 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82493.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 23:09:26,650 INFO [train2.py:809] (3/4) Epoch 21, batch 2850, loss[ctc_loss=0.09178, att_loss=0.2443, loss=0.2138, over 16761.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.00691, over 48.00 utterances.], tot_loss[ctc_loss=0.07567, att_loss=0.2371, loss=0.2048, over 3264128.16 frames. utt_duration=1188 frames, utt_pad_proportion=0.07151, over 11006.55 utterances.], batch size: 48, lr: 5.11e-03, grad_scale: 8.0 2023-03-08 23:09:50,387 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 1.879e+02 2.361e+02 2.971e+02 6.221e+02, threshold=4.723e+02, percent-clipped=6.0 2023-03-08 23:10:11,737 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82554.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 23:10:32,765 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9146, 4.0840, 3.6337, 4.3403, 2.9194, 4.1619, 2.7542, 1.8698], device='cuda:3'), covar=tensor([0.0462, 0.0218, 0.0818, 0.0212, 0.1335, 0.0220, 0.1400, 0.1617], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0164, 0.0258, 0.0157, 0.0223, 0.0143, 0.0231, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:10:35,814 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82569.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:10:42,057 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0568, 4.4237, 3.9612, 4.6761, 2.8827, 4.4768, 2.7200, 1.7307], device='cuda:3'), covar=tensor([0.0453, 0.0231, 0.0782, 0.0201, 0.1473, 0.0207, 0.1409, 0.1688], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0164, 0.0257, 0.0157, 0.0222, 0.0143, 0.0231, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:10:46,922 INFO [train2.py:809] (3/4) Epoch 21, batch 2900, loss[ctc_loss=0.04245, att_loss=0.2012, loss=0.1694, over 15356.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01113, over 35.00 utterances.], tot_loss[ctc_loss=0.07536, att_loss=0.2373, loss=0.2049, over 3270649.07 frames. utt_duration=1184 frames, utt_pad_proportion=0.07003, over 11061.42 utterances.], batch size: 35, lr: 5.10e-03, grad_scale: 8.0 2023-03-08 23:11:17,958 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0163, 4.4019, 3.8692, 4.5772, 2.7193, 4.3854, 2.4319, 1.6937], device='cuda:3'), covar=tensor([0.0509, 0.0269, 0.0995, 0.0359, 0.1713, 0.0221, 0.1813, 0.1866], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0164, 0.0257, 0.0157, 0.0222, 0.0144, 0.0231, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:11:41,017 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4789, 4.3587, 4.5176, 4.4626, 5.0401, 4.5390, 4.5697, 2.2886], device='cuda:3'), covar=tensor([0.0231, 0.0379, 0.0303, 0.0306, 0.0676, 0.0227, 0.0290, 0.1817], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0188, 0.0187, 0.0205, 0.0368, 0.0158, 0.0176, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:11:59,743 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7512, 3.9626, 3.9330, 3.9690, 3.9952, 3.8309, 2.9908, 3.8912], device='cuda:3'), covar=tensor([0.0159, 0.0142, 0.0146, 0.0096, 0.0112, 0.0132, 0.0677, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0088, 0.0112, 0.0070, 0.0076, 0.0087, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:12:07,978 INFO [train2.py:809] (3/4) Epoch 21, batch 2950, loss[ctc_loss=0.09084, att_loss=0.2563, loss=0.2232, over 17124.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01467, over 56.00 utterances.], tot_loss[ctc_loss=0.07502, att_loss=0.237, loss=0.2046, over 3274117.01 frames. utt_duration=1173 frames, utt_pad_proportion=0.07166, over 11176.40 utterances.], batch size: 56, lr: 5.10e-03, grad_scale: 8.0 2023-03-08 23:12:14,617 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82630.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:12:31,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 1.787e+02 2.245e+02 2.585e+02 7.263e+02, threshold=4.491e+02, percent-clipped=3.0 2023-03-08 23:12:36,038 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82643.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:12:48,179 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1344, 2.7381, 3.2077, 4.1561, 3.7727, 3.7378, 2.9388, 2.2617], device='cuda:3'), covar=tensor([0.0811, 0.1900, 0.0885, 0.0614, 0.0819, 0.0471, 0.1399, 0.2092], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0213, 0.0190, 0.0218, 0.0222, 0.0177, 0.0201, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:12:51,403 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5090, 4.4001, 4.5420, 4.5080, 5.1032, 4.6062, 4.5761, 2.4609], device='cuda:3'), covar=tensor([0.0238, 0.0356, 0.0295, 0.0348, 0.0771, 0.0228, 0.0297, 0.1746], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0186, 0.0185, 0.0203, 0.0365, 0.0157, 0.0174, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:13:27,966 INFO [train2.py:809] (3/4) Epoch 21, batch 3000, loss[ctc_loss=0.05735, att_loss=0.2266, loss=0.1927, over 16253.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.008818, over 43.00 utterances.], tot_loss[ctc_loss=0.07485, att_loss=0.2371, loss=0.2047, over 3281307.86 frames. utt_duration=1204 frames, utt_pad_proportion=0.0636, over 10911.39 utterances.], batch size: 43, lr: 5.10e-03, grad_scale: 8.0 2023-03-08 23:13:27,966 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 23:13:41,811 INFO [train2.py:843] (3/4) Epoch 21, validation: ctc_loss=0.04141, att_loss=0.2346, loss=0.196, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 23:13:41,812 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 23:14:05,995 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82691.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:14:24,614 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82702.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:14:38,824 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82711.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:15:02,604 INFO [train2.py:809] (3/4) Epoch 21, batch 3050, loss[ctc_loss=0.08259, att_loss=0.2491, loss=0.2158, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008989, over 51.00 utterances.], tot_loss[ctc_loss=0.07389, att_loss=0.236, loss=0.2036, over 3273940.54 frames. utt_duration=1219 frames, utt_pad_proportion=0.06205, over 10758.67 utterances.], batch size: 51, lr: 5.10e-03, grad_scale: 16.0 2023-03-08 23:15:26,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.870e+01 1.786e+02 2.187e+02 2.749e+02 5.841e+02, threshold=4.374e+02, percent-clipped=4.0 2023-03-08 23:15:55,320 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82759.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:15:58,727 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6168, 3.2700, 3.7267, 3.1297, 3.6498, 4.7519, 4.5637, 3.3291], device='cuda:3'), covar=tensor([0.0402, 0.1564, 0.1185, 0.1419, 0.1046, 0.0934, 0.0594, 0.1321], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0246, 0.0282, 0.0223, 0.0267, 0.0369, 0.0264, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:16:22,503 INFO [train2.py:809] (3/4) Epoch 21, batch 3100, loss[ctc_loss=0.06849, att_loss=0.2103, loss=0.1819, over 15870.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.0102, over 39.00 utterances.], tot_loss[ctc_loss=0.07346, att_loss=0.236, loss=0.2035, over 3273582.48 frames. utt_duration=1251 frames, utt_pad_proportion=0.05353, over 10482.32 utterances.], batch size: 39, lr: 5.10e-03, grad_scale: 16.0 2023-03-08 23:16:42,255 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7013, 5.1717, 5.0308, 4.9894, 5.1838, 4.8424, 3.6110, 5.1571], device='cuda:3'), covar=tensor([0.0113, 0.0096, 0.0106, 0.0076, 0.0080, 0.0094, 0.0639, 0.0157], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0087, 0.0111, 0.0070, 0.0076, 0.0086, 0.0105, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:17:43,783 INFO [train2.py:809] (3/4) Epoch 21, batch 3150, loss[ctc_loss=0.06725, att_loss=0.249, loss=0.2126, over 16635.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004707, over 47.00 utterances.], tot_loss[ctc_loss=0.07441, att_loss=0.2371, loss=0.2046, over 3289135.07 frames. utt_duration=1256 frames, utt_pad_proportion=0.04785, over 10484.12 utterances.], batch size: 47, lr: 5.10e-03, grad_scale: 8.0 2023-03-08 23:18:09,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.159e+02 2.205e+02 2.472e+02 3.065e+02 6.005e+02, threshold=4.943e+02, percent-clipped=5.0 2023-03-08 23:18:21,024 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82849.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 23:18:22,759 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3456, 2.0935, 2.1528, 2.4924, 2.8215, 2.1797, 2.2046, 2.9177], device='cuda:3'), covar=tensor([0.1756, 0.3211, 0.2277, 0.1412, 0.1483, 0.1685, 0.2535, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0125, 0.0120, 0.0108, 0.0123, 0.0105, 0.0129, 0.0098], device='cuda:3'), out_proj_covar=tensor([8.8420e-05, 9.6599e-05, 9.5257e-05, 8.4663e-05, 9.1167e-05, 8.4738e-05, 9.6710e-05, 7.8810e-05], device='cuda:3') 2023-03-08 23:19:02,506 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1951, 5.1256, 4.9332, 3.2076, 5.0152, 4.7921, 4.5429, 3.1246], device='cuda:3'), covar=tensor([0.0102, 0.0099, 0.0304, 0.0881, 0.0090, 0.0179, 0.0263, 0.1174], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0102, 0.0104, 0.0111, 0.0085, 0.0113, 0.0099, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 23:19:03,803 INFO [train2.py:809] (3/4) Epoch 21, batch 3200, loss[ctc_loss=0.07461, att_loss=0.2226, loss=0.193, over 16175.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006642, over 41.00 utterances.], tot_loss[ctc_loss=0.07386, att_loss=0.2363, loss=0.2038, over 3288135.98 frames. utt_duration=1285 frames, utt_pad_proportion=0.04165, over 10244.78 utterances.], batch size: 41, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:19:17,159 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7606, 3.5886, 3.5828, 3.1362, 3.6594, 3.6137, 3.6376, 2.6840], device='cuda:3'), covar=tensor([0.1188, 0.1454, 0.1880, 0.2962, 0.1060, 0.2601, 0.0846, 0.3644], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0186, 0.0198, 0.0250, 0.0156, 0.0257, 0.0178, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:20:23,226 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82925.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:20:24,599 INFO [train2.py:809] (3/4) Epoch 21, batch 3250, loss[ctc_loss=0.0563, att_loss=0.2135, loss=0.182, over 15892.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.007486, over 39.00 utterances.], tot_loss[ctc_loss=0.07354, att_loss=0.2356, loss=0.2032, over 3276972.43 frames. utt_duration=1273 frames, utt_pad_proportion=0.04803, over 10308.70 utterances.], batch size: 39, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:20:50,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.807e+02 2.273e+02 2.911e+02 7.187e+02, threshold=4.546e+02, percent-clipped=3.0 2023-03-08 23:21:12,423 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8928, 4.7640, 4.6979, 2.1976, 1.9599, 2.8306, 2.3494, 3.7494], device='cuda:3'), covar=tensor([0.0819, 0.0272, 0.0241, 0.4709, 0.5724, 0.2606, 0.3665, 0.1564], device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0277, 0.0271, 0.0248, 0.0347, 0.0335, 0.0253, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-08 23:21:24,877 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82963.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:21:45,457 INFO [train2.py:809] (3/4) Epoch 21, batch 3300, loss[ctc_loss=0.07844, att_loss=0.2416, loss=0.209, over 16767.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005745, over 48.00 utterances.], tot_loss[ctc_loss=0.07366, att_loss=0.2361, loss=0.2036, over 3280875.03 frames. utt_duration=1271 frames, utt_pad_proportion=0.04685, over 10337.09 utterances.], batch size: 48, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:22:28,791 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83002.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:22:56,288 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3818, 3.0174, 3.3563, 4.4063, 4.0290, 3.8039, 3.0624, 2.3400], device='cuda:3'), covar=tensor([0.0835, 0.1908, 0.0996, 0.0713, 0.0850, 0.0530, 0.1445, 0.2298], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0213, 0.0190, 0.0219, 0.0222, 0.0178, 0.0201, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:23:04,080 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83024.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:23:07,438 INFO [train2.py:809] (3/4) Epoch 21, batch 3350, loss[ctc_loss=0.08381, att_loss=0.2474, loss=0.2147, over 17329.00 frames. utt_duration=1006 frames, utt_pad_proportion=0.05165, over 69.00 utterances.], tot_loss[ctc_loss=0.07359, att_loss=0.236, loss=0.2035, over 3275715.85 frames. utt_duration=1265 frames, utt_pad_proportion=0.04979, over 10372.19 utterances.], batch size: 69, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:23:33,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 1.819e+02 2.175e+02 2.592e+02 7.580e+02, threshold=4.350e+02, percent-clipped=2.0 2023-03-08 23:23:41,546 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2426, 5.5485, 5.8216, 5.5620, 5.7283, 6.1281, 5.4510, 6.2182], device='cuda:3'), covar=tensor([0.0584, 0.0693, 0.0708, 0.1232, 0.1572, 0.0901, 0.0620, 0.0672], device='cuda:3'), in_proj_covar=tensor([0.0867, 0.0511, 0.0596, 0.0658, 0.0869, 0.0626, 0.0486, 0.0606], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:23:46,014 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83050.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:24:28,077 INFO [train2.py:809] (3/4) Epoch 21, batch 3400, loss[ctc_loss=0.06839, att_loss=0.2332, loss=0.2002, over 16284.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006884, over 43.00 utterances.], tot_loss[ctc_loss=0.07284, att_loss=0.2356, loss=0.203, over 3274455.74 frames. utt_duration=1282 frames, utt_pad_proportion=0.04549, over 10230.97 utterances.], batch size: 43, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:24:56,825 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9528, 5.2114, 5.1618, 5.1633, 5.2540, 5.1988, 4.9245, 4.7235], device='cuda:3'), covar=tensor([0.1101, 0.0568, 0.0362, 0.0521, 0.0320, 0.0323, 0.0418, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0364, 0.0343, 0.0356, 0.0418, 0.0426, 0.0352, 0.0391], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-08 23:25:05,145 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 23:25:28,455 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-08 23:25:47,343 INFO [train2.py:809] (3/4) Epoch 21, batch 3450, loss[ctc_loss=0.07732, att_loss=0.2456, loss=0.2119, over 17361.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.04934, over 69.00 utterances.], tot_loss[ctc_loss=0.0733, att_loss=0.236, loss=0.2035, over 3276388.51 frames. utt_duration=1273 frames, utt_pad_proportion=0.04738, over 10304.03 utterances.], batch size: 69, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:25:55,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-08 23:25:57,219 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5026, 2.6610, 5.0213, 3.9490, 3.0360, 4.2661, 4.7649, 4.6102], device='cuda:3'), covar=tensor([0.0276, 0.1605, 0.0195, 0.0920, 0.1695, 0.0262, 0.0150, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0240, 0.0184, 0.0309, 0.0262, 0.0212, 0.0173, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:26:08,358 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 23:26:13,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 1.851e+02 2.238e+02 2.723e+02 5.748e+02, threshold=4.475e+02, percent-clipped=3.0 2023-03-08 23:26:24,476 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83149.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 23:27:04,260 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.22 vs. limit=5.0 2023-03-08 23:27:06,900 INFO [train2.py:809] (3/4) Epoch 21, batch 3500, loss[ctc_loss=0.08326, att_loss=0.2309, loss=0.2014, over 16024.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006242, over 40.00 utterances.], tot_loss[ctc_loss=0.07293, att_loss=0.2354, loss=0.2029, over 3274570.93 frames. utt_duration=1298 frames, utt_pad_proportion=0.04105, over 10101.38 utterances.], batch size: 40, lr: 5.09e-03, grad_scale: 8.0 2023-03-08 23:27:14,763 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3277, 4.7318, 4.9476, 4.7470, 4.8764, 5.2282, 4.8619, 5.3242], device='cuda:3'), covar=tensor([0.0755, 0.0755, 0.0751, 0.1325, 0.1731, 0.1014, 0.1182, 0.0702], device='cuda:3'), in_proj_covar=tensor([0.0870, 0.0514, 0.0599, 0.0662, 0.0875, 0.0630, 0.0486, 0.0608], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:27:39,129 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83196.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:27:40,352 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83197.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 23:28:25,787 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83225.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:28:26,900 INFO [train2.py:809] (3/4) Epoch 21, batch 3550, loss[ctc_loss=0.09021, att_loss=0.2447, loss=0.2138, over 16273.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007706, over 43.00 utterances.], tot_loss[ctc_loss=0.07262, att_loss=0.235, loss=0.2025, over 3275765.83 frames. utt_duration=1287 frames, utt_pad_proportion=0.04307, over 10192.17 utterances.], batch size: 43, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:28:37,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-08 23:28:53,070 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.836e+02 2.204e+02 2.637e+02 6.921e+02, threshold=4.407e+02, percent-clipped=5.0 2023-03-08 23:29:16,520 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:29:37,967 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.46 vs. limit=5.0 2023-03-08 23:29:41,887 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83273.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:29:47,122 INFO [train2.py:809] (3/4) Epoch 21, batch 3600, loss[ctc_loss=0.05564, att_loss=0.2121, loss=0.1808, over 16011.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007187, over 40.00 utterances.], tot_loss[ctc_loss=0.0726, att_loss=0.2347, loss=0.2023, over 3273492.59 frames. utt_duration=1272 frames, utt_pad_proportion=0.04753, over 10306.36 utterances.], batch size: 40, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:29:58,483 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8686, 4.8515, 4.4791, 2.5472, 4.6788, 4.5611, 3.9124, 2.2151], device='cuda:3'), covar=tensor([0.0173, 0.0142, 0.0385, 0.1489, 0.0137, 0.0283, 0.0558, 0.2439], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0102, 0.0104, 0.0110, 0.0084, 0.0113, 0.0099, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-08 23:30:00,086 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83284.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:30:56,584 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83319.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:31:07,601 INFO [train2.py:809] (3/4) Epoch 21, batch 3650, loss[ctc_loss=0.07496, att_loss=0.2344, loss=0.2025, over 16497.00 frames. utt_duration=1436 frames, utt_pad_proportion=0.004881, over 46.00 utterances.], tot_loss[ctc_loss=0.07362, att_loss=0.2356, loss=0.2032, over 3265239.99 frames. utt_duration=1254 frames, utt_pad_proportion=0.05353, over 10426.71 utterances.], batch size: 46, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:31:14,791 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-08 23:31:33,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 2.085e+02 2.475e+02 3.033e+02 6.314e+02, threshold=4.951e+02, percent-clipped=5.0 2023-03-08 23:31:39,088 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83345.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:32:17,077 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1566, 5.4946, 4.9299, 5.5665, 4.8794, 5.1637, 5.5542, 5.3487], device='cuda:3'), covar=tensor([0.0595, 0.0237, 0.0800, 0.0242, 0.0427, 0.0180, 0.0226, 0.0184], device='cuda:3'), in_proj_covar=tensor([0.0384, 0.0318, 0.0362, 0.0345, 0.0320, 0.0236, 0.0300, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-08 23:32:27,567 INFO [train2.py:809] (3/4) Epoch 21, batch 3700, loss[ctc_loss=0.08064, att_loss=0.2523, loss=0.218, over 17486.00 frames. utt_duration=1015 frames, utt_pad_proportion=0.04337, over 69.00 utterances.], tot_loss[ctc_loss=0.07363, att_loss=0.2352, loss=0.2029, over 3267022.65 frames. utt_duration=1268 frames, utt_pad_proportion=0.0504, over 10321.09 utterances.], batch size: 69, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:32:39,629 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9330, 5.3016, 5.4908, 5.3104, 5.4201, 5.8637, 5.2486, 5.9951], device='cuda:3'), covar=tensor([0.0685, 0.0736, 0.0840, 0.1377, 0.1746, 0.1050, 0.0725, 0.0639], device='cuda:3'), in_proj_covar=tensor([0.0866, 0.0511, 0.0596, 0.0658, 0.0869, 0.0624, 0.0482, 0.0602], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:33:49,261 INFO [train2.py:809] (3/4) Epoch 21, batch 3750, loss[ctc_loss=0.07628, att_loss=0.2406, loss=0.2077, over 17226.00 frames. utt_duration=873.9 frames, utt_pad_proportion=0.08396, over 79.00 utterances.], tot_loss[ctc_loss=0.07429, att_loss=0.2356, loss=0.2033, over 3253710.90 frames. utt_duration=1218 frames, utt_pad_proportion=0.06557, over 10697.52 utterances.], batch size: 79, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:34:15,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.925e+02 2.407e+02 3.142e+02 5.755e+02, threshold=4.814e+02, percent-clipped=4.0 2023-03-08 23:34:30,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1433, 4.5800, 4.5865, 4.8319, 2.8203, 4.6055, 2.8451, 1.6304], device='cuda:3'), covar=tensor([0.0452, 0.0263, 0.0645, 0.0243, 0.1696, 0.0238, 0.1422, 0.1896], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0165, 0.0257, 0.0157, 0.0220, 0.0145, 0.0228, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:35:09,001 INFO [train2.py:809] (3/4) Epoch 21, batch 3800, loss[ctc_loss=0.1329, att_loss=0.2706, loss=0.243, over 13812.00 frames. utt_duration=374.7 frames, utt_pad_proportion=0.3392, over 148.00 utterances.], tot_loss[ctc_loss=0.07435, att_loss=0.235, loss=0.2029, over 3243284.82 frames. utt_duration=1217 frames, utt_pad_proportion=0.06833, over 10671.09 utterances.], batch size: 148, lr: 5.08e-03, grad_scale: 8.0 2023-03-08 23:36:17,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9447, 5.0362, 4.8520, 2.0107, 1.9367, 2.7701, 2.3018, 3.7988], device='cuda:3'), covar=tensor([0.0797, 0.0274, 0.0249, 0.5701, 0.5858, 0.2792, 0.3808, 0.1679], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0277, 0.0271, 0.0245, 0.0343, 0.0334, 0.0252, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-08 23:36:29,147 INFO [train2.py:809] (3/4) Epoch 21, batch 3850, loss[ctc_loss=0.05957, att_loss=0.2222, loss=0.1897, over 16173.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007343, over 41.00 utterances.], tot_loss[ctc_loss=0.07397, att_loss=0.235, loss=0.2028, over 3242303.83 frames. utt_duration=1229 frames, utt_pad_proportion=0.06579, over 10562.34 utterances.], batch size: 41, lr: 5.07e-03, grad_scale: 8.0 2023-03-08 23:36:38,941 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1747, 3.9340, 3.9029, 3.2235, 3.8732, 3.9806, 3.9392, 2.8588], device='cuda:3'), covar=tensor([0.0979, 0.0938, 0.1598, 0.3371, 0.1333, 0.1843, 0.0819, 0.3928], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0190, 0.0200, 0.0255, 0.0159, 0.0260, 0.0181, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:36:53,732 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 2.030e+02 2.457e+02 3.139e+02 8.463e+02, threshold=4.915e+02, percent-clipped=7.0 2023-03-08 23:37:09,557 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83552.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:37:12,966 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5385, 3.0278, 3.6535, 3.1436, 3.6514, 4.6230, 4.4750, 3.4128], device='cuda:3'), covar=tensor([0.0343, 0.1559, 0.1208, 0.1192, 0.1003, 0.0831, 0.0550, 0.1085], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0242, 0.0279, 0.0219, 0.0263, 0.0365, 0.0259, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:37:46,957 INFO [train2.py:809] (3/4) Epoch 21, batch 3900, loss[ctc_loss=0.08203, att_loss=0.2515, loss=0.2176, over 17309.00 frames. utt_duration=1261 frames, utt_pad_proportion=0.01137, over 55.00 utterances.], tot_loss[ctc_loss=0.07437, att_loss=0.2348, loss=0.2027, over 3237250.70 frames. utt_duration=1212 frames, utt_pad_proportion=0.07239, over 10697.42 utterances.], batch size: 55, lr: 5.07e-03, grad_scale: 8.0 2023-03-08 23:38:23,431 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3696, 3.9947, 3.3459, 3.5799, 4.2038, 3.8678, 3.5223, 4.5710], device='cuda:3'), covar=tensor([0.0911, 0.0471, 0.1006, 0.0666, 0.0658, 0.0630, 0.0660, 0.0415], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0213, 0.0224, 0.0195, 0.0273, 0.0238, 0.0196, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-08 23:38:36,406 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1101, 3.8840, 3.8169, 3.1742, 3.8527, 3.9761, 3.8603, 2.7842], device='cuda:3'), covar=tensor([0.0858, 0.0911, 0.1846, 0.2866, 0.1016, 0.1873, 0.0706, 0.3382], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0191, 0.0202, 0.0256, 0.0160, 0.0263, 0.0181, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:38:55,367 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83619.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:39:05,538 INFO [train2.py:809] (3/4) Epoch 21, batch 3950, loss[ctc_loss=0.07159, att_loss=0.2403, loss=0.2065, over 17018.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007846, over 51.00 utterances.], tot_loss[ctc_loss=0.07354, att_loss=0.2342, loss=0.202, over 3243341.66 frames. utt_duration=1245 frames, utt_pad_proportion=0.06239, over 10431.06 utterances.], batch size: 51, lr: 5.07e-03, grad_scale: 8.0 2023-03-08 23:39:27,363 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83640.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:39:30,216 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.299e+02 1.844e+02 2.165e+02 2.793e+02 5.808e+02, threshold=4.331e+02, percent-clipped=4.0 2023-03-08 23:40:16,719 INFO [train2.py:809] (3/4) Epoch 22, batch 0, loss[ctc_loss=0.0745, att_loss=0.2441, loss=0.2102, over 16770.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006288, over 48.00 utterances.], tot_loss[ctc_loss=0.0745, att_loss=0.2441, loss=0.2102, over 16770.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006288, over 48.00 utterances.], batch size: 48, lr: 4.95e-03, grad_scale: 8.0 2023-03-08 23:40:16,719 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-08 23:40:29,652 INFO [train2.py:843] (3/4) Epoch 22, validation: ctc_loss=0.04004, att_loss=0.2341, loss=0.1953, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-08 23:40:29,653 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-08 23:40:41,999 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83667.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:41:48,586 INFO [train2.py:809] (3/4) Epoch 22, batch 50, loss[ctc_loss=0.09328, att_loss=0.2508, loss=0.2193, over 17442.00 frames. utt_duration=1109 frames, utt_pad_proportion=0.03162, over 63.00 utterances.], tot_loss[ctc_loss=0.07279, att_loss=0.2352, loss=0.2027, over 738819.78 frames. utt_duration=1278 frames, utt_pad_proportion=0.05097, over 2315.71 utterances.], batch size: 63, lr: 4.95e-03, grad_scale: 8.0 2023-03-08 23:42:40,550 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.187e+02 1.991e+02 2.354e+02 2.967e+02 6.077e+02, threshold=4.708e+02, percent-clipped=4.0 2023-03-08 23:42:43,951 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1334, 5.4053, 5.6517, 5.4324, 5.6587, 6.0631, 5.3018, 6.1653], device='cuda:3'), covar=tensor([0.0682, 0.0771, 0.0796, 0.1353, 0.1701, 0.0932, 0.0715, 0.0650], device='cuda:3'), in_proj_covar=tensor([0.0872, 0.0511, 0.0595, 0.0664, 0.0875, 0.0626, 0.0485, 0.0606], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:43:08,480 INFO [train2.py:809] (3/4) Epoch 22, batch 100, loss[ctc_loss=0.07374, att_loss=0.2432, loss=0.2093, over 16951.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.006862, over 50.00 utterances.], tot_loss[ctc_loss=0.07294, att_loss=0.235, loss=0.2026, over 1306191.73 frames. utt_duration=1321 frames, utt_pad_proportion=0.03561, over 3958.80 utterances.], batch size: 50, lr: 4.95e-03, grad_scale: 8.0 2023-03-08 23:43:10,910 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83760.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:43:16,945 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9634, 6.1844, 5.6559, 5.9326, 5.8640, 5.3349, 5.7169, 5.3824], device='cuda:3'), covar=tensor([0.1169, 0.0893, 0.0925, 0.0786, 0.0898, 0.1542, 0.2006, 0.2260], device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0602, 0.0456, 0.0452, 0.0423, 0.0463, 0.0604, 0.0517], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-08 23:44:29,204 INFO [train2.py:809] (3/4) Epoch 22, batch 150, loss[ctc_loss=0.07866, att_loss=0.2504, loss=0.216, over 17131.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01458, over 56.00 utterances.], tot_loss[ctc_loss=0.07266, att_loss=0.2362, loss=0.2035, over 1746077.61 frames. utt_duration=1257 frames, utt_pad_proportion=0.05085, over 5561.55 utterances.], batch size: 56, lr: 4.95e-03, grad_scale: 8.0 2023-03-08 23:44:48,446 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83821.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:45:21,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 1.932e+02 2.277e+02 2.985e+02 7.139e+02, threshold=4.555e+02, percent-clipped=2.0 2023-03-08 23:45:37,566 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83852.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:45:49,300 INFO [train2.py:809] (3/4) Epoch 22, batch 200, loss[ctc_loss=0.04808, att_loss=0.2097, loss=0.1774, over 15480.00 frames. utt_duration=1721 frames, utt_pad_proportion=0.01016, over 36.00 utterances.], tot_loss[ctc_loss=0.0721, att_loss=0.2352, loss=0.2025, over 2087770.74 frames. utt_duration=1296 frames, utt_pad_proportion=0.04059, over 6452.89 utterances.], batch size: 36, lr: 4.95e-03, grad_scale: 8.0 2023-03-08 23:46:24,252 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7822, 5.2201, 4.9506, 5.2227, 5.3461, 4.8845, 3.7848, 5.2018], device='cuda:3'), covar=tensor([0.0118, 0.0093, 0.0144, 0.0064, 0.0075, 0.0095, 0.0589, 0.0163], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0086, 0.0109, 0.0069, 0.0075, 0.0084, 0.0103, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:46:53,883 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83900.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:47:08,781 INFO [train2.py:809] (3/4) Epoch 22, batch 250, loss[ctc_loss=0.07159, att_loss=0.2357, loss=0.2028, over 17363.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.04881, over 69.00 utterances.], tot_loss[ctc_loss=0.07256, att_loss=0.2345, loss=0.2021, over 2342505.33 frames. utt_duration=1255 frames, utt_pad_proportion=0.05342, over 7475.59 utterances.], batch size: 69, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:47:26,982 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-03-08 23:47:57,195 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83940.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:47:59,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 1.880e+02 2.225e+02 2.668e+02 4.764e+02, threshold=4.450e+02, percent-clipped=1.0 2023-03-08 23:48:11,525 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83949.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:48:27,877 INFO [train2.py:809] (3/4) Epoch 22, batch 300, loss[ctc_loss=0.07645, att_loss=0.2156, loss=0.1877, over 15515.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.007859, over 36.00 utterances.], tot_loss[ctc_loss=0.07319, att_loss=0.2348, loss=0.2025, over 2545423.73 frames. utt_duration=1259 frames, utt_pad_proportion=0.0535, over 8097.09 utterances.], batch size: 36, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:49:09,341 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83985.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:49:13,747 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83988.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:49:51,793 INFO [train2.py:809] (3/4) Epoch 22, batch 350, loss[ctc_loss=0.05861, att_loss=0.224, loss=0.1909, over 15953.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007073, over 41.00 utterances.], tot_loss[ctc_loss=0.0727, att_loss=0.2342, loss=0.2019, over 2703557.94 frames. utt_duration=1268 frames, utt_pad_proportion=0.05148, over 8536.80 utterances.], batch size: 41, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:49:53,706 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84010.0, num_to_drop=1, layers_to_drop={0} 2023-03-08 23:49:59,931 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4440, 2.9004, 3.6943, 2.9701, 3.5913, 4.6294, 4.4502, 3.2152], device='cuda:3'), covar=tensor([0.0366, 0.1733, 0.1242, 0.1372, 0.1056, 0.0786, 0.0553, 0.1293], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0242, 0.0281, 0.0219, 0.0264, 0.0366, 0.0259, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:50:06,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-08 23:50:43,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 1.981e+02 2.404e+02 2.952e+02 5.790e+02, threshold=4.808e+02, percent-clipped=6.0 2023-03-08 23:50:50,299 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84046.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:50:51,782 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4545, 2.9743, 3.6428, 2.9919, 3.5252, 4.5667, 4.4234, 3.1302], device='cuda:3'), covar=tensor([0.0333, 0.1591, 0.1223, 0.1362, 0.1114, 0.0775, 0.0524, 0.1297], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0243, 0.0282, 0.0219, 0.0264, 0.0368, 0.0260, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:51:11,735 INFO [train2.py:809] (3/4) Epoch 22, batch 400, loss[ctc_loss=0.08363, att_loss=0.2544, loss=0.2202, over 17091.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01669, over 56.00 utterances.], tot_loss[ctc_loss=0.07282, att_loss=0.2351, loss=0.2027, over 2831993.13 frames. utt_duration=1269 frames, utt_pad_proportion=0.04793, over 8935.72 utterances.], batch size: 56, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:51:57,506 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5706, 4.9119, 4.7979, 4.8248, 5.0475, 4.6179, 3.6323, 4.8818], device='cuda:3'), covar=tensor([0.0122, 0.0121, 0.0126, 0.0084, 0.0083, 0.0113, 0.0639, 0.0189], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0086, 0.0109, 0.0069, 0.0075, 0.0084, 0.0103, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:52:12,392 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84097.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:52:31,745 INFO [train2.py:809] (3/4) Epoch 22, batch 450, loss[ctc_loss=0.06301, att_loss=0.2072, loss=0.1784, over 15356.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.009443, over 35.00 utterances.], tot_loss[ctc_loss=0.07299, att_loss=0.2354, loss=0.2029, over 2928556.52 frames. utt_duration=1247 frames, utt_pad_proportion=0.05414, over 9402.63 utterances.], batch size: 35, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:52:42,548 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84116.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:53:22,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 1.943e+02 2.393e+02 2.892e+02 4.484e+02, threshold=4.787e+02, percent-clipped=0.0 2023-03-08 23:53:39,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-08 23:53:50,184 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84158.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:53:51,321 INFO [train2.py:809] (3/4) Epoch 22, batch 500, loss[ctc_loss=0.05693, att_loss=0.2036, loss=0.1743, over 15776.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.007651, over 38.00 utterances.], tot_loss[ctc_loss=0.07322, att_loss=0.2347, loss=0.2024, over 3001384.42 frames. utt_duration=1249 frames, utt_pad_proportion=0.0554, over 9626.61 utterances.], batch size: 38, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:54:55,807 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84199.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:55:12,694 INFO [train2.py:809] (3/4) Epoch 22, batch 550, loss[ctc_loss=0.06654, att_loss=0.2291, loss=0.1966, over 16284.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007036, over 43.00 utterances.], tot_loss[ctc_loss=0.07357, att_loss=0.2354, loss=0.203, over 3069302.80 frames. utt_duration=1245 frames, utt_pad_proportion=0.05248, over 9872.63 utterances.], batch size: 43, lr: 4.94e-03, grad_scale: 8.0 2023-03-08 23:55:55,773 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84236.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:56:04,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.221e+02 1.935e+02 2.218e+02 2.633e+02 6.667e+02, threshold=4.436e+02, percent-clipped=1.0 2023-03-08 23:56:33,881 INFO [train2.py:809] (3/4) Epoch 22, batch 600, loss[ctc_loss=0.06457, att_loss=0.2344, loss=0.2004, over 16525.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006887, over 45.00 utterances.], tot_loss[ctc_loss=0.07385, att_loss=0.2362, loss=0.2037, over 3122093.55 frames. utt_duration=1221 frames, utt_pad_proportion=0.0574, over 10244.59 utterances.], batch size: 45, lr: 4.93e-03, grad_scale: 8.0 2023-03-08 23:56:35,663 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84260.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:56:44,736 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8168, 5.1664, 5.3451, 5.2612, 5.3573, 5.7629, 5.1745, 5.8610], device='cuda:3'), covar=tensor([0.0771, 0.0772, 0.0904, 0.1239, 0.1809, 0.1008, 0.0713, 0.0703], device='cuda:3'), in_proj_covar=tensor([0.0873, 0.0514, 0.0600, 0.0666, 0.0874, 0.0627, 0.0487, 0.0612], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:57:33,972 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-08 23:57:35,148 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84297.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:57:49,016 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84305.0, num_to_drop=1, layers_to_drop={2} 2023-03-08 23:57:55,134 INFO [train2.py:809] (3/4) Epoch 22, batch 650, loss[ctc_loss=0.06697, att_loss=0.2186, loss=0.1883, over 16171.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006857, over 41.00 utterances.], tot_loss[ctc_loss=0.07424, att_loss=0.2366, loss=0.2041, over 3162536.86 frames. utt_duration=1235 frames, utt_pad_proportion=0.05303, over 10258.60 utterances.], batch size: 41, lr: 4.93e-03, grad_scale: 8.0 2023-03-08 23:58:38,267 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84336.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:58:45,769 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84341.0, num_to_drop=0, layers_to_drop=set() 2023-03-08 23:58:47,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 2.036e+02 2.451e+02 3.016e+02 6.584e+02, threshold=4.901e+02, percent-clipped=3.0 2023-03-08 23:59:13,073 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4026, 3.0144, 3.7295, 2.8837, 3.5310, 4.4849, 4.3200, 3.2484], device='cuda:3'), covar=tensor([0.0373, 0.1550, 0.1121, 0.1461, 0.1051, 0.0973, 0.0612, 0.1221], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0243, 0.0281, 0.0219, 0.0264, 0.0367, 0.0261, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-08 23:59:16,006 INFO [train2.py:809] (3/4) Epoch 22, batch 700, loss[ctc_loss=0.0771, att_loss=0.2287, loss=0.1984, over 15989.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008333, over 40.00 utterances.], tot_loss[ctc_loss=0.07395, att_loss=0.2362, loss=0.2038, over 3185379.09 frames. utt_duration=1260 frames, utt_pad_proportion=0.0487, over 10126.11 utterances.], batch size: 40, lr: 4.93e-03, grad_scale: 8.0 2023-03-08 23:59:41,726 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8702, 3.6715, 3.6442, 3.0613, 3.6743, 3.7836, 3.6797, 2.6538], device='cuda:3'), covar=tensor([0.1094, 0.1288, 0.1599, 0.3993, 0.0869, 0.1807, 0.0720, 0.3876], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0190, 0.0202, 0.0257, 0.0159, 0.0262, 0.0182, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-08 23:59:51,977 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.89 vs. limit=5.0 2023-03-09 00:00:17,550 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84397.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:00:28,851 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5250, 2.8193, 5.1122, 4.1991, 3.1738, 4.3174, 4.9280, 4.7344], device='cuda:3'), covar=tensor([0.0345, 0.1502, 0.0244, 0.0876, 0.1605, 0.0272, 0.0161, 0.0269], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0244, 0.0191, 0.0315, 0.0267, 0.0218, 0.0178, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:00:37,596 INFO [train2.py:809] (3/4) Epoch 22, batch 750, loss[ctc_loss=0.07965, att_loss=0.2465, loss=0.2131, over 16602.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006619, over 47.00 utterances.], tot_loss[ctc_loss=0.07345, att_loss=0.236, loss=0.2035, over 3203424.27 frames. utt_duration=1267 frames, utt_pad_proportion=0.04914, over 10122.75 utterances.], batch size: 47, lr: 4.93e-03, grad_scale: 8.0 2023-03-09 00:00:48,987 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84416.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:00:50,423 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7472, 6.0079, 5.4676, 5.7526, 5.6850, 5.1831, 5.4196, 5.1796], device='cuda:3'), covar=tensor([0.1343, 0.0994, 0.0892, 0.0817, 0.0983, 0.1655, 0.2407, 0.2533], device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0616, 0.0462, 0.0460, 0.0430, 0.0467, 0.0612, 0.0529], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 00:01:29,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 1.936e+02 2.365e+02 2.996e+02 6.300e+02, threshold=4.730e+02, percent-clipped=3.0 2023-03-09 00:01:48,757 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84453.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:01:57,859 INFO [train2.py:809] (3/4) Epoch 22, batch 800, loss[ctc_loss=0.07711, att_loss=0.2196, loss=0.1911, over 15644.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.00883, over 37.00 utterances.], tot_loss[ctc_loss=0.07492, att_loss=0.2364, loss=0.2041, over 3216218.59 frames. utt_duration=1252 frames, utt_pad_proportion=0.05278, over 10290.34 utterances.], batch size: 37, lr: 4.93e-03, grad_scale: 8.0 2023-03-09 00:02:06,034 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84464.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:02:12,477 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4058, 4.7541, 4.7041, 4.8032, 4.8272, 4.5655, 3.1544, 4.6524], device='cuda:3'), covar=tensor([0.0167, 0.0194, 0.0209, 0.0153, 0.0155, 0.0163, 0.1048, 0.0408], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0086, 0.0109, 0.0068, 0.0075, 0.0084, 0.0102, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:03:17,419 INFO [train2.py:809] (3/4) Epoch 22, batch 850, loss[ctc_loss=0.06581, att_loss=0.2419, loss=0.2067, over 16474.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006805, over 46.00 utterances.], tot_loss[ctc_loss=0.07375, att_loss=0.2355, loss=0.2031, over 3231304.16 frames. utt_duration=1276 frames, utt_pad_proportion=0.04655, over 10144.73 utterances.], batch size: 46, lr: 4.93e-03, grad_scale: 8.0 2023-03-09 00:03:42,555 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84525.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:04:07,927 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 1.954e+02 2.376e+02 2.798e+02 7.444e+02, threshold=4.752e+02, percent-clipped=4.0 2023-03-09 00:04:30,054 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84555.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:04:36,029 INFO [train2.py:809] (3/4) Epoch 22, batch 900, loss[ctc_loss=0.05752, att_loss=0.2158, loss=0.1842, over 15360.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01188, over 35.00 utterances.], tot_loss[ctc_loss=0.07346, att_loss=0.2357, loss=0.2032, over 3246498.78 frames. utt_duration=1271 frames, utt_pad_proportion=0.04646, over 10232.08 utterances.], batch size: 35, lr: 4.93e-03, grad_scale: 8.0 2023-03-09 00:04:52,968 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9378, 5.2662, 4.8316, 5.3589, 4.6789, 5.0493, 5.3973, 5.1681], device='cuda:3'), covar=tensor([0.0631, 0.0295, 0.0766, 0.0274, 0.0440, 0.0220, 0.0227, 0.0177], device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0322, 0.0367, 0.0350, 0.0324, 0.0238, 0.0303, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 00:05:08,797 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4805, 2.9038, 3.6642, 2.8384, 3.4729, 4.6042, 4.4348, 3.1614], device='cuda:3'), covar=tensor([0.0373, 0.1877, 0.1240, 0.1459, 0.1178, 0.0900, 0.0577, 0.1410], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0244, 0.0279, 0.0219, 0.0264, 0.0366, 0.0258, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:05:18,295 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84586.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:05:22,785 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3690, 2.9172, 3.5491, 4.5205, 4.0100, 4.0051, 3.0015, 2.3050], device='cuda:3'), covar=tensor([0.0792, 0.2017, 0.0812, 0.0513, 0.0955, 0.0460, 0.1434, 0.2201], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0216, 0.0189, 0.0219, 0.0225, 0.0179, 0.0202, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:05:27,606 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84592.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:05:50,264 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84605.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:05:55,902 INFO [train2.py:809] (3/4) Epoch 22, batch 950, loss[ctc_loss=0.09117, att_loss=0.2605, loss=0.2266, over 17048.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009688, over 53.00 utterances.], tot_loss[ctc_loss=0.07368, att_loss=0.2357, loss=0.2033, over 3240989.19 frames. utt_duration=1248 frames, utt_pad_proportion=0.05583, over 10403.41 utterances.], batch size: 53, lr: 4.92e-03, grad_scale: 8.0 2023-03-09 00:06:33,745 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84633.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:06:45,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-03-09 00:06:46,135 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84641.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:06:47,276 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.899e+02 2.395e+02 3.054e+02 5.479e+02, threshold=4.790e+02, percent-clipped=1.0 2023-03-09 00:07:06,398 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84653.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:07:15,606 INFO [train2.py:809] (3/4) Epoch 22, batch 1000, loss[ctc_loss=0.1032, att_loss=0.2566, loss=0.2259, over 13397.00 frames. utt_duration=368.4 frames, utt_pad_proportion=0.3581, over 146.00 utterances.], tot_loss[ctc_loss=0.07354, att_loss=0.2356, loss=0.2032, over 3237129.15 frames. utt_duration=1232 frames, utt_pad_proportion=0.06288, over 10524.85 utterances.], batch size: 146, lr: 4.92e-03, grad_scale: 8.0 2023-03-09 00:08:02,348 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84689.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:08:07,643 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84692.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:08:10,949 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84694.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:08:34,844 INFO [train2.py:809] (3/4) Epoch 22, batch 1050, loss[ctc_loss=0.04869, att_loss=0.1988, loss=0.1688, over 15359.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.0115, over 35.00 utterances.], tot_loss[ctc_loss=0.07322, att_loss=0.2354, loss=0.203, over 3243651.22 frames. utt_duration=1234 frames, utt_pad_proportion=0.06251, over 10530.86 utterances.], batch size: 35, lr: 4.92e-03, grad_scale: 8.0 2023-03-09 00:09:25,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.985e+02 2.384e+02 2.768e+02 5.690e+02, threshold=4.768e+02, percent-clipped=2.0 2023-03-09 00:09:45,261 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:09:54,288 INFO [train2.py:809] (3/4) Epoch 22, batch 1100, loss[ctc_loss=0.09381, att_loss=0.248, loss=0.2172, over 16968.00 frames. utt_duration=687 frames, utt_pad_proportion=0.1326, over 99.00 utterances.], tot_loss[ctc_loss=0.07347, att_loss=0.2356, loss=0.2032, over 3246645.46 frames. utt_duration=1224 frames, utt_pad_proportion=0.06379, over 10623.85 utterances.], batch size: 99, lr: 4.92e-03, grad_scale: 8.0 2023-03-09 00:10:41,291 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1078, 5.3823, 5.3089, 5.2883, 5.3938, 5.3571, 5.0693, 4.8658], device='cuda:3'), covar=tensor([0.0884, 0.0472, 0.0297, 0.0533, 0.0271, 0.0283, 0.0372, 0.0326], device='cuda:3'), in_proj_covar=tensor([0.0518, 0.0365, 0.0347, 0.0358, 0.0420, 0.0429, 0.0359, 0.0395], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 00:10:59,988 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84800.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:11:01,300 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84801.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:11:13,085 INFO [train2.py:809] (3/4) Epoch 22, batch 1150, loss[ctc_loss=0.05522, att_loss=0.2176, loss=0.1851, over 12679.00 frames. utt_duration=1813 frames, utt_pad_proportion=0.05174, over 28.00 utterances.], tot_loss[ctc_loss=0.07415, att_loss=0.2363, loss=0.2039, over 3247971.85 frames. utt_duration=1231 frames, utt_pad_proportion=0.06126, over 10566.96 utterances.], batch size: 28, lr: 4.92e-03, grad_scale: 16.0 2023-03-09 00:11:20,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1918, 5.4589, 5.7577, 5.5638, 5.6617, 6.1490, 5.3992, 6.1916], device='cuda:3'), covar=tensor([0.0663, 0.0798, 0.0752, 0.1206, 0.1738, 0.0895, 0.0589, 0.0671], device='cuda:3'), in_proj_covar=tensor([0.0872, 0.0515, 0.0597, 0.0664, 0.0872, 0.0626, 0.0486, 0.0607], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:12:04,638 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 1.971e+02 2.350e+02 2.955e+02 6.874e+02, threshold=4.701e+02, percent-clipped=3.0 2023-03-09 00:12:25,857 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:12:31,662 INFO [train2.py:809] (3/4) Epoch 22, batch 1200, loss[ctc_loss=0.07019, att_loss=0.2403, loss=0.2063, over 16880.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.0067, over 49.00 utterances.], tot_loss[ctc_loss=0.07389, att_loss=0.2359, loss=0.2035, over 3255260.47 frames. utt_duration=1262 frames, utt_pad_proportion=0.05262, over 10332.79 utterances.], batch size: 49, lr: 4.92e-03, grad_scale: 16.0 2023-03-09 00:12:35,126 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84861.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:13:06,043 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84881.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:13:24,358 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84892.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:13:41,853 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84903.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:13:51,126 INFO [train2.py:809] (3/4) Epoch 22, batch 1250, loss[ctc_loss=0.07923, att_loss=0.2493, loss=0.2153, over 17094.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01596, over 56.00 utterances.], tot_loss[ctc_loss=0.07342, att_loss=0.2358, loss=0.2033, over 3260191.28 frames. utt_duration=1256 frames, utt_pad_proportion=0.05247, over 10395.19 utterances.], batch size: 56, lr: 4.92e-03, grad_scale: 16.0 2023-03-09 00:14:40,831 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84940.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:14:43,744 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.899e+02 2.318e+02 2.747e+02 6.253e+02, threshold=4.636e+02, percent-clipped=3.0 2023-03-09 00:14:58,344 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84951.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:15:00,484 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2023-03-09 00:15:09,887 INFO [train2.py:809] (3/4) Epoch 22, batch 1300, loss[ctc_loss=0.06134, att_loss=0.2306, loss=0.1967, over 16416.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006755, over 44.00 utterances.], tot_loss[ctc_loss=0.07339, att_loss=0.2363, loss=0.2037, over 3267180.21 frames. utt_duration=1255 frames, utt_pad_proportion=0.05098, over 10427.67 utterances.], batch size: 44, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:15:58,290 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84989.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:15:59,926 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9405, 3.8722, 3.2973, 3.3213, 4.1689, 3.7230, 2.7468, 4.3636], device='cuda:3'), covar=tensor([0.1176, 0.0447, 0.1077, 0.0754, 0.0595, 0.0645, 0.1041, 0.0499], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0217, 0.0227, 0.0200, 0.0278, 0.0241, 0.0202, 0.0287], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:16:02,926 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84992.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:16:29,442 INFO [train2.py:809] (3/4) Epoch 22, batch 1350, loss[ctc_loss=0.08266, att_loss=0.2442, loss=0.2119, over 16775.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005345, over 48.00 utterances.], tot_loss[ctc_loss=0.07254, att_loss=0.2356, loss=0.203, over 3265193.97 frames. utt_duration=1255 frames, utt_pad_proportion=0.05322, over 10421.82 utterances.], batch size: 48, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:16:34,366 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85012.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:17:18,550 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85040.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:17:21,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 1.871e+02 2.280e+02 2.864e+02 6.552e+02, threshold=4.561e+02, percent-clipped=2.0 2023-03-09 00:17:48,043 INFO [train2.py:809] (3/4) Epoch 22, batch 1400, loss[ctc_loss=0.1, att_loss=0.2536, loss=0.2229, over 17043.00 frames. utt_duration=864.5 frames, utt_pad_proportion=0.09287, over 79.00 utterances.], tot_loss[ctc_loss=0.07265, att_loss=0.2355, loss=0.203, over 3270709.85 frames. utt_duration=1249 frames, utt_pad_proportion=0.05246, over 10488.56 utterances.], batch size: 79, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:18:11,496 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2694, 2.8954, 3.6106, 2.8823, 3.4230, 4.3647, 4.2963, 3.1043], device='cuda:3'), covar=tensor([0.0407, 0.1599, 0.1145, 0.1337, 0.1095, 0.1012, 0.0643, 0.1283], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0240, 0.0275, 0.0215, 0.0262, 0.0362, 0.0255, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:18:19,230 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0569, 3.6541, 3.0726, 3.3848, 3.9981, 3.6013, 3.1631, 4.2153], device='cuda:3'), covar=tensor([0.1049, 0.0532, 0.1168, 0.0717, 0.0638, 0.0748, 0.0745, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0218, 0.0228, 0.0200, 0.0279, 0.0242, 0.0202, 0.0288], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:19:02,066 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1903, 5.2001, 4.8757, 2.4596, 2.0148, 3.0926, 2.6582, 3.9654], device='cuda:3'), covar=tensor([0.0674, 0.0307, 0.0315, 0.4716, 0.5545, 0.2240, 0.3296, 0.1620], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0275, 0.0266, 0.0244, 0.0339, 0.0330, 0.0253, 0.0360], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 00:19:03,506 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85107.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:19:06,356 INFO [train2.py:809] (3/4) Epoch 22, batch 1450, loss[ctc_loss=0.07362, att_loss=0.2433, loss=0.2093, over 17195.00 frames. utt_duration=871.9 frames, utt_pad_proportion=0.08415, over 79.00 utterances.], tot_loss[ctc_loss=0.07339, att_loss=0.2359, loss=0.2034, over 3272482.31 frames. utt_duration=1237 frames, utt_pad_proportion=0.05592, over 10594.12 utterances.], batch size: 79, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:19:16,279 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85115.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:19:59,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 1.922e+02 2.285e+02 2.782e+02 4.553e+02, threshold=4.570e+02, percent-clipped=0.0 2023-03-09 00:20:21,434 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85156.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:20:25,869 INFO [train2.py:809] (3/4) Epoch 22, batch 1500, loss[ctc_loss=0.07344, att_loss=0.242, loss=0.2083, over 17038.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007637, over 51.00 utterances.], tot_loss[ctc_loss=0.07284, att_loss=0.2357, loss=0.2031, over 3273945.61 frames. utt_duration=1248 frames, utt_pad_proportion=0.05436, over 10502.59 utterances.], batch size: 51, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:20:40,486 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85168.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 00:20:47,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-09 00:20:52,801 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85176.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:21:01,411 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85181.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:21:36,120 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1518, 4.1424, 4.2381, 4.1113, 4.6496, 4.2611, 4.1341, 2.3576], device='cuda:3'), covar=tensor([0.0315, 0.0442, 0.0389, 0.0354, 0.0816, 0.0254, 0.0429, 0.1953], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0192, 0.0189, 0.0206, 0.0366, 0.0159, 0.0180, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:21:46,575 INFO [train2.py:809] (3/4) Epoch 22, batch 1550, loss[ctc_loss=0.05685, att_loss=0.2219, loss=0.1889, over 16546.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006047, over 45.00 utterances.], tot_loss[ctc_loss=0.07254, att_loss=0.2359, loss=0.2032, over 3275620.07 frames. utt_duration=1234 frames, utt_pad_proportion=0.05808, over 10626.49 utterances.], batch size: 45, lr: 4.91e-03, grad_scale: 16.0 2023-03-09 00:22:19,181 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85229.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:22:35,019 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 2023-03-09 00:22:42,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 1.952e+02 2.297e+02 2.764e+02 6.264e+02, threshold=4.595e+02, percent-clipped=3.0 2023-03-09 00:23:06,928 INFO [train2.py:809] (3/4) Epoch 22, batch 1600, loss[ctc_loss=0.08565, att_loss=0.2512, loss=0.2181, over 16467.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.007181, over 46.00 utterances.], tot_loss[ctc_loss=0.07273, att_loss=0.2366, loss=0.2038, over 3279433.08 frames. utt_duration=1234 frames, utt_pad_proportion=0.05792, over 10645.29 utterances.], batch size: 46, lr: 4.91e-03, grad_scale: 8.0 2023-03-09 00:23:56,476 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85289.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:24:11,873 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3420, 3.9320, 3.4850, 3.7269, 4.2141, 3.8542, 3.2202, 4.4941], device='cuda:3'), covar=tensor([0.0864, 0.0467, 0.0897, 0.0547, 0.0583, 0.0643, 0.0776, 0.0466], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0218, 0.0227, 0.0200, 0.0280, 0.0242, 0.0203, 0.0289], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:24:23,646 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85307.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:24:27,043 INFO [train2.py:809] (3/4) Epoch 22, batch 1650, loss[ctc_loss=0.06519, att_loss=0.2358, loss=0.2017, over 16626.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005528, over 47.00 utterances.], tot_loss[ctc_loss=0.07253, att_loss=0.2363, loss=0.2035, over 3281974.79 frames. utt_duration=1224 frames, utt_pad_proportion=0.05871, over 10738.19 utterances.], batch size: 47, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:24:53,305 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-09 00:25:11,310 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85337.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:25:20,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 1.903e+02 2.201e+02 2.688e+02 4.353e+02, threshold=4.403e+02, percent-clipped=0.0 2023-03-09 00:25:44,528 INFO [train2.py:809] (3/4) Epoch 22, batch 1700, loss[ctc_loss=0.06758, att_loss=0.2343, loss=0.2009, over 16613.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006066, over 47.00 utterances.], tot_loss[ctc_loss=0.0732, att_loss=0.2362, loss=0.2036, over 3276959.07 frames. utt_duration=1233 frames, utt_pad_proportion=0.05901, over 10645.42 utterances.], batch size: 47, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:27:03,855 INFO [train2.py:809] (3/4) Epoch 22, batch 1750, loss[ctc_loss=0.06502, att_loss=0.228, loss=0.1954, over 16121.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006772, over 42.00 utterances.], tot_loss[ctc_loss=0.07443, att_loss=0.2368, loss=0.2043, over 3272657.42 frames. utt_duration=1216 frames, utt_pad_proportion=0.06338, over 10779.06 utterances.], batch size: 42, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:27:36,198 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:27:58,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.824e+02 2.227e+02 2.751e+02 5.313e+02, threshold=4.454e+02, percent-clipped=1.0 2023-03-09 00:28:04,291 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-03-09 00:28:19,235 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85456.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:28:24,088 INFO [train2.py:809] (3/4) Epoch 22, batch 1800, loss[ctc_loss=0.0747, att_loss=0.2402, loss=0.2071, over 16456.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007828, over 46.00 utterances.], tot_loss[ctc_loss=0.0745, att_loss=0.2367, loss=0.2043, over 3269597.38 frames. utt_duration=1199 frames, utt_pad_proportion=0.06993, over 10923.54 utterances.], batch size: 46, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:28:30,531 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85463.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 00:28:44,306 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85471.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:29:04,487 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85483.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:29:13,770 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2601, 2.4578, 3.0797, 2.5135, 3.0313, 3.4636, 3.3320, 2.6306], device='cuda:3'), covar=tensor([0.0553, 0.1569, 0.1065, 0.1292, 0.0901, 0.1195, 0.0800, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0243, 0.0280, 0.0219, 0.0265, 0.0367, 0.0260, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:29:13,841 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85489.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:29:37,447 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85504.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:29:45,468 INFO [train2.py:809] (3/4) Epoch 22, batch 1850, loss[ctc_loss=0.07307, att_loss=0.2438, loss=0.2097, over 16685.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.005197, over 46.00 utterances.], tot_loss[ctc_loss=0.0741, att_loss=0.2363, loss=0.2039, over 3271394.82 frames. utt_duration=1200 frames, utt_pad_proportion=0.0683, over 10922.68 utterances.], batch size: 46, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:30:19,829 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85530.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:30:39,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.961e+02 2.288e+02 2.776e+02 5.536e+02, threshold=4.576e+02, percent-clipped=1.0 2023-03-09 00:30:41,537 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:31:04,499 INFO [train2.py:809] (3/4) Epoch 22, batch 1900, loss[ctc_loss=0.0779, att_loss=0.2195, loss=0.1912, over 15996.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.007838, over 40.00 utterances.], tot_loss[ctc_loss=0.07479, att_loss=0.2365, loss=0.2042, over 3274332.87 frames. utt_duration=1191 frames, utt_pad_proportion=0.06785, over 11007.27 utterances.], batch size: 40, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:31:25,993 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5926, 3.7202, 3.8072, 2.3236, 2.4461, 2.8512, 2.4712, 3.5092], device='cuda:3'), covar=tensor([0.0774, 0.0514, 0.0453, 0.4012, 0.3747, 0.2105, 0.2695, 0.1297], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0278, 0.0267, 0.0246, 0.0340, 0.0332, 0.0255, 0.0364], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 00:31:56,023 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85591.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:32:21,728 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85607.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:32:24,418 INFO [train2.py:809] (3/4) Epoch 22, batch 1950, loss[ctc_loss=0.07699, att_loss=0.2445, loss=0.211, over 16532.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006587, over 45.00 utterances.], tot_loss[ctc_loss=0.07483, att_loss=0.2363, loss=0.204, over 3255983.15 frames. utt_duration=1175 frames, utt_pad_proportion=0.07638, over 11101.37 utterances.], batch size: 45, lr: 4.90e-03, grad_scale: 8.0 2023-03-09 00:32:51,856 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85625.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:33:19,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.791e+02 2.130e+02 2.666e+02 5.391e+02, threshold=4.261e+02, percent-clipped=1.0 2023-03-09 00:33:39,316 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85655.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:33:45,244 INFO [train2.py:809] (3/4) Epoch 22, batch 2000, loss[ctc_loss=0.05686, att_loss=0.2083, loss=0.178, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008766, over 37.00 utterances.], tot_loss[ctc_loss=0.07381, att_loss=0.2359, loss=0.2035, over 3253212.05 frames. utt_duration=1180 frames, utt_pad_proportion=0.07717, over 11044.68 utterances.], batch size: 37, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:34:29,135 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85686.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 00:35:05,241 INFO [train2.py:809] (3/4) Epoch 22, batch 2050, loss[ctc_loss=0.06664, att_loss=0.2423, loss=0.2071, over 16872.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00568, over 49.00 utterances.], tot_loss[ctc_loss=0.07383, att_loss=0.2363, loss=0.2038, over 3261666.17 frames. utt_duration=1199 frames, utt_pad_proportion=0.06862, over 10892.65 utterances.], batch size: 49, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:35:10,132 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4263, 2.6470, 4.8674, 3.7587, 2.8644, 4.2319, 4.6370, 4.5268], device='cuda:3'), covar=tensor([0.0237, 0.1581, 0.0179, 0.0999, 0.1849, 0.0257, 0.0164, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0242, 0.0193, 0.0315, 0.0266, 0.0219, 0.0180, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:36:01,005 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 1.910e+02 2.299e+02 2.633e+02 5.710e+02, threshold=4.599e+02, percent-clipped=8.0 2023-03-09 00:36:27,095 INFO [train2.py:809] (3/4) Epoch 22, batch 2100, loss[ctc_loss=0.05832, att_loss=0.2225, loss=0.1897, over 16135.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005344, over 42.00 utterances.], tot_loss[ctc_loss=0.07395, att_loss=0.2357, loss=0.2034, over 3257315.32 frames. utt_duration=1229 frames, utt_pad_proportion=0.06237, over 10615.41 utterances.], batch size: 42, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:36:33,865 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85763.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:36:40,161 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7704, 5.9805, 5.4389, 5.7472, 5.6695, 5.2013, 5.4135, 5.2193], device='cuda:3'), covar=tensor([0.1233, 0.0865, 0.0924, 0.0778, 0.0889, 0.1381, 0.2290, 0.2253], device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0613, 0.0456, 0.0456, 0.0429, 0.0462, 0.0611, 0.0526], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 00:36:48,135 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85771.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:36:55,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1783, 3.8316, 3.3207, 3.6140, 4.0886, 3.8545, 3.1740, 4.4604], device='cuda:3'), covar=tensor([0.0958, 0.0510, 0.1060, 0.0611, 0.0727, 0.0628, 0.0841, 0.0407], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0214, 0.0223, 0.0197, 0.0277, 0.0238, 0.0198, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:37:08,410 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85784.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:37:08,437 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9174, 5.2185, 5.1716, 5.0984, 5.1972, 5.2141, 4.8174, 4.6774], device='cuda:3'), covar=tensor([0.1092, 0.0554, 0.0301, 0.0522, 0.0321, 0.0337, 0.0488, 0.0359], device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0365, 0.0345, 0.0359, 0.0421, 0.0428, 0.0358, 0.0394], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:37:47,844 INFO [train2.py:809] (3/4) Epoch 22, batch 2150, loss[ctc_loss=0.08203, att_loss=0.2426, loss=0.2105, over 17038.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.009652, over 53.00 utterances.], tot_loss[ctc_loss=0.07395, att_loss=0.2353, loss=0.203, over 3249671.74 frames. utt_duration=1251 frames, utt_pad_proportion=0.05996, over 10405.17 utterances.], batch size: 53, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:37:51,097 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85811.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:38:04,888 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85819.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:38:29,199 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6621, 2.1978, 2.5002, 2.7889, 3.0061, 2.7784, 2.3720, 3.0476], device='cuda:3'), covar=tensor([0.1927, 0.2746, 0.1985, 0.1536, 0.1835, 0.1185, 0.2631, 0.1087], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0124, 0.0122, 0.0110, 0.0124, 0.0106, 0.0131, 0.0101], device='cuda:3'), out_proj_covar=tensor([8.9815e-05, 9.6797e-05, 9.6961e-05, 8.6132e-05, 9.2639e-05, 8.5485e-05, 9.8668e-05, 8.0916e-05], device='cuda:3') 2023-03-09 00:38:36,704 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85839.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:38:42,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 1.885e+02 2.205e+02 3.046e+02 7.448e+02, threshold=4.410e+02, percent-clipped=6.0 2023-03-09 00:38:43,170 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3683, 4.5960, 4.7593, 4.5400, 5.1592, 4.6347, 4.6235, 2.5290], device='cuda:3'), covar=tensor([0.0295, 0.0283, 0.0258, 0.0311, 0.0682, 0.0229, 0.0291, 0.1764], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0190, 0.0188, 0.0206, 0.0365, 0.0159, 0.0179, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:39:07,838 INFO [train2.py:809] (3/4) Epoch 22, batch 2200, loss[ctc_loss=0.0566, att_loss=0.2138, loss=0.1823, over 15648.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.00798, over 37.00 utterances.], tot_loss[ctc_loss=0.07357, att_loss=0.235, loss=0.2027, over 3251503.83 frames. utt_duration=1261 frames, utt_pad_proportion=0.05829, over 10325.96 utterances.], batch size: 37, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:39:28,053 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3696, 5.1933, 5.0771, 3.1533, 5.0591, 4.9847, 4.7356, 3.2085], device='cuda:3'), covar=tensor([0.0086, 0.0076, 0.0245, 0.0839, 0.0080, 0.0147, 0.0220, 0.1012], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0103, 0.0106, 0.0112, 0.0086, 0.0116, 0.0100, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 00:39:35,153 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85875.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:39:52,050 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85886.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:40:28,363 INFO [train2.py:809] (3/4) Epoch 22, batch 2250, loss[ctc_loss=0.08207, att_loss=0.2454, loss=0.2128, over 17041.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.0103, over 53.00 utterances.], tot_loss[ctc_loss=0.07362, att_loss=0.2355, loss=0.2031, over 3261446.80 frames. utt_duration=1287 frames, utt_pad_proportion=0.04949, over 10147.27 utterances.], batch size: 53, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:41:02,701 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85930.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 00:41:11,760 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85936.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:41:19,402 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5038, 3.1547, 3.5768, 3.0714, 3.4678, 4.5523, 4.4111, 3.2152], device='cuda:3'), covar=tensor([0.0356, 0.1517, 0.1245, 0.1218, 0.1097, 0.0848, 0.0558, 0.1216], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0243, 0.0283, 0.0219, 0.0266, 0.0370, 0.0260, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:41:22,180 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 1.895e+02 2.304e+02 2.747e+02 7.782e+02, threshold=4.607e+02, percent-clipped=2.0 2023-03-09 00:41:44,932 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85957.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:41:47,673 INFO [train2.py:809] (3/4) Epoch 22, batch 2300, loss[ctc_loss=0.05711, att_loss=0.2228, loss=0.1896, over 15955.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.0063, over 41.00 utterances.], tot_loss[ctc_loss=0.07355, att_loss=0.2357, loss=0.2033, over 3265727.20 frames. utt_duration=1286 frames, utt_pad_proportion=0.04776, over 10171.12 utterances.], batch size: 41, lr: 4.89e-03, grad_scale: 8.0 2023-03-09 00:42:00,187 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1611, 3.8109, 3.2479, 3.5514, 4.1169, 3.8323, 3.1000, 4.4567], device='cuda:3'), covar=tensor([0.1010, 0.0546, 0.1200, 0.0644, 0.0698, 0.0640, 0.0838, 0.0396], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0213, 0.0223, 0.0197, 0.0277, 0.0237, 0.0197, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:42:24,394 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85981.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 00:42:40,457 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85991.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 00:43:02,471 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86002.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:43:13,255 INFO [train2.py:809] (3/4) Epoch 22, batch 2350, loss[ctc_loss=0.08257, att_loss=0.2455, loss=0.2129, over 16474.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006147, over 46.00 utterances.], tot_loss[ctc_loss=0.0729, att_loss=0.2358, loss=0.2032, over 3277224.04 frames. utt_duration=1275 frames, utt_pad_proportion=0.04525, over 10294.44 utterances.], batch size: 46, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:43:28,539 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:43:46,305 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7853, 3.6331, 3.6465, 3.1314, 3.6240, 3.7498, 3.6929, 2.5606], device='cuda:3'), covar=tensor([0.1297, 0.1234, 0.1505, 0.3680, 0.1110, 0.2034, 0.0954, 0.3802], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0191, 0.0202, 0.0255, 0.0160, 0.0262, 0.0185, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:43:59,591 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8938, 5.2005, 4.7911, 5.2777, 4.6175, 4.9485, 5.3717, 5.1064], device='cuda:3'), covar=tensor([0.0637, 0.0344, 0.0838, 0.0360, 0.0492, 0.0283, 0.0239, 0.0224], device='cuda:3'), in_proj_covar=tensor([0.0388, 0.0321, 0.0367, 0.0351, 0.0323, 0.0237, 0.0303, 0.0286], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 00:44:07,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 1.776e+02 2.270e+02 2.670e+02 5.895e+02, threshold=4.541e+02, percent-clipped=3.0 2023-03-09 00:44:21,752 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86052.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:44:32,415 INFO [train2.py:809] (3/4) Epoch 22, batch 2400, loss[ctc_loss=0.05786, att_loss=0.2188, loss=0.1866, over 16193.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006132, over 41.00 utterances.], tot_loss[ctc_loss=0.07274, att_loss=0.2361, loss=0.2035, over 3281771.32 frames. utt_duration=1261 frames, utt_pad_proportion=0.04778, over 10423.39 utterances.], batch size: 41, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:44:39,860 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86063.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:44:49,075 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86069.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:45:12,658 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86084.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:45:50,867 INFO [train2.py:809] (3/4) Epoch 22, batch 2450, loss[ctc_loss=0.08236, att_loss=0.2408, loss=0.2091, over 16256.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.008123, over 43.00 utterances.], tot_loss[ctc_loss=0.07318, att_loss=0.2366, loss=0.2039, over 3290038.25 frames. utt_duration=1274 frames, utt_pad_proportion=0.04246, over 10345.41 utterances.], batch size: 43, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:45:57,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9811, 3.7376, 3.7061, 3.2077, 3.7571, 3.8763, 3.7640, 2.7732], device='cuda:3'), covar=tensor([0.0929, 0.1224, 0.1865, 0.2988, 0.1065, 0.1500, 0.0711, 0.3632], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0192, 0.0203, 0.0256, 0.0161, 0.0263, 0.0186, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:45:58,816 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86113.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:46:26,059 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86130.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:46:27,551 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8064, 5.0715, 4.6329, 5.2230, 4.6016, 4.8727, 5.2722, 5.0296], device='cuda:3'), covar=tensor([0.0687, 0.0373, 0.0877, 0.0354, 0.0429, 0.0298, 0.0260, 0.0222], device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0321, 0.0369, 0.0353, 0.0324, 0.0237, 0.0305, 0.0287], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 00:46:28,916 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86132.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:46:39,900 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86139.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:46:45,845 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 2.126e+02 2.601e+02 3.056e+02 8.215e+02, threshold=5.203e+02, percent-clipped=8.0 2023-03-09 00:46:58,617 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9679, 5.2775, 5.5253, 5.3227, 5.4732, 5.9091, 5.1703, 6.0248], device='cuda:3'), covar=tensor([0.0731, 0.0791, 0.0896, 0.1465, 0.1854, 0.0974, 0.0749, 0.0708], device='cuda:3'), in_proj_covar=tensor([0.0870, 0.0513, 0.0595, 0.0662, 0.0864, 0.0630, 0.0485, 0.0607], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:47:11,853 INFO [train2.py:809] (3/4) Epoch 22, batch 2500, loss[ctc_loss=0.07685, att_loss=0.2339, loss=0.2025, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007965, over 43.00 utterances.], tot_loss[ctc_loss=0.07336, att_loss=0.2363, loss=0.2038, over 3278954.40 frames. utt_duration=1283 frames, utt_pad_proportion=0.04241, over 10235.23 utterances.], batch size: 43, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:47:12,334 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5377, 2.5903, 4.9921, 3.8827, 3.0193, 4.2464, 4.7221, 4.6364], device='cuda:3'), covar=tensor([0.0234, 0.1545, 0.0178, 0.0869, 0.1717, 0.0238, 0.0134, 0.0250], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0240, 0.0192, 0.0314, 0.0264, 0.0216, 0.0179, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:47:35,899 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-09 00:47:41,702 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0453, 5.2871, 5.2447, 5.2192, 5.3094, 5.2918, 4.9775, 4.7573], device='cuda:3'), covar=tensor([0.0995, 0.0561, 0.0300, 0.0489, 0.0304, 0.0311, 0.0395, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0365, 0.0344, 0.0360, 0.0425, 0.0428, 0.0358, 0.0395], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:47:57,562 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:47:58,912 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86187.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:48:19,487 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8000, 3.6254, 3.5714, 3.0904, 3.5747, 3.7610, 3.6744, 2.6636], device='cuda:3'), covar=tensor([0.1217, 0.1366, 0.2594, 0.3798, 0.1693, 0.1664, 0.0946, 0.3734], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0193, 0.0205, 0.0259, 0.0163, 0.0266, 0.0188, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:48:33,443 INFO [train2.py:809] (3/4) Epoch 22, batch 2550, loss[ctc_loss=0.06441, att_loss=0.2131, loss=0.1833, over 15363.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01183, over 35.00 utterances.], tot_loss[ctc_loss=0.07376, att_loss=0.2366, loss=0.204, over 3274403.91 frames. utt_duration=1272 frames, utt_pad_proportion=0.04553, over 10305.90 utterances.], batch size: 35, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:49:10,448 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86231.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:49:15,106 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86234.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:49:29,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 1.974e+02 2.356e+02 2.687e+02 6.031e+02, threshold=4.712e+02, percent-clipped=2.0 2023-03-09 00:49:34,339 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86246.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:49:54,975 INFO [train2.py:809] (3/4) Epoch 22, batch 2600, loss[ctc_loss=0.06787, att_loss=0.2071, loss=0.1793, over 15531.00 frames. utt_duration=1727 frames, utt_pad_proportion=0.006741, over 36.00 utterances.], tot_loss[ctc_loss=0.07271, att_loss=0.2353, loss=0.2028, over 3256548.06 frames. utt_duration=1251 frames, utt_pad_proportion=0.05589, over 10427.56 utterances.], batch size: 36, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:49:56,699 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0996, 5.3550, 5.6695, 5.4721, 5.5832, 6.0455, 5.3116, 6.1282], device='cuda:3'), covar=tensor([0.0711, 0.0747, 0.0707, 0.1309, 0.1786, 0.0878, 0.0596, 0.0620], device='cuda:3'), in_proj_covar=tensor([0.0867, 0.0512, 0.0594, 0.0661, 0.0865, 0.0628, 0.0484, 0.0606], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 00:50:31,773 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86281.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:50:39,213 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86286.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 00:51:13,094 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86307.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:51:15,795 INFO [train2.py:809] (3/4) Epoch 22, batch 2650, loss[ctc_loss=0.04508, att_loss=0.2147, loss=0.1808, over 16283.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006975, over 43.00 utterances.], tot_loss[ctc_loss=0.07191, att_loss=0.2354, loss=0.2027, over 3265186.56 frames. utt_duration=1247 frames, utt_pad_proportion=0.0555, over 10490.32 utterances.], batch size: 43, lr: 4.88e-03, grad_scale: 8.0 2023-03-09 00:51:16,068 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1284, 5.3874, 5.3624, 5.3090, 5.4492, 5.4139, 5.0903, 4.8815], device='cuda:3'), covar=tensor([0.0971, 0.0594, 0.0325, 0.0526, 0.0290, 0.0328, 0.0423, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0525, 0.0368, 0.0346, 0.0364, 0.0428, 0.0433, 0.0361, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 00:51:23,788 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86313.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:51:39,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-09 00:51:49,591 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:51:52,995 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9836, 4.9918, 5.0793, 2.1503, 1.9823, 2.6459, 2.3969, 3.6671], device='cuda:3'), covar=tensor([0.1007, 0.0411, 0.0242, 0.5182, 0.7001, 0.3553, 0.4058, 0.2079], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0278, 0.0268, 0.0246, 0.0342, 0.0335, 0.0256, 0.0369], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 00:52:11,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.827e+02 2.241e+02 2.634e+02 5.171e+02, threshold=4.483e+02, percent-clipped=1.0 2023-03-09 00:52:37,280 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:52:38,739 INFO [train2.py:809] (3/4) Epoch 22, batch 2700, loss[ctc_loss=0.0907, att_loss=0.2458, loss=0.2147, over 17127.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01464, over 56.00 utterances.], tot_loss[ctc_loss=0.07222, att_loss=0.2353, loss=0.2027, over 3262313.57 frames. utt_duration=1224 frames, utt_pad_proportion=0.06233, over 10677.02 utterances.], batch size: 56, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:52:54,788 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3655, 4.4544, 4.6718, 4.5212, 5.2286, 4.5673, 4.4975, 2.6498], device='cuda:3'), covar=tensor([0.0329, 0.0378, 0.0372, 0.0394, 0.0753, 0.0242, 0.0372, 0.1856], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0191, 0.0190, 0.0207, 0.0366, 0.0161, 0.0181, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:53:00,954 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8516, 3.5070, 2.9281, 3.2661, 3.6863, 3.4584, 2.9711, 3.8135], device='cuda:3'), covar=tensor([0.1007, 0.0522, 0.1128, 0.0682, 0.0724, 0.0715, 0.0804, 0.0513], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0219, 0.0229, 0.0202, 0.0282, 0.0244, 0.0202, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 00:53:58,806 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86408.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:54:00,134 INFO [train2.py:809] (3/4) Epoch 22, batch 2750, loss[ctc_loss=0.07732, att_loss=0.2399, loss=0.2074, over 16845.00 frames. utt_duration=682.2 frames, utt_pad_proportion=0.1451, over 99.00 utterances.], tot_loss[ctc_loss=0.07348, att_loss=0.2365, loss=0.2039, over 3267509.40 frames. utt_duration=1193 frames, utt_pad_proportion=0.06669, over 10972.49 utterances.], batch size: 99, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:54:25,800 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86425.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:54:43,896 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-09 00:54:49,495 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86440.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:54:53,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 1.998e+02 2.389e+02 2.932e+02 8.437e+02, threshold=4.778e+02, percent-clipped=2.0 2023-03-09 00:54:57,247 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:55:20,793 INFO [train2.py:809] (3/4) Epoch 22, batch 2800, loss[ctc_loss=0.07842, att_loss=0.2263, loss=0.1967, over 15764.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.009091, over 38.00 utterances.], tot_loss[ctc_loss=0.07253, att_loss=0.2352, loss=0.2026, over 3263423.05 frames. utt_duration=1214 frames, utt_pad_proportion=0.06396, over 10764.18 utterances.], batch size: 38, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:55:57,941 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0828, 5.0192, 4.8658, 2.3706, 1.9687, 2.8864, 2.3557, 3.8346], device='cuda:3'), covar=tensor([0.0737, 0.0273, 0.0225, 0.4687, 0.5730, 0.2685, 0.3780, 0.1760], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0277, 0.0267, 0.0244, 0.0340, 0.0333, 0.0255, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 00:56:28,131 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86501.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:56:36,558 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86506.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 00:56:40,834 INFO [train2.py:809] (3/4) Epoch 22, batch 2850, loss[ctc_loss=0.05849, att_loss=0.2171, loss=0.1854, over 16015.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006862, over 40.00 utterances.], tot_loss[ctc_loss=0.07221, att_loss=0.235, loss=0.2025, over 3268238.86 frames. utt_duration=1244 frames, utt_pad_proportion=0.05566, over 10522.57 utterances.], batch size: 40, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:57:16,136 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86531.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:57:34,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.710e+01 1.849e+02 2.202e+02 2.707e+02 3.690e+02, threshold=4.403e+02, percent-clipped=0.0 2023-03-09 00:58:01,275 INFO [train2.py:809] (3/4) Epoch 22, batch 2900, loss[ctc_loss=0.04677, att_loss=0.2075, loss=0.1753, over 15775.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008897, over 38.00 utterances.], tot_loss[ctc_loss=0.07123, att_loss=0.2343, loss=0.2017, over 3275557.46 frames. utt_duration=1263 frames, utt_pad_proportion=0.04925, over 10388.15 utterances.], batch size: 38, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:58:33,088 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86579.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:58:34,902 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86580.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:58:44,203 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86586.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 00:59:09,710 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86602.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 00:59:20,649 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7914, 2.4500, 5.1182, 4.0946, 3.0640, 4.3687, 4.9783, 4.8889], device='cuda:3'), covar=tensor([0.0175, 0.1554, 0.0178, 0.0825, 0.1643, 0.0191, 0.0100, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0243, 0.0194, 0.0318, 0.0266, 0.0219, 0.0182, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 00:59:21,691 INFO [train2.py:809] (3/4) Epoch 22, batch 2950, loss[ctc_loss=0.06074, att_loss=0.2307, loss=0.1967, over 16165.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007772, over 41.00 utterances.], tot_loss[ctc_loss=0.07101, att_loss=0.2341, loss=0.2015, over 3276754.39 frames. utt_duration=1283 frames, utt_pad_proportion=0.04553, over 10231.26 utterances.], batch size: 41, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 00:59:28,873 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86613.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:00:01,861 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86634.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 01:00:12,896 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86641.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:00:15,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 1.951e+02 2.348e+02 3.080e+02 7.280e+02, threshold=4.697e+02, percent-clipped=5.0 2023-03-09 01:00:41,195 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86658.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:00:42,432 INFO [train2.py:809] (3/4) Epoch 22, batch 3000, loss[ctc_loss=0.1046, att_loss=0.261, loss=0.2297, over 14411.00 frames. utt_duration=396.4 frames, utt_pad_proportion=0.3093, over 146.00 utterances.], tot_loss[ctc_loss=0.07076, att_loss=0.2338, loss=0.2012, over 3277212.17 frames. utt_duration=1299 frames, utt_pad_proportion=0.04282, over 10106.39 utterances.], batch size: 146, lr: 4.87e-03, grad_scale: 8.0 2023-03-09 01:00:42,432 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 01:00:57,103 INFO [train2.py:843] (3/4) Epoch 22, validation: ctc_loss=0.03993, att_loss=0.2347, loss=0.1957, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 01:00:57,105 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 01:01:01,040 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86661.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:01:02,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-09 01:02:11,053 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86706.