diff --git "a/log/log-train-2022-11-15-13-11-38-3" "b/log/log-train-2022-11-15-13-11-38-3" new file mode 100644--- /dev/null +++ "b/log/log-train-2022-11-15-13-11-38-3" @@ -0,0 +1,13830 @@ +2022-11-15 13:11:38,145 INFO [train.py:944] (3/4) Training started +2022-11-15 13:11:38,146 INFO [train.py:954] (3/4) Device: cuda:3 +2022-11-15 13:11:38,149 INFO [train.py:963] (3/4) {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 100, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.21', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': 'f271e82ef30f75fecbae44b163e1244e53def116', 'k2-git-date': 'Fri Oct 28 05:02:16 2022', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu111', 'torch-cuda-available': True, 'torch-cuda-version': '11.1', 'python-version': '3.8', 'icefall-git-branch': 'ami', 'icefall-git-sha1': '65f14ba-dirty', 'icefall-git-date': 'Mon Nov 14 18:45:09 2022', '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'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 15, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp/v2'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 5000, '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', 'decoder_dim': 512, 'joiner_dim': 512, 'manifest_dir': PosixPath('data/manifests'), 'enable_musan': True, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'max_duration': 120, 'num_buckets': 50, 'on_the_fly_feats': False, 'shuffle': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'blank_id': 0, 'vocab_size': 500} +2022-11-15 13:11:38,149 INFO [train.py:965] (3/4) About to create model +2022-11-15 13:11:38,544 INFO [zipformer.py:176] (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. +2022-11-15 13:11:38,555 INFO [train.py:969] (3/4) Number of model parameters: 70369391 +2022-11-15 13:11:43,057 INFO [train.py:984] (3/4) Using DDP +2022-11-15 13:11:43,611 INFO [asr_datamodule.py:353] (3/4) About to get AMI train cuts +2022-11-15 13:11:43,615 INFO [asr_datamodule.py:201] (3/4) About to get Musan cuts +2022-11-15 13:11:45,145 INFO [asr_datamodule.py:206] (3/4) Enable MUSAN +2022-11-15 13:11:45,145 INFO [asr_datamodule.py:229] (3/4) Enable SpecAugment +2022-11-15 13:11:45,145 INFO [asr_datamodule.py:230] (3/4) Time warp factor: 80 +2022-11-15 13:11:45,145 INFO [asr_datamodule.py:243] (3/4) About to create train dataset +2022-11-15 13:11:45,145 INFO [asr_datamodule.py:256] (3/4) Using DynamicBucketingSampler. +2022-11-15 13:11:45,521 INFO [asr_datamodule.py:264] (3/4) About to create train dataloader +2022-11-15 13:11:45,522 INFO [asr_datamodule.py:385] (3/4) About to get AMI IHM dev cuts +2022-11-15 13:11:45,523 INFO [asr_datamodule.py:296] (3/4) About to create dev dataset +2022-11-15 13:11:45,881 INFO [asr_datamodule.py:311] (3/4) About to create dev dataloader +2022-11-15 13:12:20,646 INFO [train.py:876] (3/4) Epoch 1, batch 0, loss[loss=3.807, simple_loss=3.443, pruned_loss=3.639, over 4675.00 frames. ], tot_loss[loss=3.807, simple_loss=3.443, pruned_loss=3.639, over 4675.00 frames. ], batch size: 135, lr: 2.50e-02, grad_scale: 2.0 +2022-11-15 13:12:20,647 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 13:12:37,307 INFO [train.py:908] (3/4) Epoch 1, validation: loss=3.424, simple_loss=3.08, pruned_loss=3.435, over 1530663.00 frames. +2022-11-15 13:12:37,341 INFO [train.py:909] (3/4) Maximum memory allocated so far is 2606MB +2022-11-15 13:12:39,699 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:12:50,433 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:12:54,120 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=7.94 vs. limit=2.0 +2022-11-15 13:12:56,899 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=5.23 vs. limit=2.0 +2022-11-15 13:13:09,658 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=151.84 vs. limit=5.0 +2022-11-15 13:13:21,352 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=99.60 vs. limit=5.0 +2022-11-15 13:13:23,360 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:13:32,522 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.298e+01 5.514e+01 1.134e+02 1.922e+02 2.006e+03, threshold=2.268e+02, percent-clipped=0.0 +2022-11-15 13:13:32,568 INFO [train.py:876] (3/4) Epoch 1, batch 100, loss[loss=0.4422, simple_loss=0.3825, pruned_loss=0.4769, over 5708.00 frames. ], tot_loss[loss=0.7321, simple_loss=0.6577, pruned_loss=0.6768, over 429889.74 frames. ], batch size: 12, lr: 3.00e-02, grad_scale: 2.0 +2022-11-15 13:13:57,871 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:14:15,117 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=4.16 vs. limit=2.0 +2022-11-15 13:14:29,149 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.50 vs. limit=2.0 +2022-11-15 13:14:31,594 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+01 2.547e+01 3.263e+01 4.120e+01 1.011e+02, threshold=6.525e+01, percent-clipped=0.0 +2022-11-15 13:14:31,634 INFO [train.py:876] (3/4) Epoch 1, batch 200, loss[loss=0.4098, simple_loss=0.3457, pruned_loss=0.4277, over 4956.00 frames. ], tot_loss[loss=0.5524, simple_loss=0.4869, pruned_loss=0.5307, over 690825.08 frames. ], batch size: 5, lr: 3.50e-02, grad_scale: 2.0 +2022-11-15 13:15:11,486 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=16.70 vs. limit=5.0 +2022-11-15 13:15:24,767 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:15:27,879 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:15:28,297 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.173e+01 3.095e+01 4.109e+01 5.560e+01 3.461e+02, threshold=8.218e+01, percent-clipped=17.0 +2022-11-15 13:15:28,336 INFO [train.py:876] (3/4) Epoch 1, batch 300, loss[loss=0.3779, simple_loss=0.3165, pruned_loss=0.3619, over 5550.00 frames. ], tot_loss[loss=0.4826, simple_loss=0.4192, pruned_loss=0.4598, over 852566.11 frames. ], batch size: 13, lr: 4.00e-02, grad_scale: 2.0 +2022-11-15 13:15:32,673 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.73 vs. limit=5.0 +2022-11-15 13:15:37,506 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=11.65 vs. limit=5.0 +2022-11-15 13:15:52,825 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.52 vs. limit=2.0 +2022-11-15 13:16:00,393 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={0, 3} +2022-11-15 13:16:15,066 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4089, 5.4312, 5.3348, 5.4339, 5.4272, 5.4357, 5.3826, 5.4223], + device='cuda:3'), covar=tensor([0.0104, 0.0020, 0.0169, 0.0016, 0.0058, 0.0037, 0.0134, 0.0103], + 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([9.3216e-06, 9.4421e-06, 9.0540e-06, 9.2196e-06, 9.4013e-06, 9.1921e-06, + 9.2590e-06, 9.3068e-06], device='cuda:3') +2022-11-15 13:16:16,648 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:16:24,455 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.372e+01 3.559e+01 4.600e+01 6.651e+01 2.649e+02, threshold=9.199e+01, percent-clipped=13.0 +2022-11-15 13:16:24,498 INFO [train.py:876] (3/4) Epoch 1, batch 400, loss[loss=0.4022, simple_loss=0.3332, pruned_loss=0.3653, over 5629.00 frames. ], tot_loss[loss=0.4469, simple_loss=0.3823, pruned_loss=0.4185, over 949681.73 frames. ], batch size: 32, lr: 4.50e-02, grad_scale: 4.0 +2022-11-15 13:16:31,017 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9638, 3.9646, 3.9286, 3.9599, 3.9649, 3.9659, 3.9637, 3.9661], + device='cuda:3'), covar=tensor([0.0041, 0.0031, 0.0040, 0.0045, 0.0050, 0.0040, 0.0048, 0.0061], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0008, 0.0009, 0.0008, 0.0008, 0.0008], + device='cuda:3'), out_proj_covar=tensor([8.3222e-06, 8.3313e-06, 8.6680e-06, 8.2666e-06, 8.4756e-06, 8.6287e-06, + 8.5419e-06, 8.5310e-06], device='cuda:3') +2022-11-15 13:16:33,554 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.6246, 5.6226, 5.5902, 5.6077, 5.6275, 5.6257, 5.6299, 5.6163], + device='cuda:3'), covar=tensor([0.0026, 0.0045, 0.0033, 0.0028, 0.0025, 0.0033, 0.0023, 0.0032], + 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([8.7919e-06, 8.8546e-06, 8.5461e-06, 8.6159e-06, 8.7312e-06, 8.6516e-06, + 8.6562e-06, 8.8514e-06], device='cuda:3') +2022-11-15 13:16:46,793 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:16:51,926 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={0, 3} +2022-11-15 13:17:22,361 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.467e+01 3.366e+01 4.397e+01 5.953e+01 8.256e+02, threshold=8.794e+01, percent-clipped=9.0 +2022-11-15 13:17:22,404 INFO [train.py:876] (3/4) Epoch 1, batch 500, loss[loss=0.4116, simple_loss=0.3324, pruned_loss=0.3714, over 5719.00 frames. ], tot_loss[loss=0.4293, simple_loss=0.362, pruned_loss=0.3924, over 1004588.07 frames. ], batch size: 19, lr: 4.99e-02, grad_scale: 4.0 +2022-11-15 13:17:32,884 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2022-11-15 13:17:50,497 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.97 vs. limit=2.0 +2022-11-15 13:17:57,339 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:18:00,626 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:18:13,303 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=5.58 vs. limit=2.0 +2022-11-15 13:18:14,155 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:18:20,261 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:18:20,636 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.558e+01 3.991e+01 4.815e+01 6.242e+01 3.465e+02, threshold=9.630e+01, percent-clipped=11.0 +2022-11-15 13:18:20,677 INFO [train.py:876] (3/4) Epoch 1, batch 600, loss[loss=0.4434, simple_loss=0.3569, pruned_loss=0.3784, over 5755.00 frames. ], tot_loss[loss=0.4227, simple_loss=0.3514, pruned_loss=0.376, over 1039983.16 frames. ], batch size: 20, lr: 4.98e-02, grad_scale: 4.0 +2022-11-15 13:18:31,897 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1612, 3.3114, 3.5118, 3.6645, 3.7575, 3.8331, 3.3096, 3.3574], + device='cuda:3'), covar=tensor([0.2617, 0.1696, 0.1840, 0.1523, 0.1261, 0.0709, 0.1225, 0.1610], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0016, 0.0015, 0.0015, 0.0013, 0.0014, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.4578e-05, 1.3072e-05, 1.4392e-05, 1.3398e-05, 1.3315e-05, 1.2345e-05, + 1.2806e-05, 1.3804e-05], device='cuda:3') +2022-11-15 13:18:32,999 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={1, 3} +2022-11-15 13:18:36,268 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:18:47,939 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:18:49,677 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:18:50,156 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:19:02,082 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.46 vs. limit=2.0 +2022-11-15 13:19:18,392 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.943e+01 4.768e+01 6.411e+01 9.551e+01 4.417e+02, threshold=1.282e+02, percent-clipped=24.0 +2022-11-15 13:19:18,439 INFO [train.py:876] (3/4) Epoch 1, batch 700, loss[loss=0.422, simple_loss=0.3389, pruned_loss=0.3428, over 5639.00 frames. ], tot_loss[loss=0.4125, simple_loss=0.3385, pruned_loss=0.3576, over 1047100.18 frames. ], batch size: 38, lr: 4.98e-02, grad_scale: 4.0 +2022-11-15 13:19:39,850 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:19:42,030 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:19:55,241 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.45 vs. limit=5.0 +2022-11-15 13:20:07,235 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:20:09,914 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6939, 2.8180, 2.7877, 2.8031, 2.6745, 2.8252, 2.7732, 2.4191], + device='cuda:3'), covar=tensor([0.7067, 0.4885, 0.7065, 0.4028, 0.6557, 0.6042, 0.4789, 1.0745], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0019, 0.0020, 0.0019, 0.0019, 0.0020, 0.0019, 0.0021], + device='cuda:3'), out_proj_covar=tensor([1.9452e-05, 1.8611e-05, 1.9521e-05, 1.8798e-05, 1.9065e-05, 1.9562e-05, + 1.8071e-05, 2.2082e-05], device='cuda:3') +2022-11-15 13:20:15,194 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.610e+01 5.331e+01 6.709e+01 8.250e+01 2.326e+02, threshold=1.342e+02, percent-clipped=8.0 +2022-11-15 13:20:15,236 INFO [train.py:876] (3/4) Epoch 1, batch 800, loss[loss=0.284, simple_loss=0.2279, pruned_loss=0.2211, over 5061.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.3318, pruned_loss=0.3448, over 1063040.92 frames. ], batch size: 7, lr: 4.97e-02, grad_scale: 8.0 +2022-11-15 13:20:42,194 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:21:13,723 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.878e+01 6.865e+01 9.860e+01 1.395e+02 3.183e+02, threshold=1.972e+02, percent-clipped=28.0 +2022-11-15 13:21:13,767 INFO [train.py:876] (3/4) Epoch 1, batch 900, loss[loss=0.3762, simple_loss=0.3001, pruned_loss=0.2843, over 5008.00 frames. ], tot_loss[loss=0.4038, simple_loss=0.3248, pruned_loss=0.3294, over 1075095.13 frames. ], batch size: 109, lr: 4.96e-02, grad_scale: 8.0 +2022-11-15 13:21:17,936 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:21:24,319 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:21:28,508 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:21:40,752 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:21:44,162 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:21:49,468 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.95 vs. limit=2.0 +2022-11-15 13:22:05,027 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.22 vs. limit=2.0 +2022-11-15 13:22:08,809 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.19 vs. limit=2.0 +2022-11-15 13:22:12,363 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:22:12,818 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.013e+01 8.825e+01 1.129e+02 1.516e+02 3.115e+02, threshold=2.258e+02, percent-clipped=11.0 +2022-11-15 13:22:12,861 INFO [train.py:876] (3/4) Epoch 1, batch 1000, loss[loss=0.4202, simple_loss=0.3297, pruned_loss=0.3143, over 5566.00 frames. ], tot_loss[loss=0.3988, simple_loss=0.3186, pruned_loss=0.3153, over 1081490.49 frames. ], batch size: 46, lr: 4.95e-02, grad_scale: 8.0 +2022-11-15 13:22:29,005 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 +2022-11-15 13:22:37,957 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:22:55,057 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.09 vs. limit=5.0 +2022-11-15 13:23:01,386 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2022-11-15 13:23:05,512 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:23:11,682 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.943e+01 8.541e+01 1.080e+02 1.456e+02 2.899e+02, threshold=2.160e+02, percent-clipped=5.0 +2022-11-15 13:23:11,724 INFO [train.py:876] (3/4) Epoch 1, batch 1100, loss[loss=0.3901, simple_loss=0.3067, pruned_loss=0.2817, over 5771.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.3137, pruned_loss=0.3022, over 1085880.04 frames. ], batch size: 21, lr: 4.94e-02, grad_scale: 8.0 +2022-11-15 13:23:42,086 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 13:23:42,410 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:24:09,650 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.854e+01 1.222e+02 1.602e+02 1.992e+02 6.604e+02, threshold=3.204e+02, percent-clipped=19.0 +2022-11-15 13:24:09,696 INFO [train.py:876] (3/4) Epoch 1, batch 1200, loss[loss=0.4626, simple_loss=0.3634, pruned_loss=0.3252, over 5565.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.3076, pruned_loss=0.2885, over 1083962.05 frames. ], batch size: 40, lr: 4.93e-02, grad_scale: 8.0 +2022-11-15 13:24:10,875 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:24:17,638 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:24:20,302 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:24:20,522 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 13:24:23,568 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:24:36,270 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:24:48,468 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:24:51,637 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:24:57,202 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8406, 3.0964, 3.1238, 3.0119, 3.0066, 2.8879, 2.7366, 2.9327], + device='cuda:3'), covar=tensor([0.3531, 0.3573, 0.2497, 0.2648, 0.3180, 0.3288, 0.2450, 0.2778], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0028, 0.0029, 0.0029, 0.0031, 0.0031, 0.0025, 0.0027], + device='cuda:3'), out_proj_covar=tensor([2.8830e-05, 2.7967e-05, 2.6500e-05, 2.7225e-05, 2.8931e-05, 3.0335e-05, + 2.3676e-05, 2.7985e-05], device='cuda:3') +2022-11-15 13:25:04,718 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:25:08,463 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.439e+01 1.113e+02 1.512e+02 2.067e+02 6.699e+02, threshold=3.023e+02, percent-clipped=4.0 +2022-11-15 13:25:08,506 INFO [train.py:876] (3/4) Epoch 1, batch 1300, loss[loss=0.3449, simple_loss=0.2682, pruned_loss=0.2392, over 5737.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.3038, pruned_loss=0.2785, over 1081114.75 frames. ], batch size: 31, lr: 4.92e-02, grad_scale: 8.0 +2022-11-15 13:25:26,340 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.15 vs. limit=2.0 +2022-11-15 13:25:29,759 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.44 vs. limit=5.0 +2022-11-15 13:25:31,328 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 +2022-11-15 13:25:31,393 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.62 vs. limit=5.0 +2022-11-15 13:25:43,904 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7381, 4.1480, 3.4527, 3.5887, 3.8811, 3.9223, 4.0501, 3.7248], + device='cuda:3'), covar=tensor([0.2779, 0.1657, 0.2848, 0.2404, 0.3054, 0.2282, 0.1831, 0.2171], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0016, 0.0015, 0.0015, 0.0016, 0.0014, 0.0015, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.2327e-05, 1.2733e-05, 1.1440e-05, 1.1482e-05, 1.2729e-05, 1.1224e-05, + 1.2085e-05, 1.0482e-05], device='cuda:3') +2022-11-15 13:25:50,711 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4373, 3.3197, 3.3791, 3.3274, 3.1847, 3.2411, 3.5975, 3.4477], + device='cuda:3'), covar=tensor([0.1399, 0.1909, 0.2262, 0.1847, 0.2241, 0.2244, 0.1193, 0.1731], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0030, 0.0031, 0.0031, 0.0033, 0.0029, 0.0030, 0.0030], + device='cuda:3'), out_proj_covar=tensor([2.3155e-05, 2.4291e-05, 2.8501e-05, 2.6222e-05, 3.0702e-05, 2.3777e-05, + 2.4387e-05, 2.6598e-05], device='cuda:3') +2022-11-15 13:26:09,308 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.007e+01 1.313e+02 1.838e+02 2.468e+02 4.304e+02, threshold=3.675e+02, percent-clipped=9.0 +2022-11-15 13:26:09,349 INFO [train.py:876] (3/4) Epoch 1, batch 1400, loss[loss=0.3378, simple_loss=0.2782, pruned_loss=0.2149, over 5706.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.2951, pruned_loss=0.2636, over 1088622.26 frames. ], batch size: 12, lr: 4.91e-02, grad_scale: 8.0 +2022-11-15 13:26:27,747 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1802, 2.1736, 2.0602, 1.9215, 1.9857, 2.0917, 2.1797, 2.1211], + device='cuda:3'), covar=tensor([0.1678, 0.1645, 0.2321, 0.2768, 0.2484, 0.1886, 0.1341, 0.2082], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0030, 0.0032, 0.0032, 0.0032, 0.0029, 0.0030, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.3940e-05, 2.4445e-05, 2.8973e-05, 2.6782e-05, 2.9029e-05, 2.3451e-05, + 2.4610e-05, 2.7932e-05], device='cuda:3') +2022-11-15 13:26:29,284 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:26:35,289 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.07 vs. limit=5.0 +2022-11-15 13:27:05,769 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:27:09,430 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.161e+01 1.360e+02 1.948e+02 2.853e+02 7.338e+02, threshold=3.896e+02, percent-clipped=10.0 +2022-11-15 13:27:09,473 INFO [train.py:876] (3/4) Epoch 1, batch 1500, loss[loss=0.3869, simple_loss=0.3032, pruned_loss=0.2546, over 5749.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.2902, pruned_loss=0.254, over 1088124.86 frames. ], batch size: 31, lr: 4.89e-02, grad_scale: 8.0 +2022-11-15 13:27:10,787 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:27:14,379 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:27:23,273 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.74 vs. limit=5.0 +2022-11-15 13:27:40,013 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:27:49,364 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:28:05,698 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.20 vs. limit=5.0 +2022-11-15 13:28:10,465 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.61 vs. limit=2.0 +2022-11-15 13:28:10,526 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.692e+01 1.557e+02 1.959e+02 2.401e+02 7.122e+02, threshold=3.919e+02, percent-clipped=3.0 +2022-11-15 13:28:10,569 INFO [train.py:876] (3/4) Epoch 1, batch 1600, loss[loss=0.441, simple_loss=0.3261, pruned_loss=0.2992, over 5469.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.288, pruned_loss=0.2464, over 1090786.78 frames. ], batch size: 64, lr: 4.88e-02, grad_scale: 8.0 +2022-11-15 13:28:12,182 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1102, 4.0445, 4.2256, 3.8876, 4.0301, 4.2096, 3.7742, 4.3021], + device='cuda:3'), covar=tensor([0.2070, 0.1867, 0.1024, 0.2369, 0.1601, 0.1216, 0.1134, 0.1352], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0024, 0.0027, 0.0027, 0.0026, 0.0022, 0.0022], + device='cuda:3'), out_proj_covar=tensor([2.4331e-05, 2.3102e-05, 2.1066e-05, 2.5341e-05, 2.4016e-05, 2.4739e-05, + 2.0738e-05, 2.1567e-05], device='cuda:3') +2022-11-15 13:28:17,288 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:28:26,875 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:28:41,331 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 13:28:43,675 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.40 vs. limit=5.0 +2022-11-15 13:28:54,813 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:28:55,431 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6315, 3.6165, 3.2983, 3.5750, 3.8708, 3.7252, 3.1847, 3.6477], + device='cuda:3'), covar=tensor([0.1465, 0.1261, 0.1811, 0.1363, 0.0796, 0.1407, 0.2146, 0.1623], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0026, 0.0027, 0.0026, 0.0022, 0.0023, 0.0027, 0.0026], + device='cuda:3'), out_proj_covar=tensor([2.0539e-05, 2.2190e-05, 2.5335e-05, 2.2586e-05, 2.0421e-05, 1.9885e-05, + 2.3637e-05, 2.3638e-05], device='cuda:3') +2022-11-15 13:29:00,446 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 +2022-11-15 13:29:11,724 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.804e+01 1.584e+02 2.065e+02 2.557e+02 6.153e+02, threshold=4.130e+02, percent-clipped=6.0 +2022-11-15 13:29:11,766 INFO [train.py:876] (3/4) Epoch 1, batch 1700, loss[loss=0.2614, simple_loss=0.2144, pruned_loss=0.1596, over 5722.00 frames. ], tot_loss[loss=0.3632, simple_loss=0.2851, pruned_loss=0.2401, over 1093930.08 frames. ], batch size: 11, lr: 4.86e-02, grad_scale: 8.0 +2022-11-15 13:29:32,195 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 13:30:07,143 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 13:30:14,465 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.349e+01 1.628e+02 2.392e+02 3.072e+02 5.496e+02, threshold=4.784e+02, percent-clipped=8.0 +2022-11-15 13:30:14,508 INFO [train.py:876] (3/4) Epoch 1, batch 1800, loss[loss=0.3233, simple_loss=0.2494, pruned_loss=0.2045, over 5545.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.2833, pruned_loss=0.2353, over 1087295.10 frames. ], batch size: 30, lr: 4.85e-02, grad_scale: 8.0 +2022-11-15 13:30:19,414 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 13:30:37,487 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:30:47,367 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 13:30:49,385 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:30:49,477 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:30:57,352 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:31:16,025 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={1, 3} +2022-11-15 13:31:17,086 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.491e+01 1.765e+02 2.160e+02 2.889e+02 5.463e+02, threshold=4.319e+02, percent-clipped=1.0 +2022-11-15 13:31:17,126 INFO [train.py:876] (3/4) Epoch 1, batch 1900, loss[loss=0.4395, simple_loss=0.3263, pruned_loss=0.2808, over 5396.00 frames. ], tot_loss[loss=0.36, simple_loss=0.2813, pruned_loss=0.2301, over 1089270.84 frames. ], batch size: 70, lr: 4.83e-02, grad_scale: 8.0 +2022-11-15 13:31:27,851 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:31:30,129 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 13:31:33,730 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2988, 4.0428, 4.4042, 4.2699, 4.2525, 4.0986, 3.9132, 4.4763], + device='cuda:3'), covar=tensor([0.0208, 0.0269, 0.0203, 0.0333, 0.0312, 0.0207, 0.0388, 0.0235], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0033, 0.0031, 0.0035, 0.0034, 0.0034, 0.0038, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.4435e-05, 2.7301e-05, 2.5622e-05, 2.8563e-05, 2.9748e-05, 2.7355e-05, + 3.2218e-05, 2.8717e-05], device='cuda:3') +2022-11-15 13:31:35,542 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={0, 3} +2022-11-15 13:31:50,424 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 +2022-11-15 13:31:52,197 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9417, 2.7370, 2.8915, 2.6708, 2.4348, 2.5877, 2.5398, 2.6685], + device='cuda:3'), covar=tensor([0.0668, 0.0826, 0.0836, 0.0965, 0.1221, 0.1157, 0.1162, 0.0796], + device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0033, 0.0034, 0.0035, 0.0036, 0.0036, 0.0036, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.7708e-05, 2.8274e-05, 3.1397e-05, 3.0599e-05, 2.9407e-05, 3.1742e-05, + 3.0583e-05, 2.8175e-05], device='cuda:3') +2022-11-15 13:31:58,481 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:32:04,268 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.95 vs. limit=5.0 +2022-11-15 13:32:19,572 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 1.971e+02 2.700e+02 3.319e+02 5.947e+02, threshold=5.400e+02, percent-clipped=8.0 +2022-11-15 13:32:19,615 INFO [train.py:876] (3/4) Epoch 1, batch 2000, loss[loss=0.4268, simple_loss=0.3244, pruned_loss=0.2646, over 5758.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.2771, pruned_loss=0.2231, over 1094150.79 frames. ], batch size: 21, lr: 4.82e-02, grad_scale: 16.0 +2022-11-15 13:33:20,328 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:33:27,173 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.484e+01 1.719e+02 2.410e+02 2.808e+02 6.250e+02, threshold=4.821e+02, percent-clipped=3.0 +2022-11-15 13:33:27,215 INFO [train.py:876] (3/4) Epoch 1, batch 2100, loss[loss=0.3307, simple_loss=0.2581, pruned_loss=0.2017, over 5286.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.2725, pruned_loss=0.2156, over 1086261.65 frames. ], batch size: 79, lr: 4.80e-02, grad_scale: 16.0 +2022-11-15 13:33:52,293 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:34:23,973 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.51 vs. limit=5.0 +2022-11-15 13:34:29,781 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:34:34,681 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.857e+02 2.320e+02 3.014e+02 5.745e+02, threshold=4.640e+02, percent-clipped=3.0 +2022-11-15 13:34:34,723 INFO [train.py:876] (3/4) Epoch 1, batch 2200, loss[loss=0.4235, simple_loss=0.32, pruned_loss=0.2635, over 5395.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.2685, pruned_loss=0.209, over 1089924.56 frames. ], batch size: 70, lr: 4.78e-02, grad_scale: 16.0 +2022-11-15 13:34:43,173 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:34:49,366 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:34:51,883 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:35:05,911 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.61 vs. limit=5.0 +2022-11-15 13:35:19,975 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 13:35:21,885 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:35:42,976 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.350e+01 1.693e+02 2.305e+02 3.135e+02 7.690e+02, threshold=4.610e+02, percent-clipped=6.0 +2022-11-15 13:35:43,019 INFO [train.py:876] (3/4) Epoch 1, batch 2300, loss[loss=0.3389, simple_loss=0.2661, pruned_loss=0.2059, over 5559.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.2676, pruned_loss=0.2064, over 1085980.31 frames. ], batch size: 14, lr: 4.77e-02, grad_scale: 16.0 +2022-11-15 13:35:43,931 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.30 vs. limit=5.0 +2022-11-15 13:35:48,156 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.40 vs. limit=5.0 +2022-11-15 13:35:52,629 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:35:56,839 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-15 13:36:08,381 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4760, 3.5282, 3.3427, 3.2722, 3.4156, 3.4495, 3.0809, 3.6551], + device='cuda:3'), covar=tensor([0.0251, 0.0259, 0.0366, 0.0370, 0.0403, 0.0331, 0.0581, 0.0328], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0022, 0.0024, 0.0022, 0.0023, 0.0023, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.0705e-05, 1.9928e-05, 2.2383e-05, 2.0663e-05, 2.1362e-05, 2.1932e-05, + 2.4805e-05, 2.2854e-05], device='cuda:3') +2022-11-15 13:36:09,637 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:36:33,079 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 13:36:33,609 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 +2022-11-15 13:36:40,314 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6292, 2.9502, 2.8241, 2.6960, 2.8211, 3.1011, 2.4239, 3.0908], + device='cuda:3'), covar=tensor([0.0873, 0.0547, 0.0347, 0.0511, 0.0467, 0.0506, 0.0987, 0.0376], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0026, 0.0024, 0.0026, 0.0027, 0.0025, 0.0028, 0.0023], + device='cuda:3'), out_proj_covar=tensor([2.7454e-05, 2.2780e-05, 1.9944e-05, 2.2554e-05, 2.3988e-05, 2.1114e-05, + 2.6742e-05, 2.0858e-05], device='cuda:3') +2022-11-15 13:36:46,353 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.80 vs. limit=5.0 +2022-11-15 13:36:51,173 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.270e+01 1.877e+02 2.321e+02 3.011e+02 5.507e+02, threshold=4.642e+02, percent-clipped=4.0 +2022-11-15 13:36:51,213 INFO [train.py:876] (3/4) Epoch 1, batch 2400, loss[loss=0.3923, simple_loss=0.3083, pruned_loss=0.2382, over 5643.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.2674, pruned_loss=0.2045, over 1088568.98 frames. ], batch size: 29, lr: 4.75e-02, grad_scale: 16.0 +2022-11-15 13:36:51,415 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:37:04,416 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0 +2022-11-15 13:37:43,537 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1943, 4.0486, 4.1237, 4.1609, 4.0265, 4.2391, 4.1020, 4.0729], + device='cuda:3'), covar=tensor([0.0171, 0.0168, 0.0161, 0.0185, 0.0160, 0.0162, 0.0140, 0.0155], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0025, 0.0024, 0.0024, 0.0024, 0.0027, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.4311e-05, 2.2748e-05, 2.4764e-05, 2.3275e-05, 2.3535e-05, 2.2816e-05, + 2.4845e-05, 2.3963e-05], device='cuda:3') +2022-11-15 13:37:54,124 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:37:58,565 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 2.130e+02 2.557e+02 3.291e+02 7.818e+02, threshold=5.113e+02, percent-clipped=6.0 +2022-11-15 13:37:58,608 INFO [train.py:876] (3/4) Epoch 1, batch 2500, loss[loss=0.3228, simple_loss=0.2643, pruned_loss=0.1906, over 5712.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.2652, pruned_loss=0.2011, over 1092044.22 frames. ], batch size: 28, lr: 4.73e-02, grad_scale: 16.0 +2022-11-15 13:38:06,788 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:38:15,179 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:38:19,886 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.70 vs. limit=5.0 +2022-11-15 13:38:25,943 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:38:38,840 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:38:47,655 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:38:54,796 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 +2022-11-15 13:39:01,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 13:39:06,571 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.872e+02 2.338e+02 3.221e+02 9.126e+02, threshold=4.676e+02, percent-clipped=6.0 +2022-11-15 13:39:06,613 INFO [train.py:876] (3/4) Epoch 1, batch 2600, loss[loss=0.2985, simple_loss=0.2638, pruned_loss=0.1666, over 5733.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.2642, pruned_loss=0.1983, over 1096185.79 frames. ], batch size: 17, lr: 4.71e-02, grad_scale: 16.0 +2022-11-15 13:40:04,063 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:40:11,217 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:40:14,323 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 1.859e+02 2.538e+02 3.417e+02 6.213e+02, threshold=5.075e+02, percent-clipped=6.0 +2022-11-15 13:40:14,363 INFO [train.py:876] (3/4) Epoch 1, batch 2700, loss[loss=0.3526, simple_loss=0.281, pruned_loss=0.2122, over 5774.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.2621, pruned_loss=0.1956, over 1093946.65 frames. ], batch size: 14, lr: 4.69e-02, grad_scale: 16.0 +2022-11-15 13:40:32,648 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 13:40:45,517 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:40:47,438 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9348, 3.4377, 3.1861, 3.4339, 3.2845, 2.7911, 3.0116, 2.9074], + device='cuda:3'), covar=tensor([0.0452, 0.0291, 0.0290, 0.0251, 0.0301, 0.0520, 0.0335, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0032, 0.0038, 0.0032, 0.0033, 0.0039, 0.0031, 0.0032], + device='cuda:3'), out_proj_covar=tensor([3.4165e-05, 3.5630e-05, 3.8992e-05, 3.4663e-05, 3.1539e-05, 4.1270e-05, + 3.1812e-05, 3.1530e-05], device='cuda:3') +2022-11-15 13:41:23,155 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.037e+02 2.481e+02 3.360e+02 6.352e+02, threshold=4.961e+02, percent-clipped=3.0 +2022-11-15 13:41:23,198 INFO [train.py:876] (3/4) Epoch 1, batch 2800, loss[loss=0.3219, simple_loss=0.2587, pruned_loss=0.1926, over 4967.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.2608, pruned_loss=0.1939, over 1088503.22 frames. ], batch size: 109, lr: 4.67e-02, grad_scale: 16.0 +2022-11-15 13:41:39,161 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:41:55,974 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5191, 3.5920, 3.6332, 3.8941, 3.9238, 3.3904, 2.5712, 3.7723], + device='cuda:3'), covar=tensor([0.0568, 0.0591, 0.0450, 0.0289, 0.0275, 0.0660, 0.3209, 0.0344], + device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0032, 0.0029, 0.0029, 0.0027, 0.0032, 0.0052, 0.0028], + device='cuda:3'), out_proj_covar=tensor([3.3349e-05, 2.9856e-05, 2.7002e-05, 2.4507e-05, 2.1480e-05, 2.7711e-05, + 6.2556e-05, 2.3706e-05], device='cuda:3') +2022-11-15 13:42:20,491 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:42:23,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2900, 3.5400, 2.6108, 3.1909, 3.0426, 3.5326, 3.4099, 3.3959], + device='cuda:3'), covar=tensor([0.0365, 0.0309, 0.0364, 0.0404, 0.0539, 0.0407, 0.0426, 0.0345], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0023, 0.0024, 0.0024, 0.0026, 0.0024, 0.0023, 0.0022], + device='cuda:3'), out_proj_covar=tensor([2.5587e-05, 2.1649e-05, 2.0803e-05, 2.1975e-05, 2.5058e-05, 2.1431e-05, + 2.3490e-05, 2.0391e-05], device='cuda:3') +2022-11-15 13:42:30,823 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 2.042e+02 2.577e+02 3.345e+02 6.665e+02, threshold=5.155e+02, percent-clipped=6.0 +2022-11-15 13:42:30,864 INFO [train.py:876] (3/4) Epoch 1, batch 2900, loss[loss=0.3463, simple_loss=0.2798, pruned_loss=0.2064, over 5570.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.2602, pruned_loss=0.1926, over 1089237.89 frames. ], batch size: 43, lr: 4.65e-02, grad_scale: 16.0 +2022-11-15 13:42:55,181 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:42:59,424 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6529, 3.8966, 3.5054, 3.4324, 3.6005, 3.6177, 2.8103, 3.8341], + device='cuda:3'), covar=tensor([0.0199, 0.0174, 0.0191, 0.0364, 0.0190, 0.0191, 0.0350, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0017, 0.0017, 0.0022, 0.0018, 0.0016, 0.0015, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.8996e-05, 1.8918e-05, 1.7507e-05, 2.4696e-05, 1.9498e-05, 1.7926e-05, + 1.7261e-05, 1.3892e-05], device='cuda:3') +2022-11-15 13:43:35,969 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 13:43:37,253 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={0, 1} +2022-11-15 13:43:39,748 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.318e+01 2.017e+02 2.613e+02 3.220e+02 7.122e+02, threshold=5.226e+02, percent-clipped=3.0 +2022-11-15 13:43:39,792 INFO [train.py:876] (3/4) Epoch 1, batch 3000, loss[loss=0.3352, simple_loss=0.2532, pruned_loss=0.2086, over 3133.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.2607, pruned_loss=0.193, over 1089164.03 frames. ], batch size: 284, lr: 4.63e-02, grad_scale: 16.0 +2022-11-15 13:43:39,792 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 13:43:58,880 INFO [train.py:908] (3/4) Epoch 1, validation: loss=0.2736, simple_loss=0.2548, pruned_loss=0.1462, over 1530663.00 frames. +2022-11-15 13:43:58,881 INFO [train.py:909] (3/4) Maximum memory allocated so far is 3955MB +2022-11-15 13:44:11,426 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:44:26,584 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:44:28,587 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:44:43,066 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.93 vs. limit=5.0 +2022-11-15 13:44:53,882 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={0, 3} +2022-11-15 13:44:57,365 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 13:45:08,266 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 2.068e+02 2.442e+02 3.389e+02 5.023e+02, threshold=4.884e+02, percent-clipped=1.0 +2022-11-15 13:45:08,307 INFO [train.py:876] (3/4) Epoch 1, batch 3100, loss[loss=0.3179, simple_loss=0.2413, pruned_loss=0.1972, over 5168.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.2587, pruned_loss=0.1898, over 1090222.33 frames. ], batch size: 91, lr: 4.61e-02, grad_scale: 16.0 +2022-11-15 13:45:17,467 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7151, 2.9320, 2.9713, 2.8795, 2.6574, 2.2736, 2.9346, 2.7499], + device='cuda:3'), covar=tensor([0.0479, 0.0348, 0.0286, 0.0394, 0.0471, 0.0636, 0.0285, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0022, 0.0026, 0.0026, 0.0026, 0.0025, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.9029e-05, 1.8223e-05, 1.7616e-05, 2.2942e-05, 2.3502e-05, 2.4705e-05, + 2.2081e-05, 2.2523e-05], device='cuda:3') +2022-11-15 13:45:33,586 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.51 vs. limit=5.0 +2022-11-15 13:45:34,313 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 13:46:02,908 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:46:17,615 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 2.233e+02 2.604e+02 3.182e+02 6.551e+02, threshold=5.207e+02, percent-clipped=6.0 +2022-11-15 13:46:17,656 INFO [train.py:876] (3/4) Epoch 1, batch 3200, loss[loss=0.2638, simple_loss=0.2314, pruned_loss=0.1481, over 5698.00 frames. ], tot_loss[loss=0.318, simple_loss=0.2581, pruned_loss=0.189, over 1083960.68 frames. ], batch size: 11, lr: 4.59e-02, grad_scale: 16.0 +2022-11-15 13:46:21,363 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4401, 2.1453, 2.4680, 1.9134, 2.2032, 2.5237, 1.9293, 2.1745], + device='cuda:3'), covar=tensor([0.0336, 0.0598, 0.0392, 0.0466, 0.0367, 0.0364, 0.0590, 0.0375], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0030, 0.0026, 0.0030, 0.0026, 0.0028, 0.0031, 0.0027], + device='cuda:3'), out_proj_covar=tensor([2.5629e-05, 2.8904e-05, 2.5196e-05, 2.7663e-05, 2.3334e-05, 2.6324e-05, + 2.8812e-05, 2.5280e-05], device='cuda:3') +2022-11-15 13:46:27,849 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 +2022-11-15 13:47:11,198 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7515, 3.4577, 3.0690, 3.3884, 3.3628, 3.1091, 2.9139, 2.9573], + device='cuda:3'), covar=tensor([0.0552, 0.0205, 0.0307, 0.0233, 0.0190, 0.0282, 0.0330, 0.0292], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0032, 0.0041, 0.0033, 0.0034, 0.0038, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([3.8592e-05, 3.6299e-05, 4.5270e-05, 3.6475e-05, 3.4107e-05, 4.2018e-05, + 3.4917e-05, 3.4439e-05], device='cuda:3') +2022-11-15 13:47:17,413 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:47:20,612 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:47:25,676 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.790e+01 2.210e+02 2.954e+02 4.251e+02 1.287e+03, threshold=5.908e+02, percent-clipped=13.0 +2022-11-15 13:47:25,721 INFO [train.py:876] (3/4) Epoch 1, batch 3300, loss[loss=0.2617, simple_loss=0.2264, pruned_loss=0.1485, over 5719.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.2563, pruned_loss=0.187, over 1077783.13 frames. ], batch size: 13, lr: 4.57e-02, grad_scale: 16.0 +2022-11-15 13:47:28,450 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.00 vs. limit=5.0 +2022-11-15 13:47:45,555 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 13:47:53,298 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:47:54,171 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.67 vs. limit=5.0 +2022-11-15 13:47:59,048 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={1, 2} +2022-11-15 13:48:17,245 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:48:26,983 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:48:35,851 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.826e+01 1.721e+02 2.073e+02 2.750e+02 6.330e+02, threshold=4.146e+02, percent-clipped=3.0 +2022-11-15 13:48:35,892 INFO [train.py:876] (3/4) Epoch 1, batch 3400, loss[loss=0.3501, simple_loss=0.2698, pruned_loss=0.2152, over 5315.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.2561, pruned_loss=0.1855, over 1080362.09 frames. ], batch size: 79, lr: 4.55e-02, grad_scale: 16.0 +2022-11-15 13:48:48,008 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 13:48:55,281 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0523, 2.2444, 2.1344, 2.3018, 2.2148, 2.2855, 1.5617, 2.3092], + device='cuda:3'), covar=tensor([0.0841, 0.0343, 0.0600, 0.0336, 0.0452, 0.0531, 0.2538, 0.0431], + device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0046, 0.0049, 0.0044, 0.0041, 0.0057, 0.0094, 0.0047], + device='cuda:3'), out_proj_covar=tensor([5.9470e-05, 4.0683e-05, 4.3973e-05, 3.7183e-05, 3.1911e-05, 5.2322e-05, + 1.1666e-04, 3.9378e-05], device='cuda:3') +2022-11-15 13:49:06,962 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7845, 3.3127, 3.2154, 3.2511, 3.6133, 2.8599, 2.1003, 3.2366], + device='cuda:3'), covar=tensor([0.1082, 0.0396, 0.0490, 0.0342, 0.0210, 0.0889, 0.3584, 0.0406], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0048, 0.0050, 0.0045, 0.0042, 0.0059, 0.0097, 0.0048], + device='cuda:3'), out_proj_covar=tensor([6.2054e-05, 4.2256e-05, 4.4577e-05, 3.7844e-05, 3.3022e-05, 5.3985e-05, + 1.2034e-04, 4.0116e-05], device='cuda:3') +2022-11-15 13:49:11,330 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.54 vs. limit=5.0 +2022-11-15 13:49:16,249 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 +2022-11-15 13:49:28,914 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 13:49:30,565 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:49:37,383 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2059, 2.4311, 2.6902, 2.2649, 2.4741, 1.7500, 2.2757, 2.7789], + device='cuda:3'), covar=tensor([0.0364, 0.0515, 0.0251, 0.0370, 0.0592, 0.1414, 0.0445, 0.0307], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0018, 0.0018, 0.0020, 0.0021, 0.0020, 0.0018, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.1397e-05, 1.8253e-05, 1.6609e-05, 2.0325e-05, 2.1178e-05, 1.9962e-05, + 2.0008e-05, 1.7526e-05], device='cuda:3') +2022-11-15 13:49:45,041 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 1.861e+02 2.622e+02 3.661e+02 7.520e+02, threshold=5.245e+02, percent-clipped=13.0 +2022-11-15 13:49:45,094 INFO [train.py:876] (3/4) Epoch 1, batch 3500, loss[loss=0.2731, simple_loss=0.2346, pruned_loss=0.1558, over 5738.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.2553, pruned_loss=0.1843, over 1083516.34 frames. ], batch size: 20, lr: 4.53e-02, grad_scale: 16.0 +2022-11-15 13:50:04,182 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:50:17,081 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:50:21,296 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.39 vs. limit=2.0 +2022-11-15 13:50:39,037 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 13:50:43,104 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.17 vs. limit=5.0 +2022-11-15 13:50:49,952 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:50:56,145 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.435e+01 2.156e+02 2.680e+02 3.527e+02 6.618e+02, threshold=5.360e+02, percent-clipped=3.0 +2022-11-15 13:50:56,189 INFO [train.py:876] (3/4) Epoch 1, batch 3600, loss[loss=0.3376, simple_loss=0.2803, pruned_loss=0.1975, over 5570.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.2534, pruned_loss=0.1818, over 1087883.83 frames. ], batch size: 25, lr: 4.50e-02, grad_scale: 16.0 +2022-11-15 13:51:00,505 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={2, 3} +2022-11-15 13:51:13,503 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7424, 1.5549, 1.7870, 1.5901, 1.7005, 1.4383, 0.8659, 1.5727], + device='cuda:3'), covar=tensor([0.0124, 0.0141, 0.0126, 0.0153, 0.0085, 0.0127, 0.0269, 0.0189], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0023, 0.0023, 0.0021, 0.0022, 0.0023, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.1760e-05, 2.0973e-05, 2.1428e-05, 2.2068e-05, 2.0582e-05, 2.1803e-05, + 2.4117e-05, 2.0360e-05], device='cuda:3') +2022-11-15 13:51:17,629 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.31 vs. limit=5.0 +2022-11-15 13:51:24,663 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:51:27,142 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:51:31,466 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.79 vs. limit=5.0 +2022-11-15 13:51:49,281 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:51:49,972 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:51:51,789 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 13:52:07,857 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 13:52:07,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.330e+02 3.139e+02 3.973e+02 9.859e+02, threshold=6.278e+02, percent-clipped=9.0 +2022-11-15 13:52:07,983 INFO [train.py:876] (3/4) Epoch 1, batch 3700, loss[loss=0.2226, simple_loss=0.1996, pruned_loss=0.1229, over 5364.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.2536, pruned_loss=0.1821, over 1081784.84 frames. ], batch size: 9, lr: 4.48e-02, grad_scale: 16.0 +2022-11-15 13:52:24,111 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:52:30,205 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 13:52:33,972 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={0, 2} +2022-11-15 13:53:06,846 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 +2022-11-15 13:53:20,083 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.129e+02 2.584e+02 3.254e+02 9.208e+02, threshold=5.168e+02, percent-clipped=2.0 +2022-11-15 13:53:20,124 INFO [train.py:876] (3/4) Epoch 1, batch 3800, loss[loss=0.457, simple_loss=0.3332, pruned_loss=0.2904, over 5452.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.2531, pruned_loss=0.1814, over 1080801.67 frames. ], batch size: 64, lr: 4.46e-02, grad_scale: 16.0 +2022-11-15 13:53:24,874 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.49 vs. limit=5.0 +2022-11-15 13:53:43,923 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-15 13:53:52,087 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.62 vs. limit=5.0 +2022-11-15 13:54:15,709 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0954, 1.5575, 1.8029, 1.0898, 2.1757, 1.6632, 1.8146, 2.1085], + device='cuda:3'), covar=tensor([0.0515, 0.0274, 0.0268, 0.0630, 0.0299, 0.0377, 0.0290, 0.0230], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0023, 0.0024, 0.0030, 0.0025, 0.0025, 0.0025, 0.0024], + device='cuda:3'), out_proj_covar=tensor([2.7584e-05, 2.2900e-05, 2.3575e-05, 3.5768e-05, 2.4505e-05, 2.4978e-05, + 2.4154e-05, 2.6061e-05], device='cuda:3') +2022-11-15 13:54:17,368 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.17 vs. limit=5.0 +2022-11-15 13:54:19,180 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4429, 3.0369, 3.2016, 3.3378, 3.2789, 2.7224, 1.9734, 3.0203], + device='cuda:3'), covar=tensor([0.1172, 0.0343, 0.0303, 0.0196, 0.0210, 0.0822, 0.2788, 0.0322], + device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0056, 0.0058, 0.0049, 0.0051, 0.0071, 0.0109, 0.0053], + device='cuda:3'), out_proj_covar=tensor([8.2941e-05, 5.1449e-05, 5.2247e-05, 4.1267e-05, 4.1887e-05, 6.8821e-05, + 1.3376e-04, 4.4557e-05], device='cuda:3') +2022-11-15 13:54:20,182 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.96 vs. limit=5.0 +2022-11-15 13:54:31,929 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 2.219e+02 2.583e+02 3.487e+02 8.673e+02, threshold=5.166e+02, percent-clipped=10.0 +2022-11-15 13:54:31,972 INFO [train.py:876] (3/4) Epoch 1, batch 3900, loss[loss=0.3673, simple_loss=0.3014, pruned_loss=0.2166, over 5629.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.2544, pruned_loss=0.1813, over 1087539.94 frames. ], batch size: 38, lr: 4.44e-02, grad_scale: 16.0 +2022-11-15 13:54:32,685 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 13:54:38,939 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.48 vs. limit=5.0 +2022-11-15 13:54:41,396 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 13:54:50,523 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:54:56,098 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 13:55:03,822 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:55:13,568 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 13:55:15,514 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6395, 2.0585, 1.9441, 2.0259, 1.8918, 2.0126, 1.7876, 1.9009], + device='cuda:3'), covar=tensor([0.0136, 0.0088, 0.0105, 0.0137, 0.0163, 0.0176, 0.0226, 0.0135], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0016, 0.0017, 0.0018, 0.0016, 0.0017, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.8336e-05, 1.5998e-05, 1.5274e-05, 1.8062e-05, 1.9826e-05, 1.7005e-05, + 1.9907e-05, 1.7006e-05], device='cuda:3') +2022-11-15 13:55:29,697 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.50 vs. limit=5.0 +2022-11-15 13:55:34,764 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:55:36,427 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 13:55:37,092 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 +2022-11-15 13:55:38,177 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:55:38,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.26 vs. limit=2.0 +2022-11-15 13:55:44,803 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 2.234e+02 2.679e+02 3.538e+02 6.488e+02, threshold=5.359e+02, percent-clipped=3.0 +2022-11-15 13:55:44,845 INFO [train.py:876] (3/4) Epoch 1, batch 4000, loss[loss=0.2424, simple_loss=0.2056, pruned_loss=0.1396, over 5116.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.2524, pruned_loss=0.1786, over 1089119.55 frames. ], batch size: 7, lr: 4.42e-02, grad_scale: 32.0 +2022-11-15 13:55:57,756 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 13:56:02,770 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7172, 2.0114, 2.4820, 2.9372, 3.3246, 2.8313, 2.4329, 2.2570], + device='cuda:3'), covar=tensor([0.0170, 0.0631, 0.0446, 0.0143, 0.0169, 0.0264, 0.0342, 0.0303], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0045, 0.0043, 0.0026, 0.0031, 0.0032, 0.0033, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.3013e-05, 4.5582e-05, 4.0810e-05, 2.1549e-05, 2.5795e-05, 3.0151e-05, + 3.1422e-05, 2.9482e-05], device='cuda:3') +2022-11-15 13:56:07,924 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 13:56:17,945 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.15 vs. limit=2.0 +2022-11-15 13:56:24,377 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.22 vs. limit=2.0 +2022-11-15 13:56:28,159 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.88 vs. limit=5.0 +2022-11-15 13:56:28,730 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3871, 2.6052, 2.3016, 1.9339, 2.4360, 1.9476, 2.2552, 2.4925], + device='cuda:3'), covar=tensor([0.0257, 0.0324, 0.0238, 0.0363, 0.0221, 0.0290, 0.0334, 0.0187], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0020, 0.0024, 0.0021, 0.0023, 0.0025, 0.0020], + device='cuda:3'), out_proj_covar=tensor([2.3952e-05, 2.6962e-05, 2.0999e-05, 2.4816e-05, 2.1441e-05, 2.3763e-05, + 2.6136e-05, 2.1034e-05], device='cuda:3') +2022-11-15 13:56:58,560 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 2.074e+02 2.727e+02 3.371e+02 7.611e+02, threshold=5.454e+02, percent-clipped=4.0 +2022-11-15 13:56:58,603 INFO [train.py:876] (3/4) Epoch 1, batch 4100, loss[loss=0.3114, simple_loss=0.2508, pruned_loss=0.186, over 5692.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.2501, pruned_loss=0.1765, over 1088512.89 frames. ], batch size: 28, lr: 4.40e-02, grad_scale: 32.0 +2022-11-15 13:56:58,979 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.89 vs. limit=5.0 +2022-11-15 13:57:27,346 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 +2022-11-15 13:57:28,542 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:58:13,172 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 1.959e+02 2.484e+02 3.266e+02 7.504e+02, threshold=4.967e+02, percent-clipped=2.0 +2022-11-15 13:58:13,215 INFO [train.py:876] (3/4) Epoch 1, batch 4200, loss[loss=0.3457, simple_loss=0.268, pruned_loss=0.2117, over 5369.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.25, pruned_loss=0.1763, over 1084974.88 frames. ], batch size: 70, lr: 4.38e-02, grad_scale: 32.0 +2022-11-15 13:58:14,065 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:58:14,115 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:58:49,670 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 13:59:13,103 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:59:17,699 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 13:59:27,275 INFO [train.py:876] (3/4) Epoch 1, batch 4300, loss[loss=0.2596, simple_loss=0.2249, pruned_loss=0.1471, over 5489.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.249, pruned_loss=0.1754, over 1080819.45 frames. ], batch size: 10, lr: 4.35e-02, grad_scale: 16.0 +2022-11-15 13:59:27,971 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.229e+02 3.130e+02 3.930e+02 1.663e+03, threshold=6.259e+02, percent-clipped=10.0 +2022-11-15 13:59:50,087 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 13:59:51,403 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9469, 4.7728, 4.1681, 4.6882, 4.7098, 4.0681, 3.6711, 3.7968], + device='cuda:3'), covar=tensor([0.0280, 0.0231, 0.0185, 0.0258, 0.0120, 0.0322, 0.0379, 0.0227], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0034, 0.0046, 0.0036, 0.0039, 0.0038, 0.0035, 0.0035], + device='cuda:3'), out_proj_covar=tensor([4.7087e-05, 4.4216e-05, 5.8034e-05, 4.6226e-05, 4.6843e-05, 4.6138e-05, + 4.3948e-05, 4.2709e-05], device='cuda:3') +2022-11-15 13:59:55,559 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9473, 4.5816, 4.1418, 4.6182, 4.5614, 4.0102, 3.6547, 3.9129], + device='cuda:3'), covar=tensor([0.0299, 0.0204, 0.0193, 0.0169, 0.0154, 0.0411, 0.0352, 0.0252], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0034, 0.0046, 0.0037, 0.0039, 0.0038, 0.0035, 0.0035], + device='cuda:3'), out_proj_covar=tensor([4.7229e-05, 4.4607e-05, 5.8342e-05, 4.6586e-05, 4.7015e-05, 4.6520e-05, + 4.3860e-05, 4.2753e-05], device='cuda:3') +2022-11-15 14:00:02,825 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:00:25,195 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:00:31,508 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6332, 3.1955, 4.3515, 4.0753, 3.1535, 3.6664, 4.0625, 4.4013], + device='cuda:3'), covar=tensor([0.0179, 0.0899, 0.0222, 0.0287, 0.0297, 0.0328, 0.0188, 0.0127], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0058, 0.0031, 0.0039, 0.0030, 0.0032, 0.0031, 0.0029], + device='cuda:3'), out_proj_covar=tensor([2.4116e-05, 6.2643e-05, 3.0357e-05, 3.9758e-05, 2.8283e-05, 3.2164e-05, + 2.9030e-05, 2.7452e-05], device='cuda:3') +2022-11-15 14:00:38,814 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8516, 1.7071, 1.5854, 1.6673, 1.1707, 1.2897, 1.2681, 1.6881], + device='cuda:3'), covar=tensor([0.0105, 0.0124, 0.0131, 0.0145, 0.0277, 0.0244, 0.0259, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0025, 0.0025, 0.0026, 0.0024, 0.0027, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.0124e-05, 2.0273e-05, 2.5963e-05, 2.6668e-05, 2.9212e-05, 2.6998e-05, + 3.0136e-05, 2.6535e-05], device='cuda:3') +2022-11-15 14:00:41,286 INFO [train.py:876] (3/4) Epoch 1, batch 4400, loss[loss=0.3057, simple_loss=0.2592, pruned_loss=0.1761, over 5764.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.2491, pruned_loss=0.1754, over 1081313.07 frames. ], batch size: 14, lr: 4.33e-02, grad_scale: 16.0 +2022-11-15 14:00:41,970 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 1.948e+02 2.508e+02 3.167e+02 7.237e+02, threshold=5.016e+02, percent-clipped=3.0 +2022-11-15 14:01:40,965 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3351, 1.6926, 3.0742, 2.6790, 2.6310, 3.0263, 3.1294, 3.2092], + device='cuda:3'), covar=tensor([0.0091, 0.0866, 0.0154, 0.0317, 0.0236, 0.0166, 0.0149, 0.0149], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0059, 0.0032, 0.0041, 0.0031, 0.0034, 0.0032, 0.0030], + device='cuda:3'), out_proj_covar=tensor([2.5265e-05, 6.5006e-05, 3.1157e-05, 4.1773e-05, 2.9381e-05, 3.3730e-05, + 3.1321e-05, 2.9364e-05], device='cuda:3') +2022-11-15 14:01:51,587 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:01:54,770 INFO [train.py:876] (3/4) Epoch 1, batch 4500, loss[loss=0.2402, simple_loss=0.2181, pruned_loss=0.1312, over 5748.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.2489, pruned_loss=0.1746, over 1082436.59 frames. ], batch size: 15, lr: 4.31e-02, grad_scale: 16.0 +2022-11-15 14:01:55,422 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.194e+02 3.031e+02 3.828e+02 9.010e+02, threshold=6.062e+02, percent-clipped=8.0 +2022-11-15 14:01:58,635 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 +2022-11-15 14:02:07,608 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 14:02:14,355 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2382, 1.0754, 1.0527, 1.2600, 0.8265, 1.1422, 0.5030, 1.0832], + device='cuda:3'), covar=tensor([0.0101, 0.0087, 0.0124, 0.0115, 0.0153, 0.0163, 0.0232, 0.0172], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0025, 0.0026, 0.0027, 0.0027, 0.0026, 0.0028, 0.0026], + device='cuda:3'), out_proj_covar=tensor([2.9768e-05, 2.5564e-05, 2.8165e-05, 2.7296e-05, 2.8067e-05, 2.6126e-05, + 3.3812e-05, 2.8018e-05], device='cuda:3') +2022-11-15 14:02:49,993 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 14:02:54,228 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:03:08,725 INFO [train.py:876] (3/4) Epoch 1, batch 4600, loss[loss=0.3267, simple_loss=0.2596, pruned_loss=0.1969, over 5642.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.2465, pruned_loss=0.1718, over 1079749.02 frames. ], batch size: 29, lr: 4.29e-02, grad_scale: 16.0 +2022-11-15 14:03:08,895 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8657, 2.0785, 1.7765, 1.3545, 1.9786, 2.2147, 2.2756, 2.2026], + device='cuda:3'), covar=tensor([0.0178, 0.0227, 0.0343, 0.0941, 0.0437, 0.0314, 0.0391, 0.0571], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0024, 0.0026, 0.0027, 0.0025, 0.0028, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.0645e-05, 2.0848e-05, 2.6456e-05, 2.9055e-05, 3.0716e-05, 2.9586e-05, + 3.2041e-05, 2.7236e-05], device='cuda:3') +2022-11-15 14:03:09,369 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.162e+01 1.839e+02 2.747e+02 3.849e+02 7.443e+02, threshold=5.493e+02, percent-clipped=4.0 +2022-11-15 14:03:21,754 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 14:03:24,556 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.41 vs. limit=2.0 +2022-11-15 14:03:29,919 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:03:40,562 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:03:46,407 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 14:04:24,344 INFO [train.py:876] (3/4) Epoch 1, batch 4700, loss[loss=0.2581, simple_loss=0.2298, pruned_loss=0.1432, over 5705.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.2462, pruned_loss=0.1706, over 1083861.99 frames. ], batch size: 17, lr: 4.27e-02, grad_scale: 16.0 +2022-11-15 14:04:24,958 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.250e+02 2.748e+02 3.964e+02 7.433e+02, threshold=5.495e+02, percent-clipped=7.0 +2022-11-15 14:05:03,832 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 14:05:08,330 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 14:05:20,389 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.43 vs. limit=5.0 +2022-11-15 14:05:34,570 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:05:34,865 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 14:05:37,708 INFO [train.py:876] (3/4) Epoch 1, batch 4800, loss[loss=0.2902, simple_loss=0.2434, pruned_loss=0.1684, over 5733.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.2476, pruned_loss=0.1716, over 1089915.79 frames. ], batch size: 31, lr: 4.25e-02, grad_scale: 16.0 +2022-11-15 14:05:38,343 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 1.870e+02 2.529e+02 3.283e+02 6.481e+02, threshold=5.059e+02, percent-clipped=1.0 +2022-11-15 14:05:46,851 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 14:06:09,938 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:06:10,126 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7268, 2.3130, 3.1564, 2.3250, 2.9714, 2.7386, 2.9519, 3.1861], + device='cuda:3'), covar=tensor([0.0202, 0.1709, 0.0409, 0.0996, 0.0417, 0.0515, 0.0510, 0.0459], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0074, 0.0037, 0.0049, 0.0037, 0.0043, 0.0043, 0.0037], + device='cuda:3'), out_proj_covar=tensor([2.9954e-05, 8.3062e-05, 3.6761e-05, 5.1706e-05, 3.6763e-05, 4.4116e-05, + 4.4178e-05, 3.8444e-05], device='cuda:3') +2022-11-15 14:06:50,935 INFO [train.py:876] (3/4) Epoch 1, batch 4900, loss[loss=0.3447, simple_loss=0.2729, pruned_loss=0.2083, over 5360.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.2465, pruned_loss=0.1704, over 1086834.67 frames. ], batch size: 70, lr: 4.23e-02, grad_scale: 16.0 +2022-11-15 14:06:51,624 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 2.074e+02 2.835e+02 3.865e+02 7.498e+02, threshold=5.670e+02, percent-clipped=5.0 +2022-11-15 14:07:05,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.70 vs. limit=5.0 +2022-11-15 14:07:22,584 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:07:54,824 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1104, 1.6879, 1.6424, 2.7084, 2.7542, 2.2540, 2.2164, 2.8672], + device='cuda:3'), covar=tensor([0.0128, 0.1020, 0.0919, 0.0200, 0.0312, 0.0665, 0.0743, 0.0235], + device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0081, 0.0074, 0.0041, 0.0050, 0.0064, 0.0068, 0.0047], + device='cuda:3'), out_proj_covar=tensor([3.8777e-05, 8.7746e-05, 7.5288e-05, 4.0100e-05, 4.8359e-05, 6.5486e-05, + 7.1656e-05, 4.5770e-05], device='cuda:3') +2022-11-15 14:07:57,427 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:08:08,411 INFO [train.py:876] (3/4) Epoch 1, batch 5000, loss[loss=0.2944, simple_loss=0.2426, pruned_loss=0.1731, over 5262.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.2479, pruned_loss=0.1721, over 1081579.41 frames. ], batch size: 79, lr: 4.20e-02, grad_scale: 16.0 +2022-11-15 14:08:09,092 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.045e+02 2.576e+02 3.694e+02 7.012e+02, threshold=5.152e+02, percent-clipped=6.0 +2022-11-15 14:08:16,178 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0292, 1.8596, 1.6531, 1.8223, 1.8824, 2.3135, 1.3869, 2.0373], + device='cuda:3'), covar=tensor([0.0174, 0.0216, 0.0148, 0.0171, 0.0155, 0.0124, 0.0281, 0.0201], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0020, 0.0019, 0.0024, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.2354e-05, 2.3821e-05, 2.0531e-05, 2.1340e-05, 2.1913e-05, 2.0768e-05, + 2.7110e-05, 1.8680e-05], device='cuda:3') +2022-11-15 14:08:24,696 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6177, 1.3984, 0.8677, 1.3616, 1.8842, 1.4716, 0.9607, 0.8103], + device='cuda:3'), covar=tensor([0.0201, 0.0301, 0.0245, 0.0225, 0.0138, 0.0176, 0.0262, 0.0240], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0025, 0.0025, 0.0023, 0.0023, 0.0024, 0.0022], + device='cuda:3'), out_proj_covar=tensor([2.6424e-05, 2.4203e-05, 2.7917e-05, 2.6545e-05, 2.4376e-05, 2.3972e-05, + 3.2083e-05, 2.4304e-05], device='cuda:3') +2022-11-15 14:09:10,638 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7514, 2.3449, 2.2061, 2.3917, 2.0293, 2.2691, 1.6060, 2.1753], + device='cuda:3'), covar=tensor([0.0742, 0.0168, 0.0299, 0.0103, 0.0279, 0.0392, 0.1526, 0.0196], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0066, 0.0072, 0.0057, 0.0064, 0.0092, 0.0145, 0.0064], + device='cuda:3'), out_proj_covar=tensor([1.2112e-04, 6.1109e-05, 7.0091e-05, 5.3136e-05, 6.1949e-05, 9.6472e-05, + 1.6546e-04, 5.8802e-05], device='cuda:3') +2022-11-15 14:09:15,106 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.94 vs. limit=5.0 +2022-11-15 14:09:20,201 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2678, 3.6108, 3.0715, 1.9941, 3.8432, 3.1703, 2.7586, 3.5239], + device='cuda:3'), covar=tensor([0.0299, 0.0199, 0.0259, 0.1558, 0.0130, 0.0275, 0.0354, 0.0242], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0025, 0.0024, 0.0048, 0.0026, 0.0025, 0.0021, 0.0025], + device='cuda:3'), out_proj_covar=tensor([4.6226e-05, 3.7238e-05, 3.5361e-05, 7.0845e-05, 3.6740e-05, 3.7289e-05, + 3.2442e-05, 3.6792e-05], device='cuda:3') +2022-11-15 14:09:21,851 INFO [train.py:876] (3/4) Epoch 1, batch 5100, loss[loss=0.2675, simple_loss=0.2299, pruned_loss=0.1526, over 5559.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.2456, pruned_loss=0.1687, over 1083410.99 frames. ], batch size: 16, lr: 4.18e-02, grad_scale: 16.0 +2022-11-15 14:09:22,489 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 2.161e+02 2.601e+02 3.354e+02 8.150e+02, threshold=5.203e+02, percent-clipped=5.0 +2022-11-15 14:09:38,358 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4506, 1.7732, 2.0938, 1.9737, 2.4347, 2.1326, 1.9707, 1.8538], + device='cuda:3'), covar=tensor([0.0189, 0.0283, 0.0170, 0.0215, 0.0512, 0.0603, 0.0294, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0014, 0.0014, 0.0016, 0.0016, 0.0013, 0.0015, 0.0015], + device='cuda:3'), out_proj_covar=tensor([1.9053e-05, 1.6318e-05, 1.5736e-05, 1.9692e-05, 2.0158e-05, 1.6963e-05, + 1.7849e-05, 1.7856e-05], device='cuda:3') +2022-11-15 14:09:40,349 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7378, 4.3692, 4.0767, 4.2476, 4.3587, 3.8879, 3.9125, 3.5896], + device='cuda:3'), covar=tensor([0.0341, 0.0337, 0.0296, 0.0334, 0.0270, 0.0309, 0.0212, 0.0456], + device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0036, 0.0048, 0.0035, 0.0043, 0.0043, 0.0036, 0.0037], + device='cuda:3'), out_proj_covar=tensor([5.5374e-05, 5.2422e-05, 6.6746e-05, 4.8180e-05, 5.7981e-05, 5.4363e-05, + 4.8600e-05, 4.9799e-05], device='cuda:3') +2022-11-15 14:10:30,354 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 +2022-11-15 14:10:32,198 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0219, 3.0180, 2.7078, 3.2981, 2.5154, 3.3033, 3.1315, 2.1901], + device='cuda:3'), covar=tensor([0.0534, 0.0168, 0.0191, 0.0116, 0.0285, 0.0133, 0.0170, 0.0182], + device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0026, 0.0026, 0.0020, 0.0033, 0.0025, 0.0027, 0.0024], + device='cuda:3'), out_proj_covar=tensor([3.7520e-05, 2.3047e-05, 2.4539e-05, 1.8105e-05, 3.1152e-05, 2.1661e-05, + 2.4814e-05, 1.9668e-05], device='cuda:3') +2022-11-15 14:10:33,660 INFO [train.py:876] (3/4) Epoch 1, batch 5200, loss[loss=0.3144, simple_loss=0.2511, pruned_loss=0.1889, over 5488.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.2482, pruned_loss=0.1716, over 1078727.29 frames. ], batch size: 64, lr: 4.16e-02, grad_scale: 16.0 +2022-11-15 14:10:34,300 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.030e+02 2.662e+02 3.916e+02 1.299e+03, threshold=5.323e+02, percent-clipped=9.0 +2022-11-15 14:11:03,354 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3628, 1.7563, 2.4196, 2.1809, 2.1547, 1.6299, 1.9060, 1.6132], + device='cuda:3'), covar=tensor([0.0201, 0.0293, 0.0102, 0.0202, 0.0328, 0.1660, 0.0263, 0.0318], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0014, 0.0016, 0.0016, 0.0014, 0.0015, 0.0015], + device='cuda:3'), out_proj_covar=tensor([1.9965e-05, 1.8521e-05, 1.5307e-05, 2.0028e-05, 2.0276e-05, 1.7707e-05, + 1.8689e-05, 1.8307e-05], device='cuda:3') +2022-11-15 14:11:46,392 INFO [train.py:876] (3/4) Epoch 1, batch 5300, loss[loss=0.3057, simple_loss=0.2482, pruned_loss=0.1816, over 5371.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.2474, pruned_loss=0.1701, over 1084469.73 frames. ], batch size: 70, lr: 4.14e-02, grad_scale: 16.0 +2022-11-15 14:11:46,599 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7408, 1.6708, 2.7285, 2.0322, 2.9707, 2.3860, 2.6813, 3.0859], + device='cuda:3'), covar=tensor([0.0201, 0.1357, 0.0309, 0.0756, 0.0237, 0.0464, 0.0400, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0080, 0.0041, 0.0054, 0.0037, 0.0046, 0.0044, 0.0040], + device='cuda:3'), out_proj_covar=tensor([3.5756e-05, 9.1583e-05, 4.5152e-05, 5.9736e-05, 4.0093e-05, 4.9242e-05, + 4.7485e-05, 4.5402e-05], device='cuda:3') +2022-11-15 14:11:47,374 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 2.088e+02 2.621e+02 3.205e+02 6.242e+02, threshold=5.243e+02, percent-clipped=4.0 +2022-11-15 14:12:02,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7861, 2.2833, 2.2632, 2.3718, 2.0066, 2.2594, 1.6119, 2.2165], + device='cuda:3'), covar=tensor([0.0871, 0.0234, 0.0305, 0.0161, 0.0409, 0.0369, 0.1734, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0071, 0.0079, 0.0064, 0.0072, 0.0099, 0.0155, 0.0072], + device='cuda:3'), out_proj_covar=tensor([1.3754e-04, 6.6847e-05, 7.6905e-05, 5.9933e-05, 7.1488e-05, 1.0413e-04, + 1.7544e-04, 6.8441e-05], device='cuda:3') +2022-11-15 14:12:06,231 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1711, 2.0959, 1.7603, 2.5363, 2.6644, 2.8900, 2.3358, 2.2583], + device='cuda:3'), covar=tensor([0.0162, 0.0748, 0.0277, 0.0131, 0.0133, 0.0094, 0.0171, 0.0152], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0017, 0.0016, 0.0016, 0.0017, 0.0015, 0.0019, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.9731e-05, 2.0945e-05, 1.8712e-05, 1.8389e-05, 2.0231e-05, 1.7391e-05, + 2.2644e-05, 1.6979e-05], device='cuda:3') +2022-11-15 14:12:22,768 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1947, 1.6335, 1.9320, 2.5405, 1.9453, 2.5519, 2.5863, 2.5833], + device='cuda:3'), covar=tensor([0.0332, 0.0251, 0.0167, 0.0174, 0.0303, 0.0168, 0.0167, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0031, 0.0032, 0.0024, 0.0040, 0.0029, 0.0032, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.5695e-05, 2.7693e-05, 3.1080e-05, 2.2889e-05, 3.8069e-05, 2.6415e-05, + 3.0289e-05, 2.3763e-05], device='cuda:3') +2022-11-15 14:12:47,548 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8887, 4.1227, 4.0644, 4.0549, 3.8381, 4.0123, 2.9243, 3.6985], + device='cuda:3'), covar=tensor([0.0205, 0.0127, 0.0083, 0.0137, 0.0142, 0.0163, 0.0587, 0.0194], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0030, 0.0029, 0.0025, 0.0029, 0.0028, 0.0047, 0.0031], + device='cuda:3'), out_proj_covar=tensor([4.9845e-05, 4.2078e-05, 4.1269e-05, 3.3936e-05, 4.1814e-05, 3.9761e-05, + 6.6990e-05, 4.6210e-05], device='cuda:3') +2022-11-15 14:12:52,369 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.09 vs. limit=5.0 +2022-11-15 14:13:00,084 INFO [train.py:876] (3/4) Epoch 1, batch 5400, loss[loss=0.2388, simple_loss=0.2134, pruned_loss=0.1322, over 5506.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.247, pruned_loss=0.1689, over 1085063.71 frames. ], batch size: 13, lr: 4.12e-02, grad_scale: 16.0 +2022-11-15 14:13:00,710 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 2.076e+02 2.742e+02 3.471e+02 5.546e+02, threshold=5.484e+02, percent-clipped=1.0 +2022-11-15 14:13:30,511 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8491, 2.3528, 2.3060, 2.4117, 1.7688, 2.1928, 1.5127, 2.2498], + device='cuda:3'), covar=tensor([0.0941, 0.0236, 0.0267, 0.0169, 0.0490, 0.0500, 0.2037, 0.0260], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0072, 0.0077, 0.0064, 0.0072, 0.0104, 0.0159, 0.0075], + device='cuda:3'), out_proj_covar=tensor([1.4189e-04, 6.9285e-05, 7.6409e-05, 6.1882e-05, 7.2195e-05, 1.1126e-04, + 1.7754e-04, 7.3595e-05], device='cuda:3') +2022-11-15 14:13:47,126 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 14:14:12,219 INFO [train.py:876] (3/4) Epoch 1, batch 5500, loss[loss=0.3201, simple_loss=0.2654, pruned_loss=0.1874, over 5459.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.245, pruned_loss=0.1674, over 1085406.85 frames. ], batch size: 49, lr: 4.10e-02, grad_scale: 16.0 +2022-11-15 14:14:12,892 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.146e+02 2.749e+02 3.970e+02 7.189e+02, threshold=5.498e+02, percent-clipped=5.0 +2022-11-15 14:14:17,245 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 +2022-11-15 14:15:03,319 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8542, 2.3561, 2.3091, 2.4346, 2.1450, 2.2614, 1.6766, 2.2005], + device='cuda:3'), covar=tensor([0.0751, 0.0169, 0.0254, 0.0133, 0.0290, 0.0406, 0.1427, 0.0229], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0069, 0.0076, 0.0061, 0.0071, 0.0104, 0.0152, 0.0074], + device='cuda:3'), out_proj_covar=tensor([1.4023e-04, 6.6517e-05, 7.6318e-05, 5.9293e-05, 7.2010e-05, 1.1236e-04, + 1.6918e-04, 7.2621e-05], device='cuda:3') +2022-11-15 14:15:25,183 INFO [train.py:876] (3/4) Epoch 1, batch 5600, loss[loss=0.282, simple_loss=0.2431, pruned_loss=0.1604, over 5679.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.2441, pruned_loss=0.1658, over 1087325.51 frames. ], batch size: 19, lr: 4.08e-02, grad_scale: 16.0 +2022-11-15 14:15:26,174 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 2.152e+02 2.832e+02 3.606e+02 7.262e+02, threshold=5.664e+02, percent-clipped=5.0 +2022-11-15 14:15:28,035 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 +2022-11-15 14:15:29,435 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.42 vs. limit=5.0 +2022-11-15 14:15:54,500 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.60 vs. limit=5.0 +2022-11-15 14:16:00,754 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:16:37,327 INFO [train.py:876] (3/4) Epoch 1, batch 5700, loss[loss=0.3406, simple_loss=0.2845, pruned_loss=0.1984, over 5619.