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:02:15,076 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86708.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:02:16,479 INFO [train2.py:809] (3/4) Epoch 22, batch 3050, loss[ctc_loss=0.06381, att_loss=0.2233, loss=0.1914, over 16536.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006392, over 45.00 utterances.], tot_loss[ctc_loss=0.07126, att_loss=0.2343, loss=0.2017, over 3275630.20 frames. utt_duration=1273 frames, utt_pad_proportion=0.04644, over 10305.28 utterances.], batch size: 45, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:02:38,849 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4059, 2.8265, 3.3413, 4.3980, 3.8517, 3.7038, 2.8465, 2.2771], device='cuda:3'), covar=tensor([0.0695, 0.1957, 0.0837, 0.0598, 0.0919, 0.0590, 0.1549, 0.2119], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0212, 0.0188, 0.0218, 0.0224, 0.0181, 0.0199, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:02:43,497 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:03:11,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 1.829e+02 2.203e+02 2.781e+02 6.356e+02, threshold=4.407e+02, percent-clipped=2.0 2023-03-09 01:03:31,281 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86756.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:03:36,294 INFO [train2.py:809] (3/4) Epoch 22, batch 3100, loss[ctc_loss=0.05676, att_loss=0.2111, loss=0.1802, over 15495.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008631, over 36.00 utterances.], tot_loss[ctc_loss=0.07237, att_loss=0.2353, loss=0.2027, over 3273664.53 frames. utt_duration=1243 frames, utt_pad_proportion=0.05556, over 10551.31 utterances.], batch size: 36, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:03:59,962 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86773.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:04:35,420 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86796.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:04:43,664 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86801.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 01:04:57,208 INFO [train2.py:809] (3/4) Epoch 22, batch 3150, loss[ctc_loss=0.1089, att_loss=0.2558, loss=0.2264, over 17300.00 frames. utt_duration=1005 frames, utt_pad_proportion=0.04963, over 69.00 utterances.], tot_loss[ctc_loss=0.07227, att_loss=0.2349, loss=0.2024, over 3272274.76 frames. utt_duration=1251 frames, utt_pad_proportion=0.05376, over 10477.53 utterances.], batch size: 69, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:05:06,684 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86814.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:05:08,131 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9273, 5.2300, 4.8119, 5.3244, 4.7154, 4.9381, 5.3580, 5.1311], device='cuda:3'), covar=tensor([0.0569, 0.0318, 0.0805, 0.0340, 0.0429, 0.0294, 0.0234, 0.0199], device='cuda:3'), in_proj_covar=tensor([0.0384, 0.0319, 0.0364, 0.0348, 0.0320, 0.0236, 0.0300, 0.0284], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 01:05:12,873 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1035, 5.3754, 5.3422, 5.3243, 5.4089, 5.3529, 5.0905, 4.8735], device='cuda:3'), covar=tensor([0.0973, 0.0473, 0.0271, 0.0412, 0.0274, 0.0314, 0.0341, 0.0318], device='cuda:3'), in_proj_covar=tensor([0.0525, 0.0365, 0.0348, 0.0360, 0.0427, 0.0433, 0.0359, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 01:05:17,741 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5787, 2.6486, 5.0538, 4.0210, 3.0109, 4.3432, 4.9022, 4.7556], device='cuda:3'), covar=tensor([0.0278, 0.1515, 0.0262, 0.0844, 0.1694, 0.0242, 0.0166, 0.0258], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0241, 0.0193, 0.0315, 0.0263, 0.0217, 0.0181, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:05:31,710 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7206, 3.9597, 3.8864, 3.9254, 3.9980, 3.8378, 3.1245, 3.9272], device='cuda:3'), covar=tensor([0.0155, 0.0137, 0.0158, 0.0110, 0.0110, 0.0133, 0.0624, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0090, 0.0114, 0.0071, 0.0077, 0.0088, 0.0106, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:05:51,841 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 1.971e+02 2.330e+02 2.834e+02 6.258e+02, threshold=4.661e+02, percent-clipped=4.0 2023-03-09 01:06:18,418 INFO [train2.py:809] (3/4) Epoch 22, batch 3200, loss[ctc_loss=0.06678, att_loss=0.2202, loss=0.1895, over 16008.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007265, over 40.00 utterances.], tot_loss[ctc_loss=0.07209, att_loss=0.2352, loss=0.2026, over 3271843.90 frames. utt_duration=1228 frames, utt_pad_proportion=0.05965, over 10673.58 utterances.], batch size: 40, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:06:45,810 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86875.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:06:50,302 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86878.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:07:08,735 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1679, 5.4872, 5.7306, 5.5119, 5.6782, 6.0997, 5.3744, 6.2008], device='cuda:3'), covar=tensor([0.0706, 0.0649, 0.0760, 0.1224, 0.1828, 0.0972, 0.0585, 0.0637], device='cuda:3'), in_proj_covar=tensor([0.0875, 0.0516, 0.0605, 0.0665, 0.0876, 0.0638, 0.0492, 0.0614], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:07:27,704 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86902.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:07:39,433 INFO [train2.py:809] (3/4) Epoch 22, batch 3250, loss[ctc_loss=0.07068, att_loss=0.2438, loss=0.2092, over 17044.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009419, over 53.00 utterances.], tot_loss[ctc_loss=0.0718, att_loss=0.2344, loss=0.2019, over 3266620.76 frames. utt_duration=1221 frames, utt_pad_proportion=0.06295, over 10712.21 utterances.], batch size: 53, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:08:22,562 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86936.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:08:27,454 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86939.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:08:33,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.904e+02 2.296e+02 2.851e+02 6.880e+02, threshold=4.593e+02, percent-clipped=1.0 2023-03-09 01:08:44,089 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86950.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:08:59,943 INFO [train2.py:809] (3/4) Epoch 22, batch 3300, loss[ctc_loss=0.0947, att_loss=0.252, loss=0.2206, over 14318.00 frames. utt_duration=396.5 frames, utt_pad_proportion=0.3105, over 145.00 utterances.], tot_loss[ctc_loss=0.07175, att_loss=0.2349, loss=0.2023, over 3267981.87 frames. utt_duration=1220 frames, utt_pad_proportion=0.06403, over 10728.94 utterances.], batch size: 145, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:09:48,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-09 01:10:08,648 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87002.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 01:10:20,449 INFO [train2.py:809] (3/4) Epoch 22, batch 3350, loss[ctc_loss=0.06158, att_loss=0.2369, loss=0.2018, over 16478.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005891, over 46.00 utterances.], tot_loss[ctc_loss=0.07118, att_loss=0.2348, loss=0.2021, over 3273797.37 frames. utt_duration=1239 frames, utt_pad_proportion=0.05869, over 10585.82 utterances.], batch size: 46, lr: 4.86e-03, grad_scale: 8.0 2023-03-09 01:10:44,803 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9970, 4.0112, 3.7399, 2.7013, 3.7556, 3.7966, 3.5115, 2.6696], device='cuda:3'), covar=tensor([0.0145, 0.0147, 0.0358, 0.1068, 0.0153, 0.0462, 0.0382, 0.1371], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0102, 0.0106, 0.0111, 0.0086, 0.0115, 0.0100, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:11:13,697 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.863e+02 2.152e+02 2.724e+02 4.782e+02, threshold=4.303e+02, percent-clipped=1.0 2023-03-09 01:11:40,084 INFO [train2.py:809] (3/4) Epoch 22, batch 3400, loss[ctc_loss=0.06469, att_loss=0.2311, loss=0.1979, over 15954.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007026, over 41.00 utterances.], tot_loss[ctc_loss=0.07184, att_loss=0.2353, loss=0.2026, over 3271731.54 frames. utt_duration=1243 frames, utt_pad_proportion=0.05755, over 10543.65 utterances.], batch size: 41, lr: 4.85e-03, grad_scale: 8.0 2023-03-09 01:11:47,069 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87063.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:12:38,136 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87096.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:12:39,884 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4508, 2.6197, 4.9136, 3.7651, 2.9331, 4.1712, 4.6733, 4.5744], device='cuda:3'), covar=tensor([0.0272, 0.1522, 0.0198, 0.0906, 0.1734, 0.0281, 0.0161, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0238, 0.0192, 0.0313, 0.0261, 0.0215, 0.0180, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:12:46,638 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87101.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:12:59,258 INFO [train2.py:809] (3/4) Epoch 22, batch 3450, loss[ctc_loss=0.05617, att_loss=0.2219, loss=0.1888, over 16418.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.005852, over 44.00 utterances.], tot_loss[ctc_loss=0.07237, att_loss=0.2354, loss=0.2028, over 3273056.48 frames. utt_duration=1244 frames, utt_pad_proportion=0.05736, over 10535.69 utterances.], batch size: 44, lr: 4.85e-03, grad_scale: 8.0 2023-03-09 01:13:52,181 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.972e+02 2.278e+02 2.813e+02 5.019e+02, threshold=4.557e+02, percent-clipped=5.0 2023-03-09 01:13:54,604 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87144.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:14:02,273 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87149.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:14:19,572 INFO [train2.py:809] (3/4) Epoch 22, batch 3500, loss[ctc_loss=0.0675, att_loss=0.2113, loss=0.1825, over 15625.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009405, over 37.00 utterances.], tot_loss[ctc_loss=0.07294, att_loss=0.2357, loss=0.2032, over 3266586.42 frames. utt_duration=1224 frames, utt_pad_proportion=0.06251, over 10692.12 utterances.], batch size: 37, lr: 4.85e-03, grad_scale: 8.0 2023-03-09 01:14:37,055 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87170.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:15:39,628 INFO [train2.py:809] (3/4) Epoch 22, batch 3550, loss[ctc_loss=0.05631, att_loss=0.211, loss=0.1801, over 14990.00 frames. utt_duration=1819 frames, utt_pad_proportion=0.02529, over 33.00 utterances.], tot_loss[ctc_loss=0.07313, att_loss=0.2362, loss=0.2036, over 3268722.70 frames. utt_duration=1194 frames, utt_pad_proportion=0.06943, over 10962.12 utterances.], batch size: 33, lr: 4.85e-03, grad_scale: 8.0 2023-03-09 01:15:51,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-09 01:16:12,844 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87230.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:16:18,893 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87234.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:16:22,091 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87236.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:16:32,595 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.973e+02 2.339e+02 2.801e+02 5.420e+02, threshold=4.678e+02, percent-clipped=3.0 2023-03-09 01:16:59,777 INFO [train2.py:809] (3/4) Epoch 22, batch 3600, loss[ctc_loss=0.05208, att_loss=0.234, loss=0.1976, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005752, over 47.00 utterances.], tot_loss[ctc_loss=0.07236, att_loss=0.2357, loss=0.203, over 3269935.47 frames. utt_duration=1227 frames, utt_pad_proportion=0.06227, over 10670.09 utterances.], batch size: 47, lr: 4.85e-03, grad_scale: 16.0 2023-03-09 01:17:39,325 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87284.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:17:51,198 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:18:19,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-09 01:18:19,834 INFO [train2.py:809] (3/4) Epoch 22, batch 3650, loss[ctc_loss=0.05925, att_loss=0.2086, loss=0.1787, over 14478.00 frames. utt_duration=1811 frames, utt_pad_proportion=0.03112, over 32.00 utterances.], tot_loss[ctc_loss=0.0717, att_loss=0.2351, loss=0.2024, over 3272258.18 frames. utt_duration=1239 frames, utt_pad_proportion=0.05831, over 10580.04 utterances.], batch size: 32, lr: 4.85e-03, grad_scale: 16.0 2023-03-09 01:18:59,359 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 2023-03-09 01:19:13,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 1.934e+02 2.315e+02 2.927e+02 6.600e+02, threshold=4.631e+02, percent-clipped=3.0 2023-03-09 01:19:38,699 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87358.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 01:19:40,010 INFO [train2.py:809] (3/4) Epoch 22, batch 3700, loss[ctc_loss=0.07322, att_loss=0.238, loss=0.205, over 16278.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007584, over 43.00 utterances.], tot_loss[ctc_loss=0.07116, att_loss=0.2345, loss=0.2019, over 3268020.64 frames. utt_duration=1243 frames, utt_pad_proportion=0.05972, over 10533.43 utterances.], batch size: 43, lr: 4.85e-03, grad_scale: 16.0 2023-03-09 01:19:43,738 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-09 01:19:44,783 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87362.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:20:59,239 INFO [train2.py:809] (3/4) Epoch 22, batch 3750, loss[ctc_loss=0.05514, att_loss=0.2083, loss=0.1777, over 15873.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009861, over 39.00 utterances.], tot_loss[ctc_loss=0.07215, att_loss=0.2352, loss=0.2026, over 3258183.53 frames. utt_duration=1213 frames, utt_pad_proportion=0.06768, over 10761.26 utterances.], batch size: 39, lr: 4.84e-03, grad_scale: 16.0 2023-03-09 01:21:21,549 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87423.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:21:23,874 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-03-09 01:21:52,786 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.314e+02 1.951e+02 2.463e+02 3.155e+02 5.552e+02, threshold=4.927e+02, percent-clipped=3.0 2023-03-09 01:22:19,682 INFO [train2.py:809] (3/4) Epoch 22, batch 3800, loss[ctc_loss=0.05635, att_loss=0.218, loss=0.1856, over 15380.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.009331, over 35.00 utterances.], tot_loss[ctc_loss=0.07217, att_loss=0.2352, loss=0.2026, over 3251780.33 frames. utt_duration=1218 frames, utt_pad_proportion=0.06668, over 10689.61 utterances.], batch size: 35, lr: 4.84e-03, grad_scale: 16.0 2023-03-09 01:22:23,142 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87461.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:22:36,538 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-09 01:22:37,209 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87470.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:23:05,739 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1763, 2.8332, 3.5999, 3.0617, 3.4491, 4.4809, 4.3042, 3.2358], device='cuda:3'), covar=tensor([0.0485, 0.1822, 0.1299, 0.1355, 0.1184, 0.0990, 0.0641, 0.1180], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0242, 0.0281, 0.0221, 0.0265, 0.0372, 0.0263, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:23:34,646 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87505.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:23:40,700 INFO [train2.py:809] (3/4) Epoch 22, batch 3850, loss[ctc_loss=0.07352, att_loss=0.2406, loss=0.2072, over 17052.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009013, over 52.00 utterances.], tot_loss[ctc_loss=0.07229, att_loss=0.2346, loss=0.2021, over 3246670.76 frames. utt_duration=1201 frames, utt_pad_proportion=0.07301, over 10823.63 utterances.], batch size: 52, lr: 4.84e-03, grad_scale: 16.0 2023-03-09 01:23:54,830 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87518.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:24:01,329 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87522.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:24:19,715 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87534.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:24:30,434 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2425, 5.1592, 5.0378, 2.5190, 2.2607, 3.0209, 2.9681, 4.0166], device='cuda:3'), covar=tensor([0.0688, 0.0306, 0.0247, 0.4611, 0.5370, 0.2408, 0.3039, 0.1649], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0285, 0.0271, 0.0251, 0.0345, 0.0339, 0.0260, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 01:24:33,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.913e+02 2.224e+02 2.689e+02 7.925e+02, threshold=4.448e+02, percent-clipped=4.0 2023-03-09 01:24:41,146 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7019, 5.2432, 4.9793, 5.2402, 5.3175, 4.9040, 3.5820, 5.1883], device='cuda:3'), covar=tensor([0.0126, 0.0100, 0.0129, 0.0066, 0.0073, 0.0101, 0.0698, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0090, 0.0113, 0.0071, 0.0077, 0.0088, 0.0106, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:24:57,885 INFO [train2.py:809] (3/4) Epoch 22, batch 3900, loss[ctc_loss=0.07294, att_loss=0.2321, loss=0.2003, over 16535.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.005991, over 45.00 utterances.], tot_loss[ctc_loss=0.07188, att_loss=0.2346, loss=0.2021, over 3256741.45 frames. utt_duration=1215 frames, utt_pad_proportion=0.06575, over 10733.43 utterances.], batch size: 45, lr: 4.84e-03, grad_scale: 16.0 2023-03-09 01:25:08,973 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87566.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 01:25:15,060 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5657, 3.1069, 3.3252, 4.5874, 4.1938, 4.1423, 3.1287, 2.3892], device='cuda:3'), covar=tensor([0.0617, 0.1722, 0.1070, 0.0490, 0.0616, 0.0397, 0.1263, 0.1984], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0213, 0.0189, 0.0220, 0.0226, 0.0181, 0.0201, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:25:33,510 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87582.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:25:36,934 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87584.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:25:39,692 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87586.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:25:39,842 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3024, 2.8460, 3.6419, 3.1157, 3.3703, 4.5313, 4.3777, 3.1443], device='cuda:3'), covar=tensor([0.0429, 0.1787, 0.1162, 0.1219, 0.1220, 0.0900, 0.0601, 0.1263], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0244, 0.0283, 0.0223, 0.0267, 0.0375, 0.0266, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:25:44,704 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87589.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:26:09,217 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87605.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:26:14,867 INFO [train2.py:809] (3/4) Epoch 22, batch 3950, loss[ctc_loss=0.06499, att_loss=0.2388, loss=0.204, over 16330.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006141, over 45.00 utterances.], tot_loss[ctc_loss=0.07176, att_loss=0.2353, loss=0.2026, over 3262819.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.06306, over 10694.78 utterances.], batch size: 45, lr: 4.84e-03, grad_scale: 16.0 2023-03-09 01:27:30,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.383e+02 1.991e+02 2.302e+02 2.874e+02 7.527e+02, threshold=4.603e+02, percent-clipped=2.0 2023-03-09 01:27:30,800 INFO [train2.py:809] (3/4) Epoch 23, batch 0, loss[ctc_loss=0.105, att_loss=0.2664, loss=0.2341, over 14468.00 frames. utt_duration=397.7 frames, utt_pad_proportion=0.3072, over 146.00 utterances.], tot_loss[ctc_loss=0.105, att_loss=0.2664, loss=0.2341, over 14468.00 frames. utt_duration=397.7 frames, utt_pad_proportion=0.3072, over 146.00 utterances.], batch size: 146, lr: 4.73e-03, grad_scale: 16.0 2023-03-09 01:27:30,800 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 01:27:42,672 INFO [train2.py:843] (3/4) Epoch 23, validation: ctc_loss=0.04039, att_loss=0.2346, loss=0.1958, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 01:27:42,673 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 01:27:46,050 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87645.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 01:27:53,631 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87650.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:28:06,070 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87658.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 01:28:18,460 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87666.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:28:34,675 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5186, 3.0774, 3.6645, 3.1937, 3.5187, 4.6224, 4.4487, 3.3969], device='cuda:3'), covar=tensor([0.0412, 0.1600, 0.1180, 0.1285, 0.1083, 0.0854, 0.0547, 0.1156], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0244, 0.0284, 0.0223, 0.0267, 0.0374, 0.0265, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:29:02,027 INFO [train2.py:809] (3/4) Epoch 23, batch 50, loss[ctc_loss=0.07695, att_loss=0.2501, loss=0.2155, over 16959.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007076, over 50.00 utterances.], tot_loss[ctc_loss=0.07088, att_loss=0.2353, loss=0.2024, over 737450.74 frames. utt_duration=1330 frames, utt_pad_proportion=0.03394, over 2220.75 utterances.], batch size: 50, lr: 4.73e-03, grad_scale: 16.0 2023-03-09 01:29:22,818 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87706.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:29:41,154 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87718.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:30:21,899 INFO [train2.py:809] (3/4) Epoch 23, batch 100, loss[ctc_loss=0.1014, att_loss=0.2584, loss=0.227, over 13910.00 frames. utt_duration=382.5 frames, utt_pad_proportion=0.3312, over 146.00 utterances.], tot_loss[ctc_loss=0.07243, att_loss=0.2364, loss=0.2036, over 1301231.62 frames. utt_duration=1188 frames, utt_pad_proportion=0.06887, over 4386.04 utterances.], batch size: 146, lr: 4.73e-03, grad_scale: 8.0 2023-03-09 01:30:23,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.012e+02 2.393e+02 2.818e+02 6.680e+02, threshold=4.785e+02, percent-clipped=2.0 2023-03-09 01:31:28,518 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7621, 2.1314, 2.5930, 2.6139, 2.7431, 2.5810, 2.3360, 2.9134], device='cuda:3'), covar=tensor([0.1463, 0.3103, 0.1889, 0.1371, 0.1523, 0.1474, 0.2431, 0.1263], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0130, 0.0126, 0.0116, 0.0131, 0.0112, 0.0134, 0.0106], device='cuda:3'), out_proj_covar=tensor([9.5145e-05, 1.0111e-04, 1.0061e-04, 9.0637e-05, 9.7717e-05, 8.9868e-05, 1.0222e-04, 8.4808e-05], device='cuda:3') 2023-03-09 01:31:41,403 INFO [train2.py:809] (3/4) Epoch 23, batch 150, loss[ctc_loss=0.08821, att_loss=0.2472, loss=0.2154, over 17011.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.00837, over 51.00 utterances.], tot_loss[ctc_loss=0.07316, att_loss=0.2361, loss=0.2035, over 1740626.97 frames. utt_duration=1196 frames, utt_pad_proportion=0.06587, over 5830.11 utterances.], batch size: 51, lr: 4.73e-03, grad_scale: 8.0 2023-03-09 01:31:58,601 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1163, 4.4091, 4.4887, 4.6408, 2.8731, 4.5177, 2.5323, 1.9438], device='cuda:3'), covar=tensor([0.0420, 0.0255, 0.0635, 0.0194, 0.1446, 0.0206, 0.1564, 0.1583], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0172, 0.0263, 0.0161, 0.0223, 0.0152, 0.0233, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:32:20,539 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87817.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:32:45,241 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-09 01:33:02,589 INFO [train2.py:809] (3/4) Epoch 23, batch 200, loss[ctc_loss=0.07463, att_loss=0.2501, loss=0.215, over 16961.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007772, over 50.00 utterances.], tot_loss[ctc_loss=0.07235, att_loss=0.2354, loss=0.2028, over 2082229.92 frames. utt_duration=1245 frames, utt_pad_proportion=0.05261, over 6696.85 utterances.], batch size: 50, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:33:02,896 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6049, 4.6586, 4.3334, 2.5420, 4.3661, 4.3664, 3.9090, 2.3724], device='cuda:3'), covar=tensor([0.0150, 0.0122, 0.0343, 0.1305, 0.0150, 0.0269, 0.0407, 0.1679], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0103, 0.0107, 0.0113, 0.0086, 0.0116, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:33:04,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.173e+02 1.980e+02 2.350e+02 2.635e+02 6.250e+02, threshold=4.701e+02, percent-clipped=5.0 2023-03-09 01:33:31,420 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87861.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 01:33:34,469 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4894, 5.4672, 5.2742, 3.5427, 5.2525, 5.1366, 5.0468, 3.6418], device='cuda:3'), covar=tensor([0.0080, 0.0068, 0.0205, 0.0818, 0.0079, 0.0144, 0.0182, 0.0939], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0103, 0.0106, 0.0112, 0.0086, 0.0116, 0.0101, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:34:11,172 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87886.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:34:22,277 INFO [train2.py:809] (3/4) Epoch 23, batch 250, loss[ctc_loss=0.08859, att_loss=0.2421, loss=0.2114, over 16474.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006222, over 46.00 utterances.], tot_loss[ctc_loss=0.07177, att_loss=0.2359, loss=0.2031, over 2351218.23 frames. utt_duration=1241 frames, utt_pad_proportion=0.05148, over 7587.07 utterances.], batch size: 46, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:35:27,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87934.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:35:37,871 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87940.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 01:35:42,329 INFO [train2.py:809] (3/4) Epoch 23, batch 300, loss[ctc_loss=0.06897, att_loss=0.2437, loss=0.2088, over 16636.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004587, over 47.00 utterances.], tot_loss[ctc_loss=0.07148, att_loss=0.2357, loss=0.2029, over 2556962.61 frames. utt_duration=1224 frames, utt_pad_proportion=0.05635, over 8369.30 utterances.], batch size: 47, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:35:43,860 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.187e+02 1.810e+02 2.179e+02 2.816e+02 7.106e+02, threshold=4.359e+02, percent-clipped=3.0 2023-03-09 01:35:45,668 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:36:10,767 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87961.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 01:37:02,476 INFO [train2.py:809] (3/4) Epoch 23, batch 350, loss[ctc_loss=0.1085, att_loss=0.2578, loss=0.228, over 14315.00 frames. utt_duration=388.3 frames, utt_pad_proportion=0.3163, over 148.00 utterances.], tot_loss[ctc_loss=0.07152, att_loss=0.2353, loss=0.2025, over 2703815.05 frames. utt_duration=1206 frames, utt_pad_proportion=0.06554, over 8980.65 utterances.], batch size: 148, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:37:46,461 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88018.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:37:54,861 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5206, 2.4849, 4.9996, 3.8594, 2.9611, 4.1908, 4.7932, 4.6709], device='cuda:3'), covar=tensor([0.0265, 0.1682, 0.0193, 0.0960, 0.1793, 0.0241, 0.0149, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0240, 0.0193, 0.0313, 0.0262, 0.0215, 0.0182, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:38:27,599 INFO [train2.py:809] (3/4) Epoch 23, batch 400, loss[ctc_loss=0.09313, att_loss=0.2564, loss=0.2237, over 16932.00 frames. utt_duration=685.6 frames, utt_pad_proportion=0.1376, over 99.00 utterances.], tot_loss[ctc_loss=0.07207, att_loss=0.236, loss=0.2032, over 2831927.64 frames. utt_duration=1196 frames, utt_pad_proportion=0.06774, over 9485.55 utterances.], batch size: 99, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:38:29,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.944e+02 2.395e+02 3.108e+02 5.272e+02, threshold=4.790e+02, percent-clipped=4.0 2023-03-09 01:38:44,178 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8806, 3.6337, 3.5966, 3.0968, 3.6737, 3.6695, 3.6873, 2.6807], device='cuda:3'), covar=tensor([0.1141, 0.1243, 0.1792, 0.3814, 0.0960, 0.1415, 0.0822, 0.3658], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0192, 0.0204, 0.0258, 0.0163, 0.0266, 0.0188, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:39:04,806 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:39:48,104 INFO [train2.py:809] (3/4) Epoch 23, batch 450, loss[ctc_loss=0.1005, att_loss=0.2617, loss=0.2295, over 17084.00 frames. utt_duration=1316 frames, utt_pad_proportion=0.007068, over 52.00 utterances.], tot_loss[ctc_loss=0.07148, att_loss=0.2356, loss=0.2027, over 2931667.18 frames. utt_duration=1228 frames, utt_pad_proportion=0.05849, over 9562.96 utterances.], batch size: 52, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:40:00,138 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88100.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:40:26,338 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88117.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:40:45,608 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4837, 2.9162, 3.6690, 3.0574, 3.4870, 4.5911, 4.3772, 3.1436], device='cuda:3'), covar=tensor([0.0392, 0.1791, 0.1250, 0.1340, 0.1106, 0.0761, 0.0513, 0.1270], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0239, 0.0279, 0.0220, 0.0262, 0.0367, 0.0260, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 01:41:09,095 INFO [train2.py:809] (3/4) Epoch 23, batch 500, loss[ctc_loss=0.07463, att_loss=0.2372, loss=0.2047, over 17031.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.01093, over 53.00 utterances.], tot_loss[ctc_loss=0.07094, att_loss=0.235, loss=0.2022, over 3010229.91 frames. utt_duration=1252 frames, utt_pad_proportion=0.05284, over 9630.41 utterances.], batch size: 53, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:41:10,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 1.977e+02 2.397e+02 3.059e+02 5.388e+02, threshold=4.794e+02, percent-clipped=2.0 2023-03-09 01:41:37,516 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88161.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:41:37,590 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88161.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:41:44,291 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88165.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:42:08,343 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6435, 2.2755, 2.4109, 2.6599, 2.6739, 2.6062, 2.4735, 2.9206], device='cuda:3'), covar=tensor([0.1662, 0.2966, 0.2104, 0.1437, 0.1715, 0.1445, 0.2125, 0.1280], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0128, 0.0124, 0.0114, 0.0129, 0.0110, 0.0132, 0.0105], device='cuda:3'), out_proj_covar=tensor([9.3565e-05, 9.9613e-05, 9.9404e-05, 8.9266e-05, 9.6077e-05, 8.8901e-05, 1.0076e-04, 8.3962e-05], device='cuda:3') 2023-03-09 01:42:29,383 INFO [train2.py:809] (3/4) Epoch 23, batch 550, loss[ctc_loss=0.06709, att_loss=0.2374, loss=0.2034, over 17015.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008902, over 51.00 utterances.], tot_loss[ctc_loss=0.07133, att_loss=0.2353, loss=0.2025, over 3067194.89 frames. utt_duration=1245 frames, utt_pad_proportion=0.0565, over 9869.50 utterances.], batch size: 51, lr: 4.72e-03, grad_scale: 8.0 2023-03-09 01:42:54,654 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88209.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:43:06,946 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-09 01:43:46,363 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88240.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:43:51,891 INFO [train2.py:809] (3/4) Epoch 23, batch 600, loss[ctc_loss=0.06514, att_loss=0.2334, loss=0.1998, over 16159.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.007624, over 41.00 utterances.], tot_loss[ctc_loss=0.07201, att_loss=0.2357, loss=0.203, over 3118851.33 frames. utt_duration=1221 frames, utt_pad_proportion=0.05903, over 10229.38 utterances.], batch size: 41, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:43:52,311 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9335, 3.7028, 3.6867, 3.0567, 3.6818, 3.6091, 3.7163, 2.6051], device='cuda:3'), covar=tensor([0.1162, 0.1033, 0.1341, 0.3174, 0.1015, 0.2621, 0.0893, 0.4022], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0194, 0.0206, 0.0260, 0.0165, 0.0267, 0.0189, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:43:53,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 1.886e+02 2.333e+02 2.841e+02 4.176e+02, threshold=4.666e+02, percent-clipped=0.0 2023-03-09 01:43:55,480 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88245.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:44:03,944 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-09 01:44:20,766 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88261.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:44:35,385 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9750, 3.7748, 3.7032, 3.2657, 3.7521, 3.7848, 3.7292, 2.8064], device='cuda:3'), covar=tensor([0.1098, 0.1178, 0.1741, 0.3317, 0.1255, 0.1359, 0.1158, 0.3899], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0194, 0.0206, 0.0260, 0.0165, 0.0267, 0.0190, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:45:04,433 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88288.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:45:10,133 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88291.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:45:12,840 INFO [train2.py:809] (3/4) Epoch 23, batch 650, loss[ctc_loss=0.05743, att_loss=0.2209, loss=0.1882, over 16273.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007645, over 43.00 utterances.], tot_loss[ctc_loss=0.07174, att_loss=0.2354, loss=0.2027, over 3158056.20 frames. utt_duration=1226 frames, utt_pad_proportion=0.05732, over 10312.16 utterances.], batch size: 43, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:45:12,984 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88293.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:45:38,340 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88309.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:46:01,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0737, 6.3099, 5.8583, 6.0161, 6.0441, 5.5651, 5.7335, 5.5609], device='cuda:3'), covar=tensor([0.1223, 0.0934, 0.0888, 0.0839, 0.0885, 0.1442, 0.2394, 0.2254], device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0617, 0.0468, 0.0463, 0.0438, 0.0468, 0.0618, 0.0530], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:46:27,912 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7685, 2.3848, 2.3919, 2.6129, 2.6823, 2.6505, 2.5369, 2.9488], device='cuda:3'), covar=tensor([0.1457, 0.2480, 0.1841, 0.1382, 0.1448, 0.1153, 0.1653, 0.1086], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0129, 0.0125, 0.0115, 0.0129, 0.0111, 0.0133, 0.0105], device='cuda:3'), out_proj_covar=tensor([9.4020e-05, 1.0006e-04, 1.0014e-04, 8.9819e-05, 9.6424e-05, 8.9265e-05, 1.0144e-04, 8.4142e-05], device='cuda:3') 2023-03-09 01:46:34,457 INFO [train2.py:809] (3/4) Epoch 23, batch 700, loss[ctc_loss=0.08335, att_loss=0.2388, loss=0.2077, over 16335.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005928, over 45.00 utterances.], tot_loss[ctc_loss=0.07173, att_loss=0.2353, loss=0.2026, over 3189859.85 frames. utt_duration=1243 frames, utt_pad_proportion=0.05135, over 10280.96 utterances.], batch size: 45, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:46:35,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 1.968e+02 2.314e+02 2.837e+02 1.022e+03, threshold=4.627e+02, percent-clipped=4.0 2023-03-09 01:46:49,141 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88352.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:47:13,837 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-09 01:47:56,420 INFO [train2.py:809] (3/4) Epoch 23, batch 750, loss[ctc_loss=0.07109, att_loss=0.2181, loss=0.1887, over 16168.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006339, over 41.00 utterances.], tot_loss[ctc_loss=0.07174, att_loss=0.2352, loss=0.2025, over 3209343.58 frames. utt_duration=1241 frames, utt_pad_proportion=0.05263, over 10359.10 utterances.], batch size: 41, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:48:44,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-03-09 01:49:17,982 INFO [train2.py:809] (3/4) Epoch 23, batch 800, loss[ctc_loss=0.08692, att_loss=0.2586, loss=0.2242, over 17136.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.01339, over 56.00 utterances.], tot_loss[ctc_loss=0.07134, att_loss=0.2349, loss=0.2022, over 3224713.17 frames. utt_duration=1257 frames, utt_pad_proportion=0.04938, over 10271.46 utterances.], batch size: 56, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:49:19,538 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.999e+02 2.331e+02 2.822e+02 6.920e+02, threshold=4.662e+02, percent-clipped=4.0 2023-03-09 01:49:26,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-03-09 01:49:39,402 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88456.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:50:28,796 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8779, 4.8734, 4.6032, 2.7400, 4.5506, 4.5663, 3.9260, 2.5653], device='cuda:3'), covar=tensor([0.0162, 0.0105, 0.0300, 0.1221, 0.0122, 0.0219, 0.0411, 0.1455], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0101, 0.0105, 0.0111, 0.0085, 0.0114, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:50:39,454 INFO [train2.py:809] (3/4) Epoch 23, batch 850, loss[ctc_loss=0.05357, att_loss=0.2228, loss=0.1889, over 16158.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.007045, over 41.00 utterances.], tot_loss[ctc_loss=0.07156, att_loss=0.2358, loss=0.203, over 3243798.25 frames. utt_duration=1240 frames, utt_pad_proportion=0.0519, over 10476.59 utterances.], batch size: 41, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:52:02,087 INFO [train2.py:809] (3/4) Epoch 23, batch 900, loss[ctc_loss=0.07247, att_loss=0.2554, loss=0.2188, over 17065.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.00876, over 53.00 utterances.], tot_loss[ctc_loss=0.07141, att_loss=0.2357, loss=0.2029, over 3257407.07 frames. utt_duration=1245 frames, utt_pad_proportion=0.05013, over 10481.14 utterances.], batch size: 53, lr: 4.71e-03, grad_scale: 8.0 2023-03-09 01:52:03,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 1.915e+02 2.378e+02 3.006e+02 6.830e+02, threshold=4.756e+02, percent-clipped=6.0 2023-03-09 01:53:24,722 INFO [train2.py:809] (3/4) Epoch 23, batch 950, loss[ctc_loss=0.06012, att_loss=0.2426, loss=0.2061, over 17284.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01212, over 55.00 utterances.], tot_loss[ctc_loss=0.07081, att_loss=0.235, loss=0.2022, over 3254365.14 frames. utt_duration=1255 frames, utt_pad_proportion=0.05091, over 10385.59 utterances.], batch size: 55, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 01:54:46,090 INFO [train2.py:809] (3/4) Epoch 23, batch 1000, loss[ctc_loss=0.06716, att_loss=0.2308, loss=0.1981, over 16256.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.008092, over 43.00 utterances.], tot_loss[ctc_loss=0.07068, att_loss=0.2343, loss=0.2016, over 3252559.43 frames. utt_duration=1275 frames, utt_pad_proportion=0.04806, over 10219.45 utterances.], batch size: 43, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 01:54:48,311 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.864e+02 2.157e+02 2.686e+02 4.602e+02, threshold=4.313e+02, percent-clipped=0.0 2023-03-09 01:54:53,439 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88647.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 01:55:06,282 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6242, 2.2389, 2.1469, 2.6115, 2.4786, 2.5230, 2.3096, 2.9463], device='cuda:3'), covar=tensor([0.2508, 0.4417, 0.2930, 0.2444, 0.2731, 0.2391, 0.4015, 0.2817], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0132, 0.0129, 0.0118, 0.0132, 0.0114, 0.0138, 0.0108], device='cuda:3'), out_proj_covar=tensor([9.6514e-05, 1.0227e-04, 1.0255e-04, 9.2062e-05, 9.8913e-05, 9.1663e-05, 1.0456e-04, 8.6102e-05], device='cuda:3') 2023-03-09 01:56:09,550 INFO [train2.py:809] (3/4) Epoch 23, batch 1050, loss[ctc_loss=0.06469, att_loss=0.2137, loss=0.1839, over 13251.00 frames. utt_duration=1829 frames, utt_pad_proportion=0.04155, over 29.00 utterances.], tot_loss[ctc_loss=0.07096, att_loss=0.2343, loss=0.2016, over 3254573.58 frames. utt_duration=1273 frames, utt_pad_proportion=0.04931, over 10242.36 utterances.], batch size: 29, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 01:56:25,898 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3648, 3.8538, 3.3758, 3.7670, 4.1450, 3.9199, 3.2380, 4.4864], device='cuda:3'), covar=tensor([0.0892, 0.0611, 0.1050, 0.0568, 0.0658, 0.0590, 0.0790, 0.0409], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0219, 0.0230, 0.0204, 0.0283, 0.0245, 0.0202, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 01:57:15,122 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1483, 4.4802, 4.7245, 4.8681, 2.8407, 4.5574, 2.7719, 2.0241], device='cuda:3'), covar=tensor([0.0499, 0.0293, 0.0485, 0.0177, 0.1531, 0.0180, 0.1412, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0169, 0.0258, 0.0160, 0.0222, 0.0152, 0.0230, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:57:30,982 INFO [train2.py:809] (3/4) Epoch 23, batch 1100, loss[ctc_loss=0.06159, att_loss=0.2457, loss=0.2089, over 17358.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03628, over 63.00 utterances.], tot_loss[ctc_loss=0.07117, att_loss=0.2344, loss=0.2017, over 3245436.97 frames. utt_duration=1231 frames, utt_pad_proportion=0.06235, over 10562.55 utterances.], batch size: 63, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 01:57:32,494 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.060e+02 2.487e+02 3.193e+02 8.233e+02, threshold=4.974e+02, percent-clipped=4.0 2023-03-09 01:57:49,063 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2593, 2.6043, 3.1296, 4.1572, 3.7493, 3.6978, 2.6984, 2.1316], device='cuda:3'), covar=tensor([0.0766, 0.2349, 0.0999, 0.0688, 0.0897, 0.0584, 0.1760, 0.2447], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0215, 0.0189, 0.0223, 0.0228, 0.0181, 0.0203, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 01:57:52,276 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88756.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:58:07,185 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7926, 6.0062, 5.4832, 5.7733, 5.6572, 5.1865, 5.4945, 5.2745], device='cuda:3'), covar=tensor([0.1247, 0.0950, 0.1067, 0.0837, 0.1011, 0.1518, 0.2165, 0.2203], device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0625, 0.0471, 0.0466, 0.0444, 0.0473, 0.0622, 0.0536], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 01:58:52,970 INFO [train2.py:809] (3/4) Epoch 23, batch 1150, loss[ctc_loss=0.08346, att_loss=0.2535, loss=0.2195, over 17265.00 frames. utt_duration=1257 frames, utt_pad_proportion=0.0139, over 55.00 utterances.], tot_loss[ctc_loss=0.07078, att_loss=0.2342, loss=0.2015, over 3250556.73 frames. utt_duration=1244 frames, utt_pad_proportion=0.05814, over 10461.22 utterances.], batch size: 55, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 01:59:11,779 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88804.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 01:59:30,575 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 02:00:16,047 INFO [train2.py:809] (3/4) Epoch 23, batch 1200, loss[ctc_loss=0.05718, att_loss=0.2111, loss=0.1803, over 15880.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009611, over 39.00 utterances.], tot_loss[ctc_loss=0.07089, att_loss=0.2347, loss=0.2019, over 3259948.30 frames. utt_duration=1220 frames, utt_pad_proportion=0.06265, over 10700.25 utterances.], batch size: 39, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 02:00:17,521 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.914e+02 2.277e+02 2.770e+02 5.405e+02, threshold=4.555e+02, percent-clipped=1.0 2023-03-09 02:00:38,619 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3489, 2.7491, 3.2587, 4.3570, 3.8350, 3.8290, 2.9651, 2.1869], device='cuda:3'), covar=tensor([0.0701, 0.1933, 0.0919, 0.0490, 0.0885, 0.0485, 0.1476, 0.2137], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0215, 0.0189, 0.0222, 0.0229, 0.0182, 0.0204, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:01:37,934 INFO [train2.py:809] (3/4) Epoch 23, batch 1250, loss[ctc_loss=0.07214, att_loss=0.227, loss=0.196, over 16539.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006362, over 45.00 utterances.], tot_loss[ctc_loss=0.07177, att_loss=0.2353, loss=0.2026, over 3260341.58 frames. utt_duration=1209 frames, utt_pad_proportion=0.0654, over 10798.40 utterances.], batch size: 45, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 02:01:43,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-03-09 02:02:33,468 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-09 02:02:38,322 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0071, 5.1736, 4.9265, 2.2703, 2.0064, 2.9786, 2.5749, 3.8977], device='cuda:3'), covar=tensor([0.0757, 0.0317, 0.0289, 0.5253, 0.5704, 0.2296, 0.3537, 0.1710], device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0285, 0.0273, 0.0249, 0.0345, 0.0337, 0.0261, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 02:03:00,073 INFO [train2.py:809] (3/4) Epoch 23, batch 1300, loss[ctc_loss=0.08172, att_loss=0.2468, loss=0.2138, over 17277.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02508, over 59.00 utterances.], tot_loss[ctc_loss=0.07299, att_loss=0.2362, loss=0.2036, over 3265090.78 frames. utt_duration=1190 frames, utt_pad_proportion=0.06977, over 10989.06 utterances.], batch size: 59, lr: 4.70e-03, grad_scale: 8.0 2023-03-09 02:03:01,703 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.977e+02 2.384e+02 3.004e+02 6.014e+02, threshold=4.768e+02, percent-clipped=3.0 2023-03-09 02:03:06,867 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88947.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 02:04:21,365 INFO [train2.py:809] (3/4) Epoch 23, batch 1350, loss[ctc_loss=0.05723, att_loss=0.2081, loss=0.178, over 11879.00 frames. utt_duration=1829 frames, utt_pad_proportion=0.1602, over 26.00 utterances.], tot_loss[ctc_loss=0.07262, att_loss=0.236, loss=0.2034, over 3263727.59 frames. utt_duration=1217 frames, utt_pad_proportion=0.06309, over 10739.85 utterances.], batch size: 26, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:04:24,666 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88995.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 02:04:28,060 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8728, 3.4861, 3.5090, 3.0166, 3.5405, 3.6251, 3.5797, 2.5603], device='cuda:3'), covar=tensor([0.1252, 0.1714, 0.1853, 0.3633, 0.1299, 0.2589, 0.0997, 0.4064], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0194, 0.0205, 0.0261, 0.0166, 0.0268, 0.0191, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:05:17,195 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89027.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:05:43,318 INFO [train2.py:809] (3/4) Epoch 23, batch 1400, loss[ctc_loss=0.0861, att_loss=0.2467, loss=0.2146, over 17288.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01281, over 55.00 utterances.], tot_loss[ctc_loss=0.07283, att_loss=0.236, loss=0.2034, over 3261024.69 frames. utt_duration=1206 frames, utt_pad_proportion=0.06594, over 10827.90 utterances.], batch size: 55, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:05:44,870 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 1.935e+02 2.297e+02 2.917e+02 5.540e+02, threshold=4.593e+02, percent-clipped=1.0 2023-03-09 02:06:38,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1072, 4.3536, 4.2579, 4.7678, 2.6304, 4.4533, 2.6761, 1.6675], device='cuda:3'), covar=tensor([0.0489, 0.0273, 0.0861, 0.0191, 0.2077, 0.0215, 0.1776, 0.2138], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0169, 0.0261, 0.0161, 0.0223, 0.0153, 0.0232, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:06:57,066 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8722, 6.1501, 5.6727, 5.8880, 5.8133, 5.2654, 5.5824, 5.3783], device='cuda:3'), covar=tensor([0.1333, 0.0960, 0.0872, 0.0757, 0.0905, 0.1749, 0.2409, 0.2445], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0620, 0.0469, 0.0462, 0.0439, 0.0471, 0.0620, 0.0536], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:06:57,325 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89088.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:07:04,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-09 02:07:04,593 INFO [train2.py:809] (3/4) Epoch 23, batch 1450, loss[ctc_loss=0.0751, att_loss=0.2512, loss=0.216, over 16756.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.007117, over 48.00 utterances.], tot_loss[ctc_loss=0.0725, att_loss=0.2354, loss=0.2028, over 3261678.97 frames. utt_duration=1228 frames, utt_pad_proportion=0.06092, over 10641.28 utterances.], batch size: 48, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:08:26,635 INFO [train2.py:809] (3/4) Epoch 23, batch 1500, loss[ctc_loss=0.1066, att_loss=0.2607, loss=0.2299, over 14006.00 frames. utt_duration=382.5 frames, utt_pad_proportion=0.3302, over 147.00 utterances.], tot_loss[ctc_loss=0.07252, att_loss=0.2356, loss=0.203, over 3268956.30 frames. utt_duration=1227 frames, utt_pad_proportion=0.06018, over 10670.50 utterances.], batch size: 147, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:08:28,106 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 1.977e+02 2.384e+02 2.734e+02 5.349e+02, threshold=4.768e+02, percent-clipped=3.0 2023-03-09 02:09:08,680 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89169.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:09:48,304 INFO [train2.py:809] (3/4) Epoch 23, batch 1550, loss[ctc_loss=0.05335, att_loss=0.2219, loss=0.1882, over 16400.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007692, over 44.00 utterances.], tot_loss[ctc_loss=0.07199, att_loss=0.2354, loss=0.2028, over 3275090.32 frames. utt_duration=1253 frames, utt_pad_proportion=0.05254, over 10471.53 utterances.], batch size: 44, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:10:49,440 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89230.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:11:10,189 INFO [train2.py:809] (3/4) Epoch 23, batch 1600, loss[ctc_loss=0.05684, att_loss=0.2234, loss=0.1901, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.00545, over 45.00 utterances.], tot_loss[ctc_loss=0.07182, att_loss=0.2352, loss=0.2025, over 3276884.05 frames. utt_duration=1254 frames, utt_pad_proportion=0.05227, over 10463.31 utterances.], batch size: 45, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:11:11,786 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.895e+02 2.282e+02 2.881e+02 5.348e+02, threshold=4.565e+02, percent-clipped=1.0 2023-03-09 02:12:31,873 INFO [train2.py:809] (3/4) Epoch 23, batch 1650, loss[ctc_loss=0.0741, att_loss=0.2319, loss=0.2004, over 16281.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007128, over 43.00 utterances.], tot_loss[ctc_loss=0.07201, att_loss=0.2355, loss=0.2028, over 3277398.01 frames. utt_duration=1204 frames, utt_pad_proportion=0.06379, over 10901.99 utterances.], batch size: 43, lr: 4.69e-03, grad_scale: 8.0 2023-03-09 02:13:54,209 INFO [train2.py:809] (3/4) Epoch 23, batch 1700, loss[ctc_loss=0.0764, att_loss=0.2353, loss=0.2035, over 16536.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006362, over 45.00 utterances.], tot_loss[ctc_loss=0.07158, att_loss=0.2345, loss=0.2019, over 3272088.13 frames. utt_duration=1217 frames, utt_pad_proportion=0.0627, over 10764.11 utterances.], batch size: 45, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:13:55,725 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 1.978e+02 2.274e+02 2.807e+02 5.898e+02, threshold=4.548e+02, percent-clipped=3.0 2023-03-09 02:14:45,037 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89374.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:15:00,051 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89383.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:15:15,727 INFO [train2.py:809] (3/4) Epoch 23, batch 1750, loss[ctc_loss=0.06952, att_loss=0.2103, loss=0.1821, over 15508.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008194, over 36.00 utterances.], tot_loss[ctc_loss=0.071, att_loss=0.2343, loss=0.2016, over 3275702.16 frames. utt_duration=1233 frames, utt_pad_proportion=0.05797, over 10641.52 utterances.], batch size: 36, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:16:09,009 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89425.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:16:25,522 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89435.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:16:31,821 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89439.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 02:16:37,757 INFO [train2.py:809] (3/4) Epoch 23, batch 1800, loss[ctc_loss=0.0786, att_loss=0.2442, loss=0.211, over 17296.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01228, over 55.00 utterances.], tot_loss[ctc_loss=0.07183, att_loss=0.2347, loss=0.2022, over 3275014.66 frames. utt_duration=1194 frames, utt_pad_proportion=0.06763, over 10983.91 utterances.], batch size: 55, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:16:39,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 1.934e+02 2.267e+02 2.788e+02 6.296e+02, threshold=4.533e+02, percent-clipped=2.0 2023-03-09 02:16:43,444 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1824, 5.1664, 4.7210, 2.8404, 5.0040, 4.9617, 4.1512, 2.3051], device='cuda:3'), covar=tensor([0.0169, 0.0123, 0.0404, 0.1411, 0.0125, 0.0186, 0.0517, 0.2233], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0102, 0.0105, 0.0111, 0.0086, 0.0114, 0.0100, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:17:49,835 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89486.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:18:00,958 INFO [train2.py:809] (3/4) Epoch 23, batch 1850, loss[ctc_loss=0.06567, att_loss=0.2141, loss=0.1844, over 12770.00 frames. utt_duration=1826 frames, utt_pad_proportion=0.1211, over 28.00 utterances.], tot_loss[ctc_loss=0.07189, att_loss=0.2344, loss=0.2019, over 3263940.93 frames. utt_duration=1179 frames, utt_pad_proportion=0.07511, over 11085.53 utterances.], batch size: 28, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:18:12,525 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89500.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 02:18:27,088 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89509.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:18:53,259 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89525.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:19:23,208 INFO [train2.py:809] (3/4) Epoch 23, batch 1900, loss[ctc_loss=0.08637, att_loss=0.2466, loss=0.2145, over 17466.00 frames. utt_duration=885.9 frames, utt_pad_proportion=0.07237, over 79.00 utterances.], tot_loss[ctc_loss=0.07165, att_loss=0.2339, loss=0.2015, over 3265944.48 frames. utt_duration=1205 frames, utt_pad_proportion=0.0685, over 10854.19 utterances.], batch size: 79, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:19:24,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.857e+02 2.214e+02 2.613e+02 6.463e+02, threshold=4.427e+02, percent-clipped=1.0 2023-03-09 02:20:07,575 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89570.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:20:44,882 INFO [train2.py:809] (3/4) Epoch 23, batch 1950, loss[ctc_loss=0.09251, att_loss=0.2536, loss=0.2214, over 14590.00 frames. utt_duration=401.2 frames, utt_pad_proportion=0.3011, over 146.00 utterances.], tot_loss[ctc_loss=0.07196, att_loss=0.2345, loss=0.202, over 3264898.37 frames. utt_duration=1211 frames, utt_pad_proportion=0.06802, over 10793.07 utterances.], batch size: 146, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:21:55,234 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4599, 2.4832, 4.9893, 3.9333, 3.1642, 4.2674, 4.7456, 4.6253], device='cuda:3'), covar=tensor([0.0294, 0.1646, 0.0195, 0.0807, 0.1529, 0.0250, 0.0159, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0240, 0.0193, 0.0313, 0.0261, 0.0216, 0.0183, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:22:06,987 INFO [train2.py:809] (3/4) Epoch 23, batch 2000, loss[ctc_loss=0.09486, att_loss=0.2516, loss=0.2202, over 17291.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01175, over 55.00 utterances.], tot_loss[ctc_loss=0.07272, att_loss=0.2349, loss=0.2025, over 3263781.25 frames. utt_duration=1210 frames, utt_pad_proportion=0.0668, over 10799.78 utterances.], batch size: 55, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:22:08,496 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 1.989e+02 2.307e+02 2.806e+02 7.990e+02, threshold=4.613e+02, percent-clipped=4.0 2023-03-09 02:22:11,816 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1219, 5.4506, 5.7058, 5.4238, 5.6224, 6.0959, 5.3662, 6.1477], device='cuda:3'), covar=tensor([0.0748, 0.0682, 0.0747, 0.1357, 0.1874, 0.0903, 0.0597, 0.0743], device='cuda:3'), in_proj_covar=tensor([0.0872, 0.0510, 0.0609, 0.0661, 0.0874, 0.0631, 0.0491, 0.0610], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 02:23:07,542 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89680.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:23:12,492 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89683.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:23:17,764 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3234, 3.8106, 3.2719, 3.5337, 3.9995, 3.6706, 3.1551, 4.3440], device='cuda:3'), covar=tensor([0.0878, 0.0478, 0.1029, 0.0653, 0.0653, 0.0752, 0.0793, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0217, 0.0227, 0.0202, 0.0280, 0.0243, 0.0198, 0.0291], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 02:23:27,934 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-09 02:23:28,545 INFO [train2.py:809] (3/4) Epoch 23, batch 2050, loss[ctc_loss=0.05522, att_loss=0.2166, loss=0.1843, over 15954.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006315, over 41.00 utterances.], tot_loss[ctc_loss=0.07199, att_loss=0.2343, loss=0.2019, over 3261139.54 frames. utt_duration=1227 frames, utt_pad_proportion=0.06269, over 10643.27 utterances.], batch size: 41, lr: 4.68e-03, grad_scale: 8.0 2023-03-09 02:24:18,253 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1379, 4.2671, 4.2203, 4.5311, 2.7385, 4.3222, 2.9694, 1.8545], device='cuda:3'), covar=tensor([0.0423, 0.0272, 0.0764, 0.0228, 0.1676, 0.0226, 0.1329, 0.1754], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0170, 0.0262, 0.0161, 0.0222, 0.0154, 0.0232, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:24:28,913 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89730.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:24:30,399 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89731.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:24:47,316 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89741.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:24:50,038 INFO [train2.py:809] (3/4) Epoch 23, batch 2100, loss[ctc_loss=0.06845, att_loss=0.2423, loss=0.2075, over 16774.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006229, over 48.00 utterances.], tot_loss[ctc_loss=0.07146, att_loss=0.2338, loss=0.2013, over 3258975.74 frames. utt_duration=1242 frames, utt_pad_proportion=0.05879, over 10507.35 utterances.], batch size: 48, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:24:51,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 1.974e+02 2.233e+02 2.769e+02 5.380e+02, threshold=4.466e+02, percent-clipped=5.0 2023-03-09 02:25:52,098 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89781.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:26:11,807 INFO [train2.py:809] (3/4) Epoch 23, batch 2150, loss[ctc_loss=0.05952, att_loss=0.2171, loss=0.1856, over 15899.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008381, over 39.00 utterances.], tot_loss[ctc_loss=0.07157, att_loss=0.2332, loss=0.2009, over 3249753.69 frames. utt_duration=1252 frames, utt_pad_proportion=0.05864, over 10398.34 utterances.], batch size: 39, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:26:15,137 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89795.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 02:27:04,738 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89825.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:27:08,410 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-09 02:27:33,613 INFO [train2.py:809] (3/4) Epoch 23, batch 2200, loss[ctc_loss=0.08202, att_loss=0.244, loss=0.2116, over 17163.00 frames. utt_duration=870.6 frames, utt_pad_proportion=0.08741, over 79.00 utterances.], tot_loss[ctc_loss=0.07206, att_loss=0.2339, loss=0.2015, over 3256564.97 frames. utt_duration=1244 frames, utt_pad_proportion=0.05872, over 10484.96 utterances.], batch size: 79, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:27:34,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 1.973e+02 2.386e+02 3.013e+02 6.689e+02, threshold=4.771e+02, percent-clipped=5.0 2023-03-09 02:27:40,261 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5812, 2.6259, 2.3404, 2.5792, 2.6359, 2.3809, 2.3822, 2.9605], device='cuda:3'), covar=tensor([0.1353, 0.1965, 0.1689, 0.1277, 0.1387, 0.1350, 0.1759, 0.1029], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0126, 0.0123, 0.0114, 0.0129, 0.0112, 0.0134, 0.0106], device='cuda:3'), out_proj_covar=tensor([9.4737e-05, 9.8659e-05, 9.9098e-05, 8.9275e-05, 9.6855e-05, 9.0023e-05, 1.0185e-04, 8.4387e-05], device='cuda:3') 2023-03-09 02:27:44,050 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8890, 3.7366, 3.1514, 3.4113, 3.9045, 3.5653, 2.8035, 4.1069], device='cuda:3'), covar=tensor([0.1105, 0.0476, 0.1176, 0.0680, 0.0699, 0.0758, 0.0947, 0.0493], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0218, 0.0228, 0.0203, 0.0281, 0.0244, 0.0200, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 02:27:45,644 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5597, 3.0152, 3.6815, 3.0359, 3.5960, 4.6562, 4.4587, 3.4171], device='cuda:3'), covar=tensor([0.0373, 0.1818, 0.1301, 0.1532, 0.1079, 0.0715, 0.0652, 0.1209], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0248, 0.0285, 0.0225, 0.0268, 0.0378, 0.0268, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 02:27:53,561 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8528, 4.9483, 4.3816, 2.7083, 4.7335, 4.6945, 3.9182, 2.2832], device='cuda:3'), covar=tensor([0.0220, 0.0140, 0.0531, 0.1551, 0.0144, 0.0240, 0.0571, 0.2415], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0102, 0.0105, 0.0111, 0.0085, 0.0114, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:28:09,098 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:28:09,273 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:28:23,118 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89873.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:28:54,728 INFO [train2.py:809] (3/4) Epoch 23, batch 2250, loss[ctc_loss=0.06741, att_loss=0.2481, loss=0.2119, over 16872.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006646, over 49.00 utterances.], tot_loss[ctc_loss=0.07168, att_loss=0.2342, loss=0.2017, over 3266108.06 frames. utt_duration=1266 frames, utt_pad_proportion=0.05105, over 10330.55 utterances.], batch size: 49, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:29:48,468 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89926.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 02:30:14,729 INFO [train2.py:809] (3/4) Epoch 23, batch 2300, loss[ctc_loss=0.06792, att_loss=0.2438, loss=0.2087, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01171, over 55.00 utterances.], tot_loss[ctc_loss=0.07171, att_loss=0.2343, loss=0.2017, over 3262629.19 frames. utt_duration=1256 frames, utt_pad_proportion=0.05338, over 10402.93 utterances.], batch size: 55, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:30:16,395 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 1.850e+02 2.227e+02 2.792e+02 5.012e+02, threshold=4.455e+02, percent-clipped=2.0 2023-03-09 02:31:36,740 INFO [train2.py:809] (3/4) Epoch 23, batch 2350, loss[ctc_loss=0.07144, att_loss=0.2199, loss=0.1902, over 16395.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007258, over 44.00 utterances.], tot_loss[ctc_loss=0.07173, att_loss=0.2346, loss=0.202, over 3265321.57 frames. utt_duration=1250 frames, utt_pad_proportion=0.05478, over 10462.69 utterances.], batch size: 44, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:32:13,288 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8018, 3.5578, 3.5279, 3.0268, 3.5510, 3.5634, 3.6031, 2.5409], device='cuda:3'), covar=tensor([0.1265, 0.1857, 0.2103, 0.3881, 0.1142, 0.3012, 0.1063, 0.4184], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0191, 0.0205, 0.0260, 0.0164, 0.0266, 0.0190, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:32:42,524 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90030.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:32:52,449 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90036.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:33:03,244 INFO [train2.py:809] (3/4) Epoch 23, batch 2400, loss[ctc_loss=0.05655, att_loss=0.2173, loss=0.1852, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007017, over 40.00 utterances.], tot_loss[ctc_loss=0.07127, att_loss=0.2347, loss=0.202, over 3268869.54 frames. utt_duration=1244 frames, utt_pad_proportion=0.05573, over 10524.71 utterances.], batch size: 40, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:33:04,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.874e+02 2.405e+02 2.931e+02 6.452e+02, threshold=4.810e+02, percent-clipped=5.0 2023-03-09 02:34:00,250 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90078.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:34:05,191 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90081.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:34:21,545 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3649, 2.8630, 3.1365, 4.4797, 3.9331, 3.9347, 2.9121, 2.1049], device='cuda:3'), covar=tensor([0.0747, 0.1996, 0.1028, 0.0545, 0.0925, 0.0427, 0.1563, 0.2422], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0220, 0.0193, 0.0225, 0.0231, 0.0183, 0.0207, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:34:24,272 INFO [train2.py:809] (3/4) Epoch 23, batch 2450, loss[ctc_loss=0.07529, att_loss=0.2358, loss=0.2037, over 15959.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006098, over 41.00 utterances.], tot_loss[ctc_loss=0.07221, att_loss=0.2349, loss=0.2024, over 3268653.55 frames. utt_duration=1236 frames, utt_pad_proportion=0.05702, over 10590.58 utterances.], batch size: 41, lr: 4.67e-03, grad_scale: 16.0 2023-03-09 02:34:27,930 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90095.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 02:34:56,680 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-09 02:35:23,996 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90129.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:35:25,889 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90130.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:35:47,602 INFO [train2.py:809] (3/4) Epoch 23, batch 2500, loss[ctc_loss=0.07688, att_loss=0.2547, loss=0.2191, over 17084.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01728, over 56.00 utterances.], tot_loss[ctc_loss=0.07372, att_loss=0.2365, loss=0.2039, over 3258986.69 frames. utt_duration=1168 frames, utt_pad_proportion=0.07644, over 11173.12 utterances.], batch size: 56, lr: 4.66e-03, grad_scale: 16.0 2023-03-09 02:35:47,726 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90143.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 02:35:48,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 2.045e+02 2.395e+02 2.937e+02 6.523e+02, threshold=4.790e+02, percent-clipped=3.0 2023-03-09 02:36:11,424 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5499, 2.9966, 3.2943, 4.5886, 4.1708, 4.1554, 3.0985, 2.4510], device='cuda:3'), covar=tensor([0.0668, 0.2034, 0.1094, 0.0524, 0.0789, 0.0368, 0.1437, 0.2150], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0219, 0.0192, 0.0225, 0.0229, 0.0182, 0.0206, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:36:23,551 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90165.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:37:06,638 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90191.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:37:09,385 INFO [train2.py:809] (3/4) Epoch 23, batch 2550, loss[ctc_loss=0.08994, att_loss=0.2208, loss=0.1946, over 15514.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008051, over 36.00 utterances.], tot_loss[ctc_loss=0.07366, att_loss=0.236, loss=0.2035, over 3257504.68 frames. utt_duration=1172 frames, utt_pad_proportion=0.07687, over 11131.66 utterances.], batch size: 36, lr: 4.66e-03, grad_scale: 16.0 2023-03-09 02:37:41,489 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90213.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:37:50,440 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7168, 2.4892, 2.6348, 2.7305, 2.7571, 2.6323, 2.5753, 3.0546], device='cuda:3'), covar=tensor([0.1654, 0.2621, 0.1819, 0.1436, 0.1493, 0.1350, 0.2126, 0.1232], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0126, 0.0123, 0.0114, 0.0128, 0.0111, 0.0133, 0.0105], device='cuda:3'), out_proj_covar=tensor([9.3834e-05, 9.8281e-05, 9.8450e-05, 8.9151e-05, 9.5867e-05, 8.9373e-05, 1.0090e-04, 8.4091e-05], device='cuda:3') 2023-03-09 02:37:54,867 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90221.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 02:38:12,235 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90232.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:38:30,843 INFO [train2.py:809] (3/4) Epoch 23, batch 2600, loss[ctc_loss=0.06769, att_loss=0.2202, loss=0.1897, over 16123.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006218, over 42.00 utterances.], tot_loss[ctc_loss=0.07392, att_loss=0.2357, loss=0.2033, over 3261645.75 frames. utt_duration=1194 frames, utt_pad_proportion=0.0713, over 10938.43 utterances.], batch size: 42, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:38:34,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 2.020e+02 2.371e+02 2.963e+02 6.277e+02, threshold=4.742e+02, percent-clipped=3.0 2023-03-09 02:38:45,518 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0364, 3.8010, 3.8054, 3.3014, 3.7933, 3.8554, 3.8054, 2.8820], device='cuda:3'), covar=tensor([0.1075, 0.1064, 0.1989, 0.2804, 0.0971, 0.2009, 0.0831, 0.3236], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0192, 0.0205, 0.0260, 0.0165, 0.0267, 0.0190, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:39:34,332 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0351, 3.7625, 3.7794, 3.3203, 3.7441, 3.8363, 3.7557, 2.8215], device='cuda:3'), covar=tensor([0.1159, 0.1134, 0.1438, 0.2889, 0.1215, 0.2239, 0.0889, 0.3818], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0191, 0.0205, 0.0260, 0.0165, 0.0267, 0.0190, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:39:52,427 INFO [train2.py:809] (3/4) Epoch 23, batch 2650, loss[ctc_loss=0.075, att_loss=0.2504, loss=0.2153, over 16764.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.0061, over 48.00 utterances.], tot_loss[ctc_loss=0.07362, att_loss=0.2355, loss=0.2032, over 3257258.30 frames. utt_duration=1200 frames, utt_pad_proportion=0.06994, over 10874.87 utterances.], batch size: 48, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:39:52,887 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90293.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:41:03,316 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90336.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:41:14,424 INFO [train2.py:809] (3/4) Epoch 23, batch 2700, loss[ctc_loss=0.06712, att_loss=0.2464, loss=0.2105, over 17075.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007663, over 52.00 utterances.], tot_loss[ctc_loss=0.07317, att_loss=0.2351, loss=0.2027, over 3265012.09 frames. utt_duration=1219 frames, utt_pad_proportion=0.06355, over 10728.15 utterances.], batch size: 52, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:41:17,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 1.963e+02 2.237e+02 2.916e+02 6.361e+02, threshold=4.475e+02, percent-clipped=3.0 2023-03-09 02:41:27,032 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90351.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:42:21,367 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90384.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:42:35,956 INFO [train2.py:809] (3/4) Epoch 23, batch 2750, loss[ctc_loss=0.05895, att_loss=0.2106, loss=0.1803, over 15875.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009923, over 39.00 utterances.], tot_loss[ctc_loss=0.07264, att_loss=0.235, loss=0.2025, over 3273387.49 frames. utt_duration=1245 frames, utt_pad_proportion=0.05547, over 10533.30 utterances.], batch size: 39, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:43:06,572 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90412.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:43:10,131 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9087, 5.2475, 5.4746, 5.3191, 5.4323, 5.8911, 5.1980, 5.9922], device='cuda:3'), covar=tensor([0.0732, 0.0688, 0.0842, 0.1338, 0.1780, 0.0848, 0.0758, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0880, 0.0512, 0.0617, 0.0667, 0.0885, 0.0637, 0.0496, 0.0617], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 02:43:52,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1378, 5.4557, 5.0678, 5.5501, 4.9789, 5.1196, 5.5834, 5.3898], device='cuda:3'), covar=tensor([0.0522, 0.0265, 0.0723, 0.0288, 0.0348, 0.0216, 0.0211, 0.0193], device='cuda:3'), in_proj_covar=tensor([0.0395, 0.0330, 0.0371, 0.0359, 0.0327, 0.0242, 0.0313, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 02:43:58,167 INFO [train2.py:809] (3/4) Epoch 23, batch 2800, loss[ctc_loss=0.05602, att_loss=0.2136, loss=0.1821, over 15644.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008328, over 37.00 utterances.], tot_loss[ctc_loss=0.07233, att_loss=0.2349, loss=0.2024, over 3273125.34 frames. utt_duration=1245 frames, utt_pad_proportion=0.05576, over 10527.22 utterances.], batch size: 37, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:44:01,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.964e+02 2.366e+02 2.896e+02 7.050e+02, threshold=4.731e+02, percent-clipped=3.0 2023-03-09 02:44:23,935 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90459.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:44:44,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-03-09 02:44:50,025 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0991, 5.3874, 5.6716, 5.4655, 5.6215, 6.0587, 5.3338, 6.1046], device='cuda:3'), covar=tensor([0.0700, 0.0696, 0.0818, 0.1368, 0.1828, 0.0857, 0.0717, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0882, 0.0511, 0.0617, 0.0668, 0.0884, 0.0639, 0.0496, 0.0617], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 02:45:07,730 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90486.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:45:19,165 INFO [train2.py:809] (3/4) Epoch 23, batch 2850, loss[ctc_loss=0.07628, att_loss=0.2537, loss=0.2182, over 17311.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01142, over 55.00 utterances.], tot_loss[ctc_loss=0.0718, att_loss=0.2348, loss=0.2022, over 3281294.91 frames. utt_duration=1271 frames, utt_pad_proportion=0.04781, over 10337.89 utterances.], batch size: 55, lr: 4.66e-03, grad_scale: 8.0 2023-03-09 02:45:24,399 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90496.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:45:31,574 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-03-09 02:46:03,446 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90520.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:46:04,901 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90521.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 02:46:40,203 INFO [train2.py:809] (3/4) Epoch 23, batch 2900, loss[ctc_loss=0.08038, att_loss=0.2331, loss=0.2026, over 16259.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.008483, over 43.00 utterances.], tot_loss[ctc_loss=0.07169, att_loss=0.2344, loss=0.2019, over 3273809.17 frames. utt_duration=1265 frames, utt_pad_proportion=0.05191, over 10367.12 utterances.], batch size: 43, lr: 4.65e-03, grad_scale: 8.0 2023-03-09 02:46:43,384 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 2.000e+02 2.282e+02 2.820e+02 8.053e+02, threshold=4.565e+02, percent-clipped=3.0 2023-03-09 02:46:45,353 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0150, 4.9276, 4.7194, 2.9291, 4.7445, 4.5997, 4.2121, 2.7114], device='cuda:3'), covar=tensor([0.0137, 0.0107, 0.0295, 0.1021, 0.0108, 0.0229, 0.0329, 0.1346], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0103, 0.0105, 0.0111, 0.0086, 0.0115, 0.0099, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:47:02,778 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90557.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:47:22,355 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90569.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:47:31,873 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5297, 3.1949, 3.7398, 4.6166, 4.1496, 4.0982, 3.0412, 2.6570], device='cuda:3'), covar=tensor([0.0728, 0.1862, 0.0785, 0.0581, 0.0740, 0.0458, 0.1461, 0.2029], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0219, 0.0189, 0.0224, 0.0228, 0.0182, 0.0205, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:47:53,004 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90588.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:48:01,371 INFO [train2.py:809] (3/4) Epoch 23, batch 2950, loss[ctc_loss=0.0671, att_loss=0.2342, loss=0.2008, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005693, over 47.00 utterances.], tot_loss[ctc_loss=0.07191, att_loss=0.2343, loss=0.2018, over 3275086.75 frames. utt_duration=1270 frames, utt_pad_proportion=0.05078, over 10330.67 utterances.], batch size: 47, lr: 4.65e-03, grad_scale: 8.0 2023-03-09 02:48:07,155 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 02:48:21,550 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1906, 3.8437, 3.2419, 3.5264, 4.0369, 3.7595, 3.0395, 4.3777], device='cuda:3'), covar=tensor([0.0930, 0.0516, 0.1162, 0.0683, 0.0714, 0.0671, 0.0832, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0220, 0.0229, 0.0205, 0.0283, 0.0245, 0.0201, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 02:49:23,756 INFO [train2.py:809] (3/4) Epoch 23, batch 3000, loss[ctc_loss=0.08237, att_loss=0.2524, loss=0.2184, over 17048.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009905, over 53.00 utterances.], tot_loss[ctc_loss=0.07157, att_loss=0.2337, loss=0.2013, over 3261805.84 frames. utt_duration=1277 frames, utt_pad_proportion=0.05053, over 10230.65 utterances.], batch size: 53, lr: 4.65e-03, grad_scale: 8.0 2023-03-09 02:49:23,757 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 02:49:38,034 INFO [train2.py:843] (3/4) Epoch 23, validation: ctc_loss=0.03973, att_loss=0.234, loss=0.1952, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 02:49:38,035 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 02:49:41,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.311e+02 1.804e+02 2.183e+02 2.911e+02 8.359e+02, threshold=4.365e+02, percent-clipped=3.0 2023-03-09 02:51:00,735 INFO [train2.py:809] (3/4) Epoch 23, batch 3050, loss[ctc_loss=0.05829, att_loss=0.2341, loss=0.1989, over 17017.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007255, over 51.00 utterances.], tot_loss[ctc_loss=0.07092, att_loss=0.2338, loss=0.2012, over 3269494.85 frames. utt_duration=1274 frames, utt_pad_proportion=0.04859, over 10281.13 utterances.], batch size: 51, lr: 4.65e-03, grad_scale: 8.0 2023-03-09 02:51:24,182 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90707.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:52:23,332 INFO [train2.py:809] (3/4) Epoch 23, batch 3100, loss[ctc_loss=0.05111, att_loss=0.2067, loss=0.1756, over 15999.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008437, over 40.00 utterances.], tot_loss[ctc_loss=0.07059, att_loss=0.2338, loss=0.2012, over 3266485.70 frames. utt_duration=1258 frames, utt_pad_proportion=0.05394, over 10394.78 utterances.], batch size: 40, lr: 4.65e-03, grad_scale: 4.0 2023-03-09 02:52:28,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.776e+02 2.095e+02 2.519e+02 6.521e+02, threshold=4.191e+02, percent-clipped=1.0 2023-03-09 02:53:34,430 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90786.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:53:45,909 INFO [train2.py:809] (3/4) Epoch 23, batch 3150, loss[ctc_loss=0.1241, att_loss=0.2621, loss=0.2345, over 14291.00 frames. utt_duration=393.1 frames, utt_pad_proportion=0.3139, over 146.00 utterances.], tot_loss[ctc_loss=0.07088, att_loss=0.2345, loss=0.2018, over 3273110.12 frames. utt_duration=1248 frames, utt_pad_proportion=0.05506, over 10505.12 utterances.], batch size: 146, lr: 4.65e-03, grad_scale: 4.0 2023-03-09 02:54:17,559 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-09 02:54:21,923 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90815.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:54:30,671 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4401, 2.5570, 4.8188, 3.7124, 2.9906, 4.2334, 4.5646, 4.5018], device='cuda:3'), covar=tensor([0.0253, 0.1624, 0.0161, 0.0981, 0.1738, 0.0232, 0.0193, 0.0272], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0245, 0.0198, 0.0320, 0.0266, 0.0221, 0.0188, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:54:33,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9574, 2.6936, 2.8792, 3.7772, 3.4512, 3.5422, 2.6092, 2.0959], device='cuda:3'), covar=tensor([0.0873, 0.1895, 0.1073, 0.0742, 0.1066, 0.0527, 0.1627, 0.2313], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0218, 0.0189, 0.0222, 0.0229, 0.0182, 0.0204, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:54:52,451 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90834.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:54:57,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-03-09 02:55:08,168 INFO [train2.py:809] (3/4) Epoch 23, batch 3200, loss[ctc_loss=0.07892, att_loss=0.2494, loss=0.2153, over 17046.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009992, over 53.00 utterances.], tot_loss[ctc_loss=0.07133, att_loss=0.2346, loss=0.2019, over 3273122.24 frames. utt_duration=1243 frames, utt_pad_proportion=0.05731, over 10546.90 utterances.], batch size: 53, lr: 4.65e-03, grad_scale: 8.0 2023-03-09 02:55:12,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.356e+02 1.815e+02 2.300e+02 2.792e+02 4.062e+02, threshold=4.600e+02, percent-clipped=0.0 2023-03-09 02:55:18,219 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90849.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:55:22,816 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90852.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:55:52,371 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6955, 5.9336, 5.3555, 5.6771, 5.6050, 5.1255, 5.3971, 5.1258], device='cuda:3'), covar=tensor([0.1171, 0.0884, 0.1013, 0.0829, 0.0931, 0.1442, 0.2177, 0.2445], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0620, 0.0472, 0.0464, 0.0433, 0.0472, 0.0619, 0.0532], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:56:21,736 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:56:29,992 INFO [train2.py:809] (3/4) Epoch 23, batch 3250, loss[ctc_loss=0.07508, att_loss=0.2406, loss=0.2075, over 17346.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03585, over 63.00 utterances.], tot_loss[ctc_loss=0.07205, att_loss=0.2352, loss=0.2026, over 3277506.94 frames. utt_duration=1234 frames, utt_pad_proportion=0.05913, over 10633.43 utterances.], batch size: 63, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 02:56:57,381 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90910.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:57:39,814 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:57:39,863 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6238, 4.8593, 4.2930, 4.7094, 4.5276, 4.0108, 4.3992, 4.1213], device='cuda:3'), covar=tensor([0.1493, 0.1319, 0.1188, 0.1141, 0.1232, 0.1889, 0.2351, 0.2746], device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0621, 0.0474, 0.0466, 0.0435, 0.0474, 0.0623, 0.0535], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 02:57:51,514 INFO [train2.py:809] (3/4) Epoch 23, batch 3300, loss[ctc_loss=0.04912, att_loss=0.2037, loss=0.1728, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.008744, over 39.00 utterances.], tot_loss[ctc_loss=0.07175, att_loss=0.2349, loss=0.2023, over 3279188.29 frames. utt_duration=1237 frames, utt_pad_proportion=0.05788, over 10612.65 utterances.], batch size: 39, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 02:57:56,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 1.978e+02 2.311e+02 2.914e+02 8.897e+02, threshold=4.622e+02, percent-clipped=4.0 2023-03-09 02:58:06,122 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4435, 4.5290, 4.6135, 4.6200, 5.1324, 4.6859, 4.5465, 2.5595], device='cuda:3'), covar=tensor([0.0260, 0.0345, 0.0318, 0.0322, 0.0848, 0.0213, 0.0357, 0.1716], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0198, 0.0195, 0.0212, 0.0370, 0.0165, 0.0186, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 02:59:13,554 INFO [train2.py:809] (3/4) Epoch 23, batch 3350, loss[ctc_loss=0.07578, att_loss=0.2181, loss=0.1896, over 16020.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006997, over 40.00 utterances.], tot_loss[ctc_loss=0.07112, att_loss=0.2349, loss=0.2022, over 3280898.70 frames. utt_duration=1243 frames, utt_pad_proportion=0.05649, over 10569.90 utterances.], batch size: 40, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 02:59:37,350 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91007.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 02:59:47,459 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5626, 4.9089, 5.1618, 4.9634, 5.0811, 5.5076, 5.0465, 5.5679], device='cuda:3'), covar=tensor([0.0791, 0.0741, 0.0874, 0.1412, 0.1905, 0.0931, 0.0955, 0.0697], device='cuda:3'), in_proj_covar=tensor([0.0882, 0.0515, 0.0620, 0.0671, 0.0891, 0.0642, 0.0500, 0.0622], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:00:35,855 INFO [train2.py:809] (3/4) Epoch 23, batch 3400, loss[ctc_loss=0.04801, att_loss=0.216, loss=0.1824, over 16389.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007727, over 44.00 utterances.], tot_loss[ctc_loss=0.071, att_loss=0.2355, loss=0.2026, over 3288234.36 frames. utt_duration=1243 frames, utt_pad_proportion=0.05375, over 10593.29 utterances.], batch size: 44, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 03:00:40,394 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.937e+02 2.293e+02 2.658e+02 6.821e+02, threshold=4.585e+02, percent-clipped=3.0 2023-03-09 03:00:55,233 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91055.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:01:57,224 INFO [train2.py:809] (3/4) Epoch 23, batch 3450, loss[ctc_loss=0.08463, att_loss=0.2506, loss=0.2174, over 16671.00 frames. utt_duration=675.1 frames, utt_pad_proportion=0.153, over 99.00 utterances.], tot_loss[ctc_loss=0.07103, att_loss=0.2356, loss=0.2027, over 3288777.03 frames. utt_duration=1243 frames, utt_pad_proportion=0.0528, over 10597.57 utterances.], batch size: 99, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 03:02:33,682 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91115.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:02:38,333 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6109, 3.3992, 3.7309, 4.6863, 4.0811, 4.1223, 3.0800, 2.5518], device='cuda:3'), covar=tensor([0.0601, 0.1664, 0.0792, 0.0486, 0.0795, 0.0389, 0.1397, 0.1978], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0214, 0.0187, 0.0219, 0.0227, 0.0180, 0.0201, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:03:19,284 INFO [train2.py:809] (3/4) Epoch 23, batch 3500, loss[ctc_loss=0.06975, att_loss=0.233, loss=0.2003, over 16410.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006396, over 44.00 utterances.], tot_loss[ctc_loss=0.07099, att_loss=0.2355, loss=0.2026, over 3293039.48 frames. utt_duration=1250 frames, utt_pad_proportion=0.0491, over 10553.82 utterances.], batch size: 44, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 03:03:23,955 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 1.873e+02 2.204e+02 2.861e+02 4.686e+02, threshold=4.409e+02, percent-clipped=2.0 2023-03-09 03:03:33,673 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91152.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:03:51,254 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91163.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:03:53,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1105, 5.0356, 4.8354, 2.9366, 4.8642, 4.7000, 4.3791, 2.6650], device='cuda:3'), covar=tensor([0.0109, 0.0089, 0.0257, 0.1055, 0.0098, 0.0201, 0.0302, 0.1409], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0103, 0.0105, 0.0112, 0.0086, 0.0116, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 03:04:40,871 INFO [train2.py:809] (3/4) Epoch 23, batch 3550, loss[ctc_loss=0.07372, att_loss=0.2315, loss=0.2, over 16616.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.005977, over 47.00 utterances.], tot_loss[ctc_loss=0.0712, att_loss=0.2353, loss=0.2025, over 3292188.53 frames. utt_duration=1234 frames, utt_pad_proportion=0.05303, over 10681.89 utterances.], batch size: 47, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 03:04:52,819 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91200.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:05:01,553 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91205.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:05:06,630 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2730, 4.1682, 4.2163, 4.2037, 4.7263, 4.3803, 4.0930, 2.5114], device='cuda:3'), covar=tensor([0.0288, 0.0507, 0.0441, 0.0390, 0.0663, 0.0242, 0.0488, 0.1729], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0198, 0.0197, 0.0212, 0.0371, 0.0166, 0.0187, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:06:04,017 INFO [train2.py:809] (3/4) Epoch 23, batch 3600, loss[ctc_loss=0.06477, att_loss=0.2159, loss=0.1857, over 16175.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006535, over 41.00 utterances.], tot_loss[ctc_loss=0.07109, att_loss=0.2351, loss=0.2023, over 3297199.71 frames. utt_duration=1260 frames, utt_pad_proportion=0.04568, over 10476.75 utterances.], batch size: 41, lr: 4.64e-03, grad_scale: 8.0 2023-03-09 03:06:08,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 1.921e+02 2.289e+02 2.747e+02 1.018e+03, threshold=4.578e+02, percent-clipped=6.0 2023-03-09 03:07:14,033 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91285.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:07:26,450 INFO [train2.py:809] (3/4) Epoch 23, batch 3650, loss[ctc_loss=0.07499, att_loss=0.2501, loss=0.2151, over 16962.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007251, over 50.00 utterances.], tot_loss[ctc_loss=0.07054, att_loss=0.2347, loss=0.2018, over 3290431.32 frames. utt_duration=1289 frames, utt_pad_proportion=0.04017, over 10224.65 utterances.], batch size: 50, lr: 4.63e-03, grad_scale: 8.0 2023-03-09 03:08:21,002 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7551, 2.2227, 2.2181, 2.5766, 2.7977, 2.4703, 2.3622, 2.8657], device='cuda:3'), covar=tensor([0.1359, 0.2664, 0.1918, 0.1321, 0.1456, 0.1918, 0.2317, 0.1247], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0132, 0.0130, 0.0121, 0.0136, 0.0117, 0.0143, 0.0112], device='cuda:3'), out_proj_covar=tensor([9.8780e-05, 1.0358e-04, 1.0445e-04, 9.4467e-05, 1.0193e-04, 9.4033e-05, 1.0805e-04, 8.9087e-05], device='cuda:3') 2023-03-09 03:08:42,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.79 vs. limit=5.0 2023-03-09 03:08:47,986 INFO [train2.py:809] (3/4) Epoch 23, batch 3700, loss[ctc_loss=0.06188, att_loss=0.2296, loss=0.196, over 16406.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007329, over 44.00 utterances.], tot_loss[ctc_loss=0.06967, att_loss=0.2338, loss=0.201, over 3281888.73 frames. utt_duration=1313 frames, utt_pad_proportion=0.03629, over 10006.66 utterances.], batch size: 44, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:08:53,197 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91346.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:08:54,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.187e+02 1.976e+02 2.254e+02 2.803e+02 7.124e+02, threshold=4.508e+02, percent-clipped=5.0 2023-03-09 03:10:10,216 INFO [train2.py:809] (3/4) Epoch 23, batch 3750, loss[ctc_loss=0.06365, att_loss=0.2261, loss=0.1936, over 16002.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.007285, over 40.00 utterances.], tot_loss[ctc_loss=0.07039, att_loss=0.2336, loss=0.201, over 3271969.22 frames. utt_duration=1285 frames, utt_pad_proportion=0.04521, over 10194.75 utterances.], batch size: 40, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:11:32,630 INFO [train2.py:809] (3/4) Epoch 23, batch 3800, loss[ctc_loss=0.0568, att_loss=0.2367, loss=0.2007, over 16116.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006957, over 42.00 utterances.], tot_loss[ctc_loss=0.06965, att_loss=0.2335, loss=0.2007, over 3273747.96 frames. utt_duration=1295 frames, utt_pad_proportion=0.0417, over 10120.48 utterances.], batch size: 42, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:11:38,946 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 1.926e+02 2.458e+02 3.092e+02 6.476e+02, threshold=4.915e+02, percent-clipped=5.0 2023-03-09 03:11:49,459 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9103, 4.9526, 4.5325, 2.9326, 4.7301, 4.5346, 4.1462, 2.1312], device='cuda:3'), covar=tensor([0.0172, 0.0117, 0.0367, 0.1204, 0.0112, 0.0284, 0.0390, 0.2009], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0103, 0.0106, 0.0112, 0.0086, 0.0116, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 03:12:41,418 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1834, 4.4815, 4.6363, 4.8099, 3.3269, 4.5566, 2.9857, 1.9717], device='cuda:3'), covar=tensor([0.0417, 0.0259, 0.0532, 0.0194, 0.1177, 0.0205, 0.1268, 0.1568], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0171, 0.0262, 0.0164, 0.0225, 0.0156, 0.0233, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:12:55,886 INFO [train2.py:809] (3/4) Epoch 23, batch 3850, loss[ctc_loss=0.05979, att_loss=0.2361, loss=0.2008, over 16487.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005393, over 46.00 utterances.], tot_loss[ctc_loss=0.06938, att_loss=0.234, loss=0.201, over 3283113.05 frames. utt_duration=1308 frames, utt_pad_proportion=0.03641, over 10051.04 utterances.], batch size: 46, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:13:15,036 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91505.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:14:12,263 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91542.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 03:14:13,401 INFO [train2.py:809] (3/4) Epoch 23, batch 3900, loss[ctc_loss=0.08029, att_loss=0.2369, loss=0.2056, over 17344.00 frames. utt_duration=1007 frames, utt_pad_proportion=0.04921, over 69.00 utterances.], tot_loss[ctc_loss=0.07016, att_loss=0.2346, loss=0.2017, over 3271801.25 frames. utt_duration=1261 frames, utt_pad_proportion=0.05141, over 10387.15 utterances.], batch size: 69, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:14:19,548 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.985e+02 2.304e+02 2.861e+02 4.797e+02, threshold=4.607e+02, percent-clipped=0.0 2023-03-09 03:14:29,571 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91553.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:14:29,841 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7858, 3.6444, 3.0636, 3.3420, 3.8027, 3.5420, 2.7121, 3.9541], device='cuda:3'), covar=tensor([0.1095, 0.0444, 0.1075, 0.0712, 0.0748, 0.0702, 0.0946, 0.0581], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0219, 0.0225, 0.0202, 0.0278, 0.0242, 0.0199, 0.0289], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 03:14:36,164 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9632, 2.2590, 2.2829, 2.5861, 2.6721, 2.3790, 2.4952, 2.8805], device='cuda:3'), covar=tensor([0.1332, 0.2292, 0.1687, 0.1369, 0.1469, 0.1077, 0.1755, 0.1172], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0129, 0.0127, 0.0118, 0.0134, 0.0115, 0.0139, 0.0110], device='cuda:3'), out_proj_covar=tensor([9.6828e-05, 1.0143e-04, 1.0246e-04, 9.2535e-05, 1.0010e-04, 9.2275e-05, 1.0555e-04, 8.7783e-05], device='cuda:3') 2023-03-09 03:14:48,484 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2157, 4.5273, 4.7497, 4.8100, 2.8765, 4.5844, 3.0979, 1.8859], device='cuda:3'), covar=tensor([0.0416, 0.0304, 0.0549, 0.0226, 0.1603, 0.0232, 0.1265, 0.1722], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0172, 0.0264, 0.0165, 0.0225, 0.0157, 0.0234, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:15:32,015 INFO [train2.py:809] (3/4) Epoch 23, batch 3950, loss[ctc_loss=0.05667, att_loss=0.2194, loss=0.1868, over 15977.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005581, over 41.00 utterances.], tot_loss[ctc_loss=0.07128, att_loss=0.2354, loss=0.2026, over 3274008.73 frames. utt_duration=1230 frames, utt_pad_proportion=0.05901, over 10656.26 utterances.], batch size: 41, lr: 4.63e-03, grad_scale: 4.0 2023-03-09 03:15:48,497 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91603.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 03:16:00,644 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9482, 6.1474, 5.6343, 5.9189, 5.8208, 5.3079, 5.6087, 5.3908], device='cuda:3'), covar=tensor([0.1184, 0.0884, 0.0860, 0.0817, 0.0857, 0.1499, 0.2099, 0.2147], device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0615, 0.0467, 0.0461, 0.0433, 0.0474, 0.0616, 0.0533], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 03:16:52,903 INFO [train2.py:809] (3/4) Epoch 24, batch 0, loss[ctc_loss=0.07683, att_loss=0.2184, loss=0.1901, over 15373.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01101, over 35.00 utterances.], tot_loss[ctc_loss=0.07683, att_loss=0.2184, loss=0.1901, over 15373.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01101, over 35.00 utterances.], batch size: 35, lr: 4.53e-03, grad_scale: 8.0 2023-03-09 03:16:52,903 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 03:17:00,627 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5700, 3.8611, 3.9285, 2.0206, 1.9693, 2.6325, 2.1192, 3.4266], device='cuda:3'), covar=tensor([0.0775, 0.0478, 0.0429, 0.4756, 0.5108, 0.2505, 0.3487, 0.1402], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0282, 0.0271, 0.0245, 0.0341, 0.0334, 0.0257, 0.0369], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 03:17:05,993 INFO [train2.py:843] (3/4) Epoch 24, validation: ctc_loss=0.04095, att_loss=0.2349, loss=0.1961, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 03:17:05,994 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 03:17:15,560 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2023-03-09 03:17:17,335 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 03:17:28,523 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91641.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:17:37,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.024e+02 2.562e+02 3.130e+02 9.930e+02, threshold=5.124e+02, percent-clipped=6.0 2023-03-09 03:18:27,182 INFO [train2.py:809] (3/4) Epoch 24, batch 50, loss[ctc_loss=0.05749, att_loss=0.2315, loss=0.1967, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006242, over 45.00 utterances.], tot_loss[ctc_loss=0.07201, att_loss=0.2381, loss=0.2048, over 745063.88 frames. utt_duration=1120 frames, utt_pad_proportion=0.07782, over 2664.19 utterances.], batch size: 45, lr: 4.53e-03, grad_scale: 8.0 2023-03-09 03:19:47,844 INFO [train2.py:809] (3/4) Epoch 24, batch 100, loss[ctc_loss=0.0587, att_loss=0.2209, loss=0.1885, over 16126.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006264, over 42.00 utterances.], tot_loss[ctc_loss=0.07191, att_loss=0.2362, loss=0.2033, over 1308900.88 frames. utt_duration=1196 frames, utt_pad_proportion=0.06312, over 4384.15 utterances.], batch size: 42, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:20:20,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.937e+02 2.364e+02 2.777e+02 5.363e+02, threshold=4.728e+02, percent-clipped=1.0 2023-03-09 03:21:08,900 INFO [train2.py:809] (3/4) Epoch 24, batch 150, loss[ctc_loss=0.06554, att_loss=0.2381, loss=0.2036, over 17293.00 frames. utt_duration=1099 frames, utt_pad_proportion=0.03819, over 63.00 utterances.], tot_loss[ctc_loss=0.07062, att_loss=0.2349, loss=0.2021, over 1747374.87 frames. utt_duration=1186 frames, utt_pad_proportion=0.06463, over 5901.26 utterances.], batch size: 63, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:22:06,146 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1305, 5.4818, 5.0254, 5.5208, 4.8650, 5.2075, 5.6075, 5.3723], device='cuda:3'), covar=tensor([0.0569, 0.0287, 0.0812, 0.0306, 0.0455, 0.0198, 0.0261, 0.0212], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0325, 0.0367, 0.0358, 0.0329, 0.0239, 0.0307, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 03:22:07,976 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5050, 2.9092, 4.9209, 3.9731, 2.9607, 4.2779, 4.8221, 4.5887], device='cuda:3'), covar=tensor([0.0298, 0.1437, 0.0216, 0.0921, 0.1816, 0.0250, 0.0173, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0245, 0.0200, 0.0323, 0.0268, 0.0224, 0.0191, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:22:30,994 INFO [train2.py:809] (3/4) Epoch 24, batch 200, loss[ctc_loss=0.05662, att_loss=0.2115, loss=0.1806, over 15997.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008086, over 40.00 utterances.], tot_loss[ctc_loss=0.07084, att_loss=0.2345, loss=0.2018, over 2077609.29 frames. utt_duration=1195 frames, utt_pad_proportion=0.06689, over 6961.44 utterances.], batch size: 40, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:23:02,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 1.924e+02 2.312e+02 2.679e+02 4.327e+02, threshold=4.624e+02, percent-clipped=0.0 2023-03-09 03:23:10,174 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9300, 2.4017, 2.9278, 2.7377, 2.8866, 2.5491, 2.6806, 3.0558], device='cuda:3'), covar=tensor([0.1812, 0.2847, 0.1708, 0.1479, 0.2524, 0.1458, 0.1914, 0.1312], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0130, 0.0127, 0.0118, 0.0134, 0.0115, 0.0140, 0.0111], device='cuda:3'), out_proj_covar=tensor([9.7169e-05, 1.0170e-04, 1.0225e-04, 9.2438e-05, 1.0050e-04, 9.2711e-05, 1.0578e-04, 8.8324e-05], device='cuda:3') 2023-03-09 03:23:42,028 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0038, 4.8812, 5.0165, 4.9182, 5.5453, 5.0075, 4.8210, 2.4771], device='cuda:3'), covar=tensor([0.0128, 0.0204, 0.0167, 0.0222, 0.0550, 0.0125, 0.0207, 0.1718], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0199, 0.0199, 0.0215, 0.0374, 0.0168, 0.0187, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:23:51,117 INFO [train2.py:809] (3/4) Epoch 24, batch 250, loss[ctc_loss=0.03785, att_loss=0.2036, loss=0.1705, over 15358.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01138, over 35.00 utterances.], tot_loss[ctc_loss=0.07097, att_loss=0.2341, loss=0.2015, over 2329644.41 frames. utt_duration=1187 frames, utt_pad_proportion=0.07366, over 7861.96 utterances.], batch size: 35, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:24:14,491 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91891.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:24:25,480 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91898.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 03:25:11,836 INFO [train2.py:809] (3/4) Epoch 24, batch 300, loss[ctc_loss=0.06263, att_loss=0.2285, loss=0.1953, over 16169.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007604, over 41.00 utterances.], tot_loss[ctc_loss=0.07121, att_loss=0.2342, loss=0.2016, over 2543747.93 frames. utt_duration=1189 frames, utt_pad_proportion=0.06956, over 8566.14 utterances.], batch size: 41, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:25:35,412 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91941.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:25:44,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.878e+02 2.164e+02 2.654e+02 4.888e+02, threshold=4.328e+02, percent-clipped=1.0 2023-03-09 03:25:49,850 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9555, 6.1928, 5.7282, 5.9647, 5.9166, 5.4069, 5.5914, 5.3799], device='cuda:3'), covar=tensor([0.1203, 0.0831, 0.0940, 0.0787, 0.1009, 0.1586, 0.2461, 0.2273], device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0623, 0.0473, 0.0466, 0.0440, 0.0475, 0.0624, 0.0539], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 03:25:53,281 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91952.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:25:57,825 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91955.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:26:04,064 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0525, 3.7237, 3.2121, 3.2874, 3.9582, 3.6317, 2.9656, 4.3023], device='cuda:3'), covar=tensor([0.1022, 0.0492, 0.1094, 0.0812, 0.0731, 0.0709, 0.0895, 0.0385], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0221, 0.0226, 0.0204, 0.0281, 0.0244, 0.0200, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 03:26:32,684 INFO [train2.py:809] (3/4) Epoch 24, batch 350, loss[ctc_loss=0.06187, att_loss=0.2163, loss=0.1854, over 14526.00 frames. utt_duration=1817 frames, utt_pad_proportion=0.03545, over 32.00 utterances.], tot_loss[ctc_loss=0.07119, att_loss=0.2346, loss=0.2019, over 2704055.66 frames. utt_duration=1194 frames, utt_pad_proportion=0.07006, over 9073.01 utterances.], batch size: 32, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:26:39,991 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91981.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 03:26:52,907 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91989.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:27:32,661 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 2023-03-09 03:27:40,696 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92016.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:27:58,109 INFO [train2.py:809] (3/4) Epoch 24, batch 400, loss[ctc_loss=0.1211, att_loss=0.2608, loss=0.2328, over 14290.00 frames. utt_duration=390.5 frames, utt_pad_proportion=0.3162, over 147.00 utterances.], tot_loss[ctc_loss=0.07063, att_loss=0.2347, loss=0.2019, over 2835809.68 frames. utt_duration=1194 frames, utt_pad_proportion=0.06776, over 9510.56 utterances.], batch size: 147, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:28:00,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 03:28:22,898 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92042.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 03:28:30,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.943e+02 2.339e+02 2.970e+02 6.262e+02, threshold=4.678e+02, percent-clipped=8.0 2023-03-09 03:29:18,995 INFO [train2.py:809] (3/4) Epoch 24, batch 450, loss[ctc_loss=0.08749, att_loss=0.2408, loss=0.2101, over 17018.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.0115, over 53.00 utterances.], tot_loss[ctc_loss=0.07005, att_loss=0.234, loss=0.2012, over 2931456.95 frames. utt_duration=1216 frames, utt_pad_proportion=0.06195, over 9652.54 utterances.], batch size: 53, lr: 4.52e-03, grad_scale: 8.0 2023-03-09 03:30:00,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0418, 5.2533, 5.6198, 5.4210, 5.5081, 5.9715, 5.2319, 6.0397], device='cuda:3'), covar=tensor([0.0634, 0.0750, 0.0819, 0.1274, 0.1640, 0.0905, 0.0788, 0.0683], device='cuda:3'), in_proj_covar=tensor([0.0871, 0.0508, 0.0611, 0.0662, 0.0874, 0.0632, 0.0494, 0.0615], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:30:27,517 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-03-09 03:30:40,855 INFO [train2.py:809] (3/4) Epoch 24, batch 500, loss[ctc_loss=0.08061, att_loss=0.2425, loss=0.2101, over 17145.00 frames. utt_duration=694.1 frames, utt_pad_proportion=0.1302, over 99.00 utterances.], tot_loss[ctc_loss=0.07, att_loss=0.2341, loss=0.2013, over 3008233.64 frames. utt_duration=1222 frames, utt_pad_proportion=0.06053, over 9862.61 utterances.], batch size: 99, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:31:13,732 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 1.745e+02 2.110e+02 2.591e+02 7.932e+02, threshold=4.219e+02, percent-clipped=4.0 2023-03-09 03:31:18,993 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3480, 2.5069, 4.7799, 3.8149, 2.9608, 4.1513, 4.4830, 4.5055], device='cuda:3'), covar=tensor([0.0286, 0.1598, 0.0184, 0.0938, 0.1701, 0.0277, 0.0210, 0.0286], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0241, 0.0198, 0.0317, 0.0263, 0.0220, 0.0189, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:32:01,719 INFO [train2.py:809] (3/4) Epoch 24, batch 550, loss[ctc_loss=0.08397, att_loss=0.2544, loss=0.2203, over 17333.00 frames. utt_duration=1262 frames, utt_pad_proportion=0.01015, over 55.00 utterances.], tot_loss[ctc_loss=0.07058, att_loss=0.234, loss=0.2013, over 3063565.18 frames. utt_duration=1217 frames, utt_pad_proportion=0.06287, over 10077.98 utterances.], batch size: 55, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:32:18,886 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-09 03:32:35,879 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92198.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 03:33:22,317 INFO [train2.py:809] (3/4) Epoch 24, batch 600, loss[ctc_loss=0.06587, att_loss=0.2102, loss=0.1814, over 15627.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009716, over 37.00 utterances.], tot_loss[ctc_loss=0.07109, att_loss=0.2344, loss=0.2017, over 3111926.78 frames. utt_duration=1206 frames, utt_pad_proportion=0.06459, over 10335.02 utterances.], batch size: 37, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:33:53,648 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92246.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 03:33:54,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.830e+02 2.241e+02 2.667e+02 5.450e+02, threshold=4.482e+02, percent-clipped=6.0 2023-03-09 03:33:55,216 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92247.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:34:38,850 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-09 03:34:42,404 INFO [train2.py:809] (3/4) Epoch 24, batch 650, loss[ctc_loss=0.06453, att_loss=0.2339, loss=0.2001, over 16970.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007305, over 50.00 utterances.], tot_loss[ctc_loss=0.07076, att_loss=0.2343, loss=0.2016, over 3142078.82 frames. utt_duration=1227 frames, utt_pad_proportion=0.06192, over 10256.31 utterances.], batch size: 50, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:35:24,931 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3189, 4.4675, 4.4038, 4.4450, 4.9295, 4.3950, 4.3787, 2.3620], device='cuda:3'), covar=tensor([0.0306, 0.0356, 0.0458, 0.0326, 0.0871, 0.0286, 0.0404, 0.2076], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0202, 0.0202, 0.0218, 0.0380, 0.0170, 0.0190, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:35:37,896 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92311.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:36:03,498 INFO [train2.py:809] (3/4) Epoch 24, batch 700, loss[ctc_loss=0.06809, att_loss=0.2505, loss=0.214, over 17068.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.00732, over 52.00 utterances.], tot_loss[ctc_loss=0.07096, att_loss=0.235, loss=0.2022, over 3169228.90 frames. utt_duration=1220 frames, utt_pad_proportion=0.06396, over 10402.56 utterances.], batch size: 52, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:36:20,895 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92337.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 03:36:20,922 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9015, 5.1621, 5.1065, 5.1225, 5.2043, 5.1550, 4.8493, 4.6358], device='cuda:3'), covar=tensor([0.0972, 0.0555, 0.0329, 0.0528, 0.0291, 0.0368, 0.0368, 0.0347], device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0365, 0.0353, 0.0361, 0.0430, 0.0437, 0.0363, 0.0401], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 03:36:36,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 1.935e+02 2.344e+02 2.941e+02 5.112e+02, threshold=4.689e+02, percent-clipped=2.0 2023-03-09 03:37:25,726 INFO [train2.py:809] (3/4) Epoch 24, batch 750, loss[ctc_loss=0.07283, att_loss=0.2461, loss=0.2114, over 16885.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.00718, over 49.00 utterances.], tot_loss[ctc_loss=0.07068, att_loss=0.2348, loss=0.202, over 3193027.81 frames. utt_duration=1217 frames, utt_pad_proportion=0.06347, over 10511.77 utterances.], batch size: 49, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:38:21,654 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-09 03:38:46,216 INFO [train2.py:809] (3/4) Epoch 24, batch 800, loss[ctc_loss=0.06473, att_loss=0.213, loss=0.1834, over 15888.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.009113, over 39.00 utterances.], tot_loss[ctc_loss=0.07011, att_loss=0.2338, loss=0.2011, over 3205838.49 frames. utt_duration=1226 frames, utt_pad_proportion=0.06299, over 10472.24 utterances.], batch size: 39, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:39:05,219 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8997, 5.3517, 5.1536, 5.2050, 5.3026, 4.9964, 3.7997, 5.2486], device='cuda:3'), covar=tensor([0.0099, 0.0090, 0.0107, 0.0075, 0.0097, 0.0114, 0.0599, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0089, 0.0112, 0.0071, 0.0076, 0.0087, 0.0103, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:39:14,958 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6053, 4.8639, 4.4205, 4.9263, 4.3127, 4.6561, 4.9626, 4.7763], device='cuda:3'), covar=tensor([0.0607, 0.0343, 0.0905, 0.0397, 0.0501, 0.0363, 0.0275, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0324, 0.0367, 0.0357, 0.0329, 0.0239, 0.0308, 0.0287], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 03:39:19,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 1.923e+02 2.416e+02 3.239e+02 1.370e+03, threshold=4.832e+02, percent-clipped=5.0 2023-03-09 03:40:07,916 INFO [train2.py:809] (3/4) Epoch 24, batch 850, loss[ctc_loss=0.05174, att_loss=0.212, loss=0.1799, over 14531.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.03975, over 32.00 utterances.], tot_loss[ctc_loss=0.07035, att_loss=0.2335, loss=0.2009, over 3214323.59 frames. utt_duration=1222 frames, utt_pad_proportion=0.06485, over 10533.98 utterances.], batch size: 32, lr: 4.51e-03, grad_scale: 8.0 2023-03-09 03:41:29,527 INFO [train2.py:809] (3/4) Epoch 24, batch 900, loss[ctc_loss=0.08448, att_loss=0.252, loss=0.2185, over 17082.00 frames. utt_duration=1221 frames, utt_pad_proportion=0.01576, over 56.00 utterances.], tot_loss[ctc_loss=0.07021, att_loss=0.2336, loss=0.2009, over 3223611.24 frames. utt_duration=1207 frames, utt_pad_proportion=0.06947, over 10693.60 utterances.], batch size: 56, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:42:02,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.834e+02 2.155e+02 2.748e+02 5.069e+02, threshold=4.309e+02, percent-clipped=1.0 2023-03-09 03:42:02,750 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92547.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:42:29,834 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-09 03:42:50,355 INFO [train2.py:809] (3/4) Epoch 24, batch 950, loss[ctc_loss=0.06878, att_loss=0.2363, loss=0.2028, over 16536.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005706, over 45.00 utterances.], tot_loss[ctc_loss=0.07031, att_loss=0.2341, loss=0.2013, over 3238009.65 frames. utt_duration=1203 frames, utt_pad_proportion=0.06824, over 10781.03 utterances.], batch size: 45, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:42:50,620 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1819, 5.4220, 5.3812, 5.3286, 5.4823, 5.4417, 5.0775, 4.9195], device='cuda:3'), covar=tensor([0.0998, 0.0466, 0.0299, 0.0495, 0.0268, 0.0322, 0.0369, 0.0322], device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0366, 0.0356, 0.0365, 0.0430, 0.0438, 0.0363, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 03:43:20,172 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92595.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:43:46,297 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92611.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:44:08,909 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92625.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:44:11,647 INFO [train2.py:809] (3/4) Epoch 24, batch 1000, loss[ctc_loss=0.0713, att_loss=0.2432, loss=0.2088, over 16458.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007186, over 46.00 utterances.], tot_loss[ctc_loss=0.06993, att_loss=0.2333, loss=0.2006, over 3239251.97 frames. utt_duration=1235 frames, utt_pad_proportion=0.06193, over 10506.59 utterances.], batch size: 46, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:44:28,705 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92637.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 03:44:44,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 1.788e+02 2.232e+02 2.616e+02 5.460e+02, threshold=4.464e+02, percent-clipped=2.0 2023-03-09 03:45:04,282 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92659.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:45:09,420 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2635, 4.2974, 4.3658, 4.3304, 4.8557, 4.2984, 4.2673, 2.4316], device='cuda:3'), covar=tensor([0.0302, 0.0401, 0.0348, 0.0303, 0.0775, 0.0284, 0.0380, 0.1728], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0201, 0.0200, 0.0217, 0.0377, 0.0170, 0.0189, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:45:33,246 INFO [train2.py:809] (3/4) Epoch 24, batch 1050, loss[ctc_loss=0.07162, att_loss=0.2553, loss=0.2186, over 17303.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01181, over 55.00 utterances.], tot_loss[ctc_loss=0.07033, att_loss=0.2342, loss=0.2015, over 3256270.90 frames. utt_duration=1227 frames, utt_pad_proportion=0.05987, over 10627.07 utterances.], batch size: 55, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:45:47,605 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92685.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 03:45:49,250 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92686.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:46:27,011 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0613, 4.4164, 4.2754, 4.3611, 4.4648, 4.1794, 3.0904, 4.3291], device='cuda:3'), covar=tensor([0.0154, 0.0133, 0.0168, 0.0112, 0.0098, 0.0149, 0.0746, 0.0228], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0089, 0.0112, 0.0071, 0.0076, 0.0087, 0.0103, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:46:43,773 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92720.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:46:54,333 INFO [train2.py:809] (3/4) Epoch 24, batch 1100, loss[ctc_loss=0.05424, att_loss=0.23, loss=0.1949, over 16782.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005756, over 48.00 utterances.], tot_loss[ctc_loss=0.0696, att_loss=0.2337, loss=0.2009, over 3265353.22 frames. utt_duration=1232 frames, utt_pad_proportion=0.05571, over 10611.40 utterances.], batch size: 48, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:47:27,345 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.412e+02 1.778e+02 2.254e+02 2.467e+02 8.926e+02, threshold=4.507e+02, percent-clipped=3.0 2023-03-09 03:48:16,145 INFO [train2.py:809] (3/4) Epoch 24, batch 1150, loss[ctc_loss=0.07657, att_loss=0.2511, loss=0.2162, over 17338.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03561, over 63.00 utterances.], tot_loss[ctc_loss=0.06901, att_loss=0.2337, loss=0.2008, over 3271425.00 frames. utt_duration=1237 frames, utt_pad_proportion=0.05386, over 10587.90 utterances.], batch size: 63, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:48:23,643 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92781.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:48:25,257 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8087, 4.9693, 4.9456, 4.5189, 5.4076, 4.7975, 4.8204, 3.3811], device='cuda:3'), covar=tensor([0.0219, 0.0222, 0.0227, 0.0410, 0.0751, 0.0212, 0.0275, 0.1312], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0202, 0.0201, 0.0217, 0.0377, 0.0170, 0.0190, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:49:37,245 INFO [train2.py:809] (3/4) Epoch 24, batch 1200, loss[ctc_loss=0.09033, att_loss=0.2508, loss=0.2187, over 17037.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.0106, over 53.00 utterances.], tot_loss[ctc_loss=0.06924, att_loss=0.2334, loss=0.2006, over 3264969.22 frames. utt_duration=1244 frames, utt_pad_proportion=0.05443, over 10509.51 utterances.], batch size: 53, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:49:45,315 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92831.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:50:10,381 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 1.947e+02 2.306e+02 3.014e+02 1.019e+03, threshold=4.613e+02, percent-clipped=6.0 2023-03-09 03:50:27,965 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-09 03:50:59,081 INFO [train2.py:809] (3/4) Epoch 24, batch 1250, loss[ctc_loss=0.09164, att_loss=0.2517, loss=0.2197, over 17120.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01488, over 56.00 utterances.], tot_loss[ctc_loss=0.06896, att_loss=0.2334, loss=0.2005, over 3268022.18 frames. utt_duration=1252 frames, utt_pad_proportion=0.05205, over 10455.99 utterances.], batch size: 56, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:51:24,816 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:51:35,724 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92899.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:51:49,791 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-09 03:52:20,572 INFO [train2.py:809] (3/4) Epoch 24, batch 1300, loss[ctc_loss=0.05677, att_loss=0.2051, loss=0.1754, over 15757.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009562, over 38.00 utterances.], tot_loss[ctc_loss=0.06888, att_loss=0.2327, loss=0.2, over 3266569.01 frames. utt_duration=1276 frames, utt_pad_proportion=0.04647, over 10251.45 utterances.], batch size: 38, lr: 4.50e-03, grad_scale: 8.0 2023-03-09 03:52:27,986 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92931.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:52:53,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.940e+02 2.372e+02 2.718e+02 7.433e+02, threshold=4.745e+02, percent-clipped=1.0 2023-03-09 03:53:14,580 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92960.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:53:35,859 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8455, 5.0832, 5.1039, 4.9744, 5.1271, 5.0842, 4.7712, 4.6137], device='cuda:3'), covar=tensor([0.1028, 0.0494, 0.0289, 0.0518, 0.0276, 0.0330, 0.0362, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0364, 0.0350, 0.0363, 0.0427, 0.0435, 0.0361, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 03:53:39,312 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92975.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:53:39,415 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4528, 2.7162, 4.8475, 3.8070, 2.9300, 4.1184, 4.6372, 4.4916], device='cuda:3'), covar=tensor([0.0268, 0.1442, 0.0166, 0.0879, 0.1691, 0.0272, 0.0169, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0243, 0.0202, 0.0320, 0.0265, 0.0223, 0.0191, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:53:41,991 INFO [train2.py:809] (3/4) Epoch 24, batch 1350, loss[ctc_loss=0.09324, att_loss=0.2605, loss=0.2271, over 17564.00 frames. utt_duration=1020 frames, utt_pad_proportion=0.03892, over 69.00 utterances.], tot_loss[ctc_loss=0.06956, att_loss=0.2335, loss=0.2007, over 3270163.64 frames. utt_duration=1255 frames, utt_pad_proportion=0.05176, over 10435.96 utterances.], batch size: 69, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 03:53:49,250 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92981.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:54:04,018 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4729, 4.5306, 4.5724, 4.6008, 5.1295, 4.6223, 4.4708, 2.6231], device='cuda:3'), covar=tensor([0.0277, 0.0376, 0.0322, 0.0315, 0.0712, 0.0228, 0.0388, 0.1680], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0199, 0.0198, 0.0214, 0.0371, 0.0167, 0.0187, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:54:07,136 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92992.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:54:14,883 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6888, 5.0806, 4.9175, 4.9660, 5.2194, 4.8167, 3.6270, 5.0545], device='cuda:3'), covar=tensor([0.0116, 0.0107, 0.0133, 0.0084, 0.0078, 0.0120, 0.0648, 0.0184], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0089, 0.0112, 0.0070, 0.0076, 0.0087, 0.0103, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:54:25,752 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-09 03:54:36,461 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93010.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:55:03,585 INFO [train2.py:809] (3/4) Epoch 24, batch 1400, loss[ctc_loss=0.06564, att_loss=0.2392, loss=0.2045, over 17022.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.0114, over 53.00 utterances.], tot_loss[ctc_loss=0.06861, att_loss=0.233, loss=0.2001, over 3276340.63 frames. utt_duration=1269 frames, utt_pad_proportion=0.04793, over 10342.67 utterances.], batch size: 53, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 03:55:14,557 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7260, 5.9885, 5.5116, 5.7585, 5.6759, 5.1120, 5.3997, 5.1989], device='cuda:3'), covar=tensor([0.1376, 0.0962, 0.1022, 0.0830, 0.1035, 0.1708, 0.2209, 0.2269], device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0623, 0.0474, 0.0470, 0.0439, 0.0475, 0.0621, 0.0538], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 03:55:18,106 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0798, 5.0848, 4.8227, 2.2759, 2.0873, 3.0278, 2.5267, 3.7830], device='cuda:3'), covar=tensor([0.0736, 0.0358, 0.0320, 0.5082, 0.5454, 0.2414, 0.3557, 0.1824], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0281, 0.0268, 0.0244, 0.0334, 0.0331, 0.0258, 0.0366], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 03:55:19,537 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:55:19,567 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:55:36,735 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 1.820e+02 2.183e+02 2.571e+02 5.095e+02, threshold=4.367e+02, percent-clipped=2.0 2023-03-09 03:55:54,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 2023-03-09 03:56:16,566 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93071.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:56:24,064 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93076.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:56:25,389 INFO [train2.py:809] (3/4) Epoch 24, batch 1450, loss[ctc_loss=0.06534, att_loss=0.2081, loss=0.1795, over 15641.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008392, over 37.00 utterances.], tot_loss[ctc_loss=0.06816, att_loss=0.2326, loss=0.1997, over 3268281.04 frames. utt_duration=1288 frames, utt_pad_proportion=0.04381, over 10162.96 utterances.], batch size: 37, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 03:56:34,228 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9035, 4.3409, 4.3411, 4.5846, 2.5885, 4.4581, 2.9730, 2.0713], device='cuda:3'), covar=tensor([0.0549, 0.0263, 0.0695, 0.0244, 0.1721, 0.0196, 0.1274, 0.1523], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0171, 0.0264, 0.0166, 0.0223, 0.0157, 0.0232, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 03:56:57,982 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93097.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 03:57:26,928 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93115.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:57:45,164 INFO [train2.py:809] (3/4) Epoch 24, batch 1500, loss[ctc_loss=0.08784, att_loss=0.2421, loss=0.2112, over 17475.00 frames. utt_duration=1111 frames, utt_pad_proportion=0.02797, over 63.00 utterances.], tot_loss[ctc_loss=0.06889, att_loss=0.2338, loss=0.2009, over 3281535.54 frames. utt_duration=1274 frames, utt_pad_proportion=0.04385, over 10312.02 utterances.], batch size: 63, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 03:58:07,115 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3290, 2.9305, 3.5583, 2.9721, 3.5068, 4.4321, 4.2251, 3.3267], device='cuda:3'), covar=tensor([0.0360, 0.1828, 0.1372, 0.1335, 0.1103, 0.0941, 0.0683, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0245, 0.0283, 0.0219, 0.0267, 0.0372, 0.0266, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:58:17,655 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 1.858e+02 2.215e+02 2.637e+02 6.796e+02, threshold=4.431e+02, percent-clipped=2.0 2023-03-09 03:58:44,173 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8582, 5.1300, 5.4582, 5.2604, 5.3231, 5.8223, 5.1154, 5.9240], device='cuda:3'), covar=tensor([0.0762, 0.0707, 0.0782, 0.1393, 0.1935, 0.0986, 0.0800, 0.0662], device='cuda:3'), in_proj_covar=tensor([0.0900, 0.0522, 0.0624, 0.0673, 0.0898, 0.0647, 0.0508, 0.0634], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 03:59:04,761 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93176.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 03:59:05,932 INFO [train2.py:809] (3/4) Epoch 24, batch 1550, loss[ctc_loss=0.06907, att_loss=0.2528, loss=0.2161, over 16636.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004707, over 47.00 utterances.], tot_loss[ctc_loss=0.07008, att_loss=0.2348, loss=0.2018, over 3291997.11 frames. utt_duration=1283 frames, utt_pad_proportion=0.03911, over 10271.87 utterances.], batch size: 47, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 03:59:23,142 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93187.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:00:02,569 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8624, 6.0935, 5.5333, 5.8399, 5.7114, 5.2625, 5.5696, 5.3164], device='cuda:3'), covar=tensor([0.1114, 0.0952, 0.1069, 0.0865, 0.1049, 0.1656, 0.2246, 0.2317], device='cuda:3'), in_proj_covar=tensor([0.0536, 0.0627, 0.0477, 0.0474, 0.0441, 0.0478, 0.0627, 0.0544], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 04:00:26,992 INFO [train2.py:809] (3/4) Epoch 24, batch 1600, loss[ctc_loss=0.06033, att_loss=0.2287, loss=0.195, over 16552.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005657, over 45.00 utterances.], tot_loss[ctc_loss=0.06968, att_loss=0.2346, loss=0.2016, over 3300606.32 frames. utt_duration=1283 frames, utt_pad_proportion=0.03685, over 10302.90 utterances.], batch size: 45, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 04:00:59,418 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.026e+02 2.316e+02 3.152e+02 2.970e+03, threshold=4.633e+02, percent-clipped=9.0 2023-03-09 04:01:12,777 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93255.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:01:47,571 INFO [train2.py:809] (3/4) Epoch 24, batch 1650, loss[ctc_loss=0.05997, att_loss=0.2146, loss=0.1837, over 15872.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009939, over 39.00 utterances.], tot_loss[ctc_loss=0.06954, att_loss=0.2343, loss=0.2013, over 3289766.16 frames. utt_duration=1294 frames, utt_pad_proportion=0.03708, over 10178.03 utterances.], batch size: 39, lr: 4.49e-03, grad_scale: 8.0 2023-03-09 04:01:54,068 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93281.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:02:05,089 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93287.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:03:09,907 INFO [train2.py:809] (3/4) Epoch 24, batch 1700, loss[ctc_loss=0.06649, att_loss=0.248, loss=0.2117, over 16832.00 frames. utt_duration=681.4 frames, utt_pad_proportion=0.144, over 99.00 utterances.], tot_loss[ctc_loss=0.06891, att_loss=0.2338, loss=0.2008, over 3285851.40 frames. utt_duration=1301 frames, utt_pad_proportion=0.03756, over 10113.26 utterances.], batch size: 99, lr: 4.49e-03, grad_scale: 16.0 2023-03-09 04:03:13,050 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:03:16,344 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93331.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:03:42,410 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 1.807e+02 2.162e+02 2.444e+02 4.017e+02, threshold=4.325e+02, percent-clipped=0.0 2023-03-09 04:04:14,241 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93366.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:04:29,843 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:04:31,040 INFO [train2.py:809] (3/4) Epoch 24, batch 1750, loss[ctc_loss=0.05489, att_loss=0.2144, loss=0.1825, over 15644.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008751, over 37.00 utterances.], tot_loss[ctc_loss=0.06832, att_loss=0.2334, loss=0.2004, over 3281036.76 frames. utt_duration=1280 frames, utt_pad_proportion=0.04379, over 10266.51 utterances.], batch size: 37, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:04:56,267 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93392.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 04:04:56,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-09 04:05:28,887 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-09 04:05:48,345 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93424.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:05:52,975 INFO [train2.py:809] (3/4) Epoch 24, batch 1800, loss[ctc_loss=0.07091, att_loss=0.2494, loss=0.2137, over 17306.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01083, over 55.00 utterances.], tot_loss[ctc_loss=0.06871, att_loss=0.2341, loss=0.201, over 3285348.12 frames. utt_duration=1238 frames, utt_pad_proportion=0.0528, over 10626.83 utterances.], batch size: 55, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:06:25,985 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 1.826e+02 2.245e+02 2.637e+02 3.732e+02, threshold=4.489e+02, percent-clipped=0.0 2023-03-09 04:07:05,218 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93471.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:07:14,512 INFO [train2.py:809] (3/4) Epoch 24, batch 1850, loss[ctc_loss=0.07577, att_loss=0.2458, loss=0.2118, over 17006.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009455, over 51.00 utterances.], tot_loss[ctc_loss=0.06938, att_loss=0.2345, loss=0.2015, over 3276757.90 frames. utt_duration=1210 frames, utt_pad_proportion=0.06288, over 10841.42 utterances.], batch size: 51, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:07:31,217 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:07:47,734 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6495, 2.1553, 2.2084, 2.3214, 2.4123, 2.2940, 2.1518, 2.4793], device='cuda:3'), covar=tensor([0.0905, 0.2307, 0.1780, 0.1640, 0.1638, 0.1055, 0.1725, 0.1224], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0131, 0.0128, 0.0119, 0.0134, 0.0115, 0.0139, 0.0113], device='cuda:3'), out_proj_covar=tensor([9.6804e-05, 1.0241e-04, 1.0283e-04, 9.3392e-05, 1.0037e-04, 9.2422e-05, 1.0577e-04, 8.9538e-05], device='cuda:3') 2023-03-09 04:08:35,371 INFO [train2.py:809] (3/4) Epoch 24, batch 1900, loss[ctc_loss=0.05651, att_loss=0.2275, loss=0.1933, over 16410.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.007117, over 44.00 utterances.], tot_loss[ctc_loss=0.06962, att_loss=0.2342, loss=0.2013, over 3277499.20 frames. utt_duration=1224 frames, utt_pad_proportion=0.05912, over 10720.95 utterances.], batch size: 44, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:08:48,847 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93535.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:09:07,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 1.868e+02 2.241e+02 2.979e+02 6.717e+02, threshold=4.482e+02, percent-clipped=6.0 2023-03-09 04:09:21,333 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93555.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:09:55,680 INFO [train2.py:809] (3/4) Epoch 24, batch 1950, loss[ctc_loss=0.05543, att_loss=0.2255, loss=0.1915, over 16132.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.006018, over 42.00 utterances.], tot_loss[ctc_loss=0.06944, att_loss=0.2338, loss=0.2009, over 3280664.87 frames. utt_duration=1234 frames, utt_pad_proportion=0.05574, over 10649.28 utterances.], batch size: 42, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:10:12,640 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93587.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:10:38,952 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93603.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:10:52,619 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4872, 2.5507, 4.8787, 3.8922, 3.0435, 4.2600, 4.6268, 4.5899], device='cuda:3'), covar=tensor([0.0261, 0.1574, 0.0199, 0.0879, 0.1640, 0.0229, 0.0188, 0.0277], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0242, 0.0203, 0.0319, 0.0265, 0.0222, 0.0192, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:11:17,167 INFO [train2.py:809] (3/4) Epoch 24, batch 2000, loss[ctc_loss=0.0847, att_loss=0.2184, loss=0.1917, over 15502.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.0075, over 36.00 utterances.], tot_loss[ctc_loss=0.06872, att_loss=0.233, loss=0.2002, over 3280953.81 frames. utt_duration=1261 frames, utt_pad_proportion=0.04898, over 10417.17 utterances.], batch size: 36, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:11:24,599 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93631.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:11:31,261 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93635.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:11:50,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 1.872e+02 2.240e+02 2.588e+02 7.651e+02, threshold=4.480e+02, percent-clipped=4.0 2023-03-09 04:12:22,116 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93666.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:12:38,935 INFO [train2.py:809] (3/4) Epoch 24, batch 2050, loss[ctc_loss=0.05432, att_loss=0.2316, loss=0.1961, over 16274.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007193, over 43.00 utterances.], tot_loss[ctc_loss=0.06901, att_loss=0.2334, loss=0.2005, over 3278760.43 frames. utt_duration=1260 frames, utt_pad_proportion=0.05071, over 10417.28 utterances.], batch size: 43, lr: 4.48e-03, grad_scale: 16.0 2023-03-09 04:12:42,979 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93679.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:13:03,664 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93692.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:13:39,490 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93714.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:14:00,049 INFO [train2.py:809] (3/4) Epoch 24, batch 2100, loss[ctc_loss=0.04902, att_loss=0.2067, loss=0.1752, over 15868.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01028, over 39.00 utterances.], tot_loss[ctc_loss=0.06926, att_loss=0.2335, loss=0.2007, over 3260931.34 frames. utt_duration=1257 frames, utt_pad_proportion=0.0534, over 10389.79 utterances.], batch size: 39, lr: 4.48e-03, grad_scale: 8.0 2023-03-09 04:14:21,497 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93740.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:14:34,152 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 1.856e+02 2.273e+02 2.574e+02 6.013e+02, threshold=4.546e+02, percent-clipped=4.0 2023-03-09 04:15:01,027 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-03-09 04:15:09,528 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93770.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:15:11,030 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93771.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:15:19,886 INFO [train2.py:809] (3/4) Epoch 24, batch 2150, loss[ctc_loss=0.05395, att_loss=0.2258, loss=0.1915, over 16398.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007243, over 44.00 utterances.], tot_loss[ctc_loss=0.06994, att_loss=0.2345, loss=0.2016, over 3260780.46 frames. utt_duration=1243 frames, utt_pad_proportion=0.05813, over 10509.29 utterances.], batch size: 44, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:16:28,701 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93819.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:16:41,182 INFO [train2.py:809] (3/4) Epoch 24, batch 2200, loss[ctc_loss=0.08253, att_loss=0.2449, loss=0.2124, over 17318.00 frames. utt_duration=1101 frames, utt_pad_proportion=0.0375, over 63.00 utterances.], tot_loss[ctc_loss=0.06921, att_loss=0.2337, loss=0.2008, over 3257665.72 frames. utt_duration=1257 frames, utt_pad_proportion=0.05623, over 10375.67 utterances.], batch size: 63, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:16:49,224 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93831.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:17:05,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-09 04:17:16,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.790e+02 2.187e+02 2.751e+02 8.387e+02, threshold=4.374e+02, percent-clipped=4.0 2023-03-09 04:18:02,864 INFO [train2.py:809] (3/4) Epoch 24, batch 2250, loss[ctc_loss=0.05142, att_loss=0.2348, loss=0.1982, over 17002.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009673, over 51.00 utterances.], tot_loss[ctc_loss=0.06949, att_loss=0.2336, loss=0.2008, over 3254366.01 frames. utt_duration=1232 frames, utt_pad_proportion=0.06378, over 10582.74 utterances.], batch size: 51, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:18:22,986 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4599, 4.5449, 4.5542, 4.5380, 5.1729, 4.5097, 4.5119, 2.8693], device='cuda:3'), covar=tensor([0.0275, 0.0373, 0.0365, 0.0345, 0.0814, 0.0257, 0.0387, 0.1548], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0200, 0.0198, 0.0214, 0.0371, 0.0170, 0.0187, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:18:53,051 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4185, 2.5442, 4.8112, 3.7746, 2.9927, 4.1587, 4.4581, 4.5408], device='cuda:3'), covar=tensor([0.0248, 0.1544, 0.0175, 0.0844, 0.1558, 0.0270, 0.0228, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0243, 0.0203, 0.0319, 0.0267, 0.0223, 0.0193, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:19:24,949 INFO [train2.py:809] (3/4) Epoch 24, batch 2300, loss[ctc_loss=0.09606, att_loss=0.2434, loss=0.214, over 14120.00 frames. utt_duration=390.9 frames, utt_pad_proportion=0.3202, over 145.00 utterances.], tot_loss[ctc_loss=0.06989, att_loss=0.234, loss=0.2012, over 3244523.95 frames. utt_duration=1189 frames, utt_pad_proportion=0.07747, over 10929.10 utterances.], batch size: 145, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:19:45,643 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1670, 3.8321, 3.3112, 3.3710, 4.0131, 3.7247, 3.1644, 4.3734], device='cuda:3'), covar=tensor([0.0885, 0.0438, 0.0963, 0.0736, 0.0641, 0.0723, 0.0789, 0.0399], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0220, 0.0225, 0.0204, 0.0282, 0.0244, 0.0200, 0.0289], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:20:00,414 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.894e+02 2.401e+02 2.825e+02 6.941e+02, threshold=4.802e+02, percent-clipped=5.0 2023-03-09 04:20:01,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-03-09 04:20:04,029 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9711, 3.9905, 4.0578, 4.0833, 4.1298, 4.1370, 3.8888, 3.8521], device='cuda:3'), covar=tensor([0.1012, 0.0895, 0.0896, 0.0593, 0.0381, 0.0412, 0.0479, 0.0346], device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0366, 0.0356, 0.0368, 0.0431, 0.0436, 0.0364, 0.0400], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:20:47,685 INFO [train2.py:809] (3/4) Epoch 24, batch 2350, loss[ctc_loss=0.05622, att_loss=0.2131, loss=0.1817, over 15870.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.009031, over 39.00 utterances.], tot_loss[ctc_loss=0.07, att_loss=0.2344, loss=0.2015, over 3252193.49 frames. utt_duration=1188 frames, utt_pad_proportion=0.07613, over 10965.64 utterances.], batch size: 39, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:20:55,324 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93981.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:21:22,306 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2989, 3.8525, 3.3441, 3.5054, 4.0769, 3.7409, 3.1605, 4.3775], device='cuda:3'), covar=tensor([0.0851, 0.0538, 0.0957, 0.0697, 0.0668, 0.0711, 0.0841, 0.0447], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0220, 0.0225, 0.0204, 0.0282, 0.0245, 0.0200, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:21:30,310 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9224, 2.4810, 2.5662, 2.6034, 2.7732, 2.6927, 2.4234, 2.8876], device='cuda:3'), covar=tensor([0.1046, 0.2131, 0.1500, 0.1023, 0.1183, 0.0964, 0.1778, 0.0982], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0131, 0.0127, 0.0118, 0.0133, 0.0115, 0.0139, 0.0113], device='cuda:3'), out_proj_covar=tensor([9.6949e-05, 1.0253e-04, 1.0234e-04, 9.2665e-05, 1.0002e-04, 9.2539e-05, 1.0576e-04, 8.9814e-05], device='cuda:3') 2023-03-09 04:22:13,387 INFO [train2.py:809] (3/4) Epoch 24, batch 2400, loss[ctc_loss=0.08409, att_loss=0.2483, loss=0.2155, over 17281.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01272, over 55.00 utterances.], tot_loss[ctc_loss=0.07032, att_loss=0.2352, loss=0.2022, over 3262080.15 frames. utt_duration=1188 frames, utt_pad_proportion=0.0739, over 10995.76 utterances.], batch size: 55, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:22:15,445 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3765, 2.4345, 4.7946, 3.6876, 2.8206, 4.1957, 4.6249, 4.5874], device='cuda:3'), covar=tensor([0.0259, 0.1705, 0.0195, 0.0987, 0.1894, 0.0254, 0.0183, 0.0242], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0243, 0.0204, 0.0319, 0.0268, 0.0224, 0.0193, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:22:38,922 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94042.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:22:47,937 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 1.881e+02 2.121e+02 2.677e+02 6.185e+02, threshold=4.243e+02, percent-clipped=1.0 2023-03-09 04:23:05,307 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4209, 4.8338, 4.7039, 4.7986, 4.9229, 4.4960, 3.2385, 4.7799], device='cuda:3'), covar=tensor([0.0139, 0.0116, 0.0136, 0.0088, 0.0101, 0.0137, 0.0795, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0089, 0.0111, 0.0070, 0.0076, 0.0087, 0.0103, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:23:34,727 INFO [train2.py:809] (3/4) Epoch 24, batch 2450, loss[ctc_loss=0.07853, att_loss=0.2436, loss=0.2106, over 17399.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03394, over 63.00 utterances.], tot_loss[ctc_loss=0.06963, att_loss=0.2343, loss=0.2013, over 3266202.35 frames. utt_duration=1222 frames, utt_pad_proportion=0.06414, over 10706.15 utterances.], batch size: 63, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:23:38,226 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6718, 2.9898, 3.0151, 2.7148, 3.0533, 3.0122, 3.0859, 2.2427], device='cuda:3'), covar=tensor([0.1289, 0.1642, 0.1960, 0.3280, 0.1615, 0.1940, 0.1345, 0.3575], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0199, 0.0214, 0.0266, 0.0173, 0.0274, 0.0198, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:24:40,297 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0459, 5.3347, 5.6044, 5.3571, 5.5769, 6.0162, 5.2560, 6.0819], device='cuda:3'), covar=tensor([0.0694, 0.0768, 0.0857, 0.1384, 0.1727, 0.0913, 0.0739, 0.0728], device='cuda:3'), in_proj_covar=tensor([0.0893, 0.0515, 0.0617, 0.0663, 0.0887, 0.0637, 0.0507, 0.0630], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:24:55,083 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94126.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:24:56,467 INFO [train2.py:809] (3/4) Epoch 24, batch 2500, loss[ctc_loss=0.06847, att_loss=0.2201, loss=0.1898, over 15647.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008529, over 37.00 utterances.], tot_loss[ctc_loss=0.06924, att_loss=0.2342, loss=0.2012, over 3268973.54 frames. utt_duration=1227 frames, utt_pad_proportion=0.06181, over 10665.68 utterances.], batch size: 37, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:25:12,524 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6914, 3.3765, 3.8637, 3.1507, 3.6535, 4.7311, 4.5686, 3.3883], device='cuda:3'), covar=tensor([0.0315, 0.1369, 0.1228, 0.1317, 0.1068, 0.0717, 0.0542, 0.1235], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0245, 0.0285, 0.0219, 0.0267, 0.0369, 0.0266, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:25:31,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.808e+02 2.198e+02 2.566e+02 4.413e+02, threshold=4.397e+02, percent-clipped=1.0 2023-03-09 04:26:18,034 INFO [train2.py:809] (3/4) Epoch 24, batch 2550, loss[ctc_loss=0.07976, att_loss=0.2488, loss=0.215, over 17069.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007866, over 52.00 utterances.], tot_loss[ctc_loss=0.06916, att_loss=0.2335, loss=0.2007, over 3268913.78 frames. utt_duration=1255 frames, utt_pad_proportion=0.05434, over 10434.97 utterances.], batch size: 52, lr: 4.47e-03, grad_scale: 8.0 2023-03-09 04:27:32,997 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94223.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 04:27:39,576 INFO [train2.py:809] (3/4) Epoch 24, batch 2600, loss[ctc_loss=0.06867, att_loss=0.2359, loss=0.2025, over 17124.00 frames. utt_duration=693.2 frames, utt_pad_proportion=0.1248, over 99.00 utterances.], tot_loss[ctc_loss=0.06927, att_loss=0.2336, loss=0.2008, over 3278141.16 frames. utt_duration=1255 frames, utt_pad_proportion=0.05144, over 10457.81 utterances.], batch size: 99, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:28:14,364 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 1.761e+02 2.111e+02 2.578e+02 4.836e+02, threshold=4.223e+02, percent-clipped=1.0 2023-03-09 04:28:28,967 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7276, 3.4690, 3.4262, 2.8979, 3.3691, 3.4497, 3.5255, 2.3518], device='cuda:3'), covar=tensor([0.1213, 0.1112, 0.2601, 0.3482, 0.2157, 0.2489, 0.0933, 0.4639], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0197, 0.0212, 0.0264, 0.0173, 0.0273, 0.0198, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:28:40,195 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7590, 6.0115, 5.4617, 5.7211, 5.6636, 5.1824, 5.4471, 5.2366], device='cuda:3'), covar=tensor([0.1298, 0.0887, 0.0920, 0.0819, 0.0796, 0.1679, 0.2463, 0.2194], device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0618, 0.0472, 0.0468, 0.0436, 0.0474, 0.0628, 0.0535], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 04:29:01,211 INFO [train2.py:809] (3/4) Epoch 24, batch 2650, loss[ctc_loss=0.06275, att_loss=0.2362, loss=0.2015, over 16618.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005738, over 47.00 utterances.], tot_loss[ctc_loss=0.06959, att_loss=0.2344, loss=0.2014, over 3285425.16 frames. utt_duration=1237 frames, utt_pad_proportion=0.0538, over 10638.44 utterances.], batch size: 47, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:29:14,083 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94284.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 04:29:48,805 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1589, 4.2662, 4.3439, 4.5879, 2.4727, 4.5463, 2.5246, 1.4177], device='cuda:3'), covar=tensor([0.0417, 0.0261, 0.0694, 0.0219, 0.1934, 0.0196, 0.1639, 0.2087], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0172, 0.0260, 0.0165, 0.0220, 0.0157, 0.0228, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:29:53,607 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5668, 2.9692, 3.8189, 3.2905, 3.5480, 4.7312, 4.5613, 3.5610], device='cuda:3'), covar=tensor([0.0428, 0.1948, 0.1111, 0.1233, 0.1157, 0.0804, 0.0597, 0.1095], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0247, 0.0285, 0.0220, 0.0266, 0.0370, 0.0266, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:30:23,802 INFO [train2.py:809] (3/4) Epoch 24, batch 2700, loss[ctc_loss=0.07625, att_loss=0.2527, loss=0.2174, over 17023.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.009746, over 53.00 utterances.], tot_loss[ctc_loss=0.06969, att_loss=0.2347, loss=0.2017, over 3285486.56 frames. utt_duration=1227 frames, utt_pad_proportion=0.05573, over 10725.91 utterances.], batch size: 53, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:30:40,564 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94337.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:30:57,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 1.818e+02 2.150e+02 2.630e+02 4.839e+02, threshold=4.300e+02, percent-clipped=3.0 2023-03-09 04:30:58,027 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94348.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:31:45,415 INFO [train2.py:809] (3/4) Epoch 24, batch 2750, loss[ctc_loss=0.06071, att_loss=0.2074, loss=0.1781, over 15652.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008307, over 37.00 utterances.], tot_loss[ctc_loss=0.07017, att_loss=0.2347, loss=0.2018, over 3286995.05 frames. utt_duration=1239 frames, utt_pad_proportion=0.05257, over 10623.79 utterances.], batch size: 37, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:32:38,151 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94409.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 04:32:43,022 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1116, 3.7466, 3.2309, 3.5778, 4.0135, 3.7047, 3.1428, 4.3104], device='cuda:3'), covar=tensor([0.0880, 0.0500, 0.1039, 0.0594, 0.0636, 0.0634, 0.0771, 0.0385], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0221, 0.0225, 0.0204, 0.0282, 0.0245, 0.0201, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:33:06,655 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94426.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:33:07,894 INFO [train2.py:809] (3/4) Epoch 24, batch 2800, loss[ctc_loss=0.07514, att_loss=0.2492, loss=0.2144, over 17041.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.008933, over 52.00 utterances.], tot_loss[ctc_loss=0.07033, att_loss=0.2348, loss=0.2019, over 3284856.68 frames. utt_duration=1227 frames, utt_pad_proportion=0.05575, over 10717.59 utterances.], batch size: 52, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:33:40,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.912e+02 2.313e+02 2.793e+02 7.224e+02, threshold=4.627e+02, percent-clipped=4.0 2023-03-09 04:34:24,243 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94474.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:34:28,880 INFO [train2.py:809] (3/4) Epoch 24, batch 2850, loss[ctc_loss=0.07454, att_loss=0.2459, loss=0.2116, over 17112.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.0139, over 56.00 utterances.], tot_loss[ctc_loss=0.07069, att_loss=0.2353, loss=0.2024, over 3286021.00 frames. utt_duration=1202 frames, utt_pad_proportion=0.06208, over 10947.33 utterances.], batch size: 56, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:34:41,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-09 04:35:52,354 INFO [train2.py:809] (3/4) Epoch 24, batch 2900, loss[ctc_loss=0.06382, att_loss=0.2206, loss=0.1893, over 15643.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008328, over 37.00 utterances.], tot_loss[ctc_loss=0.0705, att_loss=0.2351, loss=0.2022, over 3282745.31 frames. utt_duration=1209 frames, utt_pad_proportion=0.06134, over 10871.14 utterances.], batch size: 37, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:36:26,078 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.878e+02 2.321e+02 2.758e+02 4.740e+02, threshold=4.642e+02, percent-clipped=1.0 2023-03-09 04:37:14,351 INFO [train2.py:809] (3/4) Epoch 24, batch 2950, loss[ctc_loss=0.06364, att_loss=0.2325, loss=0.1987, over 16529.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006842, over 45.00 utterances.], tot_loss[ctc_loss=0.07081, att_loss=0.2361, loss=0.203, over 3289242.12 frames. utt_duration=1186 frames, utt_pad_proportion=0.06501, over 11108.62 utterances.], batch size: 45, lr: 4.46e-03, grad_scale: 8.0 2023-03-09 04:37:17,834 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94579.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 04:37:41,578 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0840, 2.7683, 3.1375, 4.1893, 3.8141, 3.7629, 2.7708, 2.0273], device='cuda:3'), covar=tensor([0.0868, 0.2133, 0.0976, 0.0552, 0.0824, 0.0484, 0.1658, 0.2334], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0220, 0.0189, 0.0222, 0.0231, 0.0185, 0.0206, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:38:08,137 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9295, 3.6719, 3.0787, 3.3746, 3.8117, 3.5601, 2.8797, 4.1008], device='cuda:3'), covar=tensor([0.0988, 0.0481, 0.1023, 0.0685, 0.0747, 0.0687, 0.0901, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0221, 0.0225, 0.0204, 0.0283, 0.0245, 0.0201, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:38:36,142 INFO [train2.py:809] (3/4) Epoch 24, batch 3000, loss[ctc_loss=0.06456, att_loss=0.2275, loss=0.1949, over 16138.00 frames. utt_duration=1539 frames, utt_pad_proportion=0.005479, over 42.00 utterances.], tot_loss[ctc_loss=0.07022, att_loss=0.2357, loss=0.2026, over 3290480.15 frames. utt_duration=1220 frames, utt_pad_proportion=0.05631, over 10803.32 utterances.], batch size: 42, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:38:36,142 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 04:38:50,629 INFO [train2.py:843] (3/4) Epoch 24, validation: ctc_loss=0.04165, att_loss=0.2345, loss=0.196, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 04:38:50,629 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 04:39:02,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-09 04:39:06,757 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94637.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:39:09,918 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0506, 5.3561, 5.6698, 5.4904, 5.5471, 6.0123, 5.2561, 6.0603], device='cuda:3'), covar=tensor([0.0682, 0.0768, 0.0778, 0.1281, 0.1764, 0.0861, 0.0760, 0.0699], device='cuda:3'), in_proj_covar=tensor([0.0902, 0.0523, 0.0630, 0.0675, 0.0904, 0.0648, 0.0515, 0.0640], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:39:23,558 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 2.097e+02 2.380e+02 3.047e+02 5.871e+02, threshold=4.759e+02, percent-clipped=2.0 2023-03-09 04:40:11,569 INFO [train2.py:809] (3/4) Epoch 24, batch 3050, loss[ctc_loss=0.09674, att_loss=0.256, loss=0.2241, over 17286.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.0119, over 55.00 utterances.], tot_loss[ctc_loss=0.07146, att_loss=0.2363, loss=0.2033, over 3277372.55 frames. utt_duration=1159 frames, utt_pad_proportion=0.07583, over 11323.40 utterances.], batch size: 55, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:40:24,109 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94685.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:40:54,710 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94704.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 04:41:21,328 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 04:41:22,085 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5347, 2.4628, 4.9607, 3.7814, 2.8788, 4.2531, 4.7998, 4.6561], device='cuda:3'), covar=tensor([0.0267, 0.1572, 0.0179, 0.0983, 0.1813, 0.0264, 0.0161, 0.0269], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0240, 0.0203, 0.0316, 0.0266, 0.0222, 0.0192, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:41:25,597 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-09 04:41:32,541 INFO [train2.py:809] (3/4) Epoch 24, batch 3100, loss[ctc_loss=0.06541, att_loss=0.2428, loss=0.2073, over 16945.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.007953, over 50.00 utterances.], tot_loss[ctc_loss=0.07201, att_loss=0.2365, loss=0.2036, over 3277988.61 frames. utt_duration=1170 frames, utt_pad_proportion=0.07215, over 11223.11 utterances.], batch size: 50, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:41:42,126 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8830, 6.1268, 5.6113, 5.8461, 5.7407, 5.3279, 5.4970, 5.2969], device='cuda:3'), covar=tensor([0.1203, 0.0841, 0.0938, 0.0758, 0.0884, 0.1460, 0.2296, 0.2217], device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0619, 0.0471, 0.0469, 0.0437, 0.0476, 0.0624, 0.0536], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 04:42:05,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 1.910e+02 2.247e+02 2.795e+02 4.738e+02, threshold=4.495e+02, percent-clipped=0.0 2023-03-09 04:42:54,025 INFO [train2.py:809] (3/4) Epoch 24, batch 3150, loss[ctc_loss=0.1007, att_loss=0.2613, loss=0.2291, over 17107.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01497, over 56.00 utterances.], tot_loss[ctc_loss=0.07168, att_loss=0.236, loss=0.2032, over 3279546.86 frames. utt_duration=1184 frames, utt_pad_proportion=0.06829, over 11090.66 utterances.], batch size: 56, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:44:15,928 INFO [train2.py:809] (3/4) Epoch 24, batch 3200, loss[ctc_loss=0.06202, att_loss=0.23, loss=0.1964, over 16178.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.005695, over 41.00 utterances.], tot_loss[ctc_loss=0.07106, att_loss=0.2359, loss=0.2029, over 3281844.40 frames. utt_duration=1179 frames, utt_pad_proportion=0.06861, over 11150.54 utterances.], batch size: 41, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:44:33,522 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9003, 4.8810, 4.6024, 2.8285, 4.6752, 4.5016, 4.0000, 2.5201], device='cuda:3'), covar=tensor([0.0110, 0.0121, 0.0376, 0.1128, 0.0116, 0.0243, 0.0411, 0.1573], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0105, 0.0108, 0.0114, 0.0088, 0.0117, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 04:44:48,823 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 1.841e+02 2.191e+02 2.766e+02 6.449e+02, threshold=4.381e+02, percent-clipped=4.0 2023-03-09 04:45:36,260 INFO [train2.py:809] (3/4) Epoch 24, batch 3250, loss[ctc_loss=0.05482, att_loss=0.2097, loss=0.1787, over 15502.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008865, over 36.00 utterances.], tot_loss[ctc_loss=0.07101, att_loss=0.2354, loss=0.2025, over 3283906.63 frames. utt_duration=1201 frames, utt_pad_proportion=0.06259, over 10951.12 utterances.], batch size: 36, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:45:39,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94879.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 04:46:56,542 INFO [train2.py:809] (3/4) Epoch 24, batch 3300, loss[ctc_loss=0.06054, att_loss=0.2054, loss=0.1764, over 15366.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01095, over 35.00 utterances.], tot_loss[ctc_loss=0.07024, att_loss=0.2346, loss=0.2017, over 3286760.01 frames. utt_duration=1229 frames, utt_pad_proportion=0.05502, over 10710.23 utterances.], batch size: 35, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:46:56,658 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94927.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 04:47:27,777 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3625, 3.0320, 3.6267, 4.4704, 4.0205, 4.0755, 3.0142, 2.2344], device='cuda:3'), covar=tensor([0.0784, 0.1879, 0.0777, 0.0565, 0.0907, 0.0396, 0.1444, 0.2233], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0217, 0.0186, 0.0220, 0.0228, 0.0181, 0.0204, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:47:29,073 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.857e+02 2.212e+02 2.661e+02 5.199e+02, threshold=4.424e+02, percent-clipped=2.0 2023-03-09 04:48:16,305 INFO [train2.py:809] (3/4) Epoch 24, batch 3350, loss[ctc_loss=0.06429, att_loss=0.2194, loss=0.1884, over 15953.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006346, over 41.00 utterances.], tot_loss[ctc_loss=0.06948, att_loss=0.2341, loss=0.2012, over 3287709.64 frames. utt_duration=1254 frames, utt_pad_proportion=0.04848, over 10498.81 utterances.], batch size: 41, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:48:59,851 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95004.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:49:28,807 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-09 04:49:31,236 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4012, 2.8506, 3.5835, 3.1332, 3.4283, 4.4703, 4.2866, 3.2771], device='cuda:3'), covar=tensor([0.0368, 0.1804, 0.1293, 0.1140, 0.1044, 0.0808, 0.0550, 0.1126], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0248, 0.0287, 0.0222, 0.0270, 0.0374, 0.0268, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:49:36,974 INFO [train2.py:809] (3/4) Epoch 24, batch 3400, loss[ctc_loss=0.05341, att_loss=0.211, loss=0.1795, over 15893.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008708, over 39.00 utterances.], tot_loss[ctc_loss=0.06935, att_loss=0.2342, loss=0.2012, over 3283730.19 frames. utt_duration=1266 frames, utt_pad_proportion=0.04503, over 10383.31 utterances.], batch size: 39, lr: 4.45e-03, grad_scale: 8.0 2023-03-09 04:49:38,781 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8851, 5.2011, 5.4766, 5.2958, 5.4128, 5.8146, 5.1691, 5.9382], device='cuda:3'), covar=tensor([0.0659, 0.0678, 0.0735, 0.1367, 0.1657, 0.0990, 0.0830, 0.0634], device='cuda:3'), in_proj_covar=tensor([0.0898, 0.0519, 0.0627, 0.0670, 0.0898, 0.0645, 0.0510, 0.0634], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 04:50:09,911 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.325e+02 1.881e+02 2.292e+02 2.864e+02 6.221e+02, threshold=4.584e+02, percent-clipped=2.0 2023-03-09 04:50:16,278 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95052.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:50:28,796 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3192, 3.9349, 3.3833, 3.6087, 4.1204, 3.7388, 3.1405, 4.4099], device='cuda:3'), covar=tensor([0.0842, 0.0452, 0.0914, 0.0616, 0.0725, 0.0738, 0.0844, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0223, 0.0227, 0.0206, 0.0285, 0.0245, 0.0203, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:50:56,269 INFO [train2.py:809] (3/4) Epoch 24, batch 3450, loss[ctc_loss=0.07567, att_loss=0.2473, loss=0.213, over 16752.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006352, over 48.00 utterances.], tot_loss[ctc_loss=0.07083, att_loss=0.2348, loss=0.202, over 3271449.11 frames. utt_duration=1211 frames, utt_pad_proportion=0.0626, over 10817.38 utterances.], batch size: 48, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:51:00,037 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1960, 5.1676, 4.9983, 3.0976, 4.9901, 4.8056, 4.4195, 2.8959], device='cuda:3'), covar=tensor([0.0092, 0.0105, 0.0229, 0.0945, 0.0091, 0.0173, 0.0302, 0.1240], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0104, 0.0106, 0.0112, 0.0087, 0.0115, 0.0100, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 04:51:41,230 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95105.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 04:52:16,797 INFO [train2.py:809] (3/4) Epoch 24, batch 3500, loss[ctc_loss=0.07028, att_loss=0.2391, loss=0.2053, over 16321.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006612, over 45.00 utterances.], tot_loss[ctc_loss=0.07114, att_loss=0.2346, loss=0.2019, over 3268044.84 frames. utt_duration=1209 frames, utt_pad_proportion=0.06321, over 10822.08 utterances.], batch size: 45, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:52:32,465 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4187, 2.9188, 3.3116, 4.4182, 4.1125, 3.9745, 3.0448, 2.4635], device='cuda:3'), covar=tensor([0.0806, 0.2157, 0.1028, 0.0660, 0.0924, 0.0474, 0.1501, 0.2154], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0217, 0.0186, 0.0219, 0.0226, 0.0181, 0.0204, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:52:49,706 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.240e+01 1.898e+02 2.336e+02 2.890e+02 6.917e+02, threshold=4.673e+02, percent-clipped=5.0 2023-03-09 04:53:18,830 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 04:53:35,722 INFO [train2.py:809] (3/4) Epoch 24, batch 3550, loss[ctc_loss=0.07265, att_loss=0.2396, loss=0.2062, over 17396.00 frames. utt_duration=882.2 frames, utt_pad_proportion=0.07531, over 79.00 utterances.], tot_loss[ctc_loss=0.07033, att_loss=0.2338, loss=0.2011, over 3259800.65 frames. utt_duration=1241 frames, utt_pad_proportion=0.0576, over 10520.64 utterances.], batch size: 79, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:54:56,146 INFO [train2.py:809] (3/4) Epoch 24, batch 3600, loss[ctc_loss=0.07034, att_loss=0.2514, loss=0.2152, over 17137.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01402, over 56.00 utterances.], tot_loss[ctc_loss=0.07084, att_loss=0.2345, loss=0.2018, over 3265104.45 frames. utt_duration=1207 frames, utt_pad_proportion=0.06532, over 10835.88 utterances.], batch size: 56, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:55:29,603 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 1.952e+02 2.410e+02 2.837e+02 7.042e+02, threshold=4.821e+02, percent-clipped=2.0 2023-03-09 04:55:50,128 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95260.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:56:02,723 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95268.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:56:16,598 INFO [train2.py:809] (3/4) Epoch 24, batch 3650, loss[ctc_loss=0.07023, att_loss=0.2257, loss=0.1946, over 16276.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006812, over 43.00 utterances.], tot_loss[ctc_loss=0.07035, att_loss=0.2339, loss=0.2012, over 3256494.39 frames. utt_duration=1213 frames, utt_pad_proportion=0.06473, over 10754.40 utterances.], batch size: 43, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:56:42,163 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2574, 4.4672, 4.5048, 4.8922, 3.1640, 4.6573, 2.8908, 1.9752], device='cuda:3'), covar=tensor([0.0394, 0.0307, 0.0617, 0.0177, 0.1250, 0.0208, 0.1333, 0.1576], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0175, 0.0263, 0.0168, 0.0223, 0.0160, 0.0232, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:57:28,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95321.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:57:37,491 INFO [train2.py:809] (3/4) Epoch 24, batch 3700, loss[ctc_loss=0.06792, att_loss=0.2222, loss=0.1913, over 15348.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01261, over 35.00 utterances.], tot_loss[ctc_loss=0.06948, att_loss=0.2336, loss=0.2008, over 3255877.94 frames. utt_duration=1238 frames, utt_pad_proportion=0.05917, over 10528.66 utterances.], batch size: 35, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:57:39,374 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1432, 3.8414, 3.2627, 3.5066, 3.9927, 3.7482, 3.0982, 4.2396], device='cuda:3'), covar=tensor([0.1017, 0.0515, 0.1156, 0.0671, 0.0739, 0.0704, 0.0835, 0.0528], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0225, 0.0228, 0.0207, 0.0286, 0.0246, 0.0204, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 04:57:40,886 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:57:56,742 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95339.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 04:58:07,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-09 04:58:11,159 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 1.823e+02 2.114e+02 2.643e+02 5.105e+02, threshold=4.228e+02, percent-clipped=2.0 2023-03-09 04:58:29,571 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6111, 2.6310, 5.0211, 3.9640, 3.0745, 4.4036, 4.8803, 4.6956], device='cuda:3'), covar=tensor([0.0278, 0.1641, 0.0203, 0.0879, 0.1747, 0.0234, 0.0164, 0.0290], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0244, 0.0206, 0.0321, 0.0268, 0.0225, 0.0194, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 04:58:58,644 INFO [train2.py:809] (3/4) Epoch 24, batch 3750, loss[ctc_loss=0.0625, att_loss=0.226, loss=0.1933, over 16108.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.007403, over 42.00 utterances.], tot_loss[ctc_loss=0.06921, att_loss=0.2332, loss=0.2004, over 3257059.29 frames. utt_duration=1239 frames, utt_pad_proportion=0.05927, over 10525.57 utterances.], batch size: 42, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 04:59:01,980 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8425, 2.6501, 2.3671, 2.4495, 2.8744, 2.5804, 2.5179, 2.9267], device='cuda:3'), covar=tensor([0.1691, 0.2201, 0.1966, 0.1543, 0.1402, 0.1388, 0.2200, 0.1247], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0135, 0.0131, 0.0123, 0.0138, 0.0119, 0.0143, 0.0118], device='cuda:3'), out_proj_covar=tensor([1.0032e-04, 1.0562e-04, 1.0596e-04, 9.6004e-05, 1.0365e-04, 9.6280e-05, 1.0887e-04, 9.3166e-05], device='cuda:3') 2023-03-09 04:59:35,854 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95400.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:00:19,266 INFO [train2.py:809] (3/4) Epoch 24, batch 3800, loss[ctc_loss=0.07027, att_loss=0.2322, loss=0.1998, over 16334.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005943, over 45.00 utterances.], tot_loss[ctc_loss=0.06994, att_loss=0.2338, loss=0.201, over 3259735.60 frames. utt_duration=1245 frames, utt_pad_proportion=0.05819, over 10484.99 utterances.], batch size: 45, lr: 4.44e-03, grad_scale: 8.0 2023-03-09 05:00:45,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8513, 4.9956, 4.6086, 2.7215, 4.7877, 4.6002, 4.1931, 2.5121], device='cuda:3'), covar=tensor([0.0194, 0.0099, 0.0346, 0.1245, 0.0116, 0.0227, 0.0393, 0.1646], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0105, 0.0108, 0.0114, 0.0088, 0.0117, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:00:52,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 1.978e+02 2.351e+02 3.015e+02 8.965e+02, threshold=4.702e+02, percent-clipped=6.0 2023-03-09 05:01:13,128 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95461.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 05:01:32,560 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8432, 5.1360, 4.4924, 5.2970, 4.6958, 4.7968, 5.2533, 4.9786], device='cuda:3'), covar=tensor([0.0622, 0.0319, 0.1178, 0.0326, 0.0390, 0.0338, 0.0256, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0324, 0.0364, 0.0353, 0.0325, 0.0239, 0.0305, 0.0286], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 05:01:38,903 INFO [train2.py:809] (3/4) Epoch 24, batch 3850, loss[ctc_loss=0.06673, att_loss=0.2185, loss=0.1881, over 16395.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007001, over 44.00 utterances.], tot_loss[ctc_loss=0.0704, att_loss=0.2341, loss=0.2014, over 3254368.36 frames. utt_duration=1244 frames, utt_pad_proportion=0.05951, over 10479.92 utterances.], batch size: 44, lr: 4.43e-03, grad_scale: 8.0 2023-03-09 05:02:00,326 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5253, 4.5734, 4.6006, 4.5620, 5.0561, 4.4015, 4.5770, 2.3938], device='cuda:3'), covar=tensor([0.0238, 0.0314, 0.0316, 0.0280, 0.0804, 0.0257, 0.0336, 0.1800], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0202, 0.0200, 0.0217, 0.0374, 0.0172, 0.0190, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:02:18,381 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8294, 5.2791, 5.1432, 5.3017, 5.3290, 4.9674, 3.9474, 5.3363], device='cuda:3'), covar=tensor([0.0123, 0.0102, 0.0128, 0.0072, 0.0099, 0.0132, 0.0581, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0092, 0.0115, 0.0071, 0.0079, 0.0089, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:02:54,835 INFO [train2.py:809] (3/4) Epoch 24, batch 3900, loss[ctc_loss=0.06345, att_loss=0.2311, loss=0.1976, over 16192.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.005599, over 41.00 utterances.], tot_loss[ctc_loss=0.07044, att_loss=0.2339, loss=0.2012, over 3255436.44 frames. utt_duration=1266 frames, utt_pad_proportion=0.05438, over 10297.35 utterances.], batch size: 41, lr: 4.43e-03, grad_scale: 8.0 2023-03-09 05:03:26,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.865e+02 2.329e+02 2.791e+02 6.016e+02, threshold=4.658e+02, percent-clipped=3.0 2023-03-09 05:03:42,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1426, 5.1042, 4.9475, 2.5504, 2.0535, 3.1287, 2.5447, 3.9194], device='cuda:3'), covar=tensor([0.0745, 0.0384, 0.0313, 0.4967, 0.5606, 0.2317, 0.3720, 0.1853], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0288, 0.0274, 0.0250, 0.0342, 0.0336, 0.0260, 0.0374], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 05:04:10,396 INFO [train2.py:809] (3/4) Epoch 24, batch 3950, loss[ctc_loss=0.04984, att_loss=0.2145, loss=0.1816, over 15873.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009134, over 39.00 utterances.], tot_loss[ctc_loss=0.07063, att_loss=0.2341, loss=0.2014, over 3259373.69 frames. utt_duration=1256 frames, utt_pad_proportion=0.05668, over 10394.51 utterances.], batch size: 39, lr: 4.43e-03, grad_scale: 8.0 2023-03-09 05:04:33,455 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0776, 4.3063, 4.1735, 4.6262, 2.8651, 4.4545, 2.7248, 1.7742], device='cuda:3'), covar=tensor([0.0435, 0.0264, 0.0815, 0.0195, 0.1581, 0.0214, 0.1533, 0.1754], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0175, 0.0262, 0.0168, 0.0222, 0.0160, 0.0231, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:05:25,990 INFO [train2.py:809] (3/4) Epoch 25, batch 0, loss[ctc_loss=0.06323, att_loss=0.2295, loss=0.1963, over 16426.00 frames. utt_duration=1495 frames, utt_pad_proportion=0.00606, over 44.00 utterances.], tot_loss[ctc_loss=0.06323, att_loss=0.2295, loss=0.1963, over 16426.00 frames. utt_duration=1495 frames, utt_pad_proportion=0.00606, over 44.00 utterances.], batch size: 44, lr: 4.34e-03, grad_scale: 8.0 2023-03-09 05:05:25,990 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 05:05:38,244 INFO [train2.py:843] (3/4) Epoch 25, validation: ctc_loss=0.04004, att_loss=0.2344, loss=0.1955, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 05:05:38,245 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 05:05:46,413 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95616.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:05:48,098 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95617.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:05:58,881 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95624.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:06:05,680 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3198, 3.0493, 3.2280, 4.4028, 3.9664, 3.8666, 2.9010, 2.2883], device='cuda:3'), covar=tensor([0.0773, 0.1902, 0.0925, 0.0601, 0.0804, 0.0516, 0.1635, 0.2217], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0219, 0.0186, 0.0220, 0.0228, 0.0182, 0.0204, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:06:36,946 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.872e+02 2.311e+02 2.775e+02 5.959e+02, threshold=4.621e+02, percent-clipped=4.0 2023-03-09 05:06:57,660 INFO [train2.py:809] (3/4) Epoch 25, batch 50, loss[ctc_loss=0.06191, att_loss=0.2321, loss=0.198, over 16614.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005857, over 47.00 utterances.], tot_loss[ctc_loss=0.07104, att_loss=0.2349, loss=0.2022, over 736973.12 frames. utt_duration=1292 frames, utt_pad_proportion=0.04245, over 2285.14 utterances.], batch size: 47, lr: 4.34e-03, grad_scale: 8.0 2023-03-09 05:07:18,875 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0521, 5.0256, 4.8069, 2.6970, 4.8621, 4.6509, 4.2135, 2.9273], device='cuda:3'), covar=tensor([0.0117, 0.0106, 0.0289, 0.1206, 0.0105, 0.0219, 0.0366, 0.1306], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0106, 0.0109, 0.0114, 0.0089, 0.0117, 0.0102, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:07:25,638 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95678.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:07:49,668 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.04 vs. limit=5.0 2023-03-09 05:07:52,047 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95695.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:08:00,072 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7466, 5.0674, 4.9081, 4.9374, 5.1877, 4.8550, 3.5101, 5.1145], device='cuda:3'), covar=tensor([0.0116, 0.0116, 0.0143, 0.0102, 0.0105, 0.0124, 0.0727, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0092, 0.0115, 0.0072, 0.0078, 0.0089, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:08:17,155 INFO [train2.py:809] (3/4) Epoch 25, batch 100, loss[ctc_loss=0.06738, att_loss=0.221, loss=0.1903, over 15953.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007057, over 41.00 utterances.], tot_loss[ctc_loss=0.06956, att_loss=0.2343, loss=0.2014, over 1304509.01 frames. utt_duration=1349 frames, utt_pad_proportion=0.02705, over 3871.92 utterances.], batch size: 41, lr: 4.34e-03, grad_scale: 8.0 2023-03-09 05:09:04,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-09 05:09:15,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 1.992e+02 2.344e+02 2.996e+02 8.490e+02, threshold=4.688e+02, percent-clipped=4.0 2023-03-09 05:09:36,921 INFO [train2.py:809] (3/4) Epoch 25, batch 150, loss[ctc_loss=0.06985, att_loss=0.2206, loss=0.1904, over 15984.00 frames. utt_duration=1600 frames, utt_pad_proportion=0.008674, over 40.00 utterances.], tot_loss[ctc_loss=0.06968, att_loss=0.234, loss=0.2012, over 1734457.01 frames. utt_duration=1297 frames, utt_pad_proportion=0.04546, over 5355.75 utterances.], batch size: 40, lr: 4.34e-03, grad_scale: 16.0 2023-03-09 05:09:37,274 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95761.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 05:10:55,141 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95809.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 05:10:58,467 INFO [train2.py:809] (3/4) Epoch 25, batch 200, loss[ctc_loss=0.08218, att_loss=0.2495, loss=0.216, over 17533.00 frames. utt_duration=1018 frames, utt_pad_proportion=0.0398, over 69.00 utterances.], tot_loss[ctc_loss=0.06965, att_loss=0.2347, loss=0.2017, over 2085172.75 frames. utt_duration=1288 frames, utt_pad_proportion=0.04031, over 6485.45 utterances.], batch size: 69, lr: 4.34e-03, grad_scale: 16.0 2023-03-09 05:11:01,188 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95812.