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.2422, pruned_loss=0.1643, over 1087421.34 frames. ], batch size: 38, lr: 4.06e-02, grad_scale: 16.0 +2022-11-15 14:16:37,962 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 2.164e+02 2.765e+02 3.457e+02 8.983e+02, threshold=5.530e+02, percent-clipped=5.0 +2022-11-15 14:16:44,907 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 14:17:12,187 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.58 vs. limit=5.0 +2022-11-15 14:17:16,903 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:17:39,032 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0614, 3.2689, 3.3241, 3.2764, 2.9408, 3.1640, 1.9892, 3.0835], + device='cuda:3'), covar=tensor([0.0283, 0.0213, 0.0135, 0.0135, 0.0204, 0.0199, 0.0980, 0.0240], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0032, 0.0031, 0.0026, 0.0030, 0.0027, 0.0049, 0.0032], + device='cuda:3'), out_proj_covar=tensor([5.2496e-05, 4.6568e-05, 4.6113e-05, 3.6918e-05, 4.4904e-05, 3.9873e-05, + 7.1432e-05, 4.8651e-05], device='cuda:3') +2022-11-15 14:17:50,566 INFO [train.py:876] (3/4) Epoch 1, batch 5800, loss[loss=0.2236, simple_loss=0.1976, pruned_loss=0.1248, over 5417.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.2426, pruned_loss=0.1649, over 1082611.69 frames. ], batch size: 11, lr: 4.04e-02, grad_scale: 16.0 +2022-11-15 14:17:51,229 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.770e+01 1.984e+02 2.593e+02 3.696e+02 7.124e+02, threshold=5.186e+02, percent-clipped=5.0 +2022-11-15 14:17:57,935 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 +2022-11-15 14:18:01,249 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:18:31,846 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:18:42,290 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1517, 1.3654, 2.0231, 2.0399, 1.7209, 1.5808, 1.6717, 1.8322], + device='cuda:3'), covar=tensor([0.0216, 0.0329, 0.0285, 0.0224, 0.0403, 0.0474, 0.0303, 0.0489], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0026, 0.0029, 0.0026, 0.0033, 0.0028, 0.0032, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.8580e-05, 3.1920e-05, 3.6237e-05, 3.3048e-05, 4.2384e-05, 3.9994e-05, + 4.1600e-05, 3.1310e-05], device='cuda:3') +2022-11-15 14:18:49,242 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.15 vs. limit=2.0 +2022-11-15 14:19:03,093 INFO [train.py:876] (3/4) Epoch 1, batch 5900, loss[loss=0.3142, simple_loss=0.2518, pruned_loss=0.1883, over 5475.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.2416, pruned_loss=0.1643, over 1085750.02 frames. ], batch size: 49, lr: 4.02e-02, grad_scale: 16.0 +2022-11-15 14:19:03,748 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.869e+02 2.719e+02 3.366e+02 7.828e+02, threshold=5.439e+02, percent-clipped=3.0 +2022-11-15 14:19:16,329 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:19:44,079 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.35 vs. limit=2.0 +2022-11-15 14:19:59,851 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4900, 3.3333, 2.6350, 1.4141, 2.8709, 2.1725, 3.0015, 2.2722], + device='cuda:3'), covar=tensor([0.0512, 0.0181, 0.0253, 0.1743, 0.0243, 0.0467, 0.0208, 0.0453], + device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0031, 0.0028, 0.0061, 0.0033, 0.0035, 0.0026, 0.0036], + device='cuda:3'), out_proj_covar=tensor([7.2204e-05, 4.8364e-05, 4.6464e-05, 9.4787e-05, 5.1533e-05, 5.6657e-05, + 4.5353e-05, 5.7489e-05], device='cuda:3') +2022-11-15 14:20:00,644 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4173, 3.0473, 1.3672, 1.3540, 2.0277, 2.6169, 2.0337, 2.6828], + device='cuda:3'), covar=tensor([0.0408, 0.0182, 0.0632, 0.0502, 0.0224, 0.0153, 0.0182, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0046, 0.0044, 0.0051, 0.0042, 0.0033, 0.0035, 0.0041], + device='cuda:3'), out_proj_covar=tensor([7.3755e-05, 5.3726e-05, 5.7633e-05, 6.1653e-05, 5.1026e-05, 3.7723e-05, + 4.2704e-05, 4.7255e-05], device='cuda:3') +2022-11-15 14:20:07,706 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7982, 1.5935, 2.3102, 2.8414, 3.2463, 3.1920, 2.2842, 2.6793], + device='cuda:3'), covar=tensor([0.0188, 0.0683, 0.0216, 0.0121, 0.0111, 0.0080, 0.0267, 0.0119], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0016, 0.0016, 0.0018, 0.0017, 0.0015, 0.0021, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.9575e-05, 2.0517e-05, 1.9419e-05, 2.0002e-05, 2.0675e-05, 1.7129e-05, + 2.6889e-05, 1.7751e-05], device='cuda:3') +2022-11-15 14:20:07,724 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:20:15,907 INFO [train.py:876] (3/4) Epoch 1, batch 6000, loss[loss=0.3519, simple_loss=0.2732, pruned_loss=0.2153, over 4622.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.2413, pruned_loss=0.1637, over 1079981.40 frames. ], batch size: 135, lr: 4.00e-02, grad_scale: 16.0 +2022-11-15 14:20:15,907 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 14:20:34,716 INFO [train.py:908] (3/4) Epoch 1, validation: loss=0.2263, simple_loss=0.2274, pruned_loss=0.1126, over 1530663.00 frames. +2022-11-15 14:20:34,717 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4471MB +2022-11-15 14:20:35,400 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 2.347e+02 2.873e+02 3.885e+02 1.859e+03, threshold=5.746e+02, percent-clipped=5.0 +2022-11-15 14:20:38,398 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:20:55,704 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 +2022-11-15 14:21:05,050 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:21:11,069 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:21:38,071 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 14:21:47,064 INFO [train.py:876] (3/4) Epoch 1, batch 6100, loss[loss=0.3524, simple_loss=0.2993, pruned_loss=0.2028, over 5638.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.2395, pruned_loss=0.1615, over 1081070.44 frames. ], batch size: 29, lr: 3.98e-02, grad_scale: 16.0 +2022-11-15 14:21:47,727 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.266e+02 2.673e+02 3.416e+02 6.924e+02, threshold=5.346e+02, percent-clipped=3.0 +2022-11-15 14:21:49,338 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:21:54,395 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:22:11,648 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:22:37,310 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:22:43,679 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5767, 1.4195, 1.5742, 2.4727, 2.5539, 1.7323, 1.4445, 2.1884], + device='cuda:3'), covar=tensor([0.0141, 0.0957, 0.0869, 0.0178, 0.0170, 0.0667, 0.1010, 0.0253], + device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0103, 0.0097, 0.0052, 0.0063, 0.0094, 0.0108, 0.0056], + device='cuda:3'), out_proj_covar=tensor([5.1501e-05, 1.1666e-04, 1.0677e-04, 6.0938e-05, 6.6525e-05, 1.0573e-04, + 1.1892e-04, 5.8666e-05], device='cuda:3') +2022-11-15 14:22:55,822 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:22:59,790 INFO [train.py:876] (3/4) Epoch 1, batch 6200, loss[loss=0.3608, simple_loss=0.29, pruned_loss=0.2158, over 5571.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.2406, pruned_loss=0.1625, over 1084234.51 frames. ], batch size: 43, lr: 3.96e-02, grad_scale: 16.0 +2022-11-15 14:23:00,453 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.942e+02 2.617e+02 4.109e+02 1.137e+03, threshold=5.234e+02, percent-clipped=10.0 +2022-11-15 14:23:01,349 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:07,058 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:09,425 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:21,861 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:45,470 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:46,882 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:51,487 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:23:58,041 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4968, 4.1572, 3.7169, 4.1438, 4.0336, 3.6963, 3.4418, 3.3862], + device='cuda:3'), covar=tensor([0.0330, 0.0223, 0.0282, 0.0220, 0.0207, 0.0262, 0.0347, 0.0438], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0041, 0.0055, 0.0043, 0.0052, 0.0052, 0.0044, 0.0042], + device='cuda:3'), out_proj_covar=tensor([6.9361e-05, 6.8694e-05, 8.0841e-05, 6.6880e-05, 7.8762e-05, 7.3680e-05, + 6.4657e-05, 6.2460e-05], device='cuda:3') +2022-11-15 14:24:12,261 INFO [train.py:876] (3/4) Epoch 1, batch 6300, loss[loss=0.3278, simple_loss=0.2479, pruned_loss=0.2039, over 2969.00 frames. ], tot_loss[loss=0.281, simple_loss=0.2394, pruned_loss=0.1613, over 1081769.11 frames. ], batch size: 284, lr: 3.94e-02, grad_scale: 32.0 +2022-11-15 14:24:12,910 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.220e+02 2.802e+02 3.554e+02 1.076e+03, threshold=5.605e+02, percent-clipped=6.0 +2022-11-15 14:24:15,829 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:24:31,169 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:24:44,721 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:24:50,572 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:25:22,871 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:25:24,490 INFO [train.py:876] (3/4) Epoch 1, batch 6400, loss[loss=0.2869, simple_loss=0.254, pruned_loss=0.1599, over 5820.00 frames. ], tot_loss[loss=0.283, simple_loss=0.2416, pruned_loss=0.1622, over 1086095.53 frames. ], batch size: 21, lr: 3.92e-02, grad_scale: 32.0 +2022-11-15 14:25:25,178 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 2.235e+02 2.872e+02 3.964e+02 7.777e+02, threshold=5.745e+02, percent-clipped=4.0 +2022-11-15 14:25:32,020 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:25:46,932 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 14:25:53,036 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.26 vs. limit=2.0 +2022-11-15 14:26:06,809 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:09,024 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:12,753 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:30,105 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:36,862 INFO [train.py:876] (3/4) Epoch 1, batch 6500, loss[loss=0.2153, simple_loss=0.2062, pruned_loss=0.1122, over 5688.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.2418, pruned_loss=0.1627, over 1078758.40 frames. ], batch size: 19, lr: 3.90e-02, grad_scale: 32.0 +2022-11-15 14:26:37,572 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.150e+02 2.872e+02 3.674e+02 6.857e+02, threshold=5.744e+02, percent-clipped=4.0 +2022-11-15 14:26:46,478 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:53,223 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 14:26:55,171 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:26:56,668 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:27:17,794 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 +2022-11-15 14:27:18,846 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:27:20,933 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:27:24,783 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:27:27,157 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.56 vs. limit=2.0 +2022-11-15 14:27:49,421 INFO [train.py:876] (3/4) Epoch 1, batch 6600, loss[loss=0.3197, simple_loss=0.2713, pruned_loss=0.1841, over 5546.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.2405, pruned_loss=0.1614, over 1082634.75 frames. ], batch size: 17, lr: 3.89e-02, grad_scale: 32.0 +2022-11-15 14:27:50,094 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 2.099e+02 2.757e+02 3.560e+02 8.696e+02, threshold=5.514e+02, percent-clipped=5.0 +2022-11-15 14:28:04,509 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 +2022-11-15 14:28:04,928 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:28:10,138 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:28:22,503 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:28:32,390 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7985, 1.1673, 0.7016, 0.5768, 0.9321, 0.9526, 0.7714, 0.7954], + device='cuda:3'), covar=tensor([0.0249, 0.0172, 0.0253, 0.0477, 0.0539, 0.0269, 0.0414, 0.0267], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0018, 0.0020, 0.0022, 0.0017, 0.0018, 0.0021, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.5177e-05, 2.4041e-05, 2.7149e-05, 3.5337e-05, 2.5454e-05, 2.5330e-05, + 2.9576e-05, 2.4808e-05], device='cuda:3') +2022-11-15 14:28:54,046 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:28:56,693 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:29:00,629 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:29:01,856 INFO [train.py:876] (3/4) Epoch 1, batch 6700, loss[loss=0.2324, simple_loss=0.1978, pruned_loss=0.1335, over 5149.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.2389, pruned_loss=0.1594, over 1087589.92 frames. ], batch size: 8, lr: 3.87e-02, grad_scale: 16.0 +2022-11-15 14:29:03,243 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 2.211e+02 2.874e+02 3.707e+02 9.191e+02, threshold=5.749e+02, percent-clipped=7.0 +2022-11-15 14:29:17,317 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-15 14:29:33,908 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 14:29:34,899 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:29:45,632 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3969, 4.1917, 4.5218, 3.8946, 4.7005, 4.3336, 4.2036, 4.2747], + device='cuda:3'), covar=tensor([0.0513, 0.0344, 0.0450, 0.0485, 0.0613, 0.0246, 0.0280, 0.0319], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0055, 0.0044, 0.0053, 0.0056, 0.0039, 0.0045, 0.0041], + device='cuda:3'), out_proj_covar=tensor([9.7702e-05, 9.5060e-05, 7.7671e-05, 8.9749e-05, 1.1653e-04, 6.4655e-05, + 7.8951e-05, 7.3615e-05], device='cuda:3') +2022-11-15 14:30:06,624 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:30:13,990 INFO [train.py:876] (3/4) Epoch 1, batch 6800, loss[loss=0.2606, simple_loss=0.2307, pruned_loss=0.1452, over 5694.00 frames. ], tot_loss[loss=0.278, simple_loss=0.2391, pruned_loss=0.1585, over 1089644.44 frames. ], batch size: 28, lr: 3.85e-02, grad_scale: 16.0 +2022-11-15 14:30:15,299 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.052e+02 2.561e+02 3.297e+02 6.876e+02, threshold=5.122e+02, percent-clipped=2.0 +2022-11-15 14:30:23,253 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8070, 2.9068, 1.2923, 1.3215, 2.4598, 2.6363, 1.9070, 2.3679], + device='cuda:3'), covar=tensor([0.0564, 0.0158, 0.0582, 0.0614, 0.0171, 0.0161, 0.0223, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0054, 0.0047, 0.0064, 0.0042, 0.0038, 0.0040, 0.0045], + device='cuda:3'), out_proj_covar=tensor([9.5354e-05, 6.5243e-05, 6.4475e-05, 8.0888e-05, 5.2946e-05, 4.5264e-05, + 5.1047e-05, 5.4066e-05], device='cuda:3') +2022-11-15 14:30:26,402 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 14:30:30,176 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:30:32,315 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:30:41,204 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:30:56,786 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 14:30:57,025 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:30:57,695 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3115, 4.7481, 4.3301, 3.9505, 4.1623, 4.4080, 2.4012, 4.6186], + device='cuda:3'), covar=tensor([0.0260, 0.0131, 0.0168, 0.0241, 0.0197, 0.0209, 0.0959, 0.0109], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0034, 0.0032, 0.0027, 0.0032, 0.0028, 0.0055, 0.0034], + device='cuda:3'), out_proj_covar=tensor([6.0984e-05, 5.5080e-05, 5.0828e-05, 4.3669e-05, 5.0887e-05, 4.6605e-05, + 8.7130e-05, 5.5661e-05], device='cuda:3') +2022-11-15 14:31:02,562 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:07,533 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:10,667 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:31:15,879 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 14:31:24,704 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.97 vs. limit=5.0 +2022-11-15 14:31:26,639 INFO [train.py:876] (3/4) Epoch 1, batch 6900, loss[loss=0.2588, simple_loss=0.227, pruned_loss=0.1453, over 5714.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.238, pruned_loss=0.1576, over 1089289.48 frames. ], batch size: 12, lr: 3.83e-02, grad_scale: 16.0 +2022-11-15 14:31:27,996 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.317e+02 3.048e+02 4.158e+02 6.462e+02, threshold=6.096e+02, percent-clipped=10.0 +2022-11-15 14:31:30,929 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:32,402 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:36,820 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:41,775 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:31:54,350 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:32:17,042 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:32:17,179 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:32:27,736 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:32:39,433 INFO [train.py:876] (3/4) Epoch 1, batch 7000, loss[loss=0.2806, simple_loss=0.2439, pruned_loss=0.1586, over 5723.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.238, pruned_loss=0.1575, over 1083682.43 frames. ], batch size: 15, lr: 3.81e-02, grad_scale: 16.0 +2022-11-15 14:32:40,783 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 2.319e+02 2.855e+02 3.574e+02 7.700e+02, threshold=5.709e+02, percent-clipped=2.0 +2022-11-15 14:32:53,886 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3546, 1.7621, 1.3474, 1.5838, 1.6358, 1.7836, 1.4143, 1.6456], + device='cuda:3'), covar=tensor([0.0202, 0.0236, 0.0263, 0.0165, 0.0126, 0.0146, 0.0283, 0.0145], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0013, 0.0016, 0.0018, 0.0016, 0.0016, 0.0019, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.0788e-05, 1.7106e-05, 2.1004e-05, 2.1496e-05, 2.0772e-05, 1.9326e-05, + 2.5813e-05, 1.9105e-05], device='cuda:3') +2022-11-15 14:33:04,917 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4421, 4.0379, 3.8805, 4.1040, 3.7796, 3.0556, 1.9606, 3.8685], + device='cuda:3'), covar=tensor([0.1969, 0.0186, 0.0287, 0.0108, 0.0233, 0.0831, 0.3487, 0.0164], + device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0080, 0.0096, 0.0068, 0.0081, 0.0121, 0.0167, 0.0078], + device='cuda:3'), out_proj_covar=tensor([1.6108e-04, 8.3847e-05, 1.0447e-04, 7.2861e-05, 9.3034e-05, 1.3772e-04, + 1.8394e-04, 8.1929e-05], device='cuda:3') +2022-11-15 14:33:13,386 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:33:35,200 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2669, 3.9910, 3.7747, 3.8192, 3.7050, 3.1255, 1.9699, 3.8544], + device='cuda:3'), covar=tensor([0.1948, 0.0118, 0.0299, 0.0215, 0.0162, 0.0685, 0.3175, 0.0133], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0077, 0.0093, 0.0069, 0.0080, 0.0118, 0.0165, 0.0078], + device='cuda:3'), out_proj_covar=tensor([1.5877e-04, 8.1747e-05, 1.0180e-04, 7.4282e-05, 9.1888e-05, 1.3454e-04, + 1.8215e-04, 8.2148e-05], device='cuda:3') +2022-11-15 14:33:42,529 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6761, 1.5488, 1.2064, 0.7918, 0.9338, 0.5315, 0.8343, 0.6164], + device='cuda:3'), covar=tensor([0.0452, 0.0116, 0.0219, 0.0188, 0.0570, 0.0693, 0.0338, 0.0241], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0020, 0.0023, 0.0022, 0.0023, 0.0025, 0.0025, 0.0023], + device='cuda:3'), out_proj_covar=tensor([2.8656e-05, 2.4978e-05, 3.2241e-05, 2.5869e-05, 2.6983e-05, 2.9694e-05, + 3.8927e-05, 2.7272e-05], device='cuda:3') +2022-11-15 14:33:51,321 INFO [train.py:876] (3/4) Epoch 1, batch 7100, loss[loss=0.2759, simple_loss=0.2365, pruned_loss=0.1577, over 5634.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.2394, pruned_loss=0.1585, over 1090628.31 frames. ], batch size: 38, lr: 3.79e-02, grad_scale: 16.0 +2022-11-15 14:33:52,659 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 2.197e+02 2.721e+02 3.665e+02 9.993e+02, threshold=5.441e+02, percent-clipped=4.0 +2022-11-15 14:33:53,731 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.09 vs. limit=2.0 +2022-11-15 14:33:56,930 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:34:05,342 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:34:09,115 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:34:38,218 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:34:40,170 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:34:43,804 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:34:48,822 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0280, 2.3502, 2.2725, 2.1497, 2.1494, 2.1894, 1.3418, 2.1616], + device='cuda:3'), covar=tensor([0.0300, 0.0157, 0.0152, 0.0153, 0.0191, 0.0151, 0.0866, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0035, 0.0032, 0.0026, 0.0032, 0.0027, 0.0055, 0.0034], + device='cuda:3'), out_proj_covar=tensor([6.2174e-05, 5.8131e-05, 5.0695e-05, 4.2281e-05, 5.1435e-05, 4.4771e-05, + 8.8394e-05, 5.5779e-05], device='cuda:3') +2022-11-15 14:34:51,641 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7394, 1.5372, 1.2906, 1.0161, 1.4685, 0.7666, 0.8146, 0.7198], + device='cuda:3'), covar=tensor([0.0229, 0.0106, 0.0203, 0.0205, 0.0236, 0.0428, 0.0362, 0.0288], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0020, 0.0021, 0.0021, 0.0022, 0.0023, 0.0022, 0.0020], + device='cuda:3'), out_proj_covar=tensor([2.7315e-05, 2.4613e-05, 2.9010e-05, 2.5059e-05, 2.5849e-05, 2.8093e-05, + 3.5571e-05, 2.4775e-05], device='cuda:3') +2022-11-15 14:35:05,693 INFO [train.py:876] (3/4) Epoch 1, batch 7200, loss[loss=0.2425, simple_loss=0.231, pruned_loss=0.127, over 5769.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.2388, pruned_loss=0.1577, over 1090692.07 frames. ], batch size: 16, lr: 3.78e-02, grad_scale: 16.0 +2022-11-15 14:35:07,001 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 2.253e+02 2.788e+02 3.499e+02 9.174e+02, threshold=5.576e+02, percent-clipped=8.0 +2022-11-15 14:35:20,655 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8862, 1.3565, 0.9712, 0.7236, 1.4181, 0.7570, 0.8898, 0.6907], + device='cuda:3'), covar=tensor([0.0333, 0.0110, 0.0183, 0.0266, 0.0190, 0.0360, 0.0329, 0.0188], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0020, 0.0020, 0.0021, 0.0021, 0.0023, 0.0022, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.7315e-05, 2.4504e-05, 2.8030e-05, 2.4912e-05, 2.5123e-05, 2.7522e-05, + 3.5047e-05, 2.4802e-05], device='cuda:3') +2022-11-15 14:35:22,002 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:35:29,033 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 14:35:50,132 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:37:25,313 INFO [train.py:876] (3/4) Epoch 2, batch 0, loss[loss=0.2808, simple_loss=0.2289, pruned_loss=0.1663, over 5106.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.2289, pruned_loss=0.1663, over 5106.00 frames. ], batch size: 91, lr: 3.69e-02, grad_scale: 16.0 +2022-11-15 14:37:25,313 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 14:37:42,502 INFO [train.py:908] (3/4) Epoch 2, validation: loss=0.2258, simple_loss=0.228, pruned_loss=0.1118, over 1530663.00 frames. +2022-11-15 14:37:42,502 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4504MB +2022-11-15 14:37:44,107 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:37:51,155 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:38:02,130 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.73 vs. limit=5.0 +2022-11-15 14:38:04,585 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.115e+02 2.889e+02 4.195e+02 1.182e+03, threshold=5.778e+02, percent-clipped=11.0 +2022-11-15 14:38:26,542 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:38:26,623 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7566, 4.6240, 4.2855, 4.3794, 4.1676, 3.3869, 2.6778, 4.2861], + device='cuda:3'), covar=tensor([0.2147, 0.0119, 0.0355, 0.0306, 0.0160, 0.0882, 0.3291, 0.0121], + device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0077, 0.0096, 0.0072, 0.0081, 0.0124, 0.0172, 0.0078], + device='cuda:3'), out_proj_covar=tensor([1.6774e-04, 8.0989e-05, 1.0739e-04, 8.0006e-05, 9.3821e-05, 1.4243e-04, + 1.8970e-04, 8.1760e-05], device='cuda:3') +2022-11-15 14:38:28,804 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:38:48,611 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.51 vs. limit=5.0 +2022-11-15 14:38:55,140 INFO [train.py:876] (3/4) Epoch 2, batch 100, loss[loss=0.1937, simple_loss=0.191, pruned_loss=0.09825, over 5078.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.2391, pruned_loss=0.1571, over 429271.55 frames. ], batch size: 7, lr: 3.67e-02, grad_scale: 16.0 +2022-11-15 14:39:10,796 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 14:39:17,529 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.188e+01 2.195e+02 2.755e+02 3.428e+02 7.515e+02, threshold=5.510e+02, percent-clipped=5.0 +2022-11-15 14:39:18,281 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:39:20,776 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:39:36,990 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9875, 1.9009, 3.6075, 2.4987, 2.4365, 2.6962, 3.3935, 3.6627], + device='cuda:3'), covar=tensor([0.0153, 0.1372, 0.0227, 0.0837, 0.0343, 0.0616, 0.0310, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0108, 0.0058, 0.0087, 0.0051, 0.0077, 0.0065, 0.0058], + device='cuda:3'), out_proj_covar=tensor([5.7725e-05, 1.3897e-04, 7.6691e-05, 1.1077e-04, 6.7795e-05, 9.9404e-05, + 8.5219e-05, 7.6899e-05], device='cuda:3') +2022-11-15 14:39:46,287 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9539, 1.8838, 2.0902, 3.0265, 2.9294, 2.0774, 2.0691, 3.4235], + device='cuda:3'), covar=tensor([0.0203, 0.1568, 0.1489, 0.0191, 0.0310, 0.1333, 0.1496, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0137, 0.0140, 0.0068, 0.0083, 0.0129, 0.0143, 0.0070], + device='cuda:3'), out_proj_covar=tensor([7.8800e-05, 1.6241e-04, 1.6240e-04, 8.3937e-05, 9.3688e-05, 1.5411e-04, + 1.6640e-04, 7.4969e-05], device='cuda:3') +2022-11-15 14:40:05,196 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:40:08,512 INFO [train.py:876] (3/4) Epoch 2, batch 200, loss[loss=0.3628, simple_loss=0.2885, pruned_loss=0.2185, over 5674.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.2339, pruned_loss=0.1538, over 690207.60 frames. ], batch size: 34, lr: 3.66e-02, grad_scale: 16.0 +2022-11-15 14:40:27,226 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 14:40:30,170 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.136e+02 2.623e+02 3.249e+02 5.222e+02, threshold=5.245e+02, percent-clipped=0.0 +2022-11-15 14:40:41,701 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:40:48,382 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9381, 3.9048, 4.0891, 4.0718, 3.5336, 2.9013, 4.5184, 3.7943], + device='cuda:3'), covar=tensor([0.0497, 0.0809, 0.0395, 0.0566, 0.0534, 0.0589, 0.0686, 0.0445], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0071, 0.0059, 0.0065, 0.0043, 0.0039, 0.0064, 0.0051], + device='cuda:3'), out_proj_covar=tensor([8.0370e-05, 1.2615e-04, 1.0323e-04, 1.1422e-04, 7.7265e-05, 6.6729e-05, + 1.2734e-04, 8.9300e-05], device='cuda:3') +2022-11-15 14:40:52,533 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:40:55,354 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7302, 1.0083, 0.9161, 1.1357, 1.3052, 1.3092, 0.8269, 0.6967], + device='cuda:3'), covar=tensor([0.0190, 0.0091, 0.0123, 0.0112, 0.0107, 0.0105, 0.0310, 0.0279], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0021, 0.0020, 0.0021, 0.0021, 0.0021, 0.0022, 0.0023], + device='cuda:3'), out_proj_covar=tensor([2.9355e-05, 2.5176e-05, 2.8147e-05, 2.4638e-05, 2.5168e-05, 2.6190e-05, + 3.6652e-05, 2.7653e-05], device='cuda:3') +2022-11-15 14:41:00,557 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7436, 0.7936, 0.9210, 0.5464, 0.8328, 0.9289, 0.6436, 0.4873], + device='cuda:3'), covar=tensor([0.0313, 0.0251, 0.0185, 0.0593, 0.0508, 0.0197, 0.0647, 0.0330], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0019, 0.0024, 0.0019, 0.0018, 0.0021, 0.0020], + device='cuda:3'), out_proj_covar=tensor([3.0025e-05, 2.9250e-05, 2.8416e-05, 3.8791e-05, 2.8870e-05, 2.5837e-05, + 3.3406e-05, 2.7953e-05], device='cuda:3') +2022-11-15 14:41:14,938 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:41:20,233 INFO [train.py:876] (3/4) Epoch 2, batch 300, loss[loss=0.3251, simple_loss=0.2805, pruned_loss=0.1849, over 5650.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.2369, pruned_loss=0.1562, over 847434.27 frames. ], batch size: 32, lr: 3.64e-02, grad_scale: 16.0 +2022-11-15 14:41:27,233 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:41:42,306 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 2.168e+02 2.646e+02 3.466e+02 1.431e+03, threshold=5.292e+02, percent-clipped=6.0 +2022-11-15 14:41:49,771 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:41:57,611 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:42:02,266 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:42:19,449 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.70 vs. limit=5.0 +2022-11-15 14:42:33,000 INFO [train.py:876] (3/4) Epoch 2, batch 400, loss[loss=0.2573, simple_loss=0.2348, pruned_loss=0.1398, over 5806.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.2348, pruned_loss=0.1542, over 943259.73 frames. ], batch size: 21, lr: 3.62e-02, grad_scale: 16.0 +2022-11-15 14:42:41,555 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:42:43,929 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-15 14:42:54,873 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 2.314e+02 2.984e+02 3.754e+02 8.890e+02, threshold=5.969e+02, percent-clipped=7.0 +2022-11-15 14:42:55,788 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:43:08,837 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9585, 0.8832, 1.2145, 0.8280, 1.3153, 1.2613, 0.7722, 0.8382], + device='cuda:3'), covar=tensor([0.0634, 0.0178, 0.0250, 0.0152, 0.0116, 0.0125, 0.0294, 0.0197], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0020, 0.0021, 0.0021, 0.0021, 0.0024, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.8451e-05, 2.6575e-05, 2.7785e-05, 2.4141e-05, 2.4300e-05, 2.6041e-05, + 3.7828e-05, 2.7154e-05], device='cuda:3') +2022-11-15 14:43:13,951 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.20 vs. limit=2.0 +2022-11-15 14:43:16,667 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 14:43:23,439 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:43:30,222 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 14:43:37,925 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:43:44,742 INFO [train.py:876] (3/4) Epoch 2, batch 500, loss[loss=0.2606, simple_loss=0.2272, pruned_loss=0.147, over 5594.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.2342, pruned_loss=0.153, over 1005001.91 frames. ], batch size: 23, lr: 3.61e-02, grad_scale: 16.0 +2022-11-15 14:43:52,091 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 14:44:02,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4226, 3.4012, 2.2684, 1.5719, 3.1291, 1.9399, 3.1088, 2.0656], + device='cuda:3'), covar=tensor([0.0729, 0.0189, 0.0324, 0.2079, 0.0230, 0.0902, 0.0184, 0.1025], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0043, 0.0035, 0.0082, 0.0041, 0.0058, 0.0035, 0.0065], + device='cuda:3'), out_proj_covar=tensor([1.2319e-04, 7.7628e-05, 6.7721e-05, 1.4509e-04, 7.3186e-05, 1.0759e-04, + 6.9402e-05, 1.2240e-04], device='cuda:3') +2022-11-15 14:44:06,104 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:44:06,563 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.366e+02 3.140e+02 3.903e+02 7.653e+02, threshold=6.280e+02, percent-clipped=5.0 +2022-11-15 14:44:07,835 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:44:18,476 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 14:44:35,205 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1197, 0.9367, 1.1870, 0.5229, 1.0244, 0.8363, 0.7880, 1.2989], + device='cuda:3'), covar=tensor([0.0438, 0.0209, 0.0151, 0.0526, 0.0521, 0.0457, 0.0403, 0.0230], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0017, 0.0021, 0.0015, 0.0017, 0.0019, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.6895e-05, 2.5722e-05, 2.4853e-05, 3.4836e-05, 2.4232e-05, 2.4885e-05, + 2.9244e-05, 2.3733e-05], device='cuda:3') +2022-11-15 14:44:35,900 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7890, 1.1780, 1.5269, 1.8933, 1.0840, 1.2796, 1.0183, 1.8935], + device='cuda:3'), covar=tensor([0.0127, 0.0237, 0.0250, 0.0094, 0.0282, 0.0210, 0.0213, 0.0161], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0031, 0.0036, 0.0029, 0.0042, 0.0030, 0.0037, 0.0025], + device='cuda:3'), out_proj_covar=tensor([4.0124e-05, 4.4094e-05, 5.7490e-05, 4.2328e-05, 6.7298e-05, 5.0455e-05, + 5.7319e-05, 3.7839e-05], device='cuda:3') +2022-11-15 14:44:37,967 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5920, 1.5089, 1.6259, 1.2006, 1.0354, 1.3453, 1.3091, 1.2703], + device='cuda:3'), covar=tensor([0.0172, 0.0246, 0.0169, 0.0211, 0.0500, 0.1277, 0.0341, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0018, 0.0019, 0.0020, 0.0019, 0.0016, 0.0019, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.6623e-05, 2.2791e-05, 2.3817e-05, 2.7667e-05, 2.8393e-05, 2.5079e-05, + 2.6271e-05, 2.6786e-05], device='cuda:3') +2022-11-15 14:44:50,043 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:44:52,616 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:44:56,654 INFO [train.py:876] (3/4) Epoch 2, batch 600, loss[loss=0.2595, simple_loss=0.2309, pruned_loss=0.1441, over 5730.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.2354, pruned_loss=0.1542, over 1034301.81 frames. ], batch size: 16, lr: 3.59e-02, grad_scale: 16.0 +2022-11-15 14:45:05,680 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 14:45:18,185 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.116e+02 2.659e+02 3.486e+02 9.417e+02, threshold=5.318e+02, percent-clipped=5.0 +2022-11-15 14:45:38,478 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:45:53,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8711, 4.4082, 4.7918, 4.0099, 4.9410, 4.8030, 4.3314, 4.3128], + device='cuda:3'), covar=tensor([0.0393, 0.0416, 0.0497, 0.0470, 0.0471, 0.0132, 0.0255, 0.0441], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0063, 0.0050, 0.0060, 0.0058, 0.0039, 0.0047, 0.0048], + device='cuda:3'), out_proj_covar=tensor([1.1986e-04, 1.1604e-04, 9.5239e-05, 1.0906e-04, 1.2952e-04, 7.1417e-05, + 8.9076e-05, 9.3362e-05], device='cuda:3') +2022-11-15 14:46:08,129 INFO [train.py:876] (3/4) Epoch 2, batch 700, loss[loss=0.2834, simple_loss=0.2602, pruned_loss=0.1533, over 5758.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.235, pruned_loss=0.1531, over 1053302.59 frames. ], batch size: 21, lr: 3.57e-02, grad_scale: 16.0 +2022-11-15 14:46:12,995 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:46:13,702 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:46:13,924 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2022-11-15 14:46:16,360 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 14:46:29,648 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4062, 1.9704, 1.2266, 1.2732, 1.5776, 1.4021, 1.1733, 1.3062], + device='cuda:3'), covar=tensor([0.0224, 0.0216, 0.0227, 0.0206, 0.0343, 0.0488, 0.0348, 0.0256], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0018, 0.0018, 0.0020, 0.0018, 0.0015, 0.0018, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.7655e-05, 2.3079e-05, 2.2786e-05, 2.8410e-05, 2.7662e-05, 2.3796e-05, + 2.5310e-05, 2.6283e-05], device='cuda:3') +2022-11-15 14:46:30,155 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.448e+02 3.316e+02 4.282e+02 8.235e+02, threshold=6.631e+02, percent-clipped=7.0 +2022-11-15 14:46:58,085 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 14:47:08,695 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3406, 1.9373, 0.6056, 1.0089, 1.3510, 2.1489, 1.8612, 2.0334], + device='cuda:3'), covar=tensor([0.0757, 0.0338, 0.0621, 0.0883, 0.0273, 0.0109, 0.0172, 0.0151], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0071, 0.0052, 0.0081, 0.0045, 0.0044, 0.0044, 0.0052], + device='cuda:3'), out_proj_covar=tensor([1.1318e-04, 8.8860e-05, 7.3289e-05, 1.0399e-04, 5.7284e-05, 5.6374e-05, + 5.7217e-05, 6.2818e-05], device='cuda:3') +2022-11-15 14:47:09,683 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2022-11-15 14:47:13,901 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:47:19,582 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 14:47:20,641 INFO [train.py:876] (3/4) Epoch 2, batch 800, loss[loss=0.2306, simple_loss=0.2141, pruned_loss=0.1236, over 5722.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.2338, pruned_loss=0.1523, over 1068840.22 frames. ], batch size: 13, lr: 3.56e-02, grad_scale: 16.0 +2022-11-15 14:47:39,440 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:47:42,123 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.291e+02 2.781e+02 3.438e+02 1.081e+03, threshold=5.561e+02, percent-clipped=3.0 +2022-11-15 14:47:42,633 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-15 14:47:46,500 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2022-11-15 14:47:47,266 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 14:47:48,556 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:48:22,106 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:48:32,986 INFO [train.py:876] (3/4) Epoch 2, batch 900, loss[loss=0.2707, simple_loss=0.2293, pruned_loss=0.1561, over 5642.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.2361, pruned_loss=0.154, over 1074385.15 frames. ], batch size: 29, lr: 3.54e-02, grad_scale: 16.0 +2022-11-15 14:48:47,795 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-15 14:48:51,444 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7002, 1.4568, 1.3949, 1.6988, 1.2140, 1.6563, 1.5523, 1.3657], + device='cuda:3'), covar=tensor([0.0113, 0.0049, 0.0069, 0.0039, 0.0194, 0.0043, 0.0092, 0.0055], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0040, 0.0040, 0.0038, 0.0066, 0.0040, 0.0053, 0.0034], + device='cuda:3'), out_proj_covar=tensor([8.4708e-05, 5.0657e-05, 5.1133e-05, 5.4784e-05, 9.1507e-05, 4.9226e-05, + 6.7537e-05, 4.3253e-05], device='cuda:3') +2022-11-15 14:48:54,941 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 2.386e+02 2.834e+02 3.764e+02 8.164e+02, threshold=5.667e+02, percent-clipped=3.0 +2022-11-15 14:49:36,785 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8799, 2.2003, 2.1561, 1.4606, 2.9283, 2.0882, 2.2657, 2.8632], + device='cuda:3'), covar=tensor([0.1126, 0.0637, 0.0606, 0.1108, 0.0164, 0.0208, 0.0280, 0.0207], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0074, 0.0051, 0.0082, 0.0045, 0.0045, 0.0047, 0.0053], + device='cuda:3'), out_proj_covar=tensor([1.1472e-04, 9.3368e-05, 7.4027e-05, 1.0520e-04, 5.8138e-05, 5.8668e-05, + 6.1430e-05, 6.5668e-05], device='cuda:3') +2022-11-15 14:49:39,603 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:49:44,958 INFO [train.py:876] (3/4) Epoch 2, batch 1000, loss[loss=0.3195, simple_loss=0.2691, pruned_loss=0.185, over 5576.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.2346, pruned_loss=0.1518, over 1077066.24 frames. ], batch size: 50, lr: 3.53e-02, grad_scale: 16.0 +2022-11-15 14:49:50,706 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:50:03,940 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.53 vs. limit=2.0 +2022-11-15 14:50:07,378 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.298e+02 2.271e+02 2.772e+02 3.875e+02 7.231e+02, threshold=5.545e+02, percent-clipped=6.0 +2022-11-15 14:50:14,461 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5073, 0.9126, 0.6506, 0.9449, 0.7374, 0.7616, 0.5330, 0.6018], + device='cuda:3'), covar=tensor([0.0113, 0.0050, 0.0086, 0.0062, 0.0050, 0.0073, 0.0138, 0.0066], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0020, 0.0019, 0.0020, 0.0020, 0.0020, 0.0021, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.6143e-05, 2.5349e-05, 2.6667e-05, 2.4453e-05, 2.4578e-05, 2.5556e-05, + 3.5098e-05, 2.4538e-05], device='cuda:3') +2022-11-15 14:50:23,538 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:50:24,724 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:50:45,130 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-15 14:50:57,473 INFO [train.py:876] (3/4) Epoch 2, batch 1100, loss[loss=0.2362, simple_loss=0.2208, pruned_loss=0.1258, over 5785.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.2334, pruned_loss=0.1512, over 1078907.31 frames. ], batch size: 16, lr: 3.51e-02, grad_scale: 16.0 +2022-11-15 14:51:09,111 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9415, 2.9000, 2.5435, 2.8187, 2.3827, 2.6213, 2.6968, 2.9885], + device='cuda:3'), covar=tensor([0.0463, 0.0113, 0.0119, 0.0178, 0.0356, 0.0094, 0.0215, 0.0069], + device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0043, 0.0041, 0.0042, 0.0071, 0.0044, 0.0058, 0.0037], + device='cuda:3'), out_proj_covar=tensor([9.3277e-05, 5.5599e-05, 5.2397e-05, 5.9755e-05, 1.0014e-04, 5.4914e-05, + 7.6254e-05, 4.6283e-05], device='cuda:3') +2022-11-15 14:51:16,583 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:51:19,489 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.261e+02 2.575e+02 3.836e+02 7.235e+02, threshold=5.150e+02, percent-clipped=6.0 +2022-11-15 14:51:41,752 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3146, 4.2955, 4.1294, 3.8225, 4.2063, 4.2100, 2.1831, 3.5482], + device='cuda:3'), covar=tensor([0.0268, 0.0337, 0.0270, 0.0243, 0.0218, 0.0268, 0.1857, 0.0469], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0042, 0.0038, 0.0031, 0.0042, 0.0034, 0.0072, 0.0044], + device='cuda:3'), out_proj_covar=tensor([7.8204e-05, 7.6347e-05, 6.5622e-05, 5.3041e-05, 7.1974e-05, 6.0756e-05, + 1.2108e-04, 7.7426e-05], device='cuda:3') +2022-11-15 14:51:51,358 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:51:58,971 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:52:09,820 INFO [train.py:876] (3/4) Epoch 2, batch 1200, loss[loss=0.2882, simple_loss=0.2454, pruned_loss=0.1655, over 5443.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.2332, pruned_loss=0.1515, over 1075045.21 frames. ], batch size: 53, lr: 3.50e-02, grad_scale: 16.0 +2022-11-15 14:52:31,193 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.113e+02 2.806e+02 3.522e+02 6.703e+02, threshold=5.613e+02, percent-clipped=5.0 +2022-11-15 14:52:33,373 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:52:43,511 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6854, 4.2326, 2.5714, 1.0408, 2.9008, 3.8160, 2.8691, 3.4406], + device='cuda:3'), covar=tensor([0.0560, 0.0111, 0.0248, 0.0753, 0.0098, 0.0051, 0.0124, 0.0118], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0073, 0.0055, 0.0086, 0.0050, 0.0046, 0.0047, 0.0058], + device='cuda:3'), out_proj_covar=tensor([1.1996e-04, 9.3262e-05, 7.9221e-05, 1.1155e-04, 6.4702e-05, 5.8926e-05, + 6.0688e-05, 7.3344e-05], device='cuda:3') +2022-11-15 14:52:46,022 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4014, 3.0938, 3.1561, 3.2612, 2.4025, 3.5302, 2.8878, 3.0764], + device='cuda:3'), covar=tensor([0.0450, 0.0217, 0.0107, 0.0245, 0.0473, 0.0098, 0.0227, 0.0071], + device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0044, 0.0042, 0.0042, 0.0075, 0.0045, 0.0060, 0.0039], + device='cuda:3'), out_proj_covar=tensor([9.5480e-05, 5.7427e-05, 5.4526e-05, 6.1714e-05, 1.0454e-04, 5.7114e-05, + 8.0213e-05, 5.1152e-05], device='cuda:3') +2022-11-15 14:53:05,241 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 +2022-11-15 14:53:20,999 INFO [train.py:876] (3/4) Epoch 2, batch 1300, loss[loss=0.3059, simple_loss=0.2524, pruned_loss=0.1797, over 5441.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.2321, pruned_loss=0.1502, over 1075153.22 frames. ], batch size: 58, lr: 3.48e-02, grad_scale: 16.0 +2022-11-15 14:53:23,726 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2022-11-15 14:53:26,716 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1685, 4.1376, 3.7601, 3.9786, 3.4931, 2.8944, 2.0254, 3.6224], + device='cuda:3'), covar=tensor([0.1979, 0.0148, 0.0393, 0.0203, 0.0291, 0.1064, 0.3036, 0.0212], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0087, 0.0108, 0.0077, 0.0092, 0.0132, 0.0177, 0.0080], + device='cuda:3'), out_proj_covar=tensor([1.7801e-04, 9.4020e-05, 1.2615e-04, 8.6786e-05, 1.0856e-04, 1.5781e-04, + 1.9686e-04, 8.8181e-05], device='cuda:3') +2022-11-15 14:53:33,452 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9675, 2.7666, 2.9981, 2.6736, 3.1309, 2.7547, 2.8293, 2.8777], + device='cuda:3'), covar=tensor([0.0443, 0.0327, 0.0367, 0.0379, 0.0353, 0.0232, 0.0303, 0.0380], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0062, 0.0050, 0.0061, 0.0056, 0.0039, 0.0049, 0.0049], + device='cuda:3'), out_proj_covar=tensor([1.2487e-04, 1.1753e-04, 9.8890e-05, 1.1291e-04, 1.2689e-04, 7.0620e-05, + 9.3063e-05, 9.8225e-05], device='cuda:3') +2022-11-15 14:53:42,779 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 14:53:42,885 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 2.077e+02 2.771e+02 3.615e+02 8.724e+02, threshold=5.542e+02, percent-clipped=7.0 +2022-11-15 14:53:45,260 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7533, 4.2550, 4.1482, 4.2156, 3.7623, 3.1826, 2.6112, 3.7257], + device='cuda:3'), covar=tensor([0.1743, 0.0193, 0.0264, 0.0176, 0.0233, 0.0897, 0.2467, 0.0238], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0087, 0.0108, 0.0076, 0.0091, 0.0132, 0.0174, 0.0080], + device='cuda:3'), out_proj_covar=tensor([1.7822e-04, 9.4153e-05, 1.2602e-04, 8.5838e-05, 1.0749e-04, 1.5698e-04, + 1.9330e-04, 8.8487e-05], device='cuda:3') +2022-11-15 14:53:45,962 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0352, 4.2763, 4.0520, 3.9228, 4.0079, 4.0895, 2.1435, 3.8906], + device='cuda:3'), covar=tensor([0.0278, 0.0286, 0.0190, 0.0222, 0.0235, 0.0262, 0.1547, 0.0289], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0042, 0.0039, 0.0033, 0.0043, 0.0035, 0.0074, 0.0043], + device='cuda:3'), out_proj_covar=tensor([7.9588e-05, 7.5735e-05, 6.7077e-05, 5.5749e-05, 7.5807e-05, 6.2207e-05, + 1.2449e-04, 7.7203e-05], device='cuda:3') +2022-11-15 14:53:56,861 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:54:03,697 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.13 vs. limit=2.0 +2022-11-15 14:54:14,557 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 +2022-11-15 14:54:16,072 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 +2022-11-15 14:54:34,443 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2022-11-15 14:54:35,336 INFO [train.py:876] (3/4) Epoch 2, batch 1400, loss[loss=0.2029, simple_loss=0.1743, pruned_loss=0.1157, over 4570.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.2295, pruned_loss=0.148, over 1078453.33 frames. ], batch size: 5, lr: 3.46e-02, grad_scale: 32.0 +2022-11-15 14:54:42,919 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 14:54:52,864 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7856, 1.2910, 1.3726, 1.9552, 1.3417, 1.1702, 1.3106, 1.5981], + device='cuda:3'), covar=tensor([0.0157, 0.0299, 0.0342, 0.0245, 0.0308, 0.0281, 0.0284, 0.0194], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0032, 0.0035, 0.0028, 0.0040, 0.0029, 0.0039, 0.0025], + device='cuda:3'), out_proj_covar=tensor([4.3678e-05, 4.6599e-05, 5.7732e-05, 4.1460e-05, 6.9623e-05, 4.9700e-05, + 6.2003e-05, 4.1764e-05], device='cuda:3') +2022-11-15 14:54:56,930 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 2.372e+02 3.042e+02 3.801e+02 7.959e+02, threshold=6.083e+02, percent-clipped=7.0 +2022-11-15 14:55:34,699 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.29 vs. limit=5.0 +2022-11-15 14:55:35,074 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8197, 4.2043, 4.1695, 4.2739, 3.8805, 3.3962, 4.6768, 4.1168], + device='cuda:3'), covar=tensor([0.0553, 0.0734, 0.0446, 0.0540, 0.0546, 0.0393, 0.0630, 0.0457], + device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0072, 0.0061, 0.0071, 0.0046, 0.0041, 0.0070, 0.0055], + device='cuda:3'), out_proj_covar=tensor([9.1767e-05, 1.3497e-04, 1.1513e-04, 1.3048e-04, 8.8085e-05, 7.5359e-05, + 1.4758e-04, 1.0232e-04], device='cuda:3') +2022-11-15 14:55:37,213 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:55:46,191 INFO [train.py:876] (3/4) Epoch 2, batch 1500, loss[loss=0.2373, simple_loss=0.2192, pruned_loss=0.1277, over 5587.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.2297, pruned_loss=0.147, over 1084016.54 frames. ], batch size: 23, lr: 3.45e-02, grad_scale: 32.0 +2022-11-15 14:56:08,252 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.321e+02 2.844e+02 3.403e+02 6.170e+02, threshold=5.688e+02, percent-clipped=1.0 +2022-11-15 14:56:20,980 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:56:23,127 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0715, 1.9468, 3.4123, 2.3170, 2.9594, 2.2426, 3.1488, 3.4167], + device='cuda:3'), covar=tensor([0.0092, 0.0989, 0.0165, 0.0756, 0.0177, 0.0715, 0.0237, 0.0208], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0119, 0.0065, 0.0103, 0.0060, 0.0098, 0.0078, 0.0067], + device='cuda:3'), out_proj_covar=tensor([7.4685e-05, 1.6176e-04, 9.2225e-05, 1.4008e-04, 8.5332e-05, 1.3552e-04, + 1.0975e-04, 9.5684e-05], device='cuda:3') +2022-11-15 14:56:57,531 INFO [train.py:876] (3/4) Epoch 2, batch 1600, loss[loss=0.2957, simple_loss=0.2557, pruned_loss=0.1678, over 5841.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.228, pruned_loss=0.1447, over 1085255.57 frames. ], batch size: 22, lr: 3.44e-02, grad_scale: 16.0 +2022-11-15 14:57:07,968 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.4359, 1.0911, 1.1016, 1.2356, 1.1529, 1.2310, 0.7724, 0.8784], + device='cuda:3'), covar=tensor([0.0158, 0.0101, 0.0109, 0.0130, 0.0136, 0.0134, 0.0180, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0021, 0.0019, 0.0022, 0.0022, 0.0022, 0.0022, 0.0020], + device='cuda:3'), out_proj_covar=tensor([2.9468e-05, 2.6656e-05, 2.8311e-05, 2.5907e-05, 2.7579e-05, 2.7425e-05, + 3.6317e-05, 2.5920e-05], device='cuda:3') +2022-11-15 14:57:19,296 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.087e+02 2.971e+02 3.839e+02 7.053e+02, threshold=5.941e+02, percent-clipped=2.0 +2022-11-15 14:57:20,550 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 14:57:21,177 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 14:57:24,539 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 +2022-11-15 14:57:31,971 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:57:36,152 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:57:48,233 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.40 vs. limit=5.0 +2022-11-15 14:58:05,413 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:58:08,099 INFO [train.py:876] (3/4) Epoch 2, batch 1700, loss[loss=0.2554, simple_loss=0.228, pruned_loss=0.1414, over 5672.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.2299, pruned_loss=0.1466, over 1086787.81 frames. ], batch size: 19, lr: 3.42e-02, grad_scale: 16.0 +2022-11-15 14:58:17,144 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 14:58:18,145 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:58:19,536 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:58:30,484 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.407e+02 2.261e+02 2.879e+02 3.540e+02 8.492e+02, threshold=5.758e+02, percent-clipped=3.0 +2022-11-15 14:58:37,635 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:58:44,842 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.67 vs. limit=5.0 +2022-11-15 14:59:01,682 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:59:20,368 INFO [train.py:876] (3/4) Epoch 2, batch 1800, loss[loss=0.2184, simple_loss=0.2067, pruned_loss=0.115, over 5799.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.2281, pruned_loss=0.1453, over 1085805.17 frames. ], batch size: 21, lr: 3.41e-02, grad_scale: 16.0 +2022-11-15 14:59:21,216 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 14:59:35,114 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 14:59:42,158 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 2.362e+02 3.022e+02 3.932e+02 1.031e+03, threshold=6.044e+02, percent-clipped=5.0 +2022-11-15 14:59:49,560 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4038, 1.7117, 1.2433, 1.5657, 0.9314, 1.4510, 1.5523, 1.1040], + device='cuda:3'), covar=tensor([0.0227, 0.0347, 0.0195, 0.0225, 0.0454, 0.0597, 0.0349, 0.0325], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0022, 0.0023, 0.0025, 0.0024, 0.0020, 0.0022, 0.0021], + device='cuda:3'), out_proj_covar=tensor([3.7066e-05, 3.0084e-05, 3.0601e-05, 3.6096e-05, 3.8167e-05, 3.2470e-05, + 3.2119e-05, 3.1644e-05], device='cuda:3') +2022-11-15 14:59:50,855 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:00:31,253 INFO [train.py:876] (3/4) Epoch 2, batch 1900, loss[loss=0.3122, simple_loss=0.2515, pruned_loss=0.1865, over 4999.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.2274, pruned_loss=0.144, over 1091582.40 frames. ], batch size: 109, lr: 3.39e-02, grad_scale: 16.0 +2022-11-15 15:00:53,891 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 2.307e+02 3.025e+02 3.862e+02 6.126e+02, threshold=6.049e+02, percent-clipped=1.0 +2022-11-15 15:01:36,219 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 +2022-11-15 15:01:42,871 INFO [train.py:876] (3/4) Epoch 2, batch 2000, loss[loss=0.3455, simple_loss=0.2761, pruned_loss=0.2075, over 5484.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.2299, pruned_loss=0.1465, over 1083624.25 frames. ], batch size: 64, lr: 3.38e-02, grad_scale: 16.0 +2022-11-15 15:01:50,670 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:02:05,762 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.305e+02 2.275e+02 2.942e+02 3.786e+02 7.709e+02, threshold=5.884e+02, percent-clipped=5.0 +2022-11-15 15:02:18,599 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-15 15:02:32,852 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:02:46,824 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 +2022-11-15 15:02:52,308 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:02:54,976 INFO [train.py:876] (3/4) Epoch 2, batch 2100, loss[loss=0.2068, simple_loss=0.2103, pruned_loss=0.1017, over 5550.00 frames. ], tot_loss[loss=0.261, simple_loss=0.2295, pruned_loss=0.1462, over 1085033.59 frames. ], batch size: 14, lr: 3.36e-02, grad_scale: 16.0 +2022-11-15 15:03:08,320 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.3050, 0.9187, 1.0245, 1.2915, 1.0936, 1.0614, 0.5739, 1.0352], + device='cuda:3'), covar=tensor([0.0110, 0.0075, 0.0057, 0.0044, 0.0072, 0.0068, 0.0169, 0.0089], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0022, 0.0020, 0.0022, 0.0022, 0.0023, 0.0024, 0.0021], + device='cuda:3'), out_proj_covar=tensor([3.1049e-05, 2.8275e-05, 2.9701e-05, 2.6972e-05, 2.9262e-05, 2.9784e-05, + 4.0630e-05, 2.7325e-05], device='cuda:3') +2022-11-15 15:03:11,650 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0328, 1.5547, 0.7516, 1.2432, 0.6620, 1.5914, 1.3627, 1.0974], + device='cuda:3'), covar=tensor([0.0314, 0.0406, 0.0301, 0.0285, 0.0566, 0.0327, 0.0181, 0.0395], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0022, 0.0024, 0.0027, 0.0024, 0.0019, 0.0024, 0.0023], + device='cuda:3'), out_proj_covar=tensor([3.9787e-05, 3.1640e-05, 3.2999e-05, 3.8374e-05, 3.8645e-05, 3.1652e-05, + 3.4211e-05, 3.3891e-05], device='cuda:3') +2022-11-15 15:03:17,094 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 2.437e+02 2.901e+02 3.645e+02 9.793e+02, threshold=5.801e+02, percent-clipped=2.0 +2022-11-15 15:03:24,896 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:03:26,169 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:04:00,503 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:04:06,699 INFO [train.py:876] (3/4) Epoch 2, batch 2200, loss[loss=0.2345, simple_loss=0.2147, pruned_loss=0.1272, over 5747.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.2294, pruned_loss=0.146, over 1083222.23 frames. ], batch size: 14, lr: 3.35e-02, grad_scale: 16.0 +2022-11-15 15:04:08,202 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:04:28,437 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.967e+01 2.258e+02 2.836e+02 4.027e+02 8.312e+02, threshold=5.673e+02, percent-clipped=7.0 +2022-11-15 15:04:41,101 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0010, 4.2111, 4.3713, 4.2025, 3.8391, 3.4963, 4.5847, 4.0248], + device='cuda:3'), covar=tensor([0.0484, 0.0683, 0.0321, 0.0561, 0.0625, 0.0340, 0.0579, 0.0345], + device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0073, 0.0061, 0.0072, 0.0047, 0.0042, 0.0070, 0.0055], + device='cuda:3'), out_proj_covar=tensor([9.5130e-05, 1.4141e-04, 1.1632e-04, 1.3629e-04, 9.3233e-05, 7.9558e-05, + 1.5188e-04, 1.0394e-04], device='cuda:3') +2022-11-15 15:04:44,339 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 15:04:45,428 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2987, 4.0523, 4.2228, 3.8543, 4.4698, 3.9879, 4.0512, 4.2174], + device='cuda:3'), covar=tensor([0.0323, 0.0249, 0.0345, 0.0281, 0.0259, 0.0338, 0.0221, 0.0267], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0063, 0.0052, 0.0062, 0.0058, 0.0041, 0.0051, 0.0049], + device='cuda:3'), out_proj_covar=tensor([1.2913e-04, 1.2435e-04, 1.0496e-04, 1.1865e-04, 1.3333e-04, 7.7569e-05, + 1.0081e-04, 1.0202e-04], device='cuda:3') +2022-11-15 15:05:09,394 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5289, 4.1626, 3.8146, 3.8227, 3.3712, 3.0096, 2.5106, 3.9003], + device='cuda:3'), covar=tensor([0.1748, 0.0171, 0.0330, 0.0320, 0.0367, 0.0949, 0.2445, 0.0152], + device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0088, 0.0107, 0.0074, 0.0093, 0.0130, 0.0167, 0.0078], + device='cuda:3'), out_proj_covar=tensor([1.7560e-04, 9.8336e-05, 1.2744e-04, 8.6309e-05, 1.1355e-04, 1.5568e-04, + 1.8964e-04, 8.8568e-05], device='cuda:3') +2022-11-15 15:05:18,108 INFO [train.py:876] (3/4) Epoch 2, batch 2300, loss[loss=0.2793, simple_loss=0.2368, pruned_loss=0.1609, over 5543.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.2291, pruned_loss=0.1453, over 1084143.77 frames. ], batch size: 40, lr: 3.34e-02, grad_scale: 16.0 +2022-11-15 15:05:25,130 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:05:25,875 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:05:27,158 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6394, 3.6949, 4.0059, 3.9365, 3.3899, 3.3503, 4.3007, 3.5541], + device='cuda:3'), covar=tensor([0.0598, 0.0885, 0.0496, 0.0637, 0.0633, 0.0365, 0.0832, 0.0502], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0069, 0.0057, 0.0067, 0.0044, 0.0039, 0.0066, 0.0052], + device='cuda:3'), out_proj_covar=tensor([9.0492e-05, 1.3418e-04, 1.0903e-04, 1.2791e-04, 8.6881e-05, 7.4289e-05, + 1.4426e-04, 9.8121e-05], device='cuda:3') +2022-11-15 15:05:39,845 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 2.289e+02 3.013e+02 4.083e+02 8.581e+02, threshold=6.026e+02, percent-clipped=8.0 +2022-11-15 15:05:54,062 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 15:05:59,633 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:02,150 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-15 15:06:07,379 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:09,071 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 +2022-11-15 15:06:09,484 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:18,538 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:26,550 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:29,226 INFO [train.py:876] (3/4) Epoch 2, batch 2400, loss[loss=0.2694, simple_loss=0.2186, pruned_loss=0.1601, over 5006.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.2276, pruned_loss=0.1439, over 1082425.48 frames. ], batch size: 109, lr: 3.32e-02, grad_scale: 16.0 +2022-11-15 15:06:41,903 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:06:51,267 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.170e+02 2.571e+02 3.474e+02 5.585e+02, threshold=5.143e+02, percent-clipped=0.0 +2022-11-15 15:07:00,765 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:07:02,281 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:07:19,795 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:07:38,169 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:07:40,532 INFO [train.py:876] (3/4) Epoch 2, batch 2500, loss[loss=0.2246, simple_loss=0.2147, pruned_loss=0.1173, over 5758.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.2272, pruned_loss=0.1432, over 1080709.06 frames. ], batch size: 14, lr: 3.31e-02, grad_scale: 16.0 +2022-11-15 15:08:03,223 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 2.186e+02 2.866e+02 3.924e+02 6.368e+02, threshold=5.732e+02, percent-clipped=5.0 +2022-11-15 15:08:03,452 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:08:09,464 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4174, 1.6761, 1.3234, 1.2800, 1.3981, 1.3592, 1.2735, 1.3798], + device='cuda:3'), covar=tensor([0.0236, 0.0240, 0.0454, 0.0505, 0.0270, 0.0250, 0.0301, 0.0412], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0032, 0.0037, 0.0028, 0.0038, 0.0030, 0.0037, 0.0025], + device='cuda:3'), out_proj_covar=tensor([5.0137e-05, 5.2726e-05, 6.5549e-05, 4.7780e-05, 6.6635e-05, 5.5031e-05, + 6.2576e-05, 4.3480e-05], device='cuda:3') +2022-11-15 15:08:22,225 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6066, 1.4637, 1.6466, 2.0209, 1.7835, 2.0317, 0.9138, 1.7966], + device='cuda:3'), covar=tensor([0.0152, 0.0320, 0.0370, 0.0088, 0.0122, 0.0132, 0.0197, 0.0133], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0017, 0.0018, 0.0017, 0.0017, 0.0019, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.2715e-05, 2.2601e-05, 2.5329e-05, 2.3938e-05, 2.4347e-05, 2.3179e-05, + 2.6547e-05, 2.0786e-05], device='cuda:3') +2022-11-15 15:08:49,612 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0 +2022-11-15 15:08:52,034 INFO [train.py:876] (3/4) Epoch 2, batch 2600, loss[loss=0.2413, simple_loss=0.2241, pruned_loss=0.1293, over 5652.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.2283, pruned_loss=0.1437, over 1084974.81 frames. ], batch size: 29, lr: 3.30e-02, grad_scale: 16.0 +2022-11-15 15:09:14,801 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.199e+02 2.978e+02 3.710e+02 9.077e+02, threshold=5.957e+02, percent-clipped=5.0 +2022-11-15 15:09:31,466 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0675, 3.9799, 3.0158, 1.4459, 3.9565, 1.6033, 3.3800, 2.2743], + device='cuda:3'), covar=tensor([0.0814, 0.0227, 0.0331, 0.2654, 0.0192, 0.1717, 0.0204, 0.1615], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0056, 0.0048, 0.0095, 0.0055, 0.0084, 0.0042, 0.0087], + device='cuda:3'), out_proj_covar=tensor([1.8121e-04, 1.1554e-04, 1.0360e-04, 1.9254e-04, 1.1124e-04, 1.7408e-04, + 9.3402e-05, 1.8237e-04], device='cuda:3') +2022-11-15 15:09:34,948 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.19 vs. limit=2.0 +2022-11-15 15:09:40,254 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:09:40,589 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.34 vs. limit=2.0 +2022-11-15 15:10:02,858 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.34 vs. limit=5.0 +2022-11-15 15:10:03,768 INFO [train.py:876] (3/4) Epoch 2, batch 2700, loss[loss=0.361, simple_loss=0.2863, pruned_loss=0.2178, over 5408.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.2266, pruned_loss=0.1421, over 1082383.37 frames. ], batch size: 70, lr: 3.28e-02, grad_scale: 16.0 +2022-11-15 15:10:29,551 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 2.293e+02 2.993e+02 4.046e+02 1.330e+03, threshold=5.986e+02, percent-clipped=8.0 +2022-11-15 15:10:37,234 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:10:53,323 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.25 vs. limit=2.0 +2022-11-15 15:10:54,484 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:10:59,735 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-15 15:11:07,112 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-15 15:11:10,892 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5016, 3.9889, 3.6631, 4.0519, 3.4477, 3.4930, 1.8933, 3.5292], + device='cuda:3'), covar=tensor([0.0710, 0.0613, 0.0567, 0.0360, 0.0638, 0.0684, 0.3620, 0.0610], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0045, 0.0045, 0.0034, 0.0049, 0.0038, 0.0087, 0.0050], + device='cuda:3'), out_proj_covar=tensor([9.8957e-05, 8.3781e-05, 8.1331e-05, 6.2727e-05, 8.8008e-05, 6.9499e-05, + 1.5022e-04, 9.3047e-05], device='cuda:3') +2022-11-15 15:11:10,960 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 15:11:16,398 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:11:18,636 INFO [train.py:876] (3/4) Epoch 2, batch 2800, loss[loss=0.2549, simple_loss=0.2425, pruned_loss=0.1336, over 5783.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.2261, pruned_loss=0.1416, over 1083611.02 frames. ], batch size: 16, lr: 3.27e-02, grad_scale: 16.0 +2022-11-15 15:11:36,448 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:11:37,200 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 15:11:39,748 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 2.201e+02 2.831e+02 3.552e+02 8.014e+02, threshold=5.662e+02, percent-clipped=2.0 +2022-11-15 15:11:49,992 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:11:52,434 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 +2022-11-15 15:11:53,502 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 15:12:04,417 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0939, 0.8650, 0.9512, 0.9525, 0.9744, 1.3164, 0.9842, 0.9285], + device='cuda:3'), covar=tensor([0.0507, 0.0167, 0.0348, 0.0468, 0.0187, 0.0280, 0.0312, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0013, 0.0015, 0.0011, 0.0013, 0.0015, 0.0013], + device='cuda:3'), out_proj_covar=tensor([2.2876e-05, 2.1890e-05, 2.4774e-05, 3.1585e-05, 2.0904e-05, 2.2909e-05, + 2.7864e-05, 2.2357e-05], device='cuda:3') +2022-11-15 15:12:29,729 INFO [train.py:876] (3/4) Epoch 2, batch 2900, loss[loss=0.3079, simple_loss=0.2456, pruned_loss=0.1851, over 4699.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.2265, pruned_loss=0.142, over 1081120.34 frames. ], batch size: 136, lr: 3.26e-02, grad_scale: 16.0 +2022-11-15 15:12:42,342 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.38 vs. limit=2.0 +2022-11-15 15:12:52,051 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 2.141e+02 2.737e+02 3.549e+02 7.365e+02, threshold=5.475e+02, percent-clipped=2.0 +2022-11-15 15:12:52,918 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7642, 1.8433, 1.4103, 1.4410, 1.0344, 1.6473, 1.4375, 1.0233], + device='cuda:3'), covar=tensor([0.0140, 0.0161, 0.0129, 0.0240, 0.0505, 0.0859, 0.0161, 0.0238], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0022, 0.0023, 0.0025, 0.0024, 0.0019, 0.0023, 0.0022], + device='cuda:3'), out_proj_covar=tensor([3.8043e-05, 3.2139e-05, 3.0807e-05, 3.6013e-05, 3.9736e-05, 3.0974e-05, + 3.3825e-05, 3.3390e-05], device='cuda:3') +2022-11-15 15:13:15,501 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 +2022-11-15 15:13:18,138 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:13:20,696 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 15:13:39,833 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-15 15:13:41,346 INFO [train.py:876] (3/4) Epoch 2, batch 3000, loss[loss=0.2322, simple_loss=0.2139, pruned_loss=0.1252, over 5500.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.2289, pruned_loss=0.1451, over 1081070.39 frames. ], batch size: 13, lr: 3.24e-02, grad_scale: 16.0 +2022-11-15 15:13:41,347 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 15:14:00,266 INFO [train.py:908] (3/4) Epoch 2, validation: loss=0.2049, simple_loss=0.215, pruned_loss=0.09736, over 1530663.00 frames. +2022-11-15 15:14:00,267 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4508MB +2022-11-15 15:14:10,873 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:14:22,011 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 2.226e+02 2.767e+02 3.573e+02 6.449e+02, threshold=5.534e+02, percent-clipped=5.0 +2022-11-15 15:14:22,516 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.28 vs. limit=5.0 +2022-11-15 15:14:28,951 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:14:50,080 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.39 vs. limit=5.0 +2022-11-15 15:15:02,044 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.56 vs. limit=5.0 +2022-11-15 15:15:03,069 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:15:10,522 INFO [train.py:876] (3/4) Epoch 2, batch 3100, loss[loss=0.2581, simple_loss=0.2174, pruned_loss=0.1494, over 4674.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.2287, pruned_loss=0.1447, over 1081325.53 frames. ], batch size: 135, lr: 3.23e-02, grad_scale: 16.0 +2022-11-15 15:15:26,632 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 15:15:29,525 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:15:33,067 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.883e+01 2.180e+02 2.990e+02 3.781e+02 9.963e+02, threshold=5.979e+02, percent-clipped=5.0 +2022-11-15 15:15:36,993 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:15:43,099 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 15:15:43,224 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7239, 1.6850, 2.3333, 1.9831, 2.3657, 1.2766, 2.2219, 2.6045], + device='cuda:3'), covar=tensor([0.0043, 0.0420, 0.0098, 0.0240, 0.0089, 0.0384, 0.0152, 0.0099], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0137, 0.0082, 0.0129, 0.0071, 0.0118, 0.0107, 0.0081], + device='cuda:3'), out_proj_covar=tensor([9.3415e-05, 1.9564e-04, 1.2104e-04, 1.7960e-04, 1.0341e-04, 1.7036e-04, + 1.5862e-04, 1.1996e-04], device='cuda:3') +2022-11-15 15:16:04,087 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:16:08,978 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:16:20,292 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:16:22,099 INFO [train.py:876] (3/4) Epoch 2, batch 3200, loss[loss=0.2357, simple_loss=0.1924, pruned_loss=0.1395, over 4205.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.2264, pruned_loss=0.1418, over 1084096.25 frames. ], batch size: 183, lr: 3.22e-02, grad_scale: 16.0 +2022-11-15 15:16:25,539 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8345, 5.1304, 5.2703, 5.0898, 4.4729, 3.9849, 5.6532, 4.8192], + device='cuda:3'), covar=tensor([0.0500, 0.0437, 0.0260, 0.0443, 0.0351, 0.0327, 0.0433, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0070, 0.0058, 0.0069, 0.0045, 0.0040, 0.0071, 0.0054], + device='cuda:3'), out_proj_covar=tensor([9.1670e-05, 1.4225e-04, 1.1227e-04, 1.3368e-04, 9.2219e-05, 7.8895e-05, + 1.5801e-04, 1.0591e-04], device='cuda:3') +2022-11-15 15:16:44,408 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.221e+01 2.155e+02 2.904e+02 3.416e+02 7.936e+02, threshold=5.808e+02, percent-clipped=4.0 +2022-11-15 15:16:44,663 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.3233, 2.8998, 4.1043, 4.5707, 5.4933, 4.0428, 3.6699, 5.1120], + device='cuda:3'), covar=tensor([0.0014, 0.0898, 0.0543, 0.0208, 0.0028, 0.0570, 0.0584, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0171, 0.0177, 0.0096, 0.0103, 0.0189, 0.0174, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 15:16:46,852 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2980, 3.1672, 2.8471, 2.6531, 2.5266, 3.5666, 2.6325, 3.1770], + device='cuda:3'), covar=tensor([0.0246, 0.0068, 0.0094, 0.0215, 0.0265, 0.0047, 0.0170, 0.0049], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0048, 0.0055, 0.0052, 0.0092, 0.0053, 0.0072, 0.0047], + device='cuda:3'), out_proj_covar=tensor([1.2783e-04, 7.1210e-05, 7.8711e-05, 8.3189e-05, 1.4194e-04, 7.2577e-05, + 1.0687e-04, 6.5647e-05], device='cuda:3') +2022-11-15 15:16:52,748 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 15:17:33,912 INFO [train.py:876] (3/4) Epoch 2, batch 3300, loss[loss=0.1737, simple_loss=0.1699, pruned_loss=0.08871, over 5164.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.2236, pruned_loss=0.1389, over 1077524.48 frames. ], batch size: 8, lr: 3.21e-02, grad_scale: 16.0 +2022-11-15 15:17:55,742 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.974e+02 2.609e+02 3.131e+02 6.226e+02, threshold=5.219e+02, percent-clipped=2.0 +2022-11-15 15:18:14,585 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-15 15:18:18,062 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 +2022-11-15 15:18:21,599 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 15:18:35,941 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 15:18:45,894 INFO [train.py:876] (3/4) Epoch 2, batch 3400, loss[loss=0.208, simple_loss=0.2025, pruned_loss=0.1067, over 5563.00 frames. ], tot_loss[loss=0.255, simple_loss=0.2261, pruned_loss=0.1419, over 1080744.95 frames. ], batch size: 13, lr: 3.19e-02, grad_scale: 16.0 +2022-11-15 15:18:48,286 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 15:19:01,415 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:19:07,528 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.426e+02 2.941e+02 3.632e+02 1.443e+03, threshold=5.881e+02, percent-clipped=8.0 +2022-11-15 15:19:18,045 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 15:19:36,088 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:19:36,934 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:19:46,721 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0630, 0.9899, 0.9981, 0.6828, 1.5028, 1.2064, 0.9894, 1.0679], + device='cuda:3'), covar=tensor([0.0293, 0.0132, 0.0207, 0.0499, 0.0125, 0.0219, 0.0230, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0012, 0.0010, 0.0011, 0.0013, 0.0011], + device='cuda:3'), out_proj_covar=tensor([2.1060e-05, 2.0661e-05, 2.3028e-05, 2.7487e-05, 1.8622e-05, 2.1093e-05, + 2.4297e-05, 2.0757e-05], device='cuda:3') +2022-11-15 15:19:51,854 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:19:52,492 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:19:58,020 INFO [train.py:876] (3/4) Epoch 2, batch 3500, loss[loss=0.2174, simple_loss=0.1984, pruned_loss=0.1182, over 5097.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.2253, pruned_loss=0.142, over 1079471.94 frames. ], batch size: 7, lr: 3.18e-02, grad_scale: 16.0 +2022-11-15 15:19:58,852 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:20:17,446 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2035, 1.2677, 1.2543, 1.0042, 1.5056, 1.5936, 1.0725, 1.4272], + device='cuda:3'), covar=tensor([0.0121, 0.0112, 0.0259, 0.0077, 0.0072, 0.0092, 0.0157, 0.0104], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0017, 0.0017, 0.0020, 0.0019, 0.0019, 0.0023, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.6961e-05, 2.5699e-05, 2.6468e-05, 2.5524e-05, 2.7122e-05, 2.6557e-05, + 3.2881e-05, 2.5017e-05], device='cuda:3') +2022-11-15 15:20:20,009 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.227e+02 2.667e+02 3.424e+02 6.980e+02, threshold=5.333e+02, percent-clipped=5.0 +2022-11-15 15:20:20,906 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:20:24,283 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:20:42,601 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:21:08,879 INFO [train.py:876] (3/4) Epoch 2, batch 3600, loss[loss=0.2436, simple_loss=0.1997, pruned_loss=0.1438, over 4065.