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 05:11:56,661 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95847.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:11:57,807 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 1.872e+02 2.363e+02 3.018e+02 4.508e+02, threshold=4.726e+02, percent-clipped=0.0 2023-03-09 05:12:18,653 INFO [train2.py:809] (3/4) Epoch 25, batch 250, loss[ctc_loss=0.07356, att_loss=0.2541, loss=0.218, over 17401.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03316, over 63.00 utterances.], tot_loss[ctc_loss=0.06904, att_loss=0.2336, loss=0.2007, over 2342555.07 frames. utt_duration=1319 frames, utt_pad_proportion=0.03514, over 7113.43 utterances.], batch size: 63, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:12:38,754 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95873.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 05:13:34,996 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95908.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 05:13:39,387 INFO [train2.py:809] (3/4) Epoch 25, batch 300, loss[ctc_loss=0.0528, att_loss=0.2179, loss=0.1849, over 16119.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006191, over 42.00 utterances.], tot_loss[ctc_loss=0.06915, att_loss=0.2337, loss=0.2008, over 2546706.01 frames. utt_duration=1293 frames, utt_pad_proportion=0.04065, over 7887.16 utterances.], batch size: 42, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:13:48,184 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95916.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:14:00,746 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95924.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:14:38,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.855e+02 2.204e+02 2.784e+02 5.954e+02, threshold=4.408e+02, percent-clipped=3.0 2023-03-09 05:14:59,618 INFO [train2.py:809] (3/4) Epoch 25, batch 350, loss[ctc_loss=0.056, att_loss=0.2129, loss=0.1815, over 16145.00 frames. utt_duration=1539 frames, utt_pad_proportion=0.005033, over 42.00 utterances.], tot_loss[ctc_loss=0.06961, att_loss=0.2342, loss=0.2013, over 2708227.97 frames. utt_duration=1249 frames, utt_pad_proportion=0.05329, over 8685.91 utterances.], batch size: 42, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:15:04,978 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95964.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:15:17,323 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95972.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:15:19,653 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95973.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:15:30,460 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95980.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:15:54,767 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95995.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:16:24,359 INFO [train2.py:809] (3/4) Epoch 25, batch 400, loss[ctc_loss=0.05199, att_loss=0.2047, loss=0.1741, over 15517.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007192, over 36.00 utterances.], tot_loss[ctc_loss=0.06974, att_loss=0.2342, loss=0.2013, over 2828128.57 frames. utt_duration=1204 frames, utt_pad_proportion=0.06644, over 9404.95 utterances.], batch size: 36, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:17:12,334 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96041.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:17:15,165 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96043.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:17:22,747 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.820e+02 2.178e+02 2.552e+02 4.436e+02, threshold=4.357e+02, percent-clipped=1.0 2023-03-09 05:17:43,675 INFO [train2.py:809] (3/4) Epoch 25, batch 450, loss[ctc_loss=0.07191, att_loss=0.2118, loss=0.1839, over 15765.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.00839, over 38.00 utterances.], tot_loss[ctc_loss=0.0687, att_loss=0.2335, loss=0.2005, over 2932244.25 frames. utt_duration=1218 frames, utt_pad_proportion=0.06218, over 9639.59 utterances.], batch size: 38, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:18:14,116 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96080.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:19:03,649 INFO [train2.py:809] (3/4) Epoch 25, batch 500, loss[ctc_loss=0.06555, att_loss=0.2333, loss=0.1997, over 16694.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006186, over 46.00 utterances.], tot_loss[ctc_loss=0.06887, att_loss=0.2335, loss=0.2005, over 3010080.84 frames. utt_duration=1233 frames, utt_pad_proportion=0.05844, over 9778.05 utterances.], batch size: 46, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:19:24,211 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1770, 5.4581, 5.1272, 5.5336, 4.8841, 5.1080, 5.6345, 5.3293], device='cuda:3'), covar=tensor([0.0560, 0.0288, 0.0667, 0.0288, 0.0428, 0.0225, 0.0217, 0.0196], device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0330, 0.0372, 0.0361, 0.0331, 0.0244, 0.0313, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 05:19:50,696 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96141.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:20:00,864 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 1.791e+02 2.138e+02 2.685e+02 5.642e+02, threshold=4.276e+02, percent-clipped=2.0 2023-03-09 05:20:21,839 INFO [train2.py:809] (3/4) Epoch 25, batch 550, loss[ctc_loss=0.07462, att_loss=0.2385, loss=0.2058, over 16464.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007663, over 46.00 utterances.], tot_loss[ctc_loss=0.06926, att_loss=0.2339, loss=0.201, over 3077096.94 frames. utt_duration=1240 frames, utt_pad_proportion=0.05406, over 9938.34 utterances.], batch size: 46, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:20:28,369 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96165.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:20:32,703 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96168.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 05:21:27,369 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 05:21:39,961 INFO [train2.py:809] (3/4) Epoch 25, batch 600, loss[ctc_loss=0.07392, att_loss=0.2451, loss=0.2109, over 17257.00 frames. utt_duration=875 frames, utt_pad_proportion=0.08373, over 79.00 utterances.], tot_loss[ctc_loss=0.06894, att_loss=0.2338, loss=0.2008, over 3127206.49 frames. utt_duration=1256 frames, utt_pad_proportion=0.04855, over 9971.99 utterances.], batch size: 79, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:22:03,773 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96226.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:22:37,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-09 05:22:38,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.936e+02 2.176e+02 2.702e+02 4.909e+02, threshold=4.353e+02, percent-clipped=2.0 2023-03-09 05:22:59,841 INFO [train2.py:809] (3/4) Epoch 25, batch 650, loss[ctc_loss=0.0681, att_loss=0.2272, loss=0.1954, over 15780.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.008055, over 38.00 utterances.], tot_loss[ctc_loss=0.06837, att_loss=0.2335, loss=0.2005, over 3163168.49 frames. utt_duration=1255 frames, utt_pad_proportion=0.0499, over 10096.44 utterances.], batch size: 38, lr: 4.33e-03, grad_scale: 16.0 2023-03-09 05:23:19,349 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96273.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:23:30,846 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9278, 5.1632, 5.0433, 5.2294, 5.2482, 4.9097, 3.6114, 5.2341], device='cuda:3'), covar=tensor([0.0101, 0.0115, 0.0122, 0.0075, 0.0098, 0.0109, 0.0670, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0092, 0.0115, 0.0072, 0.0079, 0.0089, 0.0106, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:24:19,878 INFO [train2.py:809] (3/4) Epoch 25, batch 700, loss[ctc_loss=0.06299, att_loss=0.2353, loss=0.2008, over 16483.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005499, over 46.00 utterances.], tot_loss[ctc_loss=0.06759, att_loss=0.2326, loss=0.1996, over 3191901.29 frames. utt_duration=1277 frames, utt_pad_proportion=0.04395, over 10009.41 utterances.], batch size: 46, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:24:36,224 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96321.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:24:40,968 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:24:59,830 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96336.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:25:18,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 1.804e+02 2.150e+02 2.706e+02 4.503e+02, threshold=4.300e+02, percent-clipped=0.0 2023-03-09 05:25:38,747 INFO [train2.py:809] (3/4) Epoch 25, batch 750, loss[ctc_loss=0.05478, att_loss=0.2098, loss=0.1788, over 16005.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007528, over 40.00 utterances.], tot_loss[ctc_loss=0.0684, att_loss=0.2323, loss=0.1995, over 3199939.52 frames. utt_duration=1272 frames, utt_pad_proportion=0.04783, over 10078.17 utterances.], batch size: 40, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:26:17,539 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96385.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:26:25,334 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96390.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:26:58,843 INFO [train2.py:809] (3/4) Epoch 25, batch 800, loss[ctc_loss=0.0755, att_loss=0.2397, loss=0.2069, over 16466.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007467, over 46.00 utterances.], tot_loss[ctc_loss=0.06867, att_loss=0.2331, loss=0.2002, over 3226866.92 frames. utt_duration=1269 frames, utt_pad_proportion=0.04566, over 10183.27 utterances.], batch size: 46, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:27:38,860 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96436.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:27:57,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.925e+02 2.269e+02 2.805e+02 5.385e+02, threshold=4.538e+02, percent-clipped=5.0 2023-03-09 05:28:02,664 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96451.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:28:19,553 INFO [train2.py:809] (3/4) Epoch 25, batch 850, loss[ctc_loss=0.07602, att_loss=0.2461, loss=0.2121, over 17047.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.009361, over 52.00 utterances.], tot_loss[ctc_loss=0.06905, att_loss=0.2333, loss=0.2005, over 3241263.96 frames. utt_duration=1266 frames, utt_pad_proportion=0.04587, over 10254.47 utterances.], batch size: 52, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:28:31,883 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96468.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 05:29:16,822 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0018, 4.4542, 4.6457, 4.7482, 2.9509, 4.4895, 3.1938, 1.8647], device='cuda:3'), covar=tensor([0.0419, 0.0277, 0.0534, 0.0207, 0.1481, 0.0211, 0.1202, 0.1690], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0177, 0.0265, 0.0170, 0.0224, 0.0162, 0.0233, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:29:18,098 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7672, 6.0276, 5.4127, 5.7388, 5.6764, 5.1475, 5.4228, 5.2681], device='cuda:3'), covar=tensor([0.1206, 0.0951, 0.0970, 0.0867, 0.0996, 0.1661, 0.2375, 0.2133], device='cuda:3'), in_proj_covar=tensor([0.0545, 0.0627, 0.0476, 0.0469, 0.0443, 0.0481, 0.0627, 0.0541], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:29:28,037 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96503.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:29:41,370 INFO [train2.py:809] (3/4) Epoch 25, batch 900, loss[ctc_loss=0.07258, att_loss=0.2549, loss=0.2184, over 17032.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.01015, over 52.00 utterances.], tot_loss[ctc_loss=0.06925, att_loss=0.2333, loss=0.2005, over 3256647.31 frames. utt_duration=1272 frames, utt_pad_proportion=0.04392, over 10254.13 utterances.], batch size: 52, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:29:49,298 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96516.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 05:29:57,215 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96521.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:30:40,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 1.959e+02 2.352e+02 2.820e+02 9.770e+02, threshold=4.704e+02, percent-clipped=4.0 2023-03-09 05:30:45,076 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96551.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:31:03,292 INFO [train2.py:809] (3/4) Epoch 25, batch 950, loss[ctc_loss=0.07168, att_loss=0.2308, loss=0.199, over 16025.00 frames. utt_duration=1604 frames, utt_pad_proportion=0.006656, over 40.00 utterances.], tot_loss[ctc_loss=0.06975, att_loss=0.2339, loss=0.2011, over 3259942.94 frames. utt_duration=1273 frames, utt_pad_proportion=0.04425, over 10257.13 utterances.], batch size: 40, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:31:19,912 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3673, 2.8515, 3.6708, 3.1077, 3.4942, 4.4874, 4.3696, 3.2871], device='cuda:3'), covar=tensor([0.0376, 0.1803, 0.1138, 0.1178, 0.1017, 0.0977, 0.0573, 0.1170], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0247, 0.0285, 0.0221, 0.0265, 0.0375, 0.0267, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:32:13,827 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8495, 5.0963, 4.7498, 5.1689, 4.5806, 4.8376, 5.2460, 5.0235], device='cuda:3'), covar=tensor([0.0631, 0.0370, 0.0799, 0.0357, 0.0467, 0.0286, 0.0240, 0.0211], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0329, 0.0370, 0.0360, 0.0329, 0.0241, 0.0312, 0.0291], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 05:32:25,902 INFO [train2.py:809] (3/4) Epoch 25, batch 1000, loss[ctc_loss=0.04191, att_loss=0.2087, loss=0.1753, over 16020.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006676, over 40.00 utterances.], tot_loss[ctc_loss=0.06933, att_loss=0.2336, loss=0.2007, over 3268571.82 frames. utt_duration=1261 frames, utt_pad_proportion=0.04612, over 10378.79 utterances.], batch size: 40, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:33:05,496 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96636.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:33:23,586 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.809e+02 2.155e+02 2.476e+02 5.052e+02, threshold=4.310e+02, percent-clipped=2.0 2023-03-09 05:33:46,052 INFO [train2.py:809] (3/4) Epoch 25, batch 1050, loss[ctc_loss=0.07286, att_loss=0.235, loss=0.2026, over 17056.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.00963, over 53.00 utterances.], tot_loss[ctc_loss=0.06949, att_loss=0.2344, loss=0.2014, over 3277340.08 frames. utt_duration=1246 frames, utt_pad_proportion=0.04921, over 10537.08 utterances.], batch size: 53, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:34:17,058 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96680.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:34:23,076 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96684.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:35:05,672 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9105, 5.1435, 5.0820, 5.0352, 5.1983, 5.1707, 4.7580, 4.6514], device='cuda:3'), covar=tensor([0.1108, 0.0567, 0.0336, 0.0602, 0.0303, 0.0377, 0.0463, 0.0377], device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0379, 0.0369, 0.0379, 0.0440, 0.0448, 0.0374, 0.0410], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:35:06,971 INFO [train2.py:809] (3/4) Epoch 25, batch 1100, loss[ctc_loss=0.05903, att_loss=0.2428, loss=0.2061, over 17314.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01138, over 55.00 utterances.], tot_loss[ctc_loss=0.06892, att_loss=0.234, loss=0.201, over 3274626.75 frames. utt_duration=1253 frames, utt_pad_proportion=0.05025, over 10467.05 utterances.], batch size: 55, lr: 4.32e-03, grad_scale: 16.0 2023-03-09 05:35:07,277 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3330, 5.3459, 5.1490, 3.6489, 5.1727, 4.8842, 4.6684, 3.3722], device='cuda:3'), covar=tensor([0.0114, 0.0081, 0.0249, 0.0694, 0.0086, 0.0166, 0.0260, 0.1064], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0104, 0.0107, 0.0112, 0.0088, 0.0115, 0.0101, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:35:47,551 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96736.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:36:03,178 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96746.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:36:06,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.003e+02 2.274e+02 2.730e+02 5.677e+02, threshold=4.548e+02, percent-clipped=3.0 2023-03-09 05:36:28,283 INFO [train2.py:809] (3/4) Epoch 25, batch 1150, loss[ctc_loss=0.05738, att_loss=0.2345, loss=0.199, over 17042.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.01011, over 53.00 utterances.], tot_loss[ctc_loss=0.06903, att_loss=0.2336, loss=0.2006, over 3270446.57 frames. utt_duration=1243 frames, utt_pad_proportion=0.05462, over 10540.31 utterances.], batch size: 53, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:36:53,000 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96776.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:37:05,051 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96784.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:37:48,671 INFO [train2.py:809] (3/4) Epoch 25, batch 1200, loss[ctc_loss=0.05218, att_loss=0.228, loss=0.1928, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006614, over 48.00 utterances.], tot_loss[ctc_loss=0.06948, att_loss=0.2335, loss=0.2007, over 3263402.84 frames. utt_duration=1245 frames, utt_pad_proportion=0.05574, over 10497.19 utterances.], batch size: 48, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:38:05,775 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96821.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:38:30,789 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96837.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:38:47,154 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9616, 6.1675, 5.5958, 5.8518, 5.7933, 5.3526, 5.6621, 5.2849], device='cuda:3'), covar=tensor([0.1113, 0.0773, 0.1005, 0.0787, 0.0838, 0.1496, 0.2056, 0.2441], device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0624, 0.0476, 0.0469, 0.0441, 0.0480, 0.0625, 0.0538], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:38:48,487 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 2.088e+02 2.358e+02 2.925e+02 6.715e+02, threshold=4.715e+02, percent-clipped=3.0 2023-03-09 05:39:08,867 INFO [train2.py:809] (3/4) Epoch 25, batch 1250, loss[ctc_loss=0.0615, att_loss=0.2086, loss=0.1791, over 15519.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.007619, over 36.00 utterances.], tot_loss[ctc_loss=0.06874, att_loss=0.2328, loss=0.2, over 3268654.04 frames. utt_duration=1255 frames, utt_pad_proportion=0.05232, over 10430.63 utterances.], batch size: 36, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:39:22,158 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96869.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:40:28,114 INFO [train2.py:809] (3/4) Epoch 25, batch 1300, loss[ctc_loss=0.07261, att_loss=0.2383, loss=0.2051, over 17060.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.00947, over 53.00 utterances.], tot_loss[ctc_loss=0.06902, att_loss=0.2333, loss=0.2005, over 3275037.72 frames. utt_duration=1254 frames, utt_pad_proportion=0.05167, over 10458.66 utterances.], batch size: 53, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:41:27,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.796e+02 2.102e+02 2.487e+02 3.780e+02, threshold=4.203e+02, percent-clipped=0.0 2023-03-09 05:41:47,817 INFO [train2.py:809] (3/4) Epoch 25, batch 1350, loss[ctc_loss=0.05603, att_loss=0.225, loss=0.1912, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007285, over 43.00 utterances.], tot_loss[ctc_loss=0.06923, att_loss=0.2339, loss=0.201, over 3286065.96 frames. utt_duration=1258 frames, utt_pad_proportion=0.04833, over 10460.36 utterances.], batch size: 43, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:42:18,030 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96980.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:42:31,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1201, 5.4384, 5.6455, 5.4396, 5.5937, 6.0696, 5.3383, 6.1484], device='cuda:3'), covar=tensor([0.0716, 0.0699, 0.0835, 0.1375, 0.1786, 0.0912, 0.0688, 0.0627], device='cuda:3'), in_proj_covar=tensor([0.0912, 0.0525, 0.0639, 0.0685, 0.0914, 0.0662, 0.0520, 0.0647], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:42:51,017 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97000.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:43:07,415 INFO [train2.py:809] (3/4) Epoch 25, batch 1400, loss[ctc_loss=0.04122, att_loss=0.2138, loss=0.1792, over 16262.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007635, over 43.00 utterances.], tot_loss[ctc_loss=0.06888, att_loss=0.2337, loss=0.2007, over 3284094.37 frames. utt_duration=1265 frames, utt_pad_proportion=0.04645, over 10395.01 utterances.], batch size: 43, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:43:23,524 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4611, 3.1049, 3.6649, 4.5163, 4.0633, 3.8858, 2.9026, 2.3469], device='cuda:3'), covar=tensor([0.0778, 0.1886, 0.0736, 0.0605, 0.0839, 0.0600, 0.1840, 0.2220], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0223, 0.0189, 0.0224, 0.0233, 0.0188, 0.0209, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-09 05:43:34,018 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97028.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:43:40,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2379, 2.7687, 3.2748, 4.3412, 3.9000, 3.8179, 2.7658, 2.1405], device='cuda:3'), covar=tensor([0.0835, 0.1973, 0.0882, 0.0542, 0.0892, 0.0501, 0.1765, 0.2211], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0223, 0.0190, 0.0224, 0.0234, 0.0188, 0.0210, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-09 05:43:57,636 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9095, 3.6262, 3.5944, 3.0525, 3.7364, 3.7146, 3.7355, 2.5989], device='cuda:3'), covar=tensor([0.0931, 0.1230, 0.1959, 0.3383, 0.0739, 0.2369, 0.0662, 0.3481], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0198, 0.0210, 0.0264, 0.0173, 0.0272, 0.0196, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:44:02,826 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97046.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:44:05,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.953e+02 2.309e+02 2.789e+02 6.785e+02, threshold=4.619e+02, percent-clipped=7.0 2023-03-09 05:44:26,322 INFO [train2.py:809] (3/4) Epoch 25, batch 1450, loss[ctc_loss=0.06204, att_loss=0.212, loss=0.182, over 15633.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008993, over 37.00 utterances.], tot_loss[ctc_loss=0.06883, att_loss=0.2335, loss=0.2005, over 3282807.40 frames. utt_duration=1249 frames, utt_pad_proportion=0.05146, over 10523.47 utterances.], batch size: 37, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:44:27,326 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97061.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:45:17,669 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97094.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:45:44,477 INFO [train2.py:809] (3/4) Epoch 25, batch 1500, loss[ctc_loss=0.0508, att_loss=0.2098, loss=0.178, over 15937.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.007591, over 41.00 utterances.], tot_loss[ctc_loss=0.06819, att_loss=0.2332, loss=0.2002, over 3283733.51 frames. utt_duration=1262 frames, utt_pad_proportion=0.04859, over 10421.23 utterances.], batch size: 41, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:45:49,907 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3966, 5.0194, 5.1021, 5.0888, 4.9386, 5.0678, 4.7768, 4.6163], device='cuda:3'), covar=tensor([0.1830, 0.0823, 0.0423, 0.0608, 0.0808, 0.0466, 0.0511, 0.0424], device='cuda:3'), in_proj_covar=tensor([0.0535, 0.0375, 0.0365, 0.0374, 0.0436, 0.0442, 0.0370, 0.0406], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:46:00,639 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0111, 3.5289, 3.5152, 3.0110, 3.6301, 3.7431, 3.6538, 2.5523], device='cuda:3'), covar=tensor([0.0887, 0.1295, 0.2095, 0.2884, 0.1162, 0.1998, 0.1223, 0.3564], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0199, 0.0213, 0.0267, 0.0175, 0.0275, 0.0199, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:46:17,401 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97132.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:46:33,261 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0160, 5.3337, 4.8696, 5.3783, 4.7735, 4.9787, 5.4483, 5.2111], device='cuda:3'), covar=tensor([0.0569, 0.0338, 0.0768, 0.0383, 0.0393, 0.0284, 0.0222, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0330, 0.0371, 0.0362, 0.0328, 0.0244, 0.0314, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 05:46:43,664 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 2.011e+02 2.273e+02 2.604e+02 5.523e+02, threshold=4.546e+02, percent-clipped=2.0 2023-03-09 05:47:03,718 INFO [train2.py:809] (3/4) Epoch 25, batch 1550, loss[ctc_loss=0.06076, att_loss=0.2257, loss=0.1927, over 16265.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.008011, over 43.00 utterances.], tot_loss[ctc_loss=0.06768, att_loss=0.2325, loss=0.1996, over 3271813.64 frames. utt_duration=1275 frames, utt_pad_proportion=0.04777, over 10273.80 utterances.], batch size: 43, lr: 4.31e-03, grad_scale: 16.0 2023-03-09 05:48:23,317 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8742, 3.8298, 3.2384, 3.3354, 3.9341, 3.5671, 2.5747, 4.2492], device='cuda:3'), covar=tensor([0.1156, 0.0467, 0.0990, 0.0814, 0.0788, 0.0770, 0.1190, 0.0554], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0223, 0.0227, 0.0205, 0.0285, 0.0244, 0.0201, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 05:48:24,498 INFO [train2.py:809] (3/4) Epoch 25, batch 1600, loss[ctc_loss=0.07298, att_loss=0.2312, loss=0.1996, over 16332.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.00605, over 45.00 utterances.], tot_loss[ctc_loss=0.06792, att_loss=0.2327, loss=0.1998, over 3271229.84 frames. utt_duration=1265 frames, utt_pad_proportion=0.05048, over 10354.23 utterances.], batch size: 45, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:49:22,641 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.951e+02 2.275e+02 2.852e+02 6.769e+02, threshold=4.549e+02, percent-clipped=7.0 2023-03-09 05:49:42,297 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97260.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:49:43,461 INFO [train2.py:809] (3/4) Epoch 25, batch 1650, loss[ctc_loss=0.04938, att_loss=0.2046, loss=0.1735, over 14066.00 frames. utt_duration=1817 frames, utt_pad_proportion=0.05237, over 31.00 utterances.], tot_loss[ctc_loss=0.06859, att_loss=0.2336, loss=0.2006, over 3272578.20 frames. utt_duration=1238 frames, utt_pad_proportion=0.05654, over 10584.44 utterances.], batch size: 31, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:51:02,815 INFO [train2.py:809] (3/4) Epoch 25, batch 1700, loss[ctc_loss=0.0606, att_loss=0.2421, loss=0.2058, over 16781.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005771, over 48.00 utterances.], tot_loss[ctc_loss=0.06798, att_loss=0.2336, loss=0.2005, over 3282760.55 frames. utt_duration=1243 frames, utt_pad_proportion=0.05281, over 10580.33 utterances.], batch size: 48, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:51:03,334 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5426, 2.8169, 4.9074, 4.0093, 3.1152, 4.3551, 4.7941, 4.6995], device='cuda:3'), covar=tensor([0.0275, 0.1419, 0.0219, 0.0779, 0.1615, 0.0253, 0.0176, 0.0291], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0244, 0.0207, 0.0320, 0.0265, 0.0227, 0.0197, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:51:19,443 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97321.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:51:52,655 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97342.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:52:01,002 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97347.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:52:01,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2023-03-09 05:52:02,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.753e+02 2.134e+02 2.732e+02 7.784e+02, threshold=4.268e+02, percent-clipped=3.0 2023-03-09 05:52:15,791 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97356.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:52:23,288 INFO [train2.py:809] (3/4) Epoch 25, batch 1750, loss[ctc_loss=0.06946, att_loss=0.2346, loss=0.2016, over 16623.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005454, over 47.00 utterances.], tot_loss[ctc_loss=0.06772, att_loss=0.2333, loss=0.2001, over 3281824.18 frames. utt_duration=1241 frames, utt_pad_proportion=0.05363, over 10589.64 utterances.], batch size: 47, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:52:44,632 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97374.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:53:08,412 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0778, 5.1053, 4.8404, 3.0336, 4.8190, 4.6690, 4.2273, 2.9390], device='cuda:3'), covar=tensor([0.0109, 0.0104, 0.0253, 0.1012, 0.0108, 0.0219, 0.0361, 0.1322], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0103, 0.0105, 0.0109, 0.0086, 0.0113, 0.0099, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 05:53:25,275 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8307, 2.2922, 2.8049, 2.6135, 2.7643, 2.7325, 2.4412, 3.1261], device='cuda:3'), covar=tensor([0.1778, 0.2946, 0.1752, 0.1735, 0.2071, 0.1503, 0.2410, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0138, 0.0132, 0.0125, 0.0141, 0.0123, 0.0146, 0.0120], device='cuda:3'), out_proj_covar=tensor([1.0247e-04, 1.0837e-04, 1.0750e-04, 9.7661e-05, 1.0662e-04, 9.8720e-05, 1.1077e-04, 9.5536e-05], device='cuda:3') 2023-03-09 05:53:31,691 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97403.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 05:53:39,279 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97408.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:53:44,076 INFO [train2.py:809] (3/4) Epoch 25, batch 1800, loss[ctc_loss=0.07484, att_loss=0.2565, loss=0.2201, over 17052.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009042, over 53.00 utterances.], tot_loss[ctc_loss=0.06776, att_loss=0.2333, loss=0.2002, over 3279102.70 frames. utt_duration=1219 frames, utt_pad_proportion=0.05992, over 10771.57 utterances.], batch size: 53, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:53:46,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-09 05:53:52,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-09 05:54:13,904 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8764, 3.7760, 3.2134, 3.2760, 4.0381, 3.5539, 2.7553, 4.1716], device='cuda:3'), covar=tensor([0.1136, 0.0501, 0.1055, 0.0792, 0.0620, 0.0769, 0.1042, 0.0436], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0226, 0.0229, 0.0207, 0.0288, 0.0246, 0.0203, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 05:54:16,750 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97432.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:54:21,361 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97435.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:54:43,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.842e+02 2.140e+02 2.481e+02 6.088e+02, threshold=4.279e+02, percent-clipped=3.0 2023-03-09 05:55:03,731 INFO [train2.py:809] (3/4) Epoch 25, batch 1850, loss[ctc_loss=0.05475, att_loss=0.2218, loss=0.1884, over 16411.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006906, over 44.00 utterances.], tot_loss[ctc_loss=0.06747, att_loss=0.233, loss=0.1999, over 3275839.18 frames. utt_duration=1234 frames, utt_pad_proportion=0.05853, over 10628.46 utterances.], batch size: 44, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:55:24,889 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.21 vs. limit=5.0 2023-03-09 05:55:33,379 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97480.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:56:23,874 INFO [train2.py:809] (3/4) Epoch 25, batch 1900, loss[ctc_loss=0.07297, att_loss=0.2254, loss=0.1949, over 15760.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.00872, over 38.00 utterances.], tot_loss[ctc_loss=0.06841, att_loss=0.2335, loss=0.2004, over 3270408.00 frames. utt_duration=1212 frames, utt_pad_proportion=0.0669, over 10808.99 utterances.], batch size: 38, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:56:27,030 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9792, 5.2857, 5.5254, 5.4079, 5.4791, 6.0001, 5.2217, 6.0379], device='cuda:3'), covar=tensor([0.0705, 0.0718, 0.0866, 0.1308, 0.1774, 0.0822, 0.0813, 0.0604], device='cuda:3'), in_proj_covar=tensor([0.0908, 0.0523, 0.0635, 0.0681, 0.0909, 0.0656, 0.0514, 0.0638], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:56:48,351 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 05:57:07,492 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5578, 3.0572, 3.6989, 3.1568, 3.4997, 4.6561, 4.4878, 3.3188], device='cuda:3'), covar=tensor([0.0338, 0.1707, 0.1135, 0.1400, 0.1109, 0.0809, 0.0527, 0.1221], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0248, 0.0288, 0.0224, 0.0268, 0.0378, 0.0267, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 05:57:23,540 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 1.849e+02 2.257e+02 2.657e+02 5.129e+02, threshold=4.513e+02, percent-clipped=1.0 2023-03-09 05:57:35,284 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97555.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:57:44,362 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-09 05:57:44,863 INFO [train2.py:809] (3/4) Epoch 25, batch 1950, loss[ctc_loss=0.07672, att_loss=0.2415, loss=0.2085, over 16862.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008022, over 49.00 utterances.], tot_loss[ctc_loss=0.06899, att_loss=0.2341, loss=0.201, over 3277218.57 frames. utt_duration=1218 frames, utt_pad_proportion=0.06246, over 10779.49 utterances.], batch size: 49, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:57:48,626 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2854, 4.3498, 4.4441, 4.3562, 4.8787, 4.2708, 4.2424, 2.5288], device='cuda:3'), covar=tensor([0.0276, 0.0370, 0.0369, 0.0331, 0.0798, 0.0288, 0.0385, 0.1824], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0206, 0.0204, 0.0222, 0.0380, 0.0178, 0.0195, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 05:59:06,992 INFO [train2.py:809] (3/4) Epoch 25, batch 2000, loss[ctc_loss=0.05726, att_loss=0.2299, loss=0.1954, over 16262.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007864, over 43.00 utterances.], tot_loss[ctc_loss=0.06873, att_loss=0.234, loss=0.201, over 3275755.12 frames. utt_duration=1219 frames, utt_pad_proportion=0.06144, over 10764.47 utterances.], batch size: 43, lr: 4.30e-03, grad_scale: 16.0 2023-03-09 05:59:15,370 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:59:15,557 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 05:59:22,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.03 vs. limit=5.0 2023-03-09 06:00:08,132 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 1.752e+02 2.147e+02 2.537e+02 5.924e+02, threshold=4.294e+02, percent-clipped=3.0 2023-03-09 06:00:21,280 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97656.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:00:29,817 INFO [train2.py:809] (3/4) Epoch 25, batch 2050, loss[ctc_loss=0.06428, att_loss=0.2255, loss=0.1933, over 16531.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006692, over 45.00 utterances.], tot_loss[ctc_loss=0.06853, att_loss=0.2339, loss=0.2008, over 3281717.87 frames. utt_duration=1246 frames, utt_pad_proportion=0.05353, over 10550.36 utterances.], batch size: 45, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:00:50,814 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97674.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 06:01:30,793 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97698.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 06:01:38,578 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97703.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:01:40,008 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:01:50,803 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-09 06:01:51,393 INFO [train2.py:809] (3/4) Epoch 25, batch 2100, loss[ctc_loss=0.06248, att_loss=0.2385, loss=0.2033, over 16889.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006421, over 49.00 utterances.], tot_loss[ctc_loss=0.0679, att_loss=0.2331, loss=0.2001, over 3272214.06 frames. utt_duration=1256 frames, utt_pad_proportion=0.05394, over 10435.01 utterances.], batch size: 49, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:02:20,980 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97730.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:02:29,535 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97735.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 06:02:50,717 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.928e+02 2.285e+02 2.765e+02 5.128e+02, threshold=4.569e+02, percent-clipped=4.0 2023-03-09 06:03:11,116 INFO [train2.py:809] (3/4) Epoch 25, batch 2150, loss[ctc_loss=0.07735, att_loss=0.2475, loss=0.2135, over 17424.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.0458, over 69.00 utterances.], tot_loss[ctc_loss=0.0688, att_loss=0.234, loss=0.201, over 3273610.45 frames. utt_duration=1241 frames, utt_pad_proportion=0.05724, over 10560.44 utterances.], batch size: 69, lr: 4.29e-03, grad_scale: 32.0 2023-03-09 06:03:11,610 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3652, 4.3925, 4.5758, 4.5139, 5.0561, 4.2845, 4.4043, 2.5500], device='cuda:3'), covar=tensor([0.0308, 0.0368, 0.0330, 0.0337, 0.0814, 0.0309, 0.0363, 0.1751], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0205, 0.0203, 0.0221, 0.0378, 0.0177, 0.0194, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:04:32,653 INFO [train2.py:809] (3/4) Epoch 25, batch 2200, loss[ctc_loss=0.04313, att_loss=0.1993, loss=0.1681, over 15762.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.009233, over 38.00 utterances.], tot_loss[ctc_loss=0.06809, att_loss=0.2337, loss=0.2006, over 3275643.62 frames. utt_duration=1239 frames, utt_pad_proportion=0.05739, over 10591.50 utterances.], batch size: 38, lr: 4.29e-03, grad_scale: 32.0 2023-03-09 06:04:40,790 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1342, 4.4260, 4.3083, 4.4289, 4.5163, 4.2237, 3.0674, 4.3975], device='cuda:3'), covar=tensor([0.0146, 0.0131, 0.0155, 0.0094, 0.0124, 0.0145, 0.0767, 0.0228], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0092, 0.0114, 0.0072, 0.0078, 0.0089, 0.0105, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:05:34,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 1.901e+02 2.356e+02 2.897e+02 7.922e+02, threshold=4.712e+02, percent-clipped=5.0 2023-03-09 06:05:53,826 INFO [train2.py:809] (3/4) Epoch 25, batch 2250, loss[ctc_loss=0.07003, att_loss=0.2294, loss=0.1975, over 16335.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005852, over 45.00 utterances.], tot_loss[ctc_loss=0.06918, att_loss=0.2344, loss=0.2013, over 3272960.76 frames. utt_duration=1188 frames, utt_pad_proportion=0.06986, over 11035.80 utterances.], batch size: 45, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:06:11,445 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5511, 3.0109, 3.5804, 4.5584, 4.1884, 3.9651, 3.1190, 2.4740], device='cuda:3'), covar=tensor([0.0695, 0.1765, 0.0772, 0.0538, 0.0725, 0.0507, 0.1382, 0.1944], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0219, 0.0187, 0.0220, 0.0230, 0.0185, 0.0205, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:06:37,501 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97888.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:06:59,415 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4104, 2.2344, 2.3341, 2.5091, 2.6954, 2.1883, 2.1301, 2.6124], device='cuda:3'), covar=tensor([0.1467, 0.2385, 0.1824, 0.1137, 0.1650, 0.1279, 0.1949, 0.1151], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0135, 0.0131, 0.0124, 0.0141, 0.0122, 0.0144, 0.0119], device='cuda:3'), out_proj_covar=tensor([1.0155e-04, 1.0653e-04, 1.0627e-04, 9.6685e-05, 1.0583e-04, 9.8170e-05, 1.0962e-04, 9.4630e-05], device='cuda:3') 2023-03-09 06:07:13,166 INFO [train2.py:809] (3/4) Epoch 25, batch 2300, loss[ctc_loss=0.08563, att_loss=0.2586, loss=0.224, over 17311.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01164, over 55.00 utterances.], tot_loss[ctc_loss=0.06912, att_loss=0.2348, loss=0.2016, over 3280307.50 frames. utt_duration=1202 frames, utt_pad_proportion=0.06444, over 10932.89 utterances.], batch size: 55, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:07:13,324 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:07:17,496 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-09 06:07:21,306 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97916.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:08:14,118 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 1.837e+02 2.120e+02 2.605e+02 6.957e+02, threshold=4.241e+02, percent-clipped=1.0 2023-03-09 06:08:15,190 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97949.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:08:33,503 INFO [train2.py:809] (3/4) Epoch 25, batch 2350, loss[ctc_loss=0.05665, att_loss=0.2305, loss=0.1957, over 17394.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.0447, over 69.00 utterances.], tot_loss[ctc_loss=0.06907, att_loss=0.2348, loss=0.2017, over 3288413.09 frames. utt_duration=1197 frames, utt_pad_proportion=0.06223, over 11004.04 utterances.], batch size: 69, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:08:38,282 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97964.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:09:31,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2023-03-09 06:09:34,060 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97998.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:09:45,870 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98003.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:09:57,929 INFO [train2.py:809] (3/4) Epoch 25, batch 2400, loss[ctc_loss=0.07882, att_loss=0.2492, loss=0.2151, over 17374.00 frames. utt_duration=1009 frames, utt_pad_proportion=0.04839, over 69.00 utterances.], tot_loss[ctc_loss=0.0686, att_loss=0.2349, loss=0.2016, over 3293254.16 frames. utt_duration=1197 frames, utt_pad_proportion=0.06112, over 11016.03 utterances.], batch size: 69, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:10:13,280 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2645, 2.7229, 3.2727, 4.3772, 3.9909, 3.8176, 2.8037, 2.3168], device='cuda:3'), covar=tensor([0.0870, 0.2240, 0.0993, 0.0503, 0.0798, 0.0521, 0.1815, 0.2280], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0218, 0.0188, 0.0220, 0.0230, 0.0184, 0.0205, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:10:29,509 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98030.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 06:10:29,585 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98030.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:10:38,088 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9444, 3.6029, 3.5797, 3.1656, 3.6732, 3.6598, 3.7313, 2.6786], device='cuda:3'), covar=tensor([0.0974, 0.1465, 0.1931, 0.2627, 0.1463, 0.1684, 0.0944, 0.3117], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0197, 0.0211, 0.0264, 0.0174, 0.0270, 0.0195, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:10:55,047 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98046.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:11:00,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 1.784e+02 2.263e+02 2.686e+02 5.862e+02, threshold=4.525e+02, percent-clipped=1.0 2023-03-09 06:11:03,272 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98051.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:11:18,651 INFO [train2.py:809] (3/4) Epoch 25, batch 2450, loss[ctc_loss=0.06477, att_loss=0.2434, loss=0.2077, over 17382.00 frames. utt_duration=881.7 frames, utt_pad_proportion=0.0758, over 79.00 utterances.], tot_loss[ctc_loss=0.06783, att_loss=0.2344, loss=0.2011, over 3294317.42 frames. utt_duration=1201 frames, utt_pad_proportion=0.05902, over 10983.15 utterances.], batch size: 79, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:11:46,655 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98078.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:12:31,658 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98106.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:12:37,210 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-09 06:12:39,097 INFO [train2.py:809] (3/4) Epoch 25, batch 2500, loss[ctc_loss=0.05842, att_loss=0.2309, loss=0.1964, over 16887.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006538, over 49.00 utterances.], tot_loss[ctc_loss=0.06786, att_loss=0.2347, loss=0.2014, over 3300129.88 frames. utt_duration=1194 frames, utt_pad_proportion=0.05861, over 11067.71 utterances.], batch size: 49, lr: 4.29e-03, grad_scale: 16.0 2023-03-09 06:13:21,319 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0871, 4.9988, 4.8344, 3.1011, 4.7747, 4.7492, 4.2497, 2.6861], device='cuda:3'), covar=tensor([0.0096, 0.0108, 0.0274, 0.0975, 0.0113, 0.0182, 0.0357, 0.1376], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0104, 0.0107, 0.0111, 0.0087, 0.0115, 0.0100, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:13:32,488 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1187, 5.4106, 4.9672, 5.4725, 4.8227, 5.0642, 5.5316, 5.3128], device='cuda:3'), covar=tensor([0.0569, 0.0283, 0.0766, 0.0314, 0.0427, 0.0228, 0.0230, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0331, 0.0374, 0.0364, 0.0332, 0.0244, 0.0315, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 06:13:40,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.428e+02 1.837e+02 2.156e+02 2.644e+02 5.272e+02, threshold=4.313e+02, percent-clipped=2.0 2023-03-09 06:13:45,686 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3299, 3.7779, 3.3009, 3.4384, 3.9800, 3.6794, 3.0991, 4.3179], device='cuda:3'), covar=tensor([0.0876, 0.0595, 0.1043, 0.0742, 0.0792, 0.0736, 0.0886, 0.0453], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0226, 0.0229, 0.0206, 0.0288, 0.0247, 0.0203, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 06:13:59,459 INFO [train2.py:809] (3/4) Epoch 25, batch 2550, loss[ctc_loss=0.09671, att_loss=0.2515, loss=0.2205, over 13806.00 frames. utt_duration=382.6 frames, utt_pad_proportion=0.3358, over 145.00 utterances.], tot_loss[ctc_loss=0.06763, att_loss=0.2341, loss=0.2008, over 3289802.87 frames. utt_duration=1207 frames, utt_pad_proportion=0.05847, over 10911.92 utterances.], batch size: 145, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:14:10,125 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98167.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:14:52,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-09 06:15:20,414 INFO [train2.py:809] (3/4) Epoch 25, batch 2600, loss[ctc_loss=0.07385, att_loss=0.2288, loss=0.1978, over 16628.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005125, over 47.00 utterances.], tot_loss[ctc_loss=0.06803, att_loss=0.2335, loss=0.2004, over 3283689.21 frames. utt_duration=1205 frames, utt_pad_proportion=0.06072, over 10915.69 utterances.], batch size: 47, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:15:20,728 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98211.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:15:42,687 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5062, 2.9106, 3.6774, 3.0391, 3.4133, 4.5932, 4.4256, 3.2067], device='cuda:3'), covar=tensor([0.0352, 0.1655, 0.1146, 0.1257, 0.1085, 0.0871, 0.0502, 0.1250], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0247, 0.0286, 0.0220, 0.0265, 0.0373, 0.0265, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:16:14,212 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98244.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:16:21,477 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 1.793e+02 2.177e+02 2.680e+02 7.649e+02, threshold=4.354e+02, percent-clipped=6.0 2023-03-09 06:16:37,160 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98259.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:16:40,777 INFO [train2.py:809] (3/4) Epoch 25, batch 2650, loss[ctc_loss=0.07675, att_loss=0.2372, loss=0.2051, over 17397.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03143, over 63.00 utterances.], tot_loss[ctc_loss=0.06779, att_loss=0.2333, loss=0.2002, over 3284161.90 frames. utt_duration=1219 frames, utt_pad_proportion=0.05831, over 10793.20 utterances.], batch size: 63, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:17:19,324 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4393, 3.9936, 3.3590, 3.6047, 4.1690, 3.9273, 3.3525, 4.5193], device='cuda:3'), covar=tensor([0.0827, 0.0500, 0.1088, 0.0704, 0.0682, 0.0665, 0.0777, 0.0413], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0223, 0.0227, 0.0204, 0.0285, 0.0245, 0.0201, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 06:18:00,626 INFO [train2.py:809] (3/4) Epoch 25, batch 2700, loss[ctc_loss=0.06767, att_loss=0.2357, loss=0.2021, over 16766.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006644, over 48.00 utterances.], tot_loss[ctc_loss=0.06756, att_loss=0.2329, loss=0.1998, over 3273833.26 frames. utt_duration=1204 frames, utt_pad_proportion=0.06374, over 10888.74 utterances.], batch size: 48, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:18:31,193 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98330.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 06:18:41,810 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4862, 2.8760, 3.4474, 4.5500, 4.1038, 3.9976, 3.0150, 2.4869], device='cuda:3'), covar=tensor([0.0720, 0.2065, 0.0847, 0.0530, 0.0863, 0.0476, 0.1499, 0.1990], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0218, 0.0186, 0.0220, 0.0230, 0.0184, 0.0204, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:19:00,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 1.887e+02 2.414e+02 2.992e+02 7.902e+02, threshold=4.828e+02, percent-clipped=2.0 2023-03-09 06:19:19,945 INFO [train2.py:809] (3/4) Epoch 25, batch 2750, loss[ctc_loss=0.05064, att_loss=0.2126, loss=0.1802, over 16274.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007508, over 43.00 utterances.], tot_loss[ctc_loss=0.0675, att_loss=0.2328, loss=0.1997, over 3275323.11 frames. utt_duration=1224 frames, utt_pad_proportion=0.05759, over 10718.73 utterances.], batch size: 43, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:19:34,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5668, 2.7173, 4.9506, 3.9035, 3.2415, 4.3301, 4.8469, 4.7687], device='cuda:3'), covar=tensor([0.0291, 0.1425, 0.0240, 0.0894, 0.1486, 0.0267, 0.0168, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0242, 0.0208, 0.0319, 0.0264, 0.0228, 0.0198, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:19:47,083 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98378.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 06:19:47,363 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9083, 4.2861, 4.3056, 4.5050, 2.8141, 4.2719, 2.8528, 1.4839], device='cuda:3'), covar=tensor([0.0559, 0.0304, 0.0767, 0.0272, 0.1602, 0.0263, 0.1461, 0.1931], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0180, 0.0265, 0.0174, 0.0226, 0.0165, 0.0234, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:20:37,058 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98410.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:20:38,134 INFO [train2.py:809] (3/4) Epoch 25, batch 2800, loss[ctc_loss=0.07436, att_loss=0.2388, loss=0.2059, over 16960.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007208, over 50.00 utterances.], tot_loss[ctc_loss=0.06745, att_loss=0.2325, loss=0.1995, over 3277156.11 frames. utt_duration=1228 frames, utt_pad_proportion=0.05687, over 10683.81 utterances.], batch size: 50, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:21:07,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 2023-03-09 06:21:37,616 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.960e+02 2.287e+02 2.906e+02 4.125e+02, threshold=4.574e+02, percent-clipped=0.0 2023-03-09 06:21:53,839 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2692, 4.3364, 4.5116, 4.4720, 5.0295, 4.4333, 4.3220, 2.4865], device='cuda:3'), covar=tensor([0.0315, 0.0399, 0.0374, 0.0358, 0.0884, 0.0271, 0.0408, 0.1923], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0205, 0.0202, 0.0218, 0.0376, 0.0177, 0.0193, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:21:56,306 INFO [train2.py:809] (3/4) Epoch 25, batch 2850, loss[ctc_loss=0.0749, att_loss=0.2303, loss=0.1992, over 16118.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006068, over 42.00 utterances.], tot_loss[ctc_loss=0.06696, att_loss=0.232, loss=0.199, over 3273594.89 frames. utt_duration=1243 frames, utt_pad_proportion=0.05393, over 10544.28 utterances.], batch size: 42, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:21:57,976 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98462.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:22:12,380 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98471.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:22:51,796 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6516, 5.0999, 4.9007, 4.9969, 5.2291, 4.7278, 3.2435, 5.0015], device='cuda:3'), covar=tensor([0.0124, 0.0097, 0.0123, 0.0087, 0.0076, 0.0131, 0.0802, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0092, 0.0113, 0.0071, 0.0078, 0.0089, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:23:14,285 INFO [train2.py:809] (3/4) Epoch 25, batch 2900, loss[ctc_loss=0.0554, att_loss=0.2137, loss=0.182, over 15643.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008925, over 37.00 utterances.], tot_loss[ctc_loss=0.06728, att_loss=0.2321, loss=0.1992, over 3270067.52 frames. utt_duration=1251 frames, utt_pad_proportion=0.05341, over 10471.00 utterances.], batch size: 37, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:24:06,296 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98543.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:24:07,663 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98544.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:24:14,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.817e+02 2.098e+02 2.346e+02 4.343e+02, threshold=4.197e+02, percent-clipped=0.0 2023-03-09 06:24:34,076 INFO [train2.py:809] (3/4) Epoch 25, batch 2950, loss[ctc_loss=0.04978, att_loss=0.2099, loss=0.1778, over 16009.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.00711, over 40.00 utterances.], tot_loss[ctc_loss=0.06735, att_loss=0.2321, loss=0.1992, over 3273883.02 frames. utt_duration=1275 frames, utt_pad_proportion=0.0467, over 10285.25 utterances.], batch size: 40, lr: 4.28e-03, grad_scale: 16.0 2023-03-09 06:25:23,566 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98592.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:25:43,576 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98604.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:25:54,033 INFO [train2.py:809] (3/4) Epoch 25, batch 3000, loss[ctc_loss=0.1289, att_loss=0.2712, loss=0.2427, over 14570.00 frames. utt_duration=400.6 frames, utt_pad_proportion=0.302, over 146.00 utterances.], tot_loss[ctc_loss=0.06822, att_loss=0.2332, loss=0.2002, over 3276483.60 frames. utt_duration=1231 frames, utt_pad_proportion=0.05715, over 10655.51 utterances.], batch size: 146, lr: 4.27e-03, grad_scale: 16.0 2023-03-09 06:25:54,033 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 06:26:08,487 INFO [train2.py:843] (3/4) Epoch 25, validation: ctc_loss=0.04165, att_loss=0.235, loss=0.1963, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 06:26:08,488 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 06:27:09,501 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 1.890e+02 2.327e+02 2.970e+02 6.667e+02, threshold=4.653e+02, percent-clipped=5.0 2023-03-09 06:27:28,841 INFO [train2.py:809] (3/4) Epoch 25, batch 3050, loss[ctc_loss=0.1131, att_loss=0.27, loss=0.2386, over 17308.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02424, over 59.00 utterances.], tot_loss[ctc_loss=0.06866, att_loss=0.234, loss=0.2009, over 3279891.67 frames. utt_duration=1224 frames, utt_pad_proportion=0.05952, over 10732.21 utterances.], batch size: 59, lr: 4.27e-03, grad_scale: 16.0 2023-03-09 06:27:45,251 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4706, 2.6102, 4.9291, 3.9361, 3.1624, 4.3742, 4.7419, 4.6720], device='cuda:3'), covar=tensor([0.0313, 0.1507, 0.0217, 0.0839, 0.1559, 0.0245, 0.0186, 0.0297], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0244, 0.0209, 0.0320, 0.0265, 0.0229, 0.0199, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:27:57,811 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7554, 5.9905, 5.4616, 5.6898, 5.6792, 5.1135, 5.4806, 5.2307], device='cuda:3'), covar=tensor([0.1352, 0.0923, 0.1033, 0.0927, 0.0910, 0.1716, 0.2217, 0.2293], device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0626, 0.0475, 0.0465, 0.0434, 0.0483, 0.0626, 0.0537], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:28:18,998 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4402, 4.3660, 4.4284, 4.4904, 5.0751, 4.3722, 4.2957, 2.4732], device='cuda:3'), covar=tensor([0.0251, 0.0377, 0.0376, 0.0287, 0.0743, 0.0268, 0.0418, 0.1914], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0206, 0.0203, 0.0220, 0.0375, 0.0177, 0.0193, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:28:31,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 06:28:48,634 INFO [train2.py:809] (3/4) Epoch 25, batch 3100, loss[ctc_loss=0.07512, att_loss=0.2535, loss=0.2178, over 17324.00 frames. utt_duration=1262 frames, utt_pad_proportion=0.009762, over 55.00 utterances.], tot_loss[ctc_loss=0.06865, att_loss=0.2336, loss=0.2006, over 3268920.29 frames. utt_duration=1241 frames, utt_pad_proportion=0.05872, over 10549.61 utterances.], batch size: 55, lr: 4.27e-03, grad_scale: 16.0 2023-03-09 06:29:04,779 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0217, 5.3740, 5.5615, 5.3999, 5.5163, 5.9810, 5.3442, 6.0391], device='cuda:3'), covar=tensor([0.0794, 0.0755, 0.0843, 0.1326, 0.1865, 0.0955, 0.0699, 0.0763], device='cuda:3'), in_proj_covar=tensor([0.0917, 0.0525, 0.0639, 0.0682, 0.0915, 0.0665, 0.0518, 0.0639], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:29:48,740 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 1.858e+02 2.243e+02 2.658e+02 4.552e+02, threshold=4.486e+02, percent-clipped=0.0 2023-03-09 06:30:07,767 INFO [train2.py:809] (3/4) Epoch 25, batch 3150, loss[ctc_loss=0.0612, att_loss=0.2317, loss=0.1976, over 16766.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.00654, over 48.00 utterances.], tot_loss[ctc_loss=0.06825, att_loss=0.2327, loss=0.1998, over 3262722.89 frames. utt_duration=1236 frames, utt_pad_proportion=0.06098, over 10569.34 utterances.], batch size: 48, lr: 4.27e-03, grad_scale: 16.0 2023-03-09 06:30:09,785 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98762.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:30:15,627 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98766.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:31:26,673 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98810.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:31:28,126 INFO [train2.py:809] (3/4) Epoch 25, batch 3200, loss[ctc_loss=0.06047, att_loss=0.2114, loss=0.1812, over 14501.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.03666, over 32.00 utterances.], tot_loss[ctc_loss=0.06813, att_loss=0.2329, loss=0.1999, over 3265051.09 frames. utt_duration=1244 frames, utt_pad_proportion=0.05742, over 10514.28 utterances.], batch size: 32, lr: 4.27e-03, grad_scale: 8.0 2023-03-09 06:31:44,664 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.10 vs. limit=5.0 2023-03-09 06:32:29,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.298e+02 1.857e+02 2.201e+02 2.576e+02 7.126e+02, threshold=4.402e+02, percent-clipped=1.0 2023-03-09 06:32:46,795 INFO [train2.py:809] (3/4) Epoch 25, batch 3250, loss[ctc_loss=0.05552, att_loss=0.2207, loss=0.1877, over 15909.00 frames. utt_duration=1633 frames, utt_pad_proportion=0.007695, over 39.00 utterances.], tot_loss[ctc_loss=0.06805, att_loss=0.2327, loss=0.1998, over 3268798.25 frames. utt_duration=1261 frames, utt_pad_proportion=0.05301, over 10383.91 utterances.], batch size: 39, lr: 4.27e-03, grad_scale: 8.0 2023-03-09 06:33:33,486 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98890.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:33:41,380 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98895.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:33:47,343 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98899.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:34:05,907 INFO [train2.py:809] (3/4) Epoch 25, batch 3300, loss[ctc_loss=0.06296, att_loss=0.2287, loss=0.1955, over 16340.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005533, over 45.00 utterances.], tot_loss[ctc_loss=0.06758, att_loss=0.2325, loss=0.1995, over 3275638.58 frames. utt_duration=1258 frames, utt_pad_proportion=0.05007, over 10427.13 utterances.], batch size: 45, lr: 4.27e-03, grad_scale: 8.0 2023-03-09 06:34:10,678 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5663, 5.8601, 5.3174, 5.5726, 5.5391, 4.9856, 5.2469, 5.0762], device='cuda:3'), covar=tensor([0.1432, 0.0856, 0.0958, 0.0828, 0.0894, 0.1586, 0.2274, 0.2229], device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0628, 0.0475, 0.0468, 0.0433, 0.0482, 0.0627, 0.0540], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:35:01,498 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2021, 3.7939, 3.1872, 3.4796, 3.9689, 3.6809, 3.0048, 4.2249], device='cuda:3'), covar=tensor([0.0876, 0.0541, 0.1063, 0.0690, 0.0723, 0.0743, 0.0882, 0.0534], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0225, 0.0227, 0.0205, 0.0286, 0.0244, 0.0201, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 06:35:07,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 1.882e+02 2.265e+02 2.775e+02 6.610e+02, threshold=4.530e+02, percent-clipped=3.0 2023-03-09 06:35:09,227 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98951.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:35:17,002 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98956.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:35:24,891 INFO [train2.py:809] (3/4) Epoch 25, batch 3350, loss[ctc_loss=0.06459, att_loss=0.2381, loss=0.2034, over 16765.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005553, over 48.00 utterances.], tot_loss[ctc_loss=0.06739, att_loss=0.2327, loss=0.1996, over 3271468.61 frames. utt_duration=1269 frames, utt_pad_proportion=0.04816, over 10327.12 utterances.], batch size: 48, lr: 4.27e-03, grad_scale: 8.0 2023-03-09 06:35:30,737 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2556, 3.8936, 3.7358, 3.4406, 3.8860, 3.9401, 3.9120, 3.0259], device='cuda:3'), covar=tensor([0.0879, 0.0963, 0.2415, 0.2592, 0.1066, 0.1564, 0.0840, 0.3152], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0214, 0.0268, 0.0177, 0.0276, 0.0198, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:35:36,981 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.14 vs. limit=5.0 2023-03-09 06:35:41,919 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98971.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:36:41,609 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4586, 3.2432, 3.3016, 4.4172, 3.8774, 3.9975, 2.8416, 2.3124], device='cuda:3'), covar=tensor([0.0688, 0.1572, 0.0868, 0.0519, 0.0855, 0.0427, 0.1622, 0.2115], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0218, 0.0187, 0.0219, 0.0231, 0.0183, 0.0204, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:36:44,966 INFO [train2.py:809] (3/4) Epoch 25, batch 3400, loss[ctc_loss=0.05598, att_loss=0.2141, loss=0.1825, over 15759.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.008735, over 38.00 utterances.], tot_loss[ctc_loss=0.06799, att_loss=0.2326, loss=0.1997, over 3266598.01 frames. utt_duration=1246 frames, utt_pad_proportion=0.05694, over 10502.55 utterances.], batch size: 38, lr: 4.27e-03, grad_scale: 8.0 2023-03-09 06:37:18,887 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99032.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:37:45,688 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 1.932e+02 2.257e+02 2.819e+02 9.991e+02, threshold=4.513e+02, percent-clipped=2.0 2023-03-09 06:37:48,229 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99051.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:38:03,634 INFO [train2.py:809] (3/4) Epoch 25, batch 3450, loss[ctc_loss=0.05485, att_loss=0.2374, loss=0.2009, over 16956.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.00802, over 50.00 utterances.], tot_loss[ctc_loss=0.06871, att_loss=0.2333, loss=0.2003, over 3276856.02 frames. utt_duration=1237 frames, utt_pad_proportion=0.05637, over 10605.74 utterances.], batch size: 50, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:38:12,081 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99066.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:38:44,679 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7553, 3.9817, 3.9667, 4.0084, 4.0628, 3.8292, 2.9988, 3.9014], device='cuda:3'), covar=tensor([0.0149, 0.0115, 0.0129, 0.0087, 0.0099, 0.0137, 0.0626, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0091, 0.0113, 0.0071, 0.0078, 0.0089, 0.0104, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:39:22,271 INFO [train2.py:809] (3/4) Epoch 25, batch 3500, loss[ctc_loss=0.05438, att_loss=0.2332, loss=0.1975, over 16473.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006086, over 46.00 utterances.], tot_loss[ctc_loss=0.06854, att_loss=0.233, loss=0.2001, over 3274020.51 frames. utt_duration=1241 frames, utt_pad_proportion=0.05505, over 10568.00 utterances.], batch size: 46, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:39:24,136 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99112.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:39:27,520 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99114.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:39:27,742 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5945, 3.1563, 3.3902, 4.4472, 3.9555, 4.0238, 2.9034, 2.4259], device='cuda:3'), covar=tensor([0.0679, 0.1718, 0.0906, 0.0576, 0.0832, 0.0439, 0.1600, 0.2014], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0219, 0.0187, 0.0220, 0.0231, 0.0183, 0.0203, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:39:30,787 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9904, 5.2666, 5.1820, 5.1695, 5.2756, 5.2610, 4.8913, 4.6557], device='cuda:3'), covar=tensor([0.0985, 0.0504, 0.0319, 0.0536, 0.0317, 0.0356, 0.0485, 0.0366], device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0374, 0.0362, 0.0374, 0.0435, 0.0441, 0.0373, 0.0408], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:39:30,919 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4841, 2.8589, 3.6335, 2.9874, 3.4282, 4.6144, 4.4003, 3.1214], device='cuda:3'), covar=tensor([0.0434, 0.1931, 0.1316, 0.1454, 0.1235, 0.0942, 0.0610, 0.1369], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0247, 0.0288, 0.0222, 0.0267, 0.0377, 0.0268, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:40:23,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.663e+02 2.107e+02 2.565e+02 8.689e+02, threshold=4.214e+02, percent-clipped=2.0 2023-03-09 06:40:34,181 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.50 vs. limit=5.0 2023-03-09 06:40:40,956 INFO [train2.py:809] (3/4) Epoch 25, batch 3550, loss[ctc_loss=0.1269, att_loss=0.2612, loss=0.2343, over 13859.00 frames. utt_duration=378.5 frames, utt_pad_proportion=0.3371, over 147.00 utterances.], tot_loss[ctc_loss=0.06862, att_loss=0.2327, loss=0.1999, over 3263564.50 frames. utt_duration=1251 frames, utt_pad_proportion=0.05583, over 10448.66 utterances.], batch size: 147, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:41:40,865 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99199.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:41:59,397 INFO [train2.py:809] (3/4) Epoch 25, batch 3600, loss[ctc_loss=0.07419, att_loss=0.2179, loss=0.1892, over 12685.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.08114, over 28.00 utterances.], tot_loss[ctc_loss=0.06803, att_loss=0.2318, loss=0.199, over 3251845.68 frames. utt_duration=1267 frames, utt_pad_proportion=0.05532, over 10279.84 utterances.], batch size: 28, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:42:23,786 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0668, 2.6304, 2.8800, 3.7571, 3.4608, 3.5921, 2.7076, 2.2611], device='cuda:3'), covar=tensor([0.0785, 0.1905, 0.0922, 0.0723, 0.0892, 0.0502, 0.1558, 0.1980], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0217, 0.0186, 0.0220, 0.0231, 0.0183, 0.0203, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:42:28,552 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99229.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:42:55,191 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99246.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:42:57,332 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:43:01,837 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.837e+02 2.162e+02 2.807e+02 5.340e+02, threshold=4.325e+02, percent-clipped=4.0 2023-03-09 06:43:03,546 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99251.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:43:19,234 INFO [train2.py:809] (3/4) Epoch 25, batch 3650, loss[ctc_loss=0.07628, att_loss=0.2371, loss=0.2049, over 17346.00 frames. utt_duration=1007 frames, utt_pad_proportion=0.04979, over 69.00 utterances.], tot_loss[ctc_loss=0.06868, att_loss=0.2325, loss=0.1997, over 3249975.77 frames. utt_duration=1219 frames, utt_pad_proportion=0.06714, over 10676.04 utterances.], batch size: 69, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:43:30,901 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99268.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:43:41,085 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1904, 5.1973, 4.9914, 2.7859, 2.2041, 3.3092, 2.6777, 3.9136], device='cuda:3'), covar=tensor([0.0695, 0.0339, 0.0330, 0.4581, 0.5014, 0.2115, 0.3575, 0.1746], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0291, 0.0275, 0.0248, 0.0340, 0.0335, 0.0262, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 06:44:05,971 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99290.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:44:25,945 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0319, 4.9893, 4.7728, 2.9750, 4.7592, 4.6715, 4.3196, 2.8899], device='cuda:3'), covar=tensor([0.0103, 0.0111, 0.0304, 0.0969, 0.0110, 0.0205, 0.0301, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0105, 0.0108, 0.0111, 0.0087, 0.0115, 0.0100, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:44:40,184 INFO [train2.py:809] (3/4) Epoch 25, batch 3700, loss[ctc_loss=0.08404, att_loss=0.2424, loss=0.2107, over 16164.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007788, over 41.00 utterances.], tot_loss[ctc_loss=0.06793, att_loss=0.232, loss=0.1992, over 3252639.10 frames. utt_duration=1239 frames, utt_pad_proportion=0.06171, over 10513.22 utterances.], batch size: 41, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:44:53,516 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0331, 6.2770, 5.7962, 5.9567, 5.9519, 5.3922, 5.7724, 5.3481], device='cuda:3'), covar=tensor([0.1324, 0.0813, 0.1024, 0.0766, 0.0909, 0.1637, 0.2045, 0.2345], device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0624, 0.0476, 0.0463, 0.0433, 0.0483, 0.0626, 0.0537], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:45:01,506 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:45:05,841 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99327.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:45:09,039 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:45:41,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.855e+02 2.238e+02 2.669e+02 6.912e+02, threshold=4.475e+02, percent-clipped=5.0 2023-03-09 06:45:59,443 INFO [train2.py:809] (3/4) Epoch 25, batch 3750, loss[ctc_loss=0.06223, att_loss=0.2291, loss=0.1958, over 15934.00 frames. utt_duration=1556 frames, utt_pad_proportion=0.008503, over 41.00 utterances.], tot_loss[ctc_loss=0.06805, att_loss=0.2323, loss=0.1995, over 3253851.37 frames. utt_duration=1215 frames, utt_pad_proportion=0.06619, over 10723.35 utterances.], batch size: 41, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:45:59,985 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9403, 4.8700, 4.8214, 2.1562, 2.0310, 2.8477, 2.5281, 3.8098], device='cuda:3'), covar=tensor([0.0760, 0.0297, 0.0231, 0.5783, 0.5311, 0.2592, 0.3471, 0.1552], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0292, 0.0276, 0.0250, 0.0341, 0.0336, 0.0262, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 06:46:37,893 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99385.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:47:13,301 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99407.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:47:19,833 INFO [train2.py:809] (3/4) Epoch 25, batch 3800, loss[ctc_loss=0.07251, att_loss=0.2324, loss=0.2004, over 16704.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.00497, over 46.00 utterances.], tot_loss[ctc_loss=0.06907, att_loss=0.2329, loss=0.2001, over 3248771.54 frames. utt_duration=1178 frames, utt_pad_proportion=0.07792, over 11049.01 utterances.], batch size: 46, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:47:34,705 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99420.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:48:22,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.283e+02 1.840e+02 2.193e+02 2.491e+02 3.920e+02, threshold=4.386e+02, percent-clipped=0.0 2023-03-09 06:48:38,760 INFO [train2.py:809] (3/4) Epoch 25, batch 3850, loss[ctc_loss=0.06556, att_loss=0.2365, loss=0.2023, over 16610.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006171, over 47.00 utterances.], tot_loss[ctc_loss=0.06854, att_loss=0.233, loss=0.2001, over 3261936.58 frames. utt_duration=1198 frames, utt_pad_proportion=0.06964, over 10903.83 utterances.], batch size: 47, lr: 4.26e-03, grad_scale: 8.0 2023-03-09 06:48:49,812 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99468.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 06:49:09,251 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99481.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:49:55,235 INFO [train2.py:809] (3/4) Epoch 25, batch 3900, loss[ctc_loss=0.04964, att_loss=0.2107, loss=0.1785, over 15957.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006964, over 41.00 utterances.], tot_loss[ctc_loss=0.06914, att_loss=0.2335, loss=0.2006, over 3266774.33 frames. utt_duration=1186 frames, utt_pad_proportion=0.07058, over 11032.01 utterances.], batch size: 41, lr: 4.25e-03, grad_scale: 8.0 2023-03-09 06:50:23,082 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99529.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 06:50:48,903 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99546.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:50:54,664 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 1.849e+02 2.248e+02 2.718e+02 5.190e+02, threshold=4.496e+02, percent-clipped=3.0 2023-03-09 06:50:56,461 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99551.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:51:11,370 INFO [train2.py:809] (3/4) Epoch 25, batch 3950, loss[ctc_loss=0.1375, att_loss=0.2665, loss=0.2407, over 14144.00 frames. utt_duration=386.5 frames, utt_pad_proportion=0.3231, over 147.00 utterances.], tot_loss[ctc_loss=0.06947, att_loss=0.2339, loss=0.201, over 3272024.04 frames. utt_duration=1184 frames, utt_pad_proportion=0.06942, over 11068.15 utterances.], batch size: 147, lr: 4.25e-03, grad_scale: 8.0 2023-03-09 06:51:22,256 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99568.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:51:47,850 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99585.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:51:48,095 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5547, 4.5175, 4.6704, 4.6477, 5.2549, 4.4645, 4.5694, 2.8634], device='cuda:3'), covar=tensor([0.0288, 0.0417, 0.0379, 0.0432, 0.0871, 0.0318, 0.0400, 0.1601], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0212, 0.0211, 0.0227, 0.0386, 0.0185, 0.0199, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:51:54,094 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99589.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:51:57,101 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3920, 4.3811, 4.5371, 4.5343, 5.1183, 4.3464, 4.4037, 2.5484], device='cuda:3'), covar=tensor([0.0323, 0.0446, 0.0392, 0.0384, 0.0779, 0.0344, 0.0446, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0213, 0.0211, 0.0227, 0.0387, 0.0185, 0.0200, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 06:52:25,626 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99594.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:52:27,616 INFO [train2.py:809] (3/4) Epoch 26, batch 0, loss[ctc_loss=0.05328, att_loss=0.2105, loss=0.179, over 15644.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008909, over 37.00 utterances.], tot_loss[ctc_loss=0.05328, att_loss=0.2105, loss=0.179, over 15644.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008909, over 37.00 utterances.], batch size: 37, lr: 4.17e-03, grad_scale: 8.0 2023-03-09 06:52:27,616 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 06:52:37,210 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1276, 4.1927, 4.4545, 4.2752, 4.3537, 5.1769, 4.9664, 4.3587], device='cuda:3'), covar=tensor([0.0259, 0.0857, 0.0737, 0.0736, 0.0681, 0.0709, 0.0399, 0.0622], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0247, 0.0287, 0.0221, 0.0265, 0.0375, 0.0269, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:52:40,070 INFO [train2.py:843] (3/4) Epoch 26, validation: ctc_loss=0.04131, att_loss=0.2338, loss=0.1953, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 06:52:40,071 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 06:52:46,819 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99599.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:53:27,455 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99624.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:53:32,886 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99627.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:53:36,102 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99629.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:54:00,743 INFO [train2.py:809] (3/4) Epoch 26, batch 50, loss[ctc_loss=0.06746, att_loss=0.2185, loss=0.1883, over 16013.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007477, over 40.00 utterances.], tot_loss[ctc_loss=0.06683, att_loss=0.2336, loss=0.2003, over 738218.72 frames. utt_duration=1191 frames, utt_pad_proportion=0.0649, over 2482.65 utterances.], batch size: 40, lr: 4.17e-03, grad_scale: 8.0 2023-03-09 06:54:08,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.776e+02 2.200e+02 2.639e+02 7.025e+02, threshold=4.399e+02, percent-clipped=1.0 2023-03-09 06:54:09,449 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99650.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:54:28,357 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8402, 6.0449, 5.4449, 5.6524, 5.7115, 5.1559, 5.4991, 5.1442], device='cuda:3'), covar=tensor([0.1146, 0.0922, 0.0993, 0.0916, 0.0934, 0.1671, 0.2209, 0.2341], device='cuda:3'), in_proj_covar=tensor([0.0544, 0.0632, 0.0481, 0.0470, 0.0438, 0.0487, 0.0630, 0.0541], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:54:31,084 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-03-09 06:54:48,779 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99675.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:54:55,013 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3105, 4.6741, 4.5229, 4.5487, 4.7018, 4.4101, 3.2411, 4.5921], device='cuda:3'), covar=tensor([0.0126, 0.0106, 0.0133, 0.0099, 0.0098, 0.0114, 0.0700, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0091, 0.0115, 0.0071, 0.0078, 0.0089, 0.0105, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 06:54:56,442 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:54:56,589 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:55:19,432 INFO [train2.py:809] (3/4) Epoch 26, batch 100, loss[ctc_loss=0.065, att_loss=0.2085, loss=0.1798, over 15508.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008418, over 36.00 utterances.], tot_loss[ctc_loss=0.06805, att_loss=0.2324, loss=0.1995, over 1301426.48 frames. utt_duration=1222 frames, utt_pad_proportion=0.05632, over 4264.52 utterances.], batch size: 36, lr: 4.17e-03, grad_scale: 8.0 2023-03-09 06:55:39,217 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99707.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:56:08,692 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1103, 5.3712, 5.3331, 5.3466, 5.3937, 5.3453, 5.0141, 4.7721], device='cuda:3'), covar=tensor([0.0984, 0.0516, 0.0278, 0.0507, 0.0285, 0.0313, 0.0420, 0.0363], device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0378, 0.0364, 0.0371, 0.0434, 0.0443, 0.0372, 0.0407], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 06:56:31,539 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99741.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:56:37,957 INFO [train2.py:809] (3/4) Epoch 26, batch 150, loss[ctc_loss=0.1279, att_loss=0.2704, loss=0.2419, over 14424.00 frames. utt_duration=396.7 frames, utt_pad_proportion=0.309, over 146.00 utterances.], tot_loss[ctc_loss=0.06963, att_loss=0.2341, loss=0.2012, over 1736362.08 frames. utt_duration=1167 frames, utt_pad_proportion=0.07418, over 5956.80 utterances.], batch size: 146, lr: 4.17e-03, grad_scale: 8.0 2023-03-09 06:56:46,328 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 1.892e+02 2.300e+02 2.888e+02 7.505e+02, threshold=4.601e+02, percent-clipped=4.0 2023-03-09 06:56:54,168 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99755.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:57:27,686 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99776.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 06:57:57,205 INFO [train2.py:809] (3/4) Epoch 26, batch 200, loss[ctc_loss=0.07284, att_loss=0.2499, loss=0.2145, over 17372.00 frames. utt_duration=1180 frames, utt_pad_proportion=0.0195, over 59.00 utterances.], tot_loss[ctc_loss=0.0684, att_loss=0.234, loss=0.2009, over 2082785.15 frames. utt_duration=1195 frames, utt_pad_proportion=0.06369, over 6978.23 utterances.], batch size: 59, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 06:58:32,174 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 06:58:44,451 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99824.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 06:59:17,388 INFO [train2.py:809] (3/4) Epoch 26, batch 250, loss[ctc_loss=0.0515, att_loss=0.2167, loss=0.1836, over 16395.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007888, over 44.00 utterances.], tot_loss[ctc_loss=0.06773, att_loss=0.2333, loss=0.2002, over 2339770.65 frames. utt_duration=1208 frames, utt_pad_proportion=0.06285, over 7759.61 utterances.], batch size: 44, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 06:59:25,256 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 1.972e+02 2.288e+02 2.962e+02 9.624e+02, threshold=4.577e+02, percent-clipped=5.0 2023-03-09 07:00:19,872 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99885.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:00:35,346 INFO [train2.py:809] (3/4) Epoch 26, batch 300, loss[ctc_loss=0.09172, att_loss=0.2584, loss=0.2251, over 17018.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008123, over 51.00 utterances.], tot_loss[ctc_loss=0.06881, att_loss=0.2341, loss=0.201, over 2553136.31 frames. utt_duration=1226 frames, utt_pad_proportion=0.05663, over 8337.19 utterances.], batch size: 51, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:01:21,970 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:01:22,079 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:01:22,285 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9537, 4.9753, 4.8719, 2.0477, 1.9583, 2.9864, 2.0891, 3.8090], device='cuda:3'), covar=tensor([0.0778, 0.0269, 0.0247, 0.5582, 0.5587, 0.2385, 0.4096, 0.1650], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0292, 0.0276, 0.0250, 0.0342, 0.0334, 0.0262, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 07:01:35,964 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99933.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:01:55,227 INFO [train2.py:809] (3/4) Epoch 26, batch 350, loss[ctc_loss=0.04946, att_loss=0.2281, loss=0.1924, over 16117.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006987, over 42.00 utterances.], tot_loss[ctc_loss=0.06885, att_loss=0.2343, loss=0.2012, over 2719974.10 frames. utt_duration=1207 frames, utt_pad_proportion=0.0607, over 9024.97 utterances.], batch size: 42, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:01:55,389 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99945.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:02:02,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 1.873e+02 2.109e+02 2.728e+02 6.807e+02, threshold=4.219e+02, percent-clipped=3.0 2023-03-09 07:02:39,743 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99972.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:02:52,048 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99980.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:03:15,357 INFO [train2.py:809] (3/4) Epoch 26, batch 400, loss[ctc_loss=0.08091, att_loss=0.2559, loss=0.2209, over 17473.00 frames. utt_duration=1111 frames, utt_pad_proportion=0.02901, over 63.00 utterances.], tot_loss[ctc_loss=0.06861, att_loss=0.234, loss=0.2009, over 2849399.00 frames. utt_duration=1239 frames, utt_pad_proportion=0.05165, over 9210.91 utterances.], batch size: 63, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:03:39,412 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100007.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 07:04:11,935 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100028.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:04:24,053 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100036.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:04:38,119 INFO [train2.py:809] (3/4) Epoch 26, batch 450, loss[ctc_loss=0.05539, att_loss=0.2283, loss=0.1938, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.00748, over 50.00 utterances.], tot_loss[ctc_loss=0.06756, att_loss=0.233, loss=0.1999, over 2935308.97 frames. utt_duration=1260 frames, utt_pad_proportion=0.05112, over 9326.21 utterances.], batch size: 50, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:04:46,164 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 1.836e+02 2.145e+02 2.710e+02 6.185e+02, threshold=4.291e+02, percent-clipped=5.0 2023-03-09 07:05:13,345 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1280, 3.6381, 3.7165, 2.9261, 3.7561, 3.8204, 3.7563, 2.3923], device='cuda:3'), covar=tensor([0.0987, 0.1588, 0.2107, 0.5579, 0.1761, 0.2317, 0.0895, 0.6183], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0213, 0.0267, 0.0177, 0.0275, 0.0196, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:05:15,473 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100068.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:05:27,562 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100076.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:05:34,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-09 07:05:56,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 07:05:56,811 INFO [train2.py:809] (3/4) Epoch 26, batch 500, loss[ctc_loss=0.05352, att_loss=0.2202, loss=0.1868, over 16122.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006556, over 42.00 utterances.], tot_loss[ctc_loss=0.06771, att_loss=0.2328, loss=0.1998, over 3012195.97 frames. utt_duration=1253 frames, utt_pad_proportion=0.05224, over 9625.56 utterances.], batch size: 42, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:05:57,877 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 07:06:43,579 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100124.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:06:43,786 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100124.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 07:07:16,141 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-03-09 07:07:16,641 INFO [train2.py:809] (3/4) Epoch 26, batch 550, loss[ctc_loss=0.06582, att_loss=0.2275, loss=0.1952, over 16627.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005155, over 47.00 utterances.], tot_loss[ctc_loss=0.06732, att_loss=0.2324, loss=0.1994, over 3068190.49 frames. utt_duration=1272 frames, utt_pad_proportion=0.04931, over 9662.92 utterances.], batch size: 47, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:07:24,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 1.784e+02 2.229e+02 2.699e+02 5.667e+02, threshold=4.458e+02, percent-clipped=3.0 2023-03-09 07:08:00,402 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100172.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:08:36,932 INFO [train2.py:809] (3/4) Epoch 26, batch 600, loss[ctc_loss=0.06996, att_loss=0.2491, loss=0.2132, over 17394.00 frames. utt_duration=1181 frames, utt_pad_proportion=0.01924, over 59.00 utterances.], tot_loss[ctc_loss=0.06723, att_loss=0.2326, loss=0.1995, over 3112512.01 frames. utt_duration=1271 frames, utt_pad_proportion=0.05017, over 9803.53 utterances.], batch size: 59, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:08:58,707 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8706, 4.6578, 4.6901, 2.1641, 2.0605, 2.7982, 2.3429, 3.7360], device='cuda:3'), covar=tensor([0.0804, 0.0305, 0.0250, 0.5156, 0.5465, 0.2625, 0.3785, 0.1561], device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0294, 0.0279, 0.0251, 0.0344, 0.0336, 0.0263, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 07:09:24,060 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-09 07:09:24,636 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100224.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:09:47,259 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:09:58,210 INFO [train2.py:809] (3/4) Epoch 26, batch 650, loss[ctc_loss=0.06708, att_loss=0.2295, loss=0.197, over 16482.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006534, over 46.00 utterances.], tot_loss[ctc_loss=0.06761, att_loss=0.2326, loss=0.1996, over 3147273.99 frames. utt_duration=1280 frames, utt_pad_proportion=0.04783, over 9846.70 utterances.], batch size: 46, lr: 4.16e-03, grad_scale: 8.0 2023-03-09 07:09:58,554 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100245.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:10:06,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 1.945e+02 2.331e+02 2.893e+02 4.719e+02, threshold=4.662e+02, percent-clipped=1.0 2023-03-09 07:10:43,118 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100272.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:11:16,656 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100293.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:11:19,701 INFO [train2.py:809] (3/4) Epoch 26, batch 700, loss[ctc_loss=0.0582, att_loss=0.2308, loss=0.1963, over 16857.00 frames. utt_duration=682.5 frames, utt_pad_proportion=0.1437, over 99.00 utterances.], tot_loss[ctc_loss=0.06784, att_loss=0.2326, loss=0.1997, over 3170982.66 frames. utt_duration=1245 frames, utt_pad_proportion=0.05801, over 10197.62 utterances.], batch size: 99, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:11:26,819 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100299.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:11:42,965 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 2023-03-09 07:11:57,316 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3852, 2.6686, 3.1870, 4.3738, 3.8859, 3.7990, 2.9089, 2.3217], device='cuda:3'), covar=tensor([0.0769, 0.2145, 0.1009, 0.0542, 0.0960, 0.0562, 0.1538, 0.2268], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0218, 0.0188, 0.0221, 0.0232, 0.0186, 0.0204, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:12:26,362 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:12:37,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-03-09 07:12:40,625 INFO [train2.py:809] (3/4) Epoch 26, batch 750, loss[ctc_loss=0.1056, att_loss=0.2606, loss=0.2296, over 13650.00 frames. utt_duration=375.4 frames, utt_pad_proportion=0.3448, over 146.00 utterances.], tot_loss[ctc_loss=0.06747, att_loss=0.2324, loss=0.1994, over 3194406.80 frames. utt_duration=1266 frames, utt_pad_proportion=0.05314, over 10104.89 utterances.], batch size: 146, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:12:48,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 1.813e+02 2.158e+02 2.667e+02 1.532e+03, threshold=4.316e+02, percent-clipped=4.0 2023-03-09 07:13:10,581 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:13:37,324 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2023-03-09 07:13:43,913 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100384.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:14:01,733 INFO [train2.py:809] (3/4) Epoch 26, batch 800, loss[ctc_loss=0.07086, att_loss=0.2429, loss=0.2085, over 17445.00 frames. utt_duration=885 frames, utt_pad_proportion=0.07334, over 79.00 utterances.], tot_loss[ctc_loss=0.06722, att_loss=0.2322, loss=0.1992, over 3209947.78 frames. utt_duration=1264 frames, utt_pad_proportion=0.05214, over 10168.70 utterances.], batch size: 79, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:14:13,173 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2998, 2.8025, 3.3320, 4.3887, 3.8769, 3.7974, 2.8650, 2.3268], device='cuda:3'), covar=tensor([0.0750, 0.1901, 0.0862, 0.0497, 0.0842, 0.0493, 0.1469, 0.2097], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0218, 0.0188, 0.0221, 0.0231, 0.0186, 0.0204, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:14:13,218 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100402.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:15:21,086 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4415, 4.4265, 4.5285, 4.5438, 5.1143, 4.4074, 4.4498, 2.5463], device='cuda:3'), covar=tensor([0.0251, 0.0407, 0.0366, 0.0337, 0.0695, 0.0264, 0.0349, 0.1789], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0212, 0.0209, 0.0226, 0.0382, 0.0184, 0.0197, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:15:22,216 INFO [train2.py:809] (3/4) Epoch 26, batch 850, loss[ctc_loss=0.0603, att_loss=0.2434, loss=0.2067, over 16956.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.008035, over 50.00 utterances.], tot_loss[ctc_loss=0.06654, att_loss=0.2317, loss=0.1987, over 3223903.22 frames. utt_duration=1252 frames, utt_pad_proportion=0.05457, over 10313.72 utterances.], batch size: 50, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:15:31,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 1.778e+02 2.108e+02 2.517e+02 5.047e+02, threshold=4.216e+02, percent-clipped=2.0 2023-03-09 07:15:52,669 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100463.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 07:16:43,802 INFO [train2.py:809] (3/4) Epoch 26, batch 900, loss[ctc_loss=0.06118, att_loss=0.2425, loss=0.2062, over 16898.00 frames. utt_duration=1381 frames, utt_pad_proportion=0.005951, over 49.00 utterances.], tot_loss[ctc_loss=0.06672, att_loss=0.2321, loss=0.199, over 3241081.29 frames. utt_duration=1244 frames, utt_pad_proportion=0.05372, over 10431.64 utterances.], batch size: 49, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:17:02,469 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0840, 5.3292, 5.5015, 5.3139, 5.5429, 5.9744, 5.3208, 6.0966], device='cuda:3'), covar=tensor([0.0656, 0.0783, 0.0857, 0.1496, 0.1694, 0.0987, 0.0680, 0.0662], device='cuda:3'), in_proj_covar=tensor([0.0912, 0.0523, 0.0639, 0.0685, 0.0905, 0.0659, 0.0510, 0.0645], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:17:21,844 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1926, 5.2235, 5.0080, 3.0232, 5.0015, 4.9051, 4.4915, 2.7216], device='cuda:3'), covar=tensor([0.0111, 0.0099, 0.0269, 0.0994, 0.0095, 0.0168, 0.0293, 0.1370], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0106, 0.0109, 0.0113, 0.0088, 0.0116, 0.0102, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 07:18:03,338 INFO [train2.py:809] (3/4) Epoch 26, batch 950, loss[ctc_loss=0.05472, att_loss=0.2047, loss=0.1747, over 16011.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007616, over 40.00 utterances.], tot_loss[ctc_loss=0.06627, att_loss=0.2314, loss=0.1984, over 3247274.14 frames. utt_duration=1233 frames, utt_pad_proportion=0.05632, over 10543.48 utterances.], batch size: 40, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:18:11,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 1.882e+02 2.194e+02 2.822e+02 5.450e+02, threshold=4.387e+02, percent-clipped=2.0 2023-03-09 07:19:22,770 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100594.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:19:24,223 INFO [train2.py:809] (3/4) Epoch 26, batch 1000, loss[ctc_loss=0.08041, att_loss=0.2445, loss=0.2117, over 17064.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008171, over 52.00 utterances.], tot_loss[ctc_loss=0.06687, att_loss=0.2319, loss=0.1989, over 3247335.45 frames. utt_duration=1231 frames, utt_pad_proportion=0.05932, over 10564.75 utterances.], batch size: 52, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:20:45,245 INFO [train2.py:809] (3/4) Epoch 26, batch 1050, loss[ctc_loss=0.05378, att_loss=0.2045, loss=0.1744, over 14661.00 frames. utt_duration=1834 frames, utt_pad_proportion=0.03521, over 32.00 utterances.], tot_loss[ctc_loss=0.06701, att_loss=0.2318, loss=0.1988, over 3244035.14 frames. utt_duration=1253 frames, utt_pad_proportion=0.05521, over 10368.81 utterances.], batch size: 32, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:20:53,109 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.280e+02 1.800e+02 2.231e+02 2.686e+02 5.529e+02, threshold=4.462e+02, percent-clipped=2.0 2023-03-09 07:21:14,685 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100663.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 07:21:39,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-09 07:22:05,999 INFO [train2.py:809] (3/4) Epoch 26, batch 1100, loss[ctc_loss=0.08837, att_loss=0.2505, loss=0.2181, over 17369.00 frames. utt_duration=1009 frames, utt_pad_proportion=0.04934, over 69.00 utterances.], tot_loss[ctc_loss=0.06738, att_loss=0.233, loss=0.1998, over 3261935.00 frames. utt_duration=1252 frames, utt_pad_proportion=0.05201, over 10436.92 utterances.], batch size: 69, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:22:31,629 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100711.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 07:23:01,681 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100730.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 07:23:25,349 INFO [train2.py:809] (3/4) Epoch 26, batch 1150, loss[ctc_loss=0.07653, att_loss=0.2296, loss=0.199, over 16117.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006114, over 42.00 utterances.], tot_loss[ctc_loss=0.06746, att_loss=0.2333, loss=0.2002, over 3271758.14 frames. utt_duration=1253 frames, utt_pad_proportion=0.04988, over 10452.91 utterances.], batch size: 42, lr: 4.15e-03, grad_scale: 8.0 2023-03-09 07:23:32,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.823e+02 2.156e+02 2.583e+02 8.005e+02, threshold=4.313e+02, percent-clipped=1.0 2023-03-09 07:23:46,176 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100758.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 07:24:39,264 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100791.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:24:44,987 INFO [train2.py:809] (3/4) Epoch 26, batch 1200, loss[ctc_loss=0.05924, att_loss=0.24, loss=0.2039, over 17306.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01179, over 55.00 utterances.], tot_loss[ctc_loss=0.06796, att_loss=0.2339, loss=0.2007, over 3282075.76 frames. utt_duration=1250 frames, utt_pad_proportion=0.04997, over 10519.10 utterances.], batch size: 55, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:24:55,060 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3638, 4.2880, 4.3946, 4.3635, 4.8473, 4.3233, 4.1930, 2.5787], device='cuda:3'), covar=tensor([0.0266, 0.0444, 0.0367, 0.0344, 0.0626, 0.0273, 0.0393, 0.1735], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0211, 0.0208, 0.0224, 0.0381, 0.0183, 0.0196, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:24:55,097 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3798, 2.7171, 4.7450, 3.9019, 2.9725, 4.1104, 4.2919, 4.4406], device='cuda:3'), covar=tensor([0.0263, 0.1466, 0.0234, 0.0806, 0.1548, 0.0275, 0.0266, 0.0282], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0246, 0.0213, 0.0322, 0.0269, 0.0231, 0.0201, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:26:04,022 INFO [train2.py:809] (3/4) Epoch 26, batch 1250, loss[ctc_loss=0.06633, att_loss=0.2405, loss=0.2057, over 16768.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006481, over 48.00 utterances.], tot_loss[ctc_loss=0.06868, att_loss=0.2337, loss=0.2007, over 3273630.28 frames. utt_duration=1209 frames, utt_pad_proportion=0.0628, over 10845.28 utterances.], batch size: 48, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:26:11,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 1.779e+02 2.144e+02 2.683e+02 5.396e+02, threshold=4.288e+02, percent-clipped=4.0 2023-03-09 07:26:28,272 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9256, 4.9471, 4.8142, 2.1371, 1.9339, 2.8647, 2.3203, 3.9155], device='cuda:3'), covar=tensor([0.0829, 0.0318, 0.0309, 0.5419, 0.5672, 0.2538, 0.3836, 0.1620], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0296, 0.0278, 0.0250, 0.0340, 0.0334, 0.0262, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 07:27:22,217 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100894.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:27:23,468 INFO [train2.py:809] (3/4) Epoch 26, batch 1300, loss[ctc_loss=0.0605, att_loss=0.218, loss=0.1865, over 16005.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007466, over 40.00 utterances.], tot_loss[ctc_loss=0.06908, att_loss=0.2338, loss=0.2009, over 3273520.90 frames. utt_duration=1204 frames, utt_pad_proportion=0.06463, over 10892.71 utterances.], batch size: 40, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:27:29,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-09 07:27:37,180 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2306, 5.2410, 5.0271, 3.1722, 5.0925, 4.9368, 4.6295, 2.9193], device='cuda:3'), covar=tensor([0.0125, 0.0128, 0.0283, 0.0942, 0.0103, 0.0171, 0.0273, 0.1328], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0106, 0.0109, 0.0113, 0.0089, 0.0117, 0.0102, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 07:28:11,745 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5106, 3.1800, 3.5025, 4.4392, 3.8846, 3.8176, 3.0342, 2.4534], device='cuda:3'), covar=tensor([0.0691, 0.1672, 0.0807, 0.0582, 0.0834, 0.0540, 0.1460, 0.2091], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0221, 0.0192, 0.0225, 0.0235, 0.0191, 0.0208, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:28:38,865 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100942.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:28:43,606 INFO [train2.py:809] (3/4) Epoch 26, batch 1350, loss[ctc_loss=0.05497, att_loss=0.2359, loss=0.1997, over 16863.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.00789, over 49.00 utterances.], tot_loss[ctc_loss=0.06853, att_loss=0.2331, loss=0.2002, over 3262831.53 frames. utt_duration=1200 frames, utt_pad_proportion=0.06806, over 10889.69 utterances.], batch size: 49, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:28:51,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.935e+02 2.207e+02 2.665e+02 5.260e+02, threshold=4.414e+02, percent-clipped=2.0 2023-03-09 07:30:04,574 INFO [train2.py:809] (3/4) Epoch 26, batch 1400, loss[ctc_loss=0.04311, att_loss=0.2218, loss=0.1861, over 16876.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007782, over 49.00 utterances.], tot_loss[ctc_loss=0.0677, att_loss=0.2327, loss=0.1997, over 3269317.91 frames. utt_duration=1213 frames, utt_pad_proportion=0.06419, over 10792.31 utterances.], batch size: 49, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:30:04,991 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5588, 2.3683, 2.2004, 2.5465, 2.7628, 2.5127, 2.2922, 2.9967], device='cuda:3'), covar=tensor([0.1959, 0.2992, 0.2019, 0.1592, 0.2009, 0.1489, 0.2316, 0.1109], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0139, 0.0134, 0.0129, 0.0146, 0.0126, 0.0148, 0.0125], device='cuda:3'), out_proj_covar=tensor([1.0578e-04, 1.1003e-04, 1.0917e-04, 1.0083e-04, 1.1012e-04, 1.0119e-04, 1.1312e-04, 9.8820e-05], device='cuda:3') 2023-03-09 07:31:24,473 INFO [train2.py:809] (3/4) Epoch 26, batch 1450, loss[ctc_loss=0.06757, att_loss=0.2401, loss=0.2056, over 16485.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005559, over 46.00 utterances.], tot_loss[ctc_loss=0.06736, att_loss=0.2326, loss=0.1995, over 3273333.19 frames. utt_duration=1227 frames, utt_pad_proportion=0.06072, over 10679.83 utterances.], batch size: 46, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:31:32,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.754e+02 2.064e+02 2.618e+02 5.187e+02, threshold=4.127e+02, percent-clipped=2.0 2023-03-09 07:31:45,771 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101058.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:32:30,565 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101086.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 07:32:44,611 INFO [train2.py:809] (3/4) Epoch 26, batch 1500, loss[ctc_loss=0.09839, att_loss=0.2632, loss=0.2302, over 17277.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01273, over 55.00 utterances.], tot_loss[ctc_loss=0.06765, att_loss=0.2331, loss=0.2, over 3267197.94 frames. utt_duration=1216 frames, utt_pad_proportion=0.06368, over 10763.29 utterances.], batch size: 55, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:33:02,317 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101106.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:33:03,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-09 07:33:04,157 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101107.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:34:03,292 INFO [train2.py:809] (3/4) Epoch 26, batch 1550, loss[ctc_loss=0.06058, att_loss=0.2125, loss=0.1822, over 15874.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009814, over 39.00 utterances.], tot_loss[ctc_loss=0.06706, att_loss=0.232, loss=0.199, over 3266512.11 frames. utt_duration=1262 frames, utt_pad_proportion=0.05361, over 10364.28 utterances.], batch size: 39, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:34:11,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.859e+02 2.187e+02 2.715e+02 5.591e+02, threshold=4.373e+02, percent-clipped=3.0 2023-03-09 07:34:13,015 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0512, 3.7220, 3.0788, 3.3490, 3.8505, 3.5517, 3.0252, 4.0876], device='cuda:3'), covar=tensor([0.0904, 0.0465, 0.1071, 0.0677, 0.0774, 0.0729, 0.0766, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0225, 0.0230, 0.0206, 0.0288, 0.0247, 0.0203, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 07:34:35,219 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5916, 2.8790, 3.4959, 4.6182, 4.0955, 4.0438, 3.0874, 2.5343], device='cuda:3'), covar=tensor([0.0648, 0.2024, 0.0896, 0.0523, 0.0767, 0.0455, 0.1289, 0.2028], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0220, 0.0189, 0.0223, 0.0232, 0.0189, 0.0205, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:34:39,833 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101168.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:35:22,402 INFO [train2.py:809] (3/4) Epoch 26, batch 1600, loss[ctc_loss=0.06453, att_loss=0.2355, loss=0.2013, over 16537.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.005781, over 45.00 utterances.], tot_loss[ctc_loss=0.0669, att_loss=0.2319, loss=0.1989, over 3272464.81 frames. utt_duration=1270 frames, utt_pad_proportion=0.0501, over 10315.67 utterances.], batch size: 45, lr: 4.14e-03, grad_scale: 16.0 2023-03-09 07:36:03,863 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-09 07:36:42,608 INFO [train2.py:809] (3/4) Epoch 26, batch 1650, loss[ctc_loss=0.07642, att_loss=0.2548, loss=0.2191, over 17018.00 frames. utt_duration=1286 frames, utt_pad_proportion=0.01089, over 53.00 utterances.], tot_loss[ctc_loss=0.06698, att_loss=0.2318, loss=0.1989, over 3272610.41 frames. utt_duration=1254 frames, utt_pad_proportion=0.05311, over 10453.13 utterances.], batch size: 53, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:36:50,258 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.931e+02 2.160e+02 2.616e+02 4.577e+02, threshold=4.320e+02, percent-clipped=2.0 2023-03-09 07:38:02,612 INFO [train2.py:809] (3/4) Epoch 26, batch 1700, loss[ctc_loss=0.08506, att_loss=0.2394, loss=0.2085, over 16966.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007202, over 50.00 utterances.], tot_loss[ctc_loss=0.06736, att_loss=0.2319, loss=0.199, over 3272707.58 frames. utt_duration=1244 frames, utt_pad_proportion=0.05504, over 10536.04 utterances.], batch size: 50, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:38:18,665 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3366, 5.2557, 5.1141, 3.4938, 5.0922, 4.9365, 4.7655, 3.1209], device='cuda:3'), covar=tensor([0.0101, 0.0095, 0.0238, 0.0774, 0.0094, 0.0171, 0.0242, 0.1124], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0107, 0.0110, 0.0114, 0.0090, 0.0118, 0.0103, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 07:39:07,993 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7434, 4.9787, 4.9675, 4.9357, 5.0197, 4.9754, 4.6311, 4.5221], device='cuda:3'), covar=tensor([0.1006, 0.0589, 0.0337, 0.0539, 0.0342, 0.0370, 0.0430, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0383, 0.0368, 0.0379, 0.0442, 0.0452, 0.0377, 0.0411], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 07:39:22,759 INFO [train2.py:809] (3/4) Epoch 26, batch 1750, loss[ctc_loss=0.06536, att_loss=0.2254, loss=0.1934, over 16013.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006893, over 40.00 utterances.], tot_loss[ctc_loss=0.06747, att_loss=0.2318, loss=0.199, over 3268734.93 frames. utt_duration=1247 frames, utt_pad_proportion=0.05604, over 10494.77 utterances.], batch size: 40, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:39:30,366 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 1.899e+02 2.202e+02 2.819e+02 5.865e+02, threshold=4.404e+02, percent-clipped=4.0 2023-03-09 07:39:39,168 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6130, 4.6038, 4.7325, 4.6908, 5.2930, 4.4963, 4.6709, 2.7920], device='cuda:3'), covar=tensor([0.0252, 0.0419, 0.0312, 0.0381, 0.0681, 0.0267, 0.0340, 0.1612], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0215, 0.0210, 0.0228, 0.0385, 0.0187, 0.0199, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:40:11,212 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-09 07:40:26,921 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101386.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:40:41,617 INFO [train2.py:809] (3/4) Epoch 26, batch 1800, loss[ctc_loss=0.0928, att_loss=0.2561, loss=0.2234, over 17389.00 frames. utt_duration=881.6 frames, utt_pad_proportion=0.07195, over 79.00 utterances.], tot_loss[ctc_loss=0.06757, att_loss=0.2321, loss=0.1992, over 3274869.66 frames. utt_duration=1259 frames, utt_pad_proportion=0.05215, over 10418.12 utterances.], batch size: 79, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:41:38,747 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7834, 5.1729, 4.9452, 5.1118, 5.2321, 4.9089, 3.7013, 5.1294], device='cuda:3'), covar=tensor([0.0104, 0.0106, 0.0134, 0.0073, 0.0083, 0.0114, 0.0627, 0.0140], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0092, 0.0116, 0.0073, 0.0078, 0.0090, 0.0106, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:41:43,390 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101434.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 07:41:53,375 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6470, 4.8953, 4.5084, 4.9273, 4.4056, 4.5650, 4.9868, 4.8310], device='cuda:3'), covar=tensor([0.0625, 0.0296, 0.0793, 0.0398, 0.0464, 0.0363, 0.0267, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0338, 0.0379, 0.0373, 0.0339, 0.0247, 0.0319, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 07:42:01,375 INFO [train2.py:809] (3/4) Epoch 26, batch 1850, loss[ctc_loss=0.06292, att_loss=0.2384, loss=0.2033, over 17410.00 frames. utt_duration=1107 frames, utt_pad_proportion=0.03231, over 63.00 utterances.], tot_loss[ctc_loss=0.06711, att_loss=0.2322, loss=0.1992, over 3280711.34 frames. utt_duration=1264 frames, utt_pad_proportion=0.04927, over 10392.18 utterances.], batch size: 63, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:42:08,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 1.881e+02 2.093e+02 3.016e+02 9.107e+02, threshold=4.186e+02, percent-clipped=4.0 2023-03-09 07:42:29,674 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101463.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:43:19,785 INFO [train2.py:809] (3/4) Epoch 26, batch 1900, loss[ctc_loss=0.06478, att_loss=0.2256, loss=0.1935, over 16537.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005736, over 45.00 utterances.], tot_loss[ctc_loss=0.06833, att_loss=0.2333, loss=0.2003, over 3266825.81 frames. utt_duration=1224 frames, utt_pad_proportion=0.06277, over 10689.54 utterances.], batch size: 45, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:44:39,309 INFO [train2.py:809] (3/4) Epoch 26, batch 1950, loss[ctc_loss=0.0614, att_loss=0.2284, loss=0.195, over 16277.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007523, over 43.00 utterances.], tot_loss[ctc_loss=0.06781, att_loss=0.2328, loss=0.1998, over 3266764.69 frames. utt_duration=1218 frames, utt_pad_proportion=0.06471, over 10739.94 utterances.], batch size: 43, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:44:43,039 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-03-09 07:44:47,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.785e+02 2.199e+02 2.810e+02 5.538e+02, threshold=4.398e+02, percent-clipped=5.0 2023-03-09 07:44:50,866 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6918, 3.2246, 3.8817, 3.2646, 3.6131, 4.7162, 4.6206, 3.4931], device='cuda:3'), covar=tensor([0.0352, 0.1656, 0.1265, 0.1268, 0.1162, 0.0936, 0.0562, 0.1163], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0253, 0.0291, 0.0225, 0.0272, 0.0383, 0.0274, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:45:17,536 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5629, 4.9437, 4.8057, 4.8693, 5.0072, 4.7640, 3.4110, 4.8828], device='cuda:3'), covar=tensor([0.0137, 0.0125, 0.0134, 0.0100, 0.0099, 0.0123, 0.0743, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0092, 0.0117, 0.0072, 0.0078, 0.0090, 0.0106, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:45:57,906 INFO [train2.py:809] (3/4) Epoch 26, batch 2000, loss[ctc_loss=0.05621, att_loss=0.2109, loss=0.18, over 15871.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01011, over 39.00 utterances.], tot_loss[ctc_loss=0.06749, att_loss=0.233, loss=0.1999, over 3266120.50 frames. utt_duration=1215 frames, utt_pad_proportion=0.06567, over 10770.12 utterances.], batch size: 39, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:46:50,426 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5193, 2.4776, 4.9490, 3.9248, 3.0808, 4.1997, 4.7401, 4.6679], device='cuda:3'), covar=tensor([0.0296, 0.1654, 0.0241, 0.0882, 0.1611, 0.0270, 0.0195, 0.0243], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0248, 0.0215, 0.0324, 0.0271, 0.0234, 0.0204, 0.0231], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:46:51,836 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4031, 2.9361, 3.6891, 2.9953, 3.4340, 4.5382, 4.4106, 3.1013], device='cuda:3'), covar=tensor([0.0384, 0.1718, 0.1135, 0.1447, 0.1155, 0.0866, 0.0497, 0.1325], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0250, 0.0288, 0.0223, 0.0269, 0.0379, 0.0272, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:47:17,705 INFO [train2.py:809] (3/4) Epoch 26, batch 2050, loss[ctc_loss=0.07048, att_loss=0.2337, loss=0.2011, over 16618.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005857, over 47.00 utterances.], tot_loss[ctc_loss=0.06684, att_loss=0.2329, loss=0.1997, over 3275073.54 frames. utt_duration=1235 frames, utt_pad_proportion=0.05796, over 10616.40 utterances.], batch size: 47, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:47:26,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.755e+02 2.108e+02 2.595e+02 4.367e+02, threshold=4.215e+02, percent-clipped=0.0 2023-03-09 07:47:50,897 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2839, 2.9775, 3.6574, 3.0732, 3.3947, 4.4246, 4.3030, 3.2650], device='cuda:3'), covar=tensor([0.0408, 0.1592, 0.1166, 0.1199, 0.1079, 0.0842, 0.0532, 0.1147], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0248, 0.0286, 0.0221, 0.0267, 0.0376, 0.0270, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:47:59,093 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6649, 3.1779, 3.6565, 4.5597, 4.0560, 4.0596, 3.1486, 2.5660], device='cuda:3'), covar=tensor([0.0639, 0.1765, 0.0843, 0.0563, 0.0970, 0.0508, 0.1377, 0.2068], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0220, 0.0188, 0.0223, 0.0234, 0.0189, 0.0204, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:48:19,674 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101684.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:48:37,218 INFO [train2.py:809] (3/4) Epoch 26, batch 2100, loss[ctc_loss=0.07973, att_loss=0.2471, loss=0.2136, over 16698.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005978, over 46.00 utterances.], tot_loss[ctc_loss=0.06812, att_loss=0.2338, loss=0.2007, over 3269772.41 frames. utt_duration=1201 frames, utt_pad_proportion=0.0689, over 10900.70 utterances.], batch size: 46, lr: 4.13e-03, grad_scale: 16.0 2023-03-09 07:48:46,601 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1856, 5.1347, 5.0042, 3.1829, 5.0352, 4.7367, 4.4819, 3.0393], device='cuda:3'), covar=tensor([0.0112, 0.0105, 0.0234, 0.0899, 0.0090, 0.0189, 0.0295, 0.1131], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0107, 0.0110, 0.0113, 0.0089, 0.0118, 0.0103, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 07:49:26,819 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6212, 5.0158, 4.8320, 4.9442, 5.0529, 4.8076, 3.5155, 4.9262], device='cuda:3'), covar=tensor([0.0125, 0.0131, 0.0145, 0.0116, 0.0114, 0.0123, 0.0726, 0.0259], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0094, 0.0119, 0.0074, 0.0080, 0.0092, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-09 07:49:55,492 INFO [train2.py:809] (3/4) Epoch 26, batch 2150, loss[ctc_loss=0.06974, att_loss=0.2516, loss=0.2153, over 16995.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009054, over 51.00 utterances.], tot_loss[ctc_loss=0.06768, att_loss=0.2334, loss=0.2002, over 3275325.18 frames. utt_duration=1195 frames, utt_pad_proportion=0.06791, over 10975.50 utterances.], batch size: 51, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:49:55,862 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101745.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:50:03,017 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 1.940e+02 2.256e+02 2.810e+02 5.136e+02, threshold=4.511e+02, percent-clipped=2.0 2023-03-09 07:50:23,112 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101763.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:51:13,848 INFO [train2.py:809] (3/4) Epoch 26, batch 2200, loss[ctc_loss=0.05298, att_loss=0.2052, loss=0.1748, over 14512.00 frames. utt_duration=1816 frames, utt_pad_proportion=0.03374, over 32.00 utterances.], tot_loss[ctc_loss=0.06774, att_loss=0.2333, loss=0.2002, over 3266763.85 frames. utt_duration=1193 frames, utt_pad_proportion=0.07017, over 10968.49 utterances.], batch size: 32, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:51:33,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-09 07:51:38,370 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101811.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:51:54,478 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0962, 4.1847, 4.2178, 4.2460, 4.7227, 4.2299, 4.1072, 2.4731], device='cuda:3'), covar=tensor([0.0314, 0.0511, 0.0429, 0.0366, 0.0701, 0.0280, 0.0419, 0.1848], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0212, 0.0207, 0.0224, 0.0379, 0.0183, 0.0197, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:52:32,209 INFO [train2.py:809] (3/4) Epoch 26, batch 2250, loss[ctc_loss=0.06014, att_loss=0.2188, loss=0.1871, over 16181.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.00696, over 41.00 utterances.], tot_loss[ctc_loss=0.06861, att_loss=0.2339, loss=0.2008, over 3273183.45 frames. utt_duration=1200 frames, utt_pad_proportion=0.06685, over 10924.73 utterances.], batch size: 41, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:52:39,681 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.832e+02 2.219e+02 2.925e+02 8.188e+02, threshold=4.439e+02, percent-clipped=2.0 2023-03-09 07:52:44,863 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5362, 2.9975, 4.9802, 4.0684, 3.2900, 4.3919, 4.8665, 4.7510], device='cuda:3'), covar=tensor([0.0303, 0.1338, 0.0274, 0.0868, 0.1524, 0.0262, 0.0190, 0.0250], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0245, 0.0213, 0.0321, 0.0269, 0.0232, 0.0203, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:53:51,084 INFO [train2.py:809] (3/4) Epoch 26, batch 2300, loss[ctc_loss=0.09124, att_loss=0.2469, loss=0.2158, over 17050.00 frames. utt_duration=1288 frames, utt_pad_proportion=0.009717, over 53.00 utterances.], tot_loss[ctc_loss=0.0683, att_loss=0.2335, loss=0.2004, over 3278314.28 frames. utt_duration=1234 frames, utt_pad_proportion=0.05761, over 10643.87 utterances.], batch size: 53, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:54:30,196 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101920.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 07:54:59,224 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3618, 2.8237, 3.6827, 2.9243, 3.4644, 4.4773, 4.4092, 3.2215], device='cuda:3'), covar=tensor([0.0379, 0.1918, 0.1241, 0.1422, 0.1180, 0.1010, 0.0577, 0.1252], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0249, 0.0286, 0.0222, 0.0268, 0.0378, 0.0271, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 07:55:09,347 INFO [train2.py:809] (3/4) Epoch 26, batch 2350, loss[ctc_loss=0.09479, att_loss=0.2515, loss=0.2202, over 17265.00 frames. utt_duration=875.7 frames, utt_pad_proportion=0.08205, over 79.00 utterances.], tot_loss[ctc_loss=0.06797, att_loss=0.2329, loss=0.1999, over 3266520.70 frames. utt_duration=1213 frames, utt_pad_proportion=0.06433, over 10785.49 utterances.], batch size: 79, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:55:16,746 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 1.850e+02 2.322e+02 2.866e+02 7.956e+02, threshold=4.644e+02, percent-clipped=4.0 2023-03-09 07:55:17,245 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5962, 4.6615, 4.8059, 4.6563, 5.2885, 4.6344, 4.6166, 2.9766], device='cuda:3'), covar=tensor([0.0284, 0.0475, 0.0317, 0.0438, 0.0736, 0.0249, 0.0461, 0.1533], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0213, 0.0207, 0.0225, 0.0380, 0.0184, 0.0198, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:55:41,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-09 07:56:05,591 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101981.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 07:56:15,711 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101987.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:56:27,652 INFO [train2.py:809] (3/4) Epoch 26, batch 2400, loss[ctc_loss=0.05769, att_loss=0.2312, loss=0.1965, over 16902.00 frames. utt_duration=684.3 frames, utt_pad_proportion=0.1371, over 99.00 utterances.], tot_loss[ctc_loss=0.06704, att_loss=0.232, loss=0.199, over 3268297.70 frames. utt_duration=1243 frames, utt_pad_proportion=0.05618, over 10528.71 utterances.], batch size: 99, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:57:44,279 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102040.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:57:51,787 INFO [train2.py:809] (3/4) Epoch 26, batch 2450, loss[ctc_loss=0.06226, att_loss=0.2347, loss=0.2002, over 16960.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007018, over 50.00 utterances.], tot_loss[ctc_loss=0.06652, att_loss=0.2315, loss=0.1985, over 3268853.25 frames. utt_duration=1272 frames, utt_pad_proportion=0.04915, over 10293.14 utterances.], batch size: 50, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 07:57:53,834 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3797, 4.4600, 4.6230, 4.5858, 5.1961, 4.3794, 4.4179, 2.8328], device='cuda:3'), covar=tensor([0.0340, 0.0435, 0.0398, 0.0404, 0.0642, 0.0296, 0.0448, 0.1625], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0213, 0.0208, 0.0225, 0.0380, 0.0184, 0.0198, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 07:57:56,737 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102048.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 07:57:59,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 1.902e+02 2.246e+02 2.761e+02 8.723e+02, threshold=4.491e+02, percent-clipped=3.0 2023-03-09 07:59:11,508 INFO [train2.py:809] (3/4) Epoch 26, batch 2500, loss[ctc_loss=0.05732, att_loss=0.2332, loss=0.198, over 16483.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005725, over 46.00 utterances.], tot_loss[ctc_loss=0.0664, att_loss=0.231, loss=0.1981, over 3269175.94 frames. utt_duration=1266 frames, utt_pad_proportion=0.05081, over 10342.88 utterances.], batch size: 46, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 08:00:29,995 INFO [train2.py:809] (3/4) Epoch 26, batch 2550, loss[ctc_loss=0.07327, att_loss=0.2234, loss=0.1934, over 16009.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007203, over 40.00 utterances.], tot_loss[ctc_loss=0.06655, att_loss=0.2311, loss=0.1982, over 3269108.44 frames. utt_duration=1251 frames, utt_pad_proportion=0.05526, over 10463.43 utterances.], batch size: 40, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 08:00:37,741 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.703e+02 2.113e+02 2.564e+02 5.440e+02, threshold=4.227e+02, percent-clipped=2.0 2023-03-09 08:01:48,436 INFO [train2.py:809] (3/4) Epoch 26, batch 2600, loss[ctc_loss=0.06466, att_loss=0.2258, loss=0.1936, over 15967.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.005678, over 41.00 utterances.], tot_loss[ctc_loss=0.06578, att_loss=0.231, loss=0.198, over 3275870.28 frames. utt_duration=1272 frames, utt_pad_proportion=0.04821, over 10313.95 utterances.], batch size: 41, lr: 4.12e-03, grad_scale: 16.0 2023-03-09 08:02:09,275 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5620, 3.0494, 3.5221, 4.4710, 4.0208, 4.0419, 3.0833, 2.3538], device='cuda:3'), covar=tensor([0.0699, 0.1840, 0.0915, 0.0611, 0.0909, 0.0490, 0.1451, 0.2200], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0220, 0.0189, 0.0224, 0.0234, 0.0189, 0.0205, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:03:06,804 INFO [train2.py:809] (3/4) Epoch 26, batch 2650, loss[ctc_loss=0.08426, att_loss=0.2522, loss=0.2187, over 17280.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01233, over 55.00 utterances.], tot_loss[ctc_loss=0.06624, att_loss=0.2314, loss=0.1983, over 3274912.85 frames. utt_duration=1262 frames, utt_pad_proportion=0.05191, over 10391.86 utterances.], batch size: 55, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:03:14,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.711e+02 1.973e+02 2.429e+02 4.776e+02, threshold=3.946e+02, percent-clipped=2.0 2023-03-09 08:03:33,897 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5995, 3.1870, 3.7985, 3.4188, 3.6422, 4.6793, 4.5279, 3.5594], device='cuda:3'), covar=tensor([0.0388, 0.1532, 0.1154, 0.1073, 0.0993, 0.0888, 0.0534, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0250, 0.0289, 0.0223, 0.0269, 0.0380, 0.0272, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:03:55,843 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102276.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 08:04:18,282 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9700, 5.2474, 5.1560, 5.1872, 5.2749, 5.2132, 4.8162, 4.6928], device='cuda:3'), covar=tensor([0.1055, 0.0571, 0.0356, 0.0463, 0.0316, 0.0337, 0.0486, 0.0409], device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0376, 0.0364, 0.0374, 0.0435, 0.0444, 0.0375, 0.0406], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:04:25,661 INFO [train2.py:809] (3/4) Epoch 26, batch 2700, loss[ctc_loss=0.07205, att_loss=0.2408, loss=0.207, over 17290.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01094, over 55.00 utterances.], tot_loss[ctc_loss=0.06664, att_loss=0.2324, loss=0.1993, over 3273535.37 frames. utt_duration=1229 frames, utt_pad_proportion=0.05989, over 10665.35 utterances.], batch size: 55, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:05:37,439 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102340.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:05:41,919 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102343.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:05:44,687 INFO [train2.py:809] (3/4) Epoch 26, batch 2750, loss[ctc_loss=0.05598, att_loss=0.2354, loss=0.1995, over 17444.00 frames. utt_duration=1013 frames, utt_pad_proportion=0.04349, over 69.00 utterances.], tot_loss[ctc_loss=0.0666, att_loss=0.2323, loss=0.1992, over 3274665.94 frames. utt_duration=1234 frames, utt_pad_proportion=0.05845, over 10624.65 utterances.], batch size: 69, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:05:52,197 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.173e+02 1.911e+02 2.234e+02 2.672e+02 6.955e+02, threshold=4.468e+02, percent-clipped=5.0 2023-03-09 08:06:52,286 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102388.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:06:55,613 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4376, 4.3837, 4.5700, 4.5204, 5.0866, 4.4085, 4.3791, 2.6636], device='cuda:3'), covar=tensor([0.0304, 0.0455, 0.0423, 0.0401, 0.0758, 0.0293, 0.0417, 0.1721], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0216, 0.0211, 0.0227, 0.0384, 0.0186, 0.0201, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:07:02,737 INFO [train2.py:809] (3/4) Epoch 26, batch 2800, loss[ctc_loss=0.07323, att_loss=0.2485, loss=0.2135, over 17158.00 frames. utt_duration=694.8 frames, utt_pad_proportion=0.1272, over 99.00 utterances.], tot_loss[ctc_loss=0.06672, att_loss=0.2328, loss=0.1996, over 3278199.34 frames. utt_duration=1238 frames, utt_pad_proportion=0.0564, over 10607.77 utterances.], batch size: 99, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:07:25,022 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0136, 5.2624, 5.5301, 5.3531, 5.5389, 5.9452, 5.2824, 6.0296], device='cuda:3'), covar=tensor([0.0714, 0.0743, 0.0900, 0.1406, 0.1859, 0.0989, 0.0679, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0918, 0.0524, 0.0637, 0.0686, 0.0907, 0.0663, 0.0512, 0.0644], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:08:21,715 INFO [train2.py:809] (3/4) Epoch 26, batch 2850, loss[ctc_loss=0.05949, att_loss=0.2231, loss=0.1904, over 16008.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007342, over 40.00 utterances.], tot_loss[ctc_loss=0.06643, att_loss=0.2322, loss=0.199, over 3272922.37 frames. utt_duration=1227 frames, utt_pad_proportion=0.06005, over 10678.70 utterances.], batch size: 40, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:08:29,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.936e+02 2.281e+02 2.866e+02 5.493e+02, threshold=4.561e+02, percent-clipped=4.0 2023-03-09 08:08:57,572 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-09 08:09:19,585 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1095, 3.6983, 3.0713, 3.3530, 3.8537, 3.5655, 2.9090, 4.1198], device='cuda:3'), covar=tensor([0.0955, 0.0491, 0.1146, 0.0811, 0.0747, 0.0800, 0.0931, 0.0532], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0223, 0.0227, 0.0206, 0.0288, 0.0246, 0.0202, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 08:09:40,753 INFO [train2.py:809] (3/4) Epoch 26, batch 2900, loss[ctc_loss=0.04498, att_loss=0.2382, loss=0.1996, over 16538.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006317, over 45.00 utterances.], tot_loss[ctc_loss=0.0666, att_loss=0.2327, loss=0.1995, over 3270934.76 frames. utt_duration=1225 frames, utt_pad_proportion=0.06154, over 10696.61 utterances.], batch size: 45, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:11:00,386 INFO [train2.py:809] (3/4) Epoch 26, batch 2950, loss[ctc_loss=0.07393, att_loss=0.2158, loss=0.1875, over 15774.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.007651, over 38.00 utterances.], tot_loss[ctc_loss=0.06812, att_loss=0.2332, loss=0.2002, over 3260723.17 frames. utt_duration=1194 frames, utt_pad_proportion=0.07104, over 10934.52 utterances.], batch size: 38, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:11:08,373 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 1.971e+02 2.322e+02 2.777e+02 5.083e+02, threshold=4.643e+02, percent-clipped=3.0 2023-03-09 08:11:54,084 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102576.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 08:12:24,262 INFO [train2.py:809] (3/4) Epoch 26, batch 3000, loss[ctc_loss=0.05938, att_loss=0.2267, loss=0.1933, over 16280.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007173, over 43.00 utterances.], tot_loss[ctc_loss=0.06799, att_loss=0.2327, loss=0.1997, over 3259921.41 frames. utt_duration=1217 frames, utt_pad_proportion=0.06726, over 10729.59 utterances.], batch size: 43, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:12:24,262 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 08:12:38,598 INFO [train2.py:843] (3/4) Epoch 26, validation: ctc_loss=0.04046, att_loss=0.2348, loss=0.1959, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 08:12:38,599 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 08:13:27,789 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102624.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 08:14:00,136 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102643.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:14:03,026 INFO [train2.py:809] (3/4) Epoch 26, batch 3050, loss[ctc_loss=0.0638, att_loss=0.2366, loss=0.2021, over 17142.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01373, over 56.00 utterances.], tot_loss[ctc_loss=0.06822, att_loss=0.2331, loss=0.2001, over 3246604.15 frames. utt_duration=1198 frames, utt_pad_proportion=0.07485, over 10854.46 utterances.], batch size: 56, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:14:10,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.303e+02 1.917e+02 2.316e+02 2.727e+02 9.288e+02, threshold=4.632e+02, percent-clipped=2.0 2023-03-09 08:15:04,868 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-09 08:15:18,710 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102691.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:15:24,944 INFO [train2.py:809] (3/4) Epoch 26, batch 3100, loss[ctc_loss=0.1263, att_loss=0.2605, loss=0.2336, over 14246.00 frames. utt_duration=391.8 frames, utt_pad_proportion=0.3161, over 146.00 utterances.], tot_loss[ctc_loss=0.06807, att_loss=0.2327, loss=0.1997, over 3250939.34 frames. utt_duration=1197 frames, utt_pad_proportion=0.07481, over 10880.18 utterances.], batch size: 146, lr: 4.11e-03, grad_scale: 16.0 2023-03-09 08:16:47,148 INFO [train2.py:809] (3/4) Epoch 26, batch 3150, loss[ctc_loss=0.06148, att_loss=0.2367, loss=0.2016, over 17425.00 frames. utt_duration=1108 frames, utt_pad_proportion=0.03162, over 63.00 utterances.], tot_loss[ctc_loss=0.0673, att_loss=0.2321, loss=0.1991, over 3245786.68 frames. utt_duration=1200 frames, utt_pad_proportion=0.07497, over 10831.13 utterances.], batch size: 63, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:16:54,909 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 1.851e+02 2.183e+02 2.802e+02 6.228e+02, threshold=4.366e+02, percent-clipped=1.0 2023-03-09 08:18:06,656 INFO [train2.py:809] (3/4) Epoch 26, batch 3200, loss[ctc_loss=0.05028, att_loss=0.2368, loss=0.1995, over 17002.00 frames. utt_duration=1335 frames, utt_pad_proportion=0.009629, over 51.00 utterances.], tot_loss[ctc_loss=0.06751, att_loss=0.2316, loss=0.1988, over 3237888.07 frames. utt_duration=1198 frames, utt_pad_proportion=0.07762, over 10822.76 utterances.], batch size: 51, lr: 4.10e-03, grad_scale: 32.0 2023-03-09 08:18:22,935 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2234, 5.5208, 5.4333, 5.4028, 5.5420, 5.4705, 5.1704, 4.9711], device='cuda:3'), covar=tensor([0.0999, 0.0485, 0.0302, 0.0484, 0.0246, 0.0318, 0.0360, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0377, 0.0365, 0.0375, 0.0436, 0.0442, 0.0374, 0.0408], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:18:38,050 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0422, 5.0464, 4.9037, 2.3676, 2.0771, 3.0256, 2.3129, 3.9320], device='cuda:3'), covar=tensor([0.0735, 0.0306, 0.0259, 0.5103, 0.5687, 0.2396, 0.3843, 0.1546], device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0295, 0.0277, 0.0250, 0.0337, 0.0329, 0.0260, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 08:18:56,427 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7124, 5.1853, 4.9923, 5.0780, 5.1686, 4.7996, 3.5414, 5.1391], device='cuda:3'), covar=tensor([0.0132, 0.0113, 0.0158, 0.0117, 0.0126, 0.0138, 0.0687, 0.0227], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0095, 0.0119, 0.0074, 0.0080, 0.0092, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-09 08:19:25,814 INFO [train2.py:809] (3/4) Epoch 26, batch 3250, loss[ctc_loss=0.07289, att_loss=0.2371, loss=0.2042, over 17014.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008094, over 51.00 utterances.], tot_loss[ctc_loss=0.06759, att_loss=0.2318, loss=0.1989, over 3244940.92 frames. utt_duration=1227 frames, utt_pad_proportion=0.06929, over 10589.46 utterances.], batch size: 51, lr: 4.10e-03, grad_scale: 32.0 2023-03-09 08:19:33,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 1.810e+02 2.101e+02 2.825e+02 6.306e+02, threshold=4.202e+02, percent-clipped=5.0 2023-03-09 08:19:35,356 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102851.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:20:08,232 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-03-09 08:20:32,513 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102887.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:20:44,488 INFO [train2.py:809] (3/4) Epoch 26, batch 3300, loss[ctc_loss=0.06094, att_loss=0.2403, loss=0.2044, over 16975.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.006213, over 50.00 utterances.], tot_loss[ctc_loss=0.06731, att_loss=0.2323, loss=0.1993, over 3253260.20 frames. utt_duration=1226 frames, utt_pad_proportion=0.06687, over 10629.07 utterances.], batch size: 50, lr: 4.10e-03, grad_scale: 32.0 2023-03-09 08:21:11,632 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102912.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:22:03,745 INFO [train2.py:809] (3/4) Epoch 26, batch 3350, loss[ctc_loss=0.07861, att_loss=0.2453, loss=0.212, over 17486.00 frames. utt_duration=1112 frames, utt_pad_proportion=0.02737, over 63.00 utterances.], tot_loss[ctc_loss=0.06749, att_loss=0.2337, loss=0.2004, over 3265983.91 frames. utt_duration=1212 frames, utt_pad_proportion=0.06739, over 10796.55 utterances.], batch size: 63, lr: 4.10e-03, grad_scale: 32.0 2023-03-09 08:22:04,235 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6709, 2.7566, 3.8941, 3.4936, 3.0152, 3.6288, 3.6758, 3.7117], device='cuda:3'), covar=tensor([0.0400, 0.1239, 0.0265, 0.0839, 0.1339, 0.0365, 0.0321, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0247, 0.0215, 0.0324, 0.0271, 0.0234, 0.0207, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:22:08,722 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102948.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:22:11,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.718e+02 2.165e+02 2.625e+02 3.837e+02, threshold=4.330e+02, percent-clipped=0.0 2023-03-09 08:22:52,347 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0285, 5.3212, 5.2407, 5.2289, 5.3140, 5.2763, 4.9525, 4.7628], device='cuda:3'), covar=tensor([0.1041, 0.0472, 0.0362, 0.0482, 0.0287, 0.0310, 0.0423, 0.0354], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0376, 0.0365, 0.0375, 0.0437, 0.0442, 0.0374, 0.0408], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:23:20,928 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1920, 5.4229, 5.7117, 5.4990, 5.6655, 6.1713, 5.4206, 6.2000], device='cuda:3'), covar=tensor([0.0721, 0.0659, 0.0817, 0.1416, 0.1799, 0.0812, 0.0593, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0919, 0.0527, 0.0636, 0.0689, 0.0908, 0.0663, 0.0513, 0.0645], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:23:22,227 INFO [train2.py:809] (3/4) Epoch 26, batch 3400, loss[ctc_loss=0.05016, att_loss=0.2029, loss=0.1723, over 15388.00 frames. utt_duration=1760 frames, utt_pad_proportion=0.009422, over 35.00 utterances.], tot_loss[ctc_loss=0.06707, att_loss=0.2332, loss=0.2, over 3268593.01 frames. utt_duration=1226 frames, utt_pad_proportion=0.06384, over 10678.77 utterances.], batch size: 35, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:23:37,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-09 08:23:53,329 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-03-09 08:23:54,147 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0083, 6.2331, 5.7792, 5.8996, 5.9327, 5.3994, 5.7047, 5.3357], device='cuda:3'), covar=tensor([0.1225, 0.0878, 0.0903, 0.0802, 0.0795, 0.1571, 0.2191, 0.2346], device='cuda:3'), in_proj_covar=tensor([0.0556, 0.0637, 0.0490, 0.0483, 0.0451, 0.0489, 0.0640, 0.0548], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:24:40,908 INFO [train2.py:809] (3/4) Epoch 26, batch 3450, loss[ctc_loss=0.09161, att_loss=0.2392, loss=0.2097, over 15996.00 frames. utt_duration=1601 frames, utt_pad_proportion=0.008008, over 40.00 utterances.], tot_loss[ctc_loss=0.06682, att_loss=0.2327, loss=0.1995, over 3262992.47 frames. utt_duration=1254 frames, utt_pad_proportion=0.05828, over 10421.81 utterances.], batch size: 40, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:24:50,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 1.811e+02 2.239e+02 2.771e+02 6.342e+02, threshold=4.479e+02, percent-clipped=4.0 2023-03-09 08:25:04,758 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6580, 3.7573, 3.9290, 2.4451, 2.3864, 2.8775, 2.4554, 3.5172], device='cuda:3'), covar=tensor([0.0668, 0.0482, 0.0378, 0.3674, 0.3784, 0.2103, 0.2861, 0.1266], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0296, 0.0278, 0.0250, 0.0338, 0.0332, 0.0261, 0.0369], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 08:26:00,925 INFO [train2.py:809] (3/4) Epoch 26, batch 3500, loss[ctc_loss=0.06691, att_loss=0.2158, loss=0.1861, over 14498.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.0332, over 32.00 utterances.], tot_loss[ctc_loss=0.06664, att_loss=0.2322, loss=0.1991, over 3258932.86 frames. utt_duration=1219 frames, utt_pad_proportion=0.06724, over 10706.12 utterances.], batch size: 32, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:26:05,879 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103098.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:26:59,866 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103132.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:27:19,712 INFO [train2.py:809] (3/4) Epoch 26, batch 3550, loss[ctc_loss=0.05879, att_loss=0.2308, loss=0.1964, over 16476.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.005921, over 46.00 utterances.], tot_loss[ctc_loss=0.06608, att_loss=0.2321, loss=0.1989, over 3265560.94 frames. utt_duration=1225 frames, utt_pad_proportion=0.06283, over 10673.00 utterances.], batch size: 46, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:27:28,876 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.344e+02 1.850e+02 2.221e+02 2.819e+02 5.198e+02, threshold=4.442e+02, percent-clipped=1.0 2023-03-09 08:27:42,870 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103159.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:28:31,631 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9884, 4.9539, 4.8153, 2.2285, 2.1022, 2.7609, 2.3946, 3.7585], device='cuda:3'), covar=tensor([0.0762, 0.0326, 0.0286, 0.4756, 0.5084, 0.2590, 0.3502, 0.1702], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0297, 0.0278, 0.0249, 0.0337, 0.0331, 0.0261, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 08:28:36,286 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103193.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:28:38,762 INFO [train2.py:809] (3/4) Epoch 26, batch 3600, loss[ctc_loss=0.0626, att_loss=0.2181, loss=0.187, over 16132.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.005972, over 42.00 utterances.], tot_loss[ctc_loss=0.06689, att_loss=0.2325, loss=0.1994, over 3269416.89 frames. utt_duration=1202 frames, utt_pad_proportion=0.06764, over 10891.59 utterances.], batch size: 42, lr: 4.10e-03, grad_scale: 16.0 2023-03-09 08:28:57,785 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103207.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:29:55,730 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103243.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:29:58,626 INFO [train2.py:809] (3/4) Epoch 26, batch 3650, loss[ctc_loss=0.04733, att_loss=0.2099, loss=0.1774, over 11456.00 frames. utt_duration=1834 frames, utt_pad_proportion=0.1909, over 25.00 utterances.], tot_loss[ctc_loss=0.06618, att_loss=0.2317, loss=0.1986, over 3261819.98 frames. utt_duration=1205 frames, utt_pad_proportion=0.06894, over 10842.17 utterances.], batch size: 25, lr: 4.09e-03, grad_scale: 16.0 2023-03-09 08:30:07,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.817e+02 2.258e+02 2.684e+02 6.066e+02, threshold=4.515e+02, percent-clipped=3.0 2023-03-09 08:30:33,851 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9641, 4.9568, 4.8645, 2.2183, 1.9648, 2.7749, 2.4594, 3.8574], device='cuda:3'), covar=tensor([0.0746, 0.0308, 0.0246, 0.4525, 0.5393, 0.2556, 0.3486, 0.1510], device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0298, 0.0278, 0.0249, 0.0337, 0.0331, 0.0261, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 08:31:17,482 INFO [train2.py:809] (3/4) Epoch 26, batch 3700, loss[ctc_loss=0.05053, att_loss=0.2137, loss=0.1811, over 16541.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006167, over 45.00 utterances.], tot_loss[ctc_loss=0.06604, att_loss=0.2311, loss=0.1981, over 3260419.63 frames. utt_duration=1228 frames, utt_pad_proportion=0.06475, over 10631.02 utterances.], batch size: 45, lr: 4.09e-03, grad_scale: 16.0 2023-03-09 08:32:36,354 INFO [train2.py:809] (3/4) Epoch 26, batch 3750, loss[ctc_loss=0.07669, att_loss=0.2497, loss=0.2151, over 17107.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.0159, over 56.00 utterances.], tot_loss[ctc_loss=0.06702, att_loss=0.2328, loss=0.1996, over 3270856.02 frames. utt_duration=1202 frames, utt_pad_proportion=0.0676, over 10898.55 utterances.], batch size: 56, lr: 4.09e-03, grad_scale: 16.0 2023-03-09 08:32:45,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 1.935e+02 2.309e+02 3.024e+02 8.462e+02, threshold=4.618e+02, percent-clipped=2.0 2023-03-09 08:32:55,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0560, 4.9871, 4.7656, 2.8382, 4.7238, 4.6775, 4.2560, 2.7368], device='cuda:3'), covar=tensor([0.0111, 0.0119, 0.0291, 0.1036, 0.0131, 0.0210, 0.0328, 0.1347], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0108, 0.0111, 0.0114, 0.0091, 0.0119, 0.0103, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:33:04,534 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-09 08:33:10,150 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103366.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 08:33:31,023 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-09 08:33:56,330 INFO [train2.py:809] (3/4) Epoch 26, batch 3800, loss[ctc_loss=0.05366, att_loss=0.2341, loss=0.198, over 16484.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.006383, over 46.00 utterances.], tot_loss[ctc_loss=0.06659, att_loss=0.2323, loss=0.1992, over 3269936.50 frames. utt_duration=1210 frames, utt_pad_proportion=0.06602, over 10824.84 utterances.], batch size: 46, lr: 4.09e-03, grad_scale: 16.0 2023-03-09 08:34:04,666 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3195, 5.2026, 5.0204, 2.9496, 5.0197, 4.8666, 4.7320, 3.0892], device='cuda:3'), covar=tensor([0.0094, 0.0112, 0.0281, 0.1013, 0.0105, 0.0189, 0.0232, 0.1167], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0107, 0.0111, 0.0114, 0.0091, 0.0119, 0.0102, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:34:48,489 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103427.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 08:35:16,369 INFO [train2.py:809] (3/4) Epoch 26, batch 3850, loss[ctc_loss=0.06655, att_loss=0.24, loss=0.2053, over 16343.00 frames. utt_duration=1454 frames, utt_pad_proportion=0.005305, over 45.00 utterances.], tot_loss[ctc_loss=0.06572, att_loss=0.2312, loss=0.1981, over 3264370.33 frames. utt_duration=1237 frames, utt_pad_proportion=0.06058, over 10570.82 utterances.], batch size: 45, lr: 4.09e-03, grad_scale: 8.0 2023-03-09 08:35:26,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 1.752e+02 2.146e+02 2.591e+02 4.293e+02, threshold=4.292e+02, percent-clipped=0.0 2023-03-09 08:35:30,368 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103454.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:35:51,733 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5896, 5.0467, 4.8498, 4.9961, 5.0968, 4.7226, 3.7234, 4.9723], device='cuda:3'), covar=tensor([0.0133, 0.0126, 0.0136, 0.0085, 0.0086, 0.0123, 0.0610, 0.0231], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0094, 0.0118, 0.0074, 0.0080, 0.0091, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:36:22,168 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103488.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:36:29,723 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2851, 3.8753, 3.2969, 3.6403, 4.1138, 3.8221, 3.4921, 4.4187], device='cuda:3'), covar=tensor([0.0887, 0.0520, 0.1076, 0.0626, 0.0711, 0.0685, 0.0637, 0.0473], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0225, 0.0231, 0.0207, 0.0289, 0.0248, 0.0203, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 08:36:32,451 INFO [train2.py:809] (3/4) Epoch 26, batch 3900, loss[ctc_loss=0.06111, att_loss=0.2343, loss=0.1997, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005543, over 47.00 utterances.], tot_loss[ctc_loss=0.06655, att_loss=0.2321, loss=0.199, over 3263456.34 frames. utt_duration=1201 frames, utt_pad_proportion=0.06934, over 10885.74 utterances.], batch size: 47, lr: 4.09e-03, grad_scale: 8.0 2023-03-09 08:36:50,997 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103507.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:37:16,388 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9998, 6.2503, 5.7631, 5.9372, 5.9152, 5.3508, 5.7504, 5.4515], device='cuda:3'), covar=tensor([0.1422, 0.0906, 0.0918, 0.0780, 0.0987, 0.1546, 0.2316, 0.2392], device='cuda:3'), in_proj_covar=tensor([0.0553, 0.0631, 0.0488, 0.0476, 0.0448, 0.0485, 0.0639, 0.0547], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:37:47,349 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103543.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:37:50,085 INFO [train2.py:809] (3/4) Epoch 26, batch 3950, loss[ctc_loss=0.06955, att_loss=0.2356, loss=0.2024, over 16474.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006237, over 46.00 utterances.], tot_loss[ctc_loss=0.06604, att_loss=0.232, loss=0.1988, over 3269136.87 frames. utt_duration=1211 frames, utt_pad_proportion=0.06632, over 10814.22 utterances.], batch size: 46, lr: 4.09e-03, grad_scale: 8.0 2023-03-09 08:37:55,734 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 08:38:00,647 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.745e+02 2.083e+02 2.654e+02 4.442e+02, threshold=4.166e+02, percent-clipped=1.0 2023-03-09 08:38:05,196 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103555.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:39:05,369 INFO [train2.py:809] (3/4) Epoch 27, batch 0, loss[ctc_loss=0.05525, att_loss=0.2152, loss=0.1832, over 15974.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005242, over 41.00 utterances.], tot_loss[ctc_loss=0.05525, att_loss=0.2152, loss=0.1832, over 15974.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005242, over 41.00 utterances.], batch size: 41, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:39:05,370 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 08:39:17,375 INFO [train2.py:843] (3/4) Epoch 27, validation: ctc_loss=0.04075, att_loss=0.2342, loss=0.1955, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 08:39:17,376 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 08:39:36,997 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103591.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:40:05,624 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-09 08:40:36,385 INFO [train2.py:809] (3/4) Epoch 27, batch 50, loss[ctc_loss=0.05155, att_loss=0.2235, loss=0.1891, over 16333.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.00611, over 45.00 utterances.], tot_loss[ctc_loss=0.06654, att_loss=0.232, loss=0.1989, over 738822.84 frames. utt_duration=1177 frames, utt_pad_proportion=0.06687, over 2514.97 utterances.], batch size: 45, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:40:53,879 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-09 08:41:14,260 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.229e+02 1.910e+02 2.259e+02 2.713e+02 6.930e+02, threshold=4.518e+02, percent-clipped=6.0 2023-03-09 08:41:56,150 INFO [train2.py:809] (3/4) Epoch 27, batch 100, loss[ctc_loss=0.05591, att_loss=0.2426, loss=0.2053, over 17036.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.006827, over 51.00 utterances.], tot_loss[ctc_loss=0.06596, att_loss=0.2317, loss=0.1986, over 1302389.16 frames. utt_duration=1185 frames, utt_pad_proportion=0.06437, over 4402.77 utterances.], batch size: 51, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:42:22,308 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1776, 3.8353, 3.2588, 3.4652, 4.0418, 3.7113, 3.1803, 4.2988], device='cuda:3'), covar=tensor([0.0890, 0.0516, 0.1080, 0.0729, 0.0682, 0.0758, 0.0819, 0.0458], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0226, 0.0233, 0.0208, 0.0291, 0.0249, 0.0205, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 08:42:52,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-09 08:43:04,838 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103722.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 08:43:11,062 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0050, 4.3671, 4.3925, 4.5793, 2.9339, 4.2718, 2.7896, 1.8398], device='cuda:3'), covar=tensor([0.0546, 0.0312, 0.0641, 0.0238, 0.1428, 0.0280, 0.1362, 0.1683], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0183, 0.0261, 0.0175, 0.0221, 0.0163, 0.0230, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:43:15,221 INFO [train2.py:809] (3/4) Epoch 27, batch 150, loss[ctc_loss=0.06242, att_loss=0.2213, loss=0.1895, over 16166.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.006708, over 41.00 utterances.], tot_loss[ctc_loss=0.0666, att_loss=0.2319, loss=0.1989, over 1739130.60 frames. utt_duration=1207 frames, utt_pad_proportion=0.06144, over 5769.99 utterances.], batch size: 41, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:43:33,677 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103740.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:43:34,223 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-03-09 08:43:38,259 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5808, 4.8168, 4.8195, 4.7974, 5.4220, 4.4970, 4.8584, 2.8355], device='cuda:3'), covar=tensor([0.0278, 0.0371, 0.0288, 0.0386, 0.0649, 0.0301, 0.0298, 0.1553], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0220, 0.0216, 0.0233, 0.0389, 0.0191, 0.0204, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:43:52,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 1.770e+02 2.085e+02 2.452e+02 6.368e+02, threshold=4.170e+02, percent-clipped=1.0 2023-03-09 08:43:56,613 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103754.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:44:12,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-09 08:44:13,426 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:44:34,216 INFO [train2.py:809] (3/4) Epoch 27, batch 200, loss[ctc_loss=0.05819, att_loss=0.1956, loss=0.1682, over 15500.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008737, over 36.00 utterances.], tot_loss[ctc_loss=0.06703, att_loss=0.2325, loss=0.1994, over 2083266.59 frames. utt_duration=1188 frames, utt_pad_proportion=0.06424, over 7021.41 utterances.], batch size: 36, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:44:38,355 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7550, 3.4722, 3.5154, 3.0429, 3.5425, 3.4850, 3.5379, 2.7014], device='cuda:3'), covar=tensor([0.1244, 0.1375, 0.1515, 0.2930, 0.1256, 0.2604, 0.0984, 0.3059], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0205, 0.0220, 0.0270, 0.0179, 0.0281, 0.0200, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:44:49,605 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103788.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:45:10,693 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103801.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:45:11,941 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103802.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:45:49,630 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103826.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:45:53,753 INFO [train2.py:809] (3/4) Epoch 27, batch 250, loss[ctc_loss=0.05323, att_loss=0.2319, loss=0.1962, over 16324.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006399, over 45.00 utterances.], tot_loss[ctc_loss=0.06614, att_loss=0.2325, loss=0.1993, over 2349568.36 frames. utt_duration=1205 frames, utt_pad_proportion=0.06037, over 7806.80 utterances.], batch size: 45, lr: 4.01e-03, grad_scale: 8.0 2023-03-09 08:46:05,849 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103836.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:46:30,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.879e+02 2.239e+02 2.738e+02 5.518e+02, threshold=4.478e+02, percent-clipped=1.0 2023-03-09 08:47:08,876 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103876.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:47:13,081 INFO [train2.py:809] (3/4) Epoch 27, batch 300, loss[ctc_loss=0.04744, att_loss=0.2261, loss=0.1904, over 16385.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.008674, over 44.00 utterances.], tot_loss[ctc_loss=0.06574, att_loss=0.232, loss=0.1987, over 2555317.83 frames. utt_duration=1239 frames, utt_pad_proportion=0.0545, over 8257.73 utterances.], batch size: 44, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:47:57,125 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0808, 5.3297, 5.2708, 5.2464, 5.3540, 5.3212, 4.9570, 4.7598], device='cuda:3'), covar=tensor([0.1006, 0.0508, 0.0329, 0.0583, 0.0302, 0.0317, 0.0462, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0377, 0.0368, 0.0374, 0.0438, 0.0441, 0.0375, 0.0407], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:47:57,184 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9779, 5.0042, 4.4574, 2.4239, 4.7999, 4.7406, 4.0124, 2.1232], device='cuda:3'), covar=tensor([0.0209, 0.0168, 0.0544, 0.1700, 0.0154, 0.0257, 0.0561, 0.2501], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0107, 0.0111, 0.0113, 0.0090, 0.0118, 0.0102, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:48:32,102 INFO [train2.py:809] (3/4) Epoch 27, batch 350, loss[ctc_loss=0.0628, att_loss=0.2255, loss=0.1929, over 16279.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006645, over 43.00 utterances.], tot_loss[ctc_loss=0.06511, att_loss=0.2312, loss=0.198, over 2709367.67 frames. utt_duration=1261 frames, utt_pad_proportion=0.05093, over 8605.85 utterances.], batch size: 43, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:48:45,742 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103937.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 08:48:45,815 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1042, 5.0659, 4.9973, 2.1136, 1.9700, 2.8982, 2.2154, 3.9829], device='cuda:3'), covar=tensor([0.0701, 0.0296, 0.0212, 0.5179, 0.5575, 0.2481, 0.3860, 0.1457], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0301, 0.0280, 0.0252, 0.0341, 0.0334, 0.0264, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 08:48:54,460 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5838, 5.8845, 5.3898, 5.5474, 5.5135, 4.9246, 5.2554, 5.0120], device='cuda:3'), covar=tensor([0.1339, 0.0857, 0.0981, 0.0896, 0.1032, 0.1752, 0.2266, 0.2407], device='cuda:3'), in_proj_covar=tensor([0.0553, 0.0632, 0.0486, 0.0477, 0.0449, 0.0482, 0.0637, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:49:08,958 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 1.861e+02 2.238e+02 2.752e+02 4.825e+02, threshold=4.477e+02, percent-clipped=1.0 2023-03-09 08:49:51,234 INFO [train2.py:809] (3/4) Epoch 27, batch 400, loss[ctc_loss=0.06751, att_loss=0.2483, loss=0.2121, over 17363.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02099, over 59.00 utterances.], tot_loss[ctc_loss=0.06534, att_loss=0.2311, loss=0.198, over 2833520.95 frames. utt_duration=1281 frames, utt_pad_proportion=0.04655, over 8858.96 utterances.], batch size: 59, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:50:02,865 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8450, 3.7337, 3.1937, 3.2439, 4.0132, 3.5754, 2.6758, 4.0911], device='cuda:3'), covar=tensor([0.1263, 0.0561, 0.1085, 0.0873, 0.0726, 0.0801, 0.1148, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0224, 0.0231, 0.0206, 0.0290, 0.0248, 0.0204, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 08:50:09,019 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103990.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:50:19,677 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103997.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:51:03,267 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4542, 3.0728, 3.3974, 4.4550, 3.9260, 3.9512, 3.1031, 2.1845], device='cuda:3'), covar=tensor([0.0726, 0.1724, 0.0900, 0.0500, 0.0877, 0.0564, 0.1442, 0.2307], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0219, 0.0187, 0.0224, 0.0233, 0.0190, 0.0204, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:51:04,815 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104022.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 08:51:15,262 INFO [train2.py:809] (3/4) Epoch 27, batch 450, loss[ctc_loss=0.07011, att_loss=0.2363, loss=0.2031, over 17394.00 frames. utt_duration=1010 frames, utt_pad_proportion=0.0471, over 69.00 utterances.], tot_loss[ctc_loss=0.06612, att_loss=0.2316, loss=0.1985, over 2937905.97 frames. utt_duration=1268 frames, utt_pad_proportion=0.0473, over 9279.98 utterances.], batch size: 69, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:51:48,294 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104049.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:51:51,433 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104051.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:51:52,513 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 1.790e+02 2.279e+02 2.671e+02 6.527e+02, threshold=4.558e+02, percent-clipped=3.0 2023-03-09 08:52:02,868 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104058.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:52:20,985 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104070.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 08:52:21,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-09 08:52:27,112 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0202, 5.2819, 5.5638, 5.4076, 5.5354, 5.9830, 5.2712, 6.0264], device='cuda:3'), covar=tensor([0.0737, 0.0770, 0.0825, 0.1356, 0.1739, 0.0876, 0.0743, 0.0778], device='cuda:3'), in_proj_covar=tensor([0.0901, 0.0522, 0.0631, 0.0681, 0.0898, 0.0659, 0.0506, 0.0638], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:52:27,179 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0453, 5.3446, 4.8826, 5.4023, 4.7717, 5.0258, 5.4751, 5.2410], device='cuda:3'), covar=tensor([0.0572, 0.0289, 0.0837, 0.0277, 0.0413, 0.0244, 0.0198, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0337, 0.0380, 0.0371, 0.0337, 0.0247, 0.0318, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 08:52:34,618 INFO [train2.py:809] (3/4) Epoch 27, batch 500, loss[ctc_loss=0.06352, att_loss=0.2399, loss=0.2046, over 17031.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007307, over 51.00 utterances.], tot_loss[ctc_loss=0.06592, att_loss=0.2313, loss=0.1983, over 3016566.13 frames. utt_duration=1295 frames, utt_pad_proportion=0.04077, over 9331.72 utterances.], batch size: 51, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:53:02,123 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104096.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:53:23,995 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104110.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:53:41,007 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104121.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:53:53,122 INFO [train2.py:809] (3/4) Epoch 27, batch 550, loss[ctc_loss=0.0632, att_loss=0.2369, loss=0.2022, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006835, over 46.00 utterances.], tot_loss[ctc_loss=0.06543, att_loss=0.2312, loss=0.198, over 3076182.86 frames. utt_duration=1295 frames, utt_pad_proportion=0.04048, over 9514.97 utterances.], batch size: 46, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:54:09,476 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0970, 4.3968, 4.7481, 4.5397, 3.1242, 4.3577, 3.1605, 1.9198], device='cuda:3'), covar=tensor([0.0470, 0.0364, 0.0526, 0.0351, 0.1308, 0.0266, 0.1170, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0184, 0.0263, 0.0177, 0.0221, 0.0163, 0.0232, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 08:54:29,374 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.746e+02 2.054e+02 2.540e+02 6.808e+02, threshold=4.109e+02, percent-clipped=2.0 2023-03-09 08:54:58,969 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5527, 3.2229, 3.8132, 2.9866, 3.5323, 4.7308, 4.4995, 3.1440], device='cuda:3'), covar=tensor([0.0387, 0.1584, 0.1200, 0.1424, 0.1158, 0.0681, 0.0634, 0.1372], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0250, 0.0289, 0.0222, 0.0271, 0.0380, 0.0274, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:55:07,902 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5718, 5.8974, 5.3507, 5.5926, 5.5285, 4.9602, 5.2071, 5.0389], device='cuda:3'), covar=tensor([0.1348, 0.0895, 0.1152, 0.0844, 0.1060, 0.1697, 0.2587, 0.2399], device='cuda:3'), in_proj_covar=tensor([0.0554, 0.0633, 0.0488, 0.0478, 0.0449, 0.0483, 0.0640, 0.0547], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:55:10,802 INFO [train2.py:809] (3/4) Epoch 27, batch 600, loss[ctc_loss=0.07677, att_loss=0.2516, loss=0.2166, over 17037.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.009768, over 53.00 utterances.], tot_loss[ctc_loss=0.06547, att_loss=0.2313, loss=0.1982, over 3119939.90 frames. utt_duration=1290 frames, utt_pad_proportion=0.04327, over 9682.35 utterances.], batch size: 53, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:56:29,436 INFO [train2.py:809] (3/4) Epoch 27, batch 650, loss[ctc_loss=0.05726, att_loss=0.2124, loss=0.1814, over 15341.00 frames. utt_duration=1755 frames, utt_pad_proportion=0.0124, over 35.00 utterances.], tot_loss[ctc_loss=0.06539, att_loss=0.2311, loss=0.1979, over 3154003.67 frames. utt_duration=1316 frames, utt_pad_proportion=0.03786, over 9600.56 utterances.], batch size: 35, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:56:34,218 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104232.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 08:57:05,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.363e+02 1.860e+02 2.240e+02 2.877e+02 8.849e+02, threshold=4.479e+02, percent-clipped=3.0 2023-03-09 08:57:29,951 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5643, 4.9623, 4.8196, 4.9322, 5.0245, 4.6324, 3.4238, 4.9455], device='cuda:3'), covar=tensor([0.0138, 0.0116, 0.0132, 0.0090, 0.0094, 0.0123, 0.0767, 0.0197], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0094, 0.0118, 0.0074, 0.0079, 0.0090, 0.0108, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 08:57:47,593 INFO [train2.py:809] (3/4) Epoch 27, batch 700, loss[ctc_loss=0.06478, att_loss=0.2327, loss=0.1991, over 16533.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006126, over 45.00 utterances.], tot_loss[ctc_loss=0.06548, att_loss=0.2317, loss=0.1984, over 3185647.22 frames. utt_duration=1273 frames, utt_pad_proportion=0.04699, over 10020.67 utterances.], batch size: 45, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:58:36,056 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8288, 4.8602, 4.5926, 2.6848, 4.6586, 4.5005, 4.0856, 2.6615], device='cuda:3'), covar=tensor([0.0138, 0.0118, 0.0283, 0.1098, 0.0107, 0.0233, 0.0342, 0.1353], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0105, 0.0109, 0.0112, 0.0089, 0.0117, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 08:59:06,611 INFO [train2.py:809] (3/4) Epoch 27, batch 750, loss[ctc_loss=0.05629, att_loss=0.2037, loss=0.1742, over 15876.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.01005, over 39.00 utterances.], tot_loss[ctc_loss=0.06564, att_loss=0.2315, loss=0.1983, over 3197534.26 frames. utt_duration=1262 frames, utt_pad_proportion=0.05082, over 10145.56 utterances.], batch size: 39, lr: 4.00e-03, grad_scale: 8.0 2023-03-09 08:59:34,751 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104346.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:59:40,825 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0284, 5.3143, 4.9133, 5.3958, 4.7545, 5.0123, 5.4374, 5.2306], device='cuda:3'), covar=tensor([0.0583, 0.0289, 0.0780, 0.0309, 0.0428, 0.0256, 0.0209, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0338, 0.0381, 0.0373, 0.0339, 0.0247, 0.0318, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 08:59:43,522 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 1.900e+02 2.289e+02 2.673e+02 5.162e+02, threshold=4.578e+02, percent-clipped=1.0 2023-03-09 08:59:45,857 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 08:59:52,103 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0326, 5.1709, 4.7217, 2.6922, 4.9004, 4.8654, 4.2723, 2.3430], device='cuda:3'), covar=tensor([0.0214, 0.0122, 0.0410, 0.1414, 0.0138, 0.0216, 0.0458, 0.2310], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0105, 0.0109, 0.0112, 0.0089, 0.0117, 0.0100, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:00:26,056 INFO [train2.py:809] (3/4) Epoch 27, batch 800, loss[ctc_loss=0.06328, att_loss=0.2398, loss=0.2045, over 17149.00 frames. utt_duration=694.4 frames, utt_pad_proportion=0.1276, over 99.00 utterances.], tot_loss[ctc_loss=0.06578, att_loss=0.2317, loss=0.1985, over 3210948.42 frames. utt_duration=1238 frames, utt_pad_proportion=0.05725, over 10385.35 utterances.], batch size: 99, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:00:27,113 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6946, 2.5043, 2.4190, 2.4534, 2.7669, 2.8768, 2.3430, 3.0000], device='cuda:3'), covar=tensor([0.1542, 0.2303, 0.1975, 0.1844, 0.1873, 0.1183, 0.2479, 0.1159], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0143, 0.0137, 0.0133, 0.0149, 0.0127, 0.0151, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 09:00:54,157 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104396.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:01:08,592 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104405.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:01:33,301 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104421.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:01:46,498 INFO [train2.py:809] (3/4) Epoch 27, batch 850, loss[ctc_loss=0.06538, att_loss=0.2168, loss=0.1865, over 16289.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006792, over 43.00 utterances.], tot_loss[ctc_loss=0.06619, att_loss=0.2326, loss=0.1993, over 3221835.27 frames. utt_duration=1200 frames, utt_pad_proportion=0.06725, over 10748.54 utterances.], batch size: 43, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:01:56,677 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9958, 5.2966, 4.8823, 5.3443, 4.7709, 4.9871, 5.4269, 5.2192], device='cuda:3'), covar=tensor([0.0683, 0.0291, 0.0825, 0.0339, 0.0413, 0.0299, 0.0213, 0.0189], device='cuda:3'), in_proj_covar=tensor([0.0401, 0.0336, 0.0380, 0.0373, 0.0337, 0.0246, 0.0318, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 09:02:06,070 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9577, 5.1992, 5.4798, 5.2755, 5.4619, 5.9208, 5.2757, 6.0060], device='cuda:3'), covar=tensor([0.0726, 0.0782, 0.0856, 0.1467, 0.1880, 0.0927, 0.0704, 0.0703], device='cuda:3'), in_proj_covar=tensor([0.0907, 0.0526, 0.0638, 0.0687, 0.0903, 0.0664, 0.0510, 0.0643], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:02:10,582 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104444.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:02:23,113 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 1.830e+02 2.139e+02 2.642e+02 5.994e+02, threshold=4.278e+02, percent-clipped=1.0 2023-03-09 09:02:49,625 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104469.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:03:05,897 INFO [train2.py:809] (3/4) Epoch 27, batch 900, loss[ctc_loss=0.06783, att_loss=0.2363, loss=0.2026, over 16895.00 frames. utt_duration=1381 frames, utt_pad_proportion=0.006827, over 49.00 utterances.], tot_loss[ctc_loss=0.06683, att_loss=0.2331, loss=0.1999, over 3234759.00 frames. utt_duration=1188 frames, utt_pad_proportion=0.06968, over 10907.41 utterances.], batch size: 49, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:04:24,928 INFO [train2.py:809] (3/4) Epoch 27, batch 950, loss[ctc_loss=0.0554, att_loss=0.2176, loss=0.1852, over 15382.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.009459, over 35.00 utterances.], tot_loss[ctc_loss=0.06702, att_loss=0.2335, loss=0.2002, over 3244976.57 frames. utt_duration=1199 frames, utt_pad_proportion=0.06548, over 10842.30 utterances.], batch size: 35, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:04:30,378 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104532.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 09:05:01,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 1.806e+02 2.235e+02 2.788e+02 5.642e+02, threshold=4.470e+02, percent-clipped=2.0 2023-03-09 09:05:43,842 INFO [train2.py:809] (3/4) Epoch 27, batch 1000, loss[ctc_loss=0.06831, att_loss=0.221, loss=0.1905, over 14543.00 frames. utt_duration=1819 frames, utt_pad_proportion=0.03032, over 32.00 utterances.], tot_loss[ctc_loss=0.06704, att_loss=0.233, loss=0.1998, over 3246894.98 frames. utt_duration=1178 frames, utt_pad_proportion=0.07098, over 11037.01 utterances.], batch size: 32, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:05:45,532 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104580.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:07:03,585 INFO [train2.py:809] (3/4) Epoch 27, batch 1050, loss[ctc_loss=0.06118, att_loss=0.24, loss=0.2043, over 16783.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005504, over 48.00 utterances.], tot_loss[ctc_loss=0.06748, att_loss=0.2334, loss=0.2002, over 3250435.45 frames. utt_duration=1175 frames, utt_pad_proportion=0.07357, over 11081.99 utterances.], batch size: 48, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:07:30,291 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104646.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:07:40,081 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 1.873e+02 2.248e+02 2.692e+02 7.712e+02, threshold=4.496e+02, percent-clipped=6.0 2023-03-09 09:07:41,972 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104653.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:08:22,949 INFO [train2.py:809] (3/4) Epoch 27, batch 1100, loss[ctc_loss=0.07372, att_loss=0.2507, loss=0.2153, over 17292.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.009484, over 55.00 utterances.], tot_loss[ctc_loss=0.06683, att_loss=0.2329, loss=0.1997, over 3250386.01 frames. utt_duration=1196 frames, utt_pad_proportion=0.0695, over 10888.20 utterances.], batch size: 55, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:08:38,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-09 09:08:46,595 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104694.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:08:58,148 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104701.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:09:04,453 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104705.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:09:24,708 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9618, 5.1858, 5.1444, 5.1001, 5.2133, 5.1835, 4.8477, 4.6464], device='cuda:3'), covar=tensor([0.0963, 0.0559, 0.0304, 0.0546, 0.0323, 0.0348, 0.0442, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0536, 0.0381, 0.0371, 0.0380, 0.0442, 0.0448, 0.0379, 0.0412], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:09:42,150 INFO [train2.py:809] (3/4) Epoch 27, batch 1150, loss[ctc_loss=0.06183, att_loss=0.2333, loss=0.199, over 16753.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006545, over 48.00 utterances.], tot_loss[ctc_loss=0.06669, att_loss=0.2326, loss=0.1994, over 3252780.34 frames. utt_duration=1186 frames, utt_pad_proportion=0.07154, over 10981.71 utterances.], batch size: 48, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:09:42,608 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6640, 4.7034, 4.8182, 4.7724, 5.3219, 4.5508, 4.7057, 2.8713], device='cuda:3'), covar=tensor([0.0262, 0.0346, 0.0311, 0.0302, 0.0567, 0.0277, 0.0325, 0.1496], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0219, 0.0214, 0.0232, 0.0385, 0.0189, 0.0203, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:10:19,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.763e+02 2.147e+02 2.589e+02 3.796e+02, threshold=4.293e+02, percent-clipped=0.0 2023-03-09 09:10:20,730 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104753.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:11:01,671 INFO [train2.py:809] (3/4) Epoch 27, batch 1200, loss[ctc_loss=0.06535, att_loss=0.2484, loss=0.2118, over 17341.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03639, over 63.00 utterances.], tot_loss[ctc_loss=0.06631, att_loss=0.2328, loss=0.1995, over 3267292.13 frames. utt_duration=1208 frames, utt_pad_proportion=0.06278, over 10834.98 utterances.], batch size: 63, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:11:05,724 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4959, 4.6882, 4.3667, 4.7089, 4.2299, 4.3055, 4.7738, 4.5980], device='cuda:3'), covar=tensor([0.0535, 0.0344, 0.0708, 0.0444, 0.0448, 0.0472, 0.0290, 0.0231], device='cuda:3'), in_proj_covar=tensor([0.0405, 0.0342, 0.0383, 0.0379, 0.0342, 0.0249, 0.0322, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 09:11:22,686 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5055, 2.3401, 2.2598, 2.3107, 2.5030, 2.6076, 2.1277, 2.4860], device='cuda:3'), covar=tensor([0.1534, 0.2143, 0.1941, 0.1415, 0.1848, 0.1116, 0.1652, 0.1301], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0145, 0.0139, 0.0134, 0.0152, 0.0129, 0.0153, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 09:11:24,071 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0949, 5.3260, 5.6146, 5.4168, 5.6357, 6.0551, 5.3306, 6.1579], device='cuda:3'), covar=tensor([0.0718, 0.0760, 0.0807, 0.1413, 0.1652, 0.0856, 0.0678, 0.0644], device='cuda:3'), in_proj_covar=tensor([0.0911, 0.0528, 0.0640, 0.0689, 0.0904, 0.0664, 0.0511, 0.0642], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:11:56,814 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-09 09:12:21,906 INFO [train2.py:809] (3/4) Epoch 27, batch 1250, loss[ctc_loss=0.07526, att_loss=0.2482, loss=0.2136, over 17034.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.006958, over 51.00 utterances.], tot_loss[ctc_loss=0.06689, att_loss=0.2324, loss=0.1993, over 3255106.33 frames. utt_duration=1192 frames, utt_pad_proportion=0.06833, over 10935.75 utterances.], batch size: 51, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:12:58,971 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.878e+02 2.209e+02 2.737e+02 5.339e+02, threshold=4.418e+02, percent-clipped=1.0 2023-03-09 09:13:42,101 INFO [train2.py:809] (3/4) Epoch 27, batch 1300, loss[ctc_loss=0.08573, att_loss=0.2481, loss=0.2156, over 17303.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.0228, over 59.00 utterances.], tot_loss[ctc_loss=0.06606, att_loss=0.2316, loss=0.1985, over 3253317.59 frames. utt_duration=1216 frames, utt_pad_proportion=0.06358, over 10713.37 utterances.], batch size: 59, lr: 3.99e-03, grad_scale: 8.0 2023-03-09 09:15:01,667 INFO [train2.py:809] (3/4) Epoch 27, batch 1350, loss[ctc_loss=0.07379, att_loss=0.2273, loss=0.1966, over 16282.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006553, over 43.00 utterances.], tot_loss[ctc_loss=0.067, att_loss=0.2318, loss=0.1988, over 3247819.39 frames. utt_duration=1196 frames, utt_pad_proportion=0.07227, over 10873.01 utterances.], batch size: 43, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:15:37,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.826e+02 2.285e+02 3.088e+02 1.440e+03, threshold=4.571e+02, percent-clipped=4.0 2023-03-09 09:16:20,669 INFO [train2.py:809] (3/4) Epoch 27, batch 1400, loss[ctc_loss=0.08331, att_loss=0.2554, loss=0.221, over 17353.