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.226, pruned_loss=0.1421, over 1080124.50 frames. ], batch size: 181, lr: 3.17e-02, grad_scale: 32.0 +2022-11-15 15:21:13,643 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.28 vs. limit=2.0 +2022-11-15 15:21:18,342 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:21:27,772 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2729, 4.9927, 5.1975, 4.6623, 5.5166, 5.3122, 4.8175, 5.0441], + device='cuda:3'), covar=tensor([0.0465, 0.0280, 0.0483, 0.0354, 0.0411, 0.0101, 0.0395, 0.0285], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0073, 0.0059, 0.0072, 0.0065, 0.0048, 0.0060, 0.0057], + device='cuda:3'), out_proj_covar=tensor([1.5150e-04, 1.4460e-04, 1.2470e-04, 1.4286e-04, 1.5360e-04, 9.1717e-05, + 1.2369e-04, 1.2058e-04], device='cuda:3') +2022-11-15 15:21:31,443 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.250e+02 2.765e+02 3.840e+02 7.288e+02, threshold=5.531e+02, percent-clipped=6.0 +2022-11-15 15:22:01,318 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 15:22:18,841 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7777, 3.0682, 2.7295, 1.2047, 2.5650, 2.9598, 2.7833, 3.7186], + device='cuda:3'), covar=tensor([0.0731, 0.0341, 0.0230, 0.0878, 0.0075, 0.0092, 0.0081, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0101, 0.0071, 0.0117, 0.0065, 0.0061, 0.0059, 0.0071], + device='cuda:3'), out_proj_covar=tensor([1.6264e-04, 1.3386e-04, 1.0396e-04, 1.5947e-04, 8.7159e-05, 8.3452e-05, + 8.1481e-05, 9.1149e-05], device='cuda:3') +2022-11-15 15:22:19,996 INFO [train.py:876] (3/4) Epoch 2, batch 3700, loss[loss=0.1945, simple_loss=0.1781, pruned_loss=0.1054, over 5588.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.226, pruned_loss=0.1417, over 1084131.24 frames. ], batch size: 25, lr: 3.16e-02, grad_scale: 32.0 +2022-11-15 15:22:42,990 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.397e+02 3.169e+02 4.273e+02 6.249e+02, threshold=6.338e+02, percent-clipped=7.0 +2022-11-15 15:23:25,746 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:23:31,391 INFO [train.py:876] (3/4) Epoch 2, batch 3800, loss[loss=0.2579, simple_loss=0.2333, pruned_loss=0.1412, over 5705.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.2246, pruned_loss=0.1392, over 1086292.64 frames. ], batch size: 28, lr: 3.15e-02, grad_scale: 16.0 +2022-11-15 15:23:50,500 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:23:53,788 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2022-11-15 15:23:54,056 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.988e+01 2.162e+02 2.820e+02 3.661e+02 7.630e+02, threshold=5.641e+02, percent-clipped=4.0 +2022-11-15 15:23:57,603 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:23:59,581 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:24:06,960 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 +2022-11-15 15:24:11,536 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:24:17,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 +2022-11-15 15:24:21,565 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 15:24:26,492 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 15:24:27,462 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7165, 2.6851, 2.6644, 2.8036, 2.4925, 2.2410, 2.9746, 2.6701], + device='cuda:3'), covar=tensor([0.0498, 0.0801, 0.0563, 0.0682, 0.0725, 0.0473, 0.0854, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0071, 0.0058, 0.0070, 0.0046, 0.0040, 0.0074, 0.0053], + device='cuda:3'), out_proj_covar=tensor([9.2587e-05, 1.4536e-04, 1.1712e-04, 1.3841e-04, 9.5971e-05, 8.1666e-05, + 1.6658e-04, 1.0557e-04], device='cuda:3') +2022-11-15 15:24:31,163 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:24:31,253 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7632, 1.5299, 0.7807, 1.0374, 0.5546, 1.1125, 1.0830, 0.6555], + device='cuda:3'), covar=tensor([0.0121, 0.0058, 0.0123, 0.0112, 0.0310, 0.0094, 0.0228, 0.0175], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0019, 0.0022, 0.0023, 0.0022, 0.0019, 0.0021, 0.0021], + device='cuda:3'), out_proj_covar=tensor([3.3560e-05, 2.8164e-05, 2.9332e-05, 3.4766e-05, 3.7943e-05, 3.1155e-05, + 3.3549e-05, 3.1531e-05], device='cuda:3') +2022-11-15 15:24:32,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 15:24:32,490 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 15:24:41,929 INFO [train.py:876] (3/4) Epoch 2, batch 3900, loss[loss=0.2429, simple_loss=0.2165, pruned_loss=0.1347, over 5612.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.2251, pruned_loss=0.1392, over 1083634.81 frames. ], batch size: 23, lr: 3.13e-02, grad_scale: 16.0 +2022-11-15 15:24:43,110 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:24:54,761 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7391, 2.1533, 1.8978, 2.4949, 1.4051, 2.1478, 1.8070, 2.0583], + device='cuda:3'), covar=tensor([0.0093, 0.0161, 0.0257, 0.0208, 0.0260, 0.0288, 0.0340, 0.0446], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0031, 0.0036, 0.0025, 0.0037, 0.0033, 0.0036, 0.0026], + device='cuda:3'), out_proj_covar=tensor([4.9079e-05, 5.3028e-05, 6.9706e-05, 4.4439e-05, 6.7317e-05, 6.3937e-05, + 6.4459e-05, 4.6757e-05], device='cuda:3') +2022-11-15 15:25:04,803 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 2.353e+02 2.852e+02 3.627e+02 7.008e+02, threshold=5.704e+02, percent-clipped=3.0 +2022-11-15 15:25:26,261 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 15:25:26,628 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.4678, 1.7423, 0.4732, 1.4161, 0.5691, 1.1904, 1.4280, 1.1215], + device='cuda:3'), covar=tensor([0.0227, 0.0085, 0.0170, 0.0170, 0.0403, 0.0169, 0.0213, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0019, 0.0021, 0.0022, 0.0022, 0.0017, 0.0020, 0.0021], + device='cuda:3'), out_proj_covar=tensor([3.2420e-05, 2.7723e-05, 2.8657e-05, 3.2766e-05, 3.6697e-05, 2.9848e-05, + 3.2135e-05, 3.0989e-05], device='cuda:3') +2022-11-15 15:25:27,371 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:25:31,463 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 15:25:54,060 INFO [train.py:876] (3/4) Epoch 2, batch 4000, loss[loss=0.1824, simple_loss=0.1748, pruned_loss=0.09497, over 5445.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.2235, pruned_loss=0.138, over 1083021.84 frames. ], batch size: 10, lr: 3.12e-02, grad_scale: 16.0 +2022-11-15 15:26:05,780 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2042, 2.4168, 2.0424, 2.2000, 1.4456, 2.4331, 2.3187, 1.8115], + device='cuda:3'), covar=tensor([0.0148, 0.0143, 0.0244, 0.0113, 0.0266, 0.0130, 0.0335, 0.1110], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0031, 0.0037, 0.0026, 0.0038, 0.0032, 0.0037, 0.0025], + device='cuda:3'), out_proj_covar=tensor([4.9502e-05, 5.3772e-05, 7.0434e-05, 4.5305e-05, 6.8311e-05, 6.2812e-05, + 6.5817e-05, 4.5928e-05], device='cuda:3') +2022-11-15 15:26:16,541 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.176e+02 2.940e+02 3.819e+02 6.622e+02, threshold=5.880e+02, percent-clipped=2.0 +2022-11-15 15:26:28,786 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.66 vs. limit=5.0 +2022-11-15 15:26:46,450 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0330, 0.9406, 0.9062, 1.2006, 1.0528, 1.1299, 0.7536, 1.0352], + device='cuda:3'), covar=tensor([0.0179, 0.0124, 0.0288, 0.0105, 0.0220, 0.0087, 0.0224, 0.0311], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0019, 0.0018, 0.0019, 0.0018, 0.0018, 0.0020, 0.0017], + device='cuda:3'), out_proj_covar=tensor([2.6248e-05, 2.6446e-05, 2.7757e-05, 2.3852e-05, 2.4921e-05, 2.5116e-05, + 3.3894e-05, 2.5446e-05], device='cuda:3') +2022-11-15 15:26:48,788 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 +2022-11-15 15:26:49,379 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 15:26:58,614 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7615, 4.6894, 4.2500, 4.5560, 3.9234, 3.3989, 2.6457, 4.1822], + device='cuda:3'), covar=tensor([0.1606, 0.0136, 0.0356, 0.0202, 0.0283, 0.0684, 0.2283, 0.0109], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0097, 0.0127, 0.0089, 0.0104, 0.0141, 0.0172, 0.0088], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 15:27:04,364 INFO [train.py:876] (3/4) Epoch 2, batch 4100, loss[loss=0.2108, simple_loss=0.2082, pruned_loss=0.1067, over 5585.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.2215, pruned_loss=0.1361, over 1080117.57 frames. ], batch size: 14, lr: 3.11e-02, grad_scale: 16.0 +2022-11-15 15:27:04,503 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0164, 2.0700, 1.1290, 1.7132, 0.7403, 1.4232, 1.6821, 1.0274], + device='cuda:3'), covar=tensor([0.0173, 0.0070, 0.0113, 0.0161, 0.0470, 0.0104, 0.0228, 0.0249], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0019, 0.0022, 0.0022, 0.0023, 0.0017, 0.0020, 0.0021], + device='cuda:3'), out_proj_covar=tensor([3.3215e-05, 2.6967e-05, 2.9022e-05, 3.3865e-05, 3.8418e-05, 2.9883e-05, + 3.2176e-05, 3.1750e-05], device='cuda:3') +2022-11-15 15:27:12,105 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9818, 1.9623, 0.8730, 1.4190, 0.8047, 1.2642, 1.5292, 0.8321], + device='cuda:3'), covar=tensor([0.0166, 0.0073, 0.0145, 0.0202, 0.0430, 0.0156, 0.0239, 0.0325], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0019, 0.0022, 0.0023, 0.0023, 0.0018, 0.0021, 0.0022], + device='cuda:3'), out_proj_covar=tensor([3.3824e-05, 2.7461e-05, 2.9651e-05, 3.4419e-05, 3.9315e-05, 3.0348e-05, + 3.2828e-05, 3.2625e-05], device='cuda:3') +2022-11-15 15:27:24,545 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:27:26,571 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1867, 0.8023, 0.8282, 0.4391, 1.3669, 0.5538, 0.8176, 1.0871], + device='cuda:3'), covar=tensor([0.0171, 0.0163, 0.0369, 0.0589, 0.0181, 0.0242, 0.0252, 0.0233], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0011, 0.0010, 0.0012, 0.0010, 0.0010, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([2.0166e-05, 2.0365e-05, 2.2484e-05, 2.6417e-05, 1.9491e-05, 1.9776e-05, + 2.2084e-05, 1.8899e-05], device='cuda:3') +2022-11-15 15:27:27,789 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.309e+02 2.774e+02 3.508e+02 5.775e+02, threshold=5.548e+02, percent-clipped=0.0 +2022-11-15 15:27:46,241 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:27:56,094 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9607, 2.1940, 1.2593, 1.6831, 0.6640, 1.2792, 1.9393, 1.1694], + device='cuda:3'), covar=tensor([0.0204, 0.0082, 0.0143, 0.0223, 0.0910, 0.0250, 0.0146, 0.0529], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0019, 0.0022, 0.0023, 0.0024, 0.0018, 0.0020, 0.0023], + device='cuda:3'), out_proj_covar=tensor([3.4891e-05, 2.7810e-05, 3.0417e-05, 3.5390e-05, 4.0510e-05, 3.0759e-05, + 3.2097e-05, 3.4156e-05], device='cuda:3') +2022-11-15 15:27:58,775 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:28:07,611 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.36 vs. limit=2.0 +2022-11-15 15:28:16,214 INFO [train.py:876] (3/4) Epoch 2, batch 4200, loss[loss=0.2465, simple_loss=0.2262, pruned_loss=0.1334, over 5558.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.2211, pruned_loss=0.1347, over 1083924.88 frames. ], batch size: 16, lr: 3.10e-02, grad_scale: 16.0 +2022-11-15 15:28:19,840 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:28:39,537 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.116e+02 2.605e+02 3.416e+02 5.601e+02, threshold=5.209e+02, percent-clipped=1.0 +2022-11-15 15:28:39,960 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 15:28:56,847 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:28:58,338 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:29:04,731 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 15:29:08,709 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0725, 5.4855, 5.6035, 5.2185, 4.9574, 4.5728, 5.8547, 5.1136], + device='cuda:3'), covar=tensor([0.0303, 0.0564, 0.0210, 0.0595, 0.0324, 0.0295, 0.0573, 0.0332], + device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0073, 0.0060, 0.0071, 0.0046, 0.0041, 0.0076, 0.0054], + device='cuda:3'), out_proj_covar=tensor([9.8406e-05, 1.5177e-04, 1.2330e-04, 1.4247e-04, 9.9178e-05, 8.4876e-05, + 1.7339e-04, 1.0788e-04], device='cuda:3') +2022-11-15 15:29:24,888 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2340, 4.1973, 4.2482, 4.1012, 4.4382, 4.4638, 4.0948, 4.3983], + device='cuda:3'), covar=tensor([0.0795, 0.0554, 0.0834, 0.0581, 0.0615, 0.0277, 0.0495, 0.0525], + device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0074, 0.0057, 0.0071, 0.0064, 0.0046, 0.0057, 0.0057], + device='cuda:3'), out_proj_covar=tensor([1.4648e-04, 1.5043e-04, 1.2136e-04, 1.4093e-04, 1.4851e-04, 9.3108e-05, + 1.1829e-04, 1.2039e-04], device='cuda:3') +2022-11-15 15:29:27,548 INFO [train.py:876] (3/4) Epoch 2, batch 4300, loss[loss=0.2223, simple_loss=0.221, pruned_loss=0.1118, over 5529.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.2234, pruned_loss=0.1364, over 1091517.35 frames. ], batch size: 17, lr: 3.09e-02, grad_scale: 16.0 +2022-11-15 15:29:38,724 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 15:29:39,554 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2022-11-15 15:29:41,469 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:29:51,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 2.389e+02 3.097e+02 3.751e+02 1.482e+03, threshold=6.195e+02, percent-clipped=9.0 +2022-11-15 15:30:11,795 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:30:16,978 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.18 vs. limit=2.0 +2022-11-15 15:30:21,642 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9403, 1.5963, 1.0302, 1.8483, 0.4761, 1.7083, 1.6122, 0.8639], + device='cuda:3'), covar=tensor([0.0255, 0.0173, 0.0170, 0.0208, 0.0730, 0.0471, 0.0377, 0.0376], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0021, 0.0024, 0.0025, 0.0025, 0.0019, 0.0022, 0.0024], + device='cuda:3'), out_proj_covar=tensor([3.8256e-05, 3.0958e-05, 3.3315e-05, 3.8231e-05, 4.3458e-05, 3.2457e-05, + 3.5417e-05, 3.7365e-05], device='cuda:3') +2022-11-15 15:30:39,153 INFO [train.py:876] (3/4) Epoch 2, batch 4400, loss[loss=0.2334, simple_loss=0.2183, pruned_loss=0.1243, over 5478.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.2235, pruned_loss=0.1371, over 1088495.63 frames. ], batch size: 12, lr: 3.08e-02, grad_scale: 8.0 +2022-11-15 15:30:42,990 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.16 vs. limit=2.0 +2022-11-15 15:30:50,408 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6296, 2.2038, 2.1344, 2.3323, 1.4025, 2.1417, 1.4472, 2.1747], + device='cuda:3'), covar=tensor([0.0954, 0.0218, 0.0498, 0.0159, 0.0675, 0.0542, 0.1317, 0.0206], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0098, 0.0126, 0.0086, 0.0102, 0.0143, 0.0170, 0.0087], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 15:30:55,203 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:30:57,335 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:31:02,564 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 2.298e+02 2.735e+02 3.598e+02 7.155e+02, threshold=5.470e+02, percent-clipped=1.0 +2022-11-15 15:31:40,414 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:31:49,141 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-15 15:31:49,953 INFO [train.py:876] (3/4) Epoch 2, batch 4500, loss[loss=0.3162, simple_loss=0.2673, pruned_loss=0.1826, over 5504.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.2227, pruned_loss=0.1374, over 1083741.29 frames. ], batch size: 49, lr: 3.07e-02, grad_scale: 8.0 +2022-11-15 15:32:04,096 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2172, 4.5377, 5.0425, 4.5473, 5.1153, 5.0204, 4.6156, 4.9525], + device='cuda:3'), covar=tensor([0.0286, 0.0349, 0.0362, 0.0290, 0.0355, 0.0125, 0.0195, 0.0270], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0077, 0.0060, 0.0074, 0.0066, 0.0048, 0.0060, 0.0061], + device='cuda:3'), out_proj_covar=tensor([1.5282e-04, 1.5842e-04, 1.2571e-04, 1.4905e-04, 1.5273e-04, 9.6474e-05, + 1.2356e-04, 1.2846e-04], device='cuda:3') +2022-11-15 15:32:04,894 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8688, 1.3341, 1.5736, 1.9811, 1.1638, 1.2420, 1.2846, 1.4803], + device='cuda:3'), covar=tensor([0.0098, 0.0238, 0.0216, 0.0151, 0.0297, 0.0232, 0.0226, 0.0266], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0034, 0.0038, 0.0027, 0.0039, 0.0033, 0.0039, 0.0025], + device='cuda:3'), out_proj_covar=tensor([5.0445e-05, 5.9103e-05, 7.5514e-05, 4.8764e-05, 7.2015e-05, 6.4331e-05, + 7.1132e-05, 4.7478e-05], device='cuda:3') +2022-11-15 15:32:13,965 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.368e+02 2.947e+02 3.819e+02 5.858e+02, threshold=5.894e+02, percent-clipped=4.0 +2022-11-15 15:32:20,977 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:32:31,010 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:32:41,974 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:01,635 INFO [train.py:876] (3/4) Epoch 2, batch 4600, loss[loss=0.1951, simple_loss=0.1732, pruned_loss=0.1085, over 5743.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.2231, pruned_loss=0.1375, over 1083122.74 frames. ], batch size: 14, lr: 3.05e-02, grad_scale: 8.0 +2022-11-15 15:33:04,955 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:05,516 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:07,033 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3608, 4.0344, 3.4925, 3.0055, 2.8867, 3.8504, 3.4529, 4.2185], + device='cuda:3'), covar=tensor([0.0321, 0.0064, 0.0091, 0.0174, 0.0355, 0.0064, 0.0159, 0.0032], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0054, 0.0064, 0.0059, 0.0103, 0.0060, 0.0082, 0.0050], + device='cuda:3'), out_proj_covar=tensor([1.4770e-04, 8.3966e-05, 9.9200e-05, 1.0134e-04, 1.6478e-04, 8.6028e-05, + 1.2828e-04, 7.5453e-05], device='cuda:3') +2022-11-15 15:33:08,457 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1089, 3.2272, 2.6617, 2.6663, 2.0554, 3.1230, 2.3902, 2.9690], + device='cuda:3'), covar=tensor([0.0157, 0.0030, 0.0058, 0.0100, 0.0158, 0.0035, 0.0096, 0.0021], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0053, 0.0064, 0.0058, 0.0102, 0.0060, 0.0082, 0.0050], + device='cuda:3'), out_proj_covar=tensor([1.4678e-04, 8.3528e-05, 9.8997e-05, 1.0088e-04, 1.6379e-04, 8.5666e-05, + 1.2780e-04, 7.5151e-05], device='cuda:3') +2022-11-15 15:33:12,421 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:12,723 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.66 vs. limit=5.0 +2022-11-15 15:33:25,411 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.171e+02 2.900e+02 3.774e+02 7.017e+02, threshold=5.800e+02, percent-clipped=1.0 +2022-11-15 15:33:25,623 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:34,971 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:33:47,303 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3085, 1.5853, 1.4463, 2.2574, 1.3551, 1.5079, 1.7997, 0.8110], + device='cuda:3'), covar=tensor([0.0221, 0.0142, 0.0145, 0.0136, 0.0447, 0.0787, 0.0253, 0.0331], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0024, 0.0027, 0.0028, 0.0027, 0.0022, 0.0024, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.1925e-05, 3.5131e-05, 3.8473e-05, 4.2530e-05, 4.6888e-05, 3.7452e-05, + 3.9363e-05, 4.3909e-05], device='cuda:3') +2022-11-15 15:33:53,672 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.27 vs. limit=2.0 +2022-11-15 15:33:54,831 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:34:12,182 INFO [train.py:876] (3/4) Epoch 2, batch 4700, loss[loss=0.2976, simple_loss=0.2455, pruned_loss=0.1749, over 4730.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.2214, pruned_loss=0.136, over 1079130.06 frames. ], batch size: 135, lr: 3.04e-02, grad_scale: 8.0 +2022-11-15 15:34:13,638 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9973, 3.7073, 3.8815, 3.5918, 4.1281, 3.7116, 3.7452, 4.0101], + device='cuda:3'), covar=tensor([0.0467, 0.0306, 0.0585, 0.0322, 0.0382, 0.0361, 0.0282, 0.0324], + device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0078, 0.0062, 0.0073, 0.0070, 0.0049, 0.0061, 0.0063], + device='cuda:3'), out_proj_covar=tensor([1.5441e-04, 1.5978e-04, 1.3054e-04, 1.4759e-04, 1.6218e-04, 9.8913e-05, + 1.2628e-04, 1.3431e-04], device='cuda:3') +2022-11-15 15:34:18,294 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:34:20,686 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:34:25,387 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:34:36,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.194e+02 2.107e+02 2.730e+02 3.355e+02 8.347e+02, threshold=5.461e+02, percent-clipped=3.0 +2022-11-15 15:34:38,509 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:35:04,515 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:35:11,917 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:35:23,828 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 +2022-11-15 15:35:24,086 INFO [train.py:876] (3/4) Epoch 2, batch 4800, loss[loss=0.2827, simple_loss=0.2562, pruned_loss=0.1545, over 5499.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.2211, pruned_loss=0.1361, over 1078462.07 frames. ], batch size: 17, lr: 3.03e-02, grad_scale: 8.0 +2022-11-15 15:35:26,322 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:35:48,825 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 2.296e+02 2.933e+02 3.523e+02 8.613e+02, threshold=5.866e+02, percent-clipped=4.0 +2022-11-15 15:36:09,335 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:36:26,405 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.15 vs. limit=2.0 +2022-11-15 15:36:34,717 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:36:35,355 INFO [train.py:876] (3/4) Epoch 2, batch 4900, loss[loss=0.2449, simple_loss=0.2193, pruned_loss=0.1353, over 5562.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.2195, pruned_loss=0.1343, over 1086239.11 frames. ], batch size: 46, lr: 3.02e-02, grad_scale: 8.0 +2022-11-15 15:36:41,054 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 +2022-11-15 15:36:45,791 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:36:55,634 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:36:59,037 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.288e+02 2.922e+02 4.193e+02 1.035e+03, threshold=5.844e+02, percent-clipped=8.0 +2022-11-15 15:37:14,029 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8900, 1.0933, 1.6220, 0.3333, 1.7837, 1.6007, 1.3299, 2.1322], + device='cuda:3'), covar=tensor([0.0335, 0.0108, 0.0154, 0.0074, 0.0113, 0.0064, 0.0146, 0.0079], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0022, 0.0018, 0.0019, 0.0017, 0.0018, 0.0019, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.7061e-05, 3.0438e-05, 2.9445e-05, 2.3629e-05, 2.3838e-05, 2.4155e-05, + 3.4436e-05, 2.5766e-05], device='cuda:3') +2022-11-15 15:37:20,202 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:37:36,362 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2022-11-15 15:37:46,253 INFO [train.py:876] (3/4) Epoch 2, batch 5000, loss[loss=0.3268, simple_loss=0.2548, pruned_loss=0.1993, over 4075.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.2179, pruned_loss=0.1329, over 1080341.81 frames. ], batch size: 181, lr: 3.01e-02, grad_scale: 8.0 +2022-11-15 15:37:48,757 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:37:58,616 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2337, 1.1399, 1.2433, 0.5598, 1.3397, 1.4696, 1.1329, 1.4240], + device='cuda:3'), covar=tensor([0.1481, 0.0253, 0.0450, 0.1378, 0.1443, 0.0340, 0.0694, 0.0700], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0010, 0.0009, 0.0011, 0.0010, 0.0009, 0.0011, 0.0009], + device='cuda:3'), out_proj_covar=tensor([2.1656e-05, 1.9958e-05, 2.1152e-05, 2.5777e-05, 2.1170e-05, 1.9594e-05, + 2.3813e-05, 1.9316e-05], device='cuda:3') +2022-11-15 15:37:59,294 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:38:08,029 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:38:09,979 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 2.107e+02 2.700e+02 3.490e+02 8.758e+02, threshold=5.401e+02, percent-clipped=1.0 +2022-11-15 15:38:11,093 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 +2022-11-15 15:38:32,897 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:38:33,586 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:38:37,791 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5529, 1.8289, 2.0612, 3.2559, 3.2584, 2.2935, 1.8607, 2.9257], + device='cuda:3'), covar=tensor([0.0066, 0.1045, 0.0970, 0.0247, 0.0099, 0.1011, 0.0878, 0.0083], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0191, 0.0196, 0.0109, 0.0117, 0.0209, 0.0187, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0001], + device='cuda:3') +2022-11-15 15:38:39,816 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3604, 0.8514, 0.9875, 0.6884, 1.4689, 1.3767, 0.8262, 1.3526], + device='cuda:3'), covar=tensor([0.0475, 0.0138, 0.0305, 0.0601, 0.0506, 0.0110, 0.0225, 0.0289], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0010, 0.0008, 0.0011, 0.0010, 0.0009, 0.0011, 0.0009], + device='cuda:3'), out_proj_covar=tensor([2.0882e-05, 1.9865e-05, 2.0488e-05, 2.5817e-05, 2.1284e-05, 1.9533e-05, + 2.3305e-05, 1.9586e-05], device='cuda:3') +2022-11-15 15:38:43,943 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:38:53,607 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0025, 1.8928, 1.6426, 1.4665, 1.0651, 2.1155, 2.1037, 1.3820], + device='cuda:3'), covar=tensor([0.0133, 0.0323, 0.0379, 0.0315, 0.0315, 0.0255, 0.0222, 0.0628], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0034, 0.0042, 0.0029, 0.0041, 0.0036, 0.0041, 0.0027], + device='cuda:3'), out_proj_covar=tensor([5.4776e-05, 6.1805e-05, 8.4619e-05, 5.3362e-05, 7.6495e-05, 7.4070e-05, + 7.4477e-05, 5.1334e-05], device='cuda:3') +2022-11-15 15:38:54,520 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 +2022-11-15 15:38:56,191 INFO [train.py:876] (3/4) Epoch 2, batch 5100, loss[loss=0.2564, simple_loss=0.218, pruned_loss=0.1474, over 5127.00 frames. ], tot_loss[loss=0.241, simple_loss=0.2173, pruned_loss=0.1323, over 1078920.51 frames. ], batch size: 91, lr: 3.00e-02, grad_scale: 8.0 +2022-11-15 15:39:00,509 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 +2022-11-15 15:39:17,690 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:39:20,361 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 2.363e+02 2.989e+02 3.735e+02 9.189e+02, threshold=5.978e+02, percent-clipped=6.0 +2022-11-15 15:39:29,518 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4112, 0.9761, 1.0050, 0.8219, 1.3210, 1.2056, 0.8610, 1.0431], + device='cuda:3'), covar=tensor([0.0358, 0.0266, 0.3664, 0.0891, 0.0489, 0.0189, 0.0420, 0.0234], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0011, 0.0010, 0.0010, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([2.1544e-05, 2.0744e-05, 2.2570e-05, 2.6290e-05, 2.1337e-05, 2.0054e-05, + 2.4374e-05, 2.0353e-05], device='cuda:3') +2022-11-15 15:39:38,294 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:39:48,220 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:39:55,751 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.08 vs. limit=5.0 +2022-11-15 15:40:05,746 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:40:06,292 INFO [train.py:876] (3/4) Epoch 2, batch 5200, loss[loss=0.249, simple_loss=0.2396, pruned_loss=0.1292, over 5570.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.2184, pruned_loss=0.1322, over 1083354.35 frames. ], batch size: 16, lr: 2.99e-02, grad_scale: 8.0 +2022-11-15 15:40:21,537 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9089, 3.6718, 3.7727, 3.5561, 3.9721, 3.4951, 3.6263, 4.0194], + device='cuda:3'), covar=tensor([0.0430, 0.0275, 0.0469, 0.0284, 0.0412, 0.0421, 0.0294, 0.0223], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0073, 0.0061, 0.0073, 0.0068, 0.0049, 0.0061, 0.0063], + device='cuda:3'), out_proj_covar=tensor([1.5102e-04, 1.4888e-04, 1.2953e-04, 1.4656e-04, 1.5914e-04, 9.9877e-05, + 1.2925e-04, 1.3437e-04], device='cuda:3') +2022-11-15 15:40:25,423 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2022-11-15 15:40:27,644 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:40:30,896 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 2.247e+02 2.853e+02 3.509e+02 7.106e+02, threshold=5.707e+02, percent-clipped=3.0 +2022-11-15 15:40:31,127 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:40:39,938 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:41:01,894 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:41:16,830 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.12 vs. limit=2.0 +2022-11-15 15:41:17,840 INFO [train.py:876] (3/4) Epoch 2, batch 5300, loss[loss=0.2387, simple_loss=0.211, pruned_loss=0.1332, over 5742.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.2196, pruned_loss=0.1335, over 1079872.55 frames. ], batch size: 11, lr: 2.98e-02, grad_scale: 8.0 +2022-11-15 15:41:20,142 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:41:40,673 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:41:42,528 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.286e+02 2.922e+02 3.556e+02 5.667e+02, threshold=5.844e+02, percent-clipped=0.0 +2022-11-15 15:41:54,447 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:42:05,330 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:42:12,186 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2794, 1.0561, 1.3675, 0.6485, 1.3628, 1.1142, 0.6756, 1.1642], + device='cuda:3'), covar=tensor([0.0066, 0.0036, 0.0045, 0.0046, 0.0033, 0.0041, 0.0085, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0021, 0.0021, 0.0020, 0.0021, 0.0022, 0.0020], + device='cuda:3'), out_proj_covar=tensor([3.0822e-05, 3.3274e-05, 3.2792e-05, 2.5983e-05, 2.7610e-05, 2.7243e-05, + 3.8939e-05, 2.9932e-05], device='cuda:3') +2022-11-15 15:42:13,762 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:42:29,012 INFO [train.py:876] (3/4) Epoch 2, batch 5400, loss[loss=0.2093, simple_loss=0.1992, pruned_loss=0.1098, over 5735.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.22, pruned_loss=0.1326, over 1090916.46 frames. ], batch size: 13, lr: 2.97e-02, grad_scale: 8.0 +2022-11-15 15:42:39,256 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:42:52,239 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.272e+02 2.868e+02 3.649e+02 6.503e+02, threshold=5.736e+02, percent-clipped=2.0 +2022-11-15 15:43:07,596 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4999, 0.5800, 1.3681, 1.3062, 1.6369, 1.8092, 1.7703, 1.4822], + device='cuda:3'), covar=tensor([0.0110, 0.0543, 0.0335, 0.0217, 0.0146, 0.0087, 0.0138, 0.0183], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0017, 0.0017, 0.0019, 0.0020, 0.0018, 0.0021, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.4368e-05, 2.5571e-05, 2.6633e-05, 2.5235e-05, 2.8160e-05, 2.3156e-05, + 2.7576e-05, 2.6215e-05], device='cuda:3') +2022-11-15 15:43:10,914 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:43:23,765 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.33 vs. limit=5.0 +2022-11-15 15:43:28,535 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 +2022-11-15 15:43:32,315 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:43:40,146 INFO [train.py:876] (3/4) Epoch 2, batch 5500, loss[loss=0.2001, simple_loss=0.1885, pruned_loss=0.1059, over 5589.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.22, pruned_loss=0.1326, over 1090095.24 frames. ], batch size: 18, lr: 2.96e-02, grad_scale: 8.0 +2022-11-15 15:43:45,535 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:43:46,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 +2022-11-15 15:44:01,675 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:44:05,122 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.098e+02 2.585e+02 3.345e+02 6.540e+02, threshold=5.170e+02, percent-clipped=1.0 +2022-11-15 15:44:18,803 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:44:51,045 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 +2022-11-15 15:44:53,052 INFO [train.py:876] (3/4) Epoch 2, batch 5600, loss[loss=0.3147, simple_loss=0.2652, pruned_loss=0.182, over 5430.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.2197, pruned_loss=0.1326, over 1087823.43 frames. ], batch size: 49, lr: 2.95e-02, grad_scale: 8.0 +2022-11-15 15:45:17,179 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 2.173e+02 2.659e+02 3.655e+02 7.963e+02, threshold=5.318e+02, percent-clipped=7.0 +2022-11-15 15:45:48,896 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8785, 4.6325, 4.2921, 4.4507, 3.8028, 3.4661, 2.7515, 4.0245], + device='cuda:3'), covar=tensor([0.1478, 0.0163, 0.0312, 0.0345, 0.0312, 0.0710, 0.1979, 0.0162], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0097, 0.0135, 0.0089, 0.0111, 0.0149, 0.0176, 0.0089], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 15:45:50,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5679, 3.7972, 3.8005, 3.8003, 3.7897, 3.5409, 1.3504, 3.5248], + device='cuda:3'), covar=tensor([0.0360, 0.0344, 0.0164, 0.0137, 0.0260, 0.0308, 0.3088, 0.0323], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0053, 0.0052, 0.0040, 0.0060, 0.0048, 0.0105, 0.0065], + device='cuda:3'), out_proj_covar=tensor([1.3421e-04, 1.0604e-04, 1.0142e-04, 7.8261e-05, 1.1883e-04, 9.6308e-05, + 1.8717e-04, 1.2773e-04], device='cuda:3') +2022-11-15 15:46:03,226 INFO [train.py:876] (3/4) Epoch 2, batch 5700, loss[loss=0.2847, simple_loss=0.2526, pruned_loss=0.1584, over 5558.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.2203, pruned_loss=0.1327, over 1078925.20 frames. ], batch size: 30, lr: 2.94e-02, grad_scale: 8.0 +2022-11-15 15:46:04,071 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1965, 0.8627, 0.9505, 0.6842, 1.1240, 1.0106, 0.7374, 0.7813], + device='cuda:3'), covar=tensor([0.0071, 0.0042, 0.0036, 0.0054, 0.0066, 0.0128, 0.0076, 0.0083], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0021, 0.0021, 0.0020, 0.0019, 0.0020, 0.0022, 0.0019], + device='cuda:3'), out_proj_covar=tensor([3.0587e-05, 3.0742e-05, 3.2738e-05, 2.5364e-05, 2.5868e-05, 2.7086e-05, + 3.7743e-05, 2.8467e-05], device='cuda:3') +2022-11-15 15:46:28,142 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.403e+02 3.143e+02 3.984e+02 1.020e+03, threshold=6.287e+02, percent-clipped=10.0 +2022-11-15 15:46:39,451 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 15:47:14,320 INFO [train.py:876] (3/4) Epoch 2, batch 5800, loss[loss=0.266, simple_loss=0.2468, pruned_loss=0.1426, over 5586.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.2182, pruned_loss=0.1311, over 1079114.44 frames. ], batch size: 23, lr: 2.93e-02, grad_scale: 8.0 +2022-11-15 15:47:33,359 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.23 vs. limit=5.0 +2022-11-15 15:47:35,093 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:47:38,322 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.022e+02 2.861e+02 3.567e+02 9.909e+02, threshold=5.722e+02, percent-clipped=2.0 +2022-11-15 15:47:46,544 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 15:47:53,048 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2022-11-15 15:48:08,629 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:48:12,441 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9085, 3.3527, 2.7910, 3.3264, 2.5336, 2.5387, 1.7486, 2.7243], + device='cuda:3'), covar=tensor([0.1650, 0.0196, 0.0604, 0.0255, 0.0501, 0.0815, 0.2150, 0.0228], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0102, 0.0140, 0.0093, 0.0116, 0.0156, 0.0184, 0.0093], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 15:48:24,659 INFO [train.py:876] (3/4) Epoch 2, batch 5900, loss[loss=0.3736, simple_loss=0.3019, pruned_loss=0.2226, over 5430.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.2186, pruned_loss=0.1315, over 1087208.27 frames. ], batch size: 58, lr: 2.92e-02, grad_scale: 8.0 +2022-11-15 15:48:39,697 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1236, 1.4912, 1.5213, 1.9105, 2.0373, 1.5881, 1.2307, 2.1444], + device='cuda:3'), covar=tensor([0.0067, 0.1039, 0.0737, 0.0154, 0.0152, 0.0598, 0.0668, 0.0059], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0203, 0.0208, 0.0118, 0.0133, 0.0220, 0.0197, 0.0102], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0001], + device='cuda:3') +2022-11-15 15:48:44,632 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6608, 1.8302, 1.6305, 1.1658, 1.9390, 0.8160, 1.9737, 1.3152], + device='cuda:3'), covar=tensor([0.0455, 0.0198, 0.0294, 0.1086, 0.0165, 0.1325, 0.0118, 0.0877], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0071, 0.0069, 0.0111, 0.0072, 0.0118, 0.0062, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0001, 0.0003], + device='cuda:3') +2022-11-15 15:48:49,672 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 2.159e+02 2.799e+02 3.502e+02 4.935e+02, threshold=5.598e+02, percent-clipped=0.0 +2022-11-15 15:48:52,960 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 15:48:53,887 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4513, 1.1674, 1.6341, 1.6869, 0.4918, 1.3897, 0.9600, 1.6274], + device='cuda:3'), covar=tensor([0.0165, 0.0349, 0.0242, 0.0142, 0.0440, 0.0436, 0.0307, 0.0267], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0034, 0.0041, 0.0029, 0.0043, 0.0035, 0.0043, 0.0028], + device='cuda:3'), out_proj_covar=tensor([5.3154e-05, 6.2383e-05, 8.6093e-05, 5.5002e-05, 8.3356e-05, 7.3722e-05, + 7.8945e-05, 5.2185e-05], device='cuda:3') +2022-11-15 15:49:21,103 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1037, 1.8973, 1.5951, 1.2376, 0.5520, 1.5587, 1.2590, 1.0179], + device='cuda:3'), covar=tensor([0.0144, 0.0107, 0.0229, 0.0292, 0.0633, 0.0514, 0.0210, 0.0277], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0024, 0.0028, 0.0031, 0.0028, 0.0020, 0.0023, 0.0027], + device='cuda:3'), out_proj_covar=tensor([4.4489e-05, 3.2791e-05, 4.0658e-05, 4.6802e-05, 4.8264e-05, 3.6292e-05, + 3.7544e-05, 4.2423e-05], device='cuda:3') +2022-11-15 15:49:36,278 INFO [train.py:876] (3/4) Epoch 2, batch 6000, loss[loss=0.2421, simple_loss=0.2276, pruned_loss=0.1284, over 5594.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.2195, pruned_loss=0.1326, over 1085327.50 frames. ], batch size: 22, lr: 2.91e-02, grad_scale: 8.0 +2022-11-15 15:49:36,278 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 15:49:41,587 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7640, 1.4940, 1.8071, 2.1142, 0.8520, 1.6986, 1.3876, 1.9864], + device='cuda:3'), covar=tensor([0.0250, 0.0288, 0.0426, 0.0157, 0.0403, 0.0354, 0.0279, 0.0125], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0034, 0.0041, 0.0028, 0.0043, 0.0034, 0.0043, 0.0027], + device='cuda:3'), out_proj_covar=tensor([5.3841e-05, 6.2601e-05, 8.6249e-05, 5.4152e-05, 8.3868e-05, 7.1806e-05, + 7.9858e-05, 5.2334e-05], device='cuda:3') +2022-11-15 15:49:49,477 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4440, 2.2375, 1.5615, 2.5634, 1.8951, 0.9104, 1.4284, 0.8314], + device='cuda:3'), covar=tensor([0.0244, 0.0138, 0.0274, 0.0237, 0.0408, 0.1031, 0.0561, 0.0302], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0024, 0.0028, 0.0031, 0.0028, 0.0021, 0.0023, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.5460e-05, 3.3216e-05, 4.1313e-05, 4.7483e-05, 4.9232e-05, 3.7141e-05, + 3.7831e-05, 4.3446e-05], device='cuda:3') +2022-11-15 15:49:54,756 INFO [train.py:908] (3/4) Epoch 2, validation: loss=0.1945, simple_loss=0.208, pruned_loss=0.09052, over 1530663.00 frames. +2022-11-15 15:49:54,757 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4525MB +2022-11-15 15:50:07,994 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1845, 2.1688, 2.2102, 1.2374, 2.2006, 2.7563, 2.4277, 2.6405], + device='cuda:3'), covar=tensor([0.0876, 0.0441, 0.0312, 0.0907, 0.0132, 0.0111, 0.0117, 0.0118], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0113, 0.0083, 0.0130, 0.0079, 0.0071, 0.0071, 0.0076], + device='cuda:3'), out_proj_covar=tensor([1.7972e-04, 1.5198e-04, 1.2595e-04, 1.7935e-04, 1.1188e-04, 9.9836e-05, + 1.0104e-04, 1.0324e-04], device='cuda:3') +2022-11-15 15:50:18,567 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1575, 1.4020, 1.1602, 1.1834, 2.5604, 2.0663, 1.4078, 1.8611], + device='cuda:3'), covar=tensor([0.0056, 0.0343, 0.0458, 0.0095, 0.0087, 0.0061, 0.0140, 0.0072], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0016, 0.0018, 0.0017, 0.0018, 0.0021, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.4194e-05, 2.4368e-05, 2.5610e-05, 2.4525e-05, 2.4539e-05, 2.3009e-05, + 2.7763e-05, 2.5052e-05], device='cuda:3') +2022-11-15 15:50:19,048 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.427e+02 3.017e+02 4.086e+02 8.174e+02, threshold=6.035e+02, percent-clipped=9.0 +2022-11-15 15:50:27,262 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7084, 4.5343, 4.3837, 3.9136, 4.5128, 4.2252, 1.5654, 4.5022], + device='cuda:3'), covar=tensor([0.0145, 0.0319, 0.0196, 0.0349, 0.0187, 0.0426, 0.2540, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0052, 0.0051, 0.0041, 0.0058, 0.0044, 0.0103, 0.0065], + device='cuda:3'), out_proj_covar=tensor([1.3075e-04, 1.0357e-04, 1.0079e-04, 8.0386e-05, 1.1566e-04, 9.0140e-05, + 1.8397e-04, 1.2727e-04], device='cuda:3') +2022-11-15 15:51:06,258 INFO [train.py:876] (3/4) Epoch 2, batch 6100, loss[loss=0.1608, simple_loss=0.1688, pruned_loss=0.0764, over 5830.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2185, pruned_loss=0.131, over 1088593.51 frames. ], batch size: 22, lr: 2.90e-02, grad_scale: 8.0 +2022-11-15 15:51:28,267 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-15 15:51:29,934 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.166e+02 2.663e+02 3.329e+02 7.571e+02, threshold=5.325e+02, percent-clipped=1.0 +2022-11-15 15:51:38,753 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:51:38,766 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8852, 1.8219, 1.1204, 1.4401, 0.5050, 2.1383, 1.4343, 0.8347], + device='cuda:3'), covar=tensor([0.0205, 0.0100, 0.0200, 0.0213, 0.0872, 0.0178, 0.0284, 0.0246], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0025, 0.0029, 0.0032, 0.0030, 0.0021, 0.0025, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.8687e-05, 3.4721e-05, 4.2519e-05, 4.8972e-05, 5.1732e-05, 3.7479e-05, + 3.9679e-05, 4.4630e-05], device='cuda:3') +2022-11-15 15:52:01,697 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 +2022-11-15 15:52:03,738 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.32 vs. limit=5.0 +2022-11-15 15:52:12,695 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:52:17,514 INFO [train.py:876] (3/4) Epoch 2, batch 6200, loss[loss=0.2089, simple_loss=0.1909, pruned_loss=0.1135, over 5509.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2181, pruned_loss=0.1313, over 1087343.09 frames. ], batch size: 10, lr: 2.89e-02, grad_scale: 8.0 +2022-11-15 15:52:26,838 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 +2022-11-15 15:52:37,535 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 15:52:41,224 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 2.479e+02 3.223e+02 3.937e+02 9.344e+02, threshold=6.446e+02, percent-clipped=9.0 +2022-11-15 15:52:45,390 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0339, 5.0494, 5.6267, 5.4742, 4.8133, 4.6177, 5.9834, 5.4227], + device='cuda:3'), covar=tensor([0.0326, 0.1327, 0.0243, 0.0726, 0.0336, 0.0224, 0.0706, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0070, 0.0058, 0.0066, 0.0046, 0.0040, 0.0074, 0.0050], + device='cuda:3'), out_proj_covar=tensor([1.0548e-04, 1.4884e-04, 1.2642e-04, 1.4124e-04, 1.0221e-04, 8.5833e-05, + 1.7812e-04, 1.0765e-04], device='cuda:3') +2022-11-15 15:52:49,033 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3678, 1.0357, 1.0755, 1.0839, 2.1044, 1.6947, 1.1977, 1.6389], + device='cuda:3'), covar=tensor([0.0702, 0.0205, 0.0534, 0.0508, 0.0463, 0.0156, 0.0653, 0.0234], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010, 0.0008, 0.0009, 0.0011, 0.0009], + device='cuda:3'), out_proj_covar=tensor([2.0843e-05, 2.0814e-05, 2.0332e-05, 2.4944e-05, 1.9343e-05, 1.9723e-05, + 2.4825e-05, 1.9238e-05], device='cuda:3') +2022-11-15 15:53:12,348 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9639, 4.4276, 4.6415, 4.2516, 5.0586, 4.6457, 4.4088, 4.8509], + device='cuda:3'), covar=tensor([0.0302, 0.0277, 0.0463, 0.0320, 0.0276, 0.0172, 0.0298, 0.0311], + device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0073, 0.0062, 0.0075, 0.0070, 0.0048, 0.0063, 0.0064], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 15:53:27,903 INFO [train.py:876] (3/4) Epoch 2, batch 6300, loss[loss=0.2726, simple_loss=0.2424, pruned_loss=0.1513, over 5592.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.2184, pruned_loss=0.132, over 1083143.46 frames. ], batch size: 22, lr: 2.88e-02, grad_scale: 8.0 +2022-11-15 15:53:52,434 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.381e+02 3.023e+02 3.897e+02 7.097e+02, threshold=6.046e+02, percent-clipped=2.0 +2022-11-15 15:53:56,662 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0318, 0.8852, 0.9863, 0.9795, 1.6366, 1.1127, 1.0393, 1.1961], + device='cuda:3'), covar=tensor([0.0633, 0.0242, 0.0913, 0.0602, 0.0366, 0.0595, 0.0481, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010, 0.0008, 0.0009, 0.0011, 0.0009], + device='cuda:3'), out_proj_covar=tensor([2.1342e-05, 2.1567e-05, 2.0958e-05, 2.5196e-05, 1.9197e-05, 2.0055e-05, + 2.4460e-05, 1.9616e-05], device='cuda:3') +2022-11-15 15:54:08,304 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0386, 1.2657, 1.6000, 1.3735, 1.9727, 1.6981, 1.1206, 1.4827], + device='cuda:3'), covar=tensor([0.1749, 0.0244, 0.0328, 0.1283, 0.0270, 0.0408, 0.0717, 0.0194], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010, 0.0008, 0.0009, 0.0010, 0.0008], + device='cuda:3'), out_proj_covar=tensor([2.0968e-05, 2.1375e-05, 2.1031e-05, 2.5024e-05, 1.9327e-05, 1.9795e-05, + 2.4128e-05, 1.9307e-05], device='cuda:3') +2022-11-15 15:54:28,637 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 15:54:39,265 INFO [train.py:876] (3/4) Epoch 2, batch 6400, loss[loss=0.266, simple_loss=0.2385, pruned_loss=0.1467, over 5705.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.2189, pruned_loss=0.1324, over 1081710.71 frames. ], batch size: 28, lr: 2.87e-02, grad_scale: 16.0 +2022-11-15 15:55:03,886 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.247e+02 2.645e+02 3.306e+02 5.348e+02, threshold=5.289e+02, percent-clipped=0.0 +2022-11-15 15:55:19,575 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.18 vs. limit=2.0 +2022-11-15 15:55:24,664 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:55:39,404 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 15:55:39,548 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.13 vs. limit=5.0 +2022-11-15 15:55:47,230 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6704, 4.1858, 3.6896, 4.2637, 4.3555, 3.6497, 3.6547, 3.4812], + device='cuda:3'), covar=tensor([0.0401, 0.0327, 0.0516, 0.0234, 0.0247, 0.0349, 0.0373, 0.0398], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0075, 0.0102, 0.0071, 0.0093, 0.0089, 0.0079, 0.0071], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 15:55:50,602 INFO [train.py:876] (3/4) Epoch 2, batch 6500, loss[loss=0.2864, simple_loss=0.2488, pruned_loss=0.162, over 5573.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.217, pruned_loss=0.1304, over 1084869.93 frames. ], batch size: 46, lr: 2.86e-02, grad_scale: 16.0 +2022-11-15 15:55:51,764 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2022-11-15 15:56:09,396 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 15:56:15,581 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.395e+02 2.916e+02 4.086e+02 9.118e+02, threshold=5.832e+02, percent-clipped=10.0 +2022-11-15 15:56:19,590 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 15:57:02,021 INFO [train.py:876] (3/4) Epoch 2, batch 6600, loss[loss=0.2618, simple_loss=0.2418, pruned_loss=0.1409, over 5549.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.2159, pruned_loss=0.1292, over 1083430.79 frames. ], batch size: 21, lr: 2.85e-02, grad_scale: 16.0 +2022-11-15 15:57:25,735 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.160e+02 2.814e+02 3.632e+02 6.196e+02, threshold=5.627e+02, percent-clipped=2.0 +2022-11-15 15:57:41,263 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 15:58:12,915 INFO [train.py:876] (3/4) Epoch 2, batch 6700, loss[loss=0.1944, simple_loss=0.2055, pruned_loss=0.09163, over 5505.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.2166, pruned_loss=0.1302, over 1078446.48 frames. ], batch size: 17, lr: 2.85e-02, grad_scale: 16.0 +2022-11-15 15:58:36,324 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 2.069e+02 2.722e+02 3.403e+02 5.968e+02, threshold=5.443e+02, percent-clipped=2.0 +2022-11-15 15:59:23,058 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9817, 1.9600, 4.0966, 2.6949, 3.8924, 3.0320, 3.7672, 3.9806], + device='cuda:3'), covar=tensor([0.0035, 0.0478, 0.0065, 0.0406, 0.0046, 0.0271, 0.0175, 0.0104], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0161, 0.0110, 0.0170, 0.0096, 0.0148, 0.0154, 0.0117], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 15:59:24,077 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:59:24,581 INFO [train.py:876] (3/4) Epoch 2, batch 6800, loss[loss=0.2965, simple_loss=0.255, pruned_loss=0.169, over 5450.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.2161, pruned_loss=0.1299, over 1079122.92 frames. ], batch size: 58, lr: 2.84e-02, grad_scale: 16.0 +2022-11-15 15:59:34,447 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.17 vs. limit=2.0 +2022-11-15 15:59:38,919 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 15:59:45,933 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 15:59:48,497 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.309e+02 2.876e+02 3.823e+02 9.866e+02, threshold=5.752e+02, percent-clipped=3.0 +2022-11-15 15:59:53,394 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5307, 3.9029, 3.7001, 3.4072, 3.6551, 3.5805, 1.3915, 3.4132], + device='cuda:3'), covar=tensor([0.0291, 0.0217, 0.0231, 0.0263, 0.0309, 0.0274, 0.2452, 0.0352], + device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0059, 0.0061, 0.0047, 0.0067, 0.0052, 0.0116, 0.0075], + device='cuda:3'), out_proj_covar=tensor([1.4714e-04, 1.1681e-04, 1.2010e-04, 9.1995e-05, 1.3371e-04, 1.0688e-04, + 2.0903e-04, 1.4748e-04], device='cuda:3') +2022-11-15 15:59:58,377 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:00:06,009 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2873, 1.7543, 1.8236, 2.1894, 2.4402, 1.7231, 1.3955, 2.5902], + device='cuda:3'), covar=tensor([0.0058, 0.0928, 0.0738, 0.0158, 0.0118, 0.0703, 0.0666, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0212, 0.0214, 0.0128, 0.0144, 0.0229, 0.0200, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:00:08,027 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:00:18,339 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 16:00:18,879 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6051, 3.9310, 2.9879, 1.7735, 3.8539, 1.3574, 3.7909, 2.0450], + device='cuda:3'), covar=tensor([0.0730, 0.0089, 0.0450, 0.1570, 0.0099, 0.1707, 0.0100, 0.1356], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0072, 0.0070, 0.0111, 0.0075, 0.0117, 0.0063, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:00:29,248 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:00:35,848 INFO [train.py:876] (3/4) Epoch 2, batch 6900, loss[loss=0.2948, simple_loss=0.232, pruned_loss=0.1789, over 4133.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.2166, pruned_loss=0.1304, over 1075198.38 frames. ], batch size: 181, lr: 2.83e-02, grad_scale: 16.0 +2022-11-15 16:00:41,937 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 16:01:00,097 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.412e+02 3.170e+02 4.129e+02 8.263e+02, threshold=6.339e+02, percent-clipped=8.0 +2022-11-15 16:01:02,893 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5575, 4.1646, 3.3971, 2.1011, 3.9995, 1.3776, 3.9163, 2.2702], + device='cuda:3'), covar=tensor([0.1134, 0.0110, 0.0480, 0.1758, 0.0115, 0.1778, 0.0137, 0.1434], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0075, 0.0072, 0.0114, 0.0077, 0.0117, 0.0065, 0.0113], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:01:36,368 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 16:01:47,610 INFO [train.py:876] (3/4) Epoch 2, batch 7000, loss[loss=0.2981, simple_loss=0.2538, pruned_loss=0.1712, over 5420.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.2185, pruned_loss=0.1317, over 1082193.65 frames. ], batch size: 58, lr: 2.82e-02, grad_scale: 16.0 +2022-11-15 16:02:11,792 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 2.311e+02 3.068e+02 3.846e+02 6.793e+02, threshold=6.137e+02, percent-clipped=2.0 +2022-11-15 16:02:22,961 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 16:02:24,522 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0590, 3.9644, 4.3882, 4.3266, 4.1050, 2.9110, 4.6800, 4.0204], + device='cuda:3'), covar=tensor([0.0359, 0.0939, 0.0299, 0.0460, 0.0389, 0.0496, 0.0730, 0.0399], + device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0071, 0.0059, 0.0068, 0.0048, 0.0043, 0.0078, 0.0051], + device='cuda:3'), out_proj_covar=tensor([1.0925e-04, 1.5522e-04, 1.3183e-04, 1.4922e-04, 1.0671e-04, 9.5233e-05, + 1.9174e-04, 1.1167e-04], device='cuda:3') +2022-11-15 16:02:30,020 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1544, 4.2346, 3.6043, 4.1860, 3.3764, 3.1087, 2.2333, 3.5131], + device='cuda:3'), covar=tensor([0.1703, 0.0143, 0.0518, 0.0158, 0.0407, 0.0624, 0.2117, 0.0208], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0103, 0.0140, 0.0093, 0.0121, 0.0150, 0.0176, 0.0094], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:02:30,070 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0771, 1.0368, 0.9808, 0.5021, 0.9856, 1.1812, 0.7529, 1.3189], + device='cuda:3'), covar=tensor([0.0220, 0.0085, 0.0082, 0.0060, 0.0050, 0.0076, 0.0159, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0019, 0.0020, 0.0019, 0.0018, 0.0017, 0.0019, 0.0015], + device='cuda:3'), out_proj_covar=tensor([2.9441e-05, 2.8450e-05, 2.9442e-05, 2.3543e-05, 2.3016e-05, 2.3118e-05, + 3.4378e-05, 2.2403e-05], device='cuda:3') +2022-11-15 16:02:55,070 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6632, 2.5385, 2.5542, 2.4683, 2.7899, 2.5366, 2.5313, 2.6884], + device='cuda:3'), covar=tensor([0.0738, 0.0799, 0.0911, 0.0733, 0.0679, 0.0424, 0.0752, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0080, 0.0063, 0.0077, 0.0072, 0.0049, 0.0065, 0.0063], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 16:02:58,389 INFO [train.py:876] (3/4) Epoch 2, batch 7100, loss[loss=0.1277, simple_loss=0.1268, pruned_loss=0.06432, over 4675.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.2178, pruned_loss=0.13, over 1080355.61 frames. ], batch size: 5, lr: 2.81e-02, grad_scale: 16.0 +2022-11-15 16:03:13,675 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:03:17,831 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5986, 1.4467, 1.7508, 1.5631, 1.1992, 1.3609, 1.1076, 1.9337], + device='cuda:3'), covar=tensor([0.0103, 0.0209, 0.0263, 0.0137, 0.0304, 0.0184, 0.0286, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0040, 0.0045, 0.0033, 0.0049, 0.0038, 0.0046, 0.0030], + device='cuda:3'), out_proj_covar=tensor([6.3921e-05, 7.8025e-05, 9.8137e-05, 6.5318e-05, 9.8034e-05, 8.0566e-05, + 9.0789e-05, 5.9160e-05], device='cuda:3') +2022-11-15 16:03:23,297 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 2.305e+02 2.820e+02 3.726e+02 7.277e+02, threshold=5.640e+02, percent-clipped=2.0 +2022-11-15 16:03:27,747 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:03:31,958 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2022-11-15 16:03:37,784 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:03:47,213 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:03:56,660 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-15 16:04:00,332 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:04:09,794 INFO [train.py:876] (3/4) Epoch 2, batch 7200, loss[loss=0.2288, simple_loss=0.2091, pruned_loss=0.1243, over 5634.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.218, pruned_loss=0.1304, over 1082455.48 frames. ], batch size: 38, lr: 2.80e-02, grad_scale: 16.0 +2022-11-15 16:04:10,627 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:04:11,848 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:04:19,432 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:04:24,249 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-15 16:04:33,141 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.188e+02 2.926e+02 3.996e+02 7.445e+02, threshold=5.852e+02, percent-clipped=7.0 +2022-11-15 16:04:37,152 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3392, 1.8243, 1.7934, 1.8241, 1.1286, 1.6610, 1.2793, 1.7737], + device='cuda:3'), covar=tensor([0.0545, 0.0138, 0.0339, 0.0163, 0.0428, 0.0361, 0.0798, 0.0147], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0103, 0.0143, 0.0097, 0.0121, 0.0158, 0.0180, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:04:37,958 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.87 vs. limit=5.0 +2022-11-15 16:04:50,439 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4904, 3.9247, 2.9186, 2.5413, 3.0141, 3.3881, 3.4428, 3.5470], + device='cuda:3'), covar=tensor([0.0625, 0.0229, 0.0189, 0.0503, 0.0074, 0.0077, 0.0085, 0.0060], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0133, 0.0097, 0.0146, 0.0090, 0.0080, 0.0081, 0.0090], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 16:05:50,250 INFO [train.py:876] (3/4) Epoch 3, batch 0, loss[loss=0.2511, simple_loss=0.2323, pruned_loss=0.135, over 5564.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.2323, pruned_loss=0.135, over 5564.00 frames. ], batch size: 24, lr: 2.66e-02, grad_scale: 16.0 +2022-11-15 16:05:50,251 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 16:05:59,546 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9053, 1.2441, 1.5891, 2.0975, 1.1582, 1.5568, 1.0200, 2.1027], + device='cuda:3'), covar=tensor([0.0089, 0.0301, 0.0162, 0.0072, 0.0351, 0.0438, 0.0281, 0.0072], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0037, 0.0042, 0.0031, 0.0048, 0.0037, 0.0044, 0.0028], + device='cuda:3'), out_proj_covar=tensor([6.1677e-05, 7.2798e-05, 9.3187e-05, 6.1843e-05, 9.7840e-05, 7.9605e-05, + 8.7051e-05, 5.5923e-05], device='cuda:3') +2022-11-15 16:06:07,522 INFO [train.py:908] (3/4) Epoch 3, validation: loss=0.1917, simple_loss=0.2065, pruned_loss=0.08845, over 1530663.00 frames. +2022-11-15 16:06:07,523 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4525MB +2022-11-15 16:06:09,681 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:06:20,066 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 16:06:20,944 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 16:06:26,592 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8697, 5.0855, 3.9160, 2.4244, 5.0338, 2.1852, 4.2144, 3.1366], + device='cuda:3'), covar=tensor([0.0561, 0.0099, 0.0335, 0.1629, 0.0076, 0.1362, 0.0160, 0.1118], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0077, 0.0076, 0.0116, 0.0081, 0.0120, 0.0068, 0.0116], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:06:42,641 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 +2022-11-15 16:06:52,131 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 2.390e+02 2.831e+02 3.708e+02 1.001e+03, threshold=5.662e+02, percent-clipped=6.0 +2022-11-15 16:07:19,306 INFO [train.py:876] (3/4) Epoch 3, batch 100, loss[loss=0.2163, simple_loss=0.2046, pruned_loss=0.1141, over 5167.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.2138, pruned_loss=0.1263, over 432411.24 frames. ], batch size: 8, lr: 2.65e-02, grad_scale: 16.0 +2022-11-15 16:07:37,633 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1242, 4.1263, 4.2155, 4.3094, 3.8470, 3.2849, 4.6674, 3.9334], + device='cuda:3'), covar=tensor([0.0390, 0.0594, 0.0369, 0.0458, 0.0491, 0.0394, 0.0579, 0.0432], + device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0072, 0.0060, 0.0069, 0.0050, 0.0043, 0.0079, 0.0052], + device='cuda:3'), out_proj_covar=tensor([1.1317e-04, 1.6014e-04, 1.3332e-04, 1.5159e-04, 1.1319e-04, 9.6336e-05, + 1.9466e-04, 1.1469e-04], device='cuda:3') +2022-11-15 16:07:42,201 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 16:07:50,099 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5422, 1.8292, 1.5244, 1.7968, 1.0905, 1.3141, 1.2685, 2.0707], + device='cuda:3'), covar=tensor([0.0230, 0.0227, 0.0406, 0.0667, 0.0450, 0.0536, 0.0527, 0.0278], + device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0037, 0.0045, 0.0032, 0.0050, 0.0038, 0.0047, 0.0029], + device='cuda:3'), out_proj_covar=tensor([6.4270e-05, 7.3113e-05, 9.8987e-05, 6.4770e-05, 1.0119e-04, 8.1867e-05, + 9.2616e-05, 5.8701e-05], device='cuda:3') +2022-11-15 16:08:03,405 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 2.280e+02 2.660e+02 3.506e+02 7.201e+02, threshold=5.320e+02, percent-clipped=1.0 +2022-11-15 16:08:18,656 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:08:20,109 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2468, 2.8258, 3.7207, 4.5535, 4.7310, 3.8869, 2.8803, 4.6208], + device='cuda:3'), covar=tensor([0.0039, 0.1294, 0.0894, 0.0322, 0.0095, 0.0635, 0.0907, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0211, 0.0219, 0.0135, 0.0147, 0.0216, 0.0203, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:08:27,270 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:08:30,531 INFO [train.py:876] (3/4) Epoch 3, batch 200, loss[loss=0.2949, simple_loss=0.2465, pruned_loss=0.1717, over 5422.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.2153, pruned_loss=0.127, over 697025.50 frames. ], batch size: 58, lr: 2.64e-02, grad_scale: 16.0 +2022-11-15 16:08:41,084 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:08:48,052 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:08:53,272 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:08:53,361 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:08:53,815 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 16:09:03,869 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 +2022-11-15 16:09:07,708 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 16:09:11,204 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:09:15,919 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 2.295e+02 2.977e+02 3.816e+02 8.125e+02, threshold=5.953e+02, percent-clipped=6.0 +2022-11-15 16:09:16,006 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:09:27,546 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:09:41,501 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:09:42,824 INFO [train.py:876] (3/4) Epoch 3, batch 300, loss[loss=0.2264, simple_loss=0.2172, pruned_loss=0.1179, over 5631.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.2139, pruned_loss=0.1273, over 845324.29 frames. ], batch size: 32, lr: 2.63e-02, grad_scale: 16.0 +2022-11-15 16:09:44,589 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.37 vs. limit=2.0 +2022-11-15 16:10:27,145 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.433e+02 2.202e+02 2.979e+02 3.584e+02 7.564e+02, threshold=5.959e+02, percent-clipped=7.0 +2022-11-15 16:10:36,815 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5296, 2.2950, 3.7439, 4.7703, 4.7892, 3.5261, 3.0882, 4.4977], + device='cuda:3'), covar=tensor([0.0044, 0.1228, 0.0567, 0.0099, 0.0039, 0.0733, 0.0664, 0.0033], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0213, 0.0224, 0.0137, 0.0146, 0.0219, 0.0201, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:10:55,020 INFO [train.py:876] (3/4) Epoch 3, batch 400, loss[loss=0.1799, simple_loss=0.1809, pruned_loss=0.0894, over 5673.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.2127, pruned_loss=0.1252, over 942989.84 frames. ], batch size: 11, lr: 2.62e-02, grad_scale: 16.0 +2022-11-15 16:11:02,804 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2265, 1.5369, 0.7609, 0.6272, 1.2284, 1.2492, 0.8705, 1.3370], + device='cuda:3'), covar=tensor([0.0057, 0.0036, 0.0130, 0.0060, 0.0040, 0.0121, 0.0119, 0.0246], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0019, 0.0021, 0.0022, 0.0020, 0.0019, 0.0021, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.7946e-05, 2.6459e-05, 3.0089e-05, 2.6909e-05, 2.5178e-05, 2.5270e-05, + 3.7255e-05, 2.5130e-05], device='cuda:3') +2022-11-15 16:11:43,139 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.230e+02 2.415e+02 2.867e+02 3.380e+02 8.857e+02, threshold=5.735e+02, percent-clipped=3.0 +2022-11-15 16:11:45,026 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 16:12:10,979 INFO [train.py:876] (3/4) Epoch 3, batch 500, loss[loss=0.1463, simple_loss=0.1561, pruned_loss=0.06825, over 4947.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.2124, pruned_loss=0.1244, over 997776.38 frames. ], batch size: 5, lr: 2.62e-02, grad_scale: 16.0 +2022-11-15 16:12:27,968 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:12:42,368 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 16:12:48,142 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:12:48,259 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5893, 3.7051, 3.1153, 1.5796, 3.1843, 4.2135, 3.3889, 4.3652], + device='cuda:3'), covar=tensor([0.0641, 0.0308, 0.0238, 0.0771, 0.0059, 0.0052, 0.0076, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0130, 0.0100, 0.0158, 0.0096, 0.0088, 0.0084, 0.0094], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 16:12:55,529 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.220e+02 2.761e+02 3.566e+02 7.983e+02, threshold=5.522e+02, percent-clipped=3.0 +2022-11-15 16:13:02,512 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:13:16,845 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2022-11-15 16:13:21,498 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:13:22,752 INFO [train.py:876] (3/4) Epoch 3, batch 600, loss[loss=0.2754, simple_loss=0.2362, pruned_loss=0.1573, over 5267.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.2124, pruned_loss=0.1245, over 1036015.79 frames. ], batch size: 79, lr: 2.61e-02, grad_scale: 16.0 +2022-11-15 16:13:25,324 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1635, 2.9939, 3.0067, 2.9138, 3.3018, 3.1754, 3.1697, 3.1545], + device='cuda:3'), covar=tensor([0.0689, 0.0663, 0.0786, 0.0656, 0.0567, 0.0379, 0.0515, 0.0642], + device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0081, 0.0066, 0.0078, 0.0075, 0.0050, 0.0066, 0.0065], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 16:13:50,617 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4929, 1.3387, 1.0247, 0.5829, 1.1185, 1.3961, 0.6387, 1.5693], + device='cuda:3'), covar=tensor([0.0032, 0.0052, 0.0047, 0.0052, 0.0048, 0.0078, 0.0048, 0.0023], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0019, 0.0019, 0.0019, 0.0018, 0.0016, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.3296e-05, 2.6522e-05, 2.6812e-05, 2.3294e-05, 2.2487e-05, 2.2090e-05, + 3.1319e-05, 2.2247e-05], device='cuda:3') +2022-11-15 16:13:55,649 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:13:58,626 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:14:02,261 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8910, 3.8544, 3.1164, 3.8266, 3.1573, 2.5483, 2.0889, 3.2152], + device='cuda:3'), covar=tensor([0.1993, 0.0167, 0.0746, 0.0257, 0.0476, 0.1044, 0.2315, 0.0204], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0104, 0.0145, 0.0101, 0.0127, 0.0161, 0.0184, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:14:07,258 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.220e+02 2.668e+02 3.263e+02 7.749e+02, threshold=5.336e+02, percent-clipped=2.0 +2022-11-15 16:14:25,314 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0327, 1.1332, 0.9959, 0.6623, 1.2065, 1.3753, 0.5783, 1.6393], + device='cuda:3'), covar=tensor([0.0051, 0.0044, 0.0040, 0.0036, 0.0034, 0.0126, 0.0068, 0.0042], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0019, 0.0019, 0.0018, 0.0016, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.4287e-05, 2.8037e-05, 2.6620e-05, 2.3540e-05, 2.2556e-05, 2.2384e-05, + 3.1800e-05, 2.2172e-05], device='cuda:3') +2022-11-15 16:14:33,897 INFO [train.py:876] (3/4) Epoch 3, batch 700, loss[loss=0.2986, simple_loss=0.2531, pruned_loss=0.1721, over 4981.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.2129, pruned_loss=0.1247, over 1046898.83 frames. ], batch size: 109, lr: 2.60e-02, grad_scale: 16.0 +2022-11-15 16:14:42,129 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:15:18,348 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 2.381e+02 2.880e+02 4.091e+02 8.657e+02, threshold=5.760e+02, percent-clipped=7.0 +2022-11-15 16:15:23,042 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2022-11-15 16:15:44,960 INFO [train.py:876] (3/4) Epoch 3, batch 800, loss[loss=0.2415, simple_loss=0.2228, pruned_loss=0.1301, over 5708.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.2136, pruned_loss=0.1255, over 1067652.05 frames. ], batch size: 17, lr: 2.59e-02, grad_scale: 16.0 +2022-11-15 16:15:49,036 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 16:16:21,735 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:16:30,195 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 2.163e+02 2.793e+02 3.391e+02 6.505e+02, threshold=5.586e+02, percent-clipped=1.0 +2022-11-15 16:16:56,552 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:16:56,910 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 16:16:57,170 INFO [train.py:876] (3/4) Epoch 3, batch 900, loss[loss=0.2048, simple_loss=0.194, pruned_loss=0.1078, over 5461.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.2147, pruned_loss=0.1256, over 1076054.61 frames. ], batch size: 11, lr: 2.59e-02, grad_scale: 16.0 +2022-11-15 16:17:12,918 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0788, 1.3482, 1.1382, 0.6973, 1.4786, 1.2249, 1.2511, 1.3208], + device='cuda:3'), covar=tensor([0.0353, 0.0206, 0.0320, 0.0864, 0.0138, 0.0319, 0.0248, 0.0243], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0009, 0.0008, 0.0008, 0.0010, 0.0008], + device='cuda:3'), out_proj_covar=tensor([2.1435e-05, 2.3895e-05, 2.1261e-05, 2.6063e-05, 2.0287e-05, 2.0921e-05, + 2.4990e-05, 2.1281e-05], device='cuda:3') +2022-11-15 16:17:41,564 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.326e+02 2.765e+02 3.451e+02 5.869e+02, threshold=5.530e+02, percent-clipped=2.0 +2022-11-15 16:17:58,786 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3195, 3.0525, 2.2040, 1.7189, 2.9031, 1.2706, 2.8021, 1.7273], + device='cuda:3'), covar=tensor([0.0716, 0.0141, 0.0517, 0.1164, 0.0185, 0.1335, 0.0180, 0.1151], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0078, 0.0074, 0.0113, 0.0082, 0.0121, 0.0069, 0.0116], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:18:09,248 INFO [train.py:876] (3/4) Epoch 3, batch 1000, loss[loss=0.2053, simple_loss=0.1933, pruned_loss=0.1087, over 5462.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.2138, pruned_loss=0.1252, over 1079036.02 frames. ], batch size: 12, lr: 2.58e-02, grad_scale: 16.0 +2022-11-15 16:18:13,445 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:18:13,813 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.11 vs. limit=5.0 +2022-11-15 16:18:47,571 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6228, 2.5357, 1.6267, 1.0650, 1.7435, 2.8739, 2.1593, 2.9398], + device='cuda:3'), covar=tensor([0.0886, 0.0435, 0.0574, 0.1077, 0.0187, 0.0129, 0.0133, 0.0111], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0146, 0.0106, 0.0171, 0.0107, 0.0092, 0.0093, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 16:18:53,835 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.311e+02 2.148e+02 2.818e+02 3.595e+02 6.939e+02, threshold=5.636e+02, percent-clipped=2.0 +2022-11-15 16:19:07,245 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-15 16:19:07,974 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 +2022-11-15 16:19:12,078 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.7210, 5.3476, 5.6730, 5.1716, 5.6600, 5.8713, 5.0924, 5.6219], + device='cuda:3'), covar=tensor([0.0401, 0.0247, 0.0380, 0.0231, 0.0497, 0.0064, 0.0190, 0.0187], + device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0084, 0.0070, 0.0082, 0.0080, 0.0051, 0.0068, 0.0072], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:19:20,894 INFO [train.py:876] (3/4) Epoch 3, batch 1100, loss[loss=0.2404, simple_loss=0.218, pruned_loss=0.1314, over 5780.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.213, pruned_loss=0.124, over 1082894.36 frames. ], batch size: 26, lr: 2.57e-02, grad_scale: 32.0 +2022-11-15 16:20:05,751 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.101e+02 2.719e+02 3.370e+02 6.510e+02, threshold=5.438e+02, percent-clipped=5.0 +2022-11-15 16:20:25,876 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.20 vs. limit=2.0 +2022-11-15 16:20:27,994 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 +2022-11-15 16:20:32,642 INFO [train.py:876] (3/4) Epoch 3, batch 1200, loss[loss=0.3272, simple_loss=0.2656, pruned_loss=0.1944, over 4687.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2118, pruned_loss=0.1232, over 1079862.54 frames. ], batch size: 135, lr: 2.56e-02, grad_scale: 16.0 +2022-11-15 16:20:46,267 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9330, 5.1407, 4.5158, 4.2218, 4.9323, 4.5154, 1.7821, 4.8703], + device='cuda:3'), covar=tensor([0.0157, 0.0110, 0.0241, 0.0296, 0.0144, 0.0189, 0.2004, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0060, 0.0064, 0.0050, 0.0072, 0.0052, 0.0118, 0.0079], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 16:20:49,312 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 16:21:17,794 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.202e+02 2.751e+02 3.254e+02 5.977e+02, threshold=5.502e+02, percent-clipped=4.0 +2022-11-15 16:21:40,188 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.25 vs. limit=5.0 +2022-11-15 16:21:43,802 INFO [train.py:876] (3/4) Epoch 3, batch 1300, loss[loss=0.2019, simple_loss=0.195, pruned_loss=0.1044, over 5265.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2101, pruned_loss=0.1207, over 1087145.02 frames. ], batch size: 8, lr: 2.56e-02, grad_scale: 16.0 +2022-11-15 16:21:48,898 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:22:11,352 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:22:22,441 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:22:29,934 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.010e+02 2.682e+02 3.501e+02 1.988e+03, threshold=5.365e+02, percent-clipped=5.0 +2022-11-15 16:22:37,320 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 16:22:39,961 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7688, 3.9882, 3.9514, 3.6638, 3.0275, 4.3801, 3.2493, 3.9685], + device='cuda:3'), covar=tensor([0.0148, 0.0101, 0.0051, 0.0104, 0.0164, 0.0023, 0.0084, 0.0020], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0064, 0.0084, 0.0080, 0.0123, 0.0079, 0.0105, 0.0070], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:22:54,789 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:22:55,281 INFO [train.py:876] (3/4) Epoch 3, batch 1400, loss[loss=0.2915, simple_loss=0.2453, pruned_loss=0.1689, over 5347.