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02174, over 59.00 utterances.], tot_loss[ctc_loss=0.06691, att_loss=0.2324, loss=0.1993, over 3251432.18 frames. utt_duration=1193 frames, utt_pad_proportion=0.07212, over 10917.58 utterances.], batch size: 59, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:16:23,181 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8750, 3.3471, 3.8852, 3.4492, 3.7882, 4.8438, 4.6850, 3.6659], device='cuda:3'), covar=tensor([0.0306, 0.1606, 0.1185, 0.1196, 0.1082, 0.0938, 0.0546, 0.1020], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0249, 0.0288, 0.0222, 0.0272, 0.0381, 0.0273, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:16:28,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 09:17:40,904 INFO [train2.py:809] (3/4) Epoch 27, batch 1450, loss[ctc_loss=0.07856, att_loss=0.2503, loss=0.216, over 17128.00 frames. utt_duration=1225 frames, utt_pad_proportion=0.0145, over 56.00 utterances.], tot_loss[ctc_loss=0.06641, att_loss=0.232, loss=0.1989, over 3252821.94 frames. utt_duration=1202 frames, utt_pad_proportion=0.06933, over 10839.08 utterances.], batch size: 56, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:17:55,088 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3393, 4.4086, 4.5612, 4.5192, 5.0758, 4.2981, 4.3869, 2.6469], device='cuda:3'), covar=tensor([0.0324, 0.0423, 0.0378, 0.0349, 0.0848, 0.0343, 0.0412, 0.1739], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0221, 0.0216, 0.0234, 0.0388, 0.0191, 0.0205, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:18:16,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 1.845e+02 2.127e+02 2.675e+02 5.625e+02, threshold=4.253e+02, percent-clipped=2.0 2023-03-09 09:18:59,154 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7952, 6.0636, 5.5490, 5.8103, 5.7580, 5.2198, 5.5848, 5.2657], device='cuda:3'), covar=tensor([0.1438, 0.0977, 0.0935, 0.0815, 0.0962, 0.1638, 0.2290, 0.2244], device='cuda:3'), in_proj_covar=tensor([0.0552, 0.0630, 0.0481, 0.0474, 0.0446, 0.0478, 0.0635, 0.0544], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:19:00,492 INFO [train2.py:809] (3/4) Epoch 27, batch 1500, loss[ctc_loss=0.06977, att_loss=0.241, loss=0.2067, over 16964.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007728, over 50.00 utterances.], tot_loss[ctc_loss=0.06646, att_loss=0.2326, loss=0.1993, over 3254840.26 frames. utt_duration=1196 frames, utt_pad_proportion=0.06877, over 10897.66 utterances.], batch size: 50, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:20:18,686 INFO [train2.py:809] (3/4) Epoch 27, batch 1550, loss[ctc_loss=0.0647, att_loss=0.2419, loss=0.2065, over 17333.00 frames. utt_duration=879 frames, utt_pad_proportion=0.07769, over 79.00 utterances.], tot_loss[ctc_loss=0.06698, att_loss=0.2325, loss=0.1994, over 3257409.61 frames. utt_duration=1186 frames, utt_pad_proportion=0.07122, over 11004.08 utterances.], batch size: 79, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:20:55,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.873e+02 2.333e+02 2.814e+02 5.399e+02, threshold=4.665e+02, percent-clipped=5.0 2023-03-09 09:21:38,746 INFO [train2.py:809] (3/4) Epoch 27, batch 1600, loss[ctc_loss=0.09033, att_loss=0.2461, loss=0.2149, over 16948.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.008488, over 50.00 utterances.], tot_loss[ctc_loss=0.06676, att_loss=0.2326, loss=0.1994, over 3260474.85 frames. utt_duration=1194 frames, utt_pad_proportion=0.06849, over 10934.92 utterances.], batch size: 50, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:22:44,829 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-09 09:22:47,352 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7720, 2.6034, 2.3120, 2.7964, 3.0503, 2.9465, 2.5349, 3.3152], device='cuda:3'), covar=tensor([0.1067, 0.1864, 0.1573, 0.1112, 0.1285, 0.0960, 0.1992, 0.0788], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0143, 0.0138, 0.0134, 0.0152, 0.0128, 0.0152, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 09:22:58,649 INFO [train2.py:809] (3/4) Epoch 27, batch 1650, loss[ctc_loss=0.05702, att_loss=0.2348, loss=0.1993, over 16892.00 frames. utt_duration=677.2 frames, utt_pad_proportion=0.1384, over 100.00 utterances.], tot_loss[ctc_loss=0.06731, att_loss=0.2328, loss=0.1997, over 3266224.25 frames. utt_duration=1191 frames, utt_pad_proportion=0.06665, over 10980.35 utterances.], batch size: 100, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:23:34,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.028e+02 2.397e+02 2.825e+02 1.110e+03, threshold=4.794e+02, percent-clipped=4.0 2023-03-09 09:24:17,723 INFO [train2.py:809] (3/4) Epoch 27, batch 1700, loss[ctc_loss=0.07093, att_loss=0.2471, loss=0.2119, over 17291.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01252, over 55.00 utterances.], tot_loss[ctc_loss=0.06681, att_loss=0.2319, loss=0.1989, over 3264968.68 frames. utt_duration=1240 frames, utt_pad_proportion=0.05654, over 10546.59 utterances.], batch size: 55, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:24:57,598 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0059, 3.7171, 3.7479, 3.2834, 3.7540, 3.8269, 3.8135, 2.8546], device='cuda:3'), covar=tensor([0.0979, 0.1190, 0.1410, 0.2794, 0.0876, 0.1832, 0.0697, 0.2893], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0204, 0.0219, 0.0269, 0.0181, 0.0281, 0.0201, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:25:35,433 INFO [train2.py:809] (3/4) Epoch 27, batch 1750, loss[ctc_loss=0.05397, att_loss=0.2333, loss=0.1974, over 17022.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.007861, over 51.00 utterances.], tot_loss[ctc_loss=0.06666, att_loss=0.2318, loss=0.1988, over 3270672.58 frames. utt_duration=1245 frames, utt_pad_proportion=0.05369, over 10524.48 utterances.], batch size: 51, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:26:11,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.890e+02 2.121e+02 2.571e+02 3.982e+02, threshold=4.242e+02, percent-clipped=0.0 2023-03-09 09:26:54,623 INFO [train2.py:809] (3/4) Epoch 27, batch 1800, loss[ctc_loss=0.06362, att_loss=0.2385, loss=0.2036, over 16523.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006992, over 45.00 utterances.], tot_loss[ctc_loss=0.06619, att_loss=0.2318, loss=0.1987, over 3270432.07 frames. utt_duration=1242 frames, utt_pad_proportion=0.0552, over 10543.89 utterances.], batch size: 45, lr: 3.98e-03, grad_scale: 8.0 2023-03-09 09:27:08,768 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105388.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:28:15,234 INFO [train2.py:809] (3/4) Epoch 27, batch 1850, loss[ctc_loss=0.07396, att_loss=0.2463, loss=0.2119, over 17303.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02339, over 59.00 utterances.], tot_loss[ctc_loss=0.06601, att_loss=0.2317, loss=0.1985, over 3266561.92 frames. utt_duration=1247 frames, utt_pad_proportion=0.0556, over 10494.00 utterances.], batch size: 59, lr: 3.97e-03, grad_scale: 16.0 2023-03-09 09:28:27,876 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8355, 6.1361, 5.5680, 5.8287, 5.8157, 5.2243, 5.5724, 5.3001], device='cuda:3'), covar=tensor([0.1339, 0.0891, 0.0988, 0.0890, 0.0880, 0.1747, 0.2312, 0.2510], device='cuda:3'), in_proj_covar=tensor([0.0562, 0.0644, 0.0492, 0.0481, 0.0455, 0.0488, 0.0649, 0.0554], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:28:47,147 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105449.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 09:28:51,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 1.738e+02 2.150e+02 2.856e+02 4.488e+02, threshold=4.299e+02, percent-clipped=4.0 2023-03-09 09:29:34,708 INFO [train2.py:809] (3/4) Epoch 27, batch 1900, loss[ctc_loss=0.05355, att_loss=0.2428, loss=0.2049, over 16750.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.007324, over 48.00 utterances.], tot_loss[ctc_loss=0.0661, att_loss=0.2318, loss=0.1986, over 3264438.33 frames. utt_duration=1239 frames, utt_pad_proportion=0.05785, over 10550.52 utterances.], batch size: 48, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:29:45,590 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105486.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:29:49,178 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-09 09:30:11,367 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105502.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:30:54,636 INFO [train2.py:809] (3/4) Epoch 27, batch 1950, loss[ctc_loss=0.06445, att_loss=0.2434, loss=0.2076, over 17138.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01329, over 56.00 utterances.], tot_loss[ctc_loss=0.06604, att_loss=0.2315, loss=0.1984, over 3268728.43 frames. utt_duration=1239 frames, utt_pad_proportion=0.05612, over 10563.31 utterances.], batch size: 56, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:30:57,366 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 09:31:23,666 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105547.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:31:32,461 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 1.913e+02 2.076e+02 2.767e+02 9.695e+02, threshold=4.152e+02, percent-clipped=4.0 2023-03-09 09:31:49,363 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105563.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:31:55,496 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7150, 5.1228, 4.9747, 5.0973, 5.1463, 4.8613, 3.8955, 5.0506], device='cuda:3'), covar=tensor([0.0131, 0.0114, 0.0128, 0.0088, 0.0107, 0.0105, 0.0587, 0.0238], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0091, 0.0116, 0.0073, 0.0079, 0.0089, 0.0106, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:32:14,217 INFO [train2.py:809] (3/4) Epoch 27, batch 2000, loss[ctc_loss=0.05073, att_loss=0.2171, loss=0.1838, over 15780.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007992, over 38.00 utterances.], tot_loss[ctc_loss=0.06473, att_loss=0.2307, loss=0.1975, over 3268111.04 frames. utt_duration=1267 frames, utt_pad_proportion=0.0492, over 10328.96 utterances.], batch size: 38, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:32:23,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2474, 5.1385, 4.9533, 2.9612, 4.9498, 4.8626, 4.4827, 3.0266], device='cuda:3'), covar=tensor([0.0109, 0.0118, 0.0296, 0.1025, 0.0112, 0.0182, 0.0274, 0.1212], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0106, 0.0111, 0.0113, 0.0090, 0.0118, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:32:47,410 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6376, 5.0344, 4.7829, 5.0297, 5.0711, 4.7020, 3.5934, 4.9939], device='cuda:3'), covar=tensor([0.0134, 0.0125, 0.0168, 0.0087, 0.0123, 0.0120, 0.0700, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0091, 0.0116, 0.0073, 0.0079, 0.0089, 0.0106, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:33:34,516 INFO [train2.py:809] (3/4) Epoch 27, batch 2050, loss[ctc_loss=0.06241, att_loss=0.2423, loss=0.2063, over 17429.00 frames. utt_duration=1012 frames, utt_pad_proportion=0.04621, over 69.00 utterances.], tot_loss[ctc_loss=0.06533, att_loss=0.2314, loss=0.1981, over 3265790.14 frames. utt_duration=1250 frames, utt_pad_proportion=0.05567, over 10464.10 utterances.], batch size: 69, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:34:11,424 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.673e+02 2.035e+02 2.438e+02 4.266e+02, threshold=4.071e+02, percent-clipped=1.0 2023-03-09 09:34:17,083 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4952, 3.1380, 3.7181, 3.1480, 3.6494, 4.5348, 4.3946, 3.5404], device='cuda:3'), covar=tensor([0.0400, 0.1680, 0.1203, 0.1312, 0.1067, 0.0995, 0.0640, 0.1037], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0251, 0.0293, 0.0222, 0.0273, 0.0384, 0.0276, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:34:53,529 INFO [train2.py:809] (3/4) Epoch 27, batch 2100, loss[ctc_loss=0.05919, att_loss=0.2036, loss=0.1747, over 15392.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.009802, over 35.00 utterances.], tot_loss[ctc_loss=0.06597, att_loss=0.2319, loss=0.1987, over 3269229.90 frames. utt_duration=1243 frames, utt_pad_proportion=0.05646, over 10531.58 utterances.], batch size: 35, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:36:13,363 INFO [train2.py:809] (3/4) Epoch 27, batch 2150, loss[ctc_loss=0.05272, att_loss=0.2219, loss=0.1881, over 16322.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006551, over 45.00 utterances.], tot_loss[ctc_loss=0.06548, att_loss=0.2317, loss=0.1984, over 3268513.44 frames. utt_duration=1237 frames, utt_pad_proportion=0.05823, over 10584.57 utterances.], batch size: 45, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:36:37,235 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105744.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 09:36:51,250 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.815e+02 2.188e+02 2.690e+02 5.528e+02, threshold=4.376e+02, percent-clipped=1.0 2023-03-09 09:37:32,706 INFO [train2.py:809] (3/4) Epoch 27, batch 2200, loss[ctc_loss=0.07688, att_loss=0.2522, loss=0.2171, over 17067.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008041, over 52.00 utterances.], tot_loss[ctc_loss=0.06603, att_loss=0.2326, loss=0.1993, over 3276904.49 frames. utt_duration=1233 frames, utt_pad_proportion=0.05733, over 10645.90 utterances.], batch size: 52, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:37:53,018 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105792.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:38:04,363 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105799.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:38:52,246 INFO [train2.py:809] (3/4) Epoch 27, batch 2250, loss[ctc_loss=0.06546, att_loss=0.2357, loss=0.2017, over 17501.00 frames. utt_duration=887.6 frames, utt_pad_proportion=0.07057, over 79.00 utterances.], tot_loss[ctc_loss=0.06619, att_loss=0.2322, loss=0.199, over 3269633.09 frames. utt_duration=1236 frames, utt_pad_proportion=0.05878, over 10591.27 utterances.], batch size: 79, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:39:13,146 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105842.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:39:30,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 1.815e+02 2.086e+02 2.504e+02 6.374e+02, threshold=4.173e+02, percent-clipped=3.0 2023-03-09 09:39:31,339 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105853.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:39:39,848 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105858.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:39:43,215 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105860.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:40:12,504 INFO [train2.py:809] (3/4) Epoch 27, batch 2300, loss[ctc_loss=0.05573, att_loss=0.2085, loss=0.1779, over 15655.00 frames. utt_duration=1694 frames, utt_pad_proportion=0.008038, over 37.00 utterances.], tot_loss[ctc_loss=0.06588, att_loss=0.2318, loss=0.1986, over 3268119.53 frames. utt_duration=1242 frames, utt_pad_proportion=0.058, over 10541.43 utterances.], batch size: 37, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:40:50,273 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-09 09:41:31,988 INFO [train2.py:809] (3/4) Epoch 27, batch 2350, loss[ctc_loss=0.08241, att_loss=0.2547, loss=0.2202, over 17039.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009695, over 52.00 utterances.], tot_loss[ctc_loss=0.06637, att_loss=0.2319, loss=0.1988, over 3272933.46 frames. utt_duration=1247 frames, utt_pad_proportion=0.05452, over 10510.67 utterances.], batch size: 52, lr: 3.97e-03, grad_scale: 8.0 2023-03-09 09:42:09,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 1.971e+02 2.209e+02 2.522e+02 5.529e+02, threshold=4.418e+02, percent-clipped=2.0 2023-03-09 09:42:10,097 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3214, 4.5633, 4.6622, 4.7513, 2.9671, 4.5919, 2.9526, 1.9859], device='cuda:3'), covar=tensor([0.0432, 0.0308, 0.0636, 0.0224, 0.1586, 0.0211, 0.1388, 0.1760], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0186, 0.0268, 0.0179, 0.0225, 0.0166, 0.0233, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:42:51,496 INFO [train2.py:809] (3/4) Epoch 27, batch 2400, loss[ctc_loss=0.08525, att_loss=0.2451, loss=0.2131, over 16398.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007828, over 44.00 utterances.], tot_loss[ctc_loss=0.06665, att_loss=0.2326, loss=0.1994, over 3274929.58 frames. utt_duration=1236 frames, utt_pad_proportion=0.05749, over 10612.07 utterances.], batch size: 44, lr: 3.96e-03, grad_scale: 8.0 2023-03-09 09:42:56,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.70 vs. limit=5.0 2023-03-09 09:44:16,151 INFO [train2.py:809] (3/4) Epoch 27, batch 2450, loss[ctc_loss=0.06874, att_loss=0.2439, loss=0.2089, over 17077.00 frames. utt_duration=691.5 frames, utt_pad_proportion=0.1324, over 99.00 utterances.], tot_loss[ctc_loss=0.06631, att_loss=0.2316, loss=0.1986, over 3272413.83 frames. utt_duration=1235 frames, utt_pad_proportion=0.05844, over 10614.94 utterances.], batch size: 99, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:44:40,208 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106044.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:44:55,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 1.876e+02 2.325e+02 2.817e+02 5.594e+02, threshold=4.649e+02, percent-clipped=2.0 2023-03-09 09:45:09,327 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0428, 3.7295, 3.2327, 3.3122, 4.0047, 3.6788, 2.7332, 4.1833], device='cuda:3'), covar=tensor([0.1002, 0.0536, 0.0981, 0.0719, 0.0673, 0.0652, 0.1023, 0.0475], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0225, 0.0229, 0.0205, 0.0287, 0.0247, 0.0203, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 09:45:20,630 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8350, 5.1287, 5.3335, 5.0974, 5.3399, 5.8059, 5.1068, 5.8799], device='cuda:3'), covar=tensor([0.0735, 0.0762, 0.0873, 0.1385, 0.1711, 0.0832, 0.0808, 0.0624], device='cuda:3'), in_proj_covar=tensor([0.0910, 0.0523, 0.0639, 0.0682, 0.0907, 0.0660, 0.0507, 0.0640], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:45:33,213 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106077.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:45:35,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-09 09:45:35,775 INFO [train2.py:809] (3/4) Epoch 27, batch 2500, loss[ctc_loss=0.08787, att_loss=0.2497, loss=0.2173, over 17359.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02033, over 59.00 utterances.], tot_loss[ctc_loss=0.0656, att_loss=0.2312, loss=0.1981, over 3270363.69 frames. utt_duration=1256 frames, utt_pad_proportion=0.05389, over 10430.16 utterances.], batch size: 59, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:45:45,657 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8775, 3.6003, 3.5829, 3.0645, 3.5133, 3.7072, 3.6271, 2.7427], device='cuda:3'), covar=tensor([0.1041, 0.1225, 0.1814, 0.3627, 0.1403, 0.1935, 0.0982, 0.3092], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0206, 0.0220, 0.0271, 0.0182, 0.0281, 0.0204, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:45:57,171 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106092.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:46:03,550 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1898, 5.4508, 4.8855, 5.2633, 5.1215, 4.5838, 4.9150, 4.7168], device='cuda:3'), covar=tensor([0.1387, 0.0996, 0.0974, 0.0911, 0.1071, 0.1724, 0.2272, 0.2280], device='cuda:3'), in_proj_covar=tensor([0.0558, 0.0633, 0.0483, 0.0476, 0.0451, 0.0479, 0.0643, 0.0543], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:46:56,808 INFO [train2.py:809] (3/4) Epoch 27, batch 2550, loss[ctc_loss=0.05577, att_loss=0.2242, loss=0.1905, over 16955.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007339, over 50.00 utterances.], tot_loss[ctc_loss=0.06538, att_loss=0.231, loss=0.1979, over 3264105.27 frames. utt_duration=1234 frames, utt_pad_proportion=0.0619, over 10596.67 utterances.], batch size: 50, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:47:12,045 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106138.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:47:16,544 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7144, 2.4381, 2.5320, 2.6575, 2.7840, 2.8991, 2.5618, 3.1574], device='cuda:3'), covar=tensor([0.1844, 0.2525, 0.1960, 0.1499, 0.1602, 0.1201, 0.1992, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0145, 0.0141, 0.0135, 0.0153, 0.0130, 0.0152, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 09:47:18,013 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106142.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:47:27,649 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106148.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:47:36,810 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.781e+02 2.163e+02 2.660e+02 6.949e+02, threshold=4.326e+02, percent-clipped=5.0 2023-03-09 09:47:39,260 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106155.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:47:39,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-09 09:47:44,057 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106158.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:48:16,576 INFO [train2.py:809] (3/4) Epoch 27, batch 2600, loss[ctc_loss=0.0673, att_loss=0.2366, loss=0.2027, over 16329.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006384, over 45.00 utterances.], tot_loss[ctc_loss=0.06531, att_loss=0.2317, loss=0.1984, over 3263088.49 frames. utt_duration=1211 frames, utt_pad_proportion=0.06666, over 10790.03 utterances.], batch size: 45, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:48:33,838 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106190.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:49:01,052 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106206.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:49:37,256 INFO [train2.py:809] (3/4) Epoch 27, batch 2650, loss[ctc_loss=0.0595, att_loss=0.2415, loss=0.2051, over 16972.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007144, over 50.00 utterances.], tot_loss[ctc_loss=0.06504, att_loss=0.2304, loss=0.1973, over 3255579.36 frames. utt_duration=1248 frames, utt_pad_proportion=0.06014, over 10447.57 utterances.], batch size: 50, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:49:48,580 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0146, 4.3805, 4.3607, 4.5273, 2.8060, 4.3068, 2.6294, 1.8311], device='cuda:3'), covar=tensor([0.0512, 0.0278, 0.0657, 0.0249, 0.1520, 0.0258, 0.1491, 0.1677], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0183, 0.0266, 0.0178, 0.0222, 0.0164, 0.0231, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:50:17,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.781e+02 2.178e+02 2.674e+02 4.972e+02, threshold=4.357e+02, percent-clipped=1.0 2023-03-09 09:50:34,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-09 09:50:39,617 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106268.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:50:55,952 INFO [train2.py:809] (3/4) Epoch 27, batch 2700, loss[ctc_loss=0.09116, att_loss=0.2552, loss=0.2224, over 17018.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008698, over 51.00 utterances.], tot_loss[ctc_loss=0.0657, att_loss=0.2305, loss=0.1975, over 3257695.27 frames. utt_duration=1259 frames, utt_pad_proportion=0.05632, over 10359.75 utterances.], batch size: 51, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:51:05,584 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5692, 3.0754, 3.6832, 2.9165, 3.5420, 4.6045, 4.4312, 3.2100], device='cuda:3'), covar=tensor([0.0320, 0.1510, 0.1176, 0.1400, 0.0987, 0.0853, 0.0595, 0.1237], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0249, 0.0290, 0.0221, 0.0270, 0.0382, 0.0275, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 09:51:19,452 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-03-09 09:52:15,261 INFO [train2.py:809] (3/4) Epoch 27, batch 2750, loss[ctc_loss=0.08851, att_loss=0.2455, loss=0.2141, over 16766.00 frames. utt_duration=1398 frames, utt_pad_proportion=0.006041, over 48.00 utterances.], tot_loss[ctc_loss=0.06614, att_loss=0.2316, loss=0.1985, over 3268516.02 frames. utt_duration=1247 frames, utt_pad_proportion=0.05634, over 10495.74 utterances.], batch size: 48, lr: 3.96e-03, grad_scale: 4.0 2023-03-09 09:52:15,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:52:27,144 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106336.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:52:32,242 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6742, 4.8823, 4.3769, 4.7176, 4.5281, 4.0765, 4.4199, 4.0843], device='cuda:3'), covar=tensor([0.1385, 0.1268, 0.1090, 0.1052, 0.1523, 0.1734, 0.2274, 0.2781], device='cuda:3'), in_proj_covar=tensor([0.0557, 0.0633, 0.0483, 0.0475, 0.0451, 0.0478, 0.0641, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 09:52:55,847 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.858e+02 2.130e+02 2.638e+02 1.636e+03, threshold=4.261e+02, percent-clipped=5.0 2023-03-09 09:53:27,398 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6674, 2.8433, 3.2692, 4.4814, 4.0597, 3.9369, 3.1919, 2.5606], device='cuda:3'), covar=tensor([0.0622, 0.2048, 0.1114, 0.0562, 0.0823, 0.0511, 0.1402, 0.1997], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0217, 0.0187, 0.0225, 0.0232, 0.0191, 0.0203, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:53:34,595 INFO [train2.py:809] (3/4) Epoch 27, batch 2800, loss[ctc_loss=0.06086, att_loss=0.2234, loss=0.1909, over 16394.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007964, over 44.00 utterances.], tot_loss[ctc_loss=0.06644, att_loss=0.2316, loss=0.1986, over 3268944.92 frames. utt_duration=1246 frames, utt_pad_proportion=0.05636, over 10506.10 utterances.], batch size: 44, lr: 3.96e-03, grad_scale: 8.0 2023-03-09 09:54:04,697 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:54:09,668 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4205, 2.5796, 4.9137, 3.9594, 3.1277, 4.2680, 4.8039, 4.6135], device='cuda:3'), covar=tensor([0.0324, 0.1563, 0.0268, 0.0897, 0.1656, 0.0263, 0.0197, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0249, 0.0223, 0.0327, 0.0273, 0.0240, 0.0214, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 09:54:55,106 INFO [train2.py:809] (3/4) Epoch 27, batch 2850, loss[ctc_loss=0.0694, att_loss=0.2357, loss=0.2024, over 16323.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006521, over 45.00 utterances.], tot_loss[ctc_loss=0.06608, att_loss=0.232, loss=0.1988, over 3274935.40 frames. utt_duration=1219 frames, utt_pad_proportion=0.061, over 10763.45 utterances.], batch size: 45, lr: 3.96e-03, grad_scale: 8.0 2023-03-09 09:55:01,772 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106433.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:55:05,631 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8848, 3.6166, 3.0068, 3.2427, 3.7487, 3.5331, 2.8056, 3.9099], device='cuda:3'), covar=tensor([0.1093, 0.0541, 0.1076, 0.0780, 0.0775, 0.0745, 0.0945, 0.0644], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0226, 0.0230, 0.0206, 0.0288, 0.0249, 0.0203, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 09:55:26,912 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106448.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:55:30,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-09 09:55:36,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.809e+02 2.032e+02 2.602e+02 5.247e+02, threshold=4.065e+02, percent-clipped=2.0 2023-03-09 09:55:36,454 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106454.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:55:38,013 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106455.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:56:15,402 INFO [train2.py:809] (3/4) Epoch 27, batch 2900, loss[ctc_loss=0.05925, att_loss=0.2378, loss=0.2021, over 16868.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.00742, over 49.00 utterances.], tot_loss[ctc_loss=0.06587, att_loss=0.2315, loss=0.1984, over 3260116.12 frames. utt_duration=1199 frames, utt_pad_proportion=0.07029, over 10889.54 utterances.], batch size: 49, lr: 3.96e-03, grad_scale: 8.0 2023-03-09 09:56:42,919 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106496.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:56:53,788 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106503.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:57:13,224 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106515.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 09:57:34,533 INFO [train2.py:809] (3/4) Epoch 27, batch 2950, loss[ctc_loss=0.06287, att_loss=0.243, loss=0.207, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008064, over 51.00 utterances.], tot_loss[ctc_loss=0.0658, att_loss=0.2315, loss=0.1984, over 3265689.20 frames. utt_duration=1194 frames, utt_pad_proportion=0.06936, over 10956.15 utterances.], batch size: 51, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 09:58:15,060 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.110e+02 2.427e+02 2.893e+02 4.636e+02, threshold=4.854e+02, percent-clipped=4.0 2023-03-09 09:58:53,989 INFO [train2.py:809] (3/4) Epoch 27, batch 3000, loss[ctc_loss=0.05728, att_loss=0.2293, loss=0.1949, over 16622.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005678, over 47.00 utterances.], tot_loss[ctc_loss=0.06594, att_loss=0.2316, loss=0.1984, over 3259196.25 frames. utt_duration=1187 frames, utt_pad_proportion=0.07352, over 10992.96 utterances.], batch size: 47, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 09:58:53,989 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 09:59:07,589 INFO [train2.py:843] (3/4) Epoch 27, validation: ctc_loss=0.04056, att_loss=0.2346, loss=0.1958, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 09:59:07,589 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 10:00:20,318 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106624.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:00:28,227 INFO [train2.py:809] (3/4) Epoch 27, batch 3050, loss[ctc_loss=0.06054, att_loss=0.2425, loss=0.2061, over 17097.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01556, over 56.00 utterances.], tot_loss[ctc_loss=0.06584, att_loss=0.2321, loss=0.1988, over 3267648.02 frames. utt_duration=1195 frames, utt_pad_proportion=0.06929, over 10950.08 utterances.], batch size: 56, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:01:08,157 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 1.917e+02 2.259e+02 2.641e+02 5.465e+02, threshold=4.518e+02, percent-clipped=3.0 2023-03-09 10:01:31,856 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7613, 5.1718, 4.9420, 5.0307, 5.2065, 4.8954, 3.8050, 5.1822], device='cuda:3'), covar=tensor([0.0117, 0.0100, 0.0137, 0.0109, 0.0088, 0.0107, 0.0594, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0092, 0.0117, 0.0073, 0.0079, 0.0090, 0.0107, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:01:46,963 INFO [train2.py:809] (3/4) Epoch 27, batch 3100, loss[ctc_loss=0.05315, att_loss=0.2029, loss=0.173, over 15366.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.009196, over 35.00 utterances.], tot_loss[ctc_loss=0.06579, att_loss=0.2317, loss=0.1985, over 3264734.98 frames. utt_duration=1199 frames, utt_pad_proportion=0.07029, over 10907.39 utterances.], batch size: 35, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:02:08,959 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106692.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:03:06,634 INFO [train2.py:809] (3/4) Epoch 27, batch 3150, loss[ctc_loss=0.09181, att_loss=0.2582, loss=0.225, over 17282.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.024, over 59.00 utterances.], tot_loss[ctc_loss=0.06547, att_loss=0.2315, loss=0.1983, over 3265689.83 frames. utt_duration=1192 frames, utt_pad_proportion=0.07103, over 10968.37 utterances.], batch size: 59, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:03:13,710 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106733.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:03:47,287 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 1.911e+02 2.328e+02 2.721e+02 4.436e+02, threshold=4.656e+02, percent-clipped=0.0 2023-03-09 10:04:26,574 INFO [train2.py:809] (3/4) Epoch 27, batch 3200, loss[ctc_loss=0.04451, att_loss=0.2273, loss=0.1907, over 16335.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005958, over 45.00 utterances.], tot_loss[ctc_loss=0.06506, att_loss=0.2321, loss=0.1987, over 3273845.35 frames. utt_duration=1198 frames, utt_pad_proportion=0.06685, over 10947.39 utterances.], batch size: 45, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:04:30,373 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106781.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:05:16,600 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-09 10:05:17,432 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:05:26,200 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-09 10:05:47,163 INFO [train2.py:809] (3/4) Epoch 27, batch 3250, loss[ctc_loss=0.07109, att_loss=0.2403, loss=0.2064, over 17338.00 frames. utt_duration=1262 frames, utt_pad_proportion=0.009954, over 55.00 utterances.], tot_loss[ctc_loss=0.06539, att_loss=0.2326, loss=0.1992, over 3285250.25 frames. utt_duration=1189 frames, utt_pad_proportion=0.06552, over 11062.48 utterances.], batch size: 55, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:06:13,948 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-09 10:06:21,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-09 10:06:27,323 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 1.765e+02 2.095e+02 2.388e+02 3.844e+02, threshold=4.190e+02, percent-clipped=0.0 2023-03-09 10:07:07,568 INFO [train2.py:809] (3/4) Epoch 27, batch 3300, loss[ctc_loss=0.06101, att_loss=0.232, loss=0.1978, over 16787.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005563, over 48.00 utterances.], tot_loss[ctc_loss=0.06568, att_loss=0.2325, loss=0.1991, over 3284330.70 frames. utt_duration=1198 frames, utt_pad_proportion=0.06385, over 10979.07 utterances.], batch size: 48, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:07:17,908 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3350, 2.6781, 3.2060, 4.2840, 3.7557, 3.7708, 2.7562, 2.0784], device='cuda:3'), covar=tensor([0.0717, 0.1891, 0.0859, 0.0484, 0.0827, 0.0489, 0.1491, 0.2229], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0215, 0.0187, 0.0223, 0.0231, 0.0190, 0.0202, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:08:19,443 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106924.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:08:27,389 INFO [train2.py:809] (3/4) Epoch 27, batch 3350, loss[ctc_loss=0.07014, att_loss=0.2452, loss=0.2102, over 17363.00 frames. utt_duration=1104 frames, utt_pad_proportion=0.03412, over 63.00 utterances.], tot_loss[ctc_loss=0.06627, att_loss=0.2329, loss=0.1996, over 3281314.22 frames. utt_duration=1182 frames, utt_pad_proportion=0.06908, over 11114.26 utterances.], batch size: 63, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:09:07,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.952e+02 2.382e+02 3.093e+02 6.808e+02, threshold=4.763e+02, percent-clipped=6.0 2023-03-09 10:09:25,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-03-09 10:09:35,729 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106972.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:09:47,293 INFO [train2.py:809] (3/4) Epoch 27, batch 3400, loss[ctc_loss=0.07313, att_loss=0.2437, loss=0.2096, over 17247.00 frames. utt_duration=874.7 frames, utt_pad_proportion=0.0841, over 79.00 utterances.], tot_loss[ctc_loss=0.06632, att_loss=0.2325, loss=0.1993, over 3269997.29 frames. utt_duration=1175 frames, utt_pad_proportion=0.07347, over 11144.24 utterances.], batch size: 79, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:10:09,133 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106992.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:11:07,807 INFO [train2.py:809] (3/4) Epoch 27, batch 3450, loss[ctc_loss=0.07984, att_loss=0.2304, loss=0.2003, over 16398.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007198, over 44.00 utterances.], tot_loss[ctc_loss=0.06578, att_loss=0.2317, loss=0.1985, over 3268831.98 frames. utt_duration=1206 frames, utt_pad_proportion=0.06546, over 10856.51 utterances.], batch size: 44, lr: 3.95e-03, grad_scale: 8.0 2023-03-09 10:11:25,846 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107040.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:11:48,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 1.835e+02 2.250e+02 2.844e+02 5.706e+02, threshold=4.501e+02, percent-clipped=2.0 2023-03-09 10:12:26,816 INFO [train2.py:809] (3/4) Epoch 27, batch 3500, loss[ctc_loss=0.05192, att_loss=0.2306, loss=0.1949, over 16874.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007884, over 49.00 utterances.], tot_loss[ctc_loss=0.06561, att_loss=0.2313, loss=0.1981, over 3268389.79 frames. utt_duration=1225 frames, utt_pad_proportion=0.06139, over 10684.69 utterances.], batch size: 49, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:12:37,312 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7928, 3.6217, 3.5575, 3.0871, 3.5782, 3.5592, 3.6559, 2.5889], device='cuda:3'), covar=tensor([0.1051, 0.0876, 0.1448, 0.3047, 0.0904, 0.1523, 0.0658, 0.3068], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0207, 0.0223, 0.0272, 0.0183, 0.0284, 0.0206, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:13:15,830 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107110.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:13:45,761 INFO [train2.py:809] (3/4) Epoch 27, batch 3550, loss[ctc_loss=0.07548, att_loss=0.2445, loss=0.2107, over 17287.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02534, over 59.00 utterances.], tot_loss[ctc_loss=0.06566, att_loss=0.2316, loss=0.1984, over 3272447.17 frames. utt_duration=1247 frames, utt_pad_proportion=0.05586, over 10511.49 utterances.], batch size: 59, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:13:51,081 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0782, 5.2874, 5.6059, 5.3587, 5.5591, 6.0158, 5.2795, 6.1052], device='cuda:3'), covar=tensor([0.0737, 0.0758, 0.0866, 0.1447, 0.1837, 0.0971, 0.0759, 0.0738], device='cuda:3'), in_proj_covar=tensor([0.0914, 0.0526, 0.0644, 0.0683, 0.0910, 0.0667, 0.0511, 0.0644], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:14:03,583 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107140.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 10:14:22,410 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107152.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:14:24,972 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.809e+01 1.862e+02 2.158e+02 2.587e+02 5.069e+02, threshold=4.316e+02, percent-clipped=3.0 2023-03-09 10:14:31,586 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107158.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:15:05,784 INFO [train2.py:809] (3/4) Epoch 27, batch 3600, loss[ctc_loss=0.07867, att_loss=0.2391, loss=0.207, over 16612.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006186, over 47.00 utterances.], tot_loss[ctc_loss=0.06623, att_loss=0.2327, loss=0.1994, over 3273952.16 frames. utt_duration=1215 frames, utt_pad_proportion=0.06332, over 10790.32 utterances.], batch size: 47, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:15:41,181 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107201.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 10:15:59,534 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107213.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 10:16:24,612 INFO [train2.py:809] (3/4) Epoch 27, batch 3650, loss[ctc_loss=0.04961, att_loss=0.2345, loss=0.1975, over 17513.00 frames. utt_duration=888.2 frames, utt_pad_proportion=0.07097, over 79.00 utterances.], tot_loss[ctc_loss=0.06592, att_loss=0.2327, loss=0.1993, over 3282277.34 frames. utt_duration=1229 frames, utt_pad_proportion=0.05667, over 10694.07 utterances.], batch size: 79, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:17:03,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.892e+02 2.154e+02 2.728e+02 3.735e+02, threshold=4.308e+02, percent-clipped=0.0 2023-03-09 10:17:32,448 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0666, 4.2974, 4.2425, 4.4506, 2.6938, 4.2854, 2.7330, 1.9391], device='cuda:3'), covar=tensor([0.0436, 0.0285, 0.0680, 0.0287, 0.1538, 0.0256, 0.1397, 0.1581], device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0186, 0.0268, 0.0181, 0.0224, 0.0167, 0.0233, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:17:43,439 INFO [train2.py:809] (3/4) Epoch 27, batch 3700, loss[ctc_loss=0.07668, att_loss=0.2372, loss=0.2051, over 17244.00 frames. utt_duration=874.7 frames, utt_pad_proportion=0.08026, over 79.00 utterances.], tot_loss[ctc_loss=0.0655, att_loss=0.2324, loss=0.199, over 3284379.89 frames. utt_duration=1242 frames, utt_pad_proportion=0.05165, over 10592.13 utterances.], batch size: 79, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:19:01,341 INFO [train2.py:809] (3/4) Epoch 27, batch 3750, loss[ctc_loss=0.06253, att_loss=0.2321, loss=0.1982, over 16324.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006521, over 45.00 utterances.], tot_loss[ctc_loss=0.06591, att_loss=0.2324, loss=0.1991, over 3283997.57 frames. utt_duration=1237 frames, utt_pad_proportion=0.05334, over 10632.69 utterances.], batch size: 45, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:19:40,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 1.917e+02 2.215e+02 2.702e+02 1.402e+03, threshold=4.430e+02, percent-clipped=2.0 2023-03-09 10:20:19,855 INFO [train2.py:809] (3/4) Epoch 27, batch 3800, loss[ctc_loss=0.07891, att_loss=0.2558, loss=0.2204, over 17341.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02207, over 59.00 utterances.], tot_loss[ctc_loss=0.06626, att_loss=0.232, loss=0.1989, over 3264083.61 frames. utt_duration=1233 frames, utt_pad_proportion=0.05913, over 10602.50 utterances.], batch size: 59, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:21:21,296 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8998, 3.6120, 3.0389, 3.2327, 3.8169, 3.5318, 2.8582, 3.9223], device='cuda:3'), covar=tensor([0.0998, 0.0498, 0.1059, 0.0750, 0.0698, 0.0717, 0.0845, 0.0466], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0225, 0.0229, 0.0206, 0.0288, 0.0246, 0.0202, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 10:21:25,856 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2572, 5.5531, 5.7770, 5.5501, 5.7918, 6.2115, 5.3590, 6.2564], device='cuda:3'), covar=tensor([0.0679, 0.0648, 0.0844, 0.1383, 0.1632, 0.0836, 0.0694, 0.0699], device='cuda:3'), in_proj_covar=tensor([0.0920, 0.0527, 0.0646, 0.0685, 0.0915, 0.0670, 0.0512, 0.0643], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:21:39,086 INFO [train2.py:809] (3/4) Epoch 27, batch 3850, loss[ctc_loss=0.04944, att_loss=0.2193, loss=0.1853, over 16477.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006177, over 46.00 utterances.], tot_loss[ctc_loss=0.06611, att_loss=0.2322, loss=0.199, over 3271555.88 frames. utt_duration=1244 frames, utt_pad_proportion=0.05319, over 10536.23 utterances.], batch size: 46, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:22:17,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 1.947e+02 2.251e+02 2.768e+02 8.015e+02, threshold=4.503e+02, percent-clipped=5.0 2023-03-09 10:22:32,662 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0069, 5.3703, 4.9587, 5.3842, 4.8218, 5.0074, 5.5043, 5.2659], device='cuda:3'), covar=tensor([0.0664, 0.0245, 0.0706, 0.0319, 0.0390, 0.0242, 0.0187, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0401, 0.0343, 0.0379, 0.0383, 0.0339, 0.0248, 0.0321, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 10:22:55,307 INFO [train2.py:809] (3/4) Epoch 27, batch 3900, loss[ctc_loss=0.0524, att_loss=0.2494, loss=0.21, over 16627.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005349, over 47.00 utterances.], tot_loss[ctc_loss=0.06496, att_loss=0.231, loss=0.1978, over 3269719.02 frames. utt_duration=1275 frames, utt_pad_proportion=0.04712, over 10272.08 utterances.], batch size: 47, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:22:58,890 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6729, 4.6545, 4.4527, 2.8359, 4.4848, 4.4197, 3.9405, 2.7334], device='cuda:3'), covar=tensor([0.0123, 0.0124, 0.0257, 0.1024, 0.0116, 0.0249, 0.0350, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0108, 0.0112, 0.0114, 0.0091, 0.0120, 0.0103, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 10:23:00,397 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5961, 4.9972, 4.8038, 4.8488, 5.1232, 4.7304, 3.5637, 4.9584], device='cuda:3'), covar=tensor([0.0143, 0.0114, 0.0155, 0.0101, 0.0083, 0.0118, 0.0685, 0.0178], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0092, 0.0116, 0.0073, 0.0078, 0.0089, 0.0105, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:23:22,178 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107496.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 10:23:29,996 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0139, 4.4179, 4.4351, 4.4913, 2.9437, 4.3326, 3.0094, 1.9309], device='cuda:3'), covar=tensor([0.0543, 0.0278, 0.0626, 0.0277, 0.1403, 0.0253, 0.1240, 0.1587], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0186, 0.0267, 0.0180, 0.0222, 0.0166, 0.0233, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:23:40,533 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107508.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 10:23:43,615 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107510.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:23:45,161 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8887, 2.5960, 2.6779, 2.7043, 2.9131, 2.9124, 2.4763, 3.1271], device='cuda:3'), covar=tensor([0.1534, 0.2139, 0.1996, 0.1362, 0.1573, 0.1062, 0.2251, 0.1125], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0142, 0.0139, 0.0133, 0.0150, 0.0129, 0.0149, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 10:24:12,009 INFO [train2.py:809] (3/4) Epoch 27, batch 3950, loss[ctc_loss=0.06821, att_loss=0.24, loss=0.2056, over 16481.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005649, over 46.00 utterances.], tot_loss[ctc_loss=0.06549, att_loss=0.2315, loss=0.1983, over 3275119.10 frames. utt_duration=1264 frames, utt_pad_proportion=0.04918, over 10378.29 utterances.], batch size: 46, lr: 3.94e-03, grad_scale: 8.0 2023-03-09 10:24:49,845 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 1.912e+02 2.157e+02 2.759e+02 5.743e+02, threshold=4.314e+02, percent-clipped=3.0 2023-03-09 10:25:20,652 INFO [train2.py:809] (3/4) Epoch 28, batch 0, loss[ctc_loss=0.07678, att_loss=0.252, loss=0.2169, over 17071.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008513, over 53.00 utterances.], tot_loss[ctc_loss=0.07678, att_loss=0.252, loss=0.2169, over 17071.00 frames. utt_duration=1290 frames, utt_pad_proportion=0.008513, over 53.00 utterances.], batch size: 53, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:25:20,653 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 10:25:32,827 INFO [train2.py:843] (3/4) Epoch 28, validation: ctc_loss=0.04041, att_loss=0.2344, loss=0.1956, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 10:25:32,828 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 10:25:47,327 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107571.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:26:28,579 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7242, 3.3729, 3.3711, 2.9456, 3.3006, 3.3046, 3.3942, 2.5239], device='cuda:3'), covar=tensor([0.1022, 0.1416, 0.1652, 0.2865, 0.2124, 0.1973, 0.0976, 0.2981], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0206, 0.0222, 0.0270, 0.0182, 0.0281, 0.0205, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:26:52,867 INFO [train2.py:809] (3/4) Epoch 28, batch 50, loss[ctc_loss=0.04217, att_loss=0.2048, loss=0.1723, over 15490.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.0096, over 36.00 utterances.], tot_loss[ctc_loss=0.06305, att_loss=0.2308, loss=0.1973, over 748000.87 frames. utt_duration=1261 frames, utt_pad_proportion=0.03425, over 2375.31 utterances.], batch size: 36, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:28:00,266 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.779e+02 2.036e+02 2.554e+02 4.414e+02, threshold=4.071e+02, percent-clipped=2.0 2023-03-09 10:28:08,372 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107659.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:28:12,551 INFO [train2.py:809] (3/4) Epoch 28, batch 100, loss[ctc_loss=0.08054, att_loss=0.2568, loss=0.2216, over 17052.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009485, over 53.00 utterances.], tot_loss[ctc_loss=0.0644, att_loss=0.2323, loss=0.1987, over 1313685.27 frames. utt_duration=1319 frames, utt_pad_proportion=0.02585, over 3989.46 utterances.], batch size: 53, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:28:29,388 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 2023-03-09 10:29:32,148 INFO [train2.py:809] (3/4) Epoch 28, batch 150, loss[ctc_loss=0.07501, att_loss=0.2463, loss=0.2121, over 17400.00 frames. utt_duration=1106 frames, utt_pad_proportion=0.03302, over 63.00 utterances.], tot_loss[ctc_loss=0.06563, att_loss=0.2325, loss=0.1991, over 1742294.97 frames. utt_duration=1280 frames, utt_pad_proportion=0.04343, over 5451.39 utterances.], batch size: 63, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:29:45,375 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107720.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:30:11,849 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107737.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:30:38,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 1.937e+02 2.331e+02 2.865e+02 6.453e+02, threshold=4.662e+02, percent-clipped=4.0 2023-03-09 10:30:51,739 INFO [train2.py:809] (3/4) Epoch 28, batch 200, loss[ctc_loss=0.06741, att_loss=0.2235, loss=0.1923, over 16184.00 frames. utt_duration=1580 frames, utt_pad_proportion=0.006699, over 41.00 utterances.], tot_loss[ctc_loss=0.06582, att_loss=0.2325, loss=0.1991, over 2082265.33 frames. utt_duration=1261 frames, utt_pad_proportion=0.04896, over 6612.79 utterances.], batch size: 41, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:31:45,227 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107796.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 10:31:48,981 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107798.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 10:32:05,093 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107808.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:32:10,871 INFO [train2.py:809] (3/4) Epoch 28, batch 250, loss[ctc_loss=0.04558, att_loss=0.2212, loss=0.1861, over 16124.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006449, over 42.00 utterances.], tot_loss[ctc_loss=0.06676, att_loss=0.2335, loss=0.2001, over 2350657.08 frames. utt_duration=1224 frames, utt_pad_proportion=0.05685, over 7688.82 utterances.], batch size: 42, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:33:01,125 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107844.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 10:33:16,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.778e+02 2.118e+02 2.564e+02 4.844e+02, threshold=4.236e+02, percent-clipped=1.0 2023-03-09 10:33:20,620 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107856.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:33:29,919 INFO [train2.py:809] (3/4) Epoch 28, batch 300, loss[ctc_loss=0.06209, att_loss=0.2395, loss=0.2041, over 17037.00 frames. utt_duration=1287 frames, utt_pad_proportion=0.009753, over 53.00 utterances.], tot_loss[ctc_loss=0.06588, att_loss=0.2333, loss=0.1998, over 2562757.19 frames. utt_duration=1236 frames, utt_pad_proportion=0.05185, over 8303.79 utterances.], batch size: 53, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:33:36,250 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107866.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:34:49,121 INFO [train2.py:809] (3/4) Epoch 28, batch 350, loss[ctc_loss=0.05795, att_loss=0.2228, loss=0.1898, over 16386.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007954, over 44.00 utterances.], tot_loss[ctc_loss=0.06574, att_loss=0.233, loss=0.1996, over 2723025.16 frames. utt_duration=1234 frames, utt_pad_proportion=0.05309, over 8840.42 utterances.], batch size: 44, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:34:55,383 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9739, 6.2101, 5.6562, 5.9296, 5.9090, 5.3807, 5.7310, 5.4361], device='cuda:3'), covar=tensor([0.1033, 0.0767, 0.0982, 0.0686, 0.0875, 0.1482, 0.1827, 0.2046], device='cuda:3'), in_proj_covar=tensor([0.0558, 0.0638, 0.0487, 0.0475, 0.0452, 0.0481, 0.0642, 0.0542], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 10:34:59,490 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4066, 2.2160, 4.8529, 3.8175, 2.9039, 4.0750, 4.6323, 4.5328], device='cuda:3'), covar=tensor([0.0303, 0.1825, 0.0217, 0.0884, 0.1748, 0.0281, 0.0194, 0.0293], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0248, 0.0223, 0.0323, 0.0269, 0.0239, 0.0213, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:35:55,813 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.778e+02 2.080e+02 2.502e+02 5.851e+02, threshold=4.160e+02, percent-clipped=2.0 2023-03-09 10:36:02,649 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1241, 5.3484, 5.6301, 5.4345, 5.6176, 6.0526, 5.3123, 6.0916], device='cuda:3'), covar=tensor([0.0723, 0.0813, 0.0945, 0.1381, 0.1812, 0.0974, 0.0709, 0.0811], device='cuda:3'), in_proj_covar=tensor([0.0912, 0.0524, 0.0645, 0.0684, 0.0913, 0.0665, 0.0510, 0.0644], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:36:08,361 INFO [train2.py:809] (3/4) Epoch 28, batch 400, loss[ctc_loss=0.06405, att_loss=0.2293, loss=0.1963, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005723, over 47.00 utterances.], tot_loss[ctc_loss=0.06581, att_loss=0.2329, loss=0.1995, over 2851929.11 frames. utt_duration=1246 frames, utt_pad_proportion=0.04911, over 9165.04 utterances.], batch size: 47, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:36:30,800 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2050, 5.1378, 4.9570, 2.9659, 4.9317, 4.8524, 4.5167, 2.9068], device='cuda:3'), covar=tensor([0.0115, 0.0114, 0.0251, 0.0993, 0.0105, 0.0180, 0.0263, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0107, 0.0112, 0.0114, 0.0091, 0.0120, 0.0102, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 10:37:33,050 INFO [train2.py:809] (3/4) Epoch 28, batch 450, loss[ctc_loss=0.05891, att_loss=0.2434, loss=0.2065, over 16891.00 frames. utt_duration=1381 frames, utt_pad_proportion=0.006053, over 49.00 utterances.], tot_loss[ctc_loss=0.06537, att_loss=0.2322, loss=0.1988, over 2941828.99 frames. utt_duration=1247 frames, utt_pad_proportion=0.05152, over 9444.37 utterances.], batch size: 49, lr: 3.86e-03, grad_scale: 8.0 2023-03-09 10:37:37,597 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108015.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:38:20,539 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1699, 4.4157, 4.4221, 4.4619, 4.4647, 4.3164, 3.2544, 4.3879], device='cuda:3'), covar=tensor([0.0146, 0.0142, 0.0157, 0.0095, 0.0121, 0.0119, 0.0687, 0.0227], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0091, 0.0116, 0.0072, 0.0078, 0.0089, 0.0104, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:38:39,469 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.861e+02 2.240e+02 2.601e+02 4.871e+02, threshold=4.480e+02, percent-clipped=1.0 2023-03-09 10:38:51,890 INFO [train2.py:809] (3/4) Epoch 28, batch 500, loss[ctc_loss=0.05079, att_loss=0.2088, loss=0.1772, over 15949.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006626, over 41.00 utterances.], tot_loss[ctc_loss=0.06569, att_loss=0.2317, loss=0.1985, over 3000711.54 frames. utt_duration=1212 frames, utt_pad_proportion=0.06539, over 9916.03 utterances.], batch size: 41, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:38:55,277 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1069, 2.7135, 3.4771, 2.8925, 3.3632, 4.2361, 4.0920, 3.0095], device='cuda:3'), covar=tensor([0.0437, 0.1854, 0.1235, 0.1199, 0.1142, 0.0965, 0.0671, 0.1307], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0249, 0.0289, 0.0218, 0.0269, 0.0379, 0.0272, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:39:05,350 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8582, 3.6650, 3.6687, 3.1616, 3.6656, 3.7338, 3.6678, 2.7992], device='cuda:3'), covar=tensor([0.1089, 0.1161, 0.1925, 0.3184, 0.1578, 0.1736, 0.0922, 0.3198], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0206, 0.0222, 0.0270, 0.0182, 0.0280, 0.0206, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:39:41,180 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108093.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 10:39:58,071 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9865, 5.0292, 4.8825, 2.3726, 2.0779, 2.8654, 2.3513, 3.8537], device='cuda:3'), covar=tensor([0.0750, 0.0291, 0.0262, 0.4375, 0.5280, 0.2419, 0.3813, 0.1656], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0302, 0.0282, 0.0254, 0.0341, 0.0335, 0.0265, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 10:40:05,605 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6660, 3.4270, 3.4364, 2.9743, 3.3533, 3.3200, 3.3720, 2.5339], device='cuda:3'), covar=tensor([0.1063, 0.1095, 0.1461, 0.2894, 0.1177, 0.1521, 0.1153, 0.2887], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0206, 0.0222, 0.0270, 0.0182, 0.0280, 0.0206, 0.0227], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:40:11,190 INFO [train2.py:809] (3/4) Epoch 28, batch 550, loss[ctc_loss=0.05264, att_loss=0.2138, loss=0.1816, over 16012.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.007001, over 40.00 utterances.], tot_loss[ctc_loss=0.06533, att_loss=0.2315, loss=0.1983, over 3065299.11 frames. utt_duration=1221 frames, utt_pad_proportion=0.06188, over 10051.15 utterances.], batch size: 40, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:40:43,371 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7391, 6.0284, 5.4696, 5.7099, 5.7430, 5.1399, 5.4147, 5.2445], device='cuda:3'), covar=tensor([0.1290, 0.0904, 0.1073, 0.0938, 0.0940, 0.1622, 0.2243, 0.2561], device='cuda:3'), in_proj_covar=tensor([0.0563, 0.0643, 0.0490, 0.0478, 0.0457, 0.0485, 0.0645, 0.0550], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 10:41:18,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 1.757e+02 2.157e+02 2.611e+02 5.249e+02, threshold=4.315e+02, percent-clipped=2.0 2023-03-09 10:41:30,580 INFO [train2.py:809] (3/4) Epoch 28, batch 600, loss[ctc_loss=0.04759, att_loss=0.2059, loss=0.1743, over 15879.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009861, over 39.00 utterances.], tot_loss[ctc_loss=0.0649, att_loss=0.2304, loss=0.1973, over 3103114.82 frames. utt_duration=1232 frames, utt_pad_proportion=0.06174, over 10084.17 utterances.], batch size: 39, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:41:37,587 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108166.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:42:49,655 INFO [train2.py:809] (3/4) Epoch 28, batch 650, loss[ctc_loss=0.07106, att_loss=0.2321, loss=0.1999, over 16124.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006264, over 42.00 utterances.], tot_loss[ctc_loss=0.06496, att_loss=0.2305, loss=0.1974, over 3135384.46 frames. utt_duration=1244 frames, utt_pad_proportion=0.06014, over 10092.16 utterances.], batch size: 42, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:42:53,330 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:43:56,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.814e+02 2.237e+02 2.665e+02 5.467e+02, threshold=4.474e+02, percent-clipped=5.0 2023-03-09 10:44:08,567 INFO [train2.py:809] (3/4) Epoch 28, batch 700, loss[ctc_loss=0.0616, att_loss=0.2406, loss=0.2048, over 17500.00 frames. utt_duration=1016 frames, utt_pad_proportion=0.04263, over 69.00 utterances.], tot_loss[ctc_loss=0.06409, att_loss=0.2299, loss=0.1968, over 3168446.22 frames. utt_duration=1271 frames, utt_pad_proportion=0.05194, over 9981.47 utterances.], batch size: 69, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:44:08,949 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7624, 2.5157, 2.5298, 2.6634, 2.8635, 2.8345, 2.3860, 2.9930], device='cuda:3'), covar=tensor([0.1484, 0.2228, 0.1790, 0.1262, 0.1777, 0.1224, 0.2264, 0.1331], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0147, 0.0142, 0.0137, 0.0154, 0.0132, 0.0153, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 10:44:46,758 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2894, 2.9513, 3.5701, 2.9632, 3.4410, 4.4139, 4.2605, 3.2327], device='cuda:3'), covar=tensor([0.0467, 0.1755, 0.1282, 0.1381, 0.1164, 0.1180, 0.0696, 0.1239], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0251, 0.0293, 0.0221, 0.0273, 0.0383, 0.0275, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 10:44:50,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2023-03-09 10:45:27,466 INFO [train2.py:809] (3/4) Epoch 28, batch 750, loss[ctc_loss=0.07526, att_loss=0.2428, loss=0.2093, over 17368.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.04983, over 69.00 utterances.], tot_loss[ctc_loss=0.06459, att_loss=0.2298, loss=0.1968, over 3185402.12 frames. utt_duration=1265 frames, utt_pad_proportion=0.05364, over 10082.69 utterances.], batch size: 69, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:45:32,969 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108315.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:45:48,274 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108325.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:46:33,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.877e+02 2.184e+02 2.631e+02 4.478e+02, threshold=4.368e+02, percent-clipped=1.0 2023-03-09 10:46:46,000 INFO [train2.py:809] (3/4) Epoch 28, batch 800, loss[ctc_loss=0.06092, att_loss=0.22, loss=0.1882, over 16291.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.006488, over 43.00 utterances.], tot_loss[ctc_loss=0.0648, att_loss=0.2296, loss=0.1966, over 3205909.61 frames. utt_duration=1274 frames, utt_pad_proportion=0.05072, over 10078.48 utterances.], batch size: 43, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:46:47,658 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108363.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:47:10,578 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5088, 2.8632, 4.9393, 3.9336, 3.0194, 4.2412, 4.7692, 4.6653], device='cuda:3'), covar=tensor([0.0289, 0.1392, 0.0241, 0.0877, 0.1721, 0.0278, 0.0221, 0.0271], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0250, 0.0227, 0.0329, 0.0273, 0.0242, 0.0217, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:47:23,506 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108386.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:47:34,714 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108393.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:48:04,323 INFO [train2.py:809] (3/4) Epoch 28, batch 850, loss[ctc_loss=0.06042, att_loss=0.2349, loss=0.2, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005454, over 47.00 utterances.], tot_loss[ctc_loss=0.06518, att_loss=0.2306, loss=0.1975, over 3223149.49 frames. utt_duration=1242 frames, utt_pad_proportion=0.05773, over 10390.88 utterances.], batch size: 47, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:48:27,381 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1077, 6.3033, 5.8759, 5.9546, 5.9916, 5.4572, 5.9210, 5.5013], device='cuda:3'), covar=tensor([0.1244, 0.0867, 0.0997, 0.0881, 0.0929, 0.1597, 0.2042, 0.2589], device='cuda:3'), in_proj_covar=tensor([0.0564, 0.0646, 0.0492, 0.0481, 0.0459, 0.0487, 0.0648, 0.0552], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 10:48:35,121 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108432.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:48:49,724 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108441.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:49:03,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0 2023-03-09 10:49:10,180 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.210e+02 1.750e+02 2.196e+02 2.771e+02 5.305e+02, threshold=4.392e+02, percent-clipped=2.0 2023-03-09 10:49:23,198 INFO [train2.py:809] (3/4) Epoch 28, batch 900, loss[ctc_loss=0.05082, att_loss=0.2088, loss=0.1772, over 15854.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01115, over 39.00 utterances.], tot_loss[ctc_loss=0.06475, att_loss=0.2302, loss=0.1971, over 3237662.13 frames. utt_duration=1269 frames, utt_pad_proportion=0.04987, over 10214.41 utterances.], batch size: 39, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:50:12,175 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108493.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 10:50:31,040 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4602, 4.5107, 4.6514, 4.5907, 5.2027, 4.3931, 4.5185, 2.6477], device='cuda:3'), covar=tensor([0.0310, 0.0411, 0.0376, 0.0386, 0.0686, 0.0295, 0.0402, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0221, 0.0216, 0.0233, 0.0379, 0.0192, 0.0207, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:50:42,017 INFO [train2.py:809] (3/4) Epoch 28, batch 950, loss[ctc_loss=0.07593, att_loss=0.2406, loss=0.2077, over 17034.00 frames. utt_duration=1338 frames, utt_pad_proportion=0.007045, over 51.00 utterances.], tot_loss[ctc_loss=0.06554, att_loss=0.2309, loss=0.1978, over 3241256.69 frames. utt_duration=1248 frames, utt_pad_proportion=0.05749, over 10399.94 utterances.], batch size: 51, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:51:36,944 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4973, 2.9834, 3.5209, 4.4418, 3.9577, 3.9291, 2.9966, 2.5281], device='cuda:3'), covar=tensor([0.0703, 0.1780, 0.0799, 0.0577, 0.0894, 0.0530, 0.1425, 0.1903], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0221, 0.0188, 0.0228, 0.0236, 0.0194, 0.0206, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:51:47,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.763e+02 2.072e+02 2.486e+02 7.743e+02, threshold=4.143e+02, percent-clipped=3.0 2023-03-09 10:52:00,758 INFO [train2.py:809] (3/4) Epoch 28, batch 1000, loss[ctc_loss=0.05149, att_loss=0.2001, loss=0.1704, over 15485.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009744, over 36.00 utterances.], tot_loss[ctc_loss=0.06542, att_loss=0.2313, loss=0.1982, over 3260465.10 frames. utt_duration=1254 frames, utt_pad_proportion=0.05139, over 10411.31 utterances.], batch size: 36, lr: 3.85e-03, grad_scale: 16.0 2023-03-09 10:52:24,751 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108577.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:53:06,571 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6208, 3.7194, 3.5293, 3.7997, 2.6804, 3.8334, 2.8707, 2.0941], device='cuda:3'), covar=tensor([0.0522, 0.0367, 0.0778, 0.0350, 0.1306, 0.0291, 0.1208, 0.1413], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0188, 0.0268, 0.0183, 0.0224, 0.0169, 0.0237, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:53:20,718 INFO [train2.py:809] (3/4) Epoch 28, batch 1050, loss[ctc_loss=0.04849, att_loss=0.1987, loss=0.1686, over 14055.00 frames. utt_duration=1815 frames, utt_pad_proportion=0.04259, over 31.00 utterances.], tot_loss[ctc_loss=0.06556, att_loss=0.2318, loss=0.1985, over 3260971.23 frames. utt_duration=1227 frames, utt_pad_proportion=0.05782, over 10639.81 utterances.], batch size: 31, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 10:54:02,760 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108638.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:54:22,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-09 10:54:27,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.905e+02 2.250e+02 2.754e+02 7.007e+02, threshold=4.500e+02, percent-clipped=3.0 2023-03-09 10:54:31,505 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.4702, 2.3699, 2.2392, 2.3825, 2.7727, 2.5060, 2.2483, 2.8990], device='cuda:3'), covar=tensor([0.1365, 0.2109, 0.1642, 0.1349, 0.1553, 0.1156, 0.1875, 0.1302], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0148, 0.0143, 0.0138, 0.0155, 0.0132, 0.0154, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 10:54:40,461 INFO [train2.py:809] (3/4) Epoch 28, batch 1100, loss[ctc_loss=0.06667, att_loss=0.2381, loss=0.2038, over 16777.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005978, over 48.00 utterances.], tot_loss[ctc_loss=0.06571, att_loss=0.232, loss=0.1987, over 3262842.07 frames. utt_duration=1230 frames, utt_pad_proportion=0.05761, over 10624.47 utterances.], batch size: 48, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 10:55:10,683 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108681.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:56:00,173 INFO [train2.py:809] (3/4) Epoch 28, batch 1150, loss[ctc_loss=0.06085, att_loss=0.2319, loss=0.1977, over 16473.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006835, over 46.00 utterances.], tot_loss[ctc_loss=0.06582, att_loss=0.2319, loss=0.1987, over 3268269.95 frames. utt_duration=1232 frames, utt_pad_proportion=0.0566, over 10622.62 utterances.], batch size: 46, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 10:56:32,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-09 10:57:07,178 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 1.925e+02 2.338e+02 2.824e+02 7.096e+02, threshold=4.677e+02, percent-clipped=6.0 2023-03-09 10:57:19,311 INFO [train2.py:809] (3/4) Epoch 28, batch 1200, loss[ctc_loss=0.07345, att_loss=0.2424, loss=0.2086, over 17360.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02136, over 59.00 utterances.], tot_loss[ctc_loss=0.06526, att_loss=0.2317, loss=0.1984, over 3274420.30 frames. utt_duration=1239 frames, utt_pad_proportion=0.05291, over 10581.46 utterances.], batch size: 59, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 10:57:27,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-09 10:57:35,944 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-09 10:57:59,936 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-09 10:58:00,772 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108788.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 10:58:05,462 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108791.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:58:06,994 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4177, 2.5661, 4.8196, 3.7948, 2.9121, 4.0537, 4.5989, 4.4989], device='cuda:3'), covar=tensor([0.0306, 0.1651, 0.0238, 0.0940, 0.1808, 0.0304, 0.0248, 0.0311], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0247, 0.0224, 0.0324, 0.0268, 0.0239, 0.0214, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 10:58:39,217 INFO [train2.py:809] (3/4) Epoch 28, batch 1250, loss[ctc_loss=0.06374, att_loss=0.204, loss=0.176, over 15496.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008471, over 36.00 utterances.], tot_loss[ctc_loss=0.06472, att_loss=0.2309, loss=0.1977, over 3270984.07 frames. utt_duration=1248 frames, utt_pad_proportion=0.05189, over 10496.69 utterances.], batch size: 36, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 10:59:42,574 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108852.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 10:59:45,115 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.829e+02 2.128e+02 2.551e+02 3.725e+02, threshold=4.255e+02, percent-clipped=0.0 2023-03-09 10:59:51,520 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-03-09 10:59:58,173 INFO [train2.py:809] (3/4) Epoch 28, batch 1300, loss[ctc_loss=0.07008, att_loss=0.227, loss=0.1956, over 15946.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007524, over 41.00 utterances.], tot_loss[ctc_loss=0.06392, att_loss=0.2307, loss=0.1974, over 3281402.34 frames. utt_duration=1258 frames, utt_pad_proportion=0.0469, over 10445.71 utterances.], batch size: 41, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:00:35,145 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8303, 5.1350, 5.0088, 5.1494, 5.2930, 4.9056, 3.5133, 5.1567], device='cuda:3'), covar=tensor([0.0111, 0.0127, 0.0148, 0.0086, 0.0082, 0.0128, 0.0697, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0092, 0.0116, 0.0073, 0.0079, 0.0089, 0.0105, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:01:13,689 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1880, 4.6814, 4.7708, 4.7516, 3.0161, 4.6580, 2.8329, 2.0637], device='cuda:3'), covar=tensor([0.0474, 0.0250, 0.0535, 0.0289, 0.1318, 0.0218, 0.1347, 0.1508], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0187, 0.0268, 0.0183, 0.0224, 0.0169, 0.0236, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:01:17,645 INFO [train2.py:809] (3/4) Epoch 28, batch 1350, loss[ctc_loss=0.06155, att_loss=0.2119, loss=0.1819, over 14600.00 frames. utt_duration=1826 frames, utt_pad_proportion=0.04022, over 32.00 utterances.], tot_loss[ctc_loss=0.06313, att_loss=0.2301, loss=0.1967, over 3278168.67 frames. utt_duration=1284 frames, utt_pad_proportion=0.04184, over 10222.73 utterances.], batch size: 32, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:01:23,448 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-09 11:01:51,237 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108933.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:02:24,362 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.787e+02 2.090e+02 2.740e+02 6.253e+02, threshold=4.180e+02, percent-clipped=4.0 2023-03-09 11:02:36,720 INFO [train2.py:809] (3/4) Epoch 28, batch 1400, loss[ctc_loss=0.06069, att_loss=0.2305, loss=0.1966, over 16988.00 frames. utt_duration=687.8 frames, utt_pad_proportion=0.137, over 99.00 utterances.], tot_loss[ctc_loss=0.06368, att_loss=0.2304, loss=0.197, over 3275328.57 frames. utt_duration=1263 frames, utt_pad_proportion=0.04858, over 10385.27 utterances.], batch size: 99, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:03:07,547 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108981.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:03:26,036 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1323, 4.5784, 4.8121, 4.6886, 2.8024, 4.5617, 3.0569, 1.9686], device='cuda:3'), covar=tensor([0.0484, 0.0281, 0.0487, 0.0261, 0.1509, 0.0244, 0.1260, 0.1610], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0189, 0.0269, 0.0184, 0.0226, 0.0171, 0.0237, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:03:56,727 INFO [train2.py:809] (3/4) Epoch 28, batch 1450, loss[ctc_loss=0.05438, att_loss=0.2108, loss=0.1795, over 16290.00 frames. utt_duration=1517 frames, utt_pad_proportion=0.005974, over 43.00 utterances.], tot_loss[ctc_loss=0.06401, att_loss=0.2303, loss=0.197, over 3263910.69 frames. utt_duration=1233 frames, utt_pad_proportion=0.06089, over 10597.78 utterances.], batch size: 43, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:04:24,447 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109029.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:04:36,941 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109037.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 11:04:38,423 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8325, 3.5758, 3.5666, 3.1400, 3.5633, 3.6139, 3.6413, 2.7238], device='cuda:3'), covar=tensor([0.1051, 0.0924, 0.1611, 0.2525, 0.0826, 0.1901, 0.0805, 0.2822], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0209, 0.0225, 0.0275, 0.0185, 0.0286, 0.0207, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:05:03,519 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.756e+02 2.072e+02 2.557e+02 4.820e+02, threshold=4.143e+02, percent-clipped=2.0 2023-03-09 11:05:06,500 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-09 11:05:16,005 INFO [train2.py:809] (3/4) Epoch 28, batch 1500, loss[ctc_loss=0.06628, att_loss=0.2429, loss=0.2076, over 17088.00 frames. utt_duration=1291 frames, utt_pad_proportion=0.007628, over 53.00 utterances.], tot_loss[ctc_loss=0.06466, att_loss=0.2307, loss=0.1975, over 3268688.26 frames. utt_duration=1254 frames, utt_pad_proportion=0.05514, over 10437.70 utterances.], batch size: 53, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:05:30,592 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1974, 5.4847, 5.7498, 5.6122, 5.7085, 6.1833, 5.3918, 6.2128], device='cuda:3'), covar=tensor([0.0803, 0.0704, 0.0852, 0.1459, 0.2034, 0.0930, 0.0644, 0.0768], device='cuda:3'), in_proj_covar=tensor([0.0924, 0.0531, 0.0651, 0.0688, 0.0923, 0.0670, 0.0517, 0.0652], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:05:57,492 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109088.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:06:13,825 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109098.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 11:06:34,720 INFO [train2.py:809] (3/4) Epoch 28, batch 1550, loss[ctc_loss=0.06275, att_loss=0.2209, loss=0.1893, over 16006.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.007435, over 40.00 utterances.], tot_loss[ctc_loss=0.06416, att_loss=0.2301, loss=0.1969, over 3272185.71 frames. utt_duration=1271 frames, utt_pad_proportion=0.04961, over 10310.86 utterances.], batch size: 40, lr: 3.84e-03, grad_scale: 16.0 2023-03-09 11:07:05,330 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9518, 4.9490, 4.8604, 2.2189, 2.0846, 3.1381, 2.4786, 3.7748], device='cuda:3'), covar=tensor([0.0754, 0.0335, 0.0281, 0.4996, 0.5175, 0.2164, 0.3899, 0.1749], device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0303, 0.0281, 0.0255, 0.0342, 0.0334, 0.0266, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 11:07:12,460 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109136.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:07:30,576 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109147.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:07:41,456 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.952e+02 2.217e+02 2.637e+02 4.191e+02, threshold=4.435e+02, percent-clipped=1.0 2023-03-09 11:07:53,856 INFO [train2.py:809] (3/4) Epoch 28, batch 1600, loss[ctc_loss=0.06523, att_loss=0.216, loss=0.1859, over 15350.00 frames. utt_duration=1756 frames, utt_pad_proportion=0.01253, over 35.00 utterances.], tot_loss[ctc_loss=0.06447, att_loss=0.2311, loss=0.1978, over 3271318.20 frames. utt_duration=1252 frames, utt_pad_proportion=0.0545, over 10466.90 utterances.], batch size: 35, lr: 3.83e-03, grad_scale: 16.0 2023-03-09 11:08:44,703 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109194.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:09:12,896 INFO [train2.py:809] (3/4) Epoch 28, batch 1650, loss[ctc_loss=0.06827, att_loss=0.2483, loss=0.2123, over 16948.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.007895, over 50.00 utterances.], tot_loss[ctc_loss=0.06468, att_loss=0.2313, loss=0.198, over 3274954.22 frames. utt_duration=1273 frames, utt_pad_proportion=0.0496, over 10303.44 utterances.], batch size: 50, lr: 3.83e-03, grad_scale: 16.0 2023-03-09 11:09:16,217 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1452, 5.4110, 5.3516, 5.4145, 5.4924, 5.4116, 5.0850, 4.8851], device='cuda:3'), covar=tensor([0.1060, 0.0544, 0.0339, 0.0494, 0.0254, 0.0314, 0.0410, 0.0326], device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0384, 0.0377, 0.0380, 0.0444, 0.0450, 0.0380, 0.0417], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:09:34,194 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7868, 3.5235, 3.4528, 3.0594, 3.4930, 3.5357, 3.5179, 2.5863], device='cuda:3'), covar=tensor([0.1155, 0.1151, 0.2372, 0.2726, 0.1235, 0.2503, 0.1065, 0.3252], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0208, 0.0225, 0.0274, 0.0185, 0.0286, 0.0208, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:09:47,311 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109233.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:09:47,329 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5665, 2.9955, 3.3321, 4.4923, 3.9674, 4.0373, 3.0598, 2.5633], device='cuda:3'), covar=tensor([0.0704, 0.1744, 0.0901, 0.0472, 0.0713, 0.0462, 0.1302, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0219, 0.0188, 0.0226, 0.0234, 0.0192, 0.0205, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:10:22,242 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.344e+02 1.779e+02 2.256e+02 2.612e+02 6.095e+02, threshold=4.513e+02, percent-clipped=4.0 2023-03-09 11:10:22,736 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109255.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:10:33,283 INFO [train2.py:809] (3/4) Epoch 28, batch 1700, loss[ctc_loss=0.08525, att_loss=0.251, loss=0.2179, over 17120.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01507, over 56.00 utterances.], tot_loss[ctc_loss=0.06392, att_loss=0.2307, loss=0.1973, over 3281699.35 frames. utt_duration=1288 frames, utt_pad_proportion=0.04386, over 10203.74 utterances.], batch size: 56, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:11:03,717 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109281.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:11:23,981 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109294.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:11:52,253 INFO [train2.py:809] (3/4) Epoch 28, batch 1750, loss[ctc_loss=0.08932, att_loss=0.254, loss=0.2211, over 17066.00 frames. utt_duration=1221 frames, utt_pad_proportion=0.01799, over 56.00 utterances.], tot_loss[ctc_loss=0.06487, att_loss=0.2313, loss=0.198, over 3271899.10 frames. utt_duration=1234 frames, utt_pad_proportion=0.05965, over 10622.77 utterances.], batch size: 56, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:12:02,569 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109318.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:12:55,001 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7936, 3.5496, 3.4994, 3.0692, 3.5300, 3.5681, 3.5904, 2.6693], device='cuda:3'), covar=tensor([0.1022, 0.1076, 0.1976, 0.2842, 0.1199, 0.2786, 0.0976, 0.3187], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0206, 0.0223, 0.0273, 0.0184, 0.0284, 0.0206, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:13:00,660 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 1.738e+02 2.027e+02 2.395e+02 5.989e+02, threshold=4.055e+02, percent-clipped=1.0 2023-03-09 11:13:01,130 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109355.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:13:06,012 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8202, 3.8585, 3.7622, 4.1546, 2.4966, 4.0135, 2.8028, 1.9308], device='cuda:3'), covar=tensor([0.0515, 0.0294, 0.0837, 0.0301, 0.1779, 0.0305, 0.1404, 0.1648], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0189, 0.0268, 0.0183, 0.0226, 0.0171, 0.0235, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:13:11,725 INFO [train2.py:809] (3/4) Epoch 28, batch 1800, loss[ctc_loss=0.09973, att_loss=0.2218, loss=0.1974, over 15524.00 frames. utt_duration=1726 frames, utt_pad_proportion=0.00722, over 36.00 utterances.], tot_loss[ctc_loss=0.06467, att_loss=0.2312, loss=0.1979, over 3277926.28 frames. utt_duration=1248 frames, utt_pad_proportion=0.05471, over 10518.67 utterances.], batch size: 36, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:13:37,822 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0591, 5.3575, 5.5960, 5.4644, 5.5715, 5.9582, 5.2786, 6.0929], device='cuda:3'), covar=tensor([0.0720, 0.0876, 0.0877, 0.1384, 0.1778, 0.1073, 0.0718, 0.0695], device='cuda:3'), in_proj_covar=tensor([0.0927, 0.0533, 0.0654, 0.0690, 0.0921, 0.0670, 0.0517, 0.0654], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:13:39,597 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109379.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 11:14:01,820 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109393.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 11:14:31,016 INFO [train2.py:809] (3/4) Epoch 28, batch 1850, loss[ctc_loss=0.05609, att_loss=0.2182, loss=0.1858, over 16114.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.00691, over 42.00 utterances.], tot_loss[ctc_loss=0.06404, att_loss=0.2304, loss=0.1971, over 3272410.83 frames. utt_duration=1269 frames, utt_pad_proportion=0.05114, over 10328.22 utterances.], batch size: 42, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:15:13,204 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6760, 2.5525, 2.7781, 2.7471, 2.9086, 2.9043, 2.4977, 3.1453], device='cuda:3'), covar=tensor([0.1916, 0.2301, 0.1584, 0.1410, 0.1915, 0.1350, 0.2069, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0149, 0.0144, 0.0139, 0.0155, 0.0133, 0.0154, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 11:15:26,817 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109447.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:15:39,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.724e+02 2.158e+02 2.874e+02 5.021e+02, threshold=4.317e+02, percent-clipped=1.0 2023-03-09 11:15:42,593 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9418, 5.2828, 5.4619, 5.2342, 5.4198, 5.9224, 5.2308, 5.9768], device='cuda:3'), covar=tensor([0.0787, 0.0808, 0.0965, 0.1627, 0.2036, 0.0991, 0.0712, 0.0783], device='cuda:3'), in_proj_covar=tensor([0.0931, 0.0534, 0.0654, 0.0692, 0.0923, 0.0669, 0.0517, 0.0653], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:15:50,152 INFO [train2.py:809] (3/4) Epoch 28, batch 1900, loss[ctc_loss=0.08193, att_loss=0.2467, loss=0.2138, over 17082.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.0164, over 56.00 utterances.], tot_loss[ctc_loss=0.06418, att_loss=0.2303, loss=0.197, over 3262882.48 frames. utt_duration=1273 frames, utt_pad_proportion=0.05219, over 10262.26 utterances.], batch size: 56, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:16:07,930 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 11:16:12,025 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1484, 2.4756, 2.7199, 3.8209, 3.5590, 3.6385, 2.6559, 2.4237], device='cuda:3'), covar=tensor([0.0693, 0.1897, 0.1108, 0.0560, 0.0817, 0.0454, 0.1400, 0.1660], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0218, 0.0187, 0.0224, 0.0234, 0.0192, 0.0206, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:16:23,181 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-09 11:16:43,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109495.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:17:09,536 INFO [train2.py:809] (3/4) Epoch 28, batch 1950, loss[ctc_loss=0.04599, att_loss=0.1965, loss=0.1664, over 10524.00 frames. utt_duration=1832 frames, utt_pad_proportion=0.1218, over 23.00 utterances.], tot_loss[ctc_loss=0.06416, att_loss=0.2305, loss=0.1972, over 3262588.18 frames. utt_duration=1271 frames, utt_pad_proportion=0.0508, over 10279.29 utterances.], batch size: 23, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:18:01,993 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.34 vs. limit=5.0 2023-03-09 11:18:10,599 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109550.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:18:17,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 1.926e+02 2.337e+02 2.750e+02 5.694e+02, threshold=4.673e+02, percent-clipped=4.0 2023-03-09 11:18:28,475 INFO [train2.py:809] (3/4) Epoch 28, batch 2000, loss[ctc_loss=0.07033, att_loss=0.2431, loss=0.2086, over 17067.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.00791, over 52.00 utterances.], tot_loss[ctc_loss=0.06423, att_loss=0.2312, loss=0.1978, over 3272446.91 frames. utt_duration=1270 frames, utt_pad_proportion=0.04694, over 10321.03 utterances.], batch size: 52, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:18:55,935 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109579.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:18:57,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-03-09 11:19:47,005 INFO [train2.py:809] (3/4) Epoch 28, batch 2050, loss[ctc_loss=0.07529, att_loss=0.2532, loss=0.2176, over 17318.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02372, over 59.00 utterances.], tot_loss[ctc_loss=0.06376, att_loss=0.2309, loss=0.1975, over 3272907.70 frames. utt_duration=1276 frames, utt_pad_proportion=0.0461, over 10273.53 utterances.], batch size: 59, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:20:20,235 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 11:20:33,251 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109640.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:20:48,886 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109650.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:20:56,143 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 1.837e+02 2.171e+02 2.517e+02 4.514e+02, threshold=4.342e+02, percent-clipped=0.0 2023-03-09 11:21:02,535 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109659.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:21:06,772 INFO [train2.py:809] (3/4) Epoch 28, batch 2100, loss[ctc_loss=0.08504, att_loss=0.2492, loss=0.2163, over 17031.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.01003, over 52.00 utterances.], tot_loss[ctc_loss=0.06359, att_loss=0.231, loss=0.1975, over 3275780.06 frames. utt_duration=1284 frames, utt_pad_proportion=0.04468, over 10213.05 utterances.], batch size: 52, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:21:07,022 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.8171, 5.1391, 5.3097, 5.1873, 5.2767, 5.7933, 5.0869, 5.8719], device='cuda:3'), covar=tensor([0.0767, 0.0805, 0.0901, 0.1455, 0.2010, 0.0930, 0.0869, 0.0685], device='cuda:3'), in_proj_covar=tensor([0.0916, 0.0528, 0.0646, 0.0684, 0.0917, 0.0664, 0.0511, 0.0643], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:21:18,100 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1603, 5.5067, 5.0536, 5.5512, 4.8776, 5.1139, 5.6107, 5.3813], device='cuda:3'), covar=tensor([0.0551, 0.0298, 0.0784, 0.0332, 0.0392, 0.0216, 0.0238, 0.0210], device='cuda:3'), in_proj_covar=tensor([0.0408, 0.0346, 0.0386, 0.0384, 0.0342, 0.0250, 0.0327, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0006, 0.0005], device='cuda:3') 2023-03-09 11:21:21,872 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0849, 5.0400, 4.7977, 3.0592, 4.8812, 4.7099, 4.3887, 2.6816], device='cuda:3'), covar=tensor([0.0106, 0.0108, 0.0292, 0.0984, 0.0102, 0.0197, 0.0307, 0.1447], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0106, 0.0112, 0.0112, 0.0089, 0.0118, 0.0101, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:21:26,408 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109674.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 11:21:57,877 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109693.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 11:22:26,687 INFO [train2.py:809] (3/4) Epoch 28, batch 2150, loss[ctc_loss=0.05054, att_loss=0.2072, loss=0.1758, over 15502.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008594, over 36.00 utterances.], tot_loss[ctc_loss=0.06388, att_loss=0.2304, loss=0.1971, over 3268353.84 frames. utt_duration=1313 frames, utt_pad_proportion=0.03942, over 9969.56 utterances.], batch size: 36, lr: 3.83e-03, grad_scale: 8.0 2023-03-09 11:22:40,773 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109720.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:23:12,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-09 11:23:13,195 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109741.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 11:23:34,197 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.864e+02 2.379e+02 2.851e+02 5.187e+02, threshold=4.758e+02, percent-clipped=2.0 2023-03-09 11:23:45,127 INFO [train2.py:809] (3/4) Epoch 28, batch 2200, loss[ctc_loss=0.06099, att_loss=0.2349, loss=0.2001, over 17057.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008621, over 52.00 utterances.], tot_loss[ctc_loss=0.06433, att_loss=0.2307, loss=0.1974, over 3272349.39 frames. utt_duration=1286 frames, utt_pad_proportion=0.04512, over 10187.18 utterances.], batch size: 52, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:25:03,657 INFO [train2.py:809] (3/4) Epoch 28, batch 2250, loss[ctc_loss=0.08677, att_loss=0.2554, loss=0.2217, over 17375.00 frames. utt_duration=1179 frames, utt_pad_proportion=0.02045, over 59.00 utterances.], tot_loss[ctc_loss=0.06439, att_loss=0.2301, loss=0.197, over 3268205.52 frames. utt_duration=1307 frames, utt_pad_proportion=0.04091, over 10014.98 utterances.], batch size: 59, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:26:03,227 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109850.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:26:10,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 1.849e+02 2.212e+02 2.625e+02 5.541e+02, threshold=4.423e+02, percent-clipped=2.0 2023-03-09 11:26:21,267 INFO [train2.py:809] (3/4) Epoch 28, batch 2300, loss[ctc_loss=0.1443, att_loss=0.271, loss=0.2457, over 14129.00 frames. utt_duration=391.3 frames, utt_pad_proportion=0.3195, over 145.00 utterances.], tot_loss[ctc_loss=0.06451, att_loss=0.2304, loss=0.1972, over 3266201.46 frames. utt_duration=1288 frames, utt_pad_proportion=0.04571, over 10153.82 utterances.], batch size: 145, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:26:51,271 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2444, 2.6127, 4.6394, 3.7432, 3.1688, 4.0196, 4.2030, 4.2771], device='cuda:3'), covar=tensor([0.0268, 0.1451, 0.0222, 0.0840, 0.1408, 0.0295, 0.0276, 0.0336], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0248, 0.0226, 0.0325, 0.0270, 0.0242, 0.0217, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:27:17,569 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109898.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:27:38,455 INFO [train2.py:809] (3/4) Epoch 28, batch 2350, loss[ctc_loss=0.05896, att_loss=0.2245, loss=0.1914, over 16382.00 frames. utt_duration=1491 frames, utt_pad_proportion=0.007954, over 44.00 utterances.], tot_loss[ctc_loss=0.06487, att_loss=0.231, loss=0.1978, over 3273683.01 frames. utt_duration=1278 frames, utt_pad_proportion=0.04591, over 10258.53 utterances.], batch size: 44, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:27:47,053 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8843, 5.0088, 4.8047, 2.1072, 2.0240, 2.9892, 2.3488, 3.9007], device='cuda:3'), covar=tensor([0.0767, 0.0320, 0.0322, 0.5080, 0.5400, 0.2273, 0.3834, 0.1482], device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0300, 0.0278, 0.0251, 0.0338, 0.0331, 0.0263, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 11:27:47,643 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2023-03-09 11:28:16,094 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109935.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:28:23,955 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9697, 4.3170, 4.1351, 4.5736, 2.6550, 4.2753, 2.6375, 1.7125], device='cuda:3'), covar=tensor([0.0522, 0.0299, 0.0791, 0.0257, 0.1613, 0.0277, 0.1579, 0.1798], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0188, 0.0265, 0.0181, 0.0222, 0.0170, 0.0233, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:28:39,550 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109950.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:28:46,860 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.356e+02 1.839e+02 2.224e+02 2.680e+02 5.661e+02, threshold=4.448e+02, percent-clipped=2.0 2023-03-09 11:28:57,735 INFO [train2.py:809] (3/4) Epoch 28, batch 2400, loss[ctc_loss=0.06624, att_loss=0.2534, loss=0.2159, over 17055.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.007974, over 52.00 utterances.], tot_loss[ctc_loss=0.06476, att_loss=0.2311, loss=0.1978, over 3272726.68 frames. utt_duration=1261 frames, utt_pad_proportion=0.05007, over 10393.47 utterances.], batch size: 52, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:29:17,583 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109974.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:29:41,996 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7915, 3.5463, 3.4994, 3.1360, 3.6024, 3.6083, 3.6032, 2.5863], device='cuda:3'), covar=tensor([0.1108, 0.1235, 0.2028, 0.2900, 0.0843, 0.1588, 0.0833, 0.3318], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0208, 0.0226, 0.0276, 0.0186, 0.0287, 0.0208, 0.0233], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:29:48,452 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5269, 2.4959, 4.9288, 3.9869, 3.0660, 4.2168, 4.7760, 4.5846], device='cuda:3'), covar=tensor([0.0294, 0.1592, 0.0267, 0.0869, 0.1708, 0.0271, 0.0208, 0.0325], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0250, 0.0228, 0.0328, 0.0272, 0.0243, 0.0218, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:29:55,817 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109998.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:30:22,868 INFO [train2.py:809] (3/4) Epoch 28, batch 2450, loss[ctc_loss=0.06207, att_loss=0.2479, loss=0.2107, over 17307.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01066, over 55.00 utterances.], tot_loss[ctc_loss=0.06413, att_loss=0.2306, loss=0.1973, over 3276001.77 frames. utt_duration=1260 frames, utt_pad_proportion=0.04925, over 10408.62 utterances.], batch size: 55, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:30:28,558 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110015.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:30:28,636 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5546, 4.9146, 4.7356, 4.8651, 4.9641, 4.6757, 3.7344, 4.9223], device='cuda:3'), covar=tensor([0.0129, 0.0122, 0.0150, 0.0088, 0.0119, 0.0113, 0.0580, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0092, 0.0117, 0.0073, 0.0080, 0.0090, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:30:31,954 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110017.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:30:39,403 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110022.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:30:41,135 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5392, 4.9118, 4.7472, 4.8914, 5.0116, 4.6963, 3.5716, 4.9396], device='cuda:3'), covar=tensor([0.0124, 0.0121, 0.0135, 0.0089, 0.0080, 0.0104, 0.0641, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0093, 0.0118, 0.0073, 0.0080, 0.0090, 0.0105, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:31:12,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-09 11:31:14,789 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110044.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:31:30,264 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1467, 3.7851, 3.2538, 3.3822, 3.9547, 3.6387, 3.1012, 4.1910], device='cuda:3'), covar=tensor([0.0955, 0.0476, 0.1030, 0.0801, 0.0761, 0.0768, 0.0862, 0.0482], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0229, 0.0233, 0.0210, 0.0291, 0.0252, 0.0205, 0.0299], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 11:31:31,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.872e+02 2.152e+02 2.645e+02 5.223e+02, threshold=4.304e+02, percent-clipped=1.0 2023-03-09 11:31:41,822 INFO [train2.py:809] (3/4) Epoch 28, batch 2500, loss[ctc_loss=0.07889, att_loss=0.241, loss=0.2086, over 16456.00 frames. utt_duration=1432 frames, utt_pad_proportion=0.007979, over 46.00 utterances.], tot_loss[ctc_loss=0.06487, att_loss=0.2314, loss=0.1981, over 3274586.60 frames. utt_duration=1222 frames, utt_pad_proportion=0.05901, over 10728.75 utterances.], batch size: 46, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:32:07,604 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110078.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:32:50,575 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:32:58,107 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110110.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:33:01,504 INFO [train2.py:809] (3/4) Epoch 28, batch 2550, loss[ctc_loss=0.06359, att_loss=0.224, loss=0.1919, over 15883.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.00876, over 39.00 utterances.], tot_loss[ctc_loss=0.06478, att_loss=0.2313, loss=0.198, over 3269164.63 frames. utt_duration=1219 frames, utt_pad_proportion=0.0615, over 10740.91 utterances.], batch size: 39, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:33:07,179 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110115.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:33:41,656 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0262, 4.4718, 4.5109, 4.6880, 2.9108, 4.4198, 2.8604, 1.9694], device='cuda:3'), covar=tensor([0.0613, 0.0334, 0.0721, 0.0266, 0.1483, 0.0283, 0.1407, 0.1638], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0191, 0.0270, 0.0184, 0.0226, 0.0172, 0.0236, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:34:08,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 1.726e+02 2.160e+02 2.628e+02 5.720e+02, threshold=4.319e+02, percent-clipped=3.0 2023-03-09 11:34:19,906 INFO [train2.py:809] (3/4) Epoch 28, batch 2600, loss[ctc_loss=0.05588, att_loss=0.2027, loss=0.1733, over 15748.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.008913, over 38.00 utterances.], tot_loss[ctc_loss=0.06429, att_loss=0.2305, loss=0.1973, over 3261653.17 frames. utt_duration=1250 frames, utt_pad_proportion=0.05626, over 10453.93 utterances.], batch size: 38, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:34:34,630 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110171.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:34:42,169 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110176.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:35:38,098 INFO [train2.py:809] (3/4) Epoch 28, batch 2650, loss[ctc_loss=0.05156, att_loss=0.233, loss=0.1967, over 16969.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007465, over 50.00 utterances.], tot_loss[ctc_loss=0.06416, att_loss=0.2301, loss=0.1969, over 3261987.36 frames. utt_duration=1278 frames, utt_pad_proportion=0.05023, over 10218.53 utterances.], batch size: 50, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:36:02,627 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9469, 4.2539, 4.1689, 4.5422, 2.5275, 4.2173, 2.6591, 1.5804], device='cuda:3'), covar=tensor([0.0564, 0.0355, 0.0807, 0.0278, 0.1743, 0.0298, 0.1606, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0190, 0.0268, 0.0183, 0.0225, 0.0171, 0.0234, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:36:16,247 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110235.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:36:46,739 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 1.822e+02 2.163e+02 2.650e+02 7.141e+02, threshold=4.325e+02, percent-clipped=4.0 2023-03-09 11:36:47,018 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1890, 5.4564, 5.3662, 5.3760, 5.5013, 5.4889, 5.1253, 4.9527], device='cuda:3'), covar=tensor([0.1033, 0.0544, 0.0310, 0.0560, 0.0263, 0.0264, 0.0401, 0.0305], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0375, 0.0371, 0.0378, 0.0438, 0.0444, 0.0376, 0.0411], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:36:57,431 INFO [train2.py:809] (3/4) Epoch 28, batch 2700, loss[ctc_loss=0.05111, att_loss=0.21, loss=0.1782, over 16134.00 frames. utt_duration=1538 frames, utt_pad_proportion=0.00519, over 42.00 utterances.], tot_loss[ctc_loss=0.06376, att_loss=0.2305, loss=0.1972, over 3272471.65 frames. utt_duration=1275 frames, utt_pad_proportion=0.0481, over 10277.77 utterances.], batch size: 42, lr: 3.82e-03, grad_scale: 8.0 2023-03-09 11:37:32,613 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:37:32,891 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:38:03,455 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110303.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:38:16,815 INFO [train2.py:809] (3/4) Epoch 28, batch 2750, loss[ctc_loss=0.06354, att_loss=0.2404, loss=0.2051, over 16612.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.005398, over 47.00 utterances.], tot_loss[ctc_loss=0.06324, att_loss=0.2304, loss=0.197, over 3284342.87 frames. utt_duration=1275 frames, utt_pad_proportion=0.04436, over 10317.99 utterances.], batch size: 47, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:38:22,993 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110315.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:38:27,619 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110318.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:38:36,820 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0257, 4.3023, 4.3505, 4.4875, 2.6073, 4.3570, 2.6779, 1.8618], device='cuda:3'), covar=tensor([0.0466, 0.0296, 0.0670, 0.0281, 0.1654, 0.0255, 0.1522, 0.1736], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0190, 0.0268, 0.0182, 0.0224, 0.0171, 0.0233, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:39:08,942 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110344.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:39:13,593 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.2407, 2.2032, 2.4598, 2.5194, 2.7233, 2.4544, 2.0929, 2.5152], device='cuda:3'), covar=tensor([0.1848, 0.2216, 0.1722, 0.1259, 0.1364, 0.1266, 0.1839, 0.1686], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0149, 0.0144, 0.0139, 0.0156, 0.0134, 0.0154, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 11:39:25,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 1.842e+02 2.112e+02 2.497e+02 5.270e+02, threshold=4.224e+02, percent-clipped=1.0 2023-03-09 11:39:34,249 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.04 vs. limit=5.0 2023-03-09 11:39:36,861 INFO [train2.py:809] (3/4) Epoch 28, batch 2800, loss[ctc_loss=0.1035, att_loss=0.254, loss=0.2239, over 14064.00 frames. utt_duration=389.6 frames, utt_pad_proportion=0.3236, over 145.00 utterances.], tot_loss[ctc_loss=0.06405, att_loss=0.2308, loss=0.1974, over 3278329.43 frames. utt_duration=1234 frames, utt_pad_proportion=0.05605, over 10637.12 utterances.], batch size: 145, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:39:39,133 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110363.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:39:40,863 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110364.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:39:54,237 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110373.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:40:04,286 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:40:37,360 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110400.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:40:55,888 INFO [train2.py:809] (3/4) Epoch 28, batch 2850, loss[ctc_loss=0.07361, att_loss=0.2342, loss=0.2021, over 16688.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.005654, over 46.00 utterances.], tot_loss[ctc_loss=0.06382, att_loss=0.2304, loss=0.1971, over 3269583.78 frames. utt_duration=1263 frames, utt_pad_proportion=0.05021, over 10364.52 utterances.], batch size: 46, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:42:02,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.889e+02 2.295e+02 2.895e+02 8.128e+02, threshold=4.591e+02, percent-clipped=8.0 2023-03-09 11:42:14,321 INFO [train2.py:809] (3/4) Epoch 28, batch 2900, loss[ctc_loss=0.06872, att_loss=0.247, loss=0.2113, over 17356.00 frames. utt_duration=1103 frames, utt_pad_proportion=0.03556, over 63.00 utterances.], tot_loss[ctc_loss=0.0645, att_loss=0.231, loss=0.1977, over 3270273.63 frames. utt_duration=1234 frames, utt_pad_proportion=0.05707, over 10612.74 utterances.], batch size: 63, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:42:21,228 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110466.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:42:28,818 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110471.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:42:35,102 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.3947, 2.3906, 2.4208, 2.4551, 2.8403, 2.5815, 2.3260, 2.7411], device='cuda:3'), covar=tensor([0.1425, 0.2209, 0.1713, 0.1118, 0.1425, 0.1123, 0.1430, 0.1333], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0146, 0.0142, 0.0137, 0.0154, 0.0132, 0.0153, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 11:43:32,999 INFO [train2.py:809] (3/4) Epoch 28, batch 2950, loss[ctc_loss=0.05063, att_loss=0.2215, loss=0.1873, over 15957.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006269, over 41.00 utterances.], tot_loss[ctc_loss=0.06493, att_loss=0.2315, loss=0.1982, over 3280246.82 frames. utt_duration=1223 frames, utt_pad_proportion=0.05649, over 10745.54 utterances.], batch size: 41, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:44:31,441 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1948, 5.4382, 5.3385, 5.3906, 5.4897, 5.4671, 5.1341, 4.9408], device='cuda:3'), covar=tensor([0.0990, 0.0567, 0.0328, 0.0487, 0.0274, 0.0307, 0.0371, 0.0290], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0374, 0.0370, 0.0376, 0.0436, 0.0440, 0.0374, 0.0409], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:44:39,991 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.946e+01 1.771e+02 2.102e+02 2.638e+02 6.862e+02, threshold=4.204e+02, percent-clipped=2.0 2023-03-09 11:44:52,459 INFO [train2.py:809] (3/4) Epoch 28, batch 3000, loss[ctc_loss=0.08128, att_loss=0.2554, loss=0.2206, over 17318.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.02351, over 59.00 utterances.], tot_loss[ctc_loss=0.06516, att_loss=0.2315, loss=0.1982, over 3277569.85 frames. utt_duration=1220 frames, utt_pad_proportion=0.05858, over 10755.43 utterances.], batch size: 59, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:44:52,460 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 11:45:06,793 INFO [train2.py:843] (3/4) Epoch 28, validation: ctc_loss=0.04082, att_loss=0.2346, loss=0.1958, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 11:45:06,793 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 11:45:33,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-09 11:46:25,788 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110611.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:46:26,958 INFO [train2.py:809] (3/4) Epoch 28, batch 3050, loss[ctc_loss=0.05401, att_loss=0.2161, loss=0.1837, over 15858.00 frames. utt_duration=1628 frames, utt_pad_proportion=0.01081, over 39.00 utterances.], tot_loss[ctc_loss=0.06447, att_loss=0.2311, loss=0.1977, over 3278297.17 frames. utt_duration=1208 frames, utt_pad_proportion=0.06105, over 10872.70 utterances.], batch size: 39, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:47:10,745 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110639.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:47:35,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 1.799e+02 2.141e+02 2.649e+02 5.384e+02, threshold=4.281e+02, percent-clipped=1.0 2023-03-09 11:47:42,927 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110659.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:47:47,514 INFO [train2.py:809] (3/4) Epoch 28, batch 3100, loss[ctc_loss=0.06164, att_loss=0.2291, loss=0.1956, over 17028.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.007511, over 51.00 utterances.], tot_loss[ctc_loss=0.06451, att_loss=0.2313, loss=0.198, over 3275079.43 frames. utt_duration=1209 frames, utt_pad_proportion=0.06139, over 10845.13 utterances.], batch size: 51, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:48:03,639 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110672.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:48:05,211 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110673.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:48:05,271 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3446, 3.9067, 3.4296, 3.6035, 4.1862, 3.7776, 3.3717, 4.4529], device='cuda:3'), covar=tensor([0.0919, 0.0507, 0.0990, 0.0673, 0.0647, 0.0730, 0.0711, 0.0400], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0229, 0.0232, 0.0210, 0.0290, 0.0250, 0.0205, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2023-03-09 11:48:06,659 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110674.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:48:11,648 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5520, 4.5757, 4.7033, 4.5994, 5.3174, 4.4858, 4.6103, 2.8032], device='cuda:3'), covar=tensor([0.0319, 0.0419, 0.0341, 0.0417, 0.0692, 0.0286, 0.0366, 0.1567], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0226, 0.0223, 0.0238, 0.0385, 0.0195, 0.0212, 0.0222], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 11:48:48,161 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110700.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:49:07,620 INFO [train2.py:809] (3/4) Epoch 28, batch 3150, loss[ctc_loss=0.05397, att_loss=0.2096, loss=0.1785, over 16393.00 frames. utt_duration=1492 frames, utt_pad_proportion=0.007334, over 44.00 utterances.], tot_loss[ctc_loss=0.06446, att_loss=0.2312, loss=0.1979, over 3265628.55 frames. utt_duration=1216 frames, utt_pad_proportion=0.06301, over 10755.64 utterances.], batch size: 44, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:49:14,605 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-09 11:49:21,680 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110721.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:49:46,559 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110736.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:50:04,805 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110748.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:50:15,627 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 1.871e+02 2.189e+02 2.637e+02 5.370e+02, threshold=4.378e+02, percent-clipped=2.0 2023-03-09 11:50:27,802 INFO [train2.py:809] (3/4) Epoch 28, batch 3200, loss[ctc_loss=0.04608, att_loss=0.2193, loss=0.1846, over 15786.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007804, over 38.00 utterances.], tot_loss[ctc_loss=0.06437, att_loss=0.2313, loss=0.1979, over 3265000.21 frames. utt_duration=1210 frames, utt_pad_proportion=0.06549, over 10810.14 utterances.], batch size: 38, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:50:34,395 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110766.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:50:42,170 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110771.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:50:45,162 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5588, 5.2551, 5.2950, 5.3769, 5.2437, 5.3285, 5.0953, 4.8343], device='cuda:3'), covar=tensor([0.1598, 0.0738, 0.0381, 0.0501, 0.0617, 0.0419, 0.0421, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0375, 0.0371, 0.0375, 0.0436, 0.0442, 0.0375, 0.0409], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:51:08,230 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1299, 5.3701, 5.2654, 5.2938, 5.4287, 5.3972, 5.0383, 4.8988], device='cuda:3'), covar=tensor([0.0942, 0.0550, 0.0296, 0.0574, 0.0263, 0.0319, 0.0403, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0375, 0.0371, 0.0375, 0.0436, 0.0443, 0.0375, 0.0409], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 11:51:23,935 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110797.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 11:51:48,163 INFO [train2.py:809] (3/4) Epoch 28, batch 3250, loss[ctc_loss=0.04447, att_loss=0.2164, loss=0.182, over 15969.00 frames. utt_duration=1559 frames, utt_pad_proportion=0.005538, over 41.00 utterances.], tot_loss[ctc_loss=0.0637, att_loss=0.2312, loss=0.1977, over 3270270.68 frames. utt_duration=1246 frames, utt_pad_proportion=0.05589, over 10507.39 utterances.], batch size: 41, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:51:51,349 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110814.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:51:59,336 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110819.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:52:55,399 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.746e+02 2.093e+02 2.617e+02 5.212e+02, threshold=4.186e+02, percent-clipped=3.0 2023-03-09 11:53:07,980 INFO [train2.py:809] (3/4) Epoch 28, batch 3300, loss[ctc_loss=0.083, att_loss=0.2465, loss=0.2138, over 17362.00 frames. utt_duration=1008 frames, utt_pad_proportion=0.04919, over 69.00 utterances.], tot_loss[ctc_loss=0.06384, att_loss=0.2307, loss=0.1973, over 3262633.21 frames. utt_duration=1244 frames, utt_pad_proportion=0.05715, over 10499.43 utterances.], batch size: 69, lr: 3.81e-03, grad_scale: 8.0 2023-03-09 11:53:17,674 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:54:27,399 INFO [train2.py:809] (3/4) Epoch 28, batch 3350, loss[ctc_loss=0.1012, att_loss=0.2527, loss=0.2224, over 17277.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01335, over 55.00 utterances.], tot_loss[ctc_loss=0.06441, att_loss=0.2312, loss=0.1978, over 3267953.49 frames. utt_duration=1257 frames, utt_pad_proportion=0.05304, over 10413.61 utterances.], batch size: 55, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 11:54:54,367 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110929.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:55:10,312 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110939.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:55:36,025 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 1.730e+02 2.154e+02 2.752e+02 9.084e+02, threshold=4.307e+02, percent-clipped=4.0 2023-03-09 11:55:41,069 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:55:42,572 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110959.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:55:46,720 INFO [train2.py:809] (3/4) Epoch 28, batch 3400, loss[ctc_loss=0.06301, att_loss=0.2247, loss=0.1923, over 16544.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005435, over 45.00 utterances.], tot_loss[ctc_loss=0.06454, att_loss=0.2311, loss=0.1978, over 3262509.12 frames. utt_duration=1225 frames, utt_pad_proportion=0.06253, over 10663.33 utterances.], batch size: 45, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 11:55:54,612 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110967.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:56:05,573 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110974.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:56:24,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-09 11:56:26,762 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110987.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:56:58,895 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111007.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:57:06,335 INFO [train2.py:809] (3/4) Epoch 28, batch 3450, loss[ctc_loss=0.06069, att_loss=0.2368, loss=0.2015, over 16947.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.007968, over 50.00 utterances.], tot_loss[ctc_loss=0.06461, att_loss=0.2314, loss=0.1981, over 3270070.81 frames. utt_duration=1241 frames, utt_pad_proportion=0.05717, over 10555.13 utterances.], batch size: 50, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 11:57:17,585 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111019.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:57:21,940 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111022.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 11:58:14,707 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 1.799e+02 2.104e+02 2.788e+02 6.212e+02, threshold=4.209e+02, percent-clipped=4.0 2023-03-09 11:58:25,586 INFO [train2.py:809] (3/4) Epoch 28, batch 3500, loss[ctc_loss=0.05548, att_loss=0.2407, loss=0.2036, over 16767.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.006673, over 48.00 utterances.], tot_loss[ctc_loss=0.06465, att_loss=0.2317, loss=0.1983, over 3270103.06 frames. utt_duration=1232 frames, utt_pad_proportion=0.05993, over 10626.75 utterances.], batch size: 48, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 11:58:33,881 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7847, 3.9669, 3.9524, 3.9977, 4.0587, 3.8046, 3.0959, 3.9214], device='cuda:3'), covar=tensor([0.0141, 0.0129, 0.0153, 0.0090, 0.0102, 0.0138, 0.0643, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0093, 0.0118, 0.0073, 0.0080, 0.0091, 0.0106, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 11:59:14,532 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111092.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 11:59:46,738 INFO [train2.py:809] (3/4) Epoch 28, batch 3550, loss[ctc_loss=0.1077, att_loss=0.2595, loss=0.2291, over 14337.00 frames. utt_duration=394.5 frames, utt_pad_proportion=0.3139, over 146.00 utterances.], tot_loss[ctc_loss=0.0642, att_loss=0.2317, loss=0.1982, over 3269628.29 frames. utt_duration=1227 frames, utt_pad_proportion=0.06017, over 10668.39 utterances.], batch size: 146, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 12:00:55,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.811e+02 2.154e+02 2.613e+02 6.278e+02, threshold=4.308e+02, percent-clipped=2.0 2023-03-09 12:01:05,922 INFO [train2.py:809] (3/4) Epoch 28, batch 3600, loss[ctc_loss=0.08936, att_loss=0.2559, loss=0.2226, over 17037.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.009637, over 52.00 utterances.], tot_loss[ctc_loss=0.06388, att_loss=0.2313, loss=0.1978, over 3277644.98 frames. utt_duration=1249 frames, utt_pad_proportion=0.05387, over 10508.38 utterances.], batch size: 52, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 12:01:11,549 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8633, 5.1416, 4.6534, 5.2213, 4.6111, 4.8457, 5.2465, 5.0500], device='cuda:3'), covar=tensor([0.0615, 0.0303, 0.0823, 0.0341, 0.0406, 0.0316, 0.0254, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0403, 0.0339, 0.0378, 0.0380, 0.0338, 0.0245, 0.0319, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 12:01:22,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-09 12:02:25,882 INFO [train2.py:809] (3/4) Epoch 28, batch 3650, loss[ctc_loss=0.09407, att_loss=0.2364, loss=0.208, over 16253.00 frames. utt_duration=1513 frames, utt_pad_proportion=0.00826, over 43.00 utterances.], tot_loss[ctc_loss=0.06472, att_loss=0.2314, loss=0.1981, over 3269098.36 frames. utt_duration=1213 frames, utt_pad_proportion=0.06607, over 10790.33 utterances.], batch size: 43, lr: 3.80e-03, grad_scale: 8.0 2023-03-09 12:02:45,384 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111224.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:02:47,122 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111225.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:03:34,857 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 1.792e+02 2.128e+02 2.694e+02 5.056e+02, threshold=4.257e+02, percent-clipped=2.0 2023-03-09 12:03:45,878 INFO [train2.py:809] (3/4) Epoch 28, batch 3700, loss[ctc_loss=0.06455, att_loss=0.2454, loss=0.2093, over 17293.00 frames. utt_duration=1174 frames, utt_pad_proportion=0.02409, over 59.00 utterances.], tot_loss[ctc_loss=0.06485, att_loss=0.2314, loss=0.1981, over 3263946.89 frames. utt_duration=1194 frames, utt_pad_proportion=0.07199, over 10952.55 utterances.], batch size: 59, lr: 3.80e-03, grad_scale: 16.0 2023-03-09 12:03:54,505 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111267.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:04:13,850 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5031, 2.7484, 4.9705, 3.8116, 3.1242, 4.2456, 4.7498, 4.6242], device='cuda:3'), covar=tensor([0.0308, 0.1443, 0.0218, 0.1165, 0.1676, 0.0287, 0.0208, 0.0317], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0252, 0.0231, 0.0330, 0.0275, 0.0246, 0.0222, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:04:24,253 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111286.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:05:06,059 INFO [train2.py:809] (3/4) Epoch 28, batch 3750, loss[ctc_loss=0.08276, att_loss=0.2388, loss=0.2076, over 16695.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006126, over 46.00 utterances.], tot_loss[ctc_loss=0.06501, att_loss=0.2315, loss=0.1982, over 3273263.51 frames. utt_duration=1208 frames, utt_pad_proportion=0.06596, over 10851.55 utterances.], batch size: 46, lr: 3.80e-03, grad_scale: 16.0 2023-03-09 12:05:09,376 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111314.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:05:10,844 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111315.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:05:45,320 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-09 12:05:58,597 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1819, 4.3116, 4.4370, 4.4343, 4.9149, 4.2775, 4.3718, 2.5832], device='cuda:3'), covar=tensor([0.0396, 0.0464, 0.0411, 0.0393, 0.0853, 0.0325, 0.0415, 0.1709], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0227, 0.0224, 0.0239, 0.0385, 0.0196, 0.0215, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:06:13,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.228e+02 1.865e+02 2.265e+02 2.758e+02 6.117e+02, threshold=4.530e+02, percent-clipped=1.0 2023-03-09 12:06:25,560 INFO [train2.py:809] (3/4) Epoch 28, batch 3800, loss[ctc_loss=0.06106, att_loss=0.2358, loss=0.2009, over 17005.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008457, over 51.00 utterances.], tot_loss[ctc_loss=0.06472, att_loss=0.2313, loss=0.198, over 3277383.95 frames. utt_duration=1226 frames, utt_pad_proportion=0.06012, over 10702.30 utterances.], batch size: 51, lr: 3.80e-03, grad_scale: 16.0 2023-03-09 12:07:13,981 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111392.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 12:07:44,858 INFO [train2.py:809] (3/4) Epoch 28, batch 3850, loss[ctc_loss=0.05384, att_loss=0.2316, loss=0.196, over 16770.00 frames. utt_duration=1399 frames, utt_pad_proportion=0.005656, over 48.00 utterances.], tot_loss[ctc_loss=0.06422, att_loss=0.2308, loss=0.1975, over 3273654.59 frames. utt_duration=1239 frames, utt_pad_proportion=0.0569, over 10582.07 utterances.], batch size: 48, lr: 3.80e-03, grad_scale: 16.0 2023-03-09 12:08:26,915 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.5954, 2.9222, 3.6082, 4.5984, 4.0969, 4.0600, 3.0354, 2.5850], device='cuda:3'), covar=tensor([0.0689, 0.1948, 0.0736, 0.0567, 0.0772, 0.0536, 0.1469, 0.1912], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0221, 0.0186, 0.0230, 0.0236, 0.0194, 0.0208, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:08:28,208 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111440.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:08:51,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 1.814e+02 2.141e+02 2.602e+02 6.153e+02, threshold=4.282e+02, percent-clipped=2.0 2023-03-09 12:08:56,873 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9501, 3.5194, 3.7141, 2.7890, 3.6205, 3.7247, 3.7055, 2.2627], device='cuda:3'), covar=tensor([0.1192, 0.1545, 0.1228, 0.5242, 0.1076, 0.1841, 0.0828, 0.5695], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0211, 0.0228, 0.0278, 0.0187, 0.0287, 0.0209, 0.0234], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:09:02,534 INFO [train2.py:809] (3/4) Epoch 28, batch 3900, loss[ctc_loss=0.05538, att_loss=0.2277, loss=0.1932, over 16875.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.006484, over 49.00 utterances.], tot_loss[ctc_loss=0.06395, att_loss=0.2305, loss=0.1972, over 3274396.61 frames. utt_duration=1239 frames, utt_pad_proportion=0.05653, over 10585.32 utterances.], batch size: 49, lr: 3.80e-03, grad_scale: 16.0 2023-03-09 12:09:12,042 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111468.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:10:19,197 INFO [train2.py:809] (3/4) Epoch 28, batch 3950, loss[ctc_loss=0.0529, att_loss=0.2087, loss=0.1775, over 15887.00 frames. utt_duration=1631 frames, utt_pad_proportion=0.0093, over 39.00 utterances.], tot_loss[ctc_loss=0.06419, att_loss=0.2302, loss=0.197, over 3270517.44 frames. utt_duration=1243 frames, utt_pad_proportion=0.05703, over 10540.49 utterances.], batch size: 39, lr: 3.79e-03, grad_scale: 16.0 2023-03-09 12:10:37,693 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111524.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:10:40,783 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0659, 4.3533, 4.1899, 4.4783, 2.8579, 4.4316, 2.7016, 1.7285], device='cuda:3'), covar=tensor([0.0531, 0.0322, 0.0841, 0.0327, 0.1525, 0.0245, 0.1473, 0.1705], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0195, 0.0276, 0.0187, 0.0231, 0.0176, 0.0239, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:10:45,244 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111529.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:11:24,278 INFO [train2.py:809] (3/4) Epoch 29, batch 0, loss[ctc_loss=0.06084, att_loss=0.2262, loss=0.1931, over 16528.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006396, over 45.00 utterances.], tot_loss[ctc_loss=0.06084, att_loss=0.2262, loss=0.1931, over 16528.00 frames. utt_duration=1471 frames, utt_pad_proportion=0.006396, over 45.00 utterances.], batch size: 45, lr: 3.73e-03, grad_scale: 8.0 2023-03-09 12:11:24,278 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 12:11:36,591 INFO [train2.py:843] (3/4) Epoch 29, validation: ctc_loss=0.04125, att_loss=0.2346, loss=0.1959, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 12:11:36,592 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 12:11:54,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.789e+02 2.164e+02 2.545e+02 5.866e+02, threshold=4.328e+02, percent-clipped=2.0 2023-03-09 12:12:19,811 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111572.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:12:34,358 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111581.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:12:55,750 INFO [train2.py:809] (3/4) Epoch 29, batch 50, loss[ctc_loss=0.06728, att_loss=0.247, loss=0.2111, over 17260.00 frames. utt_duration=1257 frames, utt_pad_proportion=0.01329, over 55.00 utterances.], tot_loss[ctc_loss=0.06619, att_loss=0.2328, loss=0.1995, over 735560.35 frames. utt_duration=1235 frames, utt_pad_proportion=0.06542, over 2384.37 utterances.], batch size: 55, lr: 3.73e-03, grad_scale: 8.0 2023-03-09 12:13:26,778 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111614.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:14:15,534 INFO [train2.py:809] (3/4) Epoch 29, batch 100, loss[ctc_loss=0.07635, att_loss=0.2351, loss=0.2033, over 16412.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006306, over 44.00 utterances.], tot_loss[ctc_loss=0.06569, att_loss=0.2333, loss=0.1998, over 1308806.22 frames. utt_duration=1295 frames, utt_pad_proportion=0.04046, over 4047.09 utterances.], batch size: 44, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:14:33,658 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.812e+02 2.121e+02 2.575e+02 9.495e+02, threshold=4.242e+02, percent-clipped=2.0 2023-03-09 12:14:42,933 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111662.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:14:44,848 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7759, 3.3823, 3.3648, 2.9098, 3.3343, 3.3848, 3.4023, 2.4027], device='cuda:3'), covar=tensor([0.0906, 0.0998, 0.1582, 0.2681, 0.1113, 0.1463, 0.0883, 0.3042], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0207, 0.0225, 0.0275, 0.0184, 0.0283, 0.0206, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:15:35,266 INFO [train2.py:809] (3/4) Epoch 29, batch 150, loss[ctc_loss=0.06275, att_loss=0.2125, loss=0.1825, over 15513.00 frames. utt_duration=1725 frames, utt_pad_proportion=0.008051, over 36.00 utterances.], tot_loss[ctc_loss=0.06528, att_loss=0.232, loss=0.1986, over 1734362.19 frames. utt_duration=1238 frames, utt_pad_proportion=0.06071, over 5611.26 utterances.], batch size: 36, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:16:57,989 INFO [train2.py:809] (3/4) Epoch 29, batch 200, loss[ctc_loss=0.09265, att_loss=0.2508, loss=0.2192, over 13802.00 frames. utt_duration=377.1 frames, utt_pad_proportion=0.3396, over 147.00 utterances.], tot_loss[ctc_loss=0.06447, att_loss=0.2314, loss=0.198, over 2069739.32 frames. utt_duration=1206 frames, utt_pad_proportion=0.07027, over 6872.26 utterances.], batch size: 147, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:17:15,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 1.846e+02 2.300e+02 2.696e+02 5.049e+02, threshold=4.600e+02, percent-clipped=1.0 2023-03-09 12:18:12,024 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111791.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 12:18:17,864 INFO [train2.py:809] (3/4) Epoch 29, batch 250, loss[ctc_loss=0.06208, att_loss=0.2387, loss=0.2034, over 16859.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.008022, over 49.00 utterances.], tot_loss[ctc_loss=0.06402, att_loss=0.2312, loss=0.1978, over 2346289.51 frames. utt_duration=1253 frames, utt_pad_proportion=0.05488, over 7501.51 utterances.], batch size: 49, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:18:48,321 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5994, 2.5344, 2.3323, 2.5555, 2.8542, 2.6431, 2.4678, 2.7636], device='cuda:3'), covar=tensor([0.1812, 0.2028, 0.1830, 0.1381, 0.1691, 0.1183, 0.1897, 0.1599], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0148, 0.0145, 0.0139, 0.0157, 0.0135, 0.0157, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 12:19:05,348 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111824.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:19:16,791 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-09 12:19:28,761 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9696, 4.0281, 4.0075, 4.3087, 2.6507, 4.2519, 2.5857, 1.6799], device='cuda:3'), covar=tensor([0.0548, 0.0319, 0.0733, 0.0276, 0.1566, 0.0259, 0.1511, 0.1786], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0190, 0.0267, 0.0182, 0.0224, 0.0172, 0.0232, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:19:37,440 INFO [train2.py:809] (3/4) Epoch 29, batch 300, loss[ctc_loss=0.05561, att_loss=0.2113, loss=0.1802, over 15868.00 frames. utt_duration=1629 frames, utt_pad_proportion=0.0102, over 39.00 utterances.], tot_loss[ctc_loss=0.06451, att_loss=0.2319, loss=0.1984, over 2551139.71 frames. utt_duration=1231 frames, utt_pad_proportion=0.05867, over 8298.65 utterances.], batch size: 39, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:19:49,980 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111852.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 12:19:55,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.741e+02 2.135e+02 2.604e+02 1.132e+03, threshold=4.270e+02, percent-clipped=2.0 2023-03-09 12:20:35,429 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111881.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:20:56,613 INFO [train2.py:809] (3/4) Epoch 29, batch 350, loss[ctc_loss=0.07013, att_loss=0.2461, loss=0.2109, over 16631.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005005, over 47.00 utterances.], tot_loss[ctc_loss=0.06442, att_loss=0.232, loss=0.1985, over 2714567.64 frames. utt_duration=1224 frames, utt_pad_proportion=0.05902, over 8880.46 utterances.], batch size: 47, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:21:12,119 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.89 vs. limit=5.0 2023-03-09 12:21:50,817 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111929.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:22:16,060 INFO [train2.py:809] (3/4) Epoch 29, batch 400, loss[ctc_loss=0.06691, att_loss=0.2469, loss=0.2109, over 16484.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005499, over 46.00 utterances.], tot_loss[ctc_loss=0.06439, att_loss=0.2312, loss=0.1979, over 2831194.56 frames. utt_duration=1242 frames, utt_pad_proportion=0.05754, over 9131.91 utterances.], batch size: 46, lr: 3.72e-03, grad_scale: 8.0 2023-03-09 12:22:34,681 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.870e+02 2.171e+02 2.633e+02 6.586e+02, threshold=4.343e+02, percent-clipped=4.0 2023-03-09 12:23:38,453 INFO [train2.py:809] (3/4) Epoch 29, batch 450, loss[ctc_loss=0.06372, att_loss=0.2376, loss=0.2029, over 17049.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008955, over 52.00 utterances.], tot_loss[ctc_loss=0.06403, att_loss=0.2313, loss=0.1979, over 2933595.89 frames. utt_duration=1265 frames, utt_pad_proportion=0.05104, over 9285.26 utterances.], batch size: 52, lr: 3.72e-03, grad_scale: 4.0 2023-03-09 12:25:02,539 INFO [train2.py:809] (3/4) Epoch 29, batch 500, loss[ctc_loss=0.06229, att_loss=0.2392, loss=0.2038, over 16891.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.00547, over 49.00 utterances.], tot_loss[ctc_loss=0.06418, att_loss=0.2312, loss=0.1978, over 3008068.74 frames. utt_duration=1271 frames, utt_pad_proportion=0.04926, over 9476.24 utterances.], batch size: 49, lr: 3.72e-03, grad_scale: 4.0 2023-03-09 12:25:22,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 1.941e+02 2.244e+02 2.736e+02 5.323e+02, threshold=4.488e+02, percent-clipped=4.0 2023-03-09 12:25:25,635 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6732, 5.1763, 5.3183, 5.4105, 5.2408, 5.3431, 5.0493, 4.7626], device='cuda:3'), covar=tensor([0.1890, 0.1007, 0.0430, 0.0508, 0.0731, 0.0425, 0.0545, 0.0456], device='cuda:3'), in_proj_covar=tensor([0.0545, 0.0385, 0.0379, 0.0384, 0.0449, 0.0453, 0.0385, 0.0422], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 12:25:37,190 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112066.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:26:22,046 INFO [train2.py:809] (3/4) Epoch 29, batch 550, loss[ctc_loss=0.06679, att_loss=0.2388, loss=0.2044, over 17071.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007881, over 52.00 utterances.], tot_loss[ctc_loss=0.06392, att_loss=0.2309, loss=0.1975, over 3070723.64 frames. utt_duration=1286 frames, utt_pad_proportion=0.0437, over 9561.45 utterances.], batch size: 52, lr: 3.72e-03, grad_scale: 4.0 2023-03-09 12:27:10,034 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112124.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:27:14,838 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112127.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:27:42,788 INFO [train2.py:809] (3/4) Epoch 29, batch 600, loss[ctc_loss=0.0787, att_loss=0.2522, loss=0.2175, over 17103.00 frames. utt_duration=1223 frames, utt_pad_proportion=0.01595, over 56.00 utterances.], tot_loss[ctc_loss=0.06404, att_loss=0.2303, loss=0.197, over 3115415.90 frames. utt_duration=1280 frames, utt_pad_proportion=0.04629, over 9749.94 utterances.], batch size: 56, lr: 3.72e-03, grad_scale: 4.0 2023-03-09 12:27:45,964 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112147.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 12:28:01,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.016e+02 2.437e+02 2.973e+02 9.680e+02, threshold=4.873e+02, percent-clipped=6.0 2023-03-09 12:28:17,009 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9649, 4.1960, 4.1487, 4.5196, 2.5280, 4.2747, 2.7099, 1.6294], device='cuda:3'), covar=tensor([0.0545, 0.0343, 0.0716, 0.0280, 0.1704, 0.0287, 0.1409, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0192, 0.0269, 0.0182, 0.0226, 0.0174, 0.0234, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:28:21,713 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6013, 2.5802, 2.6297, 2.4128, 2.5845, 2.5292, 2.6447, 1.9522], device='cuda:3'), covar=tensor([0.1038, 0.1417, 0.1884, 0.3368, 0.1307, 0.1844, 0.1376, 0.3472], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0208, 0.0224, 0.0274, 0.0185, 0.0283, 0.0207, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:28:26,108 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112172.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:28:54,093 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-09 12:29:03,069 INFO [train2.py:809] (3/4) Epoch 29, batch 650, loss[ctc_loss=0.05377, att_loss=0.2192, loss=0.1861, over 16164.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007148, over 41.00 utterances.], tot_loss[ctc_loss=0.06413, att_loss=0.2306, loss=0.1973, over 3152069.06 frames. utt_duration=1259 frames, utt_pad_proportion=0.05195, over 10027.94 utterances.], batch size: 41, lr: 3.72e-03, grad_scale: 4.0 2023-03-09 12:30:22,504 INFO [train2.py:809] (3/4) Epoch 29, batch 700, loss[ctc_loss=0.06692, att_loss=0.2144, loss=0.1849, over 15615.00 frames. utt_duration=1690 frames, utt_pad_proportion=0.01018, over 37.00 utterances.], tot_loss[ctc_loss=0.06466, att_loss=0.231, loss=0.1978, over 3179134.69 frames. utt_duration=1250 frames, utt_pad_proportion=0.0534, over 10183.47 utterances.], batch size: 37, lr: 3.71e-03, grad_scale: 4.0 2023-03-09 12:30:41,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 1.769e+02 2.156e+02 2.659e+02 7.287e+02, threshold=4.312e+02, percent-clipped=5.0 2023-03-09 12:31:41,970 INFO [train2.py:809] (3/4) Epoch 29, batch 750, loss[ctc_loss=0.06044, att_loss=0.203, loss=0.1745, over 15487.00 frames. utt_duration=1722 frames, utt_pad_proportion=0.009648, over 36.00 utterances.], tot_loss[ctc_loss=0.06481, att_loss=0.2313, loss=0.198, over 3205552.07 frames. utt_duration=1255 frames, utt_pad_proportion=0.05159, over 10229.67 utterances.], batch size: 36, lr: 3.71e-03, grad_scale: 4.0 2023-03-09 12:32:30,287 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6330, 2.4083, 2.4304, 2.4778, 2.9395, 2.8461, 2.3599, 3.1354], device='cuda:3'), covar=tensor([0.1632, 0.2122, 0.1845, 0.1230, 0.1463, 0.0967, 0.1795, 0.1157], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0150, 0.0147, 0.0141, 0.0158, 0.0136, 0.0158, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 12:33:02,102 INFO [train2.py:809] (3/4) Epoch 29, batch 800, loss[ctc_loss=0.06706, att_loss=0.2221, loss=0.1911, over 15946.00 frames. utt_duration=1557 frames, utt_pad_proportion=0.007508, over 41.00 utterances.], tot_loss[ctc_loss=0.06496, att_loss=0.2311, loss=0.1979, over 3212894.12 frames. utt_duration=1245 frames, utt_pad_proportion=0.05782, over 10338.65 utterances.], batch size: 41, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:33:05,718 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.9844, 4.8955, 4.8088, 2.4092, 1.9394, 2.8345, 2.2269, 3.8607], device='cuda:3'), covar=tensor([0.0745, 0.0289, 0.0272, 0.4393, 0.5357, 0.2510, 0.3842, 0.1488], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0303, 0.0278, 0.0250, 0.0339, 0.0332, 0.0265, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 12:33:21,112 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 1.905e+02 2.158e+02 2.747e+02 4.535e+02, threshold=4.315e+02, percent-clipped=2.0 2023-03-09 12:34:13,342 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112389.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:34:22,116 INFO [train2.py:809] (3/4) Epoch 29, batch 850, loss[ctc_loss=0.06562, att_loss=0.2122, loss=0.1829, over 15760.00 frames. utt_duration=1660 frames, utt_pad_proportion=0.009405, over 38.00 utterances.], tot_loss[ctc_loss=0.06496, att_loss=0.2311, loss=0.1978, over 3227828.12 frames. utt_duration=1252 frames, utt_pad_proportion=0.05596, over 10325.86 utterances.], batch size: 38, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:35:06,131 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112422.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:35:42,814 INFO [train2.py:809] (3/4) Epoch 29, batch 900, loss[ctc_loss=0.05675, att_loss=0.2203, loss=0.1876, over 16187.00 frames. utt_duration=1581 frames, utt_pad_proportion=0.006577, over 41.00 utterances.], tot_loss[ctc_loss=0.06486, att_loss=0.2308, loss=0.1976, over 3237623.72 frames. utt_duration=1259 frames, utt_pad_proportion=0.05487, over 10302.55 utterances.], batch size: 41, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:35:43,880 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 12:35:46,221 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112447.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 12:35:51,290 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112450.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:36:02,469 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.980e+02 2.255e+02 2.971e+02 7.744e+02, threshold=4.510e+02, percent-clipped=6.0 2023-03-09 12:37:03,515 INFO [train2.py:809] (3/4) Epoch 29, batch 950, loss[ctc_loss=0.06586, att_loss=0.247, loss=0.2108, over 16997.00 frames. utt_duration=1334 frames, utt_pad_proportion=0.009345, over 51.00 utterances.], tot_loss[ctc_loss=0.06448, att_loss=0.2307, loss=0.1975, over 3243891.95 frames. utt_duration=1238 frames, utt_pad_proportion=0.05959, over 10495.82 utterances.], batch size: 51, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:37:03,629 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112495.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 12:38:23,730 INFO [train2.py:809] (3/4) Epoch 29, batch 1000, loss[ctc_loss=0.0606, att_loss=0.2334, loss=0.1989, over 16527.00 frames. utt_duration=1470 frames, utt_pad_proportion=0.006426, over 45.00 utterances.], tot_loss[ctc_loss=0.06369, att_loss=0.2308, loss=0.1974, over 3254830.18 frames. utt_duration=1235 frames, utt_pad_proportion=0.05809, over 10557.49 utterances.], batch size: 45, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:38:42,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 1.873e+02 2.288e+02 2.757e+02 5.242e+02, threshold=4.577e+02, percent-clipped=3.0 2023-03-09 12:38:51,781 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112562.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:39:14,981 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112577.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:39:42,799 INFO [train2.py:809] (3/4) Epoch 29, batch 1050, loss[ctc_loss=0.07923, att_loss=0.2377, loss=0.206, over 16673.00 frames. utt_duration=1451 frames, utt_pad_proportion=0.006666, over 46.00 utterances.], tot_loss[ctc_loss=0.06337, att_loss=0.2299, loss=0.1966, over 3258300.25 frames. utt_duration=1251 frames, utt_pad_proportion=0.05465, over 10433.42 utterances.], batch size: 46, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:40:19,293 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112617.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:40:28,661 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112623.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:40:52,280 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112638.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 12:41:03,006 INFO [train2.py:809] (3/4) Epoch 29, batch 1100, loss[ctc_loss=0.04974, att_loss=0.2298, loss=0.1938, over 16886.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.00718, over 49.00 utterances.], tot_loss[ctc_loss=0.06357, att_loss=0.2303, loss=0.197, over 3265209.94 frames. utt_duration=1251 frames, utt_pad_proportion=0.05389, over 10451.97 utterances.], batch size: 49, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:41:22,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 1.863e+02 2.140e+02 2.769e+02 5.111e+02, threshold=4.279e+02, percent-clipped=1.0 2023-03-09 12:41:43,380 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8070, 3.4360, 2.8836, 3.1247, 3.5654, 3.3349, 2.7191, 3.5277], device='cuda:3'), covar=tensor([0.1044, 0.0512, 0.1067, 0.0797, 0.0754, 0.0707, 0.0932, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0232, 0.0234, 0.0212, 0.0293, 0.0252, 0.0208, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:3') 2023-03-09 12:41:55,479 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112678.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 12:42:21,112 INFO [train2.py:809] (3/4) Epoch 29, batch 1150, loss[ctc_loss=0.07063, att_loss=0.2473, loss=0.212, over 16958.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007991, over 50.00 utterances.], tot_loss[ctc_loss=0.06384, att_loss=0.2302, loss=0.1969, over 3260534.42 frames. utt_duration=1250 frames, utt_pad_proportion=0.05556, over 10448.71 utterances.], batch size: 50, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:43:01,937 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9134, 4.7912, 4.9219, 4.8443, 5.3585, 4.7646, 4.8219, 2.4075], device='cuda:3'), covar=tensor([0.0148, 0.0263, 0.0176, 0.0220, 0.0509, 0.0159, 0.0170, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0230, 0.0225, 0.0243, 0.0391, 0.0200, 0.0216, 0.0226], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 12:43:04,799 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112722.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:43:40,476 INFO [train2.py:809] (3/4) Epoch 29, batch 1200, loss[ctc_loss=0.05766, att_loss=0.2347, loss=0.1993, over 17059.00 frames. utt_duration=1289 frames, utt_pad_proportion=0.009311, over 53.00 utterances.], tot_loss[ctc_loss=0.06389, att_loss=0.2304, loss=0.1971, over 3262301.92 frames. utt_duration=1227 frames, utt_pad_proportion=0.06126, over 10650.70 utterances.], batch size: 53, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:43:40,697 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112745.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:43:59,913 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.786e+02 2.116e+02 2.747e+02 6.920e+02, threshold=4.232e+02, percent-clipped=7.0 2023-03-09 12:44:21,250 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112770.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:45:00,866 INFO [train2.py:809] (3/4) Epoch 29, batch 1250, loss[ctc_loss=0.06534, att_loss=0.2337, loss=0.2001, over 16471.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006116, over 46.00 utterances.], tot_loss[ctc_loss=0.06436, att_loss=0.2307, loss=0.1975, over 3263876.90 frames. utt_duration=1235 frames, utt_pad_proportion=0.05851, over 10585.93 utterances.], batch size: 46, lr: 3.71e-03, grad_scale: 8.0 2023-03-09 12:46:12,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-09 12:46:15,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-09 12:46:20,451 INFO [train2.py:809] (3/4) Epoch 29, batch 1300, loss[ctc_loss=0.04027, att_loss=0.2099, loss=0.176, over 16163.00 frames. utt_duration=1578 frames, utt_pad_proportion=0.007486, over 41.00 utterances.], tot_loss[ctc_loss=0.06402, att_loss=0.2313, loss=0.1978, over 3272015.24 frames. utt_duration=1230 frames, utt_pad_proportion=0.0569, over 10649.86 utterances.], batch size: 41, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:46:40,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.906e+02 2.330e+02 2.870e+02 1.155e+03, threshold=4.661e+02, percent-clipped=5.0 2023-03-09 12:47:02,913 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8397, 4.8421, 4.7261, 2.2941, 1.9507, 2.9623, 2.3658, 3.8685], device='cuda:3'), covar=tensor([0.0845, 0.0307, 0.0292, 0.4822, 0.5539, 0.2504, 0.3978, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0310, 0.0284, 0.0257, 0.0348, 0.0339, 0.0270, 0.0376], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 12:47:40,658 INFO [train2.py:809] (3/4) Epoch 29, batch 1350, loss[ctc_loss=0.07115, att_loss=0.2245, loss=0.1938, over 15999.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.008282, over 40.00 utterances.], tot_loss[ctc_loss=0.0646, att_loss=0.2316, loss=0.1982, over 3271497.05 frames. utt_duration=1207 frames, utt_pad_proportion=0.06227, over 10856.26 utterances.], batch size: 40, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:48:17,516 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112918.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:48:22,275 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112921.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:48:41,459 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112933.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 12:49:00,133 INFO [train2.py:809] (3/4) Epoch 29, batch 1400, loss[ctc_loss=0.05916, att_loss=0.2385, loss=0.2026, over 16686.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.006632, over 46.00 utterances.], tot_loss[ctc_loss=0.0644, att_loss=0.2312, loss=0.1979, over 3268233.32 frames. utt_duration=1228 frames, utt_pad_proportion=0.05977, over 10659.96 utterances.], batch size: 46, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:49:19,015 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.843e+02 2.238e+02 2.535e+02 6.230e+02, threshold=4.475e+02, percent-clipped=2.0 2023-03-09 12:49:41,606 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9646, 5.0061, 4.6741, 2.9685, 4.7094, 4.5557, 4.2732, 2.6699], device='cuda:3'), covar=tensor([0.0162, 0.0125, 0.0297, 0.1036, 0.0129, 0.0250, 0.0340, 0.1446], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0106, 0.0112, 0.0113, 0.0091, 0.0120, 0.0102, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 12:49:44,598 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112973.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 12:49:59,375 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112982.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:50:19,501 INFO [train2.py:809] (3/4) Epoch 29, batch 1450, loss[ctc_loss=0.1008, att_loss=0.2593, loss=0.2276, over 14416.00 frames. utt_duration=396.4 frames, utt_pad_proportion=0.307, over 146.00 utterances.], tot_loss[ctc_loss=0.06486, att_loss=0.2318, loss=0.1984, over 3268534.05 frames. utt_duration=1223 frames, utt_pad_proportion=0.05945, over 10704.42 utterances.], batch size: 146, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:51:39,662 INFO [train2.py:809] (3/4) Epoch 29, batch 1500, loss[ctc_loss=0.07312, att_loss=0.2439, loss=0.2097, over 16621.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005439, over 47.00 utterances.], tot_loss[ctc_loss=0.06529, att_loss=0.2316, loss=0.1983, over 3263485.00 frames. utt_duration=1202 frames, utt_pad_proportion=0.06697, over 10869.60 utterances.], batch size: 47, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:51:39,993 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113045.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:51:58,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.764e+02 2.089e+02 2.547e+02 5.820e+02, threshold=4.178e+02, percent-clipped=2.0 2023-03-09 12:52:55,238 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113093.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:52:58,841 INFO [train2.py:809] (3/4) Epoch 29, batch 1550, loss[ctc_loss=0.05008, att_loss=0.2123, loss=0.1799, over 16019.00 frames. utt_duration=1603 frames, utt_pad_proportion=0.006045, over 40.00 utterances.], tot_loss[ctc_loss=0.06499, att_loss=0.2313, loss=0.1981, over 3271565.70 frames. utt_duration=1235 frames, utt_pad_proportion=0.05727, over 10611.89 utterances.], batch size: 40, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:54:19,353 INFO [train2.py:809] (3/4) Epoch 29, batch 1600, loss[ctc_loss=0.06753, att_loss=0.235, loss=0.2015, over 17053.00 frames. utt_duration=1313 frames, utt_pad_proportion=0.008999, over 52.00 utterances.], tot_loss[ctc_loss=0.06505, att_loss=0.2312, loss=0.198, over 3265018.52 frames. utt_duration=1204 frames, utt_pad_proportion=0.06787, over 10858.73 utterances.], batch size: 52, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:54:38,746 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.848e+02 2.168e+02 2.655e+02 4.318e+02, threshold=4.336e+02, percent-clipped=2.0 2023-03-09 12:55:38,884 INFO [train2.py:809] (3/4) Epoch 29, batch 1650, loss[ctc_loss=0.05167, att_loss=0.2077, loss=0.1765, over 15629.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.009057, over 37.00 utterances.], tot_loss[ctc_loss=0.06402, att_loss=0.2303, loss=0.197, over 3258627.75 frames. utt_duration=1244 frames, utt_pad_proportion=0.05877, over 10493.36 utterances.], batch size: 37, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:56:14,960 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113218.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:56:38,816 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113233.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:56:40,409 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113234.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:56:58,265 INFO [train2.py:809] (3/4) Epoch 29, batch 1700, loss[ctc_loss=0.05784, att_loss=0.2113, loss=0.1806, over 15379.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01077, over 35.00 utterances.], tot_loss[ctc_loss=0.06398, att_loss=0.2311, loss=0.1977, over 3264181.11 frames. utt_duration=1239 frames, utt_pad_proportion=0.05963, over 10551.08 utterances.], batch size: 35, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:57:17,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 1.796e+02 2.106e+02 2.556e+02 4.663e+02, threshold=4.212e+02, percent-clipped=1.0 2023-03-09 12:57:31,630 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113266.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:57:43,302 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113273.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 12:57:49,298 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113277.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:57:55,549 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113281.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:58:18,083 INFO [train2.py:809] (3/4) Epoch 29, batch 1750, loss[ctc_loss=0.05016, att_loss=0.2197, loss=0.1858, over 15872.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.01002, over 39.00 utterances.], tot_loss[ctc_loss=0.06429, att_loss=0.2315, loss=0.1981, over 3267825.47 frames. utt_duration=1225 frames, utt_pad_proportion=0.05997, over 10682.31 utterances.], batch size: 39, lr: 3.70e-03, grad_scale: 8.0 2023-03-09 12:58:18,488 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:58:37,284 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1243, 5.4218, 4.9805, 5.4700, 4.8188, 5.0712, 5.5488, 5.3299], device='cuda:3'), covar=tensor([0.0540, 0.0257, 0.0712, 0.0317, 0.0365, 0.0281, 0.0198, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0407, 0.0343, 0.0383, 0.0386, 0.0341, 0.0248, 0.0325, 0.0306], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 12:58:59,095 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113321.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 12:59:06,399 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0 2023-03-09 12:59:37,381 INFO [train2.py:809] (3/4) Epoch 29, batch 1800, loss[ctc_loss=0.06173, att_loss=0.2304, loss=0.1967, over 16125.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006326, over 42.00 utterances.], tot_loss[ctc_loss=0.06399, att_loss=0.2312, loss=0.1977, over 3269613.28 frames. utt_duration=1227 frames, utt_pad_proportion=0.05938, over 10670.71 utterances.], batch size: 42, lr: 3.70e-03, grad_scale: 4.0 2023-03-09 12:59:39,121 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-09 12:59:57,720 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.881e+02 2.263e+02 2.651e+02 5.025e+02, threshold=4.526e+02, percent-clipped=2.0 2023-03-09 13:00:07,795 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-09 13:00:18,147 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7476, 5.1822, 4.2346, 5.2969, 4.6016, 4.9707, 5.1985, 5.0468], device='cuda:3'), covar=tensor([0.0726, 0.0374, 0.1332, 0.0401, 0.0430, 0.0384, 0.0362, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0342, 0.0383, 0.0385, 0.0340, 0.0247, 0.0324, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 13:00:37,692 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0680, 3.7850, 3.7663, 3.2074, 3.7651, 3.9003, 3.8075, 2.7830], device='cuda:3'), covar=tensor([0.1102, 0.1130, 0.1519, 0.2944, 0.1108, 0.1828, 0.0814, 0.3097], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0210, 0.0227, 0.0277, 0.0186, 0.0288, 0.0209, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:00:55,303 INFO [train2.py:809] (3/4) Epoch 29, batch 1850, loss[ctc_loss=0.05684, att_loss=0.2152, loss=0.1835, over 14618.00 frames. utt_duration=1829 frames, utt_pad_proportion=0.03652, over 32.00 utterances.], tot_loss[ctc_loss=0.06447, att_loss=0.2313, loss=0.1979, over 3261069.68 frames. utt_duration=1227 frames, utt_pad_proportion=0.06172, over 10642.49 utterances.], batch size: 32, lr: 3.70e-03, grad_scale: 4.0 2023-03-09 13:01:43,052 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0070, 6.2022, 5.7054, 5.9171, 5.9050, 5.3740, 5.7237, 5.4580], device='cuda:3'), covar=tensor([0.1088, 0.0919, 0.0931, 0.0852, 0.0892, 0.1487, 0.2124, 0.2074], device='cuda:3'), in_proj_covar=tensor([0.0567, 0.0643, 0.0494, 0.0476, 0.0458, 0.0485, 0.0646, 0.0547], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 13:02:15,435 INFO [train2.py:809] (3/4) Epoch 29, batch 1900, loss[ctc_loss=0.05417, att_loss=0.1976, loss=0.169, over 12779.00 frames. utt_duration=1827 frames, utt_pad_proportion=0.1182, over 28.00 utterances.], tot_loss[ctc_loss=0.06423, att_loss=0.2313, loss=0.1979, over 3265684.74 frames. utt_duration=1243 frames, utt_pad_proportion=0.05693, over 10521.71 utterances.], batch size: 28, lr: 3.70e-03, grad_scale: 4.0 2023-03-09 13:02:35,443 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.919e+02 2.347e+02 2.814e+02 3.658e+03, threshold=4.695e+02, percent-clipped=4.0 2023-03-09 13:03:34,135 INFO [train2.py:809] (3/4) Epoch 29, batch 1950, loss[ctc_loss=0.05772, att_loss=0.231, loss=0.1964, over 16554.00 frames. utt_duration=1473 frames, utt_pad_proportion=0.005597, over 45.00 utterances.], tot_loss[ctc_loss=0.06374, att_loss=0.231, loss=0.1975, over 3266679.12 frames. utt_duration=1257 frames, utt_pad_proportion=0.05174, over 10408.76 utterances.], batch size: 45, lr: 3.69e-03, grad_scale: 4.0 2023-03-09 13:04:41,257 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.6633, 2.4998, 2.3540, 2.5951, 2.8098, 2.7883, 2.5045, 3.0023], device='cuda:3'), covar=tensor([0.2234, 0.2209, 0.1841, 0.1589, 0.1471, 0.1193, 0.1995, 0.1141], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0147, 0.0145, 0.0140, 0.0156, 0.0134, 0.0157, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 13:04:54,514 INFO [train2.py:809] (3/4) Epoch 29, batch 2000, loss[ctc_loss=0.06008, att_loss=0.2491, loss=0.2113, over 17316.00 frames. utt_duration=1175 frames, utt_pad_proportion=0.02373, over 59.00 utterances.], tot_loss[ctc_loss=0.06348, att_loss=0.2308, loss=0.1973, over 3269945.31 frames. utt_duration=1275 frames, utt_pad_proportion=0.04627, over 10270.19 utterances.], batch size: 59, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:05:15,511 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.751e+02 2.072e+02 2.530e+02 5.087e+02, threshold=4.144e+02, percent-clipped=1.0 2023-03-09 13:05:45,783 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113577.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:05:53,361 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8925, 5.1773, 4.7851, 5.2247, 4.5961, 4.8396, 5.2953, 5.1016], device='cuda:3'), covar=tensor([0.0552, 0.0303, 0.0764, 0.0338, 0.0414, 0.0325, 0.0212, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0339, 0.0379, 0.0381, 0.0337, 0.0244, 0.0322, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 13:06:06,340 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113590.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:06:14,833 INFO [train2.py:809] (3/4) Epoch 29, batch 2050, loss[ctc_loss=0.05516, att_loss=0.2152, loss=0.1832, over 14499.00 frames. utt_duration=1814 frames, utt_pad_proportion=0.03213, over 32.00 utterances.], tot_loss[ctc_loss=0.06313, att_loss=0.2306, loss=0.1971, over 3265075.48 frames. utt_duration=1269 frames, utt_pad_proportion=0.05025, over 10301.06 utterances.], batch size: 32, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:06:49,212 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113617.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:07:00,878 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2721, 4.2307, 4.2621, 4.3004, 4.7948, 4.1988, 4.1926, 2.5321], device='cuda:3'), covar=tensor([0.0375, 0.0511, 0.0469, 0.0438, 0.0660, 0.0338, 0.0405, 0.1666], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0227, 0.0223, 0.0238, 0.0386, 0.0197, 0.0215, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:07:02,080 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113625.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:07:35,391 INFO [train2.py:809] (3/4) Epoch 29, batch 2100, loss[ctc_loss=0.07159, att_loss=0.2141, loss=0.1856, over 15744.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.01062, over 38.00 utterances.], tot_loss[ctc_loss=0.06352, att_loss=0.2306, loss=0.1972, over 3262332.63 frames. utt_duration=1241 frames, utt_pad_proportion=0.0587, over 10526.39 utterances.], batch size: 38, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:07:37,286 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113646.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:07:55,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 1.850e+02 2.211e+02 2.646e+02 5.386e+02, threshold=4.422e+02, percent-clipped=4.0 2023-03-09 13:08:03,845 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-09 13:08:26,925 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113678.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:08:36,646 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9151, 5.2345, 5.4270, 5.2458, 5.4127, 5.8658, 5.1469, 5.9706], device='cuda:3'), covar=tensor([0.0742, 0.0831, 0.0857, 0.1447, 0.1697, 0.0886, 0.0826, 0.0695], device='cuda:3'), in_proj_covar=tensor([0.0922, 0.0537, 0.0645, 0.0684, 0.0913, 0.0669, 0.0517, 0.0640], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:08:55,295 INFO [train2.py:809] (3/4) Epoch 29, batch 2150, loss[ctc_loss=0.0488, att_loss=0.2172, loss=0.1835, over 15959.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006778, over 41.00 utterances.], tot_loss[ctc_loss=0.0639, att_loss=0.2309, loss=0.1975, over 3261031.96 frames. utt_duration=1259 frames, utt_pad_proportion=0.0549, over 10375.76 utterances.], batch size: 41, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:09:13,936 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113707.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:09:33,052 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-03-09 13:09:46,723 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-09 13:10:13,176 INFO [train2.py:809] (3/4) Epoch 29, batch 2200, loss[ctc_loss=0.06474, att_loss=0.2346, loss=0.2006, over 16952.00 frames. utt_duration=1358 frames, utt_pad_proportion=0.007588, over 50.00 utterances.], tot_loss[ctc_loss=0.06461, att_loss=0.2308, loss=0.1976, over 3244620.93 frames. utt_duration=1220 frames, utt_pad_proportion=0.0683, over 10648.41 utterances.], batch size: 50, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:10:32,606 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 1.987e+02 2.347e+02 2.821e+02 4.899e+02, threshold=4.694e+02, percent-clipped=4.0 2023-03-09 13:11:07,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-09 13:11:07,913 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.4300, 4.7397, 5.1835, 4.6831, 4.6752, 5.2670, 4.8310, 5.2092], device='cuda:3'), covar=tensor([0.1347, 0.1775, 0.1364, 0.2456, 0.3317, 0.1768, 0.1505, 0.1603], device='cuda:3'), in_proj_covar=tensor([0.0922, 0.0537, 0.0646, 0.0682, 0.0913, 0.0668, 0.0516, 0.0641], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:11:32,258 INFO [train2.py:809] (3/4) Epoch 29, batch 2250, loss[ctc_loss=0.07545, att_loss=0.2503, loss=0.2153, over 16864.00 frames. utt_duration=1378 frames, utt_pad_proportion=0.007655, over 49.00 utterances.], tot_loss[ctc_loss=0.06517, att_loss=0.2316, loss=0.1983, over 3245036.87 frames. utt_duration=1196 frames, utt_pad_proportion=0.07176, over 10862.60 utterances.], batch size: 49, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:12:13,557 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-09 13:12:23,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-09 13:12:51,817 INFO [train2.py:809] (3/4) Epoch 29, batch 2300, loss[ctc_loss=0.04159, att_loss=0.2008, loss=0.169, over 15495.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008567, over 36.00 utterances.], tot_loss[ctc_loss=0.06434, att_loss=0.231, loss=0.1977, over 3258730.62 frames. utt_duration=1222 frames, utt_pad_proportion=0.06252, over 10683.29 utterances.], batch size: 36, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:13:12,812 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.846e+02 2.164e+02 2.611e+02 9.969e+02, threshold=4.329e+02, percent-clipped=2.0 2023-03-09 13:14:03,910 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113890.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:14:11,233 INFO [train2.py:809] (3/4) Epoch 29, batch 2350, loss[ctc_loss=0.06843, att_loss=0.2381, loss=0.2042, over 17292.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01165, over 55.00 utterances.], tot_loss[ctc_loss=0.06435, att_loss=0.2315, loss=0.1981, over 3255570.15 frames. utt_duration=1200 frames, utt_pad_proportion=0.06893, over 10868.99 utterances.], batch size: 55, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:14:56,168 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-09 13:15:19,207 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113938.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:15:29,857 INFO [train2.py:809] (3/4) Epoch 29, batch 2400, loss[ctc_loss=0.05799, att_loss=0.237, loss=0.2012, over 17353.00 frames. utt_duration=1178 frames, utt_pad_proportion=0.02161, over 59.00 utterances.], tot_loss[ctc_loss=0.06454, att_loss=0.2317, loss=0.1982, over 3269687.83 frames. utt_duration=1221 frames, utt_pad_proportion=0.06018, over 10726.84 utterances.], batch size: 59, lr: 3.69e-03, grad_scale: 8.0 2023-03-09 13:15:50,422 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 1.796e+02 2.124e+02 2.708e+02 5.253e+02, threshold=4.249e+02, percent-clipped=4.0 2023-03-09 13:15:54,332 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-09 13:16:14,214 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113973.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:16:27,485 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1395, 4.4040, 4.4962, 4.6926, 2.8984, 4.5714, 2.9746, 1.8950], device='cuda:3'), covar=tensor([0.0470, 0.0317, 0.0578, 0.0218, 0.1404, 0.0204, 0.1268, 0.1538], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0192, 0.0265, 0.0181, 0.0222, 0.0172, 0.0233, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:16:49,313 INFO [train2.py:809] (3/4) Epoch 29, batch 2450, loss[ctc_loss=0.06001, att_loss=0.2212, loss=0.189, over 14080.00 frames. utt_duration=1818 frames, utt_pad_proportion=0.05202, over 31.00 utterances.], tot_loss[ctc_loss=0.06416, att_loss=0.2314, loss=0.198, over 3274671.27 frames. utt_duration=1242 frames, utt_pad_proportion=0.05333, over 10562.71 utterances.], batch size: 31, lr: 3.69e-03, grad_scale: 4.0 2023-03-09 13:17:04,654 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114002.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:18:12,392 INFO [train2.py:809] (3/4) Epoch 29, batch 2500, loss[ctc_loss=0.09223, att_loss=0.2428, loss=0.2127, over 16460.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.006347, over 46.00 utterances.], tot_loss[ctc_loss=0.06416, att_loss=0.2314, loss=0.1979, over 3259272.88 frames. utt_duration=1214 frames, utt_pad_proportion=0.06362, over 10752.05 utterances.], batch size: 46, lr: 3.69e-03, grad_scale: 4.0 2023-03-09 13:18:34,823 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 1.923e+02 2.146e+02 2.615e+02 3.828e+02, threshold=4.291e+02, percent-clipped=0.0 2023-03-09 13:19:32,907 INFO [train2.py:809] (3/4) Epoch 29, batch 2550, loss[ctc_loss=0.05542, att_loss=0.2253, loss=0.1913, over 17479.00 frames. utt_duration=1015 frames, utt_pad_proportion=0.04366, over 69.00 utterances.], tot_loss[ctc_loss=0.06388, att_loss=0.2309, loss=0.1975, over 3266773.56 frames. utt_duration=1227 frames, utt_pad_proportion=0.05917, over 10665.52 utterances.], batch size: 69, lr: 3.68e-03, grad_scale: 4.0 2023-03-09 13:20:12,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2023-03-09 13:20:23,979 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114127.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:20:52,079 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-09 13:20:52,674 INFO [train2.py:809] (3/4) Epoch 29, batch 2600, loss[ctc_loss=0.06084, att_loss=0.2336, loss=0.199, over 16773.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.005978, over 48.00 utterances.], tot_loss[ctc_loss=0.06341, att_loss=0.2312, loss=0.1976, over 3279169.49 frames. utt_duration=1239 frames, utt_pad_proportion=0.05252, over 10600.08 utterances.], batch size: 48, lr: 3.68e-03, grad_scale: 4.0 2023-03-09 13:20:56,014 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2559, 2.9508, 3.1390, 4.3220, 3.8657, 3.8468, 2.8548, 2.3019], device='cuda:3'), covar=tensor([0.0850, 0.1828, 0.1017, 0.0545, 0.0916, 0.0538, 0.1543, 0.2146], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0222, 0.0187, 0.0231, 0.0238, 0.0195, 0.0207, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:21:15,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.450e+01 1.839e+02 2.178e+02 2.659e+02 5.439e+02, threshold=4.356e+02, percent-clipped=3.0 2023-03-09 13:21:32,018 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2538, 4.5027, 4.8665, 4.8059, 3.0139, 4.7236, 3.2300, 2.1714], device='cuda:3'), covar=tensor([0.0474, 0.0331, 0.0495, 0.0240, 0.1335, 0.0205, 0.1122, 0.1493], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0192, 0.0264, 0.0181, 0.0223, 0.0172, 0.0232, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:22:01,517 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114188.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:22:11,801 INFO [train2.py:809] (3/4) Epoch 29, batch 2650, loss[ctc_loss=0.07935, att_loss=0.2418, loss=0.2093, over 17072.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007823, over 52.00 utterances.], tot_loss[ctc_loss=0.06358, att_loss=0.2307, loss=0.1972, over 3270564.50 frames. utt_duration=1244 frames, utt_pad_proportion=0.05373, over 10530.50 utterances.], batch size: 52, lr: 3.68e-03, grad_scale: 4.0 2023-03-09 13:23:27,968 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-09 13:23:31,713 INFO [train2.py:809] (3/4) Epoch 29, batch 2700, loss[ctc_loss=0.05177, att_loss=0.2323, loss=0.1962, over 16977.00 frames. utt_duration=1360 frames, utt_pad_proportion=0.006764, over 50.00 utterances.], tot_loss[ctc_loss=0.06411, att_loss=0.2314, loss=0.1979, over 3276278.37 frames. utt_duration=1228 frames, utt_pad_proportion=0.05576, over 10681.76 utterances.], batch size: 50, lr: 3.68e-03, grad_scale: 4.0 2023-03-09 13:23:53,985 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 1.816e+02 2.102e+02 2.530e+02 4.306e+02, threshold=4.205e+02, percent-clipped=0.0 2023-03-09 13:24:16,613 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114273.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:24:16,685 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2108, 3.8611, 3.8775, 3.3349, 3.8208, 4.0041, 3.8924, 2.9500], device='cuda:3'), covar=tensor([0.1049, 0.1024, 0.1585, 0.2827, 0.1611, 0.1142, 0.1427, 0.2683], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0209, 0.0225, 0.0275, 0.0186, 0.0286, 0.0208, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:24:22,903 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 13:24:51,548 INFO [train2.py:809] (3/4) Epoch 29, batch 2750, loss[ctc_loss=0.0454, att_loss=0.1998, loss=0.1689, over 15660.00 frames. utt_duration=1695 frames, utt_pad_proportion=0.007722, over 37.00 utterances.], tot_loss[ctc_loss=0.06358, att_loss=0.2309, loss=0.1975, over 3270198.71 frames. utt_duration=1241 frames, utt_pad_proportion=0.05563, over 10552.93 utterances.], batch size: 37, lr: 3.68e-03, grad_scale: 4.0 2023-03-09 13:25:03,012 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114302.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:25:32,586 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114321.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:26:11,320 INFO [train2.py:809] (3/4) Epoch 29, batch 2800, loss[ctc_loss=0.04911, att_loss=0.2247, loss=0.1895, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006207, over 46.00 utterances.], tot_loss[ctc_loss=0.06359, att_loss=0.2314, loss=0.1979, over 3275164.83 frames. utt_duration=1234 frames, utt_pad_proportion=0.05661, over 10626.81 utterances.], batch size: 46, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:26:19,151 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:26:33,215 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.822e+02 2.190e+02 2.901e+02 6.134e+02, threshold=4.379e+02, percent-clipped=9.0 2023-03-09 13:27:30,407 INFO [train2.py:809] (3/4) Epoch 29, batch 2850, loss[ctc_loss=0.0886, att_loss=0.2594, loss=0.2252, over 17339.00 frames. utt_duration=1102 frames, utt_pad_proportion=0.03644, over 63.00 utterances.], tot_loss[ctc_loss=0.06437, att_loss=0.232, loss=0.1985, over 3283201.36 frames. utt_duration=1244 frames, utt_pad_proportion=0.05243, over 10567.94 utterances.], batch size: 63, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:28:50,608 INFO [train2.py:809] (3/4) Epoch 29, batch 2900, loss[ctc_loss=0.04852, att_loss=0.2092, loss=0.1771, over 14578.00 frames. utt_duration=1824 frames, utt_pad_proportion=0.03851, over 32.00 utterances.], tot_loss[ctc_loss=0.06339, att_loss=0.231, loss=0.1975, over 3282730.28 frames. utt_duration=1265 frames, utt_pad_proportion=0.0469, over 10396.58 utterances.], batch size: 32, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:29:08,622 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2547, 2.9201, 3.5395, 2.9418, 3.4846, 4.4434, 4.2585, 3.0057], device='cuda:3'), covar=tensor([0.0399, 0.1654, 0.1249, 0.1313, 0.1018, 0.0775, 0.0576, 0.1348], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0250, 0.0293, 0.0220, 0.0273, 0.0383, 0.0275, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:29:12,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 1.896e+02 2.215e+02 2.552e+02 6.599e+02, threshold=4.430e+02, percent-clipped=4.0 2023-03-09 13:29:52,212 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114483.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:30:10,932 INFO [train2.py:809] (3/4) Epoch 29, batch 2950, loss[ctc_loss=0.04134, att_loss=0.2187, loss=0.1832, over 16540.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006107, over 45.00 utterances.], tot_loss[ctc_loss=0.06337, att_loss=0.2308, loss=0.1973, over 3279007.44 frames. utt_duration=1271 frames, utt_pad_proportion=0.04613, over 10331.27 utterances.], batch size: 45, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:31:30,959 INFO [train2.py:809] (3/4) Epoch 29, batch 3000, loss[ctc_loss=0.05672, att_loss=0.2371, loss=0.201, over 17306.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01161, over 55.00 utterances.], tot_loss[ctc_loss=0.06405, att_loss=0.2315, loss=0.198, over 3276790.02 frames. utt_duration=1261 frames, utt_pad_proportion=0.05075, over 10410.10 utterances.], batch size: 55, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:31:30,959 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 13:31:44,954 INFO [train2.py:843] (3/4) Epoch 29, validation: ctc_loss=0.04115, att_loss=0.2348, loss=0.1961, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 13:31:44,955 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 13:32:06,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 1.877e+02 2.271e+02 2.617e+02 4.276e+02, threshold=4.543e+02, percent-clipped=0.0 2023-03-09 13:33:04,605 INFO [train2.py:809] (3/4) Epoch 29, batch 3050, loss[ctc_loss=0.05373, att_loss=0.2091, loss=0.178, over 15872.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.01, over 39.00 utterances.], tot_loss[ctc_loss=0.06484, att_loss=0.2321, loss=0.1987, over 3275896.08 frames. utt_duration=1249 frames, utt_pad_proportion=0.05401, over 10507.93 utterances.], batch size: 39, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:33:31,719 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1331, 4.3401, 4.2490, 4.5545, 2.8740, 4.4431, 2.7914, 1.7188], device='cuda:3'), covar=tensor([0.0552, 0.0314, 0.0735, 0.0288, 0.1541, 0.0274, 0.1420, 0.1745], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0191, 0.0264, 0.0181, 0.0222, 0.0173, 0.0232, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:34:17,259 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-09 13:34:18,152 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1319, 5.4649, 5.6215, 5.5091, 5.6510, 6.0709, 5.3491, 6.1567], device='cuda:3'), covar=tensor([0.0722, 0.0719, 0.0872, 0.1286, 0.1796, 0.0832, 0.0751, 0.0686], device='cuda:3'), in_proj_covar=tensor([0.0913, 0.0530, 0.0641, 0.0675, 0.0905, 0.0663, 0.0510, 0.0636], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:34:24,749 INFO [train2.py:809] (3/4) Epoch 29, batch 3100, loss[ctc_loss=0.1167, att_loss=0.2583, loss=0.23, over 14159.00 frames. utt_duration=389.5 frames, utt_pad_proportion=0.3215, over 146.00 utterances.], tot_loss[ctc_loss=0.0646, att_loss=0.2321, loss=0.1986, over 3277288.91 frames. utt_duration=1234 frames, utt_pad_proportion=0.05772, over 10635.40 utterances.], batch size: 146, lr: 3.68e-03, grad_scale: 8.0 2023-03-09 13:34:46,590 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 1.785e+02 2.252e+02 2.835e+02 6.156e+02, threshold=4.505e+02, percent-clipped=2.0 2023-03-09 13:35:04,584 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0194, 5.1957, 5.2250, 5.1948, 5.2700, 5.2595, 4.8486, 4.6676], device='cuda:3'), covar=tensor([0.1035, 0.0644, 0.0355, 0.0558, 0.0321, 0.0326, 0.0484, 0.0388], device='cuda:3'), in_proj_covar=tensor([0.0535, 0.0383, 0.0376, 0.0382, 0.0445, 0.0452, 0.0382, 0.0421], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 13:35:43,655 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2583, 3.0026, 3.5572, 3.0152, 3.4079, 4.3630, 4.1927, 3.1152], device='cuda:3'), covar=tensor([0.0383, 0.1534, 0.1222, 0.1297, 0.1162, 0.0858, 0.0612, 0.1314], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0250, 0.0292, 0.0220, 0.0273, 0.0383, 0.0274, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:35:44,840 INFO [train2.py:809] (3/4) Epoch 29, batch 3150, loss[ctc_loss=0.04759, att_loss=0.2272, loss=0.1913, over 17419.00 frames. utt_duration=1011 frames, utt_pad_proportion=0.04605, over 69.00 utterances.], tot_loss[ctc_loss=0.06478, att_loss=0.2325, loss=0.199, over 3283987.23 frames. utt_duration=1220 frames, utt_pad_proportion=0.05898, over 10783.85 utterances.], batch size: 69, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:35:47,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-09 13:36:18,451 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114716.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:36:34,215 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2920, 2.7173, 2.7446, 4.3139, 4.0438, 4.0039, 3.0231, 2.1446], device='cuda:3'), covar=tensor([0.0751, 0.2140, 0.1380, 0.0591, 0.0715, 0.0417, 0.1335, 0.2272], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0222, 0.0189, 0.0233, 0.0239, 0.0196, 0.0209, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-09 13:37:04,184 INFO [train2.py:809] (3/4) Epoch 29, batch 3200, loss[ctc_loss=0.05323, att_loss=0.2383, loss=0.2012, over 16746.00 frames. utt_duration=1397 frames, utt_pad_proportion=0.007768, over 48.00 utterances.], tot_loss[ctc_loss=0.06413, att_loss=0.2318, loss=0.1983, over 3281279.13 frames. utt_duration=1238 frames, utt_pad_proportion=0.05478, over 10618.00 utterances.], batch size: 48, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:37:27,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.814e+02 2.180e+02 2.558e+02 5.727e+02, threshold=4.361e+02, percent-clipped=1.0 2023-03-09 13:37:54,518 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114777.