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2096, pruned_loss=0.1209, over 1084700.00 frames. ], batch size: 70, lr: 2.55e-02, grad_scale: 8.0 +2022-11-15 16:23:14,024 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2288, 2.7192, 2.5708, 1.4089, 2.1629, 1.3545, 0.9663, 1.6000], + device='cuda:3'), covar=tensor([0.0033, 0.0022, 0.0021, 0.0035, 0.0016, 0.0028, 0.0030, 0.0027], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0021, 0.0020, 0.0018, 0.0021, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.6566e-05, 3.1502e-05, 2.9047e-05, 2.6401e-05, 2.4842e-05, 2.3484e-05, + 3.6150e-05, 2.3861e-05], device='cuda:3') +2022-11-15 16:23:36,108 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:23:42,148 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 2.137e+02 2.778e+02 3.586e+02 7.557e+02, threshold=5.556e+02, percent-clipped=2.0 +2022-11-15 16:23:55,321 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 16:23:57,118 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0541, 4.9095, 4.2514, 4.8611, 4.9792, 4.3437, 4.2882, 4.1132], + device='cuda:3'), covar=tensor([0.0225, 0.0282, 0.0597, 0.0271, 0.0173, 0.0300, 0.0275, 0.0573], + device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0117, 0.0083, 0.0105, 0.0099, 0.0087, 0.0080], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:24:04,412 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9529, 1.8271, 1.9286, 3.0281, 2.9941, 2.2502, 1.7951, 3.1984], + device='cuda:3'), covar=tensor([0.0058, 0.1118, 0.0915, 0.0228, 0.0107, 0.0713, 0.0839, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0216, 0.0230, 0.0150, 0.0150, 0.0227, 0.0206, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:24:07,319 INFO [train.py:876] (3/4) Epoch 3, batch 1500, loss[loss=0.2161, simple_loss=0.2106, pruned_loss=0.1108, over 5585.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2103, pruned_loss=0.1217, over 1087369.41 frames. ], batch size: 23, lr: 2.54e-02, grad_scale: 8.0 +2022-11-15 16:24:19,074 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:24:21,727 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 +2022-11-15 16:24:35,756 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-15 16:24:36,026 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:24:52,644 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.433e+02 2.944e+02 3.861e+02 5.407e+02, threshold=5.888e+02, percent-clipped=0.0 +2022-11-15 16:25:14,069 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9372, 2.1271, 1.3346, 2.6332, 1.6621, 1.7858, 1.8292, 2.3457], + device='cuda:3'), covar=tensor([0.0250, 0.0274, 0.0644, 0.0360, 0.0492, 0.0405, 0.0352, 0.1015], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0040, 0.0045, 0.0034, 0.0049, 0.0040, 0.0047, 0.0033], + device='cuda:3'), out_proj_covar=tensor([7.2629e-05, 8.0554e-05, 1.0715e-04, 7.1275e-05, 1.0361e-04, 8.9546e-05, + 9.5487e-05, 6.7610e-05], device='cuda:3') +2022-11-15 16:25:18,685 INFO [train.py:876] (3/4) Epoch 3, batch 1600, loss[loss=0.2118, simple_loss=0.2015, pruned_loss=0.1111, over 5673.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2087, pruned_loss=0.1211, over 1074549.23 frames. ], batch size: 36, lr: 2.53e-02, grad_scale: 8.0 +2022-11-15 16:25:19,568 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:25:21,620 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:26:04,923 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.236e+02 2.729e+02 3.361e+02 8.285e+02, threshold=5.459e+02, percent-clipped=7.0 +2022-11-15 16:26:05,778 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:26:26,175 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:26:30,133 INFO [train.py:876] (3/4) Epoch 3, batch 1700, loss[loss=0.1827, simple_loss=0.1843, pruned_loss=0.09059, over 5534.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2103, pruned_loss=0.1223, over 1078074.79 frames. ], batch size: 21, lr: 2.53e-02, grad_scale: 8.0 +2022-11-15 16:27:15,067 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.221e+02 2.797e+02 3.532e+02 1.022e+03, threshold=5.594e+02, percent-clipped=2.0 +2022-11-15 16:27:17,132 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.86 vs. limit=5.0 +2022-11-15 16:27:41,527 INFO [train.py:876] (3/4) Epoch 3, batch 1800, loss[loss=0.2242, simple_loss=0.2042, pruned_loss=0.1221, over 5271.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2097, pruned_loss=0.121, over 1084629.59 frames. ], batch size: 79, lr: 2.52e-02, grad_scale: 8.0 +2022-11-15 16:27:49,684 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:27:59,870 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:28:22,192 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1997, 3.5710, 2.9632, 3.0453, 2.1505, 3.5282, 2.3993, 3.2505], + device='cuda:3'), covar=tensor([0.0177, 0.0053, 0.0075, 0.0127, 0.0218, 0.0031, 0.0145, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0067, 0.0087, 0.0083, 0.0127, 0.0079, 0.0107, 0.0074], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:28:26,562 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.508e+02 2.970e+02 3.858e+02 6.298e+02, threshold=5.940e+02, percent-clipped=3.0 +2022-11-15 16:28:30,163 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8612, 2.0866, 2.0149, 1.2938, 2.2369, 1.9806, 2.5682, 1.9706], + device='cuda:3'), covar=tensor([0.0022, 0.0058, 0.0084, 0.0042, 0.0014, 0.0040, 0.0021, 0.0026], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0012, 0.0015, 0.0013, 0.0014, 0.0015, 0.0013], + device='cuda:3'), out_proj_covar=tensor([1.8253e-05, 1.9450e-05, 1.7582e-05, 1.9360e-05, 1.7495e-05, 1.8034e-05, + 1.9526e-05, 1.7695e-05], device='cuda:3') +2022-11-15 16:28:42,871 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:28:47,014 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9481, 2.1011, 3.2045, 4.2197, 4.0096, 2.9415, 2.3129, 4.0635], + device='cuda:3'), covar=tensor([0.0054, 0.1571, 0.0879, 0.0201, 0.0102, 0.1134, 0.1195, 0.0067], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0220, 0.0230, 0.0155, 0.0151, 0.0233, 0.0213, 0.0116], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:28:49,198 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:28:51,926 INFO [train.py:876] (3/4) Epoch 3, batch 1900, loss[loss=0.2273, simple_loss=0.2173, pruned_loss=0.1187, over 5530.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.2102, pruned_loss=0.1211, over 1093033.94 frames. ], batch size: 14, lr: 2.51e-02, grad_scale: 8.0 +2022-11-15 16:29:34,093 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:29:37,431 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.066e+02 2.615e+02 3.411e+02 7.424e+02, threshold=5.231e+02, percent-clipped=4.0 +2022-11-15 16:29:58,078 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:30:02,025 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.58 vs. limit=2.0 +2022-11-15 16:30:02,345 INFO [train.py:876] (3/4) Epoch 3, batch 2000, loss[loss=0.2301, simple_loss=0.214, pruned_loss=0.1231, over 5655.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2079, pruned_loss=0.1184, over 1096874.90 frames. ], batch size: 29, lr: 2.51e-02, grad_scale: 8.0 +2022-11-15 16:30:23,373 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.56 vs. limit=5.0 +2022-11-15 16:30:32,354 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:30:48,525 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 2.235e+02 2.805e+02 3.719e+02 8.798e+02, threshold=5.609e+02, percent-clipped=4.0 +2022-11-15 16:31:14,088 INFO [train.py:876] (3/4) Epoch 3, batch 2100, loss[loss=0.1591, simple_loss=0.1636, pruned_loss=0.07734, over 5506.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2076, pruned_loss=0.1181, over 1090340.13 frames. ], batch size: 12, lr: 2.50e-02, grad_scale: 8.0 +2022-11-15 16:31:22,844 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:31:56,397 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:31:57,921 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7101, 2.2665, 2.3767, 1.1819, 2.6786, 2.8148, 2.6841, 3.3452], + device='cuda:3'), covar=tensor([0.0740, 0.0516, 0.0271, 0.0962, 0.0112, 0.0144, 0.0122, 0.0097], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0153, 0.0109, 0.0174, 0.0102, 0.0096, 0.0095, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:31:59,088 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.519e+01 2.019e+02 2.564e+02 3.147e+02 5.259e+02, threshold=5.129e+02, percent-clipped=0.0 +2022-11-15 16:32:12,151 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:32:22,624 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:32:25,236 INFO [train.py:876] (3/4) Epoch 3, batch 2200, loss[loss=0.2525, simple_loss=0.2334, pruned_loss=0.1358, over 5617.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2086, pruned_loss=0.1191, over 1087225.42 frames. ], batch size: 43, lr: 2.49e-02, grad_scale: 8.0 +2022-11-15 16:32:44,090 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.18 vs. limit=2.0 +2022-11-15 16:32:56,103 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0736, 4.4176, 4.3457, 4.6040, 4.2477, 3.4213, 4.9708, 4.3481], + device='cuda:3'), covar=tensor([0.0488, 0.0823, 0.0427, 0.0576, 0.0467, 0.0403, 0.0686, 0.0329], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0076, 0.0065, 0.0072, 0.0053, 0.0044, 0.0086, 0.0057], + device='cuda:3'), out_proj_covar=tensor([1.2061e-04, 1.7409e-04, 1.4914e-04, 1.5963e-04, 1.1981e-04, 9.8860e-05, + 2.1378e-04, 1.2762e-04], device='cuda:3') +2022-11-15 16:32:56,717 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:33:01,075 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.20 vs. limit=2.0 +2022-11-15 16:33:07,908 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:33:10,455 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.121e+02 2.774e+02 3.414e+02 1.004e+03, threshold=5.548e+02, percent-clipped=7.0 +2022-11-15 16:33:36,645 INFO [train.py:876] (3/4) Epoch 3, batch 2300, loss[loss=0.2748, simple_loss=0.2222, pruned_loss=0.1637, over 3086.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.208, pruned_loss=0.1189, over 1086132.02 frames. ], batch size: 284, lr: 2.49e-02, grad_scale: 8.0 +2022-11-15 16:33:42,410 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:33:47,550 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7294, 2.8548, 2.3887, 2.4308, 1.4594, 2.7031, 1.8980, 1.2189], + device='cuda:3'), covar=tensor([0.0103, 0.0020, 0.0046, 0.0052, 0.0112, 0.0026, 0.0075, 0.0040], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0067, 0.0086, 0.0083, 0.0122, 0.0081, 0.0105, 0.0071], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:33:53,217 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:34:22,209 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:34:24,019 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.301e+02 2.850e+02 3.623e+02 1.087e+03, threshold=5.699e+02, percent-clipped=3.0 +2022-11-15 16:34:25,418 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0278, 4.2305, 4.3614, 4.3581, 4.1852, 3.4803, 4.8907, 4.2638], + device='cuda:3'), covar=tensor([0.0572, 0.0906, 0.0379, 0.0752, 0.0401, 0.0374, 0.0795, 0.0368], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0078, 0.0067, 0.0075, 0.0054, 0.0046, 0.0088, 0.0059], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:34:38,470 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 16:34:39,459 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1411, 4.1102, 4.1638, 3.8530, 4.1099, 3.9321, 1.4252, 3.9415], + device='cuda:3'), covar=tensor([0.0236, 0.0201, 0.0204, 0.0194, 0.0248, 0.0187, 0.3210, 0.0317], + device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0060, 0.0062, 0.0051, 0.0074, 0.0054, 0.0119, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 16:34:40,229 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7682, 2.1084, 1.4573, 1.9639, 1.2426, 1.0225, 1.4352, 1.8138], + device='cuda:3'), covar=tensor([0.0136, 0.0254, 0.0547, 0.0328, 0.0670, 0.0644, 0.0376, 0.0416], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0042, 0.0045, 0.0033, 0.0050, 0.0041, 0.0048, 0.0032], + device='cuda:3'), out_proj_covar=tensor([7.2571e-05, 8.4753e-05, 1.0824e-04, 7.1801e-05, 1.0730e-04, 9.3077e-05, + 9.7575e-05, 7.0043e-05], device='cuda:3') +2022-11-15 16:34:44,009 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 16:34:50,018 INFO [train.py:876] (3/4) Epoch 3, batch 2400, loss[loss=0.2222, simple_loss=0.2053, pruned_loss=0.1196, over 5557.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.209, pruned_loss=0.1199, over 1084692.96 frames. ], batch size: 30, lr: 2.48e-02, grad_scale: 8.0 +2022-11-15 16:35:05,734 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:35:30,870 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 16:35:35,037 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2022-11-15 16:35:36,095 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.211e+02 2.830e+02 3.499e+02 8.414e+02, threshold=5.661e+02, percent-clipped=2.0 +2022-11-15 16:35:48,483 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:35:57,526 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2483, 3.2242, 3.5720, 1.4627, 3.5326, 4.1422, 3.0364, 4.0043], + device='cuda:3'), covar=tensor([0.0661, 0.0386, 0.0131, 0.0843, 0.0137, 0.0053, 0.0110, 0.0082], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0153, 0.0108, 0.0166, 0.0104, 0.0093, 0.0096, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:36:01,856 INFO [train.py:876] (3/4) Epoch 3, batch 2500, loss[loss=0.2184, simple_loss=0.2102, pruned_loss=0.1133, over 5712.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2102, pruned_loss=0.1208, over 1082842.09 frames. ], batch size: 28, lr: 2.47e-02, grad_scale: 8.0 +2022-11-15 16:36:07,312 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 +2022-11-15 16:36:22,798 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:36:47,308 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 2.290e+02 2.908e+02 3.668e+02 6.942e+02, threshold=5.815e+02, percent-clipped=2.0 +2022-11-15 16:36:50,206 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8907, 4.0798, 3.8065, 3.8259, 3.9525, 3.9254, 1.4822, 3.9848], + device='cuda:3'), covar=tensor([0.0197, 0.0197, 0.0215, 0.0303, 0.0212, 0.0178, 0.2338, 0.0196], + device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0061, 0.0061, 0.0051, 0.0073, 0.0054, 0.0115, 0.0079], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 16:36:52,343 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2144, 1.2130, 1.1541, 1.6520, 1.0334, 1.1249, 0.6897, 1.4529], + device='cuda:3'), covar=tensor([0.0133, 0.0280, 0.0167, 0.0087, 0.0306, 0.0379, 0.0326, 0.0105], + device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0042, 0.0046, 0.0033, 0.0050, 0.0039, 0.0047, 0.0033], + device='cuda:3'), out_proj_covar=tensor([7.0969e-05, 8.5480e-05, 1.0887e-04, 7.1923e-05, 1.0841e-04, 9.1397e-05, + 9.8053e-05, 7.2046e-05], device='cuda:3') +2022-11-15 16:36:54,838 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.13 vs. limit=5.0 +2022-11-15 16:37:00,527 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3370, 1.8108, 1.2054, 1.6875, 1.7102, 1.3042, 1.8167, 1.6938], + device='cuda:3'), covar=tensor([0.0027, 0.0032, 0.0105, 0.0031, 0.0021, 0.0048, 0.0026, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0014, 0.0013, 0.0016, 0.0014, 0.0016, 0.0016, 0.0014], + device='cuda:3'), out_proj_covar=tensor([2.0292e-05, 1.9967e-05, 1.9423e-05, 2.0158e-05, 1.7801e-05, 1.9026e-05, + 2.0888e-05, 1.8996e-05], device='cuda:3') +2022-11-15 16:37:00,727 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.28 vs. limit=5.0 +2022-11-15 16:37:12,393 INFO [train.py:876] (3/4) Epoch 3, batch 2600, loss[loss=0.2883, simple_loss=0.2541, pruned_loss=0.1613, over 5660.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2089, pruned_loss=0.1196, over 1077051.45 frames. ], batch size: 38, lr: 2.47e-02, grad_scale: 8.0 +2022-11-15 16:37:43,051 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0558, 1.5112, 1.6660, 1.0549, 0.2957, 1.6136, 1.2619, 1.0929], + device='cuda:3'), covar=tensor([0.0174, 0.0074, 0.0091, 0.0171, 0.0513, 0.0078, 0.0239, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0026, 0.0027, 0.0025, 0.0022, 0.0023, 0.0026], + device='cuda:3'), out_proj_covar=tensor([4.3743e-05, 3.7810e-05, 3.8289e-05, 4.3463e-05, 4.5056e-05, 3.9629e-05, + 3.8812e-05, 4.3024e-05], device='cuda:3') +2022-11-15 16:37:43,080 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:37:47,043 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7578, 3.8718, 3.7637, 4.0644, 3.5692, 3.3729, 4.4847, 3.6593], + device='cuda:3'), covar=tensor([0.0410, 0.0778, 0.0408, 0.0610, 0.0468, 0.0347, 0.0513, 0.0382], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0076, 0.0064, 0.0072, 0.0053, 0.0045, 0.0084, 0.0057], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:37:53,724 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0578, 3.1218, 2.7121, 2.6915, 2.0149, 3.0501, 2.1529, 1.9375], + device='cuda:3'), covar=tensor([0.0094, 0.0019, 0.0044, 0.0042, 0.0099, 0.0020, 0.0065, 0.0029], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0073, 0.0093, 0.0091, 0.0130, 0.0086, 0.0109, 0.0074], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:37:57,338 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.163e+02 2.684e+02 3.226e+02 5.923e+02, threshold=5.368e+02, percent-clipped=1.0 +2022-11-15 16:38:08,606 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:38:18,793 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3005, 4.8890, 3.8730, 4.8822, 3.8500, 3.4906, 2.5912, 4.4930], + device='cuda:3'), covar=tensor([0.1624, 0.0081, 0.0461, 0.0108, 0.0304, 0.0622, 0.1542, 0.0126], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0110, 0.0158, 0.0106, 0.0133, 0.0168, 0.0185, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 16:38:22,230 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8579, 2.3382, 1.5805, 2.2276, 1.3417, 1.2789, 1.6964, 2.4683], + device='cuda:3'), covar=tensor([0.0160, 0.0239, 0.0630, 0.0611, 0.0595, 0.0589, 0.0342, 0.1542], + device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0040, 0.0043, 0.0032, 0.0049, 0.0039, 0.0046, 0.0034], + device='cuda:3'), out_proj_covar=tensor([6.8991e-05, 8.3775e-05, 1.0488e-04, 7.1383e-05, 1.0867e-04, 9.1145e-05, + 9.7117e-05, 7.3152e-05], device='cuda:3') +2022-11-15 16:38:22,718 INFO [train.py:876] (3/4) Epoch 3, batch 2700, loss[loss=0.2066, simple_loss=0.1976, pruned_loss=0.1078, over 5567.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2093, pruned_loss=0.1201, over 1084167.83 frames. ], batch size: 16, lr: 2.46e-02, grad_scale: 8.0 +2022-11-15 16:38:26,432 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:38:34,859 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:39:07,892 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.430e+02 3.020e+02 3.987e+02 6.630e+02, threshold=6.040e+02, percent-clipped=4.0 +2022-11-15 16:39:33,471 INFO [train.py:876] (3/4) Epoch 3, batch 2800, loss[loss=0.2001, simple_loss=0.188, pruned_loss=0.1061, over 5719.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2092, pruned_loss=0.1202, over 1079048.95 frames. ], batch size: 17, lr: 2.45e-02, grad_scale: 8.0 +2022-11-15 16:39:45,081 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:40:00,286 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:40:18,908 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.246e+02 2.631e+02 3.087e+02 4.773e+02, threshold=5.262e+02, percent-clipped=0.0 +2022-11-15 16:40:26,387 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 16:40:28,885 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:40:44,642 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 16:40:45,103 INFO [train.py:876] (3/4) Epoch 3, batch 2900, loss[loss=0.2042, simple_loss=0.1948, pruned_loss=0.1068, over 5748.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.2094, pruned_loss=0.1209, over 1075618.96 frames. ], batch size: 15, lr: 2.45e-02, grad_scale: 8.0 +2022-11-15 16:41:00,279 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2450, 1.0730, 1.1398, 1.3507, 1.6637, 1.0304, 1.0932, 1.5695], + device='cuda:3'), covar=tensor([0.0543, 0.0318, 0.0436, 0.0649, 0.0258, 0.1646, 0.0510, 0.0709], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0011, 0.0009, 0.0009, 0.0009, 0.0010, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([2.5525e-05, 2.8909e-05, 2.6743e-05, 2.8529e-05, 2.5082e-05, 2.6889e-05, + 2.7890e-05, 2.4809e-05], device='cuda:3') +2022-11-15 16:41:08,687 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0021, 2.4369, 1.6357, 2.7950, 1.4269, 1.5756, 1.9594, 2.6389], + device='cuda:3'), covar=tensor([0.0192, 0.0308, 0.0602, 0.0200, 0.0674, 0.0335, 0.0413, 0.0810], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0043, 0.0046, 0.0034, 0.0051, 0.0042, 0.0049, 0.0034], + device='cuda:3'), out_proj_covar=tensor([7.2883e-05, 8.9321e-05, 1.0982e-04, 7.5366e-05, 1.1224e-04, 9.7333e-05, + 1.0323e-04, 7.5894e-05], device='cuda:3') +2022-11-15 16:41:26,030 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5890, 2.4855, 2.8983, 3.9819, 4.3863, 3.1164, 2.6388, 4.2948], + device='cuda:3'), covar=tensor([0.0070, 0.1797, 0.1256, 0.0849, 0.0099, 0.1395, 0.1230, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0222, 0.0234, 0.0182, 0.0161, 0.0241, 0.0210, 0.0118], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0004, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:41:30,294 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.189e+02 2.747e+02 3.512e+02 7.345e+02, threshold=5.495e+02, percent-clipped=3.0 +2022-11-15 16:41:40,751 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:41:54,811 INFO [train.py:876] (3/4) Epoch 3, batch 3000, loss[loss=0.3044, simple_loss=0.2428, pruned_loss=0.183, over 2945.00 frames. ], tot_loss[loss=0.227, simple_loss=0.2103, pruned_loss=0.1219, over 1080896.81 frames. ], batch size: 285, lr: 2.44e-02, grad_scale: 8.0 +2022-11-15 16:41:54,811 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 16:42:12,231 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4204, 2.0220, 2.0596, 2.4210, 2.5185, 2.5702, 2.8217, 2.6397], + device='cuda:3'), covar=tensor([0.0475, 0.0915, 0.0498, 0.0684, 0.0454, 0.0306, 0.0680, 0.0408], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0077, 0.0065, 0.0074, 0.0057, 0.0045, 0.0085, 0.0058], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:42:13,676 INFO [train.py:908] (3/4) Epoch 3, validation: loss=0.1847, simple_loss=0.2015, pruned_loss=0.08391, over 1530663.00 frames. +2022-11-15 16:42:13,677 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4728MB +2022-11-15 16:42:13,754 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:42:16,621 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6063, 2.1070, 2.8729, 3.8332, 3.9633, 2.7783, 2.5153, 4.1070], + device='cuda:3'), covar=tensor([0.0056, 0.1771, 0.1098, 0.0703, 0.0124, 0.1144, 0.1081, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0219, 0.0231, 0.0180, 0.0160, 0.0237, 0.0207, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:42:25,432 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:42:28,215 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0068, 1.8826, 2.2065, 2.7406, 2.8709, 2.0106, 1.5962, 3.0930], + device='cuda:3'), covar=tensor([0.0082, 0.1225, 0.1046, 0.0361, 0.0162, 0.1054, 0.0996, 0.0069], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0219, 0.0231, 0.0181, 0.0160, 0.0239, 0.0208, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 16:42:32,288 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:42:59,054 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 2.309e+02 3.014e+02 4.120e+02 1.212e+03, threshold=6.029e+02, percent-clipped=7.0 +2022-11-15 16:43:00,217 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:43:24,411 INFO [train.py:876] (3/4) Epoch 3, batch 3100, loss[loss=0.2457, simple_loss=0.2354, pruned_loss=0.128, over 5624.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.2106, pruned_loss=0.1211, over 1084636.34 frames. ], batch size: 24, lr: 2.43e-02, grad_scale: 8.0 +2022-11-15 16:43:53,413 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6662, 2.3348, 2.1385, 1.5294, 2.6021, 2.5600, 2.7126, 2.8207], + device='cuda:3'), covar=tensor([0.0866, 0.0673, 0.0442, 0.1041, 0.0150, 0.0136, 0.0123, 0.0132], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0154, 0.0109, 0.0167, 0.0107, 0.0096, 0.0096, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:44:04,527 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:44:10,835 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 2.237e+02 2.895e+02 3.474e+02 7.762e+02, threshold=5.791e+02, percent-clipped=3.0 +2022-11-15 16:44:16,895 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:44:25,301 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:44:31,329 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:44:35,379 INFO [train.py:876] (3/4) Epoch 3, batch 3200, loss[loss=0.1867, simple_loss=0.1867, pruned_loss=0.09339, over 5681.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2109, pruned_loss=0.1219, over 1080476.27 frames. ], batch size: 12, lr: 2.43e-02, grad_scale: 8.0 +2022-11-15 16:44:47,648 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:45:05,836 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7539, 3.1798, 2.8456, 3.1515, 3.2066, 2.7862, 2.9516, 2.6579], + device='cuda:3'), covar=tensor([0.1297, 0.0437, 0.0784, 0.0365, 0.0439, 0.0461, 0.0384, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0097, 0.0138, 0.0091, 0.0120, 0.0108, 0.0099, 0.0091], + 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') +2022-11-15 16:45:08,555 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:45:12,658 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5666, 3.2725, 2.3000, 1.8235, 3.1120, 1.2727, 3.2484, 1.6347], + device='cuda:3'), covar=tensor([0.0723, 0.0143, 0.0519, 0.1648, 0.0185, 0.1808, 0.0131, 0.1674], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0085, 0.0083, 0.0116, 0.0087, 0.0128, 0.0073, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:45:21,431 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 2.197e+02 2.889e+02 3.896e+02 6.814e+02, threshold=5.779e+02, percent-clipped=2.0 +2022-11-15 16:45:47,449 INFO [train.py:876] (3/4) Epoch 3, batch 3300, loss[loss=0.2062, simple_loss=0.212, pruned_loss=0.1002, over 5572.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.209, pruned_loss=0.1198, over 1081680.04 frames. ], batch size: 18, lr: 2.42e-02, grad_scale: 8.0 +2022-11-15 16:45:47,593 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:45:59,901 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8212, 2.6034, 2.7305, 2.8002, 2.7036, 2.5588, 3.0371, 2.7761], + device='cuda:3'), covar=tensor([0.0413, 0.0829, 0.0422, 0.0624, 0.0568, 0.0336, 0.0714, 0.0537], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0075, 0.0063, 0.0075, 0.0056, 0.0044, 0.0086, 0.0057], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:46:08,542 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5775, 2.3234, 2.1759, 1.0921, 2.1691, 2.6347, 2.3767, 2.9897], + device='cuda:3'), covar=tensor([0.0813, 0.0415, 0.0283, 0.0968, 0.0147, 0.0127, 0.0096, 0.0112], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0157, 0.0110, 0.0168, 0.0107, 0.0097, 0.0095, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:46:21,797 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:46:32,462 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 2.187e+02 2.686e+02 3.512e+02 8.499e+02, threshold=5.372e+02, percent-clipped=2.0 +2022-11-15 16:46:53,095 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4457, 3.4614, 2.5845, 1.6898, 3.2643, 1.3312, 3.4054, 1.7927], + device='cuda:3'), covar=tensor([0.0871, 0.0181, 0.0538, 0.1753, 0.0200, 0.1862, 0.0137, 0.1678], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0086, 0.0083, 0.0114, 0.0086, 0.0127, 0.0074, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:46:58,450 INFO [train.py:876] (3/4) Epoch 3, batch 3400, loss[loss=0.157, simple_loss=0.1587, pruned_loss=0.07769, over 5702.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2091, pruned_loss=0.1203, over 1082755.77 frames. ], batch size: 12, lr: 2.41e-02, grad_scale: 16.0 +2022-11-15 16:47:08,861 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:47:31,081 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2658, 1.8543, 1.5890, 2.0892, 0.9642, 1.4728, 1.2491, 1.7541], + device='cuda:3'), covar=tensor([0.0373, 0.0358, 0.0747, 0.0363, 0.0686, 0.0333, 0.0500, 0.0531], + device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0043, 0.0048, 0.0033, 0.0051, 0.0042, 0.0050, 0.0034], + device='cuda:3'), out_proj_covar=tensor([8.1690e-05, 9.3974e-05, 1.1303e-04, 7.5373e-05, 1.1286e-04, 9.7888e-05, + 1.0627e-04, 7.6998e-05], device='cuda:3') +2022-11-15 16:47:43,699 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.311e+02 2.080e+02 2.540e+02 3.324e+02 6.629e+02, threshold=5.080e+02, percent-clipped=5.0 +2022-11-15 16:47:46,163 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.39 vs. limit=5.0 +2022-11-15 16:47:48,310 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 16:47:50,080 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:47:52,131 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:48:04,399 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:48:05,486 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6660, 2.3288, 1.7456, 1.1362, 1.5020, 2.5681, 2.1126, 2.4438], + device='cuda:3'), covar=tensor([0.0700, 0.0412, 0.0422, 0.0848, 0.0231, 0.0124, 0.0109, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0159, 0.0116, 0.0173, 0.0110, 0.0099, 0.0099, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:48:09,011 INFO [train.py:876] (3/4) Epoch 3, batch 3500, loss[loss=0.1856, simple_loss=0.1958, pruned_loss=0.08771, over 5513.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2074, pruned_loss=0.1174, over 1082322.36 frames. ], batch size: 17, lr: 2.41e-02, grad_scale: 16.0 +2022-11-15 16:48:17,491 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:48:24,391 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:48:26,579 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5765, 0.8528, 1.0191, 1.0405, 0.9160, 1.3560, 0.9249, 1.2308], + device='cuda:3'), covar=tensor([0.0043, 0.0030, 0.0027, 0.0024, 0.0034, 0.0025, 0.0023, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0018, 0.0017, 0.0019, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([2.4674e-05, 2.6233e-05, 2.2661e-05, 1.8689e-05, 2.1039e-05, 1.8978e-05, + 2.7850e-05, 2.2444e-05], device='cuda:3') +2022-11-15 16:48:38,162 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:48:38,774 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 16:48:44,424 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2022-11-15 16:48:55,271 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 2.252e+02 2.672e+02 3.554e+02 6.890e+02, threshold=5.344e+02, percent-clipped=4.0 +2022-11-15 16:49:00,347 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.69 vs. limit=5.0 +2022-11-15 16:49:04,506 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3158, 3.3596, 3.1712, 3.3030, 2.2204, 3.5283, 2.4226, 3.2235], + device='cuda:3'), covar=tensor([0.0187, 0.0069, 0.0093, 0.0070, 0.0197, 0.0037, 0.0124, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0073, 0.0094, 0.0091, 0.0130, 0.0090, 0.0111, 0.0074], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 16:49:04,757 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2022-11-15 16:49:20,260 INFO [train.py:876] (3/4) Epoch 3, batch 3600, loss[loss=0.1413, simple_loss=0.1512, pruned_loss=0.06576, over 5482.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2061, pruned_loss=0.116, over 1088907.84 frames. ], batch size: 10, lr: 2.40e-02, grad_scale: 16.0 +2022-11-15 16:49:32,579 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6410, 2.5043, 2.0349, 1.3270, 2.6464, 3.1854, 2.3526, 3.3112], + device='cuda:3'), covar=tensor([0.1027, 0.0565, 0.0512, 0.1051, 0.0129, 0.0092, 0.0109, 0.0078], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0160, 0.0117, 0.0174, 0.0109, 0.0099, 0.0099, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:49:41,606 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:49:42,299 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:50:06,555 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.223e+02 2.647e+02 3.378e+02 5.745e+02, threshold=5.295e+02, percent-clipped=2.0 +2022-11-15 16:50:25,768 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:50:26,458 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:50:27,154 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3940, 0.8844, 0.7692, 0.6683, 0.8167, 1.3341, 0.9964, 1.1522], + device='cuda:3'), covar=tensor([0.0638, 0.0495, 0.1430, 0.2640, 0.2000, 0.0525, 0.1564, 0.3068], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0012, 0.0011, 0.0010, 0.0011, 0.0011, 0.0012, 0.0010], + device='cuda:3'), out_proj_covar=tensor([2.8653e-05, 3.1143e-05, 3.1436e-05, 3.1600e-05, 3.0937e-05, 3.0491e-05, + 3.2908e-05, 2.9523e-05], device='cuda:3') +2022-11-15 16:50:31,509 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2022-11-15 16:50:32,399 INFO [train.py:876] (3/4) Epoch 3, batch 3700, loss[loss=0.2263, simple_loss=0.2159, pruned_loss=0.1184, over 5589.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2073, pruned_loss=0.1182, over 1084475.96 frames. ], batch size: 24, lr: 2.40e-02, grad_scale: 16.0 +2022-11-15 16:50:52,304 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0311, 3.5921, 3.1361, 3.6094, 3.5643, 3.1401, 3.2890, 2.8911], + device='cuda:3'), covar=tensor([0.0775, 0.0294, 0.0799, 0.0278, 0.0357, 0.0399, 0.0331, 0.0515], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0101, 0.0148, 0.0096, 0.0127, 0.0113, 0.0103, 0.0096], + 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') +2022-11-15 16:51:04,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 +2022-11-15 16:51:17,618 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.280e+02 2.074e+02 2.681e+02 3.314e+02 6.700e+02, threshold=5.362e+02, percent-clipped=3.0 +2022-11-15 16:51:17,777 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8511, 4.6458, 4.5727, 3.4543, 4.5223, 4.6987, 1.4374, 4.6214], + device='cuda:3'), covar=tensor([0.0382, 0.0368, 0.0447, 0.0406, 0.0370, 0.0320, 0.3355, 0.0411], + device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0060, 0.0063, 0.0051, 0.0077, 0.0060, 0.0115, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 16:51:22,594 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:51:43,707 INFO [train.py:876] (3/4) Epoch 3, batch 3800, loss[loss=0.2374, simple_loss=0.2284, pruned_loss=0.1232, over 5570.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.207, pruned_loss=0.1179, over 1083362.52 frames. ], batch size: 30, lr: 2.39e-02, grad_scale: 16.0 +2022-11-15 16:51:51,403 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:52:12,764 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:52:25,089 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:52:28,473 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 2.160e+02 2.623e+02 3.183e+02 6.237e+02, threshold=5.245e+02, percent-clipped=1.0 +2022-11-15 16:52:46,227 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:52:54,049 INFO [train.py:876] (3/4) Epoch 3, batch 3900, loss[loss=0.2094, simple_loss=0.1879, pruned_loss=0.1154, over 4990.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2074, pruned_loss=0.1176, over 1085453.12 frames. ], batch size: 109, lr: 2.38e-02, grad_scale: 16.0 +2022-11-15 16:53:10,346 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 16:53:11,496 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7079, 2.7616, 2.5151, 1.2264, 2.7144, 3.0531, 2.8237, 3.2637], + device='cuda:3'), covar=tensor([0.0956, 0.0515, 0.0374, 0.1075, 0.0124, 0.0107, 0.0114, 0.0090], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0156, 0.0115, 0.0175, 0.0108, 0.0103, 0.0100, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2022-11-15 16:53:16,957 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7594, 2.0306, 1.7642, 1.3949, 0.4669, 2.1631, 1.4732, 1.4453], + device='cuda:3'), covar=tensor([0.0135, 0.0114, 0.0190, 0.0241, 0.0551, 0.0274, 0.0215, 0.0215], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0027, 0.0029, 0.0029, 0.0028, 0.0023, 0.0024, 0.0027], + device='cuda:3'), out_proj_covar=tensor([4.4788e-05, 4.0603e-05, 4.3516e-05, 4.7106e-05, 4.8878e-05, 4.2461e-05, + 4.0163e-05, 4.4719e-05], device='cuda:3') +2022-11-15 16:53:26,661 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:53:39,714 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 16:53:40,176 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.213e+02 2.841e+02 3.897e+02 8.408e+02, threshold=5.683e+02, percent-clipped=7.0 +2022-11-15 16:53:54,785 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:53:55,476 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:54:05,589 INFO [train.py:876] (3/4) Epoch 3, batch 4000, loss[loss=0.2153, simple_loss=0.2026, pruned_loss=0.114, over 5693.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2077, pruned_loss=0.1177, over 1085155.90 frames. ], batch size: 19, lr: 2.38e-02, grad_scale: 16.0 +2022-11-15 16:54:10,069 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:54:22,441 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:54:42,593 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 16:54:49,456 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2022-11-15 16:54:51,285 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.156e+02 2.771e+02 3.318e+02 9.148e+02, threshold=5.542e+02, percent-clipped=3.0 +2022-11-15 16:54:56,234 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:55:08,781 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.18 vs. limit=5.0 +2022-11-15 16:55:11,069 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6612, 3.9781, 4.2599, 4.0781, 4.7436, 4.6348, 3.8634, 4.4398], + device='cuda:3'), covar=tensor([0.0704, 0.0715, 0.0978, 0.0696, 0.0706, 0.0268, 0.0634, 0.0819], + device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0084, 0.0071, 0.0091, 0.0086, 0.0055, 0.0073, 0.0078], + 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') +2022-11-15 16:55:16,129 INFO [train.py:876] (3/4) Epoch 3, batch 4100, loss[loss=0.2494, simple_loss=0.2171, pruned_loss=0.1409, over 4716.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2067, pruned_loss=0.118, over 1076090.47 frames. ], batch size: 135, lr: 2.37e-02, grad_scale: 16.0 +2022-11-15 16:55:29,466 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.70 vs. limit=5.0 +2022-11-15 16:55:29,478 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 16:55:29,860 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:55:36,235 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2022-11-15 16:55:37,271 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:56:01,442 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 2.068e+02 2.779e+02 3.549e+02 8.529e+02, threshold=5.559e+02, percent-clipped=5.0 +2022-11-15 16:56:20,125 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:56:25,022 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6197, 0.4364, 0.4162, 0.6367, 0.6085, 0.7333, 0.6354, 0.5196], + device='cuda:3'), covar=tensor([0.0202, 0.0237, 0.0195, 0.0421, 0.0325, 0.0305, 0.0320, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0013, 0.0010, 0.0011, 0.0011, 0.0010, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([2.9419e-05, 3.2532e-05, 3.0929e-05, 3.3115e-05, 3.0422e-05, 2.9909e-05, + 3.3301e-05, 2.9654e-05], device='cuda:3') +2022-11-15 16:56:26,895 INFO [train.py:876] (3/4) Epoch 3, batch 4200, loss[loss=0.2687, simple_loss=0.2242, pruned_loss=0.1566, over 3012.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2051, pruned_loss=0.1169, over 1073034.43 frames. ], batch size: 285, lr: 2.37e-02, grad_scale: 16.0 +2022-11-15 16:56:27,003 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2606, 2.9800, 2.3243, 1.5775, 2.8453, 1.0091, 2.8776, 1.7504], + device='cuda:3'), covar=tensor([0.0745, 0.0134, 0.0455, 0.1269, 0.0193, 0.1595, 0.0153, 0.1216], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0084, 0.0086, 0.0113, 0.0088, 0.0126, 0.0074, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 16:57:12,962 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 2.085e+02 2.508e+02 3.197e+02 8.828e+02, threshold=5.016e+02, percent-clipped=2.0 +2022-11-15 16:57:28,693 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:57:29,367 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:57:38,761 INFO [train.py:876] (3/4) Epoch 3, batch 4300, loss[loss=0.2601, simple_loss=0.2291, pruned_loss=0.1455, over 5015.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2061, pruned_loss=0.1168, over 1080607.19 frames. ], batch size: 110, lr: 2.36e-02, grad_scale: 16.0 +2022-11-15 16:57:39,539 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:57:52,367 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 16:58:02,524 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:58:03,539 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:58:11,037 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2022-11-15 16:58:24,292 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.437e+02 2.955e+02 3.758e+02 8.175e+02, threshold=5.909e+02, percent-clipped=8.0 +2022-11-15 16:58:24,930 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 16:58:31,132 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 16:58:40,186 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:58:42,215 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1915, 1.4230, 1.5230, 1.3354, 0.2869, 1.6836, 1.4686, 1.2426], + device='cuda:3'), covar=tensor([0.0205, 0.0199, 0.0182, 0.0367, 0.0949, 0.1565, 0.0252, 0.0283], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0028, 0.0029, 0.0031, 0.0027, 0.0024, 0.0023, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.5225e-05, 4.2276e-05, 4.3927e-05, 5.0319e-05, 4.9037e-05, 4.4210e-05, + 3.7999e-05, 4.5803e-05], device='cuda:3') +2022-11-15 16:58:43,866 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 +2022-11-15 16:58:50,709 INFO [train.py:876] (3/4) Epoch 3, batch 4400, loss[loss=0.2135, simple_loss=0.2026, pruned_loss=0.1123, over 5571.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2056, pruned_loss=0.1163, over 1076008.46 frames. ], batch size: 40, lr: 2.35e-02, grad_scale: 16.0 +2022-11-15 16:59:14,564 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-15 16:59:23,544 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 16:59:33,780 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4850, 1.5466, 1.7481, 1.7153, 0.5540, 2.1771, 1.6549, 1.6210], + device='cuda:3'), covar=tensor([0.0165, 0.0222, 0.0308, 0.0488, 0.0849, 0.0460, 0.0180, 0.0567], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0028, 0.0028, 0.0030, 0.0026, 0.0023, 0.0023, 0.0028], + device='cuda:3'), out_proj_covar=tensor([4.4163e-05, 4.1163e-05, 4.2706e-05, 4.7912e-05, 4.7295e-05, 4.2692e-05, + 3.7707e-05, 4.5712e-05], device='cuda:3') +2022-11-15 16:59:35,491 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 16:59:36,220 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.090e+02 2.682e+02 3.252e+02 5.703e+02, threshold=5.365e+02, percent-clipped=0.0 +2022-11-15 16:59:51,550 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:00:02,099 INFO [train.py:876] (3/4) Epoch 3, batch 4500, loss[loss=0.1861, simple_loss=0.1835, pruned_loss=0.0944, over 5420.00 frames. ], tot_loss[loss=0.218, simple_loss=0.205, pruned_loss=0.1154, over 1084164.25 frames. ], batch size: 11, lr: 2.35e-02, grad_scale: 16.0 +2022-11-15 17:00:07,126 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2022-11-15 17:00:26,578 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 +2022-11-15 17:00:48,080 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 1.944e+02 2.562e+02 3.151e+02 4.952e+02, threshold=5.125e+02, percent-clipped=0.0 +2022-11-15 17:01:02,219 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:01:13,365 INFO [train.py:876] (3/4) Epoch 3, batch 4600, loss[loss=0.2289, simple_loss=0.215, pruned_loss=0.1214, over 5610.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2045, pruned_loss=0.1148, over 1082023.93 frames. ], batch size: 32, lr: 2.34e-02, grad_scale: 16.0 +2022-11-15 17:01:14,572 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:01:27,316 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 17:01:46,037 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:01:48,573 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:01:59,681 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 2.110e+02 2.501e+02 3.073e+02 9.500e+02, threshold=5.001e+02, percent-clipped=3.0 +2022-11-15 17:02:01,811 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:02:12,237 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2022-11-15 17:02:24,583 INFO [train.py:876] (3/4) Epoch 3, batch 4700, loss[loss=0.2412, simple_loss=0.2163, pruned_loss=0.1331, over 5279.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2049, pruned_loss=0.1149, over 1080438.40 frames. ], batch size: 79, lr: 2.34e-02, grad_scale: 16.0 +2022-11-15 17:02:53,393 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2022-11-15 17:02:53,863 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:03:10,434 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 2.139e+02 2.654e+02 3.598e+02 8.917e+02, threshold=5.308e+02, percent-clipped=4.0 +2022-11-15 17:03:25,174 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:03:33,970 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2022-11-15 17:03:35,748 INFO [train.py:876] (3/4) Epoch 3, batch 4800, loss[loss=0.2925, simple_loss=0.2503, pruned_loss=0.1674, over 5454.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2035, pruned_loss=0.1133, over 1078756.54 frames. ], batch size: 58, lr: 2.33e-02, grad_scale: 8.0 +2022-11-15 17:03:43,921 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.20 vs. limit=2.0 +2022-11-15 17:03:45,294 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.19 vs. limit=2.0 +2022-11-15 17:03:59,327 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:04:10,770 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:04:21,914 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 2.238e+02 2.787e+02 3.697e+02 9.222e+02, threshold=5.575e+02, percent-clipped=6.0 +2022-11-15 17:04:47,609 INFO [train.py:876] (3/4) Epoch 3, batch 4900, loss[loss=0.2737, simple_loss=0.2347, pruned_loss=0.1564, over 5140.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2037, pruned_loss=0.1136, over 1080223.04 frames. ], batch size: 91, lr: 2.32e-02, grad_scale: 8.0 +2022-11-15 17:04:54,595 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:05:03,718 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0177, 2.8403, 2.7904, 3.1594, 2.8623, 2.4303, 3.3522, 2.8976], + device='cuda:3'), covar=tensor([0.0384, 0.0790, 0.0512, 0.0552, 0.0523, 0.0472, 0.0651, 0.0429], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0079, 0.0065, 0.0078, 0.0058, 0.0048, 0.0092, 0.0060], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 17:05:16,058 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:05:33,093 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.112e+02 2.712e+02 3.360e+02 6.764e+02, threshold=5.425e+02, percent-clipped=2.0 +2022-11-15 17:05:50,438 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6887, 2.1758, 1.6254, 2.1406, 1.6786, 1.8222, 1.5675, 2.7777], + device='cuda:3'), covar=tensor([0.0403, 0.0598, 0.1046, 0.0642, 0.0837, 0.0803, 0.0698, 0.0581], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0044, 0.0050, 0.0035, 0.0052, 0.0043, 0.0051, 0.0034], + device='cuda:3'), out_proj_covar=tensor([8.7026e-05, 9.9313e-05, 1.2269e-04, 8.0146e-05, 1.1909e-04, 1.0449e-04, + 1.1191e-04, 7.9052e-05], device='cuda:3') +2022-11-15 17:05:50,458 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3392, 0.6814, 1.0680, 0.6327, 0.9239, 0.9161, 1.0819, 0.9660], + device='cuda:3'), covar=tensor([0.1487, 0.0630, 0.0826, 0.3641, 0.1723, 0.2967, 0.0965, 0.2391], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0010, 0.0010, 0.0009, 0.0010, 0.0010], + device='cuda:3'), out_proj_covar=tensor([2.8932e-05, 3.0929e-05, 2.8480e-05, 3.0441e-05, 2.8962e-05, 2.7403e-05, + 3.1321e-05, 2.8573e-05], device='cuda:3') +2022-11-15 17:05:50,676 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 17:05:52,085 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 17:05:58,041 INFO [train.py:876] (3/4) Epoch 3, batch 5000, loss[loss=0.2491, simple_loss=0.2263, pruned_loss=0.136, over 5749.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2051, pruned_loss=0.1149, over 1088812.61 frames. ], batch size: 31, lr: 2.32e-02, grad_scale: 8.0 +2022-11-15 17:06:08,152 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8588, 2.1291, 3.6464, 2.9684, 3.9613, 2.6469, 3.4014, 4.0688], + device='cuda:3'), covar=tensor([0.0063, 0.0435, 0.0093, 0.0343, 0.0056, 0.0279, 0.0197, 0.0077], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0172, 0.0130, 0.0181, 0.0122, 0.0161, 0.0184, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 17:06:26,454 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:06:44,100 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.122e+02 2.600e+02 3.385e+02 5.559e+02, threshold=5.201e+02, percent-clipped=2.0 +2022-11-15 17:06:50,199 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:07:01,377 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:07:09,727 INFO [train.py:876] (3/4) Epoch 3, batch 5100, loss[loss=0.3153, simple_loss=0.2561, pruned_loss=0.1872, over 3068.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2053, pruned_loss=0.1148, over 1085610.93 frames. ], batch size: 284, lr: 2.31e-02, grad_scale: 8.0 +2022-11-15 17:07:17,443 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 17:07:34,074 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:07:56,097 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.277e+02 2.877e+02 3.641e+02 9.302e+02, threshold=5.755e+02, percent-clipped=5.0 +2022-11-15 17:08:20,624 INFO [train.py:876] (3/4) Epoch 3, batch 5200, loss[loss=0.1656, simple_loss=0.1819, pruned_loss=0.07463, over 5562.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2057, pruned_loss=0.1139, over 1091929.77 frames. ], batch size: 14, lr: 2.31e-02, grad_scale: 8.0 +2022-11-15 17:08:24,898 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:08:31,775 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2363, 1.5723, 1.1328, 1.0601, 1.2810, 1.9866, 1.7048, 1.9644], + device='cuda:3'), covar=tensor([0.0824, 0.0389, 0.0714, 0.0899, 0.0285, 0.0141, 0.0167, 0.0224], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0171, 0.0123, 0.0185, 0.0118, 0.0110, 0.0105, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:08:44,241 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 +2022-11-15 17:08:47,433 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6830, 1.8226, 3.3140, 2.5985, 3.5698, 2.1799, 3.0186, 3.5434], + device='cuda:3'), covar=tensor([0.0073, 0.0636, 0.0109, 0.0451, 0.0107, 0.0401, 0.0298, 0.0143], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0178, 0.0131, 0.0181, 0.0123, 0.0161, 0.0188, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 17:08:49,760 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:09:07,140 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.129e+02 2.534e+02 3.438e+02 1.087e+03, threshold=5.069e+02, percent-clipped=3.0 +2022-11-15 17:09:07,597 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 17:09:23,653 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:09:32,161 INFO [train.py:876] (3/4) Epoch 3, batch 5300, loss[loss=0.2662, simple_loss=0.2397, pruned_loss=0.1463, over 5308.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2042, pruned_loss=0.1136, over 1079448.15 frames. ], batch size: 79, lr: 2.30e-02, grad_scale: 8.0 +2022-11-15 17:10:18,064 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.053e+02 2.554e+02 3.563e+02 7.883e+02, threshold=5.108e+02, percent-clipped=6.0 +2022-11-15 17:10:27,228 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5536, 4.0452, 4.2396, 4.2233, 4.7493, 4.7220, 3.9684, 4.6228], + device='cuda:3'), covar=tensor([0.1151, 0.0991, 0.1300, 0.1094, 0.1047, 0.0373, 0.0829, 0.1076], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0087, 0.0071, 0.0094, 0.0092, 0.0057, 0.0078, 0.0078], + 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') +2022-11-15 17:10:39,982 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2886, 1.4686, 1.8515, 2.1767, 0.5068, 1.7310, 1.6218, 1.4721], + device='cuda:3'), covar=tensor([0.0214, 0.0274, 0.0159, 0.0179, 0.0626, 0.0403, 0.0296, 0.0262], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0029, 0.0028, 0.0032, 0.0028, 0.0027, 0.0027, 0.0030], + device='cuda:3'), out_proj_covar=tensor([4.9174e-05, 4.4244e-05, 4.3155e-05, 5.2852e-05, 5.0977e-05, 4.8987e-05, + 4.3995e-05, 4.8845e-05], device='cuda:3') +2022-11-15 17:10:43,252 INFO [train.py:876] (3/4) Epoch 3, batch 5400, loss[loss=0.1116, simple_loss=0.13, pruned_loss=0.04656, over 4677.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.205, pruned_loss=0.1146, over 1080002.25 frames. ], batch size: 5, lr: 2.30e-02, grad_scale: 8.0 +2022-11-15 17:11:04,390 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:11:33,651 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 2.054e+02 2.433e+02 3.312e+02 7.375e+02, threshold=4.866e+02, percent-clipped=2.0 +2022-11-15 17:11:40,711 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:11:54,862 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1512, 0.9862, 1.6305, 1.2787, 1.0444, 1.1965, 1.3133, 1.4758], + device='cuda:3'), covar=tensor([0.0035, 0.0255, 0.0049, 0.0032, 0.0031, 0.0063, 0.0051, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0014, 0.0014, 0.0013, 0.0016, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.9284e-05, 1.6449e-05, 1.6373e-05, 1.7779e-05, 1.5119e-05, 1.6336e-05, + 1.8684e-05, 1.7437e-05], device='cuda:3') +2022-11-15 17:11:57,506 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4383, 3.4270, 3.4140, 3.2051, 3.5654, 3.4818, 1.3675, 3.4532], + device='cuda:3'), covar=tensor([0.0238, 0.0184, 0.0214, 0.0235, 0.0202, 0.0193, 0.2274, 0.0239], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0065, 0.0067, 0.0056, 0.0081, 0.0064, 0.0120, 0.0089], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:11:58,779 INFO [train.py:876] (3/4) Epoch 3, batch 5500, loss[loss=0.1919, simple_loss=0.1879, pruned_loss=0.09801, over 5704.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2046, pruned_loss=0.1138, over 1084164.62 frames. ], batch size: 11, lr: 2.29e-02, grad_scale: 8.0 +2022-11-15 17:12:02,257 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:12:24,741 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:12:30,162 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:12:36,667 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:12:45,203 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 2.322e+02 3.147e+02 3.867e+02 1.026e+03, threshold=6.293e+02, percent-clipped=11.0 +2022-11-15 17:12:46,855 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:12:47,009 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.14 vs. limit=2.0 +2022-11-15 17:12:57,588 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.66 vs. limit=5.0 +2022-11-15 17:13:02,227 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2022-11-15 17:13:05,779 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0853, 3.6411, 2.7154, 3.6411, 2.6065, 2.6095, 1.9888, 3.1406], + device='cuda:3'), covar=tensor([0.1094, 0.0115, 0.0617, 0.0127, 0.0580, 0.0679, 0.1448, 0.0161], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0109, 0.0151, 0.0105, 0.0136, 0.0169, 0.0177, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 17:13:07,441 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 17:13:10,507 INFO [train.py:876] (3/4) Epoch 3, batch 5600, loss[loss=0.1101, simple_loss=0.1307, pruned_loss=0.04475, over 5262.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2047, pruned_loss=0.1154, over 1076258.88 frames. ], batch size: 7, lr: 2.29e-02, grad_scale: 8.0 +2022-11-15 17:13:14,144 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:13:15,892 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 +2022-11-15 17:13:28,447 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7057, 3.8103, 3.6008, 3.4347, 3.7189, 3.3575, 1.3792, 3.4789], + device='cuda:3'), covar=tensor([0.0374, 0.0351, 0.0293, 0.0211, 0.0412, 0.0420, 0.2816, 0.0339], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0064, 0.0065, 0.0055, 0.0079, 0.0063, 0.0118, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:13:30,584 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:13:44,844 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 17:13:52,976 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:13:56,931 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.126e+02 2.547e+02 3.418e+02 6.941e+02, threshold=5.093e+02, percent-clipped=1.0 +2022-11-15 17:14:22,609 INFO [train.py:876] (3/4) Epoch 3, batch 5700, loss[loss=0.2492, simple_loss=0.2262, pruned_loss=0.1361, over 5612.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.205, pruned_loss=0.1151, over 1081428.29 frames. ], batch size: 24, lr: 2.28e-02, grad_scale: 8.0 +2022-11-15 17:14:24,831 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9666, 2.2000, 2.6347, 1.4849, 1.6035, 1.7799, 1.5619, 1.7143], + device='cuda:3'), covar=tensor([0.0017, 0.0058, 0.0015, 0.0020, 0.0012, 0.0030, 0.0019, 0.0018], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0014, 0.0013, 0.0013, 0.0015, 0.0013], + device='cuda:3'), out_proj_covar=tensor([1.7474e-05, 1.5753e-05, 1.4848e-05, 1.6901e-05, 1.4203e-05, 1.5410e-05, + 1.7132e-05, 1.6082e-05], device='cuda:3') +2022-11-15 17:14:36,319 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:14:42,878 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:14:52,740 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0305, 1.4242, 1.2810, 1.4594, 0.2792, 1.7335, 1.4134, 0.9886], + device='cuda:3'), covar=tensor([0.0146, 0.0131, 0.0130, 0.0190, 0.0770, 0.0166, 0.0210, 0.0426], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0028, 0.0027, 0.0032, 0.0028, 0.0025, 0.0027, 0.0031], + device='cuda:3'), out_proj_covar=tensor([4.9294e-05, 4.2966e-05, 4.2349e-05, 5.3141e-05, 5.1013e-05, 4.6134e-05, + 4.3138e-05, 5.0789e-05], device='cuda:3') +2022-11-15 17:15:08,577 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 2.183e+02 2.755e+02 3.255e+02 9.254e+02, threshold=5.510e+02, percent-clipped=4.0 +2022-11-15 17:15:16,850 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:15:21,788 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.46 vs. limit=2.0 +2022-11-15 17:15:33,291 INFO [train.py:876] (3/4) Epoch 3, batch 5800, loss[loss=0.174, simple_loss=0.1842, pruned_loss=0.08192, over 5701.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2044, pruned_loss=0.1145, over 1079593.80 frames. ], batch size: 17, lr: 2.28e-02, grad_scale: 8.0 +2022-11-15 17:15:44,131 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 17:15:55,503 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:16:04,053 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:16:19,794 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.270e+02 2.884e+02 3.519e+02 5.566e+02, threshold=5.768e+02, percent-clipped=1.0 +2022-11-15 17:16:25,593 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:16:44,673 INFO [train.py:876] (3/4) Epoch 3, batch 5900, loss[loss=0.1683, simple_loss=0.1844, pruned_loss=0.07614, over 5571.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.204, pruned_loss=0.1136, over 1078623.90 frames. ], batch size: 18, lr: 2.27e-02, grad_scale: 8.0 +2022-11-15 17:16:44,764 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:16:47,627 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:17:01,268 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:17:09,334 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:17:30,802 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 2.074e+02 2.650e+02 3.512e+02 6.887e+02, threshold=5.300e+02, percent-clipped=1.0 +2022-11-15 17:17:33,507 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3619, 3.8546, 3.6181, 3.3151, 2.3275, 4.2062, 2.2360, 3.3040], + device='cuda:3'), covar=tensor([0.0242, 0.0072, 0.0073, 0.0172, 0.0280, 0.0036, 0.0190, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0084, 0.0101, 0.0105, 0.0141, 0.0096, 0.0121, 0.0082], + 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') +2022-11-15 17:17:38,568 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3797, 3.0040, 3.2881, 1.3111, 3.1404, 3.7313, 3.5793, 3.7085], + device='cuda:3'), covar=tensor([0.1195, 0.0812, 0.0476, 0.1705, 0.0126, 0.0104, 0.0150, 0.0125], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0176, 0.0126, 0.0193, 0.0118, 0.0111, 0.0110, 0.0128], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:17:42,018 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2578, 1.5974, 1.3470, 1.9634, 1.3432, 1.3382, 1.6516, 1.9953], + device='cuda:3'), covar=tensor([0.0506, 0.0744, 0.1225, 0.0279, 0.0883, 0.1242, 0.0540, 0.0618], + device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0042, 0.0050, 0.0035, 0.0054, 0.0044, 0.0050, 0.0038], + device='cuda:3'), out_proj_covar=tensor([8.6201e-05, 9.7804e-05, 1.2651e-04, 8.3210e-05, 1.2675e-04, 1.0782e-04, + 1.1482e-04, 8.9817e-05], device='cuda:3') +2022-11-15 17:17:55,881 INFO [train.py:876] (3/4) Epoch 3, batch 6000, loss[loss=0.1907, simple_loss=0.1876, pruned_loss=0.09687, over 5604.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2019, pruned_loss=0.1117, over 1086754.66 frames. ], batch size: 24, lr: 2.27e-02, grad_scale: 8.0 +2022-11-15 17:17:55,881 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 17:18:06,802 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7392, 1.8219, 2.7080, 3.5494, 3.9449, 2.4364, 2.4656, 3.7519], + device='cuda:3'), covar=tensor([0.0149, 0.3638, 0.1927, 0.1551, 0.0271, 0.2553, 0.2131, 0.0140], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0221, 0.0223, 0.0215, 0.0184, 0.0231, 0.0209, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0002], + device='cuda:3') +2022-11-15 17:18:06,965 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9882, 2.8452, 3.0767, 1.2872, 3.1909, 3.4269, 3.4894, 3.5854], + device='cuda:3'), covar=tensor([0.1308, 0.0766, 0.0377, 0.1758, 0.0140, 0.0115, 0.0130, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0175, 0.0125, 0.0191, 0.0118, 0.0110, 0.0109, 0.0126], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:18:14,730 INFO [train.py:908] (3/4) Epoch 3, validation: loss=0.1788, simple_loss=0.1971, pruned_loss=0.08032, over 1530663.00 frames. +2022-11-15 17:18:14,732 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4728MB +2022-11-15 17:18:25,276 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:18:45,549 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:18:45,610 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1671, 3.6486, 3.2521, 3.2547, 2.2206, 3.4803, 2.2414, 2.8615], + device='cuda:3'), covar=tensor([0.0248, 0.0117, 0.0103, 0.0121, 0.0219, 0.0055, 0.0196, 0.0045], + device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0083, 0.0101, 0.0104, 0.0137, 0.0095, 0.0118, 0.0080], + 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') +2022-11-15 17:19:00,872 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.977e+02 2.420e+02 3.254e+02 5.998e+02, threshold=4.840e+02, percent-clipped=5.0 +2022-11-15 17:19:17,100 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 +2022-11-15 17:19:25,871 INFO [train.py:876] (3/4) Epoch 3, batch 6100, loss[loss=0.1898, simple_loss=0.1771, pruned_loss=0.1013, over 5749.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2035, pruned_loss=0.1129, over 1090032.75 frames. ], batch size: 14, lr: 2.26e-02, grad_scale: 8.0 +2022-11-15 17:19:28,885 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:19:41,380 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 +2022-11-15 17:19:47,914 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:19:47,947 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0892, 3.1556, 1.5355, 2.8610, 1.6788, 2.6770, 1.8476, 2.7248], + device='cuda:3'), covar=tensor([0.0272, 0.0207, 0.0967, 0.0491, 0.0661, 0.0386, 0.0553, 0.1249], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0046, 0.0056, 0.0039, 0.0059, 0.0048, 0.0056, 0.0042], + device='cuda:3'), out_proj_covar=tensor([9.1528e-05, 1.0582e-04, 1.4148e-04, 9.2552e-05, 1.3830e-04, 1.1642e-04, + 1.2869e-04, 9.8455e-05], device='cuda:3') +2022-11-15 17:20:00,187 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2022-11-15 17:20:10,296 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 17:20:11,191 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 2.120e+02 2.658e+02 3.170e+02 7.084e+02, threshold=5.316e+02, percent-clipped=6.0 +2022-11-15 17:20:15,380 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3698, 4.4919, 3.6559, 4.4102, 3.4931, 2.9384, 2.3677, 4.0470], + device='cuda:3'), covar=tensor([0.1331, 0.0123, 0.0467, 0.0156, 0.0328, 0.0825, 0.1705, 0.0118], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0115, 0.0158, 0.0110, 0.0139, 0.0174, 0.0183, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 17:20:20,757 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:20:36,131 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:20:36,748 INFO [train.py:876] (3/4) Epoch 3, batch 6200, loss[loss=0.2174, simple_loss=0.1949, pruned_loss=0.1199, over 4119.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2036, pruned_loss=0.1132, over 1085306.13 frames. ], batch size: 181, lr: 2.26e-02, grad_scale: 8.0 +2022-11-15 17:20:36,868 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:20:45,263 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9990, 1.2442, 1.0907, 1.0830, 1.5255, 1.6014, 0.9376, 1.3126], + device='cuda:3'), covar=tensor([0.0024, 0.0021, 0.0019, 0.0015, 0.0014, 0.0012, 0.0051, 0.0023], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0015, 0.0017, 0.0016, 0.0017, 0.0014, 0.0018, 0.0015], + device='cuda:3'), out_proj_covar=tensor([2.1454e-05, 1.9498e-05, 1.9367e-05, 1.6915e-05, 1.8227e-05, 1.4771e-05, + 2.6349e-05, 1.8666e-05], device='cuda:3') +2022-11-15 17:20:52,828 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:20:57,648 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:21:11,105 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:21:22,963 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.113e+02 2.498e+02 3.399e+02 6.391e+02, threshold=4.996e+02, percent-clipped=4.0 +2022-11-15 17:21:27,275 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:21:47,813 INFO [train.py:876] (3/4) Epoch 3, batch 6300, loss[loss=0.1939, simple_loss=0.1997, pruned_loss=0.09404, over 5518.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2029, pruned_loss=0.1132, over 1086073.04 frames. ], batch size: 14, lr: 2.25e-02, grad_scale: 8.0 +2022-11-15 17:21:58,673 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:22:00,820 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7544, 1.5614, 1.0532, 1.3861, 1.0073, 0.8468, 0.9580, 1.3063], + device='cuda:3'), covar=tensor([0.0634, 0.0538, 0.0469, 0.0327, 0.0922, 0.1310, 0.0778, 0.0258], + device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0047, 0.0054, 0.0040, 0.0058, 0.0047, 0.0056, 0.0041], + device='cuda:3'), out_proj_covar=tensor([9.3243e-05, 1.0820e-04, 1.3804e-04, 9.4083e-05, 1.3843e-04, 1.1492e-04, + 1.2920e-04, 9.7445e-05], device='cuda:3') +2022-11-15 17:22:32,807 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:22:34,074 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.128e+02 2.723e+02 3.616e+02 7.802e+02, threshold=5.445e+02, percent-clipped=12.0 +2022-11-15 17:22:37,365 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 +2022-11-15 17:22:58,281 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:22:58,896 INFO [train.py:876] (3/4) Epoch 3, batch 6400, loss[loss=0.2059, simple_loss=0.2041, pruned_loss=0.1038, over 5754.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2032, pruned_loss=0.1129, over 1084438.99 frames. ], batch size: 20, lr: 2.25e-02, grad_scale: 8.0 +2022-11-15 17:23:34,188 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3642, 1.5442, 1.8037, 1.3465, 1.6916, 1.6462, 2.5209, 1.9487], + device='cuda:3'), covar=tensor([0.0015, 0.0104, 0.0059, 0.0015, 0.0015, 0.0049, 0.0011, 0.0014], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0011, 0.0013, 0.0012, 0.0012, 0.0014, 0.0012], + device='cuda:3'), out_proj_covar=tensor([1.5601e-05, 1.5122e-05, 1.5039e-05, 1.5647e-05, 1.3375e-05, 1.4297e-05, + 1.6960e-05, 1.6024e-05], device='cuda:3') +2022-11-15 17:23:46,175 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.963e+02 2.483e+02 3.263e+02 5.829e+02, threshold=4.966e+02, percent-clipped=2.0 +2022-11-15 17:23:59,621 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2022-11-15 17:24:10,688 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:24:11,260 INFO [train.py:876] (3/4) Epoch 3, batch 6500, loss[loss=0.2671, simple_loss=0.2421, pruned_loss=0.1461, over 5576.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2033, pruned_loss=0.112, over 1086754.15 frames. ], batch size: 30, lr: 2.24e-02, grad_scale: 8.0 +2022-11-15 17:24:17,480 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:24:18,259 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1079, 1.4354, 1.0997, 1.0280, 1.0897, 1.8769, 1.7530, 1.6452], + device='cuda:3'), covar=tensor([0.0576, 0.0215, 0.0441, 0.0634, 0.0227, 0.0082, 0.0131, 0.0306], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0171, 0.0125, 0.0180, 0.0117, 0.0107, 0.0109, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:24:25,461 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.13 vs. limit=2.0 +2022-11-15 17:24:25,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 +2022-11-15 17:24:32,925 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:24:45,244 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:24:45,370 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0615, 1.3635, 1.2993, 1.3763, 1.0685, 1.8501, 1.0649, 1.4735], + device='cuda:3'), covar=tensor([0.0014, 0.0011, 0.0015, 0.0013, 0.0013, 0.0006, 0.0026, 0.0015], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0017, 0.0016, 0.0017, 0.0014, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.1360e-05, 2.0339e-05, 1.9782e-05, 1.6681e-05, 1.8163e-05, 1.4522e-05, + 2.6718e-05, 1.8463e-05], device='cuda:3') +2022-11-15 17:24:58,585 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.262e+02 2.845e+02 3.817e+02 7.458e+02, threshold=5.690e+02, percent-clipped=11.0 +2022-11-15 17:25:00,949 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:25:04,407 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:25:06,962 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:25:07,067 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0538, 1.7891, 1.0566, 1.6188, 0.9852, 1.1872, 1.0064, 1.8759], + device='cuda:3'), covar=tensor([0.0527, 0.0728, 0.1207, 0.0888, 0.1220, 0.0779, 0.0916, 0.0575], + device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0048, 0.0056, 0.0042, 0.0058, 0.0046, 0.0053, 0.0040], + device='cuda:3'), out_proj_covar=tensor([9.5898e-05, 1.1080e-04, 1.4039e-04, 9.8807e-05, 1.3835e-04, 1.1410e-04, + 1.2552e-04, 9.7295e-05], device='cuda:3') +2022-11-15 17:25:14,885 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1306, 1.2167, 1.0445, 1.1167, 1.1582, 2.0364, 1.0974, 1.2667], + device='cuda:3'), covar=tensor([0.0021, 0.0019, 0.0016, 0.0018, 0.0016, 0.0010, 0.0023, 0.0027], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0017, 0.0016, 0.0017, 0.0014, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.1315e-05, 2.0213e-05, 1.9783e-05, 1.7109e-05, 1.8460e-05, 1.4519e-05, + 2.6752e-05, 1.8585e-05], device='cuda:3') +2022-11-15 17:25:22,797 INFO [train.py:876] (3/4) Epoch 3, batch 6600, loss[loss=0.1787, simple_loss=0.1799, pruned_loss=0.08879, over 5773.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2057, pruned_loss=0.1144, over 1089365.28 frames. ], batch size: 14, lr: 2.23e-02, grad_scale: 8.0 +2022-11-15 17:25:47,392 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:26:08,334 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 17:26:09,147 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.333e+02 2.074e+02 2.594e+02 3.441e+02 9.119e+02, threshold=5.187e+02, percent-clipped=3.0 +2022-11-15 17:26:10,077 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8978, 2.7949, 2.4204, 1.4449, 2.5924, 2.7619, 2.6142, 2.9107], + device='cuda:3'), covar=tensor([0.1000, 0.0543, 0.0529, 0.1191, 0.0155, 0.0200, 0.0152, 0.0135], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0174, 0.0129, 0.0186, 0.0121, 0.0112, 0.0112, 0.0128], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:26:33,847 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:26:34,418 INFO [train.py:876] (3/4) Epoch 3, batch 6700, loss[loss=0.2205, simple_loss=0.197, pruned_loss=0.122, over 5008.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2051, pruned_loss=0.1146, over 1082529.48 frames. ], batch size: 109, lr: 2.23e-02, grad_scale: 8.0 +2022-11-15 17:26:44,892 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 17:26:48,452 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2022-11-15 17:27:07,741 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:27:16,046 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 17:27:20,388 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.110e+02 2.601e+02 3.417e+02 8.582e+02, threshold=5.201e+02, percent-clipped=1.0 +2022-11-15 17:27:22,022 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:27:33,703 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:27:45,487 INFO [train.py:876] (3/4) Epoch 3, batch 6800, loss[loss=0.1948, simple_loss=0.1885, pruned_loss=0.1005, over 5566.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2036, pruned_loss=0.1136, over 1077712.55 frames. ], batch size: 21, lr: 2.22e-02, grad_scale: 16.0 +2022-11-15 17:27:50,086 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2417, 1.3198, 1.0252, 1.0251, 1.4063, 1.6951, 0.8902, 1.0623], + device='cuda:3'), covar=tensor([0.0019, 0.0017, 0.0020, 0.0019, 0.0020, 0.0009, 0.0039, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0017, 0.0018, 0.0018, 0.0018, 0.0015, 0.0019, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.1633e-05, 2.2153e-05, 2.0221e-05, 1.8576e-05, 1.9723e-05, 1.4829e-05, + 2.8889e-05, 1.9218e-05], device='cuda:3') +2022-11-15 17:28:05,228 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:28:17,373 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:28:30,804 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:28:31,949 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.12 vs. limit=5.0 +2022-11-15 17:28:32,053 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 2.037e+02 2.447e+02 3.250e+02 5.677e+02, threshold=4.895e+02, percent-clipped=2.0 +2022-11-15 17:28:57,299 INFO [train.py:876] (3/4) Epoch 3, batch 6900, loss[loss=0.2504, simple_loss=0.2332, pruned_loss=0.1337, over 5597.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2055, pruned_loss=0.1156, over 1071763.89 frames. ], batch size: 18, lr: 2.22e-02, grad_scale: 16.0 +2022-11-15 17:29:02,483 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.28 vs. limit=2.0 +2022-11-15 17:29:17,881 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:29:43,557 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.216e+02 2.709e+02 3.216e+02 6.064e+02, threshold=5.419e+02, percent-clipped=2.0 +2022-11-15 17:29:58,525 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 17:30:08,649 INFO [train.py:876] (3/4) Epoch 3, batch 7000, loss[loss=0.1654, simple_loss=0.1773, pruned_loss=0.0767, over 5722.