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:38:03,535 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 13:38:04,291 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114783.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:38:07,616 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1408, 5.1131, 4.9500, 2.2477, 1.9798, 2.9468, 2.4982, 3.9742], device='cuda:3'), covar=tensor([0.0698, 0.0329, 0.0240, 0.5079, 0.5582, 0.2474, 0.3685, 0.1472], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0309, 0.0282, 0.0257, 0.0343, 0.0335, 0.0267, 0.0375], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 13:38:22,855 INFO [train2.py:809] (3/4) Epoch 29, batch 3250, loss[ctc_loss=0.05531, att_loss=0.2289, loss=0.1941, over 16830.00 frames. utt_duration=688.6 frames, utt_pad_proportion=0.1371, over 98.00 utterances.], tot_loss[ctc_loss=0.06359, att_loss=0.2313, loss=0.1978, over 3280071.92 frames. utt_duration=1242 frames, utt_pad_proportion=0.05305, over 10574.86 utterances.], batch size: 98, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:39:15,459 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9480, 5.3529, 5.4995, 5.2745, 5.4593, 5.9705, 5.2833, 6.0451], device='cuda:3'), covar=tensor([0.0748, 0.0782, 0.0936, 0.1400, 0.1838, 0.0818, 0.0771, 0.0632], device='cuda:3'), in_proj_covar=tensor([0.0913, 0.0530, 0.0640, 0.0677, 0.0906, 0.0661, 0.0511, 0.0637], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:39:20,564 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114831.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:39:20,617 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7252, 5.9635, 5.4605, 5.7179, 5.6549, 5.1599, 5.3834, 5.0991], device='cuda:3'), covar=tensor([0.1202, 0.0853, 0.0961, 0.0837, 0.1015, 0.1496, 0.2400, 0.2496], device='cuda:3'), in_proj_covar=tensor([0.0564, 0.0635, 0.0491, 0.0473, 0.0455, 0.0483, 0.0637, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 13:39:42,518 INFO [train2.py:809] (3/4) Epoch 29, batch 3300, loss[ctc_loss=0.06169, att_loss=0.2319, loss=0.1978, over 16891.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006959, over 49.00 utterances.], tot_loss[ctc_loss=0.06365, att_loss=0.2307, loss=0.1973, over 3271307.81 frames. utt_duration=1244 frames, utt_pad_proportion=0.05523, over 10535.21 utterances.], batch size: 49, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:40:05,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 1.794e+02 2.073e+02 2.411e+02 5.454e+02, threshold=4.146e+02, percent-clipped=1.0 2023-03-09 13:40:23,014 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114871.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:41:01,522 INFO [train2.py:809] (3/4) Epoch 29, batch 3350, loss[ctc_loss=0.06073, att_loss=0.2392, loss=0.2035, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008181, over 51.00 utterances.], tot_loss[ctc_loss=0.06358, att_loss=0.2304, loss=0.197, over 3272566.37 frames. utt_duration=1248 frames, utt_pad_proportion=0.05349, over 10502.95 utterances.], batch size: 51, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:42:00,790 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114932.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:42:20,954 INFO [train2.py:809] (3/4) Epoch 29, batch 3400, loss[ctc_loss=0.04728, att_loss=0.2179, loss=0.1838, over 16499.00 frames. utt_duration=1436 frames, utt_pad_proportion=0.005525, over 46.00 utterances.], tot_loss[ctc_loss=0.06349, att_loss=0.2304, loss=0.197, over 3262036.63 frames. utt_duration=1240 frames, utt_pad_proportion=0.0574, over 10532.84 utterances.], batch size: 46, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:42:29,048 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1741, 4.4325, 4.5182, 4.6738, 2.8614, 4.6334, 2.9366, 1.6902], device='cuda:3'), covar=tensor([0.0520, 0.0293, 0.0623, 0.0269, 0.1587, 0.0266, 0.1349, 0.1798], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0193, 0.0266, 0.0183, 0.0224, 0.0174, 0.0235, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:42:35,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-09 13:42:44,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.858e+02 2.229e+02 2.688e+02 7.167e+02, threshold=4.457e+02, percent-clipped=6.0 2023-03-09 13:43:41,692 INFO [train2.py:809] (3/4) Epoch 29, batch 3450, loss[ctc_loss=0.05366, att_loss=0.2406, loss=0.2032, over 16752.00 frames. utt_duration=1397 frames, utt_pad_proportion=0.007502, over 48.00 utterances.], tot_loss[ctc_loss=0.06366, att_loss=0.2309, loss=0.1974, over 3261357.55 frames. utt_duration=1211 frames, utt_pad_proportion=0.0661, over 10785.05 utterances.], batch size: 48, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:45:02,580 INFO [train2.py:809] (3/4) Epoch 29, batch 3500, loss[ctc_loss=0.07796, att_loss=0.2567, loss=0.221, over 16967.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.006681, over 50.00 utterances.], tot_loss[ctc_loss=0.06351, att_loss=0.2307, loss=0.1973, over 3259531.15 frames. utt_duration=1211 frames, utt_pad_proportion=0.06719, over 10780.35 utterances.], batch size: 50, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:45:26,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 1.733e+02 2.269e+02 2.703e+02 9.473e+02, threshold=4.539e+02, percent-clipped=4.0 2023-03-09 13:45:46,152 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115072.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:46:09,454 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115086.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:46:22,696 INFO [train2.py:809] (3/4) Epoch 29, batch 3550, loss[ctc_loss=0.06726, att_loss=0.21, loss=0.1815, over 15359.00 frames. utt_duration=1757 frames, utt_pad_proportion=0.01199, over 35.00 utterances.], tot_loss[ctc_loss=0.06379, att_loss=0.2306, loss=0.1972, over 3259486.40 frames. utt_duration=1189 frames, utt_pad_proportion=0.07266, over 10983.53 utterances.], batch size: 35, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:47:42,041 INFO [train2.py:809] (3/4) Epoch 29, batch 3600, loss[ctc_loss=0.052, att_loss=0.2293, loss=0.1938, over 16620.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.004844, over 47.00 utterances.], tot_loss[ctc_loss=0.06387, att_loss=0.2308, loss=0.1974, over 3264568.77 frames. utt_duration=1199 frames, utt_pad_proportion=0.06872, over 10904.16 utterances.], batch size: 47, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:47:45,627 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115147.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:48:05,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 1.777e+02 2.035e+02 2.403e+02 5.227e+02, threshold=4.069e+02, percent-clipped=1.0 2023-03-09 13:49:02,357 INFO [train2.py:809] (3/4) Epoch 29, batch 3650, loss[ctc_loss=0.06378, att_loss=0.2336, loss=0.1996, over 17013.00 frames. utt_duration=1336 frames, utt_pad_proportion=0.008108, over 51.00 utterances.], tot_loss[ctc_loss=0.064, att_loss=0.2305, loss=0.1972, over 3264961.54 frames. utt_duration=1210 frames, utt_pad_proportion=0.06613, over 10802.42 utterances.], batch size: 51, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:49:51,382 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6762, 4.8218, 4.7881, 4.8342, 4.9021, 4.8664, 4.5646, 4.4134], device='cuda:3'), covar=tensor([0.0920, 0.0650, 0.0468, 0.0536, 0.0296, 0.0373, 0.0435, 0.0354], device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0384, 0.0379, 0.0385, 0.0449, 0.0454, 0.0384, 0.0423], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 13:49:54,458 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115227.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:50:03,647 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.92 vs. limit=5.0 2023-03-09 13:50:23,432 INFO [train2.py:809] (3/4) Epoch 29, batch 3700, loss[ctc_loss=0.05602, att_loss=0.2269, loss=0.1927, over 16168.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.007588, over 41.00 utterances.], tot_loss[ctc_loss=0.06357, att_loss=0.2301, loss=0.1968, over 3266071.18 frames. utt_duration=1228 frames, utt_pad_proportion=0.0614, over 10656.03 utterances.], batch size: 41, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:50:47,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 1.820e+02 2.082e+02 2.598e+02 4.716e+02, threshold=4.165e+02, percent-clipped=2.0 2023-03-09 13:51:05,948 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.6548, 5.9466, 5.2815, 5.6126, 5.5762, 5.0365, 5.4063, 4.9681], device='cuda:3'), covar=tensor([0.1366, 0.0952, 0.1132, 0.0966, 0.0992, 0.1855, 0.2483, 0.2415], device='cuda:3'), in_proj_covar=tensor([0.0568, 0.0641, 0.0497, 0.0479, 0.0458, 0.0489, 0.0645, 0.0551], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 13:51:44,492 INFO [train2.py:809] (3/4) Epoch 29, batch 3750, loss[ctc_loss=0.04818, att_loss=0.1993, loss=0.1691, over 15376.00 frames. utt_duration=1759 frames, utt_pad_proportion=0.01094, over 35.00 utterances.], tot_loss[ctc_loss=0.0633, att_loss=0.2303, loss=0.1969, over 3267093.85 frames. utt_duration=1222 frames, utt_pad_proportion=0.06318, over 10710.44 utterances.], batch size: 35, lr: 3.67e-03, grad_scale: 8.0 2023-03-09 13:52:11,595 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3666, 3.9197, 3.3410, 3.5323, 4.1300, 3.8032, 3.1818, 4.3839], device='cuda:3'), covar=tensor([0.0940, 0.0473, 0.1142, 0.0734, 0.0647, 0.0802, 0.0852, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0230, 0.0233, 0.0210, 0.0292, 0.0252, 0.0207, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:3') 2023-03-09 13:53:04,020 INFO [train2.py:809] (3/4) Epoch 29, batch 3800, loss[ctc_loss=0.04697, att_loss=0.2035, loss=0.1722, over 15499.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008423, over 36.00 utterances.], tot_loss[ctc_loss=0.06316, att_loss=0.2305, loss=0.197, over 3267793.89 frames. utt_duration=1228 frames, utt_pad_proportion=0.06172, over 10656.13 utterances.], batch size: 36, lr: 3.66e-03, grad_scale: 8.0 2023-03-09 13:53:08,379 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-03-09 13:53:14,758 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-09 13:53:27,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.885e+02 2.239e+02 2.673e+02 5.789e+02, threshold=4.479e+02, percent-clipped=6.0 2023-03-09 13:53:43,980 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115370.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:53:46,190 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115371.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:53:47,650 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115372.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:54:23,179 INFO [train2.py:809] (3/4) Epoch 29, batch 3850, loss[ctc_loss=0.06852, att_loss=0.2314, loss=0.1988, over 16634.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.00505, over 47.00 utterances.], tot_loss[ctc_loss=0.06432, att_loss=0.231, loss=0.1977, over 3264451.10 frames. utt_duration=1186 frames, utt_pad_proportion=0.07345, over 11019.58 utterances.], batch size: 47, lr: 3.66e-03, grad_scale: 8.0 2023-03-09 13:54:39,223 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.2486, 4.3455, 4.4563, 4.4079, 4.9553, 4.3160, 4.4152, 2.6220], device='cuda:3'), covar=tensor([0.0379, 0.0465, 0.0400, 0.0418, 0.0716, 0.0333, 0.0421, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0231, 0.0225, 0.0243, 0.0386, 0.0199, 0.0219, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:55:01,821 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115420.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:55:18,434 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115431.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:55:20,026 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115432.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:55:34,950 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115442.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:55:39,327 INFO [train2.py:809] (3/4) Epoch 29, batch 3900, loss[ctc_loss=0.05708, att_loss=0.212, loss=0.181, over 15773.00 frames. utt_duration=1662 frames, utt_pad_proportion=0.00862, over 38.00 utterances.], tot_loss[ctc_loss=0.06406, att_loss=0.2307, loss=0.1974, over 3267510.42 frames. utt_duration=1194 frames, utt_pad_proportion=0.06943, over 10964.15 utterances.], batch size: 38, lr: 3.66e-03, grad_scale: 8.0 2023-03-09 13:56:02,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.900e+02 2.175e+02 2.636e+02 4.480e+02, threshold=4.350e+02, percent-clipped=1.0 2023-03-09 13:56:24,525 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1329, 3.7668, 3.8319, 3.2855, 3.6933, 3.8850, 3.8226, 2.7320], device='cuda:3'), covar=tensor([0.0988, 0.1068, 0.1545, 0.2633, 0.2459, 0.2094, 0.0912, 0.3318], device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0215, 0.0229, 0.0280, 0.0191, 0.0291, 0.0214, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 13:56:44,301 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5677, 3.2311, 3.7317, 3.0298, 3.6722, 4.6897, 4.5339, 3.5762], device='cuda:3'), covar=tensor([0.0374, 0.1520, 0.1260, 0.1414, 0.1056, 0.0818, 0.0574, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0254, 0.0296, 0.0223, 0.0275, 0.0387, 0.0278, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:56:55,931 INFO [train2.py:809] (3/4) Epoch 29, batch 3950, loss[ctc_loss=0.04671, att_loss=0.2296, loss=0.1931, over 16945.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.008868, over 50.00 utterances.], tot_loss[ctc_loss=0.06383, att_loss=0.2298, loss=0.1966, over 3261055.95 frames. utt_duration=1227 frames, utt_pad_proportion=0.06288, over 10642.87 utterances.], batch size: 50, lr: 3.66e-03, grad_scale: 8.0 2023-03-09 13:57:10,207 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0546, 5.1082, 4.8608, 2.1578, 2.1879, 2.8783, 2.3967, 3.8548], device='cuda:3'), covar=tensor([0.0758, 0.0334, 0.0324, 0.5727, 0.5300, 0.2557, 0.4128, 0.1637], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0307, 0.0280, 0.0253, 0.0340, 0.0332, 0.0265, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 13:57:41,629 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7000, 5.0607, 4.9138, 4.9657, 5.1532, 4.7662, 3.7767, 4.9408], device='cuda:3'), covar=tensor([0.0117, 0.0107, 0.0136, 0.0084, 0.0084, 0.0111, 0.0564, 0.0184], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0093, 0.0119, 0.0074, 0.0080, 0.0091, 0.0107, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:57:44,607 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115527.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:58:12,588 INFO [train2.py:809] (3/4) Epoch 30, batch 0, loss[ctc_loss=0.06659, att_loss=0.2355, loss=0.2017, over 16285.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006884, over 43.00 utterances.], tot_loss[ctc_loss=0.06659, att_loss=0.2355, loss=0.2017, over 16285.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.006884, over 43.00 utterances.], batch size: 43, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 13:58:12,589 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 13:58:17,420 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.7893, 4.5074, 4.5548, 2.1857, 1.9583, 2.9272, 2.1382, 3.5799], device='cuda:3'), covar=tensor([0.0723, 0.0372, 0.0304, 0.5003, 0.5407, 0.2254, 0.4134, 0.1476], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0307, 0.0280, 0.0253, 0.0340, 0.0332, 0.0265, 0.0373], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 13:58:24,678 INFO [train2.py:843] (3/4) Epoch 30, validation: ctc_loss=0.03959, att_loss=0.2341, loss=0.1952, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 13:58:24,679 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 13:59:13,838 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 1.763e+02 2.122e+02 2.494e+02 5.580e+02, threshold=4.245e+02, percent-clipped=4.0 2023-03-09 13:59:38,538 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115575.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 13:59:41,750 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1001, 5.3949, 5.6482, 5.4212, 5.6440, 6.0407, 5.3813, 6.1392], device='cuda:3'), covar=tensor([0.0702, 0.0716, 0.0866, 0.1423, 0.1623, 0.0942, 0.0706, 0.0683], device='cuda:3'), in_proj_covar=tensor([0.0921, 0.0531, 0.0645, 0.0683, 0.0909, 0.0668, 0.0516, 0.0642], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 13:59:42,899 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-09 13:59:44,598 INFO [train2.py:809] (3/4) Epoch 30, batch 50, loss[ctc_loss=0.05765, att_loss=0.2372, loss=0.2013, over 16870.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007229, over 49.00 utterances.], tot_loss[ctc_loss=0.065, att_loss=0.2316, loss=0.1983, over 732510.31 frames. utt_duration=1114 frames, utt_pad_proportion=0.08911, over 2632.89 utterances.], batch size: 49, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 13:59:55,939 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.2065, 5.4604, 5.7215, 5.5315, 5.7472, 6.0818, 5.4511, 6.2489], device='cuda:3'), covar=tensor([0.0617, 0.0745, 0.0862, 0.1318, 0.1611, 0.1001, 0.0647, 0.0594], device='cuda:3'), in_proj_covar=tensor([0.0919, 0.0530, 0.0643, 0.0681, 0.0907, 0.0666, 0.0515, 0.0640], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 14:00:15,412 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7726, 5.1462, 4.9695, 5.0482, 5.2131, 4.8147, 3.7765, 5.1168], device='cuda:3'), covar=tensor([0.0121, 0.0114, 0.0132, 0.0090, 0.0093, 0.0124, 0.0600, 0.0211], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0094, 0.0120, 0.0075, 0.0081, 0.0092, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 14:01:04,302 INFO [train2.py:809] (3/4) Epoch 30, batch 100, loss[ctc_loss=0.05665, att_loss=0.2097, loss=0.1791, over 15392.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.009818, over 35.00 utterances.], tot_loss[ctc_loss=0.06292, att_loss=0.2303, loss=0.1968, over 1296752.37 frames. utt_duration=1261 frames, utt_pad_proportion=0.05243, over 4119.76 utterances.], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 14:01:54,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.758e+02 2.072e+02 2.595e+02 5.402e+02, threshold=4.143e+02, percent-clipped=3.0 2023-03-09 14:02:23,896 INFO [train2.py:809] (3/4) Epoch 30, batch 150, loss[ctc_loss=0.05879, att_loss=0.2099, loss=0.1797, over 14883.00 frames. utt_duration=1806 frames, utt_pad_proportion=0.03285, over 33.00 utterances.], tot_loss[ctc_loss=0.06343, att_loss=0.2308, loss=0.1974, over 1726921.07 frames. utt_duration=1248 frames, utt_pad_proportion=0.05795, over 5541.60 utterances.], batch size: 33, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 14:03:39,637 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115726.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:03:41,137 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115727.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:03:44,048 INFO [train2.py:809] (3/4) Epoch 30, batch 200, loss[ctc_loss=0.04069, att_loss=0.2052, loss=0.1723, over 15363.00 frames. utt_duration=1758 frames, utt_pad_proportion=0.01143, over 35.00 utterances.], tot_loss[ctc_loss=0.06239, att_loss=0.23, loss=0.1965, over 2069225.10 frames. utt_duration=1211 frames, utt_pad_proportion=0.06623, over 6845.09 utterances.], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 14:04:04,996 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115742.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:04:32,779 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.790e+02 2.190e+02 2.640e+02 5.774e+02, threshold=4.380e+02, percent-clipped=3.0 2023-03-09 14:04:49,124 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9782, 5.1828, 5.1413, 5.1529, 5.2463, 5.1858, 4.8493, 4.6570], device='cuda:3'), covar=tensor([0.1014, 0.0639, 0.0375, 0.0581, 0.0303, 0.0357, 0.0501, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0544, 0.0388, 0.0381, 0.0386, 0.0454, 0.0456, 0.0385, 0.0425], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:04:58,750 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0148, 4.9868, 4.7221, 2.8891, 4.7785, 4.6530, 4.2119, 2.7318], device='cuda:3'), covar=tensor([0.0116, 0.0105, 0.0289, 0.1060, 0.0104, 0.0207, 0.0333, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0107, 0.0113, 0.0113, 0.0091, 0.0120, 0.0103, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:05:03,782 INFO [train2.py:809] (3/4) Epoch 30, batch 250, loss[ctc_loss=0.05752, att_loss=0.2287, loss=0.1944, over 16619.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005663, over 47.00 utterances.], tot_loss[ctc_loss=0.0634, att_loss=0.2305, loss=0.197, over 2331040.38 frames. utt_duration=1203 frames, utt_pad_proportion=0.0682, over 7759.44 utterances.], batch size: 47, lr: 3.60e-03, grad_scale: 8.0 2023-03-09 14:05:21,490 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115790.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:05:45,001 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-09 14:06:23,346 INFO [train2.py:809] (3/4) Epoch 30, batch 300, loss[ctc_loss=0.07719, att_loss=0.2375, loss=0.2055, over 14114.00 frames. utt_duration=388.2 frames, utt_pad_proportion=0.3248, over 146.00 utterances.], tot_loss[ctc_loss=0.06306, att_loss=0.2308, loss=0.1973, over 2538724.43 frames. utt_duration=1206 frames, utt_pad_proportion=0.06681, over 8427.40 utterances.], batch size: 146, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:07:12,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 1.839e+02 2.070e+02 2.435e+02 4.420e+02, threshold=4.140e+02, percent-clipped=2.0 2023-03-09 14:07:27,723 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7587, 5.1025, 4.9501, 4.9584, 5.0666, 4.7544, 3.7945, 5.0822], device='cuda:3'), covar=tensor([0.0131, 0.0130, 0.0165, 0.0095, 0.0145, 0.0138, 0.0633, 0.0230], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0094, 0.0120, 0.0075, 0.0081, 0.0092, 0.0108, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 14:07:43,746 INFO [train2.py:809] (3/4) Epoch 30, batch 350, loss[ctc_loss=0.0454, att_loss=0.205, loss=0.1731, over 15781.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.007337, over 38.00 utterances.], tot_loss[ctc_loss=0.0619, att_loss=0.2294, loss=0.1959, over 2692043.53 frames. utt_duration=1239 frames, utt_pad_proportion=0.06113, over 8698.10 utterances.], batch size: 38, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:08:05,275 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115892.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 14:09:04,299 INFO [train2.py:809] (3/4) Epoch 30, batch 400, loss[ctc_loss=0.05673, att_loss=0.2181, loss=0.1858, over 15953.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.007166, over 41.00 utterances.], tot_loss[ctc_loss=0.06221, att_loss=0.2294, loss=0.1959, over 2817954.09 frames. utt_duration=1242 frames, utt_pad_proportion=0.06148, over 9090.20 utterances.], batch size: 41, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:09:43,471 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115953.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 14:09:53,661 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 1.924e+02 2.363e+02 2.895e+02 9.842e+02, threshold=4.725e+02, percent-clipped=4.0 2023-03-09 14:10:24,231 INFO [train2.py:809] (3/4) Epoch 30, batch 450, loss[ctc_loss=0.06585, att_loss=0.2362, loss=0.2022, over 16678.00 frames. utt_duration=1452 frames, utt_pad_proportion=0.00616, over 46.00 utterances.], tot_loss[ctc_loss=0.06258, att_loss=0.23, loss=0.1965, over 2925462.08 frames. utt_duration=1233 frames, utt_pad_proportion=0.06059, over 9505.81 utterances.], batch size: 46, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:11:25,128 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116014.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:11:43,428 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116026.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:11:45,648 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116027.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:11:48,381 INFO [train2.py:809] (3/4) Epoch 30, batch 500, loss[ctc_loss=0.05318, att_loss=0.2302, loss=0.1948, over 16479.00 frames. utt_duration=1435 frames, utt_pad_proportion=0.005845, over 46.00 utterances.], tot_loss[ctc_loss=0.06261, att_loss=0.2305, loss=0.1969, over 3012734.53 frames. utt_duration=1234 frames, utt_pad_proportion=0.05604, over 9780.13 utterances.], batch size: 46, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:12:37,213 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.798e+02 2.149e+02 2.710e+02 6.262e+02, threshold=4.298e+02, percent-clipped=5.0 2023-03-09 14:12:59,515 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116074.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:13:01,057 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116075.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:13:01,339 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116075.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 14:13:07,554 INFO [train2.py:809] (3/4) Epoch 30, batch 550, loss[ctc_loss=0.06596, att_loss=0.2286, loss=0.1961, over 16281.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007204, over 43.00 utterances.], tot_loss[ctc_loss=0.06272, att_loss=0.2304, loss=0.1969, over 3070260.18 frames. utt_duration=1235 frames, utt_pad_proportion=0.05639, over 9959.18 utterances.], batch size: 43, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:13:26,481 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3780, 4.5090, 4.5405, 4.6627, 4.9829, 4.4895, 4.4676, 2.6655], device='cuda:3'), covar=tensor([0.0329, 0.0369, 0.0349, 0.0301, 0.0772, 0.0296, 0.0377, 0.1622], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0231, 0.0224, 0.0243, 0.0386, 0.0199, 0.0218, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:13:48,822 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116105.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:14:06,592 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116116.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:14:24,436 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.0325, 5.1848, 5.1963, 5.2125, 5.2722, 5.2487, 4.8765, 4.7194], device='cuda:3'), covar=tensor([0.0950, 0.0614, 0.0285, 0.0458, 0.0295, 0.0334, 0.0444, 0.0328], device='cuda:3'), in_proj_covar=tensor([0.0540, 0.0387, 0.0380, 0.0387, 0.0452, 0.0456, 0.0384, 0.0423], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:14:27,137 INFO [train2.py:809] (3/4) Epoch 30, batch 600, loss[ctc_loss=0.06954, att_loss=0.2175, loss=0.1879, over 15505.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008482, over 36.00 utterances.], tot_loss[ctc_loss=0.06214, att_loss=0.2303, loss=0.1966, over 3121627.09 frames. utt_duration=1253 frames, utt_pad_proportion=0.04974, over 9977.30 utterances.], batch size: 36, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:15:15,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.620e+01 1.813e+02 2.125e+02 2.957e+02 6.068e+02, threshold=4.250e+02, percent-clipped=9.0 2023-03-09 14:15:26,210 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116166.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:15:44,331 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116177.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:15:47,046 INFO [train2.py:809] (3/4) Epoch 30, batch 650, loss[ctc_loss=0.05491, att_loss=0.2293, loss=0.1944, over 17290.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01259, over 55.00 utterances.], tot_loss[ctc_loss=0.06194, att_loss=0.2303, loss=0.1966, over 3158881.01 frames. utt_duration=1261 frames, utt_pad_proportion=0.04784, over 10029.26 utterances.], batch size: 55, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:16:52,239 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4622, 4.7078, 4.4058, 4.7213, 4.2865, 4.4513, 4.7692, 4.5884], device='cuda:3'), covar=tensor([0.0600, 0.0328, 0.0681, 0.0417, 0.0417, 0.0374, 0.0243, 0.0237], device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0345, 0.0385, 0.0387, 0.0341, 0.0248, 0.0324, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 14:17:07,961 INFO [train2.py:809] (3/4) Epoch 30, batch 700, loss[ctc_loss=0.06404, att_loss=0.2241, loss=0.1921, over 16692.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006156, over 46.00 utterances.], tot_loss[ctc_loss=0.06214, att_loss=0.2303, loss=0.1967, over 3185344.22 frames. utt_duration=1266 frames, utt_pad_proportion=0.0489, over 10077.88 utterances.], batch size: 46, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:17:38,396 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116248.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 14:17:57,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.730e+02 1.988e+02 2.420e+02 4.405e+02, threshold=3.977e+02, percent-clipped=2.0 2023-03-09 14:18:28,319 INFO [train2.py:809] (3/4) Epoch 30, batch 750, loss[ctc_loss=0.08936, att_loss=0.2544, loss=0.2214, over 17041.00 frames. utt_duration=1312 frames, utt_pad_proportion=0.008038, over 52.00 utterances.], tot_loss[ctc_loss=0.0625, att_loss=0.2302, loss=0.1967, over 3193906.51 frames. utt_duration=1251 frames, utt_pad_proportion=0.05554, over 10228.46 utterances.], batch size: 52, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:18:37,789 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-09 14:19:38,012 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116323.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:19:38,960 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-03-09 14:19:47,385 INFO [train2.py:809] (3/4) Epoch 30, batch 800, loss[ctc_loss=0.06958, att_loss=0.2514, loss=0.215, over 17148.00 frames. utt_duration=1226 frames, utt_pad_proportion=0.01362, over 56.00 utterances.], tot_loss[ctc_loss=0.06275, att_loss=0.2301, loss=0.1967, over 3207969.72 frames. utt_duration=1245 frames, utt_pad_proportion=0.05864, over 10321.18 utterances.], batch size: 56, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:20:36,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 1.849e+02 2.173e+02 2.485e+02 4.246e+02, threshold=4.345e+02, percent-clipped=3.0 2023-03-09 14:20:52,676 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116370.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 14:21:06,897 INFO [train2.py:809] (3/4) Epoch 30, batch 850, loss[ctc_loss=0.05915, att_loss=0.2083, loss=0.1785, over 15904.00 frames. utt_duration=1632 frames, utt_pad_proportion=0.008225, over 39.00 utterances.], tot_loss[ctc_loss=0.06264, att_loss=0.2302, loss=0.1967, over 3225059.75 frames. utt_duration=1260 frames, utt_pad_proportion=0.05405, over 10253.60 utterances.], batch size: 39, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:21:15,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116384.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:22:08,327 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.1783, 3.6780, 3.8166, 3.0139, 3.7749, 3.8922, 3.8662, 2.5840], device='cuda:3'), covar=tensor([0.1091, 0.1542, 0.1515, 0.5231, 0.1093, 0.1649, 0.0969, 0.5299], device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0215, 0.0230, 0.0280, 0.0191, 0.0290, 0.0213, 0.0236], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:22:27,525 INFO [train2.py:809] (3/4) Epoch 30, batch 900, loss[ctc_loss=0.08891, att_loss=0.2459, loss=0.2145, over 13973.00 frames. utt_duration=384.2 frames, utt_pad_proportion=0.3318, over 146.00 utterances.], tot_loss[ctc_loss=0.06262, att_loss=0.2305, loss=0.1969, over 3238503.96 frames. utt_duration=1248 frames, utt_pad_proportion=0.05497, over 10388.48 utterances.], batch size: 146, lr: 3.59e-03, grad_scale: 8.0 2023-03-09 14:22:54,794 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116446.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:23:06,054 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9585, 6.2225, 5.7043, 5.9353, 5.8949, 5.3386, 5.6846, 5.4591], device='cuda:3'), covar=tensor([0.1277, 0.0978, 0.1023, 0.1005, 0.1047, 0.1725, 0.2336, 0.2157], device='cuda:3'), in_proj_covar=tensor([0.0568, 0.0640, 0.0496, 0.0478, 0.0456, 0.0486, 0.0648, 0.0547], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:23:16,726 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 1.807e+02 2.134e+02 2.722e+02 5.477e+02, threshold=4.269e+02, percent-clipped=4.0 2023-03-09 14:23:18,511 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116461.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:23:36,294 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116472.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:23:47,730 INFO [train2.py:809] (3/4) Epoch 30, batch 950, loss[ctc_loss=0.06713, att_loss=0.235, loss=0.2015, over 16416.00 frames. utt_duration=1494 frames, utt_pad_proportion=0.006785, over 44.00 utterances.], tot_loss[ctc_loss=0.06227, att_loss=0.2297, loss=0.1962, over 3238602.68 frames. utt_duration=1259 frames, utt_pad_proportion=0.05333, over 10299.67 utterances.], batch size: 44, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:24:29,914 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116505.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:24:33,191 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116507.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 14:25:07,881 INFO [train2.py:809] (3/4) Epoch 30, batch 1000, loss[ctc_loss=0.05898, att_loss=0.2105, loss=0.1802, over 15397.00 frames. utt_duration=1761 frames, utt_pad_proportion=0.009561, over 35.00 utterances.], tot_loss[ctc_loss=0.06198, att_loss=0.2298, loss=0.1963, over 3248853.39 frames. utt_duration=1270 frames, utt_pad_proportion=0.04985, over 10246.07 utterances.], batch size: 35, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:25:38,110 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116548.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 14:25:56,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.757e+02 2.029e+02 2.635e+02 6.922e+02, threshold=4.057e+02, percent-clipped=2.0 2023-03-09 14:26:02,513 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7450, 3.1414, 3.7675, 3.2934, 3.6935, 4.7523, 4.5958, 3.6011], device='cuda:3'), covar=tensor([0.0294, 0.1688, 0.1235, 0.1161, 0.1035, 0.0865, 0.0508, 0.1068], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0256, 0.0297, 0.0223, 0.0276, 0.0388, 0.0279, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 14:26:07,245 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116566.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:26:21,766 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.7227, 2.5149, 2.5603, 2.9117, 2.9791, 2.8764, 2.4858, 3.1895], device='cuda:3'), covar=tensor([0.1943, 0.2234, 0.1870, 0.1209, 0.1600, 0.1196, 0.1971, 0.1176], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0152, 0.0149, 0.0145, 0.0161, 0.0140, 0.0163, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 14:26:27,614 INFO [train2.py:809] (3/4) Epoch 30, batch 1050, loss[ctc_loss=0.06759, att_loss=0.2487, loss=0.2125, over 17280.00 frames. utt_duration=1258 frames, utt_pad_proportion=0.01145, over 55.00 utterances.], tot_loss[ctc_loss=0.06254, att_loss=0.2305, loss=0.1969, over 3249481.17 frames. utt_duration=1245 frames, utt_pad_proportion=0.05612, over 10450.87 utterances.], batch size: 55, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:26:54,766 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116596.0, num_to_drop=1, layers_to_drop={1} 2023-03-09 14:27:30,209 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116618.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:27:47,628 INFO [train2.py:809] (3/4) Epoch 30, batch 1100, loss[ctc_loss=0.06519, att_loss=0.2374, loss=0.2029, over 17106.00 frames. utt_duration=685.8 frames, utt_pad_proportion=0.1297, over 100.00 utterances.], tot_loss[ctc_loss=0.06315, att_loss=0.2309, loss=0.1973, over 3268198.69 frames. utt_duration=1250 frames, utt_pad_proportion=0.05031, over 10471.50 utterances.], batch size: 100, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:28:37,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 1.803e+02 2.274e+02 2.882e+02 6.074e+02, threshold=4.549e+02, percent-clipped=3.0 2023-03-09 14:28:54,187 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116670.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 14:29:07,782 INFO [train2.py:809] (3/4) Epoch 30, batch 1150, loss[ctc_loss=0.05521, att_loss=0.2323, loss=0.1969, over 16626.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.005155, over 47.00 utterances.], tot_loss[ctc_loss=0.06391, att_loss=0.2315, loss=0.1979, over 3271404.48 frames. utt_duration=1218 frames, utt_pad_proportion=0.0588, over 10757.05 utterances.], batch size: 47, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:29:08,031 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:29:08,199 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:29:47,310 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116704.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:30:09,499 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116718.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:30:26,871 INFO [train2.py:809] (3/4) Epoch 30, batch 1200, loss[ctc_loss=0.06972, att_loss=0.2379, loss=0.2043, over 17283.00 frames. utt_duration=1173 frames, utt_pad_proportion=0.02549, over 59.00 utterances.], tot_loss[ctc_loss=0.06368, att_loss=0.2317, loss=0.1981, over 3277737.39 frames. utt_duration=1212 frames, utt_pad_proportion=0.05926, over 10829.53 utterances.], batch size: 59, lr: 3.58e-03, grad_scale: 8.0 2023-03-09 14:30:42,691 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1435, 4.3218, 4.0553, 4.5822, 2.7531, 4.3955, 2.3769, 1.9799], device='cuda:3'), covar=tensor([0.0496, 0.0282, 0.0874, 0.0259, 0.1820, 0.0277, 0.1831, 0.1762], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0198, 0.0270, 0.0187, 0.0228, 0.0178, 0.0238, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:31:16,204 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.678e+02 2.013e+02 2.421e+02 9.091e+02, threshold=4.026e+02, percent-clipped=5.0 2023-03-09 14:31:18,129 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116761.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:31:25,104 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116765.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:31:36,027 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:31:47,171 INFO [train2.py:809] (3/4) Epoch 30, batch 1250, loss[ctc_loss=0.06485, att_loss=0.2196, loss=0.1886, over 15955.00 frames. utt_duration=1558 frames, utt_pad_proportion=0.006269, over 41.00 utterances.], tot_loss[ctc_loss=0.06401, att_loss=0.2322, loss=0.1985, over 3289719.52 frames. utt_duration=1226 frames, utt_pad_proportion=0.05384, over 10750.61 utterances.], batch size: 41, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:32:23,516 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116802.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 14:32:23,663 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116802.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:32:35,290 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116809.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:32:53,101 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:33:07,494 INFO [train2.py:809] (3/4) Epoch 30, batch 1300, loss[ctc_loss=0.05438, att_loss=0.2256, loss=0.1914, over 16544.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.006047, over 45.00 utterances.], tot_loss[ctc_loss=0.06393, att_loss=0.2319, loss=0.1983, over 3279141.43 frames. utt_duration=1211 frames, utt_pad_proportion=0.06077, over 10840.80 utterances.], batch size: 45, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:33:38,002 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1201, 5.4332, 4.9407, 5.4888, 4.9162, 5.0846, 5.5076, 5.2586], device='cuda:3'), covar=tensor([0.0526, 0.0287, 0.0757, 0.0335, 0.0341, 0.0227, 0.0253, 0.0202], device='cuda:3'), in_proj_covar=tensor([0.0405, 0.0340, 0.0381, 0.0383, 0.0335, 0.0244, 0.0320, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 14:33:57,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 1.796e+02 2.133e+02 2.719e+02 8.820e+02, threshold=4.267e+02, percent-clipped=2.0 2023-03-09 14:33:58,895 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116861.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:34:02,743 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:34:28,216 INFO [train2.py:809] (3/4) Epoch 30, batch 1350, loss[ctc_loss=0.08012, att_loss=0.2482, loss=0.2146, over 16616.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005962, over 47.00 utterances.], tot_loss[ctc_loss=0.06349, att_loss=0.2319, loss=0.1982, over 3283012.60 frames. utt_duration=1234 frames, utt_pad_proportion=0.05514, over 10659.09 utterances.], batch size: 47, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:34:29,982 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5976, 5.8973, 5.3800, 5.6301, 5.5500, 5.0037, 5.3011, 5.0321], device='cuda:3'), covar=tensor([0.1505, 0.0910, 0.1079, 0.0919, 0.1278, 0.1739, 0.2475, 0.2447], device='cuda:3'), in_proj_covar=tensor([0.0570, 0.0646, 0.0498, 0.0480, 0.0461, 0.0489, 0.0653, 0.0550], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:35:48,164 INFO [train2.py:809] (3/4) Epoch 30, batch 1400, loss[ctc_loss=0.06264, att_loss=0.2235, loss=0.1913, over 16281.00 frames. utt_duration=1516 frames, utt_pad_proportion=0.007173, over 43.00 utterances.], tot_loss[ctc_loss=0.06399, att_loss=0.2324, loss=0.1987, over 3282060.94 frames. utt_duration=1206 frames, utt_pad_proportion=0.06198, over 10898.52 utterances.], batch size: 43, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:36:37,550 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 1.819e+02 2.157e+02 2.707e+02 5.079e+02, threshold=4.315e+02, percent-clipped=2.0 2023-03-09 14:37:01,015 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116974.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:37:08,671 INFO [train2.py:809] (3/4) Epoch 30, batch 1450, loss[ctc_loss=0.0501, att_loss=0.2152, loss=0.1821, over 16264.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007574, over 43.00 utterances.], tot_loss[ctc_loss=0.06377, att_loss=0.2315, loss=0.198, over 3280591.04 frames. utt_duration=1225 frames, utt_pad_proportion=0.05815, over 10725.03 utterances.], batch size: 43, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:37:08,993 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116979.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:37:54,475 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-09 14:38:00,659 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0929, 4.3546, 4.5152, 4.7365, 3.0579, 4.4172, 3.0314, 2.2380], device='cuda:3'), covar=tensor([0.0475, 0.0318, 0.0577, 0.0231, 0.1405, 0.0253, 0.1231, 0.1472], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0197, 0.0268, 0.0185, 0.0226, 0.0177, 0.0237, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:38:25,653 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117027.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:38:28,585 INFO [train2.py:809] (3/4) Epoch 30, batch 1500, loss[ctc_loss=0.07603, att_loss=0.2457, loss=0.2118, over 17331.00 frames. utt_duration=1177 frames, utt_pad_proportion=0.02025, over 59.00 utterances.], tot_loss[ctc_loss=0.06352, att_loss=0.2317, loss=0.1981, over 3284948.96 frames. utt_duration=1218 frames, utt_pad_proportion=0.05861, over 10798.52 utterances.], batch size: 59, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:39:18,346 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.806e+02 2.089e+02 2.584e+02 4.922e+02, threshold=4.178e+02, percent-clipped=1.0 2023-03-09 14:39:18,628 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117060.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:39:48,449 INFO [train2.py:809] (3/4) Epoch 30, batch 1550, loss[ctc_loss=0.05337, att_loss=0.2319, loss=0.1962, over 16614.00 frames. utt_duration=1415 frames, utt_pad_proportion=0.006141, over 47.00 utterances.], tot_loss[ctc_loss=0.06352, att_loss=0.2315, loss=0.1979, over 3277585.01 frames. utt_duration=1178 frames, utt_pad_proportion=0.07007, over 11141.10 utterances.], batch size: 47, lr: 3.58e-03, grad_scale: 16.0 2023-03-09 14:39:58,205 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.6120, 3.4750, 3.4883, 3.8095, 2.7680, 3.6809, 2.8469, 2.2294], device='cuda:3'), covar=tensor([0.0577, 0.0489, 0.0821, 0.0379, 0.1425, 0.0373, 0.1284, 0.1509], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0198, 0.0270, 0.0187, 0.0228, 0.0178, 0.0239, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:40:25,860 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117102.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:41:08,937 INFO [train2.py:809] (3/4) Epoch 30, batch 1600, loss[ctc_loss=0.06003, att_loss=0.2169, loss=0.1856, over 16113.00 frames. utt_duration=1536 frames, utt_pad_proportion=0.006561, over 42.00 utterances.], tot_loss[ctc_loss=0.06328, att_loss=0.2311, loss=0.1976, over 3273100.59 frames. utt_duration=1197 frames, utt_pad_proportion=0.06709, over 10952.47 utterances.], batch size: 42, lr: 3.57e-03, grad_scale: 16.0 2023-03-09 14:41:13,104 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 14:41:43,255 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117150.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:41:56,323 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117158.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:41:59,200 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.817e+02 2.106e+02 2.742e+02 6.475e+02, threshold=4.211e+02, percent-clipped=3.0 2023-03-09 14:42:00,963 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117161.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:42:14,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-09 14:42:25,180 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0460, 4.3302, 4.2882, 4.7084, 2.9784, 4.3679, 2.8890, 2.1360], device='cuda:3'), covar=tensor([0.0541, 0.0294, 0.0649, 0.0234, 0.1402, 0.0281, 0.1333, 0.1609], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0199, 0.0270, 0.0187, 0.0228, 0.0178, 0.0239, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:42:29,624 INFO [train2.py:809] (3/4) Epoch 30, batch 1650, loss[ctc_loss=0.05329, att_loss=0.2242, loss=0.19, over 15877.00 frames. utt_duration=1630 frames, utt_pad_proportion=0.009212, over 39.00 utterances.], tot_loss[ctc_loss=0.06317, att_loss=0.2309, loss=0.1973, over 3273109.32 frames. utt_duration=1206 frames, utt_pad_proportion=0.06523, over 10866.80 utterances.], batch size: 39, lr: 3.57e-03, grad_scale: 16.0 2023-03-09 14:42:54,847 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.84 vs. limit=5.0 2023-03-09 14:43:19,244 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117209.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:43:31,939 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6539, 4.9131, 4.3896, 4.7544, 4.5736, 4.0743, 4.4212, 4.1939], device='cuda:3'), covar=tensor([0.1447, 0.1236, 0.1290, 0.1111, 0.1295, 0.1878, 0.2440, 0.2333], device='cuda:3'), in_proj_covar=tensor([0.0567, 0.0644, 0.0498, 0.0482, 0.0458, 0.0488, 0.0649, 0.0546], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:43:39,155 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117221.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:43:51,228 INFO [train2.py:809] (3/4) Epoch 30, batch 1700, loss[ctc_loss=0.05757, att_loss=0.2377, loss=0.2017, over 17289.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.01263, over 55.00 utterances.], tot_loss[ctc_loss=0.06299, att_loss=0.231, loss=0.1974, over 3279768.63 frames. utt_duration=1206 frames, utt_pad_proportion=0.06379, over 10891.79 utterances.], batch size: 55, lr: 3.57e-03, grad_scale: 16.0 2023-03-09 14:44:43,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.821e+02 2.138e+02 2.705e+02 3.860e+02, threshold=4.276e+02, percent-clipped=0.0 2023-03-09 14:45:03,570 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.5270, 2.9253, 4.9098, 4.0239, 3.1813, 4.2887, 4.8008, 4.7123], device='cuda:3'), covar=tensor([0.0322, 0.1309, 0.0269, 0.0799, 0.1573, 0.0275, 0.0211, 0.0290], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0249, 0.0237, 0.0326, 0.0274, 0.0249, 0.0231, 0.0251], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 14:45:04,991 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117274.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:45:12,573 INFO [train2.py:809] (3/4) Epoch 30, batch 1750, loss[ctc_loss=0.0426, att_loss=0.2032, loss=0.1711, over 15766.00 frames. utt_duration=1661 frames, utt_pad_proportion=0.009138, over 38.00 utterances.], tot_loss[ctc_loss=0.0625, att_loss=0.2299, loss=0.1965, over 3264416.68 frames. utt_duration=1206 frames, utt_pad_proportion=0.06931, over 10841.68 utterances.], batch size: 38, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:45:17,643 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117282.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:46:23,172 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117322.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:46:34,461 INFO [train2.py:809] (3/4) Epoch 30, batch 1800, loss[ctc_loss=0.06343, att_loss=0.243, loss=0.2071, over 17089.00 frames. utt_duration=1222 frames, utt_pad_proportion=0.01672, over 56.00 utterances.], tot_loss[ctc_loss=0.06199, att_loss=0.2297, loss=0.1961, over 3269122.24 frames. utt_duration=1241 frames, utt_pad_proportion=0.05907, over 10552.79 utterances.], batch size: 56, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:46:34,855 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:46:35,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-09 14:47:17,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-09 14:47:25,092 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117360.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:47:26,234 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.761e+02 2.042e+02 2.458e+02 5.519e+02, threshold=4.084e+02, percent-clipped=2.0 2023-03-09 14:47:54,421 INFO [train2.py:809] (3/4) Epoch 30, batch 1850, loss[ctc_loss=0.05829, att_loss=0.2114, loss=0.1808, over 15778.00 frames. utt_duration=1663 frames, utt_pad_proportion=0.00626, over 38.00 utterances.], tot_loss[ctc_loss=0.06251, att_loss=0.2299, loss=0.1964, over 3265183.74 frames. utt_duration=1237 frames, utt_pad_proportion=0.06177, over 10573.34 utterances.], batch size: 38, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:48:12,244 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117390.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:48:41,558 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117408.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:49:14,297 INFO [train2.py:809] (3/4) Epoch 30, batch 1900, loss[ctc_loss=0.05525, att_loss=0.232, loss=0.1966, over 16623.00 frames. utt_duration=1416 frames, utt_pad_proportion=0.005603, over 47.00 utterances.], tot_loss[ctc_loss=0.06258, att_loss=0.2306, loss=0.197, over 3273356.48 frames. utt_duration=1215 frames, utt_pad_proportion=0.06468, over 10792.93 utterances.], batch size: 47, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:49:23,011 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-09 14:49:23,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8530, 4.8306, 4.5577, 3.0515, 4.6470, 4.6074, 4.1721, 2.6261], device='cuda:3'), covar=tensor([0.0122, 0.0127, 0.0312, 0.0939, 0.0120, 0.0223, 0.0321, 0.1478], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0109, 0.0113, 0.0115, 0.0091, 0.0121, 0.0103, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:49:54,996 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-09 14:50:00,426 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117458.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:50:04,885 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.786e+02 2.185e+02 2.601e+02 6.892e+02, threshold=4.370e+02, percent-clipped=2.0 2023-03-09 14:50:33,476 INFO [train2.py:809] (3/4) Epoch 30, batch 1950, loss[ctc_loss=0.04642, att_loss=0.2169, loss=0.1828, over 16408.00 frames. utt_duration=1493 frames, utt_pad_proportion=0.006668, over 44.00 utterances.], tot_loss[ctc_loss=0.06299, att_loss=0.2307, loss=0.1971, over 3272651.04 frames. utt_duration=1199 frames, utt_pad_proportion=0.0685, over 10929.97 utterances.], batch size: 44, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:51:17,143 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117506.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:51:36,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.83 vs. limit=5.0 2023-03-09 14:51:53,202 INFO [train2.py:809] (3/4) Epoch 30, batch 2000, loss[ctc_loss=0.06038, att_loss=0.2392, loss=0.2034, over 16774.00 frames. utt_duration=1400 frames, utt_pad_proportion=0.006007, over 48.00 utterances.], tot_loss[ctc_loss=0.06241, att_loss=0.23, loss=0.1965, over 3272316.03 frames. utt_duration=1214 frames, utt_pad_proportion=0.06406, over 10797.28 utterances.], batch size: 48, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:52:43,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.738e+02 2.282e+02 2.861e+02 7.444e+02, threshold=4.565e+02, percent-clipped=5.0 2023-03-09 14:53:08,576 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117577.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:53:11,454 INFO [train2.py:809] (3/4) Epoch 30, batch 2050, loss[ctc_loss=0.05497, att_loss=0.228, loss=0.1934, over 16632.00 frames. utt_duration=1417 frames, utt_pad_proportion=0.004975, over 47.00 utterances.], tot_loss[ctc_loss=0.06278, att_loss=0.2302, loss=0.1967, over 3273947.33 frames. utt_duration=1176 frames, utt_pad_proportion=0.07131, over 11146.73 utterances.], batch size: 47, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:53:26,549 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.7339, 3.4582, 3.8882, 3.5236, 3.8102, 4.7491, 4.6247, 3.7178], device='cuda:3'), covar=tensor([0.0306, 0.1262, 0.1161, 0.1019, 0.0926, 0.0900, 0.0552, 0.0930], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0251, 0.0291, 0.0217, 0.0272, 0.0381, 0.0274, 0.0237], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 14:53:57,602 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9845, 5.1878, 5.1630, 5.1469, 5.2917, 5.2315, 4.9233, 4.6855], device='cuda:3'), covar=tensor([0.1091, 0.0593, 0.0349, 0.0612, 0.0303, 0.0343, 0.0414, 0.0387], device='cuda:3'), in_proj_covar=tensor([0.0543, 0.0387, 0.0382, 0.0389, 0.0449, 0.0456, 0.0385, 0.0423], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:54:32,200 INFO [train2.py:809] (3/4) Epoch 30, batch 2100, loss[ctc_loss=0.06346, att_loss=0.2156, loss=0.1851, over 16269.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007859, over 43.00 utterances.], tot_loss[ctc_loss=0.0631, att_loss=0.2301, loss=0.1967, over 3268102.68 frames. utt_duration=1175 frames, utt_pad_proportion=0.07421, over 11138.61 utterances.], batch size: 43, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:55:24,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 1.716e+02 1.992e+02 2.490e+02 6.469e+02, threshold=3.983e+02, percent-clipped=1.0 2023-03-09 14:55:52,883 INFO [train2.py:809] (3/4) Epoch 30, batch 2150, loss[ctc_loss=0.07062, att_loss=0.2313, loss=0.1992, over 16545.00 frames. utt_duration=1472 frames, utt_pad_proportion=0.005927, over 45.00 utterances.], tot_loss[ctc_loss=0.06333, att_loss=0.231, loss=0.1974, over 3273472.45 frames. utt_duration=1171 frames, utt_pad_proportion=0.07382, over 11193.06 utterances.], batch size: 45, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:56:02,891 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117685.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 14:57:01,409 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.0136, 6.2489, 5.7650, 5.9625, 5.9235, 5.3982, 5.8305, 5.4475], device='cuda:3'), covar=tensor([0.1432, 0.0994, 0.1103, 0.0822, 0.0836, 0.1542, 0.2101, 0.2358], device='cuda:3'), in_proj_covar=tensor([0.0567, 0.0643, 0.0496, 0.0481, 0.0460, 0.0488, 0.0650, 0.0550], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:57:12,174 INFO [train2.py:809] (3/4) Epoch 30, batch 2200, loss[ctc_loss=0.08202, att_loss=0.2361, loss=0.2053, over 16119.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006083, over 42.00 utterances.], tot_loss[ctc_loss=0.06322, att_loss=0.2312, loss=0.1976, over 3273458.96 frames. utt_duration=1179 frames, utt_pad_proportion=0.07256, over 11120.86 utterances.], batch size: 42, lr: 3.57e-03, grad_scale: 8.0 2023-03-09 14:57:47,241 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2493, 5.5328, 5.4852, 5.4074, 5.5821, 5.4895, 5.1988, 4.9417], device='cuda:3'), covar=tensor([0.1000, 0.0470, 0.0276, 0.0528, 0.0251, 0.0295, 0.0371, 0.0361], device='cuda:3'), in_proj_covar=tensor([0.0544, 0.0385, 0.0382, 0.0388, 0.0448, 0.0455, 0.0385, 0.0422], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 14:58:03,080 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.299e+02 1.825e+02 2.078e+02 2.480e+02 5.999e+02, threshold=4.155e+02, percent-clipped=3.0 2023-03-09 14:58:30,569 INFO [train2.py:809] (3/4) Epoch 30, batch 2250, loss[ctc_loss=0.05544, att_loss=0.2328, loss=0.1973, over 16475.00 frames. utt_duration=1434 frames, utt_pad_proportion=0.006071, over 46.00 utterances.], tot_loss[ctc_loss=0.06319, att_loss=0.2309, loss=0.1974, over 3267049.52 frames. utt_duration=1177 frames, utt_pad_proportion=0.07363, over 11116.82 utterances.], batch size: 46, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 14:59:50,947 INFO [train2.py:809] (3/4) Epoch 30, batch 2300, loss[ctc_loss=0.05104, att_loss=0.2131, loss=0.1807, over 15631.00 frames. utt_duration=1692 frames, utt_pad_proportion=0.008993, over 37.00 utterances.], tot_loss[ctc_loss=0.06377, att_loss=0.2317, loss=0.1981, over 3261351.76 frames. utt_duration=1146 frames, utt_pad_proportion=0.08083, over 11393.94 utterances.], batch size: 37, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:00:39,538 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0713, 4.3110, 4.2369, 4.6837, 3.0271, 4.4624, 2.9425, 1.8479], device='cuda:3'), covar=tensor([0.0509, 0.0335, 0.0709, 0.0233, 0.1326, 0.0254, 0.1340, 0.1684], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0197, 0.0266, 0.0184, 0.0224, 0.0175, 0.0234, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:00:42,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 1.803e+02 2.047e+02 2.475e+02 6.158e+02, threshold=4.093e+02, percent-clipped=3.0 2023-03-09 15:01:07,134 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117877.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:01:09,866 INFO [train2.py:809] (3/4) Epoch 30, batch 2350, loss[ctc_loss=0.06766, att_loss=0.2151, loss=0.1856, over 15504.00 frames. utt_duration=1724 frames, utt_pad_proportion=0.008386, over 36.00 utterances.], tot_loss[ctc_loss=0.06441, att_loss=0.2319, loss=0.1984, over 3265628.40 frames. utt_duration=1158 frames, utt_pad_proportion=0.07769, over 11299.67 utterances.], batch size: 36, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:01:27,764 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.5394, 5.7906, 5.2734, 5.5615, 5.4102, 4.9571, 5.2823, 5.0456], device='cuda:3'), covar=tensor([0.1313, 0.0974, 0.0991, 0.0877, 0.1145, 0.1586, 0.2156, 0.2222], device='cuda:3'), in_proj_covar=tensor([0.0567, 0.0644, 0.0496, 0.0484, 0.0460, 0.0488, 0.0647, 0.0550], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:02:23,397 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117925.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:02:29,493 INFO [train2.py:809] (3/4) Epoch 30, batch 2400, loss[ctc_loss=0.06011, att_loss=0.2221, loss=0.1897, over 15609.00 frames. utt_duration=1689 frames, utt_pad_proportion=0.01061, over 37.00 utterances.], tot_loss[ctc_loss=0.06373, att_loss=0.231, loss=0.1975, over 3270615.48 frames. utt_duration=1179 frames, utt_pad_proportion=0.06989, over 11108.68 utterances.], batch size: 37, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:02:43,423 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0074, 3.6976, 3.6895, 3.1750, 3.6652, 3.7819, 3.7974, 2.7928], device='cuda:3'), covar=tensor([0.1232, 0.1232, 0.2213, 0.3163, 0.1603, 0.1730, 0.0874, 0.3045], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0219, 0.0232, 0.0283, 0.0194, 0.0295, 0.0215, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:02:58,327 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-09 15:03:20,929 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 1.808e+02 2.132e+02 2.494e+02 4.997e+02, threshold=4.263e+02, percent-clipped=2.0 2023-03-09 15:03:30,703 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.0992, 4.3262, 4.6605, 4.6768, 3.2742, 4.4935, 3.1841, 2.2837], device='cuda:3'), covar=tensor([0.0475, 0.0387, 0.0527, 0.0316, 0.1168, 0.0262, 0.1171, 0.1491], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0196, 0.0265, 0.0184, 0.0224, 0.0174, 0.0234, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:03:48,710 INFO [train2.py:809] (3/4) Epoch 30, batch 2450, loss[ctc_loss=0.06971, att_loss=0.2468, loss=0.2114, over 16948.00 frames. utt_duration=1357 frames, utt_pad_proportion=0.007865, over 50.00 utterances.], tot_loss[ctc_loss=0.06367, att_loss=0.2313, loss=0.1978, over 3278094.80 frames. utt_duration=1195 frames, utt_pad_proportion=0.06383, over 10983.10 utterances.], batch size: 50, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:03:52,801 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6583, 5.0379, 4.9141, 5.0089, 5.0504, 4.7790, 3.4456, 4.9243], device='cuda:3'), covar=tensor([0.0136, 0.0110, 0.0139, 0.0073, 0.0106, 0.0111, 0.0746, 0.0209], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0095, 0.0122, 0.0076, 0.0082, 0.0094, 0.0109, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:3') 2023-03-09 15:03:59,599 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117985.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:05:10,870 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8038, 5.1814, 5.0588, 5.1477, 5.1877, 4.9252, 3.8068, 5.1363], device='cuda:3'), covar=tensor([0.0123, 0.0116, 0.0143, 0.0073, 0.0095, 0.0108, 0.0624, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0094, 0.0121, 0.0075, 0.0082, 0.0093, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:05:13,629 INFO [train2.py:809] (3/4) Epoch 30, batch 2500, loss[ctc_loss=0.08781, att_loss=0.25, loss=0.2175, over 17302.00 frames. utt_duration=1260 frames, utt_pad_proportion=0.01194, over 55.00 utterances.], tot_loss[ctc_loss=0.06333, att_loss=0.2309, loss=0.1974, over 3266589.95 frames. utt_duration=1191 frames, utt_pad_proportion=0.06828, over 10988.10 utterances.], batch size: 55, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:05:21,131 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118033.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:05:32,794 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118040.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:06:07,288 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.938e+02 2.187e+02 2.706e+02 8.579e+02, threshold=4.374e+02, percent-clipped=4.0 2023-03-09 15:06:35,074 INFO [train2.py:809] (3/4) Epoch 30, batch 2550, loss[ctc_loss=0.05317, att_loss=0.2303, loss=0.1949, over 17030.00 frames. utt_duration=1337 frames, utt_pad_proportion=0.006512, over 51.00 utterances.], tot_loss[ctc_loss=0.06366, att_loss=0.2315, loss=0.198, over 3260342.18 frames. utt_duration=1178 frames, utt_pad_proportion=0.07118, over 11080.28 utterances.], batch size: 51, lr: 3.56e-03, grad_scale: 4.0 2023-03-09 15:06:36,844 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5645, 2.5235, 2.6354, 2.2705, 2.5534, 2.4575, 2.5949, 1.8952], device='cuda:3'), covar=tensor([0.1315, 0.1720, 0.1892, 0.4013, 0.1428, 0.1651, 0.1468, 0.4472], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0219, 0.0233, 0.0283, 0.0195, 0.0294, 0.0216, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:07:11,148 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118101.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:07:54,647 INFO [train2.py:809] (3/4) Epoch 30, batch 2600, loss[ctc_loss=0.07838, att_loss=0.2474, loss=0.2136, over 16697.00 frames. utt_duration=1453 frames, utt_pad_proportion=0.006037, over 46.00 utterances.], tot_loss[ctc_loss=0.06354, att_loss=0.2314, loss=0.1979, over 3268162.18 frames. utt_duration=1203 frames, utt_pad_proportion=0.06408, over 10876.14 utterances.], batch size: 46, lr: 3.56e-03, grad_scale: 4.0 2023-03-09 15:08:46,945 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.796e+02 2.196e+02 2.749e+02 4.358e+02, threshold=4.391e+02, percent-clipped=0.0 2023-03-09 15:08:56,503 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118168.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:09:14,092 INFO [train2.py:809] (3/4) Epoch 30, batch 2650, loss[ctc_loss=0.09035, att_loss=0.2556, loss=0.2225, over 14810.00 frames. utt_duration=407.2 frames, utt_pad_proportion=0.2893, over 146.00 utterances.], tot_loss[ctc_loss=0.06388, att_loss=0.2315, loss=0.198, over 3271730.60 frames. utt_duration=1201 frames, utt_pad_proportion=0.06502, over 10912.19 utterances.], batch size: 146, lr: 3.56e-03, grad_scale: 4.0 2023-03-09 15:09:45,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2023-03-09 15:10:33,463 INFO [train2.py:809] (3/4) Epoch 30, batch 2700, loss[ctc_loss=0.07286, att_loss=0.2354, loss=0.2029, over 16975.00 frames. utt_duration=687.4 frames, utt_pad_proportion=0.1354, over 99.00 utterances.], tot_loss[ctc_loss=0.06373, att_loss=0.2307, loss=0.1973, over 3270486.44 frames. utt_duration=1222 frames, utt_pad_proportion=0.06059, over 10716.01 utterances.], batch size: 99, lr: 3.56e-03, grad_scale: 4.0 2023-03-09 15:10:33,872 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118229.0, num_to_drop=1, layers_to_drop={0} 2023-03-09 15:11:25,934 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.836e+02 2.266e+02 2.685e+02 4.437e+02, threshold=4.533e+02, percent-clipped=1.0 2023-03-09 15:11:35,707 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2477, 2.7329, 2.8381, 4.3929, 3.8610, 3.9344, 2.8797, 2.1082], device='cuda:3'), covar=tensor([0.0823, 0.1990, 0.1211, 0.0442, 0.0797, 0.0442, 0.1525, 0.2399], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0222, 0.0187, 0.0230, 0.0242, 0.0196, 0.0205, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:11:37,233 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118269.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:11:40,234 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.3319, 2.9376, 3.5211, 2.9564, 3.4644, 4.4280, 4.2170, 3.3044], device='cuda:3'), covar=tensor([0.0378, 0.1592, 0.1353, 0.1233, 0.1100, 0.0953, 0.0725, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0252, 0.0293, 0.0220, 0.0275, 0.0386, 0.0278, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:11:52,829 INFO [train2.py:809] (3/4) Epoch 30, batch 2750, loss[ctc_loss=0.05429, att_loss=0.2088, loss=0.1779, over 15649.00 frames. utt_duration=1693 frames, utt_pad_proportion=0.008529, over 37.00 utterances.], tot_loss[ctc_loss=0.06406, att_loss=0.2309, loss=0.1976, over 3276403.41 frames. utt_duration=1240 frames, utt_pad_proportion=0.05536, over 10581.72 utterances.], batch size: 37, lr: 3.56e-03, grad_scale: 4.0 2023-03-09 15:12:10,161 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118289.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:12:28,706 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.3097, 5.3537, 5.1839, 3.5248, 5.1875, 5.0137, 4.7080, 3.1157], device='cuda:3'), covar=tensor([0.0115, 0.0090, 0.0241, 0.0823, 0.0097, 0.0162, 0.0265, 0.1203], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0108, 0.0113, 0.0114, 0.0091, 0.0120, 0.0103, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:13:12,412 INFO [train2.py:809] (3/4) Epoch 30, batch 2800, loss[ctc_loss=0.09064, att_loss=0.2489, loss=0.2172, over 17287.00 frames. utt_duration=1259 frames, utt_pad_proportion=0.0113, over 55.00 utterances.], tot_loss[ctc_loss=0.06428, att_loss=0.2311, loss=0.1977, over 3276643.60 frames. utt_duration=1241 frames, utt_pad_proportion=0.05561, over 10570.83 utterances.], batch size: 55, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:13:12,799 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118329.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:13:15,016 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118330.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:13:47,162 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118350.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:14:05,712 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 1.819e+02 2.314e+02 2.830e+02 4.183e+02, threshold=4.629e+02, percent-clipped=0.0 2023-03-09 15:14:32,833 INFO [train2.py:809] (3/4) Epoch 30, batch 2850, loss[ctc_loss=0.05895, att_loss=0.2266, loss=0.1931, over 15972.00 frames. utt_duration=1560 frames, utt_pad_proportion=0.005398, over 41.00 utterances.], tot_loss[ctc_loss=0.06384, att_loss=0.2308, loss=0.1974, over 3274141.07 frames. utt_duration=1231 frames, utt_pad_proportion=0.05971, over 10652.27 utterances.], batch size: 41, lr: 3.56e-03, grad_scale: 8.0 2023-03-09 15:14:51,902 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118390.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:15:01,134 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118396.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:15:53,114 INFO [train2.py:809] (3/4) Epoch 30, batch 2900, loss[ctc_loss=0.05159, att_loss=0.2197, loss=0.1861, over 16125.00 frames. utt_duration=1537 frames, utt_pad_proportion=0.006464, over 42.00 utterances.], tot_loss[ctc_loss=0.06266, att_loss=0.2293, loss=0.1959, over 3269365.15 frames. utt_duration=1272 frames, utt_pad_proportion=0.05142, over 10295.53 utterances.], batch size: 42, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:16:02,004 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.0443, 3.7812, 3.2525, 3.4358, 3.9508, 3.6609, 3.0428, 4.1666], device='cuda:3'), covar=tensor([0.1115, 0.0561, 0.1012, 0.0757, 0.0770, 0.0819, 0.0931, 0.0554], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0233, 0.0236, 0.0213, 0.0296, 0.0256, 0.0209, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:3') 2023-03-09 15:16:46,258 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.677e+02 2.045e+02 2.519e+02 5.925e+02, threshold=4.089e+02, percent-clipped=1.0 2023-03-09 15:17:12,896 INFO [train2.py:809] (3/4) Epoch 30, batch 2950, loss[ctc_loss=0.06239, att_loss=0.2321, loss=0.1982, over 16468.00 frames. utt_duration=1433 frames, utt_pad_proportion=0.007332, over 46.00 utterances.], tot_loss[ctc_loss=0.06353, att_loss=0.2308, loss=0.1974, over 3276484.57 frames. utt_duration=1263 frames, utt_pad_proportion=0.0496, over 10393.22 utterances.], batch size: 46, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:17:28,478 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.4709, 3.0894, 3.4284, 4.5771, 4.0508, 4.1387, 3.1257, 2.4837], device='cuda:3'), covar=tensor([0.0732, 0.1854, 0.0874, 0.0552, 0.0879, 0.0480, 0.1408, 0.2027], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0222, 0.0187, 0.0230, 0.0241, 0.0196, 0.0205, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:18:23,847 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118524.0, num_to_drop=1, layers_to_drop={3} 2023-03-09 15:18:31,907 INFO [train2.py:809] (3/4) Epoch 30, batch 3000, loss[ctc_loss=0.05854, att_loss=0.2236, loss=0.1906, over 16882.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.007429, over 49.00 utterances.], tot_loss[ctc_loss=0.06291, att_loss=0.2302, loss=0.1967, over 3280970.53 frames. utt_duration=1287 frames, utt_pad_proportion=0.0434, over 10212.85 utterances.], batch size: 49, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:18:31,907 INFO [train2.py:834] (3/4) Computing validation loss 2023-03-09 15:18:46,178 INFO [train2.py:843] (3/4) Epoch 30, validation: ctc_loss=0.04128, att_loss=0.2346, loss=0.196, over 944034.00 frames. utt_duration=679.8 frames, utt_pad_proportion=0.1349, over 5567.00 utterances. 2023-03-09 15:18:46,179 INFO [train2.py:844] (3/4) Maximum memory allocated so far is 16114MB 2023-03-09 15:19:38,942 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.708e+02 2.016e+02 2.374e+02 4.392e+02, threshold=4.031e+02, percent-clipped=1.0 2023-03-09 15:20:05,704 INFO [train2.py:809] (3/4) Epoch 30, batch 3050, loss[ctc_loss=0.1051, att_loss=0.258, loss=0.2274, over 13751.00 frames. utt_duration=380.8 frames, utt_pad_proportion=0.3389, over 145.00 utterances.], tot_loss[ctc_loss=0.06274, att_loss=0.2301, loss=0.1966, over 3275742.57 frames. utt_duration=1243 frames, utt_pad_proportion=0.05486, over 10555.76 utterances.], batch size: 145, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:20:13,577 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.8215, 5.1729, 5.0134, 5.0848, 5.1745, 4.9026, 3.6981, 5.1684], device='cuda:3'), covar=tensor([0.0118, 0.0112, 0.0130, 0.0079, 0.0097, 0.0135, 0.0611, 0.0146], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0093, 0.0119, 0.0074, 0.0081, 0.0091, 0.0107, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:21:20,118 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118625.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:21:26,775 INFO [train2.py:809] (3/4) Epoch 30, batch 3100, loss[ctc_loss=0.06064, att_loss=0.2381, loss=0.2026, over 16871.00 frames. utt_duration=1379 frames, utt_pad_proportion=0.007258, over 49.00 utterances.], tot_loss[ctc_loss=0.0621, att_loss=0.2294, loss=0.1959, over 3265634.45 frames. utt_duration=1256 frames, utt_pad_proportion=0.05453, over 10415.91 utterances.], batch size: 49, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:21:29,131 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-09 15:21:30,283 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:21:49,156 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-09 15:21:53,102 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:22:19,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 1.896e+02 2.204e+02 2.593e+02 6.390e+02, threshold=4.408e+02, percent-clipped=1.0 2023-03-09 15:22:47,066 INFO [train2.py:809] (3/4) Epoch 30, batch 3150, loss[ctc_loss=0.06868, att_loss=0.2144, loss=0.1852, over 15535.00 frames. utt_duration=1728 frames, utt_pad_proportion=0.006341, over 36.00 utterances.], tot_loss[ctc_loss=0.0623, att_loss=0.2291, loss=0.1957, over 3271454.06 frames. utt_duration=1245 frames, utt_pad_proportion=0.05437, over 10524.04 utterances.], batch size: 36, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:22:56,742 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118685.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:23:08,825 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118692.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:23:14,664 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118696.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:23:23,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-09 15:23:47,014 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-09 15:23:55,647 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9050, 4.9210, 4.6267, 2.6846, 4.7044, 4.6016, 4.1895, 2.5343], device='cuda:3'), covar=tensor([0.0134, 0.0132, 0.0337, 0.1290, 0.0135, 0.0248, 0.0374, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0108, 0.0112, 0.0113, 0.0091, 0.0120, 0.0102, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:24:06,682 INFO [train2.py:809] (3/4) Epoch 30, batch 3200, loss[ctc_loss=0.08085, att_loss=0.2585, loss=0.2229, over 17106.00 frames. utt_duration=1224 frames, utt_pad_proportion=0.01557, over 56.00 utterances.], tot_loss[ctc_loss=0.06256, att_loss=0.2301, loss=0.1966, over 3283038.96 frames. utt_duration=1239 frames, utt_pad_proportion=0.05254, over 10614.69 utterances.], batch size: 56, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:24:30,988 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118744.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:24:59,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.732e+02 2.054e+02 2.597e+02 6.136e+02, threshold=4.108e+02, percent-clipped=5.0 2023-03-09 15:25:26,532 INFO [train2.py:809] (3/4) Epoch 30, batch 3250, loss[ctc_loss=0.04172, att_loss=0.2129, loss=0.1787, over 16169.00 frames. utt_duration=1579 frames, utt_pad_proportion=0.006918, over 41.00 utterances.], tot_loss[ctc_loss=0.0622, att_loss=0.2298, loss=0.1963, over 3282644.51 frames. utt_duration=1246 frames, utt_pad_proportion=0.05137, over 10554.62 utterances.], batch size: 41, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:25:28,417 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.5498, 2.5332, 2.6036, 2.4275, 2.5198, 2.4528, 2.5762, 2.0602], device='cuda:3'), covar=tensor([0.1274, 0.1668, 0.1664, 0.3117, 0.1658, 0.2483, 0.1635, 0.3228], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0219, 0.0231, 0.0282, 0.0196, 0.0293, 0.0217, 0.0237], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:26:13,965 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.9168, 5.2640, 4.8106, 5.3014, 4.6517, 5.0009, 5.3718, 5.1688], device='cuda:3'), covar=tensor([0.0637, 0.0311, 0.0804, 0.0389, 0.0433, 0.0309, 0.0226, 0.0206], device='cuda:3'), in_proj_covar=tensor([0.0408, 0.0344, 0.0385, 0.0389, 0.0342, 0.0249, 0.0324, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 15:26:39,463 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118824.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:26:46,842 INFO [train2.py:809] (3/4) Epoch 30, batch 3300, loss[ctc_loss=0.06221, att_loss=0.2119, loss=0.182, over 16258.00 frames. utt_duration=1514 frames, utt_pad_proportion=0.007213, over 43.00 utterances.], tot_loss[ctc_loss=0.06195, att_loss=0.2293, loss=0.1958, over 3260065.41 frames. utt_duration=1247 frames, utt_pad_proportion=0.05623, over 10471.17 utterances.], batch size: 43, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:26:57,146 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.1201, 4.2057, 4.2689, 4.3032, 4.7726, 4.2268, 4.2191, 2.6676], device='cuda:3'), covar=tensor([0.0397, 0.0502, 0.0471, 0.0401, 0.0718, 0.0340, 0.0473, 0.1559], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0236, 0.0230, 0.0247, 0.0390, 0.0204, 0.0223, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:27:31,864 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.9034, 6.1100, 5.6190, 5.7617, 5.8021, 5.2378, 5.5691, 5.1937], device='cuda:3'), covar=tensor([0.1302, 0.0945, 0.1016, 0.0925, 0.1079, 0.1523, 0.2363, 0.2549], device='cuda:3'), in_proj_covar=tensor([0.0565, 0.0644, 0.0494, 0.0481, 0.0457, 0.0481, 0.0647, 0.0549], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:27:39,412 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.286e+02 1.786e+02 1.999e+02 2.429e+02 6.941e+02, threshold=3.998e+02, percent-clipped=3.0 2023-03-09 15:27:55,370 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118872.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:27:57,154 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.8309, 2.4913, 2.6418, 3.2733, 3.0501, 3.2713, 2.6017, 2.3224], device='cuda:3'), covar=tensor([0.0785, 0.1696, 0.0939, 0.0878, 0.0972, 0.0536, 0.1374, 0.1748], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0219, 0.0185, 0.0229, 0.0239, 0.0195, 0.0204, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:28:06,047 INFO [train2.py:809] (3/4) Epoch 30, batch 3350, loss[ctc_loss=0.06102, att_loss=0.2352, loss=0.2003, over 17070.00 frames. utt_duration=1315 frames, utt_pad_proportion=0.007765, over 52.00 utterances.], tot_loss[ctc_loss=0.0622, att_loss=0.2298, loss=0.1962, over 3260520.69 frames. utt_duration=1212 frames, utt_pad_proportion=0.06578, over 10771.80 utterances.], batch size: 52, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:28:27,915 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4344, 3.0985, 3.6174, 2.9893, 3.5312, 4.5466, 4.3620, 3.3491], device='cuda:3'), covar=tensor([0.0398, 0.1604, 0.1342, 0.1381, 0.1135, 0.0803, 0.0604, 0.1214], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0252, 0.0295, 0.0222, 0.0277, 0.0387, 0.0278, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:28:34,710 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.6542, 3.2123, 3.8237, 3.2372, 3.7108, 4.7423, 4.6040, 3.5982], device='cuda:3'), covar=tensor([0.0359, 0.1517, 0.1140, 0.1218, 0.0944, 0.0809, 0.0536, 0.1031], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0252, 0.0295, 0.0222, 0.0277, 0.0388, 0.0278, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:29:20,620 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118925.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:29:26,690 INFO [train2.py:809] (3/4) Epoch 30, batch 3400, loss[ctc_loss=0.05595, att_loss=0.2261, loss=0.192, over 16274.00 frames. utt_duration=1515 frames, utt_pad_proportion=0.007737, over 43.00 utterances.], tot_loss[ctc_loss=0.06253, att_loss=0.2304, loss=0.1968, over 3269504.26 frames. utt_duration=1199 frames, utt_pad_proportion=0.06717, over 10920.01 utterances.], batch size: 43, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:29:53,490 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118945.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:30:19,284 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 1.939e+02 2.194e+02 2.759e+02 6.591e+02, threshold=4.388e+02, percent-clipped=2.0 2023-03-09 15:30:37,149 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118973.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:30:46,800 INFO [train2.py:809] (3/4) Epoch 30, batch 3450, loss[ctc_loss=0.06693, att_loss=0.2347, loss=0.2011, over 17445.00 frames. utt_duration=1013 frames, utt_pad_proportion=0.04382, over 69.00 utterances.], tot_loss[ctc_loss=0.06198, att_loss=0.2301, loss=0.1965, over 3271543.02 frames. utt_duration=1239 frames, utt_pad_proportion=0.05705, over 10578.76 utterances.], batch size: 69, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:30:57,734 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118985.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:31:00,749 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118987.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:31:09,823 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:31:09,877 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.7968, 6.0368, 5.5337, 5.7399, 5.6985, 5.1255, 5.4843, 5.1949], device='cuda:3'), covar=tensor([0.1359, 0.0918, 0.1007, 0.0926, 0.1124, 0.1690, 0.2380, 0.2309], device='cuda:3'), in_proj_covar=tensor([0.0563, 0.0642, 0.0494, 0.0480, 0.0459, 0.0482, 0.0649, 0.0549], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:32:07,446 INFO [train2.py:809] (3/4) Epoch 30, batch 3500, loss[ctc_loss=0.06928, att_loss=0.212, loss=0.1835, over 15498.00 frames. utt_duration=1723 frames, utt_pad_proportion=0.008993, over 36.00 utterances.], tot_loss[ctc_loss=0.06238, att_loss=0.2305, loss=0.1969, over 3280129.28 frames. utt_duration=1254 frames, utt_pad_proportion=0.05091, over 10475.53 utterances.], batch size: 36, lr: 3.55e-03, grad_scale: 8.0 2023-03-09 15:32:15,046 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119033.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:32:50,106 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-09 15:32:59,919 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.837e+02 2.167e+02 2.597e+02 5.563e+02, threshold=4.335e+02, percent-clipped=3.0 2023-03-09 15:33:20,417 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119075.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:33:27,089 INFO [train2.py:809] (3/4) Epoch 30, batch 3550, loss[ctc_loss=0.07703, att_loss=0.2467, loss=0.2127, over 17616.00 frames. utt_duration=1023 frames, utt_pad_proportion=0.03613, over 69.00 utterances.], tot_loss[ctc_loss=0.06304, att_loss=0.2311, loss=0.1975, over 3281202.29 frames. utt_duration=1257 frames, utt_pad_proportion=0.05041, over 10451.43 utterances.], batch size: 69, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:33:50,484 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:34:18,179 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.2412, 3.9595, 3.8942, 3.3801, 3.8861, 3.9297, 3.9266, 3.0011], device='cuda:3'), covar=tensor([0.0914, 0.0852, 0.1811, 0.2580, 0.1160, 0.2022, 0.0794, 0.2876], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0221, 0.0234, 0.0285, 0.0198, 0.0294, 0.0218, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2023-03-09 15:34:46,595 INFO [train2.py:809] (3/4) Epoch 30, batch 3600, loss[ctc_loss=0.0748, att_loss=0.2488, loss=0.214, over 17060.00 frames. utt_duration=1314 frames, utt_pad_proportion=0.008287, over 52.00 utterances.], tot_loss[ctc_loss=0.06425, att_loss=0.2313, loss=0.1979, over 3276656.63 frames. utt_duration=1241 frames, utt_pad_proportion=0.05542, over 10575.55 utterances.], batch size: 52, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:34:58,007 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119136.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:35:00,915 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.1184, 5.4287, 5.0527, 5.4409, 4.8781, 5.1283, 5.5700, 5.3525], device='cuda:3'), covar=tensor([0.0619, 0.0306, 0.0705, 0.0364, 0.0374, 0.0198, 0.0234, 0.0198], device='cuda:3'), in_proj_covar=tensor([0.0405, 0.0342, 0.0380, 0.0385, 0.0339, 0.0245, 0.0320, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-09 15:35:07,911 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.8902, 4.5796, 4.5252, 2.3432, 2.1270, 2.8064, 2.3492, 3.8156], device='cuda:3'), covar=tensor([0.0746, 0.0329, 0.0301, 0.4487, 0.5092, 0.2508, 0.3619, 0.1358], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0312, 0.0284, 0.0257, 0.0343, 0.0337, 0.0267, 0.0376], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2023-03-09 15:35:26,399 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119154.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 15:35:38,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 1.787e+02 2.179e+02 2.834e+02 1.118e+03, threshold=4.357e+02, percent-clipped=5.0 2023-03-09 15:35:54,612 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([4.4069, 2.5986, 3.6193, 2.8293, 3.4607, 4.5613, 4.4527, 3.0430], device='cuda:3'), covar=tensor([0.0499, 0.2265, 0.1091, 0.1671, 0.1068, 0.1006, 0.0547, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0253, 0.0296, 0.0222, 0.0278, 0.0390, 0.0279, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003], device='cuda:3') 2023-03-09 15:36:05,601 INFO [train2.py:809] (3/4) Epoch 30, batch 3650, loss[ctc_loss=0.07069, att_loss=0.2392, loss=0.2055, over 17320.00 frames. utt_duration=1176 frames, utt_pad_proportion=0.0216, over 59.00 utterances.], tot_loss[ctc_loss=0.06402, att_loss=0.2314, loss=0.1979, over 3279494.63 frames. utt_duration=1247 frames, utt_pad_proportion=0.05268, over 10528.50 utterances.], batch size: 59, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:36:45,719 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([2.9323, 2.5785, 2.9391, 2.8840, 3.2223, 3.0064, 2.5670, 3.1285], device='cuda:3'), covar=tensor([0.1462, 0.2156, 0.1411, 0.1191, 0.1414, 0.0921, 0.1692, 0.1542], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0149, 0.0147, 0.0145, 0.0158, 0.0137, 0.0158, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-09 15:36:52,585 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-09 15:37:24,888 INFO [train2.py:809] (3/4) Epoch 30, batch 3700, loss[ctc_loss=0.04861, att_loss=0.2115, loss=0.1789, over 16005.00 frames. utt_duration=1602 frames, utt_pad_proportion=0.006944, over 40.00 utterances.], tot_loss[ctc_loss=0.06391, att_loss=0.231, loss=0.1976, over 3269551.74 frames. utt_duration=1241 frames, utt_pad_proportion=0.05579, over 10552.73 utterances.], batch size: 40, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:38:04,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.24 vs. limit=5.0 2023-03-09 15:38:18,047 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.417e+01 1.774e+02 2.178e+02 2.630e+02 5.010e+02, threshold=4.357e+02, percent-clipped=3.0 2023-03-09 15:38:30,110 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.76 vs. limit=5.0 2023-03-09 15:38:30,949 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([5.2260, 5.4719, 5.4211, 5.4287, 5.5319, 5.4710, 5.1466, 4.9710], device='cuda:3'), covar=tensor([0.0922, 0.0488, 0.0250, 0.0391, 0.0238, 0.0305, 0.0381, 0.0280], device='cuda:3'), in_proj_covar=tensor([0.0540, 0.0384, 0.0381, 0.0386, 0.0450, 0.0456, 0.0384, 0.0423], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:38:45,465 INFO [train2.py:809] (3/4) Epoch 30, batch 3750, loss[ctc_loss=0.04609, att_loss=0.2111, loss=0.1781, over 15625.00 frames. utt_duration=1691 frames, utt_pad_proportion=0.009405, over 37.00 utterances.], tot_loss[ctc_loss=0.06307, att_loss=0.2301, loss=0.1967, over 3258211.73 frames. utt_duration=1233 frames, utt_pad_proportion=0.06247, over 10582.32 utterances.], batch size: 37, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:38:59,260 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119287.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:40:05,290 INFO [train2.py:809] (3/4) Epoch 30, batch 3800, loss[ctc_loss=0.06396, att_loss=0.2393, loss=0.2042, over 16974.00 frames. utt_duration=1359 frames, utt_pad_proportion=0.007188, over 50.00 utterances.], tot_loss[ctc_loss=0.06245, att_loss=0.2297, loss=0.1963, over 3261344.27 frames. utt_duration=1245 frames, utt_pad_proportion=0.05994, over 10491.62 utterances.], batch size: 50, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:40:15,897 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119335.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:40:20,319 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([6.1868, 6.3479, 5.8364, 6.0725, 6.0824, 5.3317, 5.7627, 5.4627], device='cuda:3'), covar=tensor([0.1180, 0.0873, 0.0892, 0.0768, 0.0921, 0.1591, 0.2190, 0.2251], device='cuda:3'), in_proj_covar=tensor([0.0562, 0.0645, 0.0494, 0.0480, 0.0456, 0.0481, 0.0648, 0.0549], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-09 15:40:58,014 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.731e+02 2.132e+02 2.565e+02 6.876e+02, threshold=4.265e+02, percent-clipped=4.0 2023-03-09 15:41:06,001 INFO [zipformer.py:1447] (3/4) attn_weights_entropy = tensor([3.3032, 3.9577, 3.3505, 3.6626, 4.1536, 3.8659, 3.3733, 4.4565], device='cuda:3'), covar=tensor([0.0971, 0.0561, 0.1108, 0.0727, 0.0743, 0.0724, 0.0785, 0.0466], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0229, 0.0234, 0.0212, 0.0294, 0.0254, 0.0209, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:3') 2023-03-09 15:41:23,135 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-09 15:41:24,908 INFO [train2.py:809] (3/4) Epoch 30, batch 3850, loss[ctc_loss=0.03884, att_loss=0.2058, loss=0.1724, over 15752.00 frames. utt_duration=1659 frames, utt_pad_proportion=0.009939, over 38.00 utterances.], tot_loss[ctc_loss=0.0621, att_loss=0.2291, loss=0.1957, over 3253777.67 frames. utt_duration=1254 frames, utt_pad_proportion=0.05941, over 10389.45 utterances.], batch size: 38, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:42:12,364 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-09 15:42:43,262 INFO [train2.py:809] (3/4) Epoch 30, batch 3900, loss[ctc_loss=0.06646, att_loss=0.2427, loss=0.2074, over 16883.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006832, over 49.00 utterances.], tot_loss[ctc_loss=0.06264, att_loss=0.2292, loss=0.1959, over 3253237.63 frames. utt_duration=1253 frames, utt_pad_proportion=0.05946, over 10394.88 utterances.], batch size: 49, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:42:46,467 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119431.0, num_to_drop=0, layers_to_drop=set() 2023-03-09 15:43:14,043 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119449.0, num_to_drop=1, layers_to_drop={2} 2023-03-09 15:43:33,708 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.815e+02 2.229e+02 2.704e+02 7.416e+02, threshold=4.458e+02, percent-clipped=4.0 2023-03-09 15:43:59,617 INFO [train2.py:809] (3/4) Epoch 30, batch 3950, loss[ctc_loss=0.05213, att_loss=0.2321, loss=0.1961, over 16888.00 frames. utt_duration=1380 frames, utt_pad_proportion=0.006435, over 49.00 utterances.], tot_loss[ctc_loss=0.06295, att_loss=0.2292, loss=0.1959, over 3250817.26 frames. utt_duration=1230 frames, utt_pad_proportion=0.06479, over 10581.22 utterances.], batch size: 49, lr: 3.54e-03, grad_scale: 8.0 2023-03-09 15:44:52,254 INFO [train2.py:1037] (3/4) Done!