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2041, pruned_loss=0.1131, over 1076223.36 frames. ], batch size: 11, lr: 2.22e-02, grad_scale: 16.0 +2022-11-15 17:30:43,777 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:30:53,832 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0914, 2.2008, 1.5817, 1.8360, 1.3531, 1.6831, 2.0559, 2.2030], + device='cuda:3'), covar=tensor([0.0765, 0.0998, 0.1452, 0.2272, 0.1483, 0.1211, 0.0911, 0.1163], + device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0046, 0.0057, 0.0043, 0.0058, 0.0048, 0.0055, 0.0041], + 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') +2022-11-15 17:30:55,041 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.332e+02 2.225e+02 2.783e+02 3.485e+02 7.501e+02, threshold=5.566e+02, percent-clipped=1.0 +2022-11-15 17:30:57,263 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:31:00,763 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5684, 3.5595, 3.5301, 3.4096, 3.6376, 3.6510, 1.1573, 3.5863], + device='cuda:3'), covar=tensor([0.0327, 0.0350, 0.0218, 0.0229, 0.0334, 0.0206, 0.2957, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0067, 0.0066, 0.0056, 0.0080, 0.0061, 0.0121, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:31:13,384 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0847, 4.2043, 4.0341, 4.1319, 3.7670, 3.3761, 4.6276, 3.8756], + device='cuda:3'), covar=tensor([0.0341, 0.0733, 0.0379, 0.0586, 0.0815, 0.0417, 0.0662, 0.0416], + device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0080, 0.0066, 0.0076, 0.0061, 0.0051, 0.0096, 0.0063], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 17:31:19,547 INFO [train.py:876] (3/4) Epoch 3, batch 7100, loss[loss=0.1784, simple_loss=0.186, pruned_loss=0.08536, over 5272.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.204, pruned_loss=0.1123, over 1078764.11 frames. ], batch size: 9, lr: 2.21e-02, grad_scale: 16.0 +2022-11-15 17:31:27,578 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:31:27,772 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2022-11-15 17:31:36,530 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:31:40,754 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:31:42,094 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4842, 2.5250, 3.5042, 3.7684, 4.3672, 3.2193, 2.3203, 4.4970], + device='cuda:3'), covar=tensor([0.0139, 0.4086, 0.2033, 0.2563, 0.0532, 0.2977, 0.2993, 0.0185], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0229, 0.0232, 0.0232, 0.0188, 0.0238, 0.0206, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 17:31:47,847 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:31:55,766 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.19 vs. limit=2.0 +2022-11-15 17:32:04,373 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:32:05,618 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 2.184e+02 2.618e+02 3.432e+02 5.543e+02, threshold=5.236e+02, percent-clipped=0.0 +2022-11-15 17:32:27,080 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4558, 4.7158, 5.3242, 4.7893, 5.4099, 5.3271, 4.6074, 5.5123], + device='cuda:3'), covar=tensor([0.0240, 0.0186, 0.0292, 0.0234, 0.0293, 0.0077, 0.0162, 0.0143], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0087, 0.0071, 0.0097, 0.0095, 0.0055, 0.0079, 0.0085], + 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') +2022-11-15 17:32:31,171 INFO [train.py:876] (3/4) Epoch 3, batch 7200, loss[loss=0.2106, simple_loss=0.2094, pruned_loss=0.1059, over 5641.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2031, pruned_loss=0.112, over 1083695.84 frames. ], batch size: 32, lr: 2.21e-02, grad_scale: 16.0 +2022-11-15 17:32:36,803 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3869, 1.2809, 1.2768, 1.6451, 0.8157, 1.2552, 1.2430, 1.4911], + device='cuda:3'), covar=tensor([0.0850, 0.1156, 0.1389, 0.0506, 0.1389, 0.1073, 0.0716, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0045, 0.0057, 0.0040, 0.0055, 0.0046, 0.0054, 0.0039], + device='cuda:3'), out_proj_covar=tensor([9.6580e-05, 1.0823e-04, 1.4397e-04, 9.8069e-05, 1.3372e-04, 1.1587e-04, + 1.2733e-04, 9.8661e-05], device='cuda:3') +2022-11-15 17:32:38,706 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:32:43,684 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:32:51,899 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:33:06,726 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2543, 2.4320, 3.2556, 3.7671, 4.1793, 3.4867, 2.6583, 4.5613], + device='cuda:3'), covar=tensor([0.0120, 0.2808, 0.1800, 0.2109, 0.0445, 0.1922, 0.1878, 0.0076], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0220, 0.0223, 0.0230, 0.0184, 0.0229, 0.0204, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0002], + device='cuda:3') +2022-11-15 17:33:16,995 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.105e+02 2.662e+02 3.428e+02 5.862e+02, threshold=5.324e+02, percent-clipped=2.0 +2022-11-15 17:34:13,369 INFO [train.py:876] (3/4) Epoch 4, batch 0, loss[loss=0.2768, simple_loss=0.2369, pruned_loss=0.1583, over 5428.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.2369, pruned_loss=0.1583, over 5428.00 frames. ], batch size: 58, lr: 2.06e-02, grad_scale: 16.0 +2022-11-15 17:34:13,370 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 17:34:23,737 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8734, 2.1442, 1.4643, 2.3637, 1.1842, 2.3096, 1.7793, 2.7462], + device='cuda:3'), covar=tensor([0.0532, 0.0756, 0.2023, 0.0436, 0.1196, 0.0695, 0.0799, 0.0273], + device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0046, 0.0060, 0.0042, 0.0058, 0.0047, 0.0055, 0.0040], + device='cuda:3'), out_proj_covar=tensor([9.9988e-05, 1.1180e-04, 1.4945e-04, 1.0184e-04, 1.3856e-04, 1.1950e-04, + 1.2985e-04, 1.0211e-04], device='cuda:3') +2022-11-15 17:34:30,839 INFO [train.py:908] (3/4) Epoch 4, validation: loss=0.1863, simple_loss=0.204, pruned_loss=0.08431, over 1530663.00 frames. +2022-11-15 17:34:30,839 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4728MB +2022-11-15 17:34:34,342 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:34:35,822 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:34:41,989 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8556, 3.8113, 3.8180, 3.0918, 3.7545, 3.5309, 1.5258, 3.8763], + device='cuda:3'), covar=tensor([0.0237, 0.0211, 0.0224, 0.0382, 0.0312, 0.0312, 0.2328, 0.0218], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0068, 0.0067, 0.0057, 0.0082, 0.0063, 0.0122, 0.0087], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:34:45,956 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.13 vs. limit=5.0 +2022-11-15 17:34:55,157 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:35:00,222 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.24 vs. limit=2.0 +2022-11-15 17:35:10,332 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2988, 3.3877, 3.2167, 2.9827, 3.2635, 3.1586, 1.3064, 3.3197], + device='cuda:3'), covar=tensor([0.0259, 0.0206, 0.0237, 0.0278, 0.0326, 0.0243, 0.2811, 0.0316], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0067, 0.0067, 0.0057, 0.0082, 0.0063, 0.0122, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:35:38,008 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.101e+02 2.586e+02 3.383e+02 7.997e+02, threshold=5.171e+02, percent-clipped=3.0 +2022-11-15 17:35:38,899 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:35:42,904 INFO [train.py:876] (3/4) Epoch 4, batch 100, loss[loss=0.1666, simple_loss=0.1726, pruned_loss=0.08026, over 5716.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.197, pruned_loss=0.107, over 431600.30 frames. ], batch size: 11, lr: 2.05e-02, grad_scale: 16.0 +2022-11-15 17:35:58,398 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 17:36:07,082 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:19,015 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:19,636 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:20,404 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:29,990 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:49,995 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.318e+02 2.900e+02 3.607e+02 8.310e+02, threshold=5.801e+02, percent-clipped=7.0 +2022-11-15 17:36:53,601 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:36:54,234 INFO [train.py:876] (3/4) Epoch 4, batch 200, loss[loss=0.1435, simple_loss=0.1498, pruned_loss=0.06856, over 5455.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.1962, pruned_loss=0.1051, over 688465.95 frames. ], batch size: 11, lr: 2.05e-02, grad_scale: 8.0 +2022-11-15 17:37:04,776 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:37:05,302 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:38:01,665 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.970e+02 2.360e+02 3.051e+02 4.623e+02, threshold=4.719e+02, percent-clipped=0.0 +2022-11-15 17:38:06,229 INFO [train.py:876] (3/4) Epoch 4, batch 300, loss[loss=0.2215, simple_loss=0.2113, pruned_loss=0.1158, over 5587.00 frames. ], tot_loss[loss=0.206, simple_loss=0.1979, pruned_loss=0.107, over 845929.62 frames. ], batch size: 22, lr: 2.05e-02, grad_scale: 8.0 +2022-11-15 17:38:07,666 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:38:11,751 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9573, 3.0067, 2.7770, 1.2806, 2.6586, 3.2542, 2.8523, 3.1623], + device='cuda:3'), covar=tensor([0.0898, 0.0611, 0.0591, 0.1277, 0.0129, 0.0165, 0.0137, 0.0115], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0176, 0.0130, 0.0181, 0.0120, 0.0111, 0.0113, 0.0127], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:38:15,910 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:38:20,954 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 17:38:25,103 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 17:38:48,225 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:38:59,424 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:09,868 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:13,074 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 2.158e+02 2.699e+02 3.433e+02 6.693e+02, threshold=5.398e+02, percent-clipped=7.0 +2022-11-15 17:39:15,895 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:17,098 INFO [train.py:876] (3/4) Epoch 4, batch 400, loss[loss=0.1837, simple_loss=0.1818, pruned_loss=0.09278, over 5758.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2, pruned_loss=0.1081, over 947122.62 frames. ], batch size: 16, lr: 2.04e-02, grad_scale: 8.0 +2022-11-15 17:39:31,518 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:31,539 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:33,309 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2022-11-15 17:39:41,798 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:54,917 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:39:57,060 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2316, 0.7855, 1.5882, 0.8030, 1.4671, 1.0728, 1.2592, 1.3928], + device='cuda:3'), covar=tensor([0.0020, 0.0027, 0.0013, 0.0018, 0.0027, 0.0011, 0.0028, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0018, 0.0019, 0.0019, 0.0019, 0.0017, 0.0021, 0.0018], + device='cuda:3'), out_proj_covar=tensor([2.3361e-05, 2.2866e-05, 2.1299e-05, 1.9727e-05, 2.0417e-05, 1.6016e-05, + 3.0105e-05, 1.9648e-05], device='cuda:3') +2022-11-15 17:39:59,922 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:40:00,672 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8534, 1.4442, 1.6028, 1.3511, 1.8023, 1.4870, 1.0241, 1.9144], + device='cuda:3'), covar=tensor([0.0248, 0.1869, 0.0881, 0.0608, 0.0509, 0.0854, 0.1244, 0.0187], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0230, 0.0232, 0.0237, 0.0193, 0.0238, 0.0208, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 17:40:15,789 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:40:16,319 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:40:25,359 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.107e+02 2.622e+02 3.333e+02 5.947e+02, threshold=5.245e+02, percent-clipped=1.0 +2022-11-15 17:40:29,491 INFO [train.py:876] (3/4) Epoch 4, batch 500, loss[loss=0.1479, simple_loss=0.1568, pruned_loss=0.06952, over 5699.00 frames. ], tot_loss[loss=0.207, simple_loss=0.1991, pruned_loss=0.1074, over 997444.86 frames. ], batch size: 12, lr: 2.04e-02, grad_scale: 8.0 +2022-11-15 17:40:29,545 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:40:35,649 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:41:17,518 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:41:23,971 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2967, 2.1814, 2.0000, 1.3932, 0.6056, 2.6741, 1.8289, 2.3046], + device='cuda:3'), covar=tensor([0.0304, 0.0186, 0.0175, 0.0562, 0.1730, 0.0489, 0.0872, 0.0177], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0029, 0.0029, 0.0033, 0.0032, 0.0025, 0.0028, 0.0030], + device='cuda:3'), out_proj_covar=tensor([5.1052e-05, 4.5052e-05, 4.4618e-05, 5.8554e-05, 5.5585e-05, 4.6265e-05, + 4.6924e-05, 4.8926e-05], device='cuda:3') +2022-11-15 17:41:36,528 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.045e+02 2.785e+02 3.827e+02 6.845e+02, threshold=5.570e+02, percent-clipped=5.0 +2022-11-15 17:41:40,691 INFO [train.py:876] (3/4) Epoch 4, batch 600, loss[loss=0.2106, simple_loss=0.2046, pruned_loss=0.1083, over 5685.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2006, pruned_loss=0.11, over 1021045.53 frames. ], batch size: 36, lr: 2.03e-02, grad_scale: 8.0 +2022-11-15 17:41:42,575 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:42:01,409 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:42:02,733 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6693, 2.4126, 1.8208, 3.1973, 1.9649, 2.7560, 2.2441, 2.7416], + device='cuda:3'), covar=tensor([0.0859, 0.0534, 0.1299, 0.0413, 0.1119, 0.0452, 0.0740, 0.5012], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0047, 0.0056, 0.0042, 0.0056, 0.0043, 0.0053, 0.0038], + device='cuda:3'), out_proj_covar=tensor([9.8786e-05, 1.1250e-04, 1.4391e-04, 1.0079e-04, 1.3637e-04, 1.1189e-04, + 1.2594e-04, 9.7809e-05], device='cuda:3') +2022-11-15 17:42:16,338 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:42:29,713 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:42:36,339 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 17:42:43,784 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:42:47,539 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 1.953e+02 2.259e+02 3.253e+02 4.972e+02, threshold=4.518e+02, percent-clipped=0.0 +2022-11-15 17:42:51,951 INFO [train.py:876] (3/4) Epoch 4, batch 700, loss[loss=0.2343, simple_loss=0.2242, pruned_loss=0.1222, over 5799.00 frames. ], tot_loss[loss=0.209, simple_loss=0.1998, pruned_loss=0.1091, over 1040728.46 frames. ], batch size: 21, lr: 2.03e-02, grad_scale: 8.0 +2022-11-15 17:43:02,567 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:43:18,015 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:43:30,244 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:43:45,945 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:43:52,562 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6987, 4.6317, 3.5994, 4.4875, 3.5216, 3.3866, 2.6125, 4.0849], + device='cuda:3'), covar=tensor([0.1225, 0.0117, 0.0496, 0.0133, 0.0327, 0.0572, 0.1580, 0.0107], + device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0114, 0.0159, 0.0110, 0.0152, 0.0175, 0.0184, 0.0118], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 17:43:55,608 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 +2022-11-15 17:43:58,687 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.071e+02 2.662e+02 3.271e+02 7.195e+02, threshold=5.325e+02, percent-clipped=7.0 +2022-11-15 17:44:03,253 INFO [train.py:876] (3/4) Epoch 4, batch 800, loss[loss=0.2528, simple_loss=0.2114, pruned_loss=0.1471, over 3085.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2001, pruned_loss=0.1085, over 1051945.03 frames. ], batch size: 284, lr: 2.02e-02, grad_scale: 8.0 +2022-11-15 17:44:09,819 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:44:19,321 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:44:43,329 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:45:01,855 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 17:45:09,602 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 2.138e+02 2.671e+02 3.427e+02 5.214e+02, threshold=5.342e+02, percent-clipped=0.0 +2022-11-15 17:45:13,840 INFO [train.py:876] (3/4) Epoch 4, batch 900, loss[loss=0.2829, simple_loss=0.2269, pruned_loss=0.1694, over 4145.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.199, pruned_loss=0.1071, over 1065017.74 frames. ], batch size: 181, lr: 2.02e-02, grad_scale: 8.0 +2022-11-15 17:45:31,214 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:45:50,280 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3320, 3.2066, 3.2263, 3.4637, 3.2896, 2.6181, 3.6210, 3.2846], + device='cuda:3'), covar=tensor([0.0377, 0.0605, 0.0342, 0.0521, 0.0452, 0.0376, 0.0801, 0.0418], + device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0077, 0.0063, 0.0073, 0.0057, 0.0048, 0.0093, 0.0060], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 17:45:50,689 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-15 17:46:03,078 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:08,948 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:20,974 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 2.200e+02 2.748e+02 3.509e+02 2.017e+03, threshold=5.496e+02, percent-clipped=6.0 +2022-11-15 17:46:21,249 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8793, 1.9296, 2.3669, 3.1320, 3.1908, 2.3179, 1.7931, 3.2121], + device='cuda:3'), covar=tensor([0.0255, 0.2361, 0.1557, 0.1479, 0.0411, 0.1704, 0.1761, 0.0101], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0217, 0.0223, 0.0238, 0.0185, 0.0224, 0.0200, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 17:46:25,455 INFO [train.py:876] (3/4) Epoch 4, batch 1000, loss[loss=0.21, simple_loss=0.2075, pruned_loss=0.1062, over 5753.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.1978, pruned_loss=0.1059, over 1076469.81 frames. ], batch size: 14, lr: 2.02e-02, grad_scale: 8.0 +2022-11-15 17:46:31,080 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:35,904 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:37,245 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:37,261 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5234, 4.4170, 4.5023, 4.5105, 4.0548, 3.6369, 5.0166, 4.3336], + device='cuda:3'), covar=tensor([0.0377, 0.0909, 0.0362, 0.0722, 0.0640, 0.0382, 0.0791, 0.0455], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0081, 0.0067, 0.0077, 0.0061, 0.0050, 0.0098, 0.0063], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 17:46:52,930 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:46:57,171 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-15 17:47:04,132 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:47:10,246 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:47:10,287 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5488, 2.6229, 2.2303, 2.5504, 2.6400, 2.2931, 2.3454, 2.2527], + device='cuda:3'), covar=tensor([0.0299, 0.0480, 0.1269, 0.0379, 0.0433, 0.0480, 0.0427, 0.0441], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0175, 0.0108, 0.0139, 0.0121, 0.0120, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:47:14,483 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:47:19,583 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:47:32,217 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 1.971e+02 2.339e+02 2.906e+02 5.624e+02, threshold=4.678e+02, percent-clipped=1.0 +2022-11-15 17:47:36,401 INFO [train.py:876] (3/4) Epoch 4, batch 1100, loss[loss=0.3078, simple_loss=0.2508, pruned_loss=0.1824, over 3119.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.197, pruned_loss=0.1044, over 1079363.77 frames. ], batch size: 284, lr: 2.01e-02, grad_scale: 8.0 +2022-11-15 17:47:37,671 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2022-11-15 17:47:37,873 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:47:53,253 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:48:31,532 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 17:48:37,994 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1911, 4.7124, 4.1200, 4.7534, 4.7675, 4.1477, 4.3636, 4.0673], + device='cuda:3'), covar=tensor([0.0238, 0.0327, 0.0817, 0.0262, 0.0230, 0.0288, 0.0244, 0.0305], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0177, 0.0109, 0.0139, 0.0119, 0.0120, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 17:48:43,817 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.126e+02 2.633e+02 3.216e+02 6.006e+02, threshold=5.267e+02, percent-clipped=5.0 +2022-11-15 17:48:47,973 INFO [train.py:876] (3/4) Epoch 4, batch 1200, loss[loss=0.1484, simple_loss=0.1583, pruned_loss=0.06925, over 5724.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.201, pruned_loss=0.1088, over 1079729.95 frames. ], batch size: 11, lr: 2.01e-02, grad_scale: 8.0 +2022-11-15 17:49:04,983 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:49:38,856 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:49:41,732 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2156, 2.1844, 1.8353, 1.2153, 1.3813, 1.7910, 1.4060, 1.9901], + device='cuda:3'), covar=tensor([0.0230, 0.0136, 0.0196, 0.0448, 0.0549, 0.1625, 0.0438, 0.0211], + device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0031, 0.0032, 0.0035, 0.0032, 0.0027, 0.0031, 0.0033], + device='cuda:3'), out_proj_covar=tensor([5.4888e-05, 4.7285e-05, 4.9463e-05, 6.4428e-05, 5.7894e-05, 4.9986e-05, + 5.2455e-05, 5.3990e-05], device='cuda:3') +2022-11-15 17:49:52,314 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7001, 3.5638, 3.7042, 3.8151, 3.3111, 3.0805, 4.1805, 3.5634], + device='cuda:3'), covar=tensor([0.0481, 0.1000, 0.0489, 0.0772, 0.0920, 0.0405, 0.0843, 0.0464], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0082, 0.0067, 0.0079, 0.0063, 0.0050, 0.0098, 0.0065], + 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') +2022-11-15 17:49:52,416 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:49:54,240 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.201e+02 2.049e+02 2.605e+02 3.403e+02 6.305e+02, threshold=5.209e+02, percent-clipped=2.0 +2022-11-15 17:49:58,727 INFO [train.py:876] (3/4) Epoch 4, batch 1300, loss[loss=0.2644, simple_loss=0.229, pruned_loss=0.1499, over 5441.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.1999, pruned_loss=0.1073, over 1087637.44 frames. ], batch size: 58, lr: 2.00e-02, grad_scale: 8.0 +2022-11-15 17:50:21,972 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:50:36,146 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:50:44,525 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:51:05,600 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.170e+01 2.060e+02 2.584e+02 3.363e+02 5.860e+02, threshold=5.168e+02, percent-clipped=2.0 +2022-11-15 17:51:09,744 INFO [train.py:876] (3/4) Epoch 4, batch 1400, loss[loss=0.1696, simple_loss=0.1696, pruned_loss=0.08486, over 5434.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.1991, pruned_loss=0.107, over 1085817.61 frames. ], batch size: 11, lr: 2.00e-02, grad_scale: 8.0 +2022-11-15 17:51:09,917 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:51:19,222 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:51:53,364 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:52:02,685 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:52:06,038 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:52:17,016 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.039e+02 2.495e+02 2.958e+02 6.410e+02, threshold=4.991e+02, percent-clipped=2.0 +2022-11-15 17:52:21,023 INFO [train.py:876] (3/4) Epoch 4, batch 1500, loss[loss=0.2103, simple_loss=0.2055, pruned_loss=0.1076, over 5627.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.1976, pruned_loss=0.1055, over 1082431.63 frames. ], batch size: 29, lr: 1.99e-02, grad_scale: 8.0 +2022-11-15 17:52:39,666 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:53:27,573 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 2.167e+02 2.612e+02 3.180e+02 6.082e+02, threshold=5.224e+02, percent-clipped=1.0 +2022-11-15 17:53:32,217 INFO [train.py:876] (3/4) Epoch 4, batch 1600, loss[loss=0.248, simple_loss=0.2049, pruned_loss=0.1455, over 5508.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.1968, pruned_loss=0.1054, over 1079492.47 frames. ], batch size: 12, lr: 1.99e-02, grad_scale: 8.0 +2022-11-15 17:53:56,064 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:54:05,586 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:54:17,873 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:54:22,754 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:54:29,480 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:54:39,499 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.935e+02 2.419e+02 3.182e+02 7.599e+02, threshold=4.837e+02, percent-clipped=7.0 +2022-11-15 17:54:43,666 INFO [train.py:876] (3/4) Epoch 4, batch 1700, loss[loss=0.1879, simple_loss=0.1919, pruned_loss=0.09193, over 5566.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.1963, pruned_loss=0.1047, over 1081984.91 frames. ], batch size: 14, lr: 1.99e-02, grad_scale: 8.0 +2022-11-15 17:54:45,201 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1947, 1.8310, 1.3620, 0.6127, 1.7259, 1.7245, 1.3512, 1.2955], + device='cuda:3'), covar=tensor([0.0030, 0.0015, 0.0015, 0.0018, 0.0013, 0.0016, 0.0018, 0.0034], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0018, 0.0017, 0.0017, 0.0018, 0.0016, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([2.2633e-05, 2.2171e-05, 1.8229e-05, 1.7763e-05, 1.8656e-05, 1.5429e-05, + 2.5936e-05, 1.6825e-05], device='cuda:3') +2022-11-15 17:54:51,898 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:55:06,127 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:55:10,438 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 17:55:23,681 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:55:32,252 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:55:42,098 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3344, 1.9840, 2.4738, 3.4561, 3.3564, 2.5482, 2.0358, 3.6019], + device='cuda:3'), covar=tensor([0.0162, 0.2741, 0.2490, 0.1650, 0.0519, 0.2232, 0.2132, 0.0168], + device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0219, 0.0231, 0.0257, 0.0196, 0.0231, 0.0207, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 17:55:51,049 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 2.088e+02 2.577e+02 3.329e+02 6.388e+02, threshold=5.153e+02, percent-clipped=6.0 +2022-11-15 17:55:55,438 INFO [train.py:876] (3/4) Epoch 4, batch 1800, loss[loss=0.2727, simple_loss=0.2215, pruned_loss=0.1619, over 3097.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.197, pruned_loss=0.1054, over 1080929.84 frames. ], batch size: 284, lr: 1.98e-02, grad_scale: 8.0 +2022-11-15 17:56:09,619 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9612, 5.1049, 3.5189, 4.7884, 3.8115, 3.6124, 2.8979, 4.3365], + device='cuda:3'), covar=tensor([0.1085, 0.0054, 0.0640, 0.0192, 0.0288, 0.0524, 0.1269, 0.0098], + device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0115, 0.0163, 0.0112, 0.0150, 0.0176, 0.0191, 0.0120], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 17:56:16,441 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3046, 1.9798, 2.8801, 2.3246, 2.8483, 2.0718, 2.7359, 3.0660], + device='cuda:3'), covar=tensor([0.0099, 0.0461, 0.0153, 0.0485, 0.0139, 0.0364, 0.0223, 0.0179], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0172, 0.0142, 0.0188, 0.0134, 0.0168, 0.0196, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 17:56:17,773 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9580, 1.7995, 2.5595, 2.1312, 2.4825, 1.8727, 2.4235, 2.7819], + device='cuda:3'), covar=tensor([0.0123, 0.0523, 0.0179, 0.0412, 0.0171, 0.0394, 0.0287, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0172, 0.0142, 0.0188, 0.0134, 0.0168, 0.0196, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 17:56:56,800 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0788, 1.2419, 1.4585, 1.0258, 1.6314, 1.8385, 1.0119, 1.5631], + device='cuda:3'), covar=tensor([0.0020, 0.0021, 0.0015, 0.0019, 0.0014, 0.0011, 0.0019, 0.0015], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0018, 0.0019, 0.0017, 0.0020, 0.0017], + device='cuda:3'), out_proj_covar=tensor([2.3824e-05, 2.4310e-05, 1.9096e-05, 1.8887e-05, 1.9275e-05, 1.6250e-05, + 2.8014e-05, 1.7646e-05], device='cuda:3') +2022-11-15 17:57:01,354 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.273e+02 2.086e+02 2.661e+02 3.167e+02 7.420e+02, threshold=5.321e+02, percent-clipped=1.0 +2022-11-15 17:57:05,425 INFO [train.py:876] (3/4) Epoch 4, batch 1900, loss[loss=0.1904, simple_loss=0.1917, pruned_loss=0.09448, over 5604.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.1949, pruned_loss=0.1027, over 1085923.38 frames. ], batch size: 24, lr: 1.98e-02, grad_scale: 8.0 +2022-11-15 17:57:28,325 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2022-11-15 17:57:39,176 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:58:03,088 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 17:58:12,700 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 2.073e+02 2.797e+02 3.748e+02 9.524e+02, threshold=5.593e+02, percent-clipped=9.0 +2022-11-15 17:58:13,476 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:58:16,810 INFO [train.py:876] (3/4) Epoch 4, batch 2000, loss[loss=0.2453, simple_loss=0.2183, pruned_loss=0.1362, over 5534.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.1951, pruned_loss=0.1025, over 1090059.69 frames. ], batch size: 43, lr: 1.97e-02, grad_scale: 8.0 +2022-11-15 17:58:17,145 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2022-11-15 17:58:36,715 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 17:58:56,907 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:59:05,884 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:59:23,777 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 2.154e+02 2.632e+02 3.272e+02 6.334e+02, threshold=5.263e+02, percent-clipped=2.0 +2022-11-15 17:59:27,899 INFO [train.py:876] (3/4) Epoch 4, batch 2100, loss[loss=0.1548, simple_loss=0.1731, pruned_loss=0.06829, over 5708.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.1955, pruned_loss=0.1031, over 1089742.30 frames. ], batch size: 15, lr: 1.97e-02, grad_scale: 8.0 +2022-11-15 17:59:31,171 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 17:59:39,170 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:00:34,708 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.500e+02 2.023e+02 2.446e+02 2.916e+02 5.131e+02, threshold=4.891e+02, percent-clipped=0.0 +2022-11-15 18:00:38,753 INFO [train.py:876] (3/4) Epoch 4, batch 2200, loss[loss=0.2019, simple_loss=0.2041, pruned_loss=0.09983, over 5688.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.1949, pruned_loss=0.1026, over 1091797.36 frames. ], batch size: 36, lr: 1.97e-02, grad_scale: 16.0 +2022-11-15 18:00:53,046 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:01:36,336 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:01:41,988 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 18:01:45,263 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.100e+02 2.713e+02 3.440e+02 7.345e+02, threshold=5.426e+02, percent-clipped=3.0 +2022-11-15 18:01:50,146 INFO [train.py:876] (3/4) Epoch 4, batch 2300, loss[loss=0.1985, simple_loss=0.2052, pruned_loss=0.09588, over 5613.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.1952, pruned_loss=0.1023, over 1092801.57 frames. ], batch size: 18, lr: 1.96e-02, grad_scale: 16.0 +2022-11-15 18:02:03,075 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0077, 1.8421, 1.9322, 1.7509, 2.0425, 1.7836, 1.9614, 1.9645], + device='cuda:3'), covar=tensor([0.0528, 0.0432, 0.0573, 0.0554, 0.0518, 0.0291, 0.0362, 0.0690], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0076, 0.0101, 0.0102, 0.0061, 0.0083, 0.0092], + 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') +2022-11-15 18:02:08,847 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:02:42,465 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:02:56,252 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.961e+02 2.462e+02 3.190e+02 6.497e+02, threshold=4.924e+02, percent-clipped=3.0 +2022-11-15 18:02:59,874 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2396, 1.0450, 1.5148, 1.4736, 1.7290, 1.0664, 1.1686, 1.2362], + device='cuda:3'), covar=tensor([0.0016, 0.0029, 0.0024, 0.0017, 0.0013, 0.0036, 0.0018, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0011, 0.0014, 0.0013, 0.0014, 0.0015, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.5106e-05, 1.6751e-05, 1.3492e-05, 1.6348e-05, 1.3279e-05, 1.6271e-05, + 1.7555e-05, 1.8181e-05], device='cuda:3') +2022-11-15 18:03:00,410 INFO [train.py:876] (3/4) Epoch 4, batch 2400, loss[loss=0.2216, simple_loss=0.2105, pruned_loss=0.1164, over 5568.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.1948, pruned_loss=0.1023, over 1093350.22 frames. ], batch size: 46, lr: 1.96e-02, grad_scale: 16.0 +2022-11-15 18:03:27,767 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 +2022-11-15 18:03:28,820 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2220, 4.9292, 4.4149, 5.0299, 4.9158, 4.2301, 4.4063, 4.1770], + device='cuda:3'), covar=tensor([0.0312, 0.0411, 0.1100, 0.0292, 0.0343, 0.0304, 0.0280, 0.0529], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0120, 0.0188, 0.0119, 0.0150, 0.0129, 0.0127, 0.0112], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:03:29,064 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.13 vs. limit=2.0 +2022-11-15 18:03:38,101 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6698, 2.7501, 2.2436, 2.6486, 1.6292, 2.1862, 1.6838, 2.5112], + device='cuda:3'), covar=tensor([0.1071, 0.0159, 0.0734, 0.0210, 0.0836, 0.0731, 0.1447, 0.0207], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0122, 0.0170, 0.0119, 0.0158, 0.0183, 0.0197, 0.0125], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:04:02,420 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1994, 4.6580, 4.9855, 4.5828, 5.2773, 5.0397, 4.5284, 5.1741], + device='cuda:3'), covar=tensor([0.0281, 0.0210, 0.0337, 0.0236, 0.0258, 0.0105, 0.0218, 0.0224], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0094, 0.0073, 0.0099, 0.0099, 0.0059, 0.0081, 0.0092], + 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') +2022-11-15 18:04:07,100 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 2.091e+02 2.435e+02 3.216e+02 5.165e+02, threshold=4.869e+02, percent-clipped=2.0 +2022-11-15 18:04:11,698 INFO [train.py:876] (3/4) Epoch 4, batch 2500, loss[loss=0.1385, simple_loss=0.1592, pruned_loss=0.05892, over 5557.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.1939, pruned_loss=0.1017, over 1087299.13 frames. ], batch size: 15, lr: 1.96e-02, grad_scale: 16.0 +2022-11-15 18:04:29,677 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2143, 3.1362, 2.8495, 2.8177, 1.8327, 3.0116, 1.9384, 2.7035], + device='cuda:3'), covar=tensor([0.0178, 0.0046, 0.0057, 0.0123, 0.0200, 0.0054, 0.0180, 0.0042], + device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0093, 0.0107, 0.0111, 0.0142, 0.0105, 0.0125, 0.0089], + 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') +2022-11-15 18:05:05,221 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:05:12,522 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.52 vs. limit=5.0 +2022-11-15 18:05:18,633 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.064e+02 2.482e+02 3.355e+02 5.467e+02, threshold=4.964e+02, percent-clipped=1.0 +2022-11-15 18:05:22,808 INFO [train.py:876] (3/4) Epoch 4, batch 2600, loss[loss=0.2234, simple_loss=0.2136, pruned_loss=0.1166, over 5272.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.1954, pruned_loss=0.1024, over 1088433.52 frames. ], batch size: 79, lr: 1.95e-02, grad_scale: 16.0 +2022-11-15 18:05:24,294 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3184, 0.1868, 1.5684, 1.5338, 1.8141, 1.3677, 1.8961, 1.4624], + device='cuda:3'), covar=tensor([0.0016, 0.0042, 0.0041, 0.0017, 0.0016, 0.0024, 0.0014, 0.0027], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0011, 0.0015, 0.0014, 0.0014, 0.0016, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.5241e-05, 1.6108e-05, 1.3506e-05, 1.7270e-05, 1.3731e-05, 1.6639e-05, + 1.7732e-05, 1.8233e-05], device='cuda:3') +2022-11-15 18:05:32,409 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2022-11-15 18:05:47,698 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 18:05:50,539 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0727, 1.0057, 1.5933, 1.6806, 1.2969, 1.4296, 1.4507, 1.2196], + device='cuda:3'), covar=tensor([0.0028, 0.0057, 0.0050, 0.0020, 0.0072, 0.0027, 0.0036, 0.0027], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0011, 0.0015, 0.0014, 0.0014, 0.0015, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.5113e-05, 1.5997e-05, 1.3451e-05, 1.6889e-05, 1.3561e-05, 1.6359e-05, + 1.7556e-05, 1.7941e-05], device='cuda:3') +2022-11-15 18:06:20,397 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-15 18:06:24,628 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5314, 3.6636, 3.5422, 3.2742, 2.5191, 4.0424, 2.2479, 3.4582], + device='cuda:3'), covar=tensor([0.0250, 0.0155, 0.0084, 0.0170, 0.0270, 0.0049, 0.0228, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0096, 0.0113, 0.0117, 0.0146, 0.0108, 0.0128, 0.0092], + 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') +2022-11-15 18:06:29,498 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.874e+02 2.327e+02 3.027e+02 6.734e+02, threshold=4.654e+02, percent-clipped=5.0 +2022-11-15 18:06:33,930 INFO [train.py:876] (3/4) Epoch 4, batch 2700, loss[loss=0.2201, simple_loss=0.2044, pruned_loss=0.1179, over 5806.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.1941, pruned_loss=0.1005, over 1089628.35 frames. ], batch size: 21, lr: 1.95e-02, grad_scale: 16.0 +2022-11-15 18:06:55,318 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:07:11,177 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:07:21,665 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6197, 4.1279, 3.5224, 4.1266, 4.1362, 3.5774, 3.5589, 3.4635], + device='cuda:3'), covar=tensor([0.0655, 0.0550, 0.1601, 0.0393, 0.0555, 0.0385, 0.0555, 0.0889], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0117, 0.0185, 0.0115, 0.0147, 0.0127, 0.0126, 0.0113], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:07:38,833 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:07:40,660 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 2.178e+02 2.525e+02 3.048e+02 6.598e+02, threshold=5.050e+02, percent-clipped=4.0 +2022-11-15 18:07:45,018 INFO [train.py:876] (3/4) Epoch 4, batch 2800, loss[loss=0.2524, simple_loss=0.2242, pruned_loss=0.1403, over 5374.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.1934, pruned_loss=0.1006, over 1089261.83 frames. ], batch size: 70, lr: 1.94e-02, grad_scale: 16.0 +2022-11-15 18:07:46,511 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5634, 1.1048, 1.1111, 0.7607, 0.5169, 1.5062, 0.9645, 0.7954], + device='cuda:3'), covar=tensor([0.0352, 0.0148, 0.0240, 0.0488, 0.0560, 0.0132, 0.0315, 0.0385], + device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0034, 0.0032, 0.0037, 0.0034, 0.0027, 0.0030, 0.0036], + device='cuda:3'), out_proj_covar=tensor([6.1178e-05, 5.2582e-05, 4.9516e-05, 6.9159e-05, 6.0416e-05, 5.1674e-05, + 5.2269e-05, 5.9172e-05], device='cuda:3') +2022-11-15 18:07:54,350 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:08:01,726 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2022-11-15 18:08:38,947 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:08:51,550 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.858e+02 2.475e+02 3.106e+02 5.971e+02, threshold=4.951e+02, percent-clipped=4.0 +2022-11-15 18:08:55,662 INFO [train.py:876] (3/4) Epoch 4, batch 2900, loss[loss=0.2176, simple_loss=0.2085, pruned_loss=0.1133, over 5590.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.1932, pruned_loss=0.1009, over 1086436.80 frames. ], batch size: 25, lr: 1.94e-02, grad_scale: 16.0 +2022-11-15 18:09:00,211 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0897, 3.6049, 3.9150, 3.6553, 4.1373, 3.6218, 3.7951, 4.0493], + device='cuda:3'), covar=tensor([0.0337, 0.0333, 0.0437, 0.0328, 0.0327, 0.0377, 0.0278, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0098, 0.0078, 0.0102, 0.0101, 0.0061, 0.0087, 0.0094], + 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') +2022-11-15 18:09:13,208 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:09:24,285 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.63 vs. limit=5.0 +2022-11-15 18:09:26,621 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-15 18:09:30,496 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:09:36,236 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0885, 1.4796, 1.8714, 1.1773, 0.5636, 1.9351, 1.1593, 1.6476], + device='cuda:3'), covar=tensor([0.0365, 0.0326, 0.0179, 0.0861, 0.1090, 0.0649, 0.0416, 0.0524], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0034, 0.0033, 0.0039, 0.0035, 0.0029, 0.0032, 0.0038], + device='cuda:3'), out_proj_covar=tensor([6.4419e-05, 5.3254e-05, 5.1191e-05, 7.2797e-05, 6.3154e-05, 5.4230e-05, + 5.6670e-05, 6.3311e-05], device='cuda:3') +2022-11-15 18:10:03,413 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.004e+02 2.362e+02 2.844e+02 4.612e+02, threshold=4.725e+02, percent-clipped=0.0 +2022-11-15 18:10:07,568 INFO [train.py:876] (3/4) Epoch 4, batch 3000, loss[loss=0.1902, simple_loss=0.1924, pruned_loss=0.09406, over 5725.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.1936, pruned_loss=0.1013, over 1085122.23 frames. ], batch size: 31, lr: 1.94e-02, grad_scale: 16.0 +2022-11-15 18:10:07,568 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 18:10:16,679 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4769, 4.7913, 3.8190, 4.6511, 3.8046, 3.7082, 2.3860, 4.3196], + device='cuda:3'), covar=tensor([0.1798, 0.0132, 0.0958, 0.0190, 0.0413, 0.0729, 0.2247, 0.0143], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0119, 0.0165, 0.0115, 0.0155, 0.0180, 0.0194, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:10:18,016 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4063, 3.7970, 3.2947, 3.3186, 2.0729, 3.4165, 2.1351, 3.1347], + device='cuda:3'), covar=tensor([0.0275, 0.0053, 0.0093, 0.0123, 0.0361, 0.0081, 0.0250, 0.0052], + device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0094, 0.0110, 0.0115, 0.0145, 0.0106, 0.0127, 0.0092], + 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') +2022-11-15 18:10:26,937 INFO [train.py:908] (3/4) Epoch 4, validation: loss=0.1712, simple_loss=0.1916, pruned_loss=0.07544, over 1530663.00 frames. +2022-11-15 18:10:26,938 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 18:10:33,610 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:10:48,981 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:10:49,887 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 +2022-11-15 18:11:28,034 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:11:32,546 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:11:34,400 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 1.907e+02 2.538e+02 3.082e+02 5.544e+02, threshold=5.077e+02, percent-clipped=4.0 +2022-11-15 18:11:37,841 INFO [train.py:876] (3/4) Epoch 4, batch 3100, loss[loss=0.2066, simple_loss=0.1908, pruned_loss=0.1112, over 5574.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.1942, pruned_loss=0.1022, over 1082295.49 frames. ], batch size: 25, lr: 1.93e-02, grad_scale: 8.0 +2022-11-15 18:11:43,323 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:11:51,477 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 18:12:00,332 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3487, 2.1198, 1.5139, 2.6047, 1.3545, 1.3273, 1.7015, 2.2348], + device='cuda:3'), covar=tensor([0.0387, 0.0644, 0.1358, 0.0310, 0.1074, 0.0942, 0.0953, 0.1064], + device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0048, 0.0061, 0.0041, 0.0056, 0.0049, 0.0056, 0.0040], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 18:12:07,168 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9398, 1.0008, 1.1123, 0.9900, 1.2829, 1.3253, 0.7479, 1.4245], + device='cuda:3'), covar=tensor([0.0024, 0.0017, 0.0012, 0.0022, 0.0014, 0.0012, 0.0030, 0.0011], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0021, 0.0020, 0.0020, 0.0021, 0.0019, 0.0020, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.6354e-05, 2.4943e-05, 1.9057e-05, 1.9623e-05, 2.1476e-05, 1.6856e-05, + 2.8066e-05, 1.8829e-05], device='cuda:3') +2022-11-15 18:12:07,784 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2577, 1.3740, 1.4125, 1.1638, 0.1829, 1.6073, 1.1044, 1.2444], + device='cuda:3'), covar=tensor([0.0300, 0.0411, 0.0275, 0.0671, 0.1724, 0.0485, 0.0423, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0032, 0.0031, 0.0036, 0.0033, 0.0029, 0.0030, 0.0036], + device='cuda:3'), out_proj_covar=tensor([6.0198e-05, 5.0023e-05, 4.8034e-05, 6.8298e-05, 5.9860e-05, 5.2260e-05, + 5.2659e-05, 5.9866e-05], device='cuda:3') +2022-11-15 18:12:33,570 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:12:48,704 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 2.138e+02 2.743e+02 3.438e+02 6.298e+02, threshold=5.486e+02, percent-clipped=1.0 +2022-11-15 18:12:52,456 INFO [train.py:876] (3/4) Epoch 4, batch 3200, loss[loss=0.2123, simple_loss=0.2042, pruned_loss=0.1102, over 5588.00 frames. ], tot_loss[loss=0.201, simple_loss=0.1954, pruned_loss=0.1033, over 1081017.13 frames. ], batch size: 24, lr: 1.93e-02, grad_scale: 8.0 +2022-11-15 18:13:20,284 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 18:13:22,542 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 18:13:35,869 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7240, 2.0888, 3.2964, 2.7368, 3.5572, 2.3758, 3.0168, 3.7500], + device='cuda:3'), covar=tensor([0.0151, 0.0684, 0.0212, 0.0602, 0.0153, 0.0509, 0.0361, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0181, 0.0152, 0.0196, 0.0145, 0.0175, 0.0206, 0.0175], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:13:59,429 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3787, 2.1397, 1.7350, 2.4456, 1.3453, 1.5060, 1.4974, 2.5200], + device='cuda:3'), covar=tensor([0.0419, 0.0968, 0.1654, 0.0601, 0.1337, 0.0784, 0.1338, 0.0526], + device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0052, 0.0065, 0.0040, 0.0058, 0.0050, 0.0059, 0.0041], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 18:14:01,374 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.332e+02 2.126e+02 2.620e+02 3.307e+02 8.462e+02, threshold=5.241e+02, percent-clipped=1.0 +2022-11-15 18:14:05,060 INFO [train.py:876] (3/4) Epoch 4, batch 3300, loss[loss=0.2925, simple_loss=0.2372, pruned_loss=0.1739, over 3019.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.1939, pruned_loss=0.101, over 1082746.42 frames. ], batch size: 284, lr: 1.93e-02, grad_scale: 8.0 +2022-11-15 18:14:08,006 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:14:08,020 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0746, 4.4870, 3.8176, 4.5258, 4.4927, 3.7373, 3.9795, 3.6415], + device='cuda:3'), covar=tensor([0.0324, 0.0436, 0.1363, 0.0349, 0.0342, 0.0432, 0.0433, 0.0906], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0119, 0.0189, 0.0118, 0.0148, 0.0132, 0.0127, 0.0113], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:14:26,229 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8186, 3.9100, 3.9445, 4.0906, 3.5333, 3.0747, 4.4674, 3.7316], + device='cuda:3'), covar=tensor([0.0484, 0.0860, 0.0409, 0.0711, 0.0652, 0.0549, 0.0678, 0.0613], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0081, 0.0068, 0.0081, 0.0062, 0.0052, 0.0100, 0.0067], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:15:07,255 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:15:08,285 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:15:13,949 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 1.803e+02 2.408e+02 3.153e+02 7.479e+02, threshold=4.816e+02, percent-clipped=4.0 +2022-11-15 18:15:17,452 INFO [train.py:876] (3/4) Epoch 4, batch 3400, loss[loss=0.2403, simple_loss=0.2228, pruned_loss=0.1289, over 5544.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.1949, pruned_loss=0.1019, over 1082290.21 frames. ], batch size: 43, lr: 1.92e-02, grad_scale: 8.0 +2022-11-15 18:15:23,193 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:15:26,068 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-15 18:15:27,382 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6891, 1.6784, 1.9549, 2.3198, 2.5420, 2.0105, 1.5423, 2.7202], + device='cuda:3'), covar=tensor([0.0214, 0.3003, 0.1780, 0.0818, 0.0483, 0.1778, 0.1797, 0.0206], + device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0220, 0.0225, 0.0275, 0.0201, 0.0227, 0.0204, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:15:28,050 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7027, 2.2045, 2.9222, 1.1641, 2.6853, 2.8834, 2.6715, 3.3386], + device='cuda:3'), covar=tensor([0.1605, 0.1300, 0.0464, 0.1823, 0.0238, 0.0252, 0.0213, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0183, 0.0132, 0.0192, 0.0131, 0.0127, 0.0122, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:15:41,065 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:15:54,831 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.7056, 5.1100, 5.4169, 4.9453, 5.7911, 5.6434, 4.7381, 5.4817], + device='cuda:3'), covar=tensor([0.0280, 0.0176, 0.0335, 0.0271, 0.0236, 0.0066, 0.0214, 0.0418], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0077, 0.0101, 0.0096, 0.0058, 0.0085, 0.0090], + 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') +2022-11-15 18:15:56,864 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:16:23,804 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.900e+02 2.223e+02 2.867e+02 4.399e+02, threshold=4.447e+02, percent-clipped=0.0 +2022-11-15 18:16:27,931 INFO [train.py:876] (3/4) Epoch 4, batch 3500, loss[loss=0.2281, simple_loss=0.2259, pruned_loss=0.1151, over 5589.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.1923, pruned_loss=0.09909, over 1081703.17 frames. ], batch size: 24, lr: 1.92e-02, grad_scale: 8.0 +2022-11-15 18:16:40,681 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6030, 1.7570, 1.6536, 0.9274, 1.4227, 1.4497, 1.4175, 2.1847], + device='cuda:3'), covar=tensor([0.0018, 0.0016, 0.0012, 0.0020, 0.0023, 0.0010, 0.0015, 0.0020], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0020, 0.0020, 0.0022, 0.0019, 0.0020, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.5068e-05, 2.5432e-05, 1.9072e-05, 2.0012e-05, 2.1818e-05, 1.6618e-05, + 2.7296e-05, 1.9295e-05], device='cuda:3') +2022-11-15 18:16:46,960 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2022-11-15 18:16:51,499 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:17:07,621 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8447, 3.1436, 2.4527, 3.1813, 3.1874, 2.8651, 2.9830, 2.8271], + device='cuda:3'), covar=tensor([0.1364, 0.0673, 0.1805, 0.0544, 0.0605, 0.0536, 0.0511, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0118, 0.0186, 0.0117, 0.0148, 0.0128, 0.0124, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:17:17,587 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2050, 0.9091, 1.2248, 0.8083, 1.2228, 1.4140, 1.2215, 1.8095], + device='cuda:3'), covar=tensor([0.0025, 0.0033, 0.0018, 0.0028, 0.0028, 0.0014, 0.0033, 0.0015], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0020, 0.0020, 0.0021, 0.0018, 0.0020, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.4501e-05, 2.4707e-05, 1.8671e-05, 1.9794e-05, 2.1229e-05, 1.6266e-05, + 2.7018e-05, 1.8793e-05], device='cuda:3') +2022-11-15 18:17:34,604 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.249e+02 2.867e+02 3.402e+02 1.165e+03, threshold=5.733e+02, percent-clipped=11.0 +2022-11-15 18:17:38,067 INFO [train.py:876] (3/4) Epoch 4, batch 3600, loss[loss=0.1395, simple_loss=0.1471, pruned_loss=0.06598, over 5768.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.192, pruned_loss=0.09951, over 1087376.59 frames. ], batch size: 14, lr: 1.91e-02, grad_scale: 8.0 +2022-11-15 18:17:41,245 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:17:46,139 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:18:10,180 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 +2022-11-15 18:18:15,274 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:18:30,218 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:18:40,599 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:18:45,860 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.020e+02 2.444e+02 3.029e+02 5.693e+02, threshold=4.888e+02, percent-clipped=0.0 +2022-11-15 18:18:49,296 INFO [train.py:876] (3/4) Epoch 4, batch 3700, loss[loss=0.1828, simple_loss=0.1869, pruned_loss=0.08936, over 5580.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.1922, pruned_loss=0.09962, over 1087740.02 frames. ], batch size: 25, lr: 1.91e-02, grad_scale: 8.0 +2022-11-15 18:19:00,760 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9707, 3.0978, 2.9721, 2.9585, 3.1602, 3.0068, 1.2061, 3.0656], + device='cuda:3'), covar=tensor([0.0280, 0.0191, 0.0236, 0.0193, 0.0223, 0.0270, 0.2547, 0.0250], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0067, 0.0069, 0.0060, 0.0083, 0.0067, 0.0123, 0.0092], + 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') +2022-11-15 18:19:14,627 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:19:57,017 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 2.082e+02 2.548e+02 3.444e+02 5.526e+02, threshold=5.097e+02, percent-clipped=3.0 +2022-11-15 18:20:00,200 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-15 18:20:00,477 INFO [train.py:876] (3/4) Epoch 4, batch 3800, loss[loss=0.198, simple_loss=0.1965, pruned_loss=0.09975, over 5574.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.1922, pruned_loss=0.09944, over 1084281.35 frames. ], batch size: 22, lr: 1.91e-02, grad_scale: 8.0 +2022-11-15 18:20:10,273 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 18:20:24,499 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:20:58,562 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:21:08,419 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.968e+01 1.976e+02 2.317e+02 2.935e+02 6.102e+02, threshold=4.634e+02, percent-clipped=2.0 +2022-11-15 18:21:11,946 INFO [train.py:876] (3/4) Epoch 4, batch 3900, loss[loss=0.1739, simple_loss=0.1752, pruned_loss=0.08624, over 5612.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.1934, pruned_loss=0.1004, over 1085454.44 frames. ], batch size: 23, lr: 1.90e-02, grad_scale: 8.0 +2022-11-15 18:21:16,810 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9587, 3.2381, 2.9918, 3.0150, 3.1362, 3.0929, 1.0989, 3.1413], + device='cuda:3'), covar=tensor([0.0389, 0.0211, 0.0267, 0.0299, 0.0333, 0.0289, 0.3449, 0.0384], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0069, 0.0069, 0.0060, 0.0085, 0.0066, 0.0125, 0.0093], + 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') +2022-11-15 18:21:25,299 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2158, 2.2122, 3.7614, 3.0807, 4.1763, 2.7829, 3.8518, 4.1305], + device='cuda:3'), covar=tensor([0.0121, 0.0633, 0.0210, 0.0638, 0.0097, 0.0515, 0.0397, 0.0229], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0178, 0.0153, 0.0193, 0.0143, 0.0173, 0.0207, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:21:32,019 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7004, 2.0744, 2.8313, 3.3980, 3.6264, 2.7038, 2.0105, 3.8305], + device='cuda:3'), covar=tensor([0.0200, 0.3294, 0.2645, 0.2428, 0.0756, 0.3076, 0.2484, 0.0185], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0211, 0.0221, 0.0276, 0.0202, 0.0226, 0.0198, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:21:59,897 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:22:20,041 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.927e+01 1.909e+02 2.418e+02 3.208e+02 6.074e+02, threshold=4.836e+02, percent-clipped=3.0 +2022-11-15 18:22:24,177 INFO [train.py:876] (3/4) Epoch 4, batch 4000, loss[loss=0.1943, simple_loss=0.1958, pruned_loss=0.09635, over 5594.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.1929, pruned_loss=0.1005, over 1080207.73 frames. ], batch size: 18, lr: 1.90e-02, grad_scale: 8.0 +2022-11-15 18:22:45,999 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7666, 0.6584, 0.7568, 0.7323, 0.7649, 0.8703, 0.7053, 0.8432], + device='cuda:3'), covar=tensor([0.0120, 0.0222, 0.0184, 0.0244, 0.0148, 0.0105, 0.0191, 0.0167], + device='cuda:3'), in_proj_covar=tensor([0.0007, 0.0011, 0.0009, 0.0009, 0.0009, 0.0008, 0.0009, 0.0008], + device='cuda:3'), out_proj_covar=tensor([2.7980e-05, 3.6213e-05, 3.1461e-05, 3.3650e-05, 3.2322e-05, 2.9599e-05, + 3.1788e-05, 3.0500e-05], device='cuda:3') +2022-11-15 18:22:52,062 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:23:30,893 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.955e+02 2.286e+02 3.069e+02 5.405e+02, threshold=4.573e+02, percent-clipped=2.0 +2022-11-15 18:23:35,102 INFO [train.py:876] (3/4) Epoch 4, batch 4100, loss[loss=0.2147, simple_loss=0.2111, pruned_loss=0.1092, over 5569.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.1946, pruned_loss=0.1014, over 1081307.24 frames. ], batch size: 25, lr: 1.90e-02, grad_scale: 8.0 +2022-11-15 18:23:37,082 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:24:08,157 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0646, 1.0057, 1.1919, 0.9770, 1.1823, 1.3295, 0.7567, 1.3160], + device='cuda:3'), covar=tensor([0.0017, 0.0013, 0.0010, 0.0013, 0.0015, 0.0008, 0.0019, 0.0011], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0020, 0.0020, 0.0020, 0.0020, 0.0018, 0.0020, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.4202e-05, 2.2999e-05, 1.8541e-05, 1.9919e-05, 2.0049e-05, 1.5880e-05, + 2.6888e-05, 1.9132e-05], device='cuda:3') +2022-11-15 18:24:08,340 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 18:24:41,171 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 18:24:42,055 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.057e+02 2.578e+02 3.231e+02 5.725e+02, threshold=5.157e+02, percent-clipped=4.0 +2022-11-15 18:24:45,827 INFO [train.py:876] (3/4) Epoch 4, batch 4200, loss[loss=0.215, simple_loss=0.2035, pruned_loss=0.1132, over 5695.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.1943, pruned_loss=0.1011, over 1082389.05 frames. ], batch size: 36, lr: 1.89e-02, grad_scale: 8.0 +2022-11-15 18:24:48,359 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9696, 4.4979, 4.5753, 3.8179, 4.7706, 4.8021, 2.3094, 4.7854], + device='cuda:3'), covar=tensor([0.0244, 0.0302, 0.0160, 0.0515, 0.0251, 0.0230, 0.2562, 0.0332], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0065, 0.0066, 0.0057, 0.0082, 0.0065, 0.0119, 0.0088], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:24:55,185 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5353, 1.9832, 2.4176, 3.2471, 3.4762, 2.5058, 2.0437, 3.5458], + device='cuda:3'), covar=tensor([0.0279, 0.2964, 0.2444, 0.3043, 0.0659, 0.2705, 0.1975, 0.0224], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0214, 0.0227, 0.0283, 0.0200, 0.0227, 0.0195, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:25:34,269 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:25:46,687 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:25:53,374 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 1.847e+02 2.412e+02 3.140e+02 6.094e+02, threshold=4.824e+02, percent-clipped=3.0 +2022-11-15 18:25:56,767 INFO [train.py:876] (3/4) Epoch 4, batch 4300, loss[loss=0.2581, simple_loss=0.2184, pruned_loss=0.1489, over 5473.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.1931, pruned_loss=0.1008, over 1082381.80 frames. ], batch size: 64, lr: 1.89e-02, grad_scale: 8.0 +2022-11-15 18:26:08,217 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:26:18,317 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5228, 3.9105, 3.3409, 3.3772, 2.5158, 4.0836, 2.3011, 3.4315], + device='cuda:3'), covar=tensor([0.0267, 0.0091, 0.0120, 0.0202, 0.0262, 0.0052, 0.0229, 0.0053], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0099, 0.0114, 0.0119, 0.0147, 0.0111, 0.0130, 0.0095], + 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') +2022-11-15 18:26:18,916 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1320, 3.9022, 2.5341, 3.7506, 2.8321, 2.5019, 2.0305, 3.2921], + device='cuda:3'), covar=tensor([0.1447, 0.0139, 0.0986, 0.0178, 0.0664, 0.1009, 0.1914, 0.0207], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0120, 0.0171, 0.0119, 0.0157, 0.0182, 0.0191, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:26:22,516 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 18:26:27,852 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 18:26:29,695 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:26:33,041 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:27:04,695 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 2.078e+02 2.562e+02 3.309e+02 6.433e+02, threshold=5.124e+02, percent-clipped=7.0 +2022-11-15 18:27:06,220 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:27:08,249 INFO [train.py:876] (3/4) Epoch 4, batch 4400, loss[loss=0.1995, simple_loss=0.1878, pruned_loss=0.1055, over 5754.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.1919, pruned_loss=0.09935, over 1080568.45 frames. ], batch size: 8, lr: 1.89e-02, grad_scale: 8.0 +2022-11-15 18:27:08,451 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8366, 2.0844, 2.6286, 3.5688, 3.7888, 2.7253, 2.1618, 3.9000], + device='cuda:3'), covar=tensor([0.0194, 0.3222, 0.2505, 0.2912, 0.0630, 0.3137, 0.2394, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0212, 0.0222, 0.0282, 0.0201, 0.0226, 0.0201, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:27:16,434 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:27:32,227 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:27:55,911 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5260, 3.6077, 2.8050, 1.6232, 3.4332, 1.4099, 3.4578, 1.7338], + device='cuda:3'), covar=tensor([0.1019, 0.0189, 0.0721, 0.2151, 0.0254, 0.2223, 0.0200, 0.1982], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0095, 0.0101, 0.0120, 0.0097, 0.0134, 0.0086, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 18:28:07,277 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3681, 2.6471, 2.5372, 2.5414, 2.4439, 2.6125, 1.1194, 2.4570], + device='cuda:3'), covar=tensor([0.0498, 0.0340, 0.0334, 0.0296, 0.0529, 0.0353, 0.3241, 0.0468], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0069, 0.0069, 0.0060, 0.0085, 0.0067, 0.0127, 0.0091], + 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') +2022-11-15 18:28:15,306 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:28:15,742 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.026e+02 2.337e+02 2.838e+02 7.809e+02, threshold=4.674e+02, percent-clipped=2.0 +2022-11-15 18:28:19,245 INFO [train.py:876] (3/4) Epoch 4, batch 4500, loss[loss=0.1563, simple_loss=0.157, pruned_loss=0.07777, over 5484.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.1923, pruned_loss=0.09847, over 1086592.36 frames. ], batch size: 12, lr: 1.88e-02, grad_scale: 8.0 +2022-11-15 18:28:21,474 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8385, 4.9554, 3.3614, 4.8387, 3.7042, 3.4187, 3.0285, 4.3333], + device='cuda:3'), covar=tensor([0.1316, 0.0117, 0.0768, 0.0135, 0.0361, 0.0701, 0.1488, 0.0102], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0121, 0.0170, 0.0119, 0.0157, 0.0179, 0.0191, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:29:04,170 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6703, 1.9837, 2.6272, 1.9212, 1.4489, 1.5827, 2.0440, 1.4611], + device='cuda:3'), covar=tensor([0.0014, 0.0024, 0.0010, 0.0015, 0.0038, 0.0029, 0.0012, 0.0024], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0014, 0.0014, 0.0013, 0.0014, 0.0014], + device='cuda:3'), out_proj_covar=tensor([1.3975e-05, 1.4878e-05, 1.3120e-05, 1.5321e-05, 1.4088e-05, 1.4898e-05, + 1.6007e-05, 1.8362e-05], device='cuda:3') +2022-11-15 18:29:10,958 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2418, 4.2116, 3.9294, 4.1173, 4.3188, 3.8390, 1.6800, 4.3794], + device='cuda:3'), covar=tensor([0.0256, 0.0282, 0.0232, 0.0214, 0.0233, 0.0374, 0.2546, 0.0217], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0068, 0.0068, 0.0059, 0.0084, 0.0066, 0.0125, 0.0090], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:29:27,463 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.298e+02 1.949e+02 2.521e+02 3.180e+02 6.694e+02, threshold=5.042e+02, percent-clipped=2.0 +2022-11-15 18:29:29,018 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:29:31,214 INFO [train.py:876] (3/4) Epoch 4, batch 4600, loss[loss=0.2634, simple_loss=0.2173, pruned_loss=0.1547, over 4064.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.192, pruned_loss=0.09862, over 1082247.80 frames. ], batch size: 181, lr: 1.88e-02, grad_scale: 8.0 +2022-11-15 18:29:36,994 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 +2022-11-15 18:29:59,375 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:30:02,517 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2194, 1.3828, 1.2649, 0.9585, 1.3288, 1.1676, 1.0689, 1.3716], + device='cuda:3'), covar=tensor([0.0029, 0.0019, 0.0013, 0.0016, 0.0018, 0.0013, 0.0047, 0.0010], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0020, 0.0021, 0.0019, 0.0020, 0.0019], + device='cuda:3'), out_proj_covar=tensor([2.5250e-05, 2.5884e-05, 1.9089e-05, 1.9895e-05, 2.0298e-05, 1.6189e-05, + 2.7343e-05, 1.8928e-05], device='cuda:3') +2022-11-15 18:30:12,147 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:30:28,019 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-15 18:30:29,489 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2022-11-15 18:30:37,113 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 1.888e+02 2.493e+02 3.125e+02 5.400e+02, threshold=4.987e+02, percent-clipped=2.0 +2022-11-15 18:30:38,714 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:30:40,596 INFO [train.py:876] (3/4) Epoch 4, batch 4700, loss[loss=0.1902, simple_loss=0.2005, pruned_loss=0.08996, over 5682.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.1898, pruned_loss=0.0972, over 1082718.01 frames. ], batch size: 36, lr: 1.88e-02, grad_scale: 8.0 +2022-11-15 18:30:46,190 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:31:12,678 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:31:19,492 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1275, 4.6660, 3.6503, 1.9649, 4.4180, 2.2671, 4.4869, 2.7510], + device='cuda:3'), covar=tensor([0.1015, 0.0104, 0.0392, 0.2260, 0.0148, 0.1652, 0.0112, 0.1370], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0097, 0.0102, 0.0121, 0.0100, 0.0136, 0.0086, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 18:31:43,665 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:31:44,178 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 18:31:47,742 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.139e+02 2.674e+02 3.379e+02 6.556e+02, threshold=5.348e+02, percent-clipped=3.0 +2022-11-15 18:31:51,249 INFO [train.py:876] (3/4) Epoch 4, batch 4800, loss[loss=0.2132, simple_loss=0.2009, pruned_loss=0.1128, over 5322.00 frames. ], tot_loss[loss=0.195, simple_loss=0.192, pruned_loss=0.09898, over 1087545.68 frames. ], batch size: 79, lr: 1.87e-02, grad_scale: 8.0 +2022-11-15 18:32:06,293 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 +2022-11-15 18:32:58,144 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.842e+02 2.215e+02 2.988e+02 4.976e+02, threshold=4.429e+02, percent-clipped=0.0 +2022-11-15 18:33:01,598 INFO [train.py:876] (3/4) Epoch 4, batch 4900, loss[loss=0.21, simple_loss=0.2083, pruned_loss=0.1059, over 5594.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.1921, pruned_loss=0.09861, over 1086823.34 frames. ], batch size: 22, lr: 1.87e-02, grad_scale: 8.0 +2022-11-15 18:33:02,463 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5452, 1.9037, 2.9896, 2.6960, 3.0288, 1.8333, 2.7653, 3.4576], + device='cuda:3'), covar=tensor([0.0125, 0.0541, 0.0207, 0.0493, 0.0160, 0.0541, 0.0354, 0.0201], + device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0187, 0.0167, 0.0201, 0.0155, 0.0179, 0.0221, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:33:31,040 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:33:39,433 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:33:42,144 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8075, 1.0564, 1.3814, 0.7148, 1.3905, 0.8829, 0.8092, 1.3289], + device='cuda:3'), covar=tensor([0.0025, 0.0018, 0.0012, 0.0021, 0.0016, 0.0018, 0.0032, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0022, 0.0023, 0.0023, 0.0021, 0.0022, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.8460e-05, 2.8150e-05, 2.0918e-05, 2.2542e-05, 2.2438e-05, 1.7951e-05, + 2.9767e-05, 2.1371e-05], device='cuda:3') +2022-11-15 18:34:05,207 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:34:09,166 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.286e+02 1.967e+02 2.433e+02 3.040e+02 5.263e+02, threshold=4.865e+02, percent-clipped=3.0 +2022-11-15 18:34:12,717 INFO [train.py:876] (3/4) Epoch 4, batch 5000, loss[loss=0.1774, simple_loss=0.1875, pruned_loss=0.08364, over 5547.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.1919, pruned_loss=0.09823, over 1092427.38 frames. ], batch size: 16, lr: 1.87e-02, grad_scale: 8.0 +2022-11-15 18:34:14,281 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7399, 1.1386, 2.0330, 0.6355, 1.3725, 0.9498, 1.1624, 1.9794], + device='cuda:3'), covar=tensor([0.0020, 0.0062, 0.0008, 0.0030, 0.0175, 0.0019, 0.0043, 0.0018], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0021, 0.0022, 0.0022, 0.0020, 0.0021, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.7020e-05, 2.6718e-05, 1.9748e-05, 2.1482e-05, 2.1365e-05, 1.6939e-05, + 2.7888e-05, 2.0487e-05], device='cuda:3') +2022-11-15 18:34:17,622 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:34:19,200 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 +2022-11-15 18:34:51,701 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:35:02,896 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.75 vs. limit=5.0 +2022-11-15 18:35:14,882 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:35:16,185 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:35:20,045 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.925e+02 2.492e+02 3.201e+02 4.823e+02, threshold=4.984e+02, percent-clipped=0.0 +2022-11-15 18:35:23,729 INFO [train.py:876] (3/4) Epoch 4, batch 5100, loss[loss=0.1273, simple_loss=0.1497, pruned_loss=0.05249, over 5511.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.1929, pruned_loss=0.09917, over 1086619.45 frames. ], batch size: 12, lr: 1.86e-02, grad_scale: 16.0 +2022-11-15 18:35:46,960 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:35:49,816 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1696, 1.6577, 1.9875, 1.1253, 0.7106, 1.9373, 1.3277, 1.3860], + device='cuda:3'), covar=tensor([0.0346, 0.0320, 0.0283, 0.0697, 0.0985, 0.1019, 0.0565, 0.0661], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0036, 0.0036, 0.0041, 0.0036, 0.0033, 0.0032, 0.0040], + device='cuda:3'), out_proj_covar=tensor([6.5115e-05, 5.7811e-05, 5.7971e-05, 7.7117e-05, 6.6635e-05, 5.9821e-05, + 5.6354e-05, 6.8487e-05], device='cuda:3') +2022-11-15 18:35:50,397 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:35:51,231 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1766, 3.9744, 2.8206, 3.8019, 2.9705, 2.7042, 1.8170, 3.2626], + device='cuda:3'), covar=tensor([0.1307, 0.0134, 0.0734, 0.0195, 0.0465, 0.0781, 0.1826, 0.0171], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0122, 0.0168, 0.0121, 0.0156, 0.0180, 0.0188, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:35:58,390 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:36:31,256 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:36:31,700 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.232e+02 2.660e+02 3.188e+02 6.115e+02, threshold=5.320e+02, percent-clipped=2.0 +2022-11-15 18:36:35,115 INFO [train.py:876] (3/4) Epoch 4, batch 5200, loss[loss=0.182, simple_loss=0.1893, pruned_loss=0.08737, over 5736.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.1936, pruned_loss=0.09997, over 1077920.65 frames. ], batch size: 15, lr: 1.86e-02, grad_scale: 16.0 +2022-11-15 18:37:12,644 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:37:26,754 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 18:37:27,709 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4989, 1.1791, 1.6430, 1.2469, 0.9492, 1.3958, 2.0694, 1.2188], + device='cuda:3'), covar=tensor([0.0019, 0.0156, 0.0056, 0.0022, 0.0065, 0.0040, 0.0016, 0.0030], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0014, 0.0013, 0.0015, 0.0015, 0.0015, 0.0016, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.5712e-05, 1.6715e-05, 1.4870e-05, 1.6733e-05, 1.6028e-05, 1.6554e-05, + 1.7460e-05, 2.0315e-05], device='cuda:3') +2022-11-15 18:37:42,559 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.845e+02 2.257e+02 3.098e+02 5.332e+02, threshold=4.514e+02, percent-clipped=1.0 +2022-11-15 18:37:45,683 INFO [train.py:876] (3/4) Epoch 4, batch 5300, loss[loss=0.2642, simple_loss=0.2327, pruned_loss=0.1479, over 4718.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.1915, pruned_loss=0.09852, over 1077380.45 frames. ], batch size: 135, lr: 1.86e-02, grad_scale: 8.0 +2022-11-15 18:37:46,367 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:37:48,493 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5881, 1.8310, 1.9634, 1.2392, 0.9001, 2.3340, 1.3341, 1.4412], + device='cuda:3'), covar=tensor([0.0360, 0.0800, 0.0450, 0.0944, 0.1205, 0.1037, 0.1011, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0035, 0.0037, 0.0040, 0.0034, 0.0031, 0.0030, 0.0038], + device='cuda:3'), out_proj_covar=tensor([6.4170e-05, 5.6017e-05, 5.8300e-05, 7.5832e-05, 6.3288e-05, 5.6942e-05, + 5.3519e-05, 6.5249e-05], device='cuda:3') +2022-11-15 18:38:07,369 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:38:49,398 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0587, 1.0067, 1.2054, 0.5518, 0.8899, 0.8491, 0.4893, 1.1644], + device='cuda:3'), covar=tensor([0.0018, 0.0012, 0.0018, 0.0018, 0.0028, 0.0018, 0.0048, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0021, 0.0022, 0.0022, 0.0020, 0.0021, 0.0021], + device='cuda:3'), out_proj_covar=tensor([2.5824e-05, 2.6329e-05, 1.9884e-05, 2.1634e-05, 2.0959e-05, 1.6888e-05, + 2.8032e-05, 2.0359e-05], device='cuda:3') +2022-11-15 18:38:50,800 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:38:53,683 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.431e+02 2.106e+02 2.706e+02 3.414e+02 5.155e+02, threshold=5.411e+02, percent-clipped=7.0 +2022-11-15 18:38:56,361 INFO [train.py:876] (3/4) Epoch 4, batch 5400, loss[loss=0.1562, simple_loss=0.1752, pruned_loss=0.0686, over 5529.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.191, pruned_loss=0.0981, over 1076098.79 frames. ], batch size: 21, lr: 1.85e-02, grad_scale: 8.0 +2022-11-15 18:39:09,473 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2022-11-15 18:39:26,845 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:39:59,508 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:40:04,134 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 2.138e+02 2.582e+02 3.197e+02 8.943e+02, threshold=5.164e+02, percent-clipped=1.0 +2022-11-15 18:40:07,247 INFO [train.py:876] (3/4) Epoch 4, batch 5500, loss[loss=0.1061, simple_loss=0.1282, pruned_loss=0.04207, over 5402.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.1907, pruned_loss=0.09727, over 1079204.63 frames. ], batch size: 9, lr: 1.85e-02, grad_scale: 8.0 +2022-11-15 18:40:16,189 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7943, 4.2546, 3.7665, 3.7344, 2.5120, 4.4804, 2.5356, 3.7107], + device='cuda:3'), covar=tensor([0.0267, 0.0073, 0.0119, 0.0188, 0.0335, 0.0056, 0.0234, 0.0057], + device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0102, 0.0119, 0.0127, 0.0150, 0.0116, 0.0135, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:40:23,284 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2022-11-15 18:41:15,201 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 2.174e+02 2.548e+02 3.213e+02 6.484e+02, threshold=5.095e+02, percent-clipped=3.0 +2022-11-15 18:41:17,888 INFO [train.py:876] (3/4) Epoch 4, batch 5600, loss[loss=0.2905, simple_loss=0.249, pruned_loss=0.166, over 5441.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.1899, pruned_loss=0.09713, over 1084094.81 frames. ], batch size: 58, lr: 1.85e-02, grad_scale: 8.0 +2022-11-15 18:41:49,554 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 +2022-11-15 18:42:19,420 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:42:25,764 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.283e+02 2.178e+02 2.574e+02 3.004e+02 8.808e+02, threshold=5.148e+02, percent-clipped=2.0 +2022-11-15 18:42:28,790 INFO [train.py:876] (3/4) Epoch 4, batch 5700, loss[loss=0.238, simple_loss=0.2151, pruned_loss=0.1304, over 5009.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.1893, pruned_loss=0.09615, over 1086080.14 frames. ], batch size: 109, lr: 1.84e-02, grad_scale: 8.0 +2022-11-15 18:42:58,708 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 +2022-11-15 18:42:59,124 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:42:59,447 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.94 vs. limit=5.0 +2022-11-15 18:43:10,885 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3239, 3.4876, 3.2853, 3.2357, 3.3252, 3.1993, 1.3240, 3.4676], + device='cuda:3'), covar=tensor([0.0373, 0.0220, 0.0256, 0.0285, 0.0346, 0.0350, 0.3204, 0.0344], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0068, 0.0068, 0.0058, 0.0087, 0.0069, 0.0125, 0.0093], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 18:43:32,366 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:43:33,706 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:43:37,001 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 1.840e+02 2.310e+02 2.841e+02 5.066e+02, threshold=4.620e+02, percent-clipped=0.0 +2022-11-15 18:43:40,072 INFO [train.py:876] (3/4) Epoch 4, batch 5800, loss[loss=0.2228, simple_loss=0.2105, pruned_loss=0.1176, over 5290.00 frames. ], tot_loss[loss=0.192, simple_loss=0.1903, pruned_loss=0.09687, over 1079948.90 frames. ], batch size: 79, lr: 1.84e-02, grad_scale: 8.0 +2022-11-15 18:43:53,938 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.25 vs. limit=2.0 +2022-11-15 18:44:06,337 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:44:16,619 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2022-11-15 18:44:31,442 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2017, 1.4068, 2.1389, 0.9623, 1.4713, 1.4948, 1.3128, 1.3361], + device='cuda:3'), covar=tensor([0.0043, 0.0049, 0.0012, 0.0017, 0.0019, 0.0020, 0.0023, 0.0021], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0014, 0.0013, 0.0015, 0.0015, 0.0014, 0.0016, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.5661e-05, 1.6264e-05, 1.4153e-05, 1.6045e-05, 1.4983e-05, 1.5877e-05, + 1.6887e-05, 1.9795e-05], device='cuda:3') +2022-11-15 18:44:47,990 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.281e+02 1.945e+02 2.573e+02 2.996e+02 4.462e+02, threshold=5.147e+02, percent-clipped=0.0 +2022-11-15 18:44:48,889 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:44:51,047 INFO [train.py:876] (3/4) Epoch 4, batch 5900, loss[loss=0.1597, simple_loss=0.1751, pruned_loss=0.07213, over 5737.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.1907, pruned_loss=0.09803, over 1080831.92 frames. ], batch size: 14, lr: 1.84e-02, grad_scale: 8.0 +2022-11-15 18:45:32,353 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:45:53,385 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:45:59,276 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.246e+02 2.057e+02 2.608e+02 3.378e+02 5.938e+02, threshold=5.215e+02, percent-clipped=4.0 +2022-11-15 18:46:02,468 INFO [train.py:876] (3/4) Epoch 4, batch 6000, loss[loss=0.1945, simple_loss=0.2076, pruned_loss=0.09071, over 5648.00 frames. ], tot_loss[loss=0.196, simple_loss=0.1927, pruned_loss=0.09969, over 1078672.82 frames. ], batch size: 18, lr: 1.83e-02, grad_scale: 8.0 +2022-11-15 18:46:02,469 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 18:46:08,608 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6422, 3.5169, 2.7129, 1.7934, 3.3637, 1.4439, 3.2609, 2.2761], + device='cuda:3'), covar=tensor([0.0997, 0.0183, 0.1047, 0.1932, 0.0180, 0.1905, 0.0245, 0.1514], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0098, 0.0102, 0.0121, 0.0098, 0.0130, 0.0086, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 18:46:20,379 INFO [train.py:908] (3/4) Epoch 4, validation: loss=0.1691, simple_loss=0.1898, pruned_loss=0.07419, over 1530663.00 frames. +2022-11-15 18:46:20,380 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 18:46:45,374 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:47:28,451 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.961e+02 2.307e+02 2.838e+02 4.570e+02, threshold=4.614e+02, percent-clipped=0.0 +2022-11-15 18:47:31,205 INFO [train.py:876] (3/4) Epoch 4, batch 6100, loss[loss=0.1589, simple_loss=0.1722, pruned_loss=0.07278, over 5573.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.1932, pruned_loss=0.09949, over 1082195.11 frames. ], batch size: 16, lr: 1.83e-02, grad_scale: 8.0 +2022-11-15 18:48:39,806 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.715e+01 1.962e+02 2.542e+02 3.499e+02 5.895e+02, threshold=5.084e+02, percent-clipped=10.0 +2022-11-15 18:48:42,563 INFO [train.py:876] (3/4) Epoch 4, batch 6200, loss[loss=0.1515, simple_loss=0.1611, pruned_loss=0.07094, over 5559.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.194, pruned_loss=0.1008, over 1075367.47 frames. ], batch size: 16, lr: 1.83e-02, grad_scale: 8.0 +2022-11-15 18:48:48,529 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3416, 1.7554, 1.8823, 1.0822, 1.0080, 2.0999, 1.6633, 1.1737], + device='cuda:3'), covar=tensor([0.0429, 0.0471, 0.0407, 0.1083, 0.0933, 0.0807, 0.0587, 0.0643], + device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0036, 0.0037, 0.0043, 0.0035, 0.0032, 0.0030, 0.0039], + device='cuda:3'), out_proj_covar=tensor([6.7559e-05, 5.9669e-05, 6.0262e-05, 8.2520e-05, 6.4799e-05, 6.0051e-05, + 5.4700e-05, 6.8576e-05], device='cuda:3') +2022-11-15 18:49:14,767 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2022-11-15 18:49:19,867 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 18:49:34,449 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9586, 1.3025, 2.2034, 1.7504, 1.4653, 1.7013, 1.7782, 1.5114], + device='cuda:3'), covar=tensor([0.0012, 0.0029, 0.0026, 0.0013, 0.0032, 0.0022, 0.0017, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0013, 0.0016, 0.0016, 0.0014, 0.0017, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.6061e-05, 1.6984e-05, 1.5104e-05, 1.6724e-05, 1.6181e-05, 1.5928e-05, + 1.7877e-05, 2.0050e-05], device='cuda:3') +2022-11-15 18:49:50,343 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.008e+02 2.464e+02 2.967e+02 7.434e+02, threshold=4.928e+02, percent-clipped=2.0 +2022-11-15 18:49:53,145 INFO [train.py:876] (3/4) Epoch 4, batch 6300, loss[loss=0.2129, simple_loss=0.2097, pruned_loss=0.1081, over 5622.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.1897, pruned_loss=0.09707, over 1075424.02 frames. ], batch size: 38, lr: 1.82e-02, grad_scale: 8.0 +2022-11-15 18:50:30,001 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 +2022-11-15 18:50:55,391 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 18:51:00,494 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 2.107e+02 2.497e+02 3.273e+02 5.114e+02, threshold=4.994e+02, percent-clipped=2.0 +2022-11-15 18:51:01,660 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 +2022-11-15 18:51:04,030 INFO [train.py:876] (3/4) Epoch 4, batch 6400, loss[loss=0.2569, simple_loss=0.2267, pruned_loss=0.1435, over 5124.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.1896, pruned_loss=0.09726, over 1078164.49 frames. ], batch size: 91, lr: 1.82e-02, grad_scale: 8.0 +2022-11-15 18:51:12,366 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:51:14,446 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0458, 2.5590, 2.6883, 2.4945, 1.6082, 2.6023, 1.7147, 1.9000], + device='cuda:3'), covar=tensor([0.0160, 0.0043, 0.0057, 0.0096, 0.0193, 0.0060, 0.0178, 0.0069], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0103, 0.0119, 0.0130, 0.0152, 0.0119, 0.0140, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:51:22,399 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2022-11-15 18:51:30,765 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:51:33,371 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5646, 1.8993, 1.6651, 1.1609, 1.8712, 0.8363, 1.9545, 1.0772], + device='cuda:3'), covar=tensor([0.0675, 0.0166, 0.0526, 0.1023, 0.0242, 0.1619, 0.0206, 0.1108], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0098, 0.0100, 0.0118, 0.0101, 0.0129, 0.0086, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 18:51:53,834 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:51:55,864 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:52:06,495 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9358, 4.4372, 4.7549, 4.4832, 4.9655, 4.7822, 4.3569, 4.9448], + device='cuda:3'), covar=tensor([0.0383, 0.0204, 0.0426, 0.0270, 0.0315, 0.0108, 0.0208, 0.0266], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0101, 0.0082, 0.0111, 0.0107, 0.0061, 0.0088, 0.0100], + 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') +2022-11-15 18:52:11,784 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 2.095e+02 2.644e+02 3.190e+02 6.758e+02, threshold=5.289e+02, percent-clipped=3.0 +2022-11-15 18:52:14,071 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 18:52:14,562 INFO [train.py:876] (3/4) Epoch 4, batch 6500, loss[loss=0.2029, simple_loss=0.2076, pruned_loss=0.09905, over 5591.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.1899, pruned_loss=0.09696, over 1085131.77 frames. ], batch size: 18, lr: 1.82e-02, grad_scale: 8.0 +2022-11-15 18:52:36,613 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:52:41,541 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 18:52:51,698 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 18:52:52,122 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 18:52:52,218 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2884, 3.8575, 3.2756, 3.0397, 2.3613, 3.4926, 2.1521, 3.2706], + device='cuda:3'), covar=tensor([0.0305, 0.0054, 0.0095, 0.0219, 0.0293, 0.0089, 0.0246, 0.0063], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0104, 0.0120, 0.0132, 0.0151, 0.0118, 0.0137, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:52:55,005 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8563, 2.1751, 3.3648, 2.8456, 3.6815, 2.3005, 3.3174, 3.7639], + device='cuda:3'), covar=tensor([0.0147, 0.0804, 0.0279, 0.0801, 0.0164, 0.0667, 0.0479, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0186, 0.0159, 0.0199, 0.0155, 0.0177, 0.0211, 0.0182], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:52:58,845 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.39 vs. limit=5.0 +2022-11-15 18:53:22,963 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 1.962e+02 2.374e+02 2.918e+02 5.741e+02, threshold=4.749e+02, percent-clipped=1.0 +2022-11-15 18:53:23,991 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3467, 4.1243, 3.5457, 3.4333, 2.2639, 3.9841, 2.1774, 3.7112], + device='cuda:3'), covar=tensor([0.0367, 0.0082, 0.0129, 0.0250, 0.0350, 0.0091, 0.0294, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0103, 0.0120, 0.0131, 0.0151, 0.0118, 0.0138, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:53:26,219 INFO [train.py:876] (3/4) Epoch 4, batch 6600, loss[loss=0.225, simple_loss=0.2158, pruned_loss=0.1171, over 5683.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.1891, pruned_loss=0.09581, over 1088275.66 frames. ], batch size: 34, lr: 1.81e-02, grad_scale: 8.0 +2022-11-15 18:53:26,946 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 18:53:48,628 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 18:53:55,643 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 +2022-11-15 18:54:14,669 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:54:19,424 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:54:33,671 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.033e+02 2.580e+02 2.965e+02 5.128e+02, threshold=5.159e+02, percent-clipped=4.0 +2022-11-15 18:54:36,432 INFO [train.py:876] (3/4) Epoch 4, batch 6700, loss[loss=0.233, simple_loss=0.204, pruned_loss=0.131, over 5018.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.189, pruned_loss=0.09708, over 1083846.97 frames. ], batch size: 110, lr: 1.81e-02, grad_scale: 8.0 +2022-11-15 18:54:57,432 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:55:02,310 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:55:24,587 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:55:39,795 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 18:55:43,579 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 18:55:44,798 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 2.021e+02 2.337e+02 3.022e+02 5.149e+02, threshold=4.674e+02, percent-clipped=0.0 +2022-11-15 18:55:47,615 INFO [train.py:876] (3/4) Epoch 4, batch 6800, loss[loss=0.2282, simple_loss=0.2095, pruned_loss=0.1234, over 5197.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.1908, pruned_loss=0.09896, over 1078020.80 frames. ], batch size: 91, lr: 1.81e-02, grad_scale: 8.0 +2022-11-15 18:56:06,106 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:56:11,011 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0100, 3.9798, 2.3555, 3.6311, 2.8996, 2.6783, 1.9543, 3.5046], + device='cuda:3'), covar=tensor([0.1692, 0.0142, 0.1218, 0.0316, 0.0556, 0.0900, 0.1987, 0.0223], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0124, 0.0171, 0.0127, 0.0156, 0.0181, 0.0188, 0.0128], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 18:56:56,415 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 2.083e+02 2.579e+02 3.189e+02 6.516e+02, threshold=5.157e+02, percent-clipped=5.0 +2022-11-15 18:56:59,135 INFO [train.py:876] (3/4) Epoch 4, batch 6900, loss[loss=0.2338, simple_loss=0.2075, pruned_loss=0.13, over 4992.00 frames. ], tot_loss[loss=0.194, simple_loss=0.191, pruned_loss=0.09852, over 1077832.26 frames. ], batch size: 109, lr: 1.80e-02, grad_scale: 8.0 +2022-11-15 18:57:29,224 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 +2022-11-15 18:57:36,372 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 18:57:45,088 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4508, 2.0204, 1.5859, 1.2306, 1.4018, 2.1214, 1.9758, 1.9783], + device='cuda:3'), covar=tensor([0.1257, 0.0873, 0.1225, 0.1700, 0.0660, 0.0361, 0.0265, 0.0528], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0188, 0.0136, 0.0193, 0.0142, 0.0137, 0.0126, 0.0154], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 18:58:07,454 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.482e+01 2.092e+02 2.563e+02 3.027e+02 6.161e+02, threshold=5.126e+02, percent-clipped=1.0 +2022-11-15 18:58:10,632 INFO [train.py:876] (3/4) Epoch 4, batch 7000, loss[loss=0.1498, simple_loss=0.1702, pruned_loss=0.06473, over 5547.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.191, pruned_loss=0.0981, over 1077651.64 frames. ], batch size: 15, lr: 1.80e-02, grad_scale: 8.0 +2022-11-15 18:58:28,005 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:58:32,054 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:58:41,453 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:58:58,295 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:59:12,507 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1848, 1.7549, 2.8063, 2.2679, 2.7232, 1.9856, 2.6635, 2.9613], + device='cuda:3'), covar=tensor([0.0146, 0.0674, 0.0217, 0.0554, 0.0204, 0.0539, 0.0359, 0.0222], + device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0185, 0.0161, 0.0199, 0.0159, 0.0180, 0.0213, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 18:59:16,482 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 18:59:17,672 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 2.242e+02 2.603e+02 3.137e+02 5.359e+02, threshold=5.206e+02, percent-clipped=2.0 +2022-11-15 18:59:21,227 INFO [train.py:876] (3/4) Epoch 4, batch 7100, loss[loss=0.2042, simple_loss=0.1786, pruned_loss=0.1149, over 4155.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.1916, pruned_loss=0.09846, over 1079111.39 frames. ], batch size: 181, lr: 1.80e-02, grad_scale: 8.0 +2022-11-15 18:59:24,890 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:59:32,541 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:59:39,673 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 18:59:50,846 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:00:13,729 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:00:21,874 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:00:22,221 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 19:00:29,545 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 2.015e+02 2.566e+02 3.105e+02 6.744e+02, threshold=5.133e+02, percent-clipped=2.0 +2022-11-15 19:00:32,330 INFO [train.py:876] (3/4) Epoch 4, batch 7200, loss[loss=0.2615, simple_loss=0.2407, pruned_loss=0.1411, over 5560.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.1882, pruned_loss=0.09521, over 1078171.16 frames. ], batch size: 46, lr: 1.80e-02, grad_scale: 8.0 +2022-11-15 19:01:03,739 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:02:11,321 INFO [train.py:876] (3/4) Epoch 5, batch 0, loss[loss=0.1253, simple_loss=0.144, pruned_loss=0.05332, over 5382.00 frames. ], tot_loss[loss=0.1253, simple_loss=0.144, pruned_loss=0.05332, over 5382.00 frames. ], batch size: 9, lr: 1.67e-02, grad_scale: 16.0 +2022-11-15 19:02:11,322 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 19:02:25,868 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2745, 1.0577, 0.7924, 0.9414, 0.9175, 1.2025, 0.8644, 0.7065], + device='cuda:3'), covar=tensor([0.0292, 0.0320, 0.0243, 0.0412, 0.0459, 0.0107, 0.0418, 0.0339], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0011, 0.0010, 0.0010, 0.0009, 0.0008, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5605e-05, 4.2313e-05, 3.7726e-05, 4.1032e-05, 3.6362e-05, 3.2453e-05, + 3.8988e-05, 3.6036e-05], device='cuda:3') +2022-11-15 19:02:28,899 INFO [train.py:908] (3/4) Epoch 5, validation: loss=0.1679, simple_loss=0.1892, pruned_loss=0.07329, over 1530663.00 frames. +2022-11-15 19:02:28,900 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 19:02:28,997 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7483, 3.3410, 3.5034, 3.2968, 3.7637, 3.6070, 3.4582, 3.7199], + device='cuda:3'), covar=tensor([0.0351, 0.0301, 0.0482, 0.0326, 0.0363, 0.0145, 0.0275, 0.0352], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0100, 0.0080, 0.0110, 0.0106, 0.0062, 0.0086, 0.0098], + 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') +2022-11-15 19:02:45,830 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 19:02:46,020 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.898e+02 2.233e+02 2.969e+02 5.666e+02, threshold=4.467e+02, percent-clipped=2.0 +2022-11-15 19:03:06,468 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:03:10,786 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:03:24,681 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3995, 1.4703, 1.0685, 1.3012, 1.5063, 1.3411, 1.3252, 1.0949], + device='cuda:3'), covar=tensor([0.0023, 0.0041, 0.0069, 0.0020, 0.0129, 0.0081, 0.0055, 0.0081], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0016, 0.0014, 0.0016, 0.0015, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.6146e-05, 1.7228e-05, 1.5507e-05, 1.7027e-05, 1.5111e-05, 1.8116e-05, + 1.8508e-05, 2.0274e-05], device='cuda:3') +2022-11-15 19:03:39,231 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2022-11-15 19:03:40,120 INFO [train.py:876] (3/4) Epoch 5, batch 100, loss[loss=0.1299, simple_loss=0.1447, pruned_loss=0.0576, over 5456.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.1875, pruned_loss=0.0956, over 431851.09 frames. ], batch size: 11, lr: 1.67e-02, grad_scale: 16.0 +2022-11-15 19:03:41,273 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:03:45,646 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:03:58,638 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.864e+02 2.257e+02 2.875e+02 4.325e+02, threshold=4.514e+02, percent-clipped=0.0 +2022-11-15 19:04:01,610 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:04:16,094 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:04:30,858 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2022-11-15 19:04:54,396 INFO [train.py:876] (3/4) Epoch 5, batch 200, loss[loss=0.1737, simple_loss=0.19, pruned_loss=0.0787, over 5789.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.1863, pruned_loss=0.0941, over 686977.07 frames. ], batch size: 21, lr: 1.66e-02, grad_scale: 16.0 +2022-11-15 19:05:00,111 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:05:04,967 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 +2022-11-15 19:05:11,288 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.852e+02 2.311e+02 2.896e+02 4.298e+02, threshold=4.622e+02, percent-clipped=0.0 +2022-11-15 19:05:43,278 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:05:50,274 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 +2022-11-15 19:05:55,556 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9770, 2.5698, 1.8801, 3.4202, 2.5866, 3.0531, 3.1148, 3.1036], + device='cuda:3'), covar=tensor([0.0692, 0.0763, 0.2077, 0.0524, 0.1790, 0.0698, 0.1812, 0.2963], + device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0057, 0.0073, 0.0046, 0.0064, 0.0053, 0.0066, 0.0046], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2022-11-15 19:06:04,996 INFO [train.py:876] (3/4) Epoch 5, batch 300, loss[loss=0.2433, simple_loss=0.2247, pruned_loss=0.131, over 5262.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.1855, pruned_loss=0.0931, over 843387.73 frames. ], batch size: 79, lr: 1.66e-02, grad_scale: 8.0 +2022-11-15 19:06:16,644 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5276, 2.0601, 1.7424, 1.7933, 1.0574, 1.6714, 1.3244, 1.8526], + device='cuda:3'), covar=tensor([0.0666, 0.0137, 0.0538, 0.0254, 0.0837, 0.0607, 0.1049, 0.0259], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0125, 0.0168, 0.0128, 0.0161, 0.0183, 0.0190, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:06:22,461 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.934e+02 2.477e+02 3.130e+02 8.486e+02, threshold=4.954e+02, percent-clipped=7.0 +2022-11-15 19:06:41,805 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9582, 2.1632, 3.4243, 2.5839, 4.1321, 2.1040, 3.4054, 3.8820], + device='cuda:3'), covar=tensor([0.0206, 0.1117, 0.0394, 0.1076, 0.0135, 0.0919, 0.0557, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0186, 0.0164, 0.0196, 0.0162, 0.0179, 0.0210, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 19:07:15,086 INFO [train.py:876] (3/4) Epoch 5, batch 400, loss[loss=0.1907, simple_loss=0.1973, pruned_loss=0.09206, over 5686.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.1858, pruned_loss=0.09228, over 943064.52 frames. ], batch size: 36, lr: 1.66e-02, grad_scale: 8.0 +2022-11-15 19:07:32,202 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 1.992e+02 2.391e+02 2.865e+02 5.865e+02, threshold=4.782e+02, percent-clipped=1.0 +2022-11-15 19:07:34,758 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:07:48,971 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 +2022-11-15 19:07:58,390 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:08:00,331 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3838, 4.3387, 4.5175, 4.6864, 3.9185, 3.5926, 5.1276, 4.3863], + device='cuda:3'), covar=tensor([0.0371, 0.0936, 0.0251, 0.0672, 0.0465, 0.0347, 0.0556, 0.0393], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0081, 0.0067, 0.0082, 0.0063, 0.0055, 0.0106, 0.0068], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:08:07,906 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:08:25,427 INFO [train.py:876] (3/4) Epoch 5, batch 500, loss[loss=0.2013, simple_loss=0.2212, pruned_loss=0.09069, over 5574.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.185, pruned_loss=0.0909, over 997745.89 frames. ], batch size: 25, lr: 1.66e-02, grad_scale: 8.0 +2022-11-15 19:08:27,528 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:08:31,240 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6041, 4.7357, 3.4022, 4.4843, 3.6228, 3.2673, 2.6800, 4.0252], + device='cuda:3'), covar=tensor([0.1511, 0.0220, 0.0887, 0.0212, 0.0406, 0.0823, 0.1796, 0.0187], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0120, 0.0163, 0.0121, 0.0154, 0.0178, 0.0184, 0.0126], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:08:36,166 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8707, 4.7488, 4.6879, 5.0223, 4.3311, 4.3021, 5.4007, 4.5130], + device='cuda:3'), covar=tensor([0.0327, 0.0590, 0.0266, 0.0619, 0.0530, 0.0241, 0.0533, 0.0422], + device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0081, 0.0068, 0.0083, 0.0064, 0.0055, 0.0106, 0.0069], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:08:41,586 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:08:42,744 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.299e+01 1.690e+02 2.176e+02 2.799e+02 4.262e+02, threshold=4.352e+02, percent-clipped=0.0 +2022-11-15 19:09:14,444 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:09:17,207 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:09:36,295 INFO [train.py:876] (3/4) Epoch 5, batch 600, loss[loss=0.2431, simple_loss=0.2174, pruned_loss=0.1344, over 4730.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.1862, pruned_loss=0.09172, over 1030758.85 frames. ], batch size: 135, lr: 1.65e-02, grad_scale: 8.0 +2022-11-15 19:09:44,374 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 +2022-11-15 19:09:48,480 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:09:53,665 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.408e+01 2.026e+02 2.597e+02 3.156e+02 6.029e+02, threshold=5.193e+02, percent-clipped=5.0 +2022-11-15 19:09:59,875 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 19:10:46,602 INFO [train.py:876] (3/4) Epoch 5, batch 700, loss[loss=0.1954, simple_loss=0.2027, pruned_loss=0.09407, over 5556.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.1872, pruned_loss=0.09258, over 1052241.37 frames. ], batch size: 25, lr: 1.65e-02, grad_scale: 8.0 +2022-11-15 19:10:56,332 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:11:04,298 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.007e+02 2.393e+02 2.831e+02 4.977e+02, threshold=4.787e+02, percent-clipped=0.0 +2022-11-15 19:11:07,538 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6763, 1.7022, 1.7325, 1.3676, 1.1201, 2.5763, 2.0690, 1.5225], + device='cuda:3'), covar=tensor([0.0492, 0.0669, 0.0579, 0.1121, 0.1542, 0.2166, 0.0570, 0.0647], + device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0040, 0.0042, 0.0046, 0.0039, 0.0036, 0.0033, 0.0041], + device='cuda:3'), out_proj_covar=tensor([7.4800e-05, 6.8109e-05, 7.0324e-05, 9.0544e-05, 7.2571e-05, 6.8978e-05, + 6.1643e-05, 7.2738e-05], device='cuda:3') +2022-11-15 19:11:17,214 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2022-11-15 19:11:39,263 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:11:57,077 INFO [train.py:876] (3/4) Epoch 5, batch 800, loss[loss=0.1935, simple_loss=0.1962, pruned_loss=0.09538, over 5583.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.1879, pruned_loss=0.09362, over 1061141.26 frames. ], batch size: 16, lr: 1.65e-02, grad_scale: 8.0 +2022-11-15 19:11:59,162 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:12:10,343 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:12:15,044 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.917e+02 2.325e+02 2.760e+02 6.321e+02, threshold=4.650e+02, percent-clipped=1.0 +2022-11-15 19:12:27,253 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8168, 4.6015, 3.4311, 2.1920, 4.4000, 1.6148, 4.1823, 2.6460], + device='cuda:3'), covar=tensor([0.1276, 0.0118, 0.0525, 0.2166, 0.0148, 0.2124, 0.0154, 0.1590], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0098, 0.0107, 0.0122, 0.0098, 0.0131, 0.0087, 0.0124], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 19:12:28,640 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3688, 1.3639, 1.4648, 1.5825, 1.3699, 1.8485, 1.3667, 1.2101], + device='cuda:3'), covar=tensor([0.0014, 0.0082, 0.0042, 0.0016, 0.0024, 0.0032, 0.0018, 0.0032], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0014, 0.0016, 0.0016, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.5439e-05, 1.7369e-05, 1.5044e-05, 1.7144e-05, 1.5874e-05, 1.7193e-05, + 1.8549e-05, 2.0342e-05], device='cuda:3') +2022-11-15 19:12:33,175 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:12:35,375 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1558, 0.9198, 1.0033, 0.8845, 1.1244, 1.3134, 0.8981, 0.8739], + device='cuda:3'), covar=tensor([0.0445, 0.0423, 0.1301, 0.1857, 0.1987, 0.0540, 0.0861, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0012, 0.0010, 0.0011, 0.0009, 0.0008, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([3.7114e-05, 4.3610e-05, 3.9597e-05, 4.3773e-05, 3.7488e-05, 3.4625e-05, + 4.1111e-05, 3.8769e-05], device='cuda:3') +2022-11-15 19:12:50,871 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3454, 3.1619, 3.3485, 1.4409, 2.8580, 3.4501, 3.3362, 3.6936], + device='cuda:3'), covar=tensor([0.1403, 0.0847, 0.0442, 0.1894, 0.0180, 0.0218, 0.0174, 0.0300], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0189, 0.0135, 0.0190, 0.0140, 0.0138, 0.0128, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 19:12:54,844 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4668, 4.6749, 3.4288, 4.3352, 3.4349, 3.1449, 2.6350, 3.8834], + device='cuda:3'), covar=tensor([0.1351, 0.0126, 0.0648, 0.0223, 0.0470, 0.0834, 0.1579, 0.0180], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0119, 0.0164, 0.0121, 0.0156, 0.0178, 0.0185, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:12:56,196 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1795, 0.8787, 0.9783, 0.8598, 1.1946, 1.5583, 0.8278, 0.8032], + device='cuda:3'), covar=tensor([0.0502, 0.0500, 0.0591, 0.3101, 0.0907, 0.0421, 0.1088, 0.1483], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0012, 0.0010, 0.0011, 0.0009, 0.0008, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([3.7669e-05, 4.3853e-05, 3.9646e-05, 4.4112e-05, 3.8178e-05, 3.4804e-05, + 4.1609e-05, 3.9479e-05], device='cuda:3') +2022-11-15 19:13:07,972 INFO [train.py:876] (3/4) Epoch 5, batch 900, loss[loss=0.1947, simple_loss=0.185, pruned_loss=0.1022, over 5612.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.1875, pruned_loss=0.09335, over 1070296.17 frames. ], batch size: 23, lr: 1.65e-02, grad_scale: 8.0 +2022-11-15 19:13:22,150 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-15 19:13:29,502 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.538e+01 1.888e+02 2.305e+02 2.835e+02 5.650e+02, threshold=4.611e+02, percent-clipped=1.0 +2022-11-15 19:13:32,277 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 19:13:37,718 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0789, 3.5936, 3.1270, 3.5943, 3.5304, 3.0706, 3.2106, 2.9993], + device='cuda:3'), covar=tensor([0.0871, 0.0400, 0.1219, 0.0371, 0.0467, 0.0452, 0.0472, 0.0607], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0137, 0.0217, 0.0135, 0.0167, 0.0141, 0.0142, 0.0128], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:13:48,341 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2611, 0.7848, 1.2671, 1.0043, 1.5617, 1.5302, 1.0823, 1.0650], + device='cuda:3'), covar=tensor([0.0898, 0.0555, 0.0525, 0.1460, 0.1269, 0.0686, 0.1331, 0.0688], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0011, 0.0009, 0.0010, 0.0009, 0.0008, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5777e-05, 4.1914e-05, 3.7582e-05, 4.1913e-05, 3.6251e-05, 3.3193e-05, + 3.9306e-05, 3.6775e-05], device='cuda:3') +2022-11-15 19:14:22,091 INFO [train.py:876] (3/4) Epoch 5, batch 1000, loss[loss=0.1899, simple_loss=0.191, pruned_loss=0.09447, over 5738.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.1879, pruned_loss=0.09292, over 1078247.53 frames. ], batch size: 27, lr: 1.64e-02, grad_scale: 8.0 +2022-11-15 19:14:23,968 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5280, 1.3950, 1.7218, 1.4510, 1.3324, 1.6572, 1.1705, 1.0492], + device='cuda:3'), covar=tensor([0.0010, 0.0038, 0.0011, 0.0016, 0.0026, 0.0034, 0.0018, 0.0035], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0014, 0.0014, 0.0016, 0.0015, 0.0015, 0.0017, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.5294e-05, 1.6107e-05, 1.4660e-05, 1.6867e-05, 1.5143e-05, 1.6486e-05, + 1.7901e-05, 1.9029e-05], device='cuda:3') +2022-11-15 19:14:39,368 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.933e+01 1.915e+02 2.282e+02 2.914e+02 7.246e+02, threshold=4.564e+02, percent-clipped=3.0 +2022-11-15 19:14:47,778 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:15:10,221 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:15:30,417 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:15:32,269 INFO [train.py:876] (3/4) Epoch 5, batch 1100, loss[loss=0.2154, simple_loss=0.2075, pruned_loss=0.1117, over 5452.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.1862, pruned_loss=0.09145, over 1081548.83 frames. ], batch size: 58, lr: 1.64e-02, grad_scale: 8.0 +2022-11-15 19:15:34,705 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 19:15:45,337 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:15:46,762 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:15:49,988 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 1.933e+02 2.265e+02 2.887e+02 4.331e+02, threshold=4.530e+02, percent-clipped=0.0 +2022-11-15 19:16:19,263 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:16:30,107 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:16:43,128 INFO [train.py:876] (3/4) Epoch 5, batch 1200, loss[loss=0.1977, simple_loss=0.2061, pruned_loss=0.09469, over 5474.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.1862, pruned_loss=0.09172, over 1086978.10 frames. ], batch size: 49, lr: 1.64e-02, grad_scale: 8.0 +2022-11-15 19:16:51,405 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0022, 3.7856, 3.7937, 3.5508, 3.8964, 3.7837, 1.4528, 4.1076], + device='cuda:3'), covar=tensor([0.0326, 0.0428, 0.0221, 0.0333, 0.0399, 0.0422, 0.3355, 0.0388], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0069, 0.0069, 0.0059, 0.0087, 0.0070, 0.0124, 0.0094], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:17:00,684 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.969e+02 2.334e+02 2.925e+02 5.027e+02, threshold=4.667e+02, percent-clipped=3.0 +2022-11-15 19:17:03,522 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:17:15,056 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.38 vs. limit=5.0 +2022-11-15 19:17:33,182 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:17:37,208 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:17:52,806 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.46 vs. limit=5.0 +2022-11-15 19:17:52,997 INFO [train.py:876] (3/4) Epoch 5, batch 1300, loss[loss=0.2242, simple_loss=0.2252, pruned_loss=0.1116, over 5543.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.1872, pruned_loss=0.09281, over 1082869.55 frames. ], batch size: 46, lr: 1.63e-02, grad_scale: 8.0 +2022-11-15 19:18:10,523 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 1.960e+02 2.546e+02 3.212e+02 9.234e+02, threshold=5.093e+02, percent-clipped=6.0 +2022-11-15 19:18:15,967 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:18:42,365 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:18:58,174 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:19:04,003 INFO [train.py:876] (3/4) Epoch 5, batch 1400, loss[loss=0.2286, simple_loss=0.2047, pruned_loss=0.1262, over 5146.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.1859, pruned_loss=0.09236, over 1075945.65 frames. ], batch size: 91, lr: 1.63e-02, grad_scale: 8.0 +2022-11-15 19:19:15,744 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:19:21,872 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.868e+02 2.246e+02 2.844e+02 4.562e+02, threshold=4.491e+02, percent-clipped=0.0 +2022-11-15 19:19:29,046 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3735, 4.5264, 4.5127, 4.8407, 3.9082, 3.2963, 5.2313, 4.6250], + device='cuda:3'), covar=tensor([0.0411, 0.0729, 0.0366, 0.0764, 0.0509, 0.0398, 0.0559, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0086, 0.0071, 0.0087, 0.0066, 0.0057, 0.0109, 0.0071], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:19:57,521 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:20:14,011 INFO [train.py:876] (3/4) Epoch 5, batch 1500, loss[loss=0.1865, simple_loss=0.1803, pruned_loss=0.09636, over 5008.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.1877, pruned_loss=0.09298, over 1082339.99 frames. ], batch size: 109, lr: 1.63e-02, grad_scale: 8.0 +2022-11-15 19:20:31,688 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.945e+02 2.500e+02 2.887e+02 6.503e+02, threshold=4.999e+02, percent-clipped=3.0 +2022-11-15 19:21:01,543 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:21:25,659 INFO [train.py:876] (3/4) Epoch 5, batch 1600, loss[loss=0.1802, simple_loss=0.1973, pruned_loss=0.08149, over 5577.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.1864, pruned_loss=0.09145, over 1089232.00 frames. ], batch size: 24, lr: 1.63e-02, grad_scale: 8.0 +2022-11-15 19:21:43,140 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 1.815e+02 2.150e+02 2.703e+02 4.613e+02, threshold=4.301e+02, percent-clipped=0.0 +2022-11-15 19:21:44,624 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:21:44,728 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:22:01,680 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2022-11-15 19:22:31,303 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:22:36,909 INFO [train.py:876] (3/4) Epoch 5, batch 1700, loss[loss=0.1728, simple_loss=0.1783, pruned_loss=0.08368, over 5589.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.1864, pruned_loss=0.0915, over 1085033.46 frames. ], batch size: 24, lr: 1.62e-02, grad_scale: 8.0 +2022-11-15 19:22:54,317 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.936e+02 2.464e+02 3.039e+02 4.896e+02, threshold=4.928e+02, percent-clipped=3.0 +2022-11-15 19:23:04,933 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:23:13,123 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5899, 1.3014, 1.2586, 1.0492, 1.2902, 1.1315, 0.7698, 1.5176], + device='cuda:3'), covar=tensor([0.0016, 0.0015, 0.0027, 0.0017, 0.0019, 0.0016, 0.0032, 0.0021], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0025, 0.0026, 0.0025, 0.0026, 0.0024, 0.0024, 0.0023], + device='cuda:3'), out_proj_covar=tensor([2.6703e-05, 2.7974e-05, 2.4565e-05, 2.3780e-05, 2.4029e-05, 1.9823e-05, + 2.8157e-05, 2.1741e-05], device='cuda:3') +2022-11-15 19:23:24,205 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.79 vs. limit=5.0 +2022-11-15 19:23:29,934 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:23:37,965 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 +2022-11-15 19:23:45,849 INFO [train.py:876] (3/4) Epoch 5, batch 1800, loss[loss=0.2162, simple_loss=0.2061, pruned_loss=0.1132, over 5733.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.1843, pruned_loss=0.08999, over 1089282.05 frames. ], batch size: 31, lr: 1.62e-02, grad_scale: 8.0 +2022-11-15 19:23:55,999 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:23:57,719 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2870, 4.1176, 3.4676, 3.3021, 2.3544, 3.8769, 2.3281, 3.4665], + device='cuda:3'), covar=tensor([0.0352, 0.0086, 0.0134, 0.0229, 0.0369, 0.0083, 0.0295, 0.0086], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0111, 0.0127, 0.0137, 0.0153, 0.0126, 0.0142, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:24:01,457 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0565, 4.4688, 3.7885, 4.3841, 4.4557, 3.6408, 3.6645, 3.5955], + device='cuda:3'), covar=tensor([0.0410, 0.0453, 0.1792, 0.0415, 0.0405, 0.0514, 0.0578, 0.0771], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0138, 0.0217, 0.0135, 0.0167, 0.0141, 0.0142, 0.0125], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:24:02,786 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:24:03,361 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.810e+02 2.217e+02 2.817e+02 4.516e+02, threshold=4.435e+02, percent-clipped=0.0 +2022-11-15 19:24:10,695 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:24:16,527 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:24:30,393 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6400, 2.6165, 2.4363, 2.7870, 2.6709, 2.3980, 3.0440, 2.6066], + device='cuda:3'), covar=tensor([0.0624, 0.0962, 0.0597, 0.0910, 0.0598, 0.0523, 0.0858, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0085, 0.0073, 0.0089, 0.0069, 0.0058, 0.0110, 0.0072], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:24:37,869 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:24:52,283 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:24:54,009 INFO [train.py:876] (3/4) Epoch 5, batch 1900, loss[loss=0.1766, simple_loss=0.168, pruned_loss=0.09264, over 5351.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.1838, pruned_loss=0.08972, over 1088840.77 frames. ], batch size: 79, lr: 1.62e-02, grad_scale: 8.0 +2022-11-15 19:24:58,426 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:25:09,014 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7483, 4.2013, 3.1088, 1.8640, 4.0368, 1.6148, 3.9354, 2.2094], + device='cuda:3'), covar=tensor([0.0982, 0.0109, 0.0510, 0.2052, 0.0146, 0.1744, 0.0135, 0.1590], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0097, 0.0108, 0.0122, 0.0100, 0.0130, 0.0086, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 19:25:10,260 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:25:10,325 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 19:25:12,009 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.895e+02 2.325e+02 2.903e+02 5.344e+02, threshold=4.651e+02, percent-clipped=5.0 +2022-11-15 19:25:13,458 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:25:14,462 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 19:25:46,275 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:25:48,487 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2022-11-15 19:25:51,709 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 19:26:02,813 INFO [train.py:876] (3/4) Epoch 5, batch 2000, loss[loss=0.1432, simple_loss=0.1646, pruned_loss=0.06084, over 5745.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.1845, pruned_loss=0.0906, over 1086418.76 frames. ], batch size: 15, lr: 1.62e-02, grad_scale: 8.0 +2022-11-15 19:26:20,066 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.737e+02 2.220e+02 2.843e+02 5.773e+02, threshold=4.440e+02, percent-clipped=3.0 +2022-11-15 19:26:22,746 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4611, 4.0853, 4.3308, 4.1285, 4.5050, 4.2158, 4.1572, 4.5523], + device='cuda:3'), covar=tensor([0.0367, 0.0234, 0.0358, 0.0277, 0.0343, 0.0270, 0.0207, 0.0235], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0107, 0.0083, 0.0114, 0.0112, 0.0068, 0.0089, 0.0103], + 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') +2022-11-15 19:26:53,742 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0366, 2.2815, 3.6402, 3.0882, 4.0709, 2.6965, 3.7449, 4.0958], + device='cuda:3'), covar=tensor([0.0201, 0.1119, 0.0321, 0.0846, 0.0196, 0.0824, 0.0507, 0.0333], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0187, 0.0173, 0.0199, 0.0169, 0.0179, 0.0215, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 19:27:10,092 INFO [train.py:876] (3/4) Epoch 5, batch 2100, loss[loss=0.1951, simple_loss=0.1778, pruned_loss=0.1062, over 4109.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.1855, pruned_loss=0.0908, over 1089677.20 frames. ], batch size: 181, lr: 1.61e-02, grad_scale: 8.0 +2022-11-15 19:27:26,925 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 1.972e+02 2.588e+02 3.306e+02 8.013e+02, threshold=5.176e+02, percent-clipped=4.0 +2022-11-15 19:27:51,776 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6765, 2.2564, 3.2372, 2.7738, 3.4624, 2.3656, 3.1708, 3.6099], + device='cuda:3'), covar=tensor([0.0278, 0.0766, 0.0409, 0.0846, 0.0265, 0.0793, 0.0636, 0.0403], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0185, 0.0173, 0.0202, 0.0172, 0.0181, 0.0217, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 19:27:58,201 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:28:11,160 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 +2022-11-15 19:28:13,296 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:28:18,979 INFO [train.py:876] (3/4) Epoch 5, batch 2200, loss[loss=0.1899, simple_loss=0.1979, pruned_loss=0.091, over 5540.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.1847, pruned_loss=0.08936, over 1087572.89 frames. ], batch size: 21, lr: 1.61e-02, grad_scale: 8.0 +2022-11-15 19:28:19,740 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:28:33,500 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:28:35,400 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.016e+02 2.479e+02 3.242e+02 5.312e+02, threshold=4.958e+02, percent-clipped=2.0 +2022-11-15 19:29:06,521 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:29:11,634 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:29:26,896 INFO [train.py:876] (3/4) Epoch 5, batch 2300, loss[loss=0.2339, simple_loss=0.2237, pruned_loss=0.122, over 5594.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.1859, pruned_loss=0.09074, over 1089969.44 frames. ], batch size: 18, lr: 1.61e-02, grad_scale: 16.0 +2022-11-15 19:29:41,114 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:29:44,268 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.900e+02 2.299e+02 2.871e+02 4.466e+02, threshold=4.598e+02, percent-clipped=0.0 +2022-11-15 19:29:49,047 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3766, 0.8911, 1.2064, 0.8499, 1.0386, 1.2953, 0.9342, 0.9721], + device='cuda:3'), covar=tensor([0.0572, 0.0186, 0.0605, 0.0973, 0.0649, 0.0254, 0.0524, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0012, 0.0009, 0.0010, 0.0010, 0.0009, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([3.7895e-05, 4.5585e-05, 3.8479e-05, 4.3240e-05, 3.9645e-05, 3.6171e-05, + 4.1274e-05, 3.9654e-05], device='cuda:3') +2022-11-15 19:30:21,024 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 19:30:22,594 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:30:33,991 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6871, 3.6978, 3.5057, 3.7386, 3.7609, 3.7035, 1.4842, 3.7419], + device='cuda:3'), covar=tensor([0.0278, 0.0249, 0.0253, 0.0195, 0.0311, 0.0300, 0.2969, 0.0345], + device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0070, 0.0072, 0.0060, 0.0086, 0.0072, 0.0126, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:30:35,177 INFO [train.py:876] (3/4) Epoch 5, batch 2400, loss[loss=0.1501, simple_loss=0.1693, pruned_loss=0.06544, over 5709.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.1858, pruned_loss=0.09057, over 1088356.97 frames. ], batch size: 17, lr: 1.61e-02, grad_scale: 16.0 +2022-11-15 19:30:37,448 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:30:53,270 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.788e+02 2.240e+02 2.771e+02 4.453e+02, threshold=4.479e+02, percent-clipped=0.0 +2022-11-15 19:31:04,526 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2750, 1.1842, 1.6984, 1.2224, 1.1726, 1.8321, 1.1205, 0.9042], + device='cuda:3'), covar=tensor([0.0018, 0.0035, 0.0013, 0.0018, 0.0036, 0.0017, 0.0018, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0015, 0.0017, 0.0017, 0.0016, 0.0018, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.6111e-05, 1.6459e-05, 1.5241e-05, 1.7698e-05, 1.7155e-05, 1.6670e-05, + 1.9518e-05, 2.0046e-05], device='cuda:3') +2022-11-15 19:31:19,173 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 19:31:23,815 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2020, 2.0083, 2.2154, 2.9620, 3.0425, 2.3586, 1.9509, 3.1783], + device='cuda:3'), covar=tensor([0.0321, 0.1942, 0.1837, 0.1363, 0.0682, 0.2066, 0.1938, 0.0260], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0208, 0.0209, 0.0302, 0.0209, 0.0223, 0.0205, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 19:31:24,383 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:31:38,735 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:31:43,846 INFO [train.py:876] (3/4) Epoch 5, batch 2500, loss[loss=0.184, simple_loss=0.1925, pruned_loss=0.08777, over 5723.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.1855, pruned_loss=0.09088, over 1079807.56 frames. ], batch size: 20, lr: 1.60e-02, grad_scale: 16.0 +2022-11-15 19:31:44,348 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 19:31:44,622 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:31:56,592 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:32:01,409 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.811e+02 2.202e+02 2.747e+02 5.680e+02, threshold=4.404e+02, percent-clipped=5.0 +2022-11-15 19:32:11,676 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:32:17,656 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:32:20,762 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 19:32:28,622 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 +2022-11-15 19:32:37,447 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:32:47,289 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8617, 0.7943, 0.9348, 0.6196, 0.7529, 0.8471, 0.4894, 0.7148], + device='cuda:3'), covar=tensor([0.0287, 0.0383, 0.0244, 0.0277, 0.0288, 0.0166, 0.0560, 0.0222], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0011, 0.0010, 0.0009, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([3.9016e-05, 4.8680e-05, 4.0388e-05, 4.5964e-05, 4.1334e-05, 3.7925e-05, + 4.4393e-05, 4.1424e-05], device='cuda:3') +2022-11-15 19:32:52,308 INFO [train.py:876] (3/4) Epoch 5, batch 2600, loss[loss=0.1793, simple_loss=0.1827, pruned_loss=0.08796, over 5600.00 frames. ], tot_loss[loss=0.183, simple_loss=0.185, pruned_loss=0.09047, over 1080864.99 frames. ], batch size: 18, lr: 1.60e-02, grad_scale: 16.0 +2022-11-15 19:33:02,101 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:33:08,478 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.912e+02 2.361e+02 2.985e+02 5.385e+02, threshold=4.723e+02, percent-clipped=4.0 +2022-11-15 19:33:09,210 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 19:33:14,949 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2022-11-15 19:33:29,679 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1949, 0.8555, 1.0429, 0.8499, 1.0025, 1.5286, 0.7864, 0.8686], + device='cuda:3'), covar=tensor([0.0769, 0.0609, 0.0501, 0.1874, 0.1773, 0.0475, 0.2972, 0.0928], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0010, 0.0010, 0.0009, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([3.8497e-05, 4.8303e-05, 3.9797e-05, 4.4606e-05, 4.0303e-05, 3.7562e-05, + 4.3917e-05, 4.1039e-05], device='cuda:3') +2022-11-15 19:33:43,345 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:33:43,442 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:34:00,070 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:34:00,582 INFO [train.py:876] (3/4) Epoch 5, batch 2700, loss[loss=0.1588, simple_loss=0.1783, pruned_loss=0.06969, over 5626.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.183, pruned_loss=0.08864, over 1088300.07 frames. ], batch size: 32, lr: 1.60e-02, grad_scale: 16.0 +2022-11-15 19:34:17,214 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.996e+02 2.440e+02 3.153e+02 7.916e+02, threshold=4.880e+02, percent-clipped=8.0 +2022-11-15 19:34:23,145 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:34:40,077 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 19:34:41,489 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:00,779 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:05,488 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:08,454 INFO [train.py:876] (3/4) Epoch 5, batch 2800, loss[loss=0.2129, simple_loss=0.1981, pruned_loss=0.1138, over 5629.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.1815, pruned_loss=0.08672, over 1079842.07 frames. ], batch size: 29, lr: 1.60e-02, grad_scale: 16.0 +2022-11-15 19:35:25,010 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.889e+02 2.395e+02 3.111e+02 5.606e+02, threshold=4.789e+02, percent-clipped=5.0 +2022-11-15 19:35:25,855 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1709, 3.0930, 3.1438, 3.1519, 3.2947, 3.2509, 1.2797, 3.3953], + device='cuda:3'), covar=tensor([0.0320, 0.0332, 0.0295, 0.0217, 0.0312, 0.0290, 0.2933, 0.0275], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0072, 0.0073, 0.0062, 0.0089, 0.0074, 0.0128, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:35:25,914 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:32,439 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9594, 1.7093, 1.6863, 1.1677, 0.9638, 2.5439, 1.7654, 1.4857], + device='cuda:3'), covar=tensor([0.0399, 0.0422, 0.0370, 0.1320, 0.1790, 0.0353, 0.0719, 0.0544], + device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0038, 0.0042, 0.0050, 0.0042, 0.0034, 0.0037, 0.0040], + device='cuda:3'), out_proj_covar=tensor([7.9304e-05, 7.0235e-05, 7.4218e-05, 9.6566e-05, 7.8549e-05, 7.0415e-05, + 6.9318e-05, 7.4264e-05], device='cuda:3') +2022-11-15 19:35:38,919 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2022-11-15 19:35:42,285 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:50,331 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:35:58,660 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6685, 4.2964, 3.7029, 4.1939, 4.3315, 3.6099, 3.7662, 3.5076], + device='cuda:3'), covar=tensor([0.0481, 0.0418, 0.1763, 0.0439, 0.0413, 0.0436, 0.0473, 0.0578], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0136, 0.0217, 0.0135, 0.0161, 0.0139, 0.0145, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:36:05,963 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2022-11-15 19:36:07,074 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:36:15,568 INFO [train.py:876] (3/4) Epoch 5, batch 2900, loss[loss=0.2203, simple_loss=0.2086, pruned_loss=0.116, over 5565.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.1831, pruned_loss=0.08882, over 1078258.11 frames. ], batch size: 43, lr: 1.59e-02, grad_scale: 16.0 +2022-11-15 19:36:31,855 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:36:33,617 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 1.876e+02 2.342e+02 2.951e+02 6.735e+02, threshold=4.684e+02, percent-clipped=4.0 +2022-11-15 19:36:34,339 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1263, 5.0397, 5.2979, 5.4132, 4.7462, 4.3266, 5.8730, 4.8863], + device='cuda:3'), covar=tensor([0.0254, 0.0967, 0.0247, 0.0893, 0.0428, 0.0208, 0.0587, 0.0319], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0084, 0.0070, 0.0086, 0.0067, 0.0056, 0.0107, 0.0070], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:37:04,962 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:37:08,615 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:37:24,195 INFO [train.py:876] (3/4) Epoch 5, batch 3000, loss[loss=0.2027, simple_loss=0.2017, pruned_loss=0.1019, over 5669.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.183, pruned_loss=0.089, over 1081451.17 frames. ], batch size: 29, lr: 1.59e-02, grad_scale: 16.0 +2022-11-15 19:37:24,196 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 19:37:30,752 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7663, 2.0992, 1.7813, 2.5004, 1.8065, 1.4262, 1.8159, 2.3138], + device='cuda:3'), covar=tensor([0.0844, 0.1190, 0.1624, 0.0449, 0.1343, 0.1838, 0.1312, 0.0416], + device='cuda:3'), in_proj_covar=tensor([0.0051, 0.0057, 0.0076, 0.0048, 0.0063, 0.0053, 0.0067, 0.0048], + 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') +2022-11-15 19:37:32,017 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8459, 1.9255, 2.1045, 1.3203, 1.7296, 2.6592, 2.6646, 2.4427], + device='cuda:3'), covar=tensor([0.0726, 0.0587, 0.0928, 0.1034, 0.0464, 0.0306, 0.0123, 0.0395], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0183, 0.0138, 0.0187, 0.0140, 0.0141, 0.0127, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 19:37:33,085 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1632, 0.9875, 1.0495, 1.0318, 1.1967, 1.5493, 0.7905, 0.9118], + device='cuda:3'), covar=tensor([0.0577, 0.0862, 0.0979, 0.1088, 0.0633, 0.0939, 0.0709, 0.1287], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0012, 0.0009, 0.0010, 0.0009, 0.0008, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5990e-05, 4.5954e-05, 3.7282e-05, 4.1189e-05, 3.7728e-05, 3.4499e-05, + 4.0361e-05, 3.8347e-05], device='cuda:3') +2022-11-15 19:37:41,542 INFO [train.py:908] (3/4) Epoch 5, validation: loss=0.1632, simple_loss=0.186, pruned_loss=0.07021, over 1530663.00 frames. +2022-11-15 19:37:41,543 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 19:37:58,635 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:37:59,205 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 2.101e+02 2.644e+02 3.414e+02 5.284e+02, threshold=5.288e+02, percent-clipped=5.0 +2022-11-15 19:38:02,002 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7971, 1.5348, 1.7776, 0.9752, 0.8572, 2.4746, 1.7710, 1.1145], + device='cuda:3'), covar=tensor([0.0501, 0.0610, 0.0351, 0.1885, 0.2385, 0.0486, 0.0587, 0.0862], + device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0038, 0.0041, 0.0051, 0.0041, 0.0035, 0.0037, 0.0041], + device='cuda:3'), out_proj_covar=tensor([7.9626e-05, 7.0367e-05, 7.3008e-05, 9.8356e-05, 7.8596e-05, 7.1214e-05, + 6.9453e-05, 7.5379e-05], device='cuda:3') +2022-11-15 19:38:19,082 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:38:21,096 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 19:38:25,365 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1496, 4.4453, 2.7438, 4.0506, 3.1694, 2.8776, 2.2087, 3.8018], + device='cuda:3'), covar=tensor([0.1565, 0.0115, 0.0991, 0.0292, 0.0540, 0.0915, 0.1925, 0.0197], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0122, 0.0167, 0.0127, 0.0165, 0.0177, 0.0184, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:38:43,330 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:38:49,858 INFO [train.py:876] (3/4) Epoch 5, batch 3100, loss[loss=0.2073, simple_loss=0.2051, pruned_loss=0.1047, over 5591.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.1837, pruned_loss=0.08959, over 1086458.83 frames. ], batch size: 22, lr: 1.59e-02, grad_scale: 16.0 +2022-11-15 19:38:53,728 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 19:39:07,482 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 2.024e+02 2.459e+02 3.147e+02 6.395e+02, threshold=4.918e+02, percent-clipped=1.0 +2022-11-15 19:39:21,169 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:39:46,917 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:39:55,090 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 +2022-11-15 19:39:58,534 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 19:39:58,835 INFO [train.py:876] (3/4) Epoch 5, batch 3200, loss[loss=0.1923, simple_loss=0.204, pruned_loss=0.09028, over 5678.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.1837, pruned_loss=0.08963, over 1088671.03 frames. ], batch size: 19, lr: 1.59e-02, grad_scale: 16.0 +2022-11-15 19:40:07,413 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8807, 0.9641, 1.3846, 1.3026, 0.9478, 1.3426, 0.7435, 0.7423], + device='cuda:3'), covar=tensor([0.0014, 0.0029, 0.0018, 0.0016, 0.0023, 0.0029, 0.0020, 0.0029], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0014, 0.0015, 0.0016, 0.0016, 0.0015, 0.0018, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.5640e-05, 1.5748e-05, 1.5359e-05, 1.6399e-05, 1.6595e-05, 1.5371e-05, + 1.9318e-05, 1.8892e-05], device='cuda:3') +2022-11-15 19:40:10,372 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:40:16,204 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 1.788e+02 2.177e+02 2.789e+02 5.595e+02, threshold=4.355e+02, percent-clipped=3.0 +2022-11-15 19:40:19,382 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 19:40:26,447 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1920, 0.8552, 1.0852, 0.9066, 0.9569, 1.6281, 0.9030, 0.9271], + device='cuda:3'), covar=tensor([0.0976, 0.2743, 0.1330, 0.2598, 0.1936, 0.0846, 0.6718, 0.2241], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0012, 0.0009, 0.0010, 0.0009, 0.0008, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5305e-05, 4.5928e-05, 3.7465e-05, 4.1812e-05, 3.7737e-05, 3.4273e-05, + 4.0035e-05, 3.8156e-05], device='cuda:3') +2022-11-15 19:40:47,300 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:40:51,869 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5814, 2.7219, 2.6688, 2.7567, 2.7302, 2.6847, 1.0510, 2.6618], + device='cuda:3'), covar=tensor([0.0622, 0.0391, 0.0430, 0.0382, 0.0542, 0.0509, 0.3739, 0.0627], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0070, 0.0071, 0.0064, 0.0087, 0.0074, 0.0128, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:40:56,423 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7802, 4.2921, 3.6641, 4.2058, 4.2250, 3.4781, 3.7733, 3.5803], + device='cuda:3'), covar=tensor([0.0501, 0.0429, 0.1556, 0.0461, 0.0459, 0.0467, 0.0570, 0.0692], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0135, 0.0216, 0.0136, 0.0164, 0.0139, 0.0147, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:41:07,143 INFO [train.py:876] (3/4) Epoch 5, batch 3300, loss[loss=0.2383, simple_loss=0.2229, pruned_loss=0.1269, over 5454.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.1847, pruned_loss=0.09082, over 1078069.06 frames. ], batch size: 58, lr: 1.58e-02, grad_scale: 16.0 +2022-11-15 19:41:09,874 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8468, 1.8088, 2.1761, 2.5690, 2.8098, 2.1555, 1.8450, 2.9542], + device='cuda:3'), covar=tensor([0.0463, 0.2296, 0.1684, 0.1471, 0.0701, 0.2013, 0.1791, 0.0325], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0214, 0.0211, 0.0308, 0.0215, 0.0222, 0.0201, 0.0169], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 19:41:19,787 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:41:24,093 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 1.817e+02 2.244e+02 2.768e+02 6.144e+02, threshold=4.488e+02, percent-clipped=4.0 +2022-11-15 19:41:28,355 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.15 vs. limit=5.0 +2022-11-15 19:41:30,205 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6854, 0.9771, 1.3761, 1.3982, 0.7375, 1.5802, 1.1626, 0.9916], + device='cuda:3'), covar=tensor([0.0765, 0.1996, 0.1807, 0.1974, 0.4772, 0.0469, 0.6414, 0.2964], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0012, 0.0009, 0.0010, 0.0009, 0.0008, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5310e-05, 4.6297e-05, 3.8002e-05, 4.1989e-05, 3.8387e-05, 3.4733e-05, + 4.0204e-05, 3.8447e-05], device='cuda:3') +2022-11-15 19:41:32,884 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 +2022-11-15 19:41:44,495 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:42:08,051 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:42:15,161 INFO [train.py:876] (3/4) Epoch 5, batch 3400, loss[loss=0.1869, simple_loss=0.2028, pruned_loss=0.08549, over 5567.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.1838, pruned_loss=0.08935, over 1081346.40 frames. ], batch size: 30, lr: 1.58e-02, grad_scale: 16.0 +2022-11-15 19:42:17,570 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:42:32,056 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.882e+02 2.362e+02 2.947e+02 5.374e+02, threshold=4.725e+02, percent-clipped=3.0 +2022-11-15 19:42:40,958 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:42:46,051 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:43:07,640 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5395, 1.3824, 1.8569, 1.2489, 1.2727, 1.9575, 0.8463, 1.0943], + device='cuda:3'), covar=tensor([0.0012, 0.0023, 0.0029, 0.0018, 0.0022, 0.0036, 0.0017, 0.0022], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0014, 0.0014, 0.0015, 0.0015, 0.0014, 0.0017, 0.0015], + device='cuda:3'), out_proj_covar=tensor([1.4708e-05, 1.5284e-05, 1.4358e-05, 1.5350e-05, 1.5494e-05, 1.4468e-05, + 1.8047e-05, 1.7862e-05], device='cuda:3') +2022-11-15 19:43:10,884 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:43:18,440 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:43:23,549 INFO [train.py:876] (3/4) Epoch 5, batch 3500, loss[loss=0.1109, simple_loss=0.1207, pruned_loss=0.05054, over 5269.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.1819, pruned_loss=0.08766, over 1077607.72 frames. ], batch size: 8, lr: 1.58e-02, grad_scale: 16.0 +2022-11-15 19:43:35,714 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:43:40,884 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 1.935e+02 2.270e+02 2.895e+02 5.164e+02, threshold=4.540e+02, percent-clipped=2.0 +2022-11-15 19:43:44,158 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:43:54,799 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4702, 3.6723, 3.4164, 3.1996, 2.2339, 3.6640, 2.2556, 3.1391], + device='cuda:3'), covar=tensor([0.0261, 0.0100, 0.0112, 0.0196, 0.0318, 0.0079, 0.0249, 0.0063], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0116, 0.0131, 0.0143, 0.0159, 0.0128, 0.0146, 0.0113], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:44:08,843 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:44:17,160 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 +2022-11-15 19:44:26,369 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2022-11-15 19:44:32,334 INFO [train.py:876] (3/4) Epoch 5, batch 3600, loss[loss=0.1683, simple_loss=0.1741, pruned_loss=0.0812, over 5763.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.1831, pruned_loss=0.08904, over 1083562.13 frames. ], batch size: 16, lr: 1.58e-02, grad_scale: 16.0 +2022-11-15 19:44:49,899 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.827e+02 2.423e+02 3.084e+02 7.397e+02, threshold=4.846e+02, percent-clipped=5.0 +2022-11-15 19:45:40,991 INFO [train.py:876] (3/4) Epoch 5, batch 3700, loss[loss=0.187, simple_loss=0.1993, pruned_loss=0.08737, over 5627.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.1838, pruned_loss=0.08948, over 1082169.40 frames. ], batch size: 29, lr: 1.58e-02, grad_scale: 16.0 +2022-11-15 19:45:57,989 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.092e+02 2.533e+02 3.307e+02 5.477e+02, threshold=5.066e+02, percent-clipped=1.0 +2022-11-15 19:46:04,139 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7314, 1.1390, 1.5933, 0.8812, 1.1599, 1.5651, 1.1645, 0.7916], + device='cuda:3'), covar=tensor([0.1896, 0.1004, 0.0626, 0.2919, 0.4937, 0.1871, 0.1111, 0.3110], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0012, 0.0009, 0.0010, 0.0009, 0.0009, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.6303e-05, 4.7254e-05, 3.8125e-05, 4.3436e-05, 4.0007e-05, 3.6841e-05, + 4.1859e-05, 4.0234e-05], device='cuda:3') +2022-11-15 19:46:39,097 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2022-11-15 19:46:42,232 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5750, 1.9874, 3.1194, 2.4966, 3.4112, 2.0218, 3.0266, 3.5341], + device='cuda:3'), covar=tensor([0.0317, 0.1077, 0.0421, 0.0894, 0.0281, 0.0884, 0.0605, 0.0436], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0194, 0.0177, 0.0206, 0.0181, 0.0188, 0.0222, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 19:46:49,250 INFO [train.py:876] (3/4) Epoch 5, batch 3800, loss[loss=0.1486, simple_loss=0.162, pruned_loss=0.06762, over 5779.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.183, pruned_loss=0.08792, over 1081858.67 frames. ], batch size: 20, lr: 1.57e-02, grad_scale: 16.0 +2022-11-15 19:47:05,910 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 1.821e+02 2.317e+02 3.245e+02 5.660e+02, threshold=4.635e+02, percent-clipped=3.0 +2022-11-15 19:47:24,483 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3045, 1.6826, 1.2564, 0.7044, 1.4273, 1.0798, 1.0279, 0.8737], + device='cuda:3'), covar=tensor([0.0018, 0.0019, 0.0020, 0.0024, 0.0029, 0.0014, 0.0022, 0.0106], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0025, 0.0026, 0.0025, 0.0026, 0.0023, 0.0025, 0.0023], + device='cuda:3'), out_proj_covar=tensor([2.6520e-05, 2.7683e-05, 2.3724e-05, 2.3355e-05, 2.4227e-05, 1.8660e-05, + 2.8747e-05, 2.2030e-05], device='cuda:3') +2022-11-15 19:47:42,297 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:47:56,112 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6925, 3.9129, 3.9096, 1.4865, 3.6320, 4.1066, 3.9179, 4.3409], + device='cuda:3'), covar=tensor([0.1541, 0.1003, 0.0327, 0.2571, 0.0189, 0.0334, 0.0208, 0.0237], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0189, 0.0145, 0.0203, 0.0150, 0.0151, 0.0136, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 19:47:57,890 INFO [train.py:876] (3/4) Epoch 5, batch 3900, loss[loss=0.1614, simple_loss=0.1744, pruned_loss=0.07416, over 5492.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.1824, pruned_loss=0.08738, over 1085219.19 frames. ], batch size: 17, lr: 1.57e-02, grad_scale: 16.0 +2022-11-15 19:48:14,592 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0237, 4.3801, 3.7550, 4.4192, 4.3270, 3.6522, 3.8227, 3.3130], + device='cuda:3'), covar=tensor([0.0374, 0.0368, 0.1414, 0.0402, 0.0431, 0.0461, 0.0407, 0.0875], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0137, 0.0217, 0.0137, 0.0167, 0.0139, 0.0147, 0.0129], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:48:15,131 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.831e+02 2.324e+02 2.912e+02 5.041e+02, threshold=4.648e+02, percent-clipped=1.0 +2022-11-15 19:48:19,143 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 19:48:24,290 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:48:52,743 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:49:06,339 INFO [train.py:876] (3/4) Epoch 5, batch 4000, loss[loss=0.2032, simple_loss=0.192, pruned_loss=0.1071, over 4696.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.1808, pruned_loss=0.0859, over 1086485.14 frames. ], batch size: 135, lr: 1.57e-02, grad_scale: 16.0 +2022-11-15 19:49:23,826 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 1.884e+02 2.400e+02 2.913e+02 6.279e+02, threshold=4.801e+02, percent-clipped=5.0 +2022-11-15 19:49:23,993 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2952, 5.4362, 4.0124, 4.9984, 4.1801, 4.0582, 3.2574, 4.6991], + device='cuda:3'), covar=tensor([0.1128, 0.0090, 0.0586, 0.0207, 0.0266, 0.0553, 0.1398, 0.0135], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0127, 0.0166, 0.0130, 0.0160, 0.0179, 0.0182, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:49:26,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9990, 3.4550, 4.0612, 3.7284, 3.8491, 3.6110, 1.4588, 3.9772], + device='cuda:3'), covar=tensor([0.0332, 0.0692, 0.0216, 0.0271, 0.0437, 0.0603, 0.3357, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0070, 0.0072, 0.0063, 0.0087, 0.0074, 0.0128, 0.0096], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:49:34,162 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:49:51,328 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:50:08,168 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5036, 0.8757, 2.1332, 1.6696, 1.2850, 2.0843, 1.5712, 1.4634], + device='cuda:3'), covar=tensor([0.0032, 0.0074, 0.0016, 0.0011, 0.0033, 0.0056, 0.0022, 0.0019], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0015, 0.0016, 0.0015, 0.0016, 0.0017, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.6226e-05, 1.6101e-05, 1.5399e-05, 1.5915e-05, 1.5679e-05, 1.6175e-05, + 1.8611e-05, 1.8300e-05], device='cuda:3') +2022-11-15 19:50:13,957 INFO [train.py:876] (3/4) Epoch 5, batch 4100, loss[loss=0.206, simple_loss=0.1858, pruned_loss=0.1131, over 4171.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.1811, pruned_loss=0.08679, over 1084393.82 frames. ], batch size: 181, lr: 1.57e-02, grad_scale: 8.0 +2022-11-15 19:50:32,026 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 1.879e+02 2.350e+02 3.001e+02 5.532e+02, threshold=4.700e+02, percent-clipped=2.0 +2022-11-15 19:50:32,851 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:51:22,796 INFO [train.py:876] (3/4) Epoch 5, batch 4200, loss[loss=0.2071, simple_loss=0.2055, pruned_loss=0.1044, over 5540.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.1823, pruned_loss=0.08735, over 1088532.92 frames. ], batch size: 40, lr: 1.56e-02, grad_scale: 8.0 +2022-11-15 19:51:24,483 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 19:51:36,075 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2283, 3.8572, 2.7884, 3.6850, 2.9662, 2.7314, 1.9684, 3.2185], + device='cuda:3'), covar=tensor([0.1428, 0.0208, 0.0918, 0.0289, 0.0606, 0.0944, 0.1946, 0.0294], + device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0133, 0.0172, 0.0133, 0.0163, 0.0187, 0.0189, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:51:40,446 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.845e+02 2.173e+02 2.647e+02 4.072e+02, threshold=4.345e+02, percent-clipped=0.0 +2022-11-15 19:51:45,060 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:51:55,266 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3534, 3.8817, 3.3779, 3.8411, 3.9081, 3.3208, 3.4511, 3.2167], + device='cuda:3'), covar=tensor([0.0644, 0.0393, 0.1498, 0.0423, 0.0344, 0.0382, 0.0403, 0.0528], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0139, 0.0219, 0.0136, 0.0166, 0.0141, 0.0145, 0.0129], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 19:52:30,460 INFO [train.py:876] (3/4) Epoch 5, batch 4300, loss[loss=0.1894, simple_loss=0.1882, pruned_loss=0.09529, over 5619.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.1815, pruned_loss=0.08652, over 1090363.94 frames. ], batch size: 38, lr: 1.56e-02, grad_scale: 8.0 +2022-11-15 19:52:49,001 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.372e+01 1.969e+02 2.435e+02 3.163e+02 9.091e+02, threshold=4.870e+02, percent-clipped=6.0 +2022-11-15 19:52:55,800 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:53:28,335 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:53:36,867 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0989, 4.3038, 2.6253, 3.9133, 3.5297, 2.9839, 2.4170, 3.5907], + device='cuda:3'), covar=tensor([0.2595, 0.0431, 0.2053, 0.0566, 0.0686, 0.1503, 0.2702, 0.0390], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0130, 0.0166, 0.0131, 0.0161, 0.0182, 0.0184, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:53:39,861 INFO [train.py:876] (3/4) Epoch 5, batch 4400, loss[loss=0.1899, simple_loss=0.1936, pruned_loss=0.0931, over 5721.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.1826, pruned_loss=0.08784, over 1084771.71 frames. ], batch size: 28, lr: 1.56e-02, grad_scale: 8.0 +2022-11-15 19:53:55,869 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:53:57,545 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 +2022-11-15 19:53:58,512 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.879e+02 2.441e+02 2.964e+02 5.680e+02, threshold=4.882e+02, percent-clipped=2.0 +2022-11-15 19:54:12,033 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:54:25,090 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4030, 2.8013, 3.0903, 1.2214, 3.1223, 3.3862, 3.3609, 3.7117], + device='cuda:3'), covar=tensor([0.1573, 0.1466, 0.0669, 0.2662, 0.0243, 0.0346, 0.0219, 0.0327], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0186, 0.0143, 0.0196, 0.0145, 0.0149, 0.0136, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 19:54:38,375 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1334, 4.1535, 2.8370, 4.0503, 3.3420, 2.8038, 2.2877, 3.3952], + device='cuda:3'), covar=tensor([0.1881, 0.0227, 0.1075, 0.0260, 0.0534, 0.1248, 0.2113, 0.0402], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0130, 0.0168, 0.0130, 0.0163, 0.0184, 0.0184, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:54:50,516 INFO [train.py:876] (3/4) Epoch 5, batch 4500, loss[loss=0.1059, simple_loss=0.1235, pruned_loss=0.04416, over 5229.00 frames. ], tot_loss[loss=0.177, simple_loss=0.1815, pruned_loss=0.08626, over 1088604.87 frames. ], batch size: 8, lr: 1.56e-02, grad_scale: 8.0 +2022-11-15 19:55:08,211 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.892e+02 2.378e+02 2.959e+02 6.563e+02, threshold=4.756e+02, percent-clipped=3.0 +2022-11-15 19:55:13,239 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:55:22,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4961, 0.9402, 1.1705, 0.8971, 1.4577, 1.2921, 1.0913, 0.9238], + device='cuda:3'), covar=tensor([0.0617, 0.0798, 0.1436, 0.2020, 0.1345, 0.1347, 0.0932, 0.2190], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0011, 0.0009, 0.0010, 0.0009, 0.0009, 0.0010, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5791e-05, 4.6086e-05, 3.8188e-05, 4.2258e-05, 4.0316e-05, 3.6488e-05, + 3.9524e-05, 3.9585e-05], device='cuda:3') +2022-11-15 19:55:45,891 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:55:50,867 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:55:51,457 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4478, 4.0408, 4.0769, 4.0951, 4.5159, 4.3266, 4.0170, 4.3730], + device='cuda:3'), covar=tensor([0.0533, 0.0714, 0.0759, 0.0744, 0.0549, 0.0342, 0.0581, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0109, 0.0082, 0.0113, 0.0113, 0.0068, 0.0094, 0.0104], + 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') +2022-11-15 19:55:58,825 INFO [train.py:876] (3/4) Epoch 5, batch 4600, loss[loss=0.1546, simple_loss=0.1712, pruned_loss=0.06901, over 5670.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.1833, pruned_loss=0.0885, over 1082055.07 frames. ], batch size: 32, lr: 1.55e-02, grad_scale: 8.0 +2022-11-15 19:56:15,738 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5701, 1.0088, 1.4405, 0.9056, 1.2362, 1.1120, 0.7545, 1.0606], + device='cuda:3'), covar=tensor([0.0129, 0.0257, 0.0183, 0.0669, 0.0551, 0.0951, 0.0595, 0.0416], + device='cuda:3'), in_proj_covar=tensor([0.0008, 0.0011, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.5463e-05, 4.5746e-05, 3.6724e-05, 4.1532e-05, 3.9356e-05, 3.6143e-05, + 3.8565e-05, 3.8606e-05], device='cuda:3') +2022-11-15 19:56:16,176 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.825e+02 2.230e+02 2.892e+02 8.047e+02, threshold=4.459e+02, percent-clipped=4.0 +2022-11-15 19:56:23,214 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:56:32,052 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 19:56:50,307 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9612, 4.2513, 3.9144, 4.3111, 3.6641, 3.1886, 4.8295, 3.8051], + device='cuda:3'), covar=tensor([0.0457, 0.0732, 0.0499, 0.0765, 0.0508, 0.0468, 0.0603, 0.0554], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0087, 0.0072, 0.0088, 0.0069, 0.0058, 0.0111, 0.0072], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:56:55,509 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:56:58,588 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4409, 2.0406, 1.6788, 1.7736, 1.1353, 1.6539, 1.3711, 1.7953], + device='cuda:3'), covar=tensor([0.0495, 0.0154, 0.0599, 0.0312, 0.0920, 0.0587, 0.0886, 0.0303], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0129, 0.0168, 0.0131, 0.0163, 0.0179, 0.0183, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:57:02,636 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8544, 2.5107, 2.8004, 3.8288, 4.1064, 2.8845, 2.4712, 3.9714], + device='cuda:3'), covar=tensor([0.0242, 0.3537, 0.2328, 0.3766, 0.0666, 0.3448, 0.2197, 0.0258], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0209, 0.0212, 0.0314, 0.0212, 0.0223, 0.0199, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 19:57:06,420 INFO [train.py:876] (3/4) Epoch 5, batch 4700, loss[loss=0.2186, simple_loss=0.203, pruned_loss=0.1171, over 5456.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.1832, pruned_loss=0.08826, over 1086796.97 frames. ], batch size: 49, lr: 1.55e-02, grad_scale: 8.0 +2022-11-15 19:57:08,983 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9697, 1.2031, 1.2683, 1.2800, 0.9985, 1.1630, 0.9727, 1.0833], + device='cuda:3'), covar=tensor([0.0016, 0.0017, 0.0019, 0.0016, 0.0018, 0.0018, 0.0016, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0016, 0.0016, 0.0016, 0.0016, 0.0017, 0.0016], + device='cuda:3'), out_proj_covar=tensor([1.6373e-05, 1.6082e-05, 1.6215e-05, 1.6758e-05, 1.6270e-05, 1.6508e-05, + 1.8242e-05, 1.8767e-05], device='cuda:3') +2022-11-15 19:57:22,557 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:57:25,038 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.755e+02 2.231e+02 2.801e+02 4.827e+02, threshold=4.463e+02, percent-clipped=3.0 +2022-11-15 19:57:33,507 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:57:55,052 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:58:06,185 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.43 vs. limit=2.0 +2022-11-15 19:58:15,453 INFO [train.py:876] (3/4) Epoch 5, batch 4800, loss[loss=0.2317, simple_loss=0.1935, pruned_loss=0.1349, over 4129.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.1826, pruned_loss=0.08902, over 1068771.38 frames. ], batch size: 181, lr: 1.55e-02, grad_scale: 8.0 +2022-11-15 19:58:23,861 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8996, 2.1389, 2.7463, 3.7896, 3.9763, 3.1125, 2.2964, 3.7271], + device='cuda:3'), covar=tensor([0.0278, 0.3788, 0.2413, 0.3236, 0.0891, 0.2659, 0.2230, 0.0363], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0212, 0.0216, 0.0317, 0.0213, 0.0224, 0.0203, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 19:58:33,232 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.864e+02 2.250e+02 2.859e+02 4.870e+02, threshold=4.500e+02, percent-clipped=2.0 +2022-11-15 19:58:49,527 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:59:12,981 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7316, 4.8195, 4.8818, 5.0663, 4.6530, 3.7345, 5.6194, 4.6913], + device='cuda:3'), covar=tensor([0.0357, 0.0713, 0.0256, 0.0833, 0.0478, 0.0342, 0.0529, 0.0325], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0090, 0.0074, 0.0089, 0.0070, 0.0060, 0.0115, 0.0074], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 19:59:23,238 INFO [train.py:876] (3/4) Epoch 5, batch 4900, loss[loss=0.151, simple_loss=0.1693, pruned_loss=0.06632, over 5499.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.182, pruned_loss=0.08821, over 1071493.11 frames. ], batch size: 12, lr: 1.55e-02, grad_scale: 8.0 +2022-11-15 19:59:31,329 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 19:59:36,291 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0018, 2.5446, 2.1275, 1.2603, 2.5329, 2.6671, 2.8138, 3.0301], + device='cuda:3'), covar=tensor([0.1352, 0.1228, 0.0799, 0.2109, 0.0282, 0.0460, 0.0196, 0.0481], + device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0179, 0.0141, 0.0187, 0.0143, 0.0148, 0.0134, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 19:59:41,366 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 1.963e+02 2.433e+02 3.223e+02 8.796e+02, threshold=4.867e+02, percent-clipped=10.0 +2022-11-15 19:59:44,589 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4544, 2.7697, 3.8413, 3.3265, 4.5096, 3.4591, 4.1894, 4.5867], + device='cuda:3'), covar=tensor([0.0198, 0.0976, 0.0377, 0.1039, 0.0172, 0.0692, 0.0544, 0.0256], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0188, 0.0177, 0.0206, 0.0176, 0.0183, 0.0220, 0.0200], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 19:59:53,938 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:00:01,251 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:00:18,160 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6953, 1.9228, 1.8802, 1.7329, 1.8615, 1.8946, 0.8727, 1.9197], + device='cuda:3'), covar=tensor([0.0357, 0.0217, 0.0215, 0.0238, 0.0323, 0.0243, 0.1781, 0.0272], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0071, 0.0072, 0.0063, 0.0089, 0.0075, 0.0127, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:00:26,528 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8512, 2.0075, 1.5525, 2.2552, 1.3563, 1.4226, 1.5878, 2.1197], + device='cuda:3'), covar=tensor([0.0980, 0.2234, 0.3168, 0.0944, 0.2496, 0.1884, 0.2321, 0.1067], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0064, 0.0076, 0.0050, 0.0064, 0.0057, 0.0069, 0.0048], + 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') +2022-11-15 20:00:32,065 INFO [train.py:876] (3/4) Epoch 5, batch 5000, loss[loss=0.241, simple_loss=0.2177, pruned_loss=0.1321, over 5360.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.1803, pruned_loss=0.0864, over 1074873.40 frames. ], batch size: 70, lr: 1.55e-02, grad_scale: 8.0 +2022-11-15 20:00:35,659 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.10 vs. limit=5.0 +2022-11-15 20:00:41,320 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4297, 0.8558, 1.5763, 1.4600, 1.2144, 1.6267, 1.2487, 1.0026], + device='cuda:3'), covar=tensor([0.0030, 0.0072, 0.0033, 0.0023, 0.0028, 0.0047, 0.0020, 0.0035], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0016, 0.0018, 0.0017, 0.0017, 0.0018, 0.0018], + device='cuda:3'), out_proj_covar=tensor([1.7600e-05, 1.7769e-05, 1.6686e-05, 1.7761e-05, 1.7505e-05, 1.7465e-05, + 1.9702e-05, 2.0553e-05], device='cuda:3') +2022-11-15 20:00:42,678 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:00:49,471 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.808e+02 2.328e+02 2.773e+02 5.652e+02, threshold=4.656e+02, percent-clipped=1.0 +2022-11-15 20:00:57,875 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2022-11-15 20:00:58,876 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:01:10,315 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:01:31,151 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:01:40,503 INFO [train.py:876] (3/4) Epoch 5, batch 5100, loss[loss=0.2354, simple_loss=0.2105, pruned_loss=0.1302, over 5557.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.18, pruned_loss=0.08606, over 1072742.85 frames. ], batch size: 43, lr: 1.54e-02, grad_scale: 8.0 +2022-11-15 20:01:52,692 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:01:58,447 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.276e+02 1.924e+02 2.234e+02 2.951e+02 5.133e+02, threshold=4.468e+02, percent-clipped=1.0 +2022-11-15 20:02:13,069 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5359, 4.6190, 3.9985, 4.1939, 4.4049, 4.3138, 1.9418, 4.6330], + device='cuda:3'), covar=tensor([0.0293, 0.0180, 0.0484, 0.0216, 0.0409, 0.0379, 0.2783, 0.0352], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0071, 0.0073, 0.0063, 0.0090, 0.0075, 0.0128, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:02:49,026 INFO [train.py:876] (3/4) Epoch 5, batch 5200, loss[loss=0.2042, simple_loss=0.2002, pruned_loss=0.1041, over 5703.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.1803, pruned_loss=0.08568, over 1079156.06 frames. ], batch size: 28, lr: 1.54e-02, grad_scale: 8.0 +2022-11-15 20:02:53,492 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:03:07,059 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.891e+02 2.372e+02 3.198e+02 5.762e+02, threshold=4.744e+02, percent-clipped=5.0 +2022-11-15 20:03:19,808 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:03:41,005 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0008, 2.7580, 3.2072, 1.6624, 2.8122, 3.0405, 3.1802, 3.8749], + device='cuda:3'), covar=tensor([0.2133, 0.1596, 0.0738, 0.2649, 0.0322, 0.0637, 0.0266, 0.0363], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0186, 0.0143, 0.0192, 0.0147, 0.0148, 0.0133, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 20:03:52,329 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:03:57,991 INFO [train.py:876] (3/4) Epoch 5, batch 5300, loss[loss=0.1228, simple_loss=0.1485, pruned_loss=0.04857, over 5465.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.18, pruned_loss=0.08544, over 1081860.84 frames. ], batch size: 12, lr: 1.54e-02, grad_scale: 8.0 +2022-11-15 20:03:58,491 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 +2022-11-15 20:04:05,524 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:04:11,171 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1467, 3.3087, 3.2640, 3.0470, 3.2178, 3.2369, 1.1016, 3.3353], + device='cuda:3'), covar=tensor([0.0337, 0.0214, 0.0250, 0.0256, 0.0331, 0.0284, 0.2997, 0.0313], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0071, 0.0074, 0.0063, 0.0091, 0.0076, 0.0129, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:04:15,583 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.782e+02 2.113e+02 2.792e+02 4.181e+02, threshold=4.226e+02, percent-clipped=0.0 +2022-11-15 20:05:00,431 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0107, 2.1390, 3.6841, 2.9437, 4.0242, 2.7515, 3.5330, 3.9709], + device='cuda:3'), covar=tensor([0.0313, 0.1336, 0.0428, 0.1221, 0.0272, 0.0891, 0.0642, 0.0449], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0191, 0.0184, 0.0206, 0.0176, 0.0187, 0.0223, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:05:06,440 INFO [train.py:876] (3/4) Epoch 5, batch 5400, loss[loss=0.1303, simple_loss=0.1474, pruned_loss=0.05663, over 5204.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.1801, pruned_loss=0.08522, over 1083519.27 frames. ], batch size: 8, lr: 1.54e-02, grad_scale: 8.0 +2022-11-15 20:05:06,823 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 20:05:14,778 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:05:17,398 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3555, 4.1823, 3.3609, 1.6107, 4.0906, 1.2618, 3.8992, 2.1385], + device='cuda:3'), covar=tensor([0.1576, 0.0161, 0.0467, 0.2703, 0.0179, 0.2379, 0.0249, 0.2008], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0101, 0.0110, 0.0121, 0.0104, 0.0133, 0.0094, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:05:24,228 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.783e+02 2.368e+02 3.183e+02 6.760e+02, threshold=4.736e+02, percent-clipped=8.0 +2022-11-15 20:05:25,058 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1597, 3.2660, 2.6069, 1.5833, 3.2090, 1.1769, 3.1233, 1.7028], + device='cuda:3'), covar=tensor([0.1233, 0.0177, 0.0698, 0.1894, 0.0215, 0.2064, 0.0219, 0.1615], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0100, 0.0109, 0.0119, 0.0103, 0.0132, 0.0094, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:05:30,266 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8526, 3.2143, 2.2238, 3.1012, 2.0460, 2.4000, 1.7770, 2.8087], + device='cuda:3'), covar=tensor([0.1172, 0.0174, 0.0951, 0.0241, 0.0930, 0.0890, 0.1535, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0129, 0.0168, 0.0129, 0.0163, 0.0180, 0.0182, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:06:14,516 INFO [train.py:876] (3/4) Epoch 5, batch 5500, loss[loss=0.2149, simple_loss=0.2035, pruned_loss=0.1131, over 5020.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.1814, pruned_loss=0.08617, over 1082070.60 frames. ], batch size: 109, lr: 1.53e-02, grad_scale: 8.0 +2022-11-15 20:06:17,332 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:06:18,646 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:06:21,923 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 20:06:32,588 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.877e+02 2.418e+02 2.886e+02 5.617e+02, threshold=4.837e+02, percent-clipped=2.0 +2022-11-15 20:06:36,704 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 20:06:37,691 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:06:40,315 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:06:51,544 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:06:59,159 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:07:00,790 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.6686, 5.2332, 5.5604, 4.9846, 5.8531, 5.7368, 5.0130, 5.6689], + device='cuda:3'), covar=tensor([0.0390, 0.0223, 0.0389, 0.0253, 0.0239, 0.0079, 0.0192, 0.0179], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0112, 0.0085, 0.0116, 0.0115, 0.0067, 0.0095, 0.0109], + 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') +2022-11-15 20:07:09,379 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2022-11-15 20:07:19,517 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 20:07:22,498 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:07:23,629 INFO [train.py:876] (3/4) Epoch 5, batch 5600, loss[loss=0.2234, simple_loss=0.2199, pruned_loss=0.1134, over 5746.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.1823, pruned_loss=0.08756, over 1084634.06 frames. ], batch size: 27, lr: 1.53e-02, grad_scale: 8.0 +2022-11-15 20:07:31,131 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:07:37,941 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:07:41,445 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.835e+02 2.169e+02 2.808e+02 5.282e+02, threshold=4.338e+02, percent-clipped=2.0 +2022-11-15 20:08:03,965 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:08:20,064 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:08:32,090 INFO [train.py:876] (3/4) Epoch 5, batch 5700, loss[loss=0.1971, simple_loss=0.1841, pruned_loss=0.1051, over 5594.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.1814, pruned_loss=0.08677, over 1082491.84 frames. ], batch size: 50, lr: 1.53e-02, grad_scale: 8.0 +2022-11-15 20:08:36,852 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1943, 1.4520, 1.3631, 0.9923, 1.0327, 1.0369, 1.1199, 0.8193], + device='cuda:3'), covar=tensor([0.0016, 0.0013, 0.0012, 0.0017, 0.0016, 0.0016, 0.0018, 0.0022], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0016, 0.0018, 0.0017, 0.0016, 0.0018, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.7632e-05, 1.6675e-05, 1.6038e-05, 1.7648e-05, 1.7659e-05, 1.6899e-05, + 1.9064e-05, 1.9159e-05], device='cuda:3') +2022-11-15 20:08:40,373 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:08:49,695 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.820e+02 2.216e+02 2.818e+02 4.619e+02, threshold=4.433e+02, percent-clipped=3.0 +2022-11-15 20:09:01,850 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7622, 1.9508, 2.1061, 2.7477, 2.7923, 2.0440, 1.5894, 3.0302], + device='cuda:3'), covar=tensor([0.0538, 0.2166, 0.1914, 0.1263, 0.0813, 0.2376, 0.1956, 0.0390], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0213, 0.0213, 0.0318, 0.0215, 0.0225, 0.0200, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:09:13,162 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:09:25,018 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-15 20:09:30,743 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 +2022-11-15 20:09:40,485 INFO [train.py:876] (3/4) Epoch 5, batch 5800, loss[loss=0.1942, simple_loss=0.1995, pruned_loss=0.0944, over 5687.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.1815, pruned_loss=0.08694, over 1078709.04 frames. ], batch size: 36, lr: 1.53e-02, grad_scale: 8.0 +2022-11-15 20:09:41,952 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5645, 1.2628, 1.2274, 0.8749, 0.8039, 1.9678, 1.4184, 1.0599], + device='cuda:3'), covar=tensor([0.0739, 0.1091, 0.0762, 0.1913, 0.2408, 0.0926, 0.1159, 0.1099], + device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0044, 0.0046, 0.0055, 0.0045, 0.0040, 0.0040, 0.0043], + device='cuda:3'), out_proj_covar=tensor([9.3090e-05, 8.3388e-05, 8.6143e-05, 1.0844e-04, 8.8702e-05, 8.2248e-05, + 7.9162e-05, 8.3359e-05], device='cuda:3') +2022-11-15 20:09:58,512 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.871e+02 2.247e+02 2.954e+02 6.973e+02, threshold=4.493e+02, percent-clipped=4.0 +2022-11-15 20:10:04,114 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 20:10:21,823 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:10:37,161 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8275, 1.4812, 1.7415, 0.9760, 1.2668, 2.5106, 1.9287, 1.3440], + device='cuda:3'), covar=tensor([0.0493, 0.0703, 0.0866, 0.1657, 0.2179, 0.0373, 0.1646, 0.0850], + device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0044, 0.0046, 0.0055, 0.0046, 0.0040, 0.0040, 0.0043], + device='cuda:3'), out_proj_covar=tensor([9.4817e-05, 8.4502e-05, 8.6689e-05, 1.0915e-04, 9.0447e-05, 8.2291e-05, + 8.0129e-05, 8.4079e-05], device='cuda:3') +2022-11-15 20:10:40,007 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.10 vs. limit=2.0 +2022-11-15 20:10:40,993 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:10:44,257 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:10:48,843 INFO [train.py:876] (3/4) Epoch 5, batch 5900, loss[loss=0.1332, simple_loss=0.1581, pruned_loss=0.05417, over 5785.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.1808, pruned_loss=0.0864, over 1077158.62 frames. ], batch size: 20, lr: 1.53e-02, grad_scale: 8.0 +2022-11-15 20:11:09,452 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 2.067e+02 2.510e+02 3.048e+02 6.634e+02, threshold=5.021e+02, percent-clipped=2.0 +2022-11-15 20:11:44,845 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 20:12:00,905 INFO [train.py:876] (3/4) Epoch 5, batch 6000, loss[loss=0.1853, simple_loss=0.1905, pruned_loss=0.09006, over 5555.00 frames. ], tot_loss[loss=0.175, simple_loss=0.1804, pruned_loss=0.08483, over 1080817.75 frames. ], batch size: 14, lr: 1.52e-02, grad_scale: 8.0 +2022-11-15 20:12:00,906 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 20:12:05,496 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6662, 1.7359, 1.5476, 1.2878, 0.6973, 2.5620, 1.6640, 1.2512], + device='cuda:3'), covar=tensor([0.0334, 0.0646, 0.0571, 0.1621, 0.4608, 0.0541, 0.1053, 0.1987], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0042, 0.0043, 0.0052, 0.0045, 0.0037, 0.0038, 0.0042], + device='cuda:3'), out_proj_covar=tensor([9.0082e-05, 8.0102e-05, 8.2148e-05, 1.0365e-04, 8.7394e-05, 7.7846e-05, + 7.6009e-05, 8.0931e-05], device='cuda:3') +2022-11-15 20:12:18,599 INFO [train.py:908] (3/4) Epoch 5, validation: loss=0.1648, simple_loss=0.1864, pruned_loss=0.07158, over 1530663.00 frames. +2022-11-15 20:12:18,600 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 20:12:20,797 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:12:21,423 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1574, 1.3933, 1.2671, 0.9766, 0.5203, 1.9974, 1.3162, 0.9161], + device='cuda:3'), covar=tensor([0.0868, 0.0568, 0.0857, 0.1424, 0.2019, 0.0311, 0.0759, 0.1416], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0041, 0.0043, 0.0052, 0.0044, 0.0037, 0.0038, 0.0041], + device='cuda:3'), out_proj_covar=tensor([8.9565e-05, 7.9562e-05, 8.1739e-05, 1.0305e-04, 8.6777e-05, 7.7511e-05, + 7.5564e-05, 8.0496e-05], device='cuda:3') +2022-11-15 20:12:32,310 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2586, 4.9340, 4.3265, 4.8759, 4.8657, 4.0428, 4.3203, 4.1169], + device='cuda:3'), covar=tensor([0.0287, 0.0347, 0.1444, 0.0377, 0.0365, 0.0408, 0.0472, 0.0473], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0139, 0.0220, 0.0137, 0.0169, 0.0143, 0.0151, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 20:12:35,733 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7201, 1.8591, 2.4951, 2.2901, 2.4925, 1.7324, 2.2816, 2.7626], + device='cuda:3'), covar=tensor([0.0280, 0.0797, 0.0427, 0.0632, 0.0346, 0.0739, 0.0525, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0186, 0.0180, 0.0206, 0.0174, 0.0185, 0.0221, 0.0201], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:12:36,105 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.732e+01 1.810e+02 2.246e+02 2.914e+02 5.187e+02, threshold=4.493e+02, percent-clipped=1.0 +2022-11-15 20:13:02,219 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4512, 1.5021, 1.5680, 1.0167, 0.6409, 2.2360, 1.7716, 1.1403], + device='cuda:3'), covar=tensor([0.0666, 0.1028, 0.0655, 0.1994, 0.3536, 0.0866, 0.1286, 0.1120], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0041, 0.0043, 0.0052, 0.0043, 0.0038, 0.0037, 0.0041], + device='cuda:3'), out_proj_covar=tensor([8.8835e-05, 7.9136e-05, 8.1583e-05, 1.0289e-04, 8.5302e-05, 7.8547e-05, + 7.4699e-05, 8.0139e-05], device='cuda:3') +2022-11-15 20:13:02,271 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:13:10,632 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4377, 1.5174, 1.5679, 1.2137, 1.2383, 1.5040, 1.2155, 0.9647], + device='cuda:3'), covar=tensor([0.0022, 0.0066, 0.0025, 0.0023, 0.0033, 0.0029, 0.0021, 0.0032], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0016, 0.0016, 0.0018, 0.0017, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.6475e-05, 1.6537e-05, 1.5810e-05, 1.7895e-05, 1.6910e-05, 1.6903e-05, + 1.8094e-05, 1.9460e-05], device='cuda:3') +2022-11-15 20:13:24,811 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4126, 3.6270, 3.7404, 1.6960, 3.5454, 3.8321, 3.5534, 4.1215], + device='cuda:3'), covar=tensor([0.1704, 0.1048, 0.0555, 0.2312, 0.0268, 0.0235, 0.0263, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0182, 0.0143, 0.0196, 0.0152, 0.0148, 0.0137, 0.0175], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 20:13:26,555 INFO [train.py:876] (3/4) Epoch 5, batch 6100, loss[loss=0.2036, simple_loss=0.1972, pruned_loss=0.105, over 4914.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.1797, pruned_loss=0.08455, over 1080513.42 frames. ], batch size: 109, lr: 1.52e-02, grad_scale: 16.0 +2022-11-15 20:13:44,499 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.904e+02 2.292e+02 2.845e+02 6.036e+02, threshold=4.585e+02, percent-clipped=4.0 +2022-11-15 20:14:07,830 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:14:20,304 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2702, 3.1467, 3.6160, 1.4532, 3.1192, 3.6742, 3.4696, 3.6828], + device='cuda:3'), covar=tensor([0.1834, 0.1210, 0.0466, 0.2571, 0.0341, 0.0220, 0.0342, 0.0548], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0183, 0.0140, 0.0193, 0.0151, 0.0148, 0.0136, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 20:14:28,470 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:14:31,240 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:14:36,327 INFO [train.py:876] (3/4) Epoch 5, batch 6200, loss[loss=0.2291, simple_loss=0.2224, pruned_loss=0.1179, over 5646.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.1805, pruned_loss=0.08547, over 1086902.73 frames. ], batch size: 32, lr: 1.52e-02, grad_scale: 16.0 +2022-11-15 20:14:36,480 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7622, 4.4637, 3.3339, 1.9873, 3.9928, 1.6074, 4.3317, 2.1807], + device='cuda:3'), covar=tensor([0.1221, 0.0125, 0.0447, 0.2268, 0.0211, 0.1994, 0.0127, 0.1821], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0102, 0.0108, 0.0119, 0.0103, 0.0131, 0.0092, 0.0120], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:14:41,930 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:14:54,868 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0 +2022-11-15 20:14:55,052 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.804e+02 2.260e+02 2.687e+02 5.215e+02, threshold=4.521e+02, percent-clipped=1.0 +2022-11-15 20:15:02,902 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:15:05,750 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:15:16,314 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:15:30,939 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:15:36,813 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:15:47,103 INFO [train.py:876] (3/4) Epoch 5, batch 6300, loss[loss=0.205, simple_loss=0.2118, pruned_loss=0.09915, over 5579.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.1784, pruned_loss=0.08347, over 1085151.57 frames. ], batch size: 22, lr: 1.52e-02, grad_scale: 16.0 +2022-11-15 20:15:53,248 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8872, 2.2594, 3.5660, 3.1621, 3.9920, 2.4752, 3.4090, 3.9554], + device='cuda:3'), covar=tensor([0.0322, 0.1245, 0.0415, 0.1082, 0.0303, 0.1033, 0.0694, 0.0408], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0186, 0.0180, 0.0207, 0.0174, 0.0183, 0.0219, 0.0200], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:15:59,513 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 20:16:04,773 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:16:05,294 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.779e+02 2.287e+02 3.039e+02 6.063e+02, threshold=4.574e+02, percent-clipped=3.0 +2022-11-15 20:16:12,645 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 20:16:20,134 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:16:28,634 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:16:50,280 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4006, 0.5799, 1.4081, 0.8189, 1.6580, 1.5288, 0.8864, 1.2584], + device='cuda:3'), covar=tensor([0.0799, 0.0601, 0.0327, 0.2437, 0.0750, 0.0560, 0.1397, 0.0760], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0011, 0.0010, 0.0010, 0.0011, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.0338e-05, 5.2872e-05, 4.2085e-05, 4.7929e-05, 4.4340e-05, 4.2108e-05, + 4.5713e-05, 4.3670e-05], device='cuda:3') +2022-11-15 20:16:57,613 INFO [train.py:876] (3/4) Epoch 5, batch 6400, loss[loss=0.1725, simple_loss=0.1763, pruned_loss=0.08432, over 5602.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.1803, pruned_loss=0.08521, over 1086701.05 frames. ], batch size: 18, lr: 1.52e-02, grad_scale: 16.0 +2022-11-15 20:17:00,349 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9229, 2.3635, 1.6782, 1.0969, 1.0223, 2.6633, 1.5597, 1.4943], + device='cuda:3'), covar=tensor([0.0477, 0.0411, 0.0600, 0.1487, 0.1963, 0.0398, 0.1722, 0.0844], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0040, 0.0042, 0.0051, 0.0042, 0.0036, 0.0038, 0.0039], + device='cuda:3'), out_proj_covar=tensor([8.7345e-05, 7.7755e-05, 8.1005e-05, 1.0141e-04, 8.4323e-05, 7.6356e-05, + 7.6352e-05, 7.8184e-05], device='cuda:3') +2022-11-15 20:17:10,253 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2022-11-15 20:17:14,788 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 1.875e+02 2.323e+02 3.297e+02 5.699e+02, threshold=4.646e+02, percent-clipped=4.0 +2022-11-15 20:17:20,525 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8171, 2.1793, 3.6032, 2.8197, 4.0227, 2.5242, 3.2761, 3.9687], + device='cuda:3'), covar=tensor([0.0426, 0.1287, 0.0512, 0.1326, 0.0224, 0.1085, 0.1045, 0.0485], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0187, 0.0181, 0.0206, 0.0176, 0.0186, 0.0222, 0.0200], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:17:25,838 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:18:05,885 INFO [train.py:876] (3/4) Epoch 5, batch 6500, loss[loss=0.275, simple_loss=0.2353, pruned_loss=0.1574, over 3139.00 frames. ], tot_loss[loss=0.177, simple_loss=0.1812, pruned_loss=0.08635, over 1086754.74 frames. ], batch size: 284, lr: 1.51e-02, grad_scale: 16.0 +2022-11-15 20:18:07,316 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:18:12,862 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 +2022-11-15 20:18:23,695 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 1.920e+02 2.418e+02 3.178e+02 5.825e+02, threshold=4.835e+02, percent-clipped=5.0 +2022-11-15 20:19:10,169 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7905, 3.0135, 3.2771, 1.1880, 2.9379, 3.2613, 3.2626, 3.4577], + device='cuda:3'), covar=tensor([0.2650, 0.1504, 0.0563, 0.3640, 0.0351, 0.0429, 0.0323, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0183, 0.0140, 0.0191, 0.0150, 0.0147, 0.0136, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 20:19:12,814 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-15 20:19:14,314 INFO [train.py:876] (3/4) Epoch 5, batch 6600, loss[loss=0.1396, simple_loss=0.1662, pruned_loss=0.05651, over 5596.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.181, pruned_loss=0.08563, over 1088541.30 frames. ], batch size: 24, lr: 1.51e-02, grad_scale: 16.0 +2022-11-15 20:19:17,847 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5604, 1.6703, 1.8357, 2.4171, 2.5601, 1.9267, 1.6203, 2.7869], + device='cuda:3'), covar=tensor([0.0645, 0.2706, 0.1771, 0.1325, 0.0670, 0.2356, 0.1981, 0.0493], + device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0212, 0.0207, 0.0313, 0.0215, 0.0220, 0.0200, 0.0175], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:19:22,588 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:19:31,774 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 1.753e+02 2.153e+02 2.756e+02 5.336e+02, threshold=4.306e+02, percent-clipped=2.0 +2022-11-15 20:19:39,445 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2367, 4.9039, 4.2924, 4.8798, 4.8916, 4.2139, 4.2188, 4.2198], + device='cuda:3'), covar=tensor([0.0314, 0.0367, 0.1331, 0.0358, 0.0346, 0.0350, 0.0411, 0.0448], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0139, 0.0224, 0.0138, 0.0171, 0.0144, 0.0152, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 20:19:42,129 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:19:54,537 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:20:09,984 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:20:16,150 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7163, 4.3600, 3.4178, 1.8197, 4.1473, 1.5641, 4.1542, 2.2944], + device='cuda:3'), covar=tensor([0.1196, 0.0142, 0.0525, 0.2353, 0.0195, 0.1899, 0.0147, 0.1603], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0103, 0.0107, 0.0120, 0.0104, 0.0131, 0.0092, 0.0120], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:20:22,005 INFO [train.py:876] (3/4) Epoch 5, batch 6700, loss[loss=0.2164, simple_loss=0.2104, pruned_loss=0.1112, over 5497.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.1796, pruned_loss=0.08454, over 1087815.11 frames. ], batch size: 49, lr: 1.51e-02, grad_scale: 16.0 +2022-11-15 20:20:27,353 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:20:38,042 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:20:40,467 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.856e+02 2.372e+02 2.960e+02 5.756e+02, threshold=4.743e+02, percent-clipped=4.0 +2022-11-15 20:20:42,554 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4109, 1.5713, 1.3472, 1.2530, 1.3562, 1.3923, 1.3349, 1.1780], + device='cuda:3'), covar=tensor([0.0019, 0.0038, 0.0037, 0.0020, 0.0020, 0.0028, 0.0018, 0.0034], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0016, 0.0016, 0.0018, 0.0016, 0.0016, 0.0018, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.6383e-05, 1.6588e-05, 1.6062e-05, 1.7665e-05, 1.6360e-05, 1.6368e-05, + 1.8307e-05, 1.9861e-05], device='cuda:3') +2022-11-15 20:20:52,207 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:21:19,384 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:21:28,461 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:21:30,772 INFO [train.py:876] (3/4) Epoch 5, batch 6800, loss[loss=0.2395, simple_loss=0.2177, pruned_loss=0.1307, over 5383.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.1801, pruned_loss=0.08489, over 1090908.71 frames. ], batch size: 70, lr: 1.51e-02, grad_scale: 16.0 +2022-11-15 20:21:40,654 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-15 20:21:42,336 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5164, 4.8041, 5.3304, 4.8000, 5.5897, 5.3769, 4.8275, 5.4161], + device='cuda:3'), covar=tensor([0.0280, 0.0221, 0.0329, 0.0283, 0.0276, 0.0090, 0.0158, 0.0203], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0113, 0.0087, 0.0113, 0.0120, 0.0070, 0.0097, 0.0109], + 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') +2022-11-15 20:21:48,382 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.967e+02 2.535e+02 3.123e+02 6.625e+02, threshold=5.070e+02, percent-clipped=2.0 +2022-11-15 20:22:38,598 INFO [train.py:876] (3/4) Epoch 5, batch 6900, loss[loss=0.1676, simple_loss=0.1767, pruned_loss=0.07924, over 5702.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.1811, pruned_loss=0.08624, over 1088824.20 frames. ], batch size: 34, lr: 1.51e-02, grad_scale: 16.0 +2022-11-15 20:22:46,848 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 20:22:56,605 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.741e+02 2.226e+02 2.702e+02 5.830e+02, threshold=4.452e+02, percent-clipped=1.0 +2022-11-15 20:23:07,224 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:23:11,261 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2625, 1.3067, 1.5303, 1.2171, 1.2962, 1.1220, 1.3495, 1.0687], + device='cuda:3'), covar=tensor([0.0024, 0.0036, 0.0034, 0.0022, 0.0049, 0.0041, 0.0020, 0.0030], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0015, 0.0017, 0.0016, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.5992e-05, 1.5536e-05, 1.5013e-05, 1.7082e-05, 1.5818e-05, 1.6162e-05, + 1.7325e-05, 1.9366e-05], device='cuda:3') +2022-11-15 20:23:19,732 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:23:21,577 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 20:23:26,684 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3140, 1.3980, 1.5836, 0.9728, 0.5730, 2.1983, 1.3262, 1.1639], + device='cuda:3'), covar=tensor([0.0620, 0.0792, 0.0648, 0.1871, 0.2193, 0.0497, 0.1353, 0.0776], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0040, 0.0042, 0.0051, 0.0044, 0.0036, 0.0038, 0.0040], + device='cuda:3'), out_proj_covar=tensor([8.9038e-05, 7.8589e-05, 8.1222e-05, 1.0166e-04, 8.7686e-05, 7.6216e-05, + 7.7547e-05, 7.8993e-05], device='cuda:3') +2022-11-15 20:23:37,984 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1657, 3.2857, 3.0741, 3.0696, 1.9580, 3.1503, 2.0044, 2.8533], + device='cuda:3'), covar=tensor([0.0298, 0.0136, 0.0129, 0.0177, 0.0322, 0.0094, 0.0306, 0.0083], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0127, 0.0137, 0.0149, 0.0164, 0.0136, 0.0152, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:23:39,889 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:23:47,621 INFO [train.py:876] (3/4) Epoch 5, batch 7000, loss[loss=0.1765, simple_loss=0.1809, pruned_loss=0.08603, over 5577.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.1802, pruned_loss=0.08505, over 1085269.06 frames. ], batch size: 43, lr: 1.50e-02, grad_scale: 16.0 +2022-11-15 20:24:05,016 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.861e+02 2.290e+02 2.878e+02 5.762e+02, threshold=4.579e+02, percent-clipped=5.0 +2022-11-15 20:24:05,108 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1061, 3.9612, 3.8579, 4.1884, 3.7612, 3.3842, 4.6429, 3.8684], + device='cuda:3'), covar=tensor([0.0472, 0.1070, 0.0497, 0.0942, 0.0687, 0.0403, 0.0796, 0.0562], + device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0088, 0.0074, 0.0091, 0.0070, 0.0059, 0.0115, 0.0075], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:24:05,808 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9812, 2.3858, 1.8087, 2.6731, 1.6522, 1.8926, 2.0301, 2.8539], + device='cuda:3'), covar=tensor([0.1077, 0.1205, 0.3182, 0.0655, 0.1957, 0.0875, 0.1979, 0.2964], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0066, 0.0084, 0.0054, 0.0068, 0.0061, 0.0075, 0.0054], + 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') +2022-11-15 20:24:13,664 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:24:21,487 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 20:24:40,741 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:24:51,026 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:24:54,729 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:24:56,577 INFO [train.py:876] (3/4) Epoch 5, batch 7100, loss[loss=0.2533, simple_loss=0.2111, pruned_loss=0.1478, over 3005.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.1795, pruned_loss=0.08485, over 1082934.72 frames. ], batch size: 284, lr: 1.50e-02, grad_scale: 16.0 +2022-11-15 20:25:08,251 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4740, 2.6676, 1.9546, 3.0804, 2.0164, 2.9440, 2.8046, 3.2671], + device='cuda:3'), covar=tensor([0.1060, 0.1917, 0.3312, 0.0847, 0.2028, 0.1456, 0.1867, 0.2150], + device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0068, 0.0088, 0.0056, 0.0071, 0.0063, 0.0078, 0.0057], + 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') +2022-11-15 20:25:14,390 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.219e+01 1.811e+02 2.272e+02 2.779e+02 4.389e+02, threshold=4.544e+02, percent-clipped=0.0 +2022-11-15 20:25:27,659 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:25:33,401 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:25:37,334 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9302, 1.1054, 1.1003, 0.9105, 0.9871, 0.9781, 0.9102, 0.5877], + device='cuda:3'), covar=tensor([0.0024, 0.0025, 0.0036, 0.0031, 0.0023, 0.0030, 0.0024, 0.0040], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0016, 0.0018, 0.0016, 0.0016, 0.0017, 0.0017], + device='cuda:3'), out_proj_covar=tensor([1.6118e-05, 1.6086e-05, 1.5780e-05, 1.7532e-05, 1.5699e-05, 1.6771e-05, + 1.7216e-05, 1.9816e-05], device='cuda:3') +2022-11-15 20:25:39,892 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3693, 4.5223, 3.9644, 3.8283, 4.2172, 4.1549, 1.6368, 4.3820], + device='cuda:3'), covar=tensor([0.0301, 0.0190, 0.0370, 0.0424, 0.0409, 0.0292, 0.3218, 0.0315], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0073, 0.0073, 0.0063, 0.0089, 0.0078, 0.0128, 0.0096], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:25:55,188 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2429, 1.2351, 1.2438, 0.9619, 0.5699, 2.0630, 1.5439, 1.0769], + device='cuda:3'), covar=tensor([0.0595, 0.0921, 0.0702, 0.1550, 0.2316, 0.0826, 0.0898, 0.0903], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0041, 0.0042, 0.0052, 0.0045, 0.0036, 0.0039, 0.0041], + device='cuda:3'), out_proj_covar=tensor([8.9681e-05, 8.0860e-05, 8.2803e-05, 1.0379e-04, 8.9697e-05, 7.6723e-05, + 7.7979e-05, 8.1417e-05], device='cuda:3') +2022-11-15 20:26:05,724 INFO [train.py:876] (3/4) Epoch 5, batch 7200, loss[loss=0.1886, simple_loss=0.1893, pruned_loss=0.09396, over 5545.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.1785, pruned_loss=0.08393, over 1080789.54 frames. ], batch size: 46, lr: 1.50e-02, grad_scale: 16.0 +2022-11-15 20:26:11,106 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8449, 4.3657, 4.7519, 4.3492, 4.8958, 4.8105, 4.2809, 4.8902], + device='cuda:3'), covar=tensor([0.0386, 0.0303, 0.0361, 0.0314, 0.0397, 0.0119, 0.0255, 0.0292], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0112, 0.0087, 0.0111, 0.0121, 0.0070, 0.0097, 0.0108], + 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') +2022-11-15 20:26:16,845 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9398, 5.4543, 4.4317, 5.1746, 5.3527, 4.6288, 5.1513, 4.9636], + device='cuda:3'), covar=tensor([0.0212, 0.0420, 0.1809, 0.0730, 0.0453, 0.0364, 0.0299, 0.0466], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0141, 0.0229, 0.0140, 0.0174, 0.0145, 0.0155, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 20:26:22,592 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.732e+01 1.770e+02 2.199e+02 2.603e+02 4.829e+02, threshold=4.399e+02, percent-clipped=1.0 +2022-11-15 20:26:43,168 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3145, 0.9059, 1.8480, 1.1501, 1.0730, 1.2642, 1.5415, 0.7502], + device='cuda:3'), covar=tensor([0.0030, 0.0052, 0.0043, 0.0028, 0.0073, 0.0064, 0.0020, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0015, 0.0016, 0.0018, 0.0016, 0.0016, 0.0017, 0.0018], + device='cuda:3'), out_proj_covar=tensor([1.6295e-05, 1.6129e-05, 1.5886e-05, 1.7632e-05, 1.5906e-05, 1.6753e-05, + 1.7515e-05, 2.0334e-05], device='cuda:3') +2022-11-15 20:26:46,009 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.09 vs. limit=5.0 +2022-11-15 20:27:38,810 INFO [train.py:876] (3/4) Epoch 6, batch 0, loss[loss=0.1445, simple_loss=0.1616, pruned_loss=0.06365, over 5558.00 frames. ], tot_loss[loss=0.1445, simple_loss=0.1616, pruned_loss=0.06365, over 5558.00 frames. ], batch size: 25, lr: 1.40e-02, grad_scale: 16.0 +2022-11-15 20:27:38,810 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 20:27:55,407 INFO [train.py:908] (3/4) Epoch 6, validation: loss=0.1637, simple_loss=0.1861, pruned_loss=0.07065, over 1530663.00 frames. +2022-11-15 20:27:55,407 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 20:28:10,989 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2245, 4.0263, 4.0771, 4.0871, 4.1039, 3.6130, 1.6843, 4.4260], + device='cuda:3'), covar=tensor([0.0262, 0.0395, 0.0235, 0.0204, 0.0302, 0.0427, 0.2685, 0.0290], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0075, 0.0075, 0.0064, 0.0091, 0.0078, 0.0130, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:28:31,820 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.895e+02 2.226e+02 2.646e+02 4.624e+02, threshold=4.452e+02, percent-clipped=2.0 +2022-11-15 20:28:40,305 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:29:03,066 INFO [train.py:876] (3/4) Epoch 6, batch 100, loss[loss=0.1941, simple_loss=0.1759, pruned_loss=0.1062, over 4044.00 frames. ], tot_loss[loss=0.17, simple_loss=0.1784, pruned_loss=0.08081, over 435452.46 frames. ], batch size: 181, lr: 1.40e-02, grad_scale: 16.0 +2022-11-15 20:29:07,137 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:29:08,499 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5449, 1.3162, 1.6886, 1.2527, 1.2914, 1.3351, 1.4897, 0.9808], + device='cuda:3'), covar=tensor([0.0014, 0.0031, 0.0021, 0.0019, 0.0023, 0.0048, 0.0015, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0016, 0.0016, 0.0018, 0.0016, 0.0016, 0.0018, 0.0018], + device='cuda:3'), out_proj_covar=tensor([1.6481e-05, 1.6403e-05, 1.6133e-05, 1.7846e-05, 1.6302e-05, 1.6964e-05, + 1.8179e-05, 2.0572e-05], device='cuda:3') +2022-11-15 20:29:12,766 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:29:40,202 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.873e+02 2.283e+02 2.904e+02 6.033e+02, threshold=4.566e+02, percent-clipped=4.0 +2022-11-15 20:29:40,287 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:29:55,303 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:29:58,059 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0234, 0.6278, 0.7246, 0.7707, 0.8891, 0.7436, 0.5742, 0.6057], + device='cuda:3'), covar=tensor([0.0479, 0.0404, 0.0508, 0.0569, 0.0475, 0.0400, 0.0909, 0.0447], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0012, 0.0009, 0.0010, 0.0010, 0.0009, 0.0011, 0.0009], + device='cuda:3'), out_proj_covar=tensor([3.8595e-05, 5.1544e-05, 4.1261e-05, 4.6406e-05, 4.3567e-05, 4.0417e-05, + 4.5513e-05, 4.2138e-05], device='cuda:3') +2022-11-15 20:30:04,907 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-15 20:30:06,935 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2022-11-15 20:30:11,764 INFO [train.py:876] (3/4) Epoch 6, batch 200, loss[loss=0.1685, simple_loss=0.1922, pruned_loss=0.07244, over 5537.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.1783, pruned_loss=0.08299, over 693383.24 frames. ], batch size: 15, lr: 1.39e-02, grad_scale: 16.0 +2022-11-15 20:30:23,092 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3833, 3.8907, 3.2477, 1.7765, 3.5533, 1.7081, 3.6394, 2.0715], + device='cuda:3'), covar=tensor([0.1891, 0.0369, 0.0547, 0.3434, 0.0416, 0.2850, 0.0343, 0.3086], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0101, 0.0109, 0.0120, 0.0102, 0.0132, 0.0095, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:30:35,168 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2728, 1.9120, 2.9462, 2.5380, 2.8118, 2.0960, 2.6720, 3.1973], + device='cuda:3'), covar=tensor([0.0351, 0.0965, 0.0404, 0.0903, 0.0336, 0.0849, 0.0571, 0.0449], + device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0192, 0.0187, 0.0213, 0.0183, 0.0192, 0.0227, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:30:42,299 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:30:42,446 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 +2022-11-15 20:30:44,374 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:30:46,960 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:30:48,739 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.665e+02 2.149e+02 2.847e+02 6.157e+02, threshold=4.299e+02, percent-clipped=2.0 +2022-11-15 20:30:55,614 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2704, 3.6919, 2.7685, 1.7566, 3.5204, 1.2455, 3.3107, 1.8865], + device='cuda:3'), covar=tensor([0.1437, 0.0188, 0.0760, 0.2065, 0.0229, 0.2428, 0.0301, 0.1922], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0101, 0.0109, 0.0122, 0.0103, 0.0133, 0.0096, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:31:20,153 INFO [train.py:876] (3/4) Epoch 6, batch 300, loss[loss=0.1254, simple_loss=0.1504, pruned_loss=0.05017, over 5761.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.1777, pruned_loss=0.08175, over 851775.27 frames. ], batch size: 14, lr: 1.39e-02, grad_scale: 16.0 +2022-11-15 20:31:23,588 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:31:25,557 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:31:28,139 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:31:34,375 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 +2022-11-15 20:31:37,891 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2022-11-15 20:31:46,956 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:31:56,890 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.229e+02 1.849e+02 2.219e+02 2.840e+02 6.377e+02, threshold=4.439e+02, percent-clipped=4.0 +2022-11-15 20:32:27,631 INFO [train.py:876] (3/4) Epoch 6, batch 400, loss[loss=0.1053, simple_loss=0.1219, pruned_loss=0.0444, over 5445.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.1762, pruned_loss=0.08126, over 937536.17 frames. ], batch size: 10, lr: 1.39e-02, grad_scale: 16.0 +2022-11-15 20:32:27,805 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:32:58,152 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:33:04,837 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.826e+02 2.118e+02 2.813e+02 4.458e+02, threshold=4.236e+02, percent-clipped=1.0 +2022-11-15 20:33:20,216 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:33:35,934 INFO [train.py:876] (3/4) Epoch 6, batch 500, loss[loss=0.2189, simple_loss=0.2055, pruned_loss=0.1161, over 5467.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.1766, pruned_loss=0.08089, over 997569.34 frames. ], batch size: 53, lr: 1.39e-02, grad_scale: 16.0 +2022-11-15 20:33:40,083 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:33:53,154 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:10,486 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:13,304 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.807e+02 2.308e+02 2.926e+02 6.442e+02, threshold=4.616e+02, percent-clipped=7.0 +2022-11-15 20:34:23,208 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5410, 1.1704, 1.3113, 1.2622, 1.3548, 1.3772, 1.2384, 1.1162], + device='cuda:3'), covar=tensor([0.0018, 0.0035, 0.0035, 0.0021, 0.0020, 0.0044, 0.0019, 0.0033], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0017, 0.0017, 0.0020, 0.0017, 0.0018, 0.0019, 0.0019], + device='cuda:3'), out_proj_covar=tensor([1.7279e-05, 1.7989e-05, 1.7432e-05, 1.9655e-05, 1.7300e-05, 1.8687e-05, + 1.9161e-05, 2.1878e-05], device='cuda:3') +2022-11-15 20:34:38,064 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.67 vs. limit=5.0 +2022-11-15 20:34:43,155 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:43,401 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-15 20:34:44,968 INFO [train.py:876] (3/4) Epoch 6, batch 600, loss[loss=0.1342, simple_loss=0.1506, pruned_loss=0.05895, over 5761.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.1749, pruned_loss=0.07887, over 1033377.46 frames. ], batch size: 20, lr: 1.39e-02, grad_scale: 16.0 +2022-11-15 20:34:45,043 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:47,035 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:49,624 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:34:52,739 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:35:15,284 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:35:22,242 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.873e+02 2.337e+02 2.752e+02 4.766e+02, threshold=4.674e+02, percent-clipped=2.0 +2022-11-15 20:35:25,072 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:35:48,577 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6814, 1.8032, 1.6925, 0.8472, 0.6868, 2.3051, 1.7610, 1.3374], + device='cuda:3'), covar=tensor([0.0512, 0.1044, 0.0463, 0.2137, 0.2798, 0.0802, 0.1348, 0.0655], + device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0044, 0.0045, 0.0055, 0.0049, 0.0040, 0.0044, 0.0044], + device='cuda:3'), out_proj_covar=tensor([1.0036e-04, 8.6455e-05, 8.9311e-05, 1.1135e-04, 9.7595e-05, 8.4852e-05, + 8.8222e-05, 8.7231e-05], device='cuda:3') +2022-11-15 20:35:51,217 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 20:35:54,386 INFO [train.py:876] (3/4) Epoch 6, batch 700, loss[loss=0.08642, simple_loss=0.115, pruned_loss=0.0289, over 5370.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.1773, pruned_loss=0.08123, over 1050473.97 frames. ], batch size: 9, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:35:57,283 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:36:26,536 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5607, 3.8148, 3.8637, 3.5826, 4.1067, 3.8781, 2.0223, 4.4731], + device='cuda:3'), covar=tensor([0.0310, 0.0606, 0.0325, 0.0423, 0.0509, 0.0487, 0.3277, 0.0317], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0074, 0.0075, 0.0065, 0.0092, 0.0078, 0.0129, 0.0096], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:36:31,398 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.989e+01 2.001e+02 2.497e+02 2.973e+02 6.578e+02, threshold=4.994e+02, percent-clipped=4.0 +2022-11-15 20:36:58,609 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9561, 4.3289, 4.6872, 4.3805, 4.9922, 4.8102, 4.3802, 4.9328], + device='cuda:3'), covar=tensor([0.0315, 0.0300, 0.0463, 0.0275, 0.0335, 0.0124, 0.0315, 0.0311], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0113, 0.0085, 0.0115, 0.0120, 0.0071, 0.0098, 0.0108], + 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') +2022-11-15 20:37:02,432 INFO [train.py:876] (3/4) Epoch 6, batch 800, loss[loss=0.1906, simple_loss=0.1859, pruned_loss=0.0977, over 5528.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.1768, pruned_loss=0.08031, over 1070585.78 frames. ], batch size: 40, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:37:02,892 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 20:37:40,547 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.784e+02 2.226e+02 2.677e+02 4.647e+02, threshold=4.452e+02, percent-clipped=0.0 +2022-11-15 20:37:44,235 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7716, 3.7928, 3.7027, 3.9117, 3.4318, 3.1133, 4.3289, 3.7991], + device='cuda:3'), covar=tensor([0.0420, 0.0846, 0.0449, 0.0991, 0.0590, 0.0525, 0.0762, 0.0493], + device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0092, 0.0078, 0.0098, 0.0074, 0.0062, 0.0122, 0.0080], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:38:11,375 INFO [train.py:876] (3/4) Epoch 6, batch 900, loss[loss=0.1176, simple_loss=0.1399, pruned_loss=0.04771, over 5741.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.1776, pruned_loss=0.08162, over 1068399.79 frames. ], batch size: 20, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:38:11,492 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:13,409 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:15,271 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:15,957 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:15,988 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:43,257 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:45,294 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:46,652 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:48,597 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.992e+02 2.315e+02 2.795e+02 5.537e+02, threshold=4.630e+02, percent-clipped=3.0 +2022-11-15 20:38:48,678 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:38:57,779 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:39:15,676 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:39:18,338 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:39:18,957 INFO [train.py:876] (3/4) Epoch 6, batch 1000, loss[loss=0.1721, simple_loss=0.1801, pruned_loss=0.08206, over 5731.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.1775, pruned_loss=0.08193, over 1069640.79 frames. ], batch size: 15, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:39:21,058 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:39:48,191 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 20:39:56,108 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.744e+01 1.732e+02 2.123e+02 2.683e+02 6.509e+02, threshold=4.246e+02, percent-clipped=3.0 +2022-11-15 20:40:02,461 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:40:26,881 INFO [train.py:876] (3/4) Epoch 6, batch 1100, loss[loss=0.2574, simple_loss=0.2087, pruned_loss=0.153, over 3074.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.1792, pruned_loss=0.08273, over 1075238.80 frames. ], batch size: 284, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:40:26,996 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:40:59,804 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 20:41:04,181 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.745e+02 2.111e+02 2.538e+02 7.660e+02, threshold=4.223e+02, percent-clipped=2.0 +2022-11-15 20:41:35,287 INFO [train.py:876] (3/4) Epoch 6, batch 1200, loss[loss=0.1322, simple_loss=0.1573, pruned_loss=0.05355, over 5457.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.1773, pruned_loss=0.08104, over 1083029.23 frames. ], batch size: 12, lr: 1.38e-02, grad_scale: 16.0 +2022-11-15 20:41:39,297 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:42:11,420 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:42:12,016 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:42:12,626 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.872e+02 2.304e+02 2.879e+02 7.161e+02, threshold=4.608e+02, percent-clipped=4.0 +2022-11-15 20:42:18,031 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:42:42,640 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:42:43,125 INFO [train.py:876] (3/4) Epoch 6, batch 1300, loss[loss=0.2544, simple_loss=0.2358, pruned_loss=0.1365, over 5593.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.1781, pruned_loss=0.08135, over 1090168.27 frames. ], batch size: 50, lr: 1.37e-02, grad_scale: 16.0 +2022-11-15 20:42:44,190 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:43:03,343 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3502, 3.4903, 3.3381, 3.3926, 3.5196, 3.3673, 1.2333, 3.6139], + device='cuda:3'), covar=tensor([0.0361, 0.0221, 0.0295, 0.0198, 0.0256, 0.0318, 0.3068, 0.0260], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0076, 0.0077, 0.0066, 0.0093, 0.0080, 0.0130, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:43:15,366 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:43:20,509 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.632e+02 1.886e+02 2.277e+02 3.768e+02, threshold=3.772e+02, percent-clipped=0.0 +2022-11-15 20:43:23,147 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:43:31,902 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 +2022-11-15 20:43:32,352 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 20:43:51,754 INFO [train.py:876] (3/4) Epoch 6, batch 1400, loss[loss=0.1674, simple_loss=0.1842, pruned_loss=0.07528, over 5721.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.177, pruned_loss=0.08101, over 1083767.46 frames. ], batch size: 17, lr: 1.37e-02, grad_scale: 16.0 +2022-11-15 20:43:54,821 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0484, 2.2704, 1.8531, 2.4997, 1.6921, 2.1007, 2.0383, 2.8093], + device='cuda:3'), covar=tensor([0.0726, 0.1445, 0.2327, 0.0626, 0.1772, 0.0746, 0.1634, 0.1094], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0072, 0.0088, 0.0058, 0.0074, 0.0064, 0.0081, 0.0058], + 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') +2022-11-15 20:44:12,444 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.74 vs. limit=5.0 +2022-11-15 20:44:31,663 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 1.794e+02 2.213e+02 2.826e+02 4.726e+02, threshold=4.425e+02, percent-clipped=5.0 +2022-11-15 20:44:37,932 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 20:44:56,552 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1437, 3.8049, 3.7734, 4.1038, 4.1307, 3.9003, 1.5122, 4.1650], + device='cuda:3'), covar=tensor([0.0297, 0.0442, 0.0441, 0.0295, 0.0384, 0.0485, 0.3493, 0.0444], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0076, 0.0075, 0.0066, 0.0092, 0.0079, 0.0128, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:45:01,665 INFO [train.py:876] (3/4) Epoch 6, batch 1500, loss[loss=0.2099, simple_loss=0.2045, pruned_loss=0.1077, over 5508.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.1765, pruned_loss=0.08043, over 1084558.07 frames. ], batch size: 49, lr: 1.37e-02, grad_scale: 16.0 +2022-11-15 20:45:38,927 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.859e+02 2.319e+02 2.620e+02 5.467e+02, threshold=4.638e+02, percent-clipped=1.0 +2022-11-15 20:45:45,052 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:45:58,756 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2022-11-15 20:46:10,114 INFO [train.py:876] (3/4) Epoch 6, batch 1600, loss[loss=0.1766, simple_loss=0.1878, pruned_loss=0.08268, over 5768.00 frames. ], tot_loss[loss=0.17, simple_loss=0.1773, pruned_loss=0.08135, over 1081126.90 frames. ], batch size: 20, lr: 1.37e-02, grad_scale: 16.0 +2022-11-15 20:46:17,403 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:46:22,795 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:46:31,243 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 +2022-11-15 20:46:37,919 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8020, 4.2946, 4.5804, 4.3578, 4.9087, 4.7235, 4.2525, 4.8331], + device='cuda:3'), covar=tensor([0.0378, 0.0287, 0.0468, 0.0285, 0.0308, 0.0129, 0.0281, 0.0248], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0115, 0.0087, 0.0117, 0.0125, 0.0074, 0.0099, 0.0109], + 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') +2022-11-15 20:46:47,655 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.872e+01 1.813e+02 2.310e+02 2.979e+02 5.455e+02, threshold=4.619e+02, percent-clipped=4.0 +2022-11-15 20:46:49,448 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2022-11-15 20:46:50,390 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:47:03,882 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:47:18,026 INFO [train.py:876] (3/4) Epoch 6, batch 1700, loss[loss=0.136, simple_loss=0.1555, pruned_loss=0.05826, over 5323.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.1765, pruned_loss=0.08096, over 1081098.29 frames. ], batch size: 9, lr: 1.37e-02, grad_scale: 16.0 +2022-11-15 20:47:22,635 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:47:55,396 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.764e+02 2.104e+02 2.685e+02 5.145e+02, threshold=4.208e+02, percent-clipped=1.0 +2022-11-15 20:48:25,387 INFO [train.py:876] (3/4) Epoch 6, batch 1800, loss[loss=0.1388, simple_loss=0.1504, pruned_loss=0.0636, over 5552.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.1761, pruned_loss=0.08063, over 1088014.64 frames. ], batch size: 14, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:48:52,093 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 +2022-11-15 20:48:57,780 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9711, 2.2889, 2.6118, 2.3643, 1.5731, 2.4363, 1.6346, 1.5360], + device='cuda:3'), covar=tensor([0.0189, 0.0076, 0.0079, 0.0118, 0.0236, 0.0077, 0.0258, 0.0134], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0128, 0.0143, 0.0159, 0.0163, 0.0140, 0.0156, 0.0129], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:49:02,063 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5343, 3.7917, 3.6711, 3.3246, 1.9829, 3.7865, 2.0408, 3.1931], + device='cuda:3'), covar=tensor([0.0350, 0.0134, 0.0113, 0.0311, 0.0444, 0.0110, 0.0389, 0.0109], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0128, 0.0144, 0.0160, 0.0164, 0.0141, 0.0157, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:49:03,152 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.837e+02 2.239e+02 2.713e+02 4.527e+02, threshold=4.477e+02, percent-clipped=1.0 +2022-11-15 20:49:33,776 INFO [train.py:876] (3/4) Epoch 6, batch 1900, loss[loss=0.1282, simple_loss=0.1408, pruned_loss=0.05781, over 5718.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.1749, pruned_loss=0.07915, over 1089389.79 frames. ], batch size: 15, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:50:06,741 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9760, 2.8905, 2.2574, 1.5224, 2.8344, 1.1287, 2.8268, 1.5329], + device='cuda:3'), covar=tensor([0.1207, 0.0181, 0.1071, 0.1773, 0.0220, 0.2048, 0.0254, 0.1661], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0100, 0.0111, 0.0123, 0.0103, 0.0130, 0.0093, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 20:50:10,565 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.862e+02 2.216e+02 2.632e+02 5.559e+02, threshold=4.433e+02, percent-clipped=2.0 +2022-11-15 20:50:24,598 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:50:40,951 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:50:41,461 INFO [train.py:876] (3/4) Epoch 6, batch 2000, loss[loss=0.1033, simple_loss=0.1221, pruned_loss=0.04226, over 5209.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.174, pruned_loss=0.07831, over 1088423.06 frames. ], batch size: 8, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:51:14,852 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-15 20:51:15,585 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 +2022-11-15 20:51:19,232 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 2.027e+02 2.499e+02 3.028e+02 6.402e+02, threshold=4.998e+02, percent-clipped=6.0 +2022-11-15 20:51:22,836 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:51:29,701 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6574, 3.7690, 3.5955, 3.5371, 2.3815, 4.0822, 2.0754, 3.2805], + device='cuda:3'), covar=tensor([0.0404, 0.0293, 0.0186, 0.0297, 0.0468, 0.0111, 0.0433, 0.0169], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0133, 0.0145, 0.0162, 0.0166, 0.0143, 0.0159, 0.0133], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:51:50,149 INFO [train.py:876] (3/4) Epoch 6, batch 2100, loss[loss=0.1209, simple_loss=0.144, pruned_loss=0.04886, over 5535.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.1752, pruned_loss=0.07971, over 1083370.24 frames. ], batch size: 10, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:51:59,736 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 +2022-11-15 20:52:10,992 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 +2022-11-15 20:52:21,091 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4176, 2.0846, 2.3856, 3.4455, 3.3198, 2.4400, 2.1281, 3.5995], + device='cuda:3'), covar=tensor([0.0433, 0.2711, 0.2295, 0.2180, 0.0904, 0.2792, 0.2141, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0203, 0.0207, 0.0324, 0.0216, 0.0219, 0.0201, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 20:52:27,550 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.554e+01 1.866e+02 2.249e+02 2.710e+02 6.180e+02, threshold=4.497e+02, percent-clipped=1.0 +2022-11-15 20:52:39,884 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5866, 2.7758, 2.7155, 2.7001, 2.6906, 2.7418, 1.1649, 2.7711], + device='cuda:3'), covar=tensor([0.0317, 0.0217, 0.0235, 0.0201, 0.0297, 0.0245, 0.2448, 0.0279], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0074, 0.0076, 0.0067, 0.0093, 0.0079, 0.0128, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:52:41,517 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.15 vs. limit=5.0 +2022-11-15 20:52:58,526 INFO [train.py:876] (3/4) Epoch 6, batch 2200, loss[loss=0.1623, simple_loss=0.1803, pruned_loss=0.07213, over 5518.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.1746, pruned_loss=0.07884, over 1083889.76 frames. ], batch size: 17, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:53:33,321 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2993, 1.2273, 1.2657, 0.9740, 1.2818, 1.0011, 1.0480, 0.7303], + device='cuda:3'), covar=tensor([0.0017, 0.0033, 0.0017, 0.0024, 0.0018, 0.0023, 0.0019, 0.0037], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0016, 0.0017, 0.0020, 0.0018, 0.0016, 0.0018, 0.0019], + device='cuda:3'), out_proj_covar=tensor([1.7340e-05, 1.6740e-05, 1.6896e-05, 1.9882e-05, 1.7860e-05, 1.6969e-05, + 1.8259e-05, 2.0706e-05], device='cuda:3') +2022-11-15 20:53:36,405 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.801e+02 2.106e+02 2.533e+02 3.933e+02, threshold=4.211e+02, percent-clipped=0.0 +2022-11-15 20:53:39,792 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8916, 2.5975, 2.2069, 2.7457, 1.8678, 2.7261, 2.5239, 3.1360], + device='cuda:3'), covar=tensor([0.0818, 0.1653, 0.3087, 0.2241, 0.2399, 0.1057, 0.1757, 0.4141], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0073, 0.0088, 0.0059, 0.0072, 0.0065, 0.0080, 0.0059], + 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') +2022-11-15 20:53:49,024 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:54:06,823 INFO [train.py:876] (3/4) Epoch 6, batch 2300, loss[loss=0.2098, simple_loss=0.2066, pruned_loss=0.1065, over 5546.00 frames. ], tot_loss[loss=0.166, simple_loss=0.1746, pruned_loss=0.07875, over 1079924.90 frames. ], batch size: 46, lr: 1.36e-02, grad_scale: 16.0 +2022-11-15 20:54:21,936 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:54:27,235 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4465, 3.6317, 3.3598, 3.1424, 2.2074, 3.4585, 2.1760, 3.0340], + device='cuda:3'), covar=tensor([0.0333, 0.0111, 0.0150, 0.0231, 0.0381, 0.0132, 0.0368, 0.0125], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0131, 0.0145, 0.0161, 0.0165, 0.0142, 0.0158, 0.0131], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:54:44,925 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.846e+02 2.203e+02 3.021e+02 7.472e+02, threshold=4.405e+02, percent-clipped=8.0 +2022-11-15 20:54:45,038 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:55:15,139 INFO [train.py:876] (3/4) Epoch 6, batch 2400, loss[loss=0.158, simple_loss=0.1754, pruned_loss=0.07031, over 5767.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.1757, pruned_loss=0.07938, over 1081177.24 frames. ], batch size: 16, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 20:55:23,242 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4323, 3.3383, 3.2573, 3.0236, 2.0453, 3.3935, 2.1519, 2.8707], + device='cuda:3'), covar=tensor([0.0277, 0.0106, 0.0119, 0.0272, 0.0326, 0.0103, 0.0279, 0.0113], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0130, 0.0143, 0.0161, 0.0165, 0.0143, 0.0158, 0.0131], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 20:55:52,457 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.810e+02 2.219e+02 2.775e+02 4.582e+02, threshold=4.438e+02, percent-clipped=1.0 +2022-11-15 20:55:57,931 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:56:23,604 INFO [train.py:876] (3/4) Epoch 6, batch 2500, loss[loss=0.1417, simple_loss=0.1663, pruned_loss=0.0585, over 5565.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.1756, pruned_loss=0.07909, over 1081813.11 frames. ], batch size: 24, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 20:56:39,902 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:56:48,520 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2752, 1.1969, 1.2779, 1.0623, 0.9381, 0.9266, 1.4027, 1.3206], + device='cuda:3'), covar=tensor([0.0054, 0.0057, 0.0080, 0.0062, 0.0091, 0.0108, 0.0019, 0.0027], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0017, 0.0017, 0.0020, 0.0019, 0.0017, 0.0019, 0.0019], + device='cuda:3'), out_proj_covar=tensor([1.7631e-05, 1.7213e-05, 1.7072e-05, 1.9949e-05, 1.8324e-05, 1.7640e-05, + 1.8972e-05, 2.0917e-05], device='cuda:3') +2022-11-15 20:56:55,844 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0318, 2.7860, 2.5311, 1.6200, 2.6664, 2.9331, 2.7565, 3.0526], + device='cuda:3'), covar=tensor([0.1703, 0.1158, 0.0730, 0.2237, 0.0374, 0.0369, 0.0354, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0194, 0.0145, 0.0195, 0.0157, 0.0159, 0.0138, 0.0182], + 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') +2022-11-15 20:57:01,241 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.730e+02 2.141e+02 2.666e+02 7.999e+02, threshold=4.283e+02, percent-clipped=2.0 +2022-11-15 20:57:31,861 INFO [train.py:876] (3/4) Epoch 6, batch 2600, loss[loss=0.1849, simple_loss=0.1952, pruned_loss=0.08727, over 5616.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.1746, pruned_loss=0.07827, over 1081519.67 frames. ], batch size: 23, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 20:57:33,350 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8131, 3.4281, 2.2870, 3.2234, 2.3770, 2.4131, 1.8729, 2.9695], + device='cuda:3'), covar=tensor([0.1514, 0.0178, 0.0961, 0.0263, 0.0901, 0.0931, 0.1698, 0.0251], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0129, 0.0167, 0.0134, 0.0166, 0.0178, 0.0178, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 20:58:09,806 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.750e+02 2.073e+02 2.691e+02 5.649e+02, threshold=4.146e+02, percent-clipped=5.0 +2022-11-15 20:58:09,973 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:58:40,660 INFO [train.py:876] (3/4) Epoch 6, batch 2700, loss[loss=0.1191, simple_loss=0.1402, pruned_loss=0.04905, over 5692.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.1742, pruned_loss=0.07757, over 1079654.11 frames. ], batch size: 19, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 20:58:42,655 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 20:59:17,244 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4797, 4.3017, 2.9309, 4.0888, 3.1728, 3.0684, 2.1638, 3.5560], + device='cuda:3'), covar=tensor([0.1190, 0.0138, 0.0861, 0.0214, 0.0577, 0.0754, 0.1698, 0.0289], + device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0128, 0.0166, 0.0132, 0.0164, 0.0176, 0.0175, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 20:59:18,425 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.741e+02 2.185e+02 2.571e+02 4.809e+02, threshold=4.370e+02, percent-clipped=1.0 +2022-11-15 20:59:49,229 INFO [train.py:876] (3/4) Epoch 6, batch 2800, loss[loss=0.1808, simple_loss=0.1897, pruned_loss=0.08593, over 5605.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.1736, pruned_loss=0.0779, over 1080233.87 frames. ], batch size: 23, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 21:00:01,219 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:00:05,538 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:00:27,552 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.875e+02 2.211e+02 2.740e+02 7.049e+02, threshold=4.422e+02, percent-clipped=6.0 +2022-11-15 21:00:31,478 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 21:00:31,820 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3943, 2.7106, 2.1570, 2.6089, 2.2237, 2.6466, 2.6870, 3.2636], + device='cuda:3'), covar=tensor([0.1063, 0.1871, 0.2977, 0.1197, 0.2099, 0.1552, 0.1873, 0.4553], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0073, 0.0088, 0.0060, 0.0072, 0.0065, 0.0080, 0.0056], + 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') +2022-11-15 21:00:47,927 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:00:58,295 INFO [train.py:876] (3/4) Epoch 6, batch 2900, loss[loss=0.1661, simple_loss=0.1787, pruned_loss=0.0767, over 5580.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.1728, pruned_loss=0.07724, over 1082587.94 frames. ], batch size: 43, lr: 1.35e-02, grad_scale: 16.0 +2022-11-15 21:01:04,580 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2022-11-15 21:01:07,819 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9642, 2.7830, 2.5373, 1.5004, 2.5987, 2.9958, 2.9935, 3.2981], + device='cuda:3'), covar=tensor([0.2144, 0.1269, 0.1205, 0.2804, 0.0464, 0.0419, 0.0288, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0194, 0.0147, 0.0199, 0.0158, 0.0160, 0.0139, 0.0185], + 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') +2022-11-15 21:01:13,863 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:01:14,487 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4928, 1.3198, 1.1985, 0.8687, 1.0389, 1.9203, 1.3562, 1.2680], + device='cuda:3'), covar=tensor([0.0861, 0.0703, 0.0844, 0.1772, 0.1367, 0.0647, 0.1353, 0.1099], + device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0042, 0.0046, 0.0057, 0.0046, 0.0040, 0.0045, 0.0047], + device='cuda:3'), out_proj_covar=tensor([1.0377e-04, 8.8227e-05, 9.4631e-05, 1.1710e-04, 9.7562e-05, 8.7458e-05, + 9.2427e-05, 9.4463e-05], device='cuda:3') +2022-11-15 21:01:26,747 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0132, 2.8128, 2.6951, 1.5119, 2.4293, 3.0302, 2.9540, 3.3804], + device='cuda:3'), covar=tensor([0.2040, 0.1481, 0.0802, 0.2729, 0.0608, 0.0541, 0.0333, 0.0556], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0196, 0.0147, 0.0200, 0.0159, 0.0162, 0.0139, 0.0187], + 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') +2022-11-15 21:01:31,515 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6117, 4.4355, 4.8545, 4.6720, 4.3670, 3.8217, 5.3183, 4.5167], + device='cuda:3'), covar=tensor([0.0411, 0.1025, 0.0312, 0.0988, 0.0460, 0.0323, 0.0650, 0.0581], + device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0093, 0.0075, 0.0097, 0.0073, 0.0062, 0.0120, 0.0079], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:01:36,727 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.016e+02 2.489e+02 3.062e+02 5.776e+02, threshold=4.978e+02, percent-clipped=1.0 +2022-11-15 21:01:46,195 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9177, 1.3750, 1.1487, 0.9817, 0.6262, 1.1213, 0.4754, 1.0787], + device='cuda:3'), covar=tensor([0.0018, 0.0013, 0.0016, 0.0023, 0.0019, 0.0018, 0.0049, 0.0024], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0028, 0.0031, 0.0031, 0.0028, 0.0027, 0.0030, 0.0025], + device='cuda:3'), out_proj_covar=tensor([2.9074e-05, 2.8948e-05, 2.7998e-05, 2.8015e-05, 2.5045e-05, 2.3306e-05, + 3.1713e-05, 2.2613e-05], device='cuda:3') +2022-11-15 21:01:55,776 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:02:04,885 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7378, 2.4972, 2.8697, 1.5571, 2.4543, 2.8963, 2.8805, 2.8316], + device='cuda:3'), covar=tensor([0.2557, 0.1831, 0.0806, 0.3378, 0.0891, 0.0613, 0.0489, 0.0986], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0192, 0.0143, 0.0197, 0.0156, 0.0159, 0.0136, 0.0181], + 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') +2022-11-15 21:02:06,701 INFO [train.py:876] (3/4) Epoch 6, batch 3000, loss[loss=0.1225, simple_loss=0.1454, pruned_loss=0.04978, over 4816.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.1747, pruned_loss=0.07907, over 1078809.15 frames. ], batch size: 5, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:02:06,701 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 21:02:13,957 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4945, 2.2149, 3.2174, 2.9336, 2.7718, 1.9761, 2.6787, 3.2593], + device='cuda:3'), covar=tensor([0.0215, 0.0644, 0.0269, 0.0445, 0.0356, 0.0674, 0.0448, 0.0435], + device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0186, 0.0186, 0.0205, 0.0188, 0.0186, 0.0220, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:02:15,922 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4384, 3.8102, 3.4196, 3.1203, 2.0776, 3.7031, 2.1171, 3.0908], + device='cuda:3'), covar=tensor([0.0370, 0.0086, 0.0165, 0.0288, 0.0445, 0.0109, 0.0378, 0.0128], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0128, 0.0141, 0.0159, 0.0161, 0.0140, 0.0154, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:02:24,359 INFO [train.py:908] (3/4) Epoch 6, validation: loss=0.1626, simple_loss=0.1844, pruned_loss=0.07046, over 1530663.00 frames. +2022-11-15 21:02:24,359 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 21:02:33,631 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9028, 3.2770, 2.3482, 3.0081, 2.2336, 2.3890, 1.8212, 2.8565], + device='cuda:3'), covar=tensor([0.1448, 0.0179, 0.0971, 0.0290, 0.0916, 0.0956, 0.1810, 0.0372], + device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0133, 0.0175, 0.0137, 0.0169, 0.0183, 0.0183, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 21:03:01,501 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.785e+02 2.182e+02 2.915e+02 6.384e+02, threshold=4.364e+02, percent-clipped=3.0 +2022-11-15 21:03:05,326 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1616, 2.5122, 3.6456, 3.2942, 4.2178, 2.6802, 3.5673, 4.3133], + device='cuda:3'), covar=tensor([0.0351, 0.1377, 0.0711, 0.1282, 0.0251, 0.1349, 0.0999, 0.0483], + device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0187, 0.0186, 0.0205, 0.0186, 0.0187, 0.0222, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:03:31,200 INFO [train.py:876] (3/4) Epoch 6, batch 3100, loss[loss=0.2002, simple_loss=0.1889, pruned_loss=0.1057, over 5490.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.1741, pruned_loss=0.07858, over 1085022.87 frames. ], batch size: 64, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:03:33,043 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:03:40,518 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2959, 3.7673, 3.1850, 3.7299, 3.7387, 3.0863, 3.2826, 3.1361], + device='cuda:3'), covar=tensor([0.0919, 0.0487, 0.1870, 0.0492, 0.0483, 0.0571, 0.0666, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0149, 0.0239, 0.0149, 0.0181, 0.0155, 0.0160, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:03:43,785 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:04:00,999 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8915, 0.6599, 0.6397, 0.5173, 0.7012, 0.6839, 0.4185, 0.5694], + device='cuda:3'), covar=tensor([0.0252, 0.0419, 0.0537, 0.0574, 0.0488, 0.0349, 0.0894, 0.0455], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0012, 0.0011, 0.0009, 0.0013, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.3851e-05, 5.6591e-05, 4.5916e-05, 5.2922e-05, 4.8940e-05, 4.3339e-05, + 5.3925e-05, 4.7218e-05], device='cuda:3') +2022-11-15 21:04:09,321 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.769e+02 2.283e+02 2.689e+02 7.122e+02, threshold=4.567e+02, percent-clipped=2.0 +2022-11-15 21:04:14,715 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:04:16,933 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:04:26,514 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:04:33,054 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7327, 4.2233, 3.1555, 2.0254, 4.0013, 1.5376, 4.0609, 2.1610], + device='cuda:3'), covar=tensor([0.1102, 0.0124, 0.0566, 0.1876, 0.0167, 0.2048, 0.0130, 0.1745], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0102, 0.0109, 0.0118, 0.0104, 0.0130, 0.0094, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:04:40,020 INFO [train.py:876] (3/4) Epoch 6, batch 3200, loss[loss=0.1946, simple_loss=0.1966, pruned_loss=0.09633, over 5741.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.1736, pruned_loss=0.07772, over 1083523.34 frames. ], batch size: 14, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:05:18,392 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.194e+02 1.784e+02 2.150e+02 2.757e+02 5.278e+02, threshold=4.299e+02, percent-clipped=1.0 +2022-11-15 21:05:19,943 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6825, 1.7960, 2.1233, 2.9453, 2.7837, 2.1753, 1.7724, 3.1610], + device='cuda:3'), covar=tensor([0.0618, 0.2499, 0.1915, 0.1777, 0.1103, 0.2355, 0.2029, 0.1019], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0213, 0.0204, 0.0329, 0.0219, 0.0218, 0.0201, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:05:33,687 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:05:34,270 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:05:45,219 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0546, 2.3390, 3.6726, 3.2584, 4.1513, 2.6433, 3.6324, 4.1394], + device='cuda:3'), covar=tensor([0.0460, 0.1508, 0.0750, 0.1719, 0.0271, 0.1410, 0.0988, 0.0628], + device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0189, 0.0188, 0.0207, 0.0188, 0.0188, 0.0223, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:05:48,168 INFO [train.py:876] (3/4) Epoch 6, batch 3300, loss[loss=0.1319, simple_loss=0.1547, pruned_loss=0.05451, over 5473.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.1745, pruned_loss=0.07819, over 1087245.13 frames. ], batch size: 12, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:05:58,643 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9214, 4.6144, 3.7041, 1.9605, 4.3523, 1.7972, 4.3582, 2.4122], + device='cuda:3'), covar=tensor([0.1184, 0.0145, 0.0373, 0.2193, 0.0128, 0.1771, 0.0122, 0.1635], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0101, 0.0109, 0.0118, 0.0103, 0.0130, 0.0095, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:06:02,592 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4321, 3.9940, 4.2221, 3.9392, 4.4809, 4.3203, 4.0564, 4.4580], + device='cuda:3'), covar=tensor([0.0331, 0.0278, 0.0445, 0.0298, 0.0313, 0.0197, 0.0242, 0.0252], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0117, 0.0089, 0.0120, 0.0127, 0.0077, 0.0103, 0.0115], + 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') +2022-11-15 21:06:10,872 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2022-11-15 21:06:15,364 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 21:06:27,137 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.777e+02 2.323e+02 2.808e+02 5.808e+02, threshold=4.646e+02, percent-clipped=1.0 +2022-11-15 21:06:54,491 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6983, 4.2630, 4.5280, 4.2969, 4.7602, 4.6275, 4.2135, 4.7404], + device='cuda:3'), covar=tensor([0.0406, 0.0266, 0.0422, 0.0299, 0.0359, 0.0163, 0.0253, 0.0279], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0117, 0.0089, 0.0119, 0.0127, 0.0076, 0.0103, 0.0116], + 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') +2022-11-15 21:06:57,086 INFO [train.py:876] (3/4) Epoch 6, batch 3400, loss[loss=0.1094, simple_loss=0.1362, pruned_loss=0.04129, over 5526.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.1736, pruned_loss=0.07731, over 1087925.95 frames. ], batch size: 10, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:07:01,535 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:07:18,001 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4445, 4.1345, 3.3491, 3.1318, 2.4794, 4.1350, 2.1339, 3.4590], + device='cuda:3'), covar=tensor([0.0459, 0.0146, 0.0242, 0.0521, 0.0516, 0.0086, 0.0466, 0.0110], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0132, 0.0143, 0.0163, 0.0164, 0.0142, 0.0157, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:07:34,877 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:07:36,023 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.758e+02 2.190e+02 2.600e+02 4.839e+02, threshold=4.379e+02, percent-clipped=2.0 +2022-11-15 21:07:37,419 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:07:42,823 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7079, 0.9834, 1.3708, 0.8401, 1.4277, 1.2379, 0.7848, 1.1072], + device='cuda:3'), covar=tensor([0.0340, 0.0453, 0.0750, 0.1150, 0.0753, 0.0441, 0.1480, 0.0538], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0012, 0.0011, 0.0010, 0.0012, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.3436e-05, 5.7084e-05, 4.4952e-05, 5.2084e-05, 4.7725e-05, 4.4027e-05, + 5.3102e-05, 4.6667e-05], device='cuda:3') +2022-11-15 21:07:43,448 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:07:46,057 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5918, 1.7957, 1.9906, 2.6359, 2.8296, 2.1335, 1.6146, 3.0685], + device='cuda:3'), covar=tensor([0.0787, 0.2615, 0.2004, 0.1546, 0.0770, 0.2264, 0.2178, 0.0430], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0212, 0.0206, 0.0326, 0.0218, 0.0218, 0.0203, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:07:52,336 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:08:04,961 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0952, 2.1898, 1.9556, 2.2355, 2.0311, 1.5781, 1.8265, 2.6139], + device='cuda:3'), covar=tensor([0.1158, 0.1650, 0.2503, 0.1280, 0.1541, 0.1465, 0.2013, 0.1449], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0072, 0.0086, 0.0060, 0.0071, 0.0064, 0.0079, 0.0057], + 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') +2022-11-15 21:08:06,130 INFO [train.py:876] (3/4) Epoch 6, batch 3500, loss[loss=0.08932, simple_loss=0.1146, pruned_loss=0.03201, over 5170.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.1726, pruned_loss=0.07604, over 1086864.09 frames. ], batch size: 8, lr: 1.34e-02, grad_scale: 16.0 +2022-11-15 21:08:16,404 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:08:24,879 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:08:43,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.932e+02 2.210e+02 2.836e+02 6.294e+02, threshold=4.421e+02, percent-clipped=3.0 +2022-11-15 21:08:59,714 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:09:13,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9243, 1.2298, 1.3075, 0.7115, 1.0060, 1.2282, 0.5907, 1.0357], + device='cuda:3'), covar=tensor([0.0022, 0.0012, 0.0014, 0.0020, 0.0017, 0.0015, 0.0031, 0.0017], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0029, 0.0032, 0.0032, 0.0030, 0.0029, 0.0031, 0.0027], + device='cuda:3'), out_proj_covar=tensor([3.0521e-05, 2.9485e-05, 2.9197e-05, 2.9504e-05, 2.6413e-05, 2.4927e-05, + 3.1933e-05, 2.3960e-05], device='cuda:3') +2022-11-15 21:09:14,036 INFO [train.py:876] (3/4) Epoch 6, batch 3600, loss[loss=0.1963, simple_loss=0.1888, pruned_loss=0.1019, over 5348.00 frames. ], tot_loss[loss=0.1616, simple_loss=0.172, pruned_loss=0.07565, over 1086043.75 frames. ], batch size: 70, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:09:19,751 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5483, 1.1757, 0.8976, 0.3204, 0.9128, 0.9247, 0.4848, 0.9407], + device='cuda:3'), covar=tensor([0.0036, 0.0012, 0.0017, 0.0027, 0.0015, 0.0018, 0.0044, 0.0020], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0029, 0.0032, 0.0032, 0.0030, 0.0029, 0.0031, 0.0027], + device='cuda:3'), out_proj_covar=tensor([3.0834e-05, 2.9810e-05, 2.9449e-05, 2.9825e-05, 2.6679e-05, 2.5160e-05, + 3.2233e-05, 2.4118e-05], device='cuda:3') +2022-11-15 21:09:32,386 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:09:36,966 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 21:09:55,016 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.820e+02 2.295e+02 3.006e+02 4.723e+02, threshold=4.590e+02, percent-clipped=2.0 +2022-11-15 21:10:00,112 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7369, 2.7004, 2.2126, 2.6721, 2.0603, 2.4874, 2.4990, 3.2920], + device='cuda:3'), covar=tensor([0.0908, 0.1668, 0.3108, 0.4646, 0.2513, 0.1990, 0.2142, 0.2274], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0072, 0.0087, 0.0061, 0.0072, 0.0065, 0.0079, 0.0058], + 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') +2022-11-15 21:10:25,134 INFO [train.py:876] (3/4) Epoch 6, batch 3700, loss[loss=0.1546, simple_loss=0.1734, pruned_loss=0.0679, over 5603.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.1726, pruned_loss=0.07634, over 1085086.64 frames. ], batch size: 23, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:10:45,871 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9179, 4.1087, 3.0485, 4.1158, 3.8669, 3.6621, 4.0601, 3.9324], + device='cuda:3'), covar=tensor([0.0466, 0.1013, 0.2916, 0.0918, 0.1312, 0.0640, 0.0758, 0.0642], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0153, 0.0240, 0.0149, 0.0181, 0.0155, 0.0164, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:11:03,523 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.767e+02 2.199e+02 2.713e+02 5.676e+02, threshold=4.399e+02, percent-clipped=2.0 +2022-11-15 21:11:04,982 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:11:07,567 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:11:21,247 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2022-11-15 21:11:33,225 INFO [train.py:876] (3/4) Epoch 6, batch 3800, loss[loss=0.1469, simple_loss=0.1644, pruned_loss=0.06475, over 5507.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.1722, pruned_loss=0.07699, over 1084602.94 frames. ], batch size: 12, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:11:37,524 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:11:40,247 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:11:49,892 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3721, 4.3202, 4.1175, 4.3653, 4.0479, 3.9168, 4.8701, 4.3545], + device='cuda:3'), covar=tensor([0.0449, 0.0880, 0.0481, 0.1007, 0.0491, 0.0343, 0.0787, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0092, 0.0076, 0.0097, 0.0073, 0.0062, 0.0123, 0.0080], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:11:57,145 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:12:11,859 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.630e+02 2.050e+02 2.494e+02 4.190e+02, threshold=4.099e+02, percent-clipped=0.0 +2022-11-15 21:12:13,029 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:12:39,067 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:12:42,144 INFO [train.py:876] (3/4) Epoch 6, batch 3900, loss[loss=0.1981, simple_loss=0.189, pruned_loss=0.1037, over 5441.00 frames. ], tot_loss[loss=0.1614, simple_loss=0.1713, pruned_loss=0.07573, over 1084307.46 frames. ], batch size: 53, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:12:49,552 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:12:54,512 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:13:05,291 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 21:13:20,314 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.159e+02 1.970e+02 2.299e+02 2.764e+02 4.240e+02, threshold=4.598e+02, percent-clipped=2.0 +2022-11-15 21:13:28,948 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:13:31,212 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:13:37,591 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 21:13:47,355 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 21:13:49,826 INFO [train.py:876] (3/4) Epoch 6, batch 4000, loss[loss=0.2049, simple_loss=0.1873, pruned_loss=0.1113, over 4716.00 frames. ], tot_loss[loss=0.1614, simple_loss=0.1717, pruned_loss=0.07555, over 1087210.39 frames. ], batch size: 135, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:14:02,827 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7997, 3.5569, 2.2289, 3.3230, 2.6185, 2.4200, 1.7650, 2.8030], + device='cuda:3'), covar=tensor([0.2251, 0.0284, 0.1669, 0.0435, 0.0971, 0.1395, 0.2595, 0.0498], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0130, 0.0168, 0.0133, 0.0164, 0.0177, 0.0179, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 21:14:10,420 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:14:28,141 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.803e+02 2.226e+02 2.834e+02 5.770e+02, threshold=4.452e+02, percent-clipped=4.0 +2022-11-15 21:14:31,399 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6946, 3.9979, 3.6676, 3.3390, 2.3190, 4.0560, 2.1795, 2.9059], + device='cuda:3'), covar=tensor([0.0398, 0.0153, 0.0141, 0.0329, 0.0433, 0.0109, 0.0411, 0.0182], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0130, 0.0145, 0.0162, 0.0164, 0.0143, 0.0160, 0.0132], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:14:32,577 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:14:35,259 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7708, 2.5647, 2.7463, 3.5529, 3.8069, 2.9247, 2.5510, 3.8267], + device='cuda:3'), covar=tensor([0.0433, 0.3367, 0.2641, 0.3944, 0.0966, 0.2881, 0.2456, 0.0478], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0204, 0.0320, 0.0218, 0.0214, 0.0196, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:14:54,615 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 21:14:58,064 INFO [train.py:876] (3/4) Epoch 6, batch 4100, loss[loss=0.1579, simple_loss=0.1653, pruned_loss=0.07526, over 5543.00 frames. ], tot_loss[loss=0.16, simple_loss=0.1701, pruned_loss=0.07495, over 1082177.90 frames. ], batch size: 21, lr: 1.33e-02, grad_scale: 16.0 +2022-11-15 21:15:04,547 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:15:04,629 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:15:19,106 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:15:36,005 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.881e+02 2.239e+02 2.711e+02 4.699e+02, threshold=4.478e+02, percent-clipped=1.0 +2022-11-15 21:15:37,413 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:15:59,205 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:16:00,605 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:16:01,606 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 21:16:05,971 INFO [train.py:876] (3/4) Epoch 6, batch 4200, loss[loss=0.1151, simple_loss=0.142, pruned_loss=0.04415, over 5579.00 frames. ], tot_loss[loss=0.1605, simple_loss=0.1711, pruned_loss=0.07497, over 1084705.32 frames. ], batch size: 14, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:16:14,946 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:16:29,487 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5771, 4.1504, 3.1420, 1.7729, 3.9513, 1.2824, 3.8079, 2.0520], + device='cuda:3'), covar=tensor([0.1157, 0.0119, 0.0535, 0.1959, 0.0170, 0.1995, 0.0164, 0.1657], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0101, 0.0111, 0.0118, 0.0103, 0.0128, 0.0095, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:16:30,870 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8691, 1.0903, 1.1435, 0.5349, 0.9867, 1.3107, 0.5871, 0.9566], + device='cuda:3'), covar=tensor([0.0044, 0.0035, 0.0041, 0.0050, 0.0034, 0.0035, 0.0067, 0.0044], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0031, 0.0033, 0.0034, 0.0030, 0.0030, 0.0032, 0.0028], + device='cuda:3'), out_proj_covar=tensor([3.1694e-05, 3.1458e-05, 3.0463e-05, 3.0942e-05, 2.7253e-05, 2.5263e-05, + 3.2635e-05, 2.5497e-05], device='cuda:3') +2022-11-15 21:16:41,010 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9559, 3.2491, 2.8411, 3.2117, 3.2108, 2.7853, 2.8033, 2.7940], + device='cuda:3'), covar=tensor([0.0972, 0.0585, 0.1619, 0.0497, 0.0503, 0.0585, 0.0912, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0154, 0.0243, 0.0149, 0.0182, 0.0156, 0.0164, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:16:44,497 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 1.657e+02 2.106e+02 2.806e+02 4.425e+02, threshold=4.213e+02, percent-clipped=0.0 +2022-11-15 21:16:52,163 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:17:08,276 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:17:14,810 INFO [train.py:876] (3/4) Epoch 6, batch 4300, loss[loss=0.1824, simple_loss=0.179, pruned_loss=0.09294, over 5411.00 frames. ], tot_loss[loss=0.1616, simple_loss=0.1722, pruned_loss=0.07552, over 1084897.75 frames. ], batch size: 70, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:17:18,277 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 +2022-11-15 21:17:32,061 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:17:49,666 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:17:52,814 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.535e+01 1.873e+02 2.249e+02 2.828e+02 5.659e+02, threshold=4.498e+02, percent-clipped=5.0 +2022-11-15 21:18:23,386 INFO [train.py:876] (3/4) Epoch 6, batch 4400, loss[loss=0.2188, simple_loss=0.205, pruned_loss=0.1162, over 4703.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.1712, pruned_loss=0.07494, over 1082543.03 frames. ], batch size: 135, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:18:55,495 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3359, 1.1895, 1.0461, 1.1972, 2.1603, 1.0035, 1.4618, 1.5587], + device='cuda:3'), covar=tensor([0.1127, 0.0749, 0.1195, 0.2662, 0.1540, 0.1630, 0.1049, 0.1160], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0013, 0.0010, 0.0011, 0.0011, 0.0010, 0.0012, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.4945e-05, 5.8271e-05, 4.5727e-05, 5.1940e-05, 4.8554e-05, 4.4570e-05, + 5.3599e-05, 4.5743e-05], device='cuda:3') +2022-11-15 21:19:01,938 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.711e+02 2.029e+02 2.549e+02 3.838e+02, threshold=4.057e+02, percent-clipped=0.0 +2022-11-15 21:19:23,027 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:19:24,987 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:19:32,349 INFO [train.py:876] (3/4) Epoch 6, batch 4500, loss[loss=0.1105, simple_loss=0.1402, pruned_loss=0.04042, over 5550.00 frames. ], tot_loss[loss=0.1608, simple_loss=0.1712, pruned_loss=0.07525, over 1081972.83 frames. ], batch size: 15, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:19:35,166 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:19:41,105 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:19:47,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0388, 2.8991, 2.3495, 1.4743, 2.8443, 1.2847, 2.8222, 1.6164], + device='cuda:3'), covar=tensor([0.1099, 0.0198, 0.0888, 0.1931, 0.0225, 0.1906, 0.0228, 0.1595], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0101, 0.0114, 0.0121, 0.0106, 0.0128, 0.0097, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:19:58,082 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:19:58,866 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:20:04,423 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3572, 4.0641, 3.1576, 1.8170, 3.9386, 1.5567, 3.7987, 1.9622], + device='cuda:3'), covar=tensor([0.1440, 0.0117, 0.0585, 0.2189, 0.0127, 0.1932, 0.0149, 0.1809], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0101, 0.0113, 0.0120, 0.0105, 0.0127, 0.0096, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:20:04,698 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 21:20:06,376 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6895, 3.0251, 2.2523, 2.8293, 1.9395, 2.2725, 1.6911, 2.6567], + device='cuda:3'), covar=tensor([0.1569, 0.0237, 0.0889, 0.0354, 0.1095, 0.0916, 0.1771, 0.0365], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0133, 0.0173, 0.0140, 0.0171, 0.0184, 0.0184, 0.0142], + 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') +2022-11-15 21:20:07,613 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8437, 4.8488, 5.0532, 4.9759, 4.4662, 4.1867, 5.5762, 4.6818], + device='cuda:3'), covar=tensor([0.0288, 0.0995, 0.0224, 0.1215, 0.0584, 0.0256, 0.0776, 0.0436], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0093, 0.0076, 0.0098, 0.0074, 0.0063, 0.0121, 0.0081], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:20:10,191 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 1.848e+02 2.167e+02 2.574e+02 4.684e+02, threshold=4.333e+02, percent-clipped=2.0 +2022-11-15 21:20:13,544 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:20:16,360 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:20:17,589 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:20:39,752 INFO [train.py:876] (3/4) Epoch 6, batch 4600, loss[loss=0.1511, simple_loss=0.1731, pruned_loss=0.06451, over 5670.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.1726, pruned_loss=0.07655, over 1082839.35 frames. ], batch size: 19, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:20:39,922 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 21:20:50,074 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:20:56,729 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:04,224 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:05,537 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3957, 3.6207, 3.5128, 3.3042, 3.5086, 3.4783, 1.2188, 3.6415], + device='cuda:3'), covar=tensor([0.0346, 0.0272, 0.0324, 0.0336, 0.0324, 0.0363, 0.3380, 0.0301], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0076, 0.0076, 0.0067, 0.0090, 0.0078, 0.0124, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:21:11,462 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:11,549 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6911, 0.4191, 0.5898, 0.5426, 0.6772, 0.6514, 0.4013, 0.6553], + device='cuda:3'), covar=tensor([0.0209, 0.0244, 0.0191, 0.0204, 0.0197, 0.0190, 0.0408, 0.0166], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0011, 0.0010, 0.0009, 0.0012, 0.0009], + device='cuda:3'), out_proj_covar=tensor([4.3928e-05, 5.6393e-05, 4.4347e-05, 5.0488e-05, 4.7263e-05, 4.2893e-05, + 5.2232e-05, 4.4253e-05], device='cuda:3') +2022-11-15 21:21:11,888 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2022-11-15 21:21:17,996 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 1.825e+02 2.193e+02 2.554e+02 5.272e+02, threshold=4.385e+02, percent-clipped=4.0 +2022-11-15 21:21:29,552 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:36,932 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:45,919 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:21:46,606 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6656, 1.9878, 3.0940, 2.7062, 3.2889, 2.1232, 3.0060, 3.6023], + device='cuda:3'), covar=tensor([0.0480, 0.1353, 0.0625, 0.1209, 0.0604, 0.1191, 0.0811, 0.0602], + device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0186, 0.0188, 0.0204, 0.0190, 0.0188, 0.0222, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:21:48,093 INFO [train.py:876] (3/4) Epoch 6, batch 4700, loss[loss=0.1135, simple_loss=0.1406, pruned_loss=0.04316, over 5548.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.173, pruned_loss=0.07737, over 1074122.68 frames. ], batch size: 14, lr: 1.32e-02, grad_scale: 16.0 +2022-11-15 21:21:59,443 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2022-11-15 21:22:04,666 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4079, 1.0035, 0.9735, 0.8233, 1.2948, 1.0928, 1.0455, 1.0595], + device='cuda:3'), covar=tensor([0.1101, 0.0460, 0.0524, 0.1482, 0.0902, 0.1307, 0.0679, 0.0554], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0011, 0.0010, 0.0009, 0.0012, 0.0009], + device='cuda:3'), out_proj_covar=tensor([4.3641e-05, 5.5610e-05, 4.4046e-05, 5.0126e-05, 4.6538e-05, 4.2328e-05, + 5.1790e-05, 4.3637e-05], device='cuda:3') +2022-11-15 21:22:07,947 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2521, 4.6205, 4.2744, 4.3390, 4.3382, 4.4231, 1.8127, 4.5857], + device='cuda:3'), covar=tensor([0.0337, 0.0249, 0.0263, 0.0183, 0.0266, 0.0269, 0.3034, 0.0268], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0076, 0.0077, 0.0067, 0.0091, 0.0079, 0.0127, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:22:16,163 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:22:18,827 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:22:26,237 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.964e+01 1.808e+02 2.339e+02 2.942e+02 4.729e+02, threshold=4.678e+02, percent-clipped=2.0 +2022-11-15 21:22:27,678 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5911, 3.7630, 3.4141, 3.4730, 3.5300, 3.6479, 1.2333, 3.7376], + device='cuda:3'), covar=tensor([0.0283, 0.0196, 0.0317, 0.0296, 0.0347, 0.0335, 0.3273, 0.0386], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0076, 0.0077, 0.0067, 0.0090, 0.0079, 0.0126, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:22:47,071 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:22:56,341 INFO [train.py:876] (3/4) Epoch 6, batch 4800, loss[loss=0.08875, simple_loss=0.1115, pruned_loss=0.03301, over 5235.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.1721, pruned_loss=0.07637, over 1075447.66 frames. ], batch size: 7, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:22:57,811 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:23:19,866 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:23:25,226 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.92 vs. limit=5.0 +2022-11-15 21:23:35,273 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.896e+02 2.366e+02 3.104e+02 6.971e+02, threshold=4.733e+02, percent-clipped=2.0 +2022-11-15 21:23:37,418 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:23:38,718 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8790, 4.1943, 3.2205, 1.9695, 4.0612, 1.5490, 3.9106, 2.3060], + device='cuda:3'), covar=tensor([0.0958, 0.0103, 0.0685, 0.1776, 0.0140, 0.1863, 0.0188, 0.1450], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0102, 0.0113, 0.0119, 0.0106, 0.0128, 0.0098, 0.0120], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:23:42,088 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4145, 2.9289, 2.9776, 1.4267, 2.7638, 3.1034, 3.2796, 3.3728], + device='cuda:3'), covar=tensor([0.1762, 0.1319, 0.0818, 0.2654, 0.0372, 0.0536, 0.0194, 0.0725], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0190, 0.0147, 0.0195, 0.0162, 0.0165, 0.0139, 0.0178], + 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') +2022-11-15 21:24:00,968 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 21:24:04,477 INFO [train.py:876] (3/4) Epoch 6, batch 4900, loss[loss=0.0959, simple_loss=0.1234, pruned_loss=0.03422, over 5118.00 frames. ], tot_loss[loss=0.1605, simple_loss=0.1709, pruned_loss=0.07502, over 1078689.64 frames. ], batch size: 7, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:24:10,511 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 21:24:35,584 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:24:43,023 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.647e+02 1.947e+02 2.444e+02 4.412e+02, threshold=3.894e+02, percent-clipped=0.0 +2022-11-15 21:25:01,408 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0862, 1.2630, 1.5293, 1.1303, 0.8965, 1.8181, 1.1774, 0.9440], + device='cuda:3'), covar=tensor([0.1311, 0.0812, 0.0857, 0.1862, 0.2415, 0.0541, 0.1309, 0.1913], + device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0045, 0.0049, 0.0061, 0.0050, 0.0040, 0.0045, 0.0050], + device='cuda:3'), out_proj_covar=tensor([1.1600e-04, 9.5728e-05, 1.0100e-04, 1.2595e-04, 1.0705e-04, 9.1414e-05, + 9.7752e-05, 1.0262e-04], device='cuda:3') +2022-11-15 21:25:07,256 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:25:08,583 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:25:12,444 INFO [train.py:876] (3/4) Epoch 6, batch 5000, loss[loss=0.1763, simple_loss=0.186, pruned_loss=0.08328, over 5749.00 frames. ], tot_loss[loss=0.1616, simple_loss=0.1718, pruned_loss=0.07569, over 1083381.70 frames. ], batch size: 31, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:25:39,906 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:25:51,843 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.916e+02 2.252e+02 2.845e+02 5.364e+02, threshold=4.504e+02, percent-clipped=7.0 +2022-11-15 21:25:58,687 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2022-11-15 21:26:18,844 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:26:20,687 INFO [train.py:876] (3/4) Epoch 6, batch 5100, loss[loss=0.2329, simple_loss=0.1954, pruned_loss=0.1352, over 4141.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.172, pruned_loss=0.07571, over 1080671.84 frames. ], batch size: 181, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:26:23,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2022-11-15 21:26:48,108 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-15 21:26:59,644 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.689e+02 2.100e+02 2.600e+02 5.796e+02, threshold=4.200e+02, percent-clipped=2.0 +2022-11-15 21:27:02,075 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:27:25,843 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:27:28,987 INFO [train.py:876] (3/4) Epoch 6, batch 5200, loss[loss=0.198, simple_loss=0.215, pruned_loss=0.09053, over 5564.00 frames. ], tot_loss[loss=0.1596, simple_loss=0.1708, pruned_loss=0.07423, over 1088192.05 frames. ], batch size: 21, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:27:34,217 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:27:35,491 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 21:27:38,113 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4302, 5.0193, 4.5196, 5.0198, 5.0557, 4.3181, 4.6568, 4.2210], + device='cuda:3'), covar=tensor([0.0256, 0.0540, 0.1422, 0.0361, 0.0328, 0.0392, 0.0541, 0.0574], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0155, 0.0241, 0.0148, 0.0185, 0.0156, 0.0165, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:27:58,403 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:28:08,059 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.794e+02 2.245e+02 2.810e+02 6.868e+02, threshold=4.491e+02, percent-clipped=3.0 +2022-11-15 21:28:32,314 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:28:33,028 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:28:35,002 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1221, 3.7824, 3.9612, 3.7235, 4.2025, 3.6650, 3.7311, 4.1172], + device='cuda:3'), covar=tensor([0.0293, 0.0283, 0.0329, 0.0332, 0.0283, 0.0461, 0.0290, 0.0285], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0117, 0.0089, 0.0119, 0.0127, 0.0075, 0.0101, 0.0115], + 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') +2022-11-15 21:28:37,522 INFO [train.py:876] (3/4) Epoch 6, batch 5300, loss[loss=0.1696, simple_loss=0.1905, pruned_loss=0.07432, over 5748.00 frames. ], tot_loss[loss=0.1604, simple_loss=0.1715, pruned_loss=0.0747, over 1094917.67 frames. ], batch size: 20, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:29:03,992 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2022-11-15 21:29:05,168 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:29:05,243 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:29:11,931 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-15 21:29:15,039 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:29:16,828 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 1.882e+02 2.155e+02 2.589e+02 5.849e+02, threshold=4.310e+02, percent-clipped=2.0 +2022-11-15 21:29:36,858 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-15 21:29:38,525 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:29:44,837 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:29:46,714 INFO [train.py:876] (3/4) Epoch 6, batch 5400, loss[loss=0.173, simple_loss=0.1803, pruned_loss=0.08283, over 5728.00 frames. ], tot_loss[loss=0.1596, simple_loss=0.1704, pruned_loss=0.07439, over 1086236.61 frames. ], batch size: 27, lr: 1.31e-02, grad_scale: 16.0 +2022-11-15 21:30:17,685 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:30:26,061 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.844e+02 2.196e+02 2.605e+02 6.100e+02, threshold=4.391e+02, percent-clipped=2.0 +2022-11-15 21:30:31,175 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.74 vs. limit=5.0 +2022-11-15 21:30:55,102 INFO [train.py:876] (3/4) Epoch 6, batch 5500, loss[loss=0.1119, simple_loss=0.1414, pruned_loss=0.0412, over 5604.00 frames. ], tot_loss[loss=0.1601, simple_loss=0.1704, pruned_loss=0.07483, over 1078947.73 frames. ], batch size: 18, lr: 1.30e-02, grad_scale: 16.0 +2022-11-15 21:31:14,798 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8529, 3.7311, 2.3541, 3.6129, 3.0146, 2.6093, 2.0293, 3.1012], + device='cuda:3'), covar=tensor([0.2459, 0.0439, 0.1731, 0.0462, 0.0838, 0.1523, 0.2460, 0.0513], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0133, 0.0169, 0.0136, 0.0169, 0.0180, 0.0180, 0.0142], + 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') +2022-11-15 21:31:20,341 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.76 vs. limit=5.0 +2022-11-15 21:31:25,962 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:31:29,689 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5101, 3.8459, 3.7851, 3.3709, 2.4863, 4.2775, 2.5524, 3.5966], + device='cuda:3'), covar=tensor([0.0383, 0.0296, 0.0128, 0.0337, 0.0444, 0.0099, 0.0337, 0.0106], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0136, 0.0149, 0.0164, 0.0168, 0.0145, 0.0160, 0.0133], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:31:33,899 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.786e+02 2.219e+02 2.876e+02 6.539e+02, threshold=4.438e+02, percent-clipped=3.0 +2022-11-15 21:31:54,707 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 21:32:03,400 INFO [train.py:876] (3/4) Epoch 6, batch 5600, loss[loss=0.1162, simple_loss=0.1371, pruned_loss=0.04767, over 5481.00 frames. ], tot_loss[loss=0.1597, simple_loss=0.1703, pruned_loss=0.07455, over 1080050.63 frames. ], batch size: 10, lr: 1.30e-02, grad_scale: 16.0 +2022-11-15 21:32:07,836 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:32:36,385 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:32:36,944 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 21:32:42,423 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.893e+02 2.253e+02 2.739e+02 4.723e+02, threshold=4.505e+02, percent-clipped=2.0 +2022-11-15 21:32:43,610 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8630, 3.2400, 2.1886, 2.9889, 2.2194, 2.4381, 1.8045, 2.8419], + device='cuda:3'), covar=tensor([0.1509, 0.0206, 0.1041, 0.0284, 0.0955, 0.0959, 0.1795, 0.0315], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0134, 0.0171, 0.0137, 0.0171, 0.0181, 0.0180, 0.0143], + 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') +2022-11-15 21:33:11,598 INFO [train.py:876] (3/4) Epoch 6, batch 5700, loss[loss=0.1275, simple_loss=0.1416, pruned_loss=0.05667, over 4552.00 frames. ], tot_loss[loss=0.1577, simple_loss=0.1695, pruned_loss=0.07297, over 1077390.33 frames. ], batch size: 5, lr: 1.30e-02, grad_scale: 16.0 +2022-11-15 21:33:25,096 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2022-11-15 21:33:51,625 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.803e+02 2.125e+02 2.518e+02 5.093e+02, threshold=4.250e+02, percent-clipped=1.0 +2022-11-15 21:34:14,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-15 21:34:21,498 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.67 vs. limit=5.0 +2022-11-15 21:34:23,229 INFO [train.py:876] (3/4) Epoch 6, batch 5800, loss[loss=0.155, simple_loss=0.1719, pruned_loss=0.06907, over 5643.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.17, pruned_loss=0.07329, over 1084279.13 frames. ], batch size: 38, lr: 1.30e-02, grad_scale: 16.0 +2022-11-15 21:34:28,937 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9108, 2.0932, 2.3658, 2.1528, 1.3856, 2.1242, 1.4309, 1.5935], + device='cuda:3'), covar=tensor([0.0146, 0.0060, 0.0062, 0.0084, 0.0187, 0.0068, 0.0192, 0.0107], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0135, 0.0149, 0.0165, 0.0170, 0.0146, 0.0160, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:34:57,162 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-15 21:35:01,870 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.784e+02 2.140e+02 2.674e+02 6.368e+02, threshold=4.280e+02, percent-clipped=2.0 +2022-11-15 21:35:31,645 INFO [train.py:876] (3/4) Epoch 6, batch 5900, loss[loss=0.1863, simple_loss=0.1751, pruned_loss=0.0987, over 4211.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.17, pruned_loss=0.07429, over 1082019.17 frames. ], batch size: 181, lr: 1.30e-02, grad_scale: 16.0 +2022-11-15 21:35:32,379 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:35:56,155 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1544, 4.6893, 4.1703, 4.7414, 4.7856, 3.8505, 4.1894, 3.8865], + device='cuda:3'), covar=tensor([0.0370, 0.0621, 0.1765, 0.0536, 0.0473, 0.0636, 0.1208, 0.1476], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0156, 0.0244, 0.0152, 0.0187, 0.0156, 0.0166, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:36:00,789 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:36:05,201 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 21:36:10,905 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.865e+02 2.230e+02 2.875e+02 6.515e+02, threshold=4.459e+02, percent-clipped=7.0 +2022-11-15 21:36:23,000 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4159, 3.9498, 3.5569, 3.3812, 2.1463, 3.8989, 2.1278, 3.1749], + device='cuda:3'), covar=tensor([0.0414, 0.0127, 0.0154, 0.0318, 0.0451, 0.0109, 0.0412, 0.0111], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0135, 0.0147, 0.0162, 0.0165, 0.0144, 0.0157, 0.0132], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:36:37,638 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:36:39,540 INFO [train.py:876] (3/4) Epoch 6, batch 6000, loss[loss=0.1542, simple_loss=0.1743, pruned_loss=0.06707, over 5561.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.171, pruned_loss=0.07477, over 1081284.97 frames. ], batch size: 16, lr: 1.30e-02, grad_scale: 8.0 +2022-11-15 21:36:39,540 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 21:36:50,453 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5140, 4.1289, 3.6132, 3.4943, 2.1735, 3.8595, 2.0691, 3.2066], + device='cuda:3'), covar=tensor([0.0439, 0.0101, 0.0171, 0.0336, 0.0479, 0.0121, 0.0485, 0.0140], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0135, 0.0147, 0.0162, 0.0166, 0.0145, 0.0158, 0.0132], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:36:57,490 INFO [train.py:908] (3/4) Epoch 6, validation: loss=0.1626, simple_loss=0.1837, pruned_loss=0.07077, over 1530663.00 frames. +2022-11-15 21:36:57,490 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 21:37:04,033 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9332, 2.0919, 3.2827, 2.8222, 3.8983, 2.2867, 3.2572, 3.8941], + device='cuda:3'), covar=tensor([0.0480, 0.1734, 0.0753, 0.1610, 0.0285, 0.1452, 0.1064, 0.0629], + device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0193, 0.0196, 0.0213, 0.0195, 0.0190, 0.0232, 0.0217], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:37:09,886 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:37:37,847 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.691e+02 2.055e+02 2.547e+02 4.668e+02, threshold=4.111e+02, percent-clipped=1.0 +2022-11-15 21:37:52,182 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:38:06,269 INFO [train.py:876] (3/4) Epoch 6, batch 6100, loss[loss=0.1353, simple_loss=0.1624, pruned_loss=0.05409, over 5618.00 frames. ], tot_loss[loss=0.1586, simple_loss=0.1702, pruned_loss=0.07351, over 1078756.93 frames. ], batch size: 24, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:38:26,907 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 21:38:35,783 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1260, 5.4661, 3.9695, 2.6021, 5.2571, 2.5517, 4.7028, 3.1812], + device='cuda:3'), covar=tensor([0.0816, 0.0075, 0.0476, 0.2039, 0.0121, 0.1388, 0.0156, 0.1555], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0102, 0.0115, 0.0120, 0.0107, 0.0128, 0.0098, 0.0121], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:38:37,143 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:38:45,892 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.891e+02 2.304e+02 2.895e+02 7.790e+02, threshold=4.608e+02, percent-clipped=4.0 +2022-11-15 21:38:46,355 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-15 21:39:14,512 INFO [train.py:876] (3/4) Epoch 6, batch 6200, loss[loss=0.166, simple_loss=0.1734, pruned_loss=0.07933, over 5698.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.1694, pruned_loss=0.07338, over 1079979.07 frames. ], batch size: 17, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:39:15,276 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:39:18,953 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:39:44,655 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 21:39:48,431 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:39:54,749 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.838e+01 1.745e+02 2.107e+02 2.704e+02 4.645e+02, threshold=4.215e+02, percent-clipped=1.0 +2022-11-15 21:40:06,654 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5074, 3.5157, 3.2999, 3.1982, 2.0586, 3.6114, 2.1714, 2.9423], + device='cuda:3'), covar=tensor([0.0253, 0.0100, 0.0133, 0.0210, 0.0314, 0.0094, 0.0367, 0.0091], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0137, 0.0150, 0.0166, 0.0168, 0.0147, 0.0163, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:40:15,369 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:40:17,233 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 21:40:23,151 INFO [train.py:876] (3/4) Epoch 6, batch 6300, loss[loss=0.08187, simple_loss=0.1104, pruned_loss=0.02667, over 5137.00 frames. ], tot_loss[loss=0.1585, simple_loss=0.1691, pruned_loss=0.07394, over 1080428.99 frames. ], batch size: 6, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:40:55,894 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:41:02,049 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.507e+01 1.912e+02 2.400e+02 2.974e+02 5.468e+02, threshold=4.801e+02, percent-clipped=3.0 +2022-11-15 21:41:13,144 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:41:28,045 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.01 vs. limit=5.0 +2022-11-15 21:41:30,146 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4030, 4.7658, 3.0058, 4.4439, 3.5186, 2.9948, 2.4023, 3.9773], + device='cuda:3'), covar=tensor([0.1636, 0.0130, 0.0889, 0.0239, 0.0568, 0.1101, 0.1941, 0.0227], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0133, 0.0170, 0.0138, 0.0173, 0.0181, 0.0183, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 21:41:31,361 INFO [train.py:876] (3/4) Epoch 6, batch 6400, loss[loss=0.1428, simple_loss=0.1675, pruned_loss=0.05904, over 5764.00 frames. ], tot_loss[loss=0.1558, simple_loss=0.1678, pruned_loss=0.07189, over 1082092.92 frames. ], batch size: 16, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:41:43,432 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 +2022-11-15 21:41:53,547 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1872, 0.9560, 0.9983, 0.8191, 1.0496, 1.1881, 0.8566, 0.8085], + device='cuda:3'), covar=tensor([0.1903, 0.0769, 0.0847, 0.1594, 0.1715, 0.1750, 0.0801, 0.0933], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0014, 0.0011, 0.0013, 0.0011, 0.0010, 0.0014, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.8515e-05, 6.3542e-05, 4.9161e-05, 5.8049e-05, 5.3065e-05, 4.7464e-05, + 5.9038e-05, 4.9088e-05], device='cuda:3') +2022-11-15 21:42:11,496 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.772e+02 2.214e+02 2.586e+02 5.921e+02, threshold=4.429e+02, percent-clipped=3.0 +2022-11-15 21:42:12,992 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9359, 1.1860, 0.9926, 0.9385, 0.8377, 1.4404, 0.9959, 0.7297], + device='cuda:3'), covar=tensor([0.1501, 0.0325, 0.1285, 0.2014, 0.1815, 0.0400, 0.1403, 0.1604], + device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0046, 0.0051, 0.0063, 0.0049, 0.0040, 0.0045, 0.0051], + device='cuda:3'), out_proj_covar=tensor([1.1842e-04, 9.6874e-05, 1.0533e-04, 1.3109e-04, 1.0725e-04, 9.1767e-05, + 9.8838e-05, 1.0587e-04], device='cuda:3') +2022-11-15 21:42:13,736 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3984, 3.2889, 3.0085, 1.6689, 2.9234, 3.3549, 3.2480, 3.8489], + device='cuda:3'), covar=tensor([0.1683, 0.1188, 0.0905, 0.2613, 0.0503, 0.0673, 0.0234, 0.0507], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0183, 0.0143, 0.0193, 0.0157, 0.0165, 0.0132, 0.0176], + 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') +2022-11-15 21:42:23,770 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2190, 1.8259, 2.1020, 2.3942, 2.6707, 2.0449, 1.6507, 2.5921], + device='cuda:3'), covar=tensor([0.1028, 0.2200, 0.1517, 0.0940, 0.0789, 0.2048, 0.1737, 0.0827], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0207, 0.0198, 0.0322, 0.0219, 0.0212, 0.0193, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:42:23,798 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0819, 2.5181, 2.7305, 2.5748, 1.6383, 2.6408, 1.6633, 2.0125], + device='cuda:3'), covar=tensor([0.0236, 0.0110, 0.0101, 0.0119, 0.0256, 0.0112, 0.0291, 0.0131], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0137, 0.0150, 0.0165, 0.0169, 0.0150, 0.0162, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:42:38,683 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:42:39,913 INFO [train.py:876] (3/4) Epoch 6, batch 6500, loss[loss=0.1145, simple_loss=0.125, pruned_loss=0.05198, over 5045.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.1704, pruned_loss=0.07392, over 1081787.30 frames. ], batch size: 7, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:42:40,664 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:42:52,374 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:43:19,261 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.824e+02 2.207e+02 2.735e+02 7.842e+02, threshold=4.414e+02, percent-clipped=4.0 +2022-11-15 21:43:20,115 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:43:33,787 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:43:44,285 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-15 21:43:47,834 INFO [train.py:876] (3/4) Epoch 6, batch 6600, loss[loss=0.1259, simple_loss=0.1532, pruned_loss=0.04932, over 5741.00 frames. ], tot_loss[loss=0.1601, simple_loss=0.1713, pruned_loss=0.07443, over 1086392.55 frames. ], batch size: 13, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:44:18,657 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:44:23,737 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-15 21:44:25,420 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7915, 1.9189, 3.2871, 2.7138, 4.0069, 2.0058, 3.1489, 3.7449], + device='cuda:3'), covar=tensor([0.0502, 0.1646, 0.0588, 0.1782, 0.0301, 0.1685, 0.1151, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0193, 0.0189, 0.0211, 0.0194, 0.0186, 0.0228, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:44:28,196 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.711e+02 2.097e+02 2.455e+02 4.370e+02, threshold=4.194e+02, percent-clipped=0.0 +2022-11-15 21:44:39,655 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:44:56,951 INFO [train.py:876] (3/4) Epoch 6, batch 6700, loss[loss=0.1454, simple_loss=0.1695, pruned_loss=0.06069, over 5765.00 frames. ], tot_loss[loss=0.161, simple_loss=0.1721, pruned_loss=0.07496, over 1085499.82 frames. ], batch size: 27, lr: 1.29e-02, grad_scale: 8.0 +2022-11-15 21:45:12,513 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:45:19,674 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-15 21:45:27,049 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 +2022-11-15 21:45:28,346 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 21:45:36,325 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.495e+01 1.976e+02 2.503e+02 2.987e+02 6.000e+02, threshold=5.005e+02, percent-clipped=7.0 +2022-11-15 21:46:00,826 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 21:46:04,886 INFO [train.py:876] (3/4) Epoch 6, batch 6800, loss[loss=0.1897, simple_loss=0.2002, pruned_loss=0.08959, over 5672.00 frames. ], tot_loss[loss=0.1598, simple_loss=0.1715, pruned_loss=0.07402, over 1091508.91 frames. ], batch size: 34, lr: 1.28e-02, grad_scale: 8.0 +2022-11-15 21:46:05,686 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:46:38,997 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:46:42,328 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:46:44,823 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.755e+02 2.083e+02 2.590e+02 5.758e+02, threshold=4.166e+02, percent-clipped=2.0 +2022-11-15 21:46:45,737 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4620, 3.6190, 3.2427, 3.2970, 2.1230, 3.5432, 2.0242, 3.0979], + device='cuda:3'), covar=tensor([0.0352, 0.0119, 0.0185, 0.0212, 0.0381, 0.0118, 0.0398, 0.0102], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0141, 0.0154, 0.0170, 0.0173, 0.0153, 0.0165, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:46:52,922 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 21:46:55,886 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:47:10,957 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2022-11-15 21:47:13,946 INFO [train.py:876] (3/4) Epoch 6, batch 6900, loss[loss=0.1537, simple_loss=0.1757, pruned_loss=0.06581, over 5540.00 frames. ], tot_loss[loss=0.1605, simple_loss=0.1723, pruned_loss=0.07438, over 1091424.32 frames. ], batch size: 14, lr: 1.28e-02, grad_scale: 8.0 +2022-11-15 21:47:21,768 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2022-11-15 21:47:44,595 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:47:46,672 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9043, 0.7517, 0.8684, 0.4847, 0.7348, 0.9452, 0.5566, 0.7447], + device='cuda:3'), covar=tensor([0.0822, 0.1511, 0.1099, 0.2914, 0.2835, 0.0795, 0.2264, 0.1040], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0014, 0.0011, 0.0013, 0.0011, 0.0010, 0.0013, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.8808e-05, 6.2885e-05, 4.9407e-05, 5.8725e-05, 5.3200e-05, 4.8960e-05, + 5.9056e-05, 4.9268e-05], device='cuda:3') +2022-11-15 21:47:53,738 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 1.786e+02 2.260e+02 2.890e+02 4.786e+02, threshold=4.520e+02, percent-clipped=4.0 +2022-11-15 21:48:13,033 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1901, 2.9403, 2.7118, 1.4214, 2.7219, 3.0710, 2.7936, 3.3851], + device='cuda:3'), covar=tensor([0.1805, 0.1430, 0.0718, 0.2667, 0.0403, 0.0668, 0.0269, 0.0507], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0186, 0.0143, 0.0188, 0.0159, 0.0166, 0.0135, 0.0176], + 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') +2022-11-15 21:48:16,707 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:48:18,483 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 21:48:22,248 INFO [train.py:876] (3/4) Epoch 6, batch 7000, loss[loss=0.1415, simple_loss=0.1633, pruned_loss=0.05988, over 5558.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.1711, pruned_loss=0.07375, over 1089479.06 frames. ], batch size: 16, lr: 1.28e-02, grad_scale: 8.0 +2022-11-15 21:48:44,359 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0945, 2.1094, 2.4157, 3.2539, 3.2064, 2.4285, 1.8769, 3.3796], + device='cuda:3'), covar=tensor([0.0595, 0.2733, 0.2448, 0.2304, 0.1008, 0.2335, 0.2246, 0.0561], + device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0213, 0.0202, 0.0325, 0.0220, 0.0215, 0.0194, 0.0200], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:48:46,604 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2022-11-15 21:49:02,128 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.834e+02 2.170e+02 2.645e+02 5.568e+02, threshold=4.340e+02, percent-clipped=1.0 +2022-11-15 21:49:04,629 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9443, 1.4773, 0.9897, 0.9571, 1.4845, 1.0789, 0.9123, 1.4126], + device='cuda:3'), covar=tensor([0.0032, 0.0026, 0.0047, 0.0035, 0.0022, 0.0023, 0.0037, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0032, 0.0035, 0.0034, 0.0031, 0.0029, 0.0034, 0.0028], + device='cuda:3'), out_proj_covar=tensor([3.3334e-05, 3.0835e-05, 3.2200e-05, 3.0926e-05, 2.7158e-05, 2.5052e-05, + 3.3468e-05, 2.4607e-05], device='cuda:3') +2022-11-15 21:49:14,192 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 +2022-11-15 21:49:22,544 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4764, 4.6598, 4.8311, 5.0048, 4.2931, 4.0416, 5.3245, 4.6353], + device='cuda:3'), covar=tensor([0.0385, 0.0919, 0.0235, 0.0788, 0.0499, 0.0256, 0.0618, 0.0422], + device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0095, 0.0080, 0.0102, 0.0078, 0.0067, 0.0128, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:49:30,778 INFO [train.py:876] (3/4) Epoch 6, batch 7100, loss[loss=0.1557, simple_loss=0.1747, pruned_loss=0.06837, over 5733.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.1709, pruned_loss=0.07401, over 1082731.99 frames. ], batch size: 27, lr: 1.28e-02, grad_scale: 8.0 +2022-11-15 21:49:37,239 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0614, 1.8935, 2.7535, 2.4340, 2.5500, 1.7840, 2.5190, 3.1148], + device='cuda:3'), covar=tensor([0.0464, 0.1205, 0.0556, 0.1090, 0.0538, 0.1201, 0.0851, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0192, 0.0192, 0.0212, 0.0195, 0.0186, 0.0226, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:49:41,430 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1822, 3.6331, 3.2272, 3.7067, 3.7379, 3.1037, 3.2715, 3.1066], + device='cuda:3'), covar=tensor([0.0919, 0.0568, 0.1523, 0.0437, 0.0430, 0.0524, 0.0557, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0154, 0.0241, 0.0149, 0.0186, 0.0154, 0.0164, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 21:49:56,789 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2813, 4.4154, 4.5454, 4.7824, 4.2649, 3.4637, 4.9697, 4.5518], + device='cuda:3'), covar=tensor([0.0419, 0.0831, 0.0359, 0.1024, 0.0480, 0.0538, 0.0907, 0.0477], + device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0095, 0.0080, 0.0102, 0.0077, 0.0066, 0.0128, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:49:57,779 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 +2022-11-15 21:50:06,739 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:07,739 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:10,160 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.966e+01 1.749e+02 2.086e+02 2.685e+02 6.877e+02, threshold=4.173e+02, percent-clipped=3.0 +2022-11-15 21:50:21,409 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:38,609 INFO [train.py:876] (3/4) Epoch 6, batch 7200, loss[loss=0.1873, simple_loss=0.1983, pruned_loss=0.08819, over 5617.00 frames. ], tot_loss[loss=0.1597, simple_loss=0.1709, pruned_loss=0.07427, over 1078324.65 frames. ], batch size: 50, lr: 1.28e-02, grad_scale: 8.0 +2022-11-15 21:50:40,025 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:48,398 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:53,858 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:50:58,309 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9421, 2.2850, 2.9644, 3.8064, 3.9946, 3.3755, 2.6137, 4.0137], + device='cuda:3'), covar=tensor([0.0522, 0.3906, 0.2808, 0.3327, 0.0917, 0.2668, 0.2392, 0.0498], + device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0213, 0.0205, 0.0326, 0.0222, 0.0216, 0.0194, 0.0204], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:51:17,828 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.746e+02 2.258e+02 2.937e+02 6.701e+02, threshold=4.517e+02, percent-clipped=5.0 +2022-11-15 21:52:13,099 INFO [train.py:876] (3/4) Epoch 7, batch 0, loss[loss=0.1119, simple_loss=0.142, pruned_loss=0.04093, over 4981.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.142, pruned_loss=0.04093, over 4981.00 frames. ], batch size: 7, lr: 1.20e-02, grad_scale: 8.0 +2022-11-15 21:52:13,099 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 21:52:28,043 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8861, 3.0911, 3.0842, 2.9797, 3.0166, 2.9298, 1.3864, 3.1169], + device='cuda:3'), covar=tensor([0.0211, 0.0138, 0.0153, 0.0126, 0.0231, 0.0232, 0.2118, 0.0193], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0077, 0.0078, 0.0068, 0.0094, 0.0082, 0.0127, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 21:52:29,702 INFO [train.py:908] (3/4) Epoch 7, validation: loss=0.1631, simple_loss=0.1871, pruned_loss=0.06958, over 1530663.00 frames. +2022-11-15 21:52:29,703 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 21:52:48,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9234, 4.7565, 3.6808, 1.9837, 4.4189, 1.7315, 4.4294, 2.2784], + device='cuda:3'), covar=tensor([0.1369, 0.0099, 0.0516, 0.2341, 0.0142, 0.2097, 0.0125, 0.1947], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0101, 0.0112, 0.0117, 0.0103, 0.0127, 0.0095, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 21:53:02,933 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-15 21:53:28,696 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.724e+02 2.075e+02 2.427e+02 4.341e+02, threshold=4.150e+02, percent-clipped=0.0 +2022-11-15 21:53:37,918 INFO [train.py:876] (3/4) Epoch 7, batch 100, loss[loss=0.1323, simple_loss=0.1617, pruned_loss=0.05149, over 5732.00 frames. ], tot_loss[loss=0.1541, simple_loss=0.166, pruned_loss=0.07112, over 430004.99 frames. ], batch size: 20, lr: 1.20e-02, grad_scale: 8.0 +2022-11-15 21:54:15,364 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:54:22,916 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4884, 1.0982, 0.9623, 0.6833, 0.9706, 1.0425, 0.6433, 0.7195], + device='cuda:3'), covar=tensor([0.0390, 0.0586, 0.0764, 0.0887, 0.0888, 0.0343, 0.1000, 0.0475], + device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0013, 0.0010, 0.0012, 0.0011, 0.0010, 0.0013, 0.0009], + device='cuda:3'), out_proj_covar=tensor([4.5717e-05, 5.9080e-05, 4.7314e-05, 5.4754e-05, 4.9972e-05, 4.6261e-05, + 5.5650e-05, 4.5631e-05], device='cuda:3') +2022-11-15 21:54:37,879 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.863e+02 2.234e+02 2.721e+02 4.230e+02, threshold=4.468e+02, percent-clipped=1.0 +2022-11-15 21:54:47,143 INFO [train.py:876] (3/4) Epoch 7, batch 200, loss[loss=0.09358, simple_loss=0.1239, pruned_loss=0.03164, over 5722.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.1675, pruned_loss=0.07134, over 691682.40 frames. ], batch size: 12, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 21:54:57,726 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:55:01,017 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4592, 0.9002, 1.5213, 1.1241, 1.3099, 1.2431, 1.1735, 1.1067], + device='cuda:3'), covar=tensor([0.0556, 0.0671, 0.0642, 0.1065, 0.1881, 0.0862, 0.0913, 0.0522], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0014, 0.0010, 0.0012, 0.0011, 0.0010, 0.0013, 0.0009], + device='cuda:3'), out_proj_covar=tensor([4.6707e-05, 6.0607e-05, 4.8530e-05, 5.5486e-05, 5.1012e-05, 4.7396e-05, + 5.6706e-05, 4.6224e-05], device='cuda:3') +2022-11-15 21:55:01,255 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 21:55:12,858 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:55:38,980 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:55:46,088 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 1.863e+02 2.345e+02 2.603e+02 4.668e+02, threshold=4.689e+02, percent-clipped=1.0 +2022-11-15 21:55:55,858 INFO [train.py:876] (3/4) Epoch 7, batch 300, loss[loss=0.1283, simple_loss=0.158, pruned_loss=0.04935, over 5520.00 frames. ], tot_loss[loss=0.1566, simple_loss=0.1688, pruned_loss=0.07217, over 847925.65 frames. ], batch size: 17, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 21:55:57,666 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 21:56:13,539 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:56:21,725 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5585, 1.0784, 1.7064, 1.2437, 1.4661, 1.5917, 1.3502, 1.1366], + device='cuda:3'), covar=tensor([0.0015, 0.0100, 0.0025, 0.0043, 0.0021, 0.0047, 0.0028, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0018, 0.0017, 0.0020, 0.0019, 0.0016, 0.0019, 0.0020], + device='cuda:3'), out_proj_covar=tensor([1.5885e-05, 1.7752e-05, 1.6441e-05, 2.0658e-05, 1.7782e-05, 1.6787e-05, + 1.8970e-05, 2.1545e-05], device='cuda:3') +2022-11-15 21:56:21,761 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:56:54,567 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 1.760e+02 2.035e+02 2.652e+02 4.590e+02, threshold=4.070e+02, percent-clipped=0.0 +2022-11-15 21:56:55,804 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:57:04,544 INFO [train.py:876] (3/4) Epoch 7, batch 400, loss[loss=0.1644, simple_loss=0.1808, pruned_loss=0.07401, over 5574.00 frames. ], tot_loss[loss=0.156, simple_loss=0.1683, pruned_loss=0.07183, over 938349.21 frames. ], batch size: 25, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 21:57:06,623 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 +2022-11-15 21:57:12,303 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5066, 1.0470, 1.0682, 0.9106, 1.1319, 1.1643, 0.9635, 1.0253], + device='cuda:3'), covar=tensor([0.0948, 0.0615, 0.1104, 0.1340, 0.1878, 0.0554, 0.1612, 0.1611], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0014, 0.0011, 0.0012, 0.0011, 0.0010, 0.0013, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.7599e-05, 6.1353e-05, 4.9247e-05, 5.7102e-05, 5.2539e-05, 4.8122e-05, + 5.7946e-05, 4.8092e-05], device='cuda:3') +2022-11-15 21:58:02,998 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.245e+02 1.796e+02 2.180e+02 2.809e+02 7.513e+02, threshold=4.360e+02, percent-clipped=4.0 +2022-11-15 21:58:13,183 INFO [train.py:876] (3/4) Epoch 7, batch 500, loss[loss=0.1335, simple_loss=0.1587, pruned_loss=0.05418, over 5440.00 frames. ], tot_loss[loss=0.1535, simple_loss=0.1677, pruned_loss=0.06962, over 999276.57 frames. ], batch size: 11, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 21:58:20,229 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:58:37,756 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:59:10,361 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:59:10,882 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.778e+02 2.186e+02 2.838e+02 5.081e+02, threshold=4.371e+02, percent-clipped=2.0 +2022-11-15 21:59:14,282 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3220, 4.0554, 2.8544, 3.9483, 3.0472, 2.9199, 2.1199, 3.4577], + device='cuda:3'), covar=tensor([0.1454, 0.0164, 0.0814, 0.0227, 0.0685, 0.1009, 0.2034, 0.0297], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0134, 0.0168, 0.0138, 0.0176, 0.0181, 0.0184, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 21:59:20,341 INFO [train.py:876] (3/4) Epoch 7, batch 600, loss[loss=0.1106, simple_loss=0.1386, pruned_loss=0.04128, over 5682.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.1686, pruned_loss=0.07083, over 1033934.40 frames. ], batch size: 17, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 21:59:25,336 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 +2022-11-15 21:59:29,624 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9665, 1.7550, 2.6677, 2.3906, 2.4533, 1.8169, 2.2509, 2.8712], + device='cuda:3'), covar=tensor([0.0346, 0.1125, 0.0471, 0.0812, 0.0522, 0.1011, 0.0731, 0.0484], + device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0190, 0.0190, 0.0207, 0.0193, 0.0181, 0.0223, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 21:59:42,363 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 21:59:43,735 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8536, 1.5149, 1.8210, 1.2340, 1.1436, 1.2656, 1.3484, 1.2010], + device='cuda:3'), covar=tensor([0.0019, 0.0038, 0.0016, 0.0028, 0.0050, 0.0067, 0.0020, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0017, 0.0020, 0.0019, 0.0017, 0.0019, 0.0020], + device='cuda:3'), out_proj_covar=tensor([1.5523e-05, 1.7285e-05, 1.6224e-05, 2.0433e-05, 1.8257e-05, 1.6852e-05, + 1.8949e-05, 2.1520e-05], device='cuda:3') +2022-11-15 21:59:48,518 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-15 22:00:16,427 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:00:18,959 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.730e+02 2.067e+02 2.557e+02 5.519e+02, threshold=4.134e+02, percent-clipped=3.0 +2022-11-15 22:00:19,158 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5455, 0.9727, 1.1342, 0.8159, 1.0983, 1.1379, 0.8256, 0.9915], + device='cuda:3'), covar=tensor([0.0289, 0.0506, 0.0491, 0.0738, 0.0519, 0.0235, 0.0757, 0.0471], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0014, 0.0011, 0.0013, 0.0012, 0.0010, 0.0014, 0.0010], + device='cuda:3'), out_proj_covar=tensor([4.9479e-05, 6.3981e-05, 5.0198e-05, 5.8470e-05, 5.4293e-05, 4.9574e-05, + 6.0205e-05, 5.0390e-05], device='cuda:3') +2022-11-15 22:00:28,128 INFO [train.py:876] (3/4) Epoch 7, batch 700, loss[loss=0.2309, simple_loss=0.2081, pruned_loss=0.1268, over 4707.00 frames. ], tot_loss[loss=0.1573, simple_loss=0.1699, pruned_loss=0.07233, over 1054079.45 frames. ], batch size: 135, lr: 1.19e-02, grad_scale: 16.0 +2022-11-15 22:00:32,353 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.67 vs. limit=5.0 +2022-11-15 22:00:36,969 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2022-11-15 22:01:01,121 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8155, 1.2211, 1.0391, 0.9880, 1.1069, 1.1285, 0.9037, 1.3244], + device='cuda:3'), covar=tensor([0.0035, 0.0035, 0.0034, 0.0040, 0.0033, 0.0023, 0.0042, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0034, 0.0036, 0.0035, 0.0032, 0.0030, 0.0034, 0.0028], + device='cuda:3'), out_proj_covar=tensor([3.3346e-05, 3.2452e-05, 3.2502e-05, 3.2436e-05, 2.8636e-05, 2.5049e-05, + 3.3842e-05, 2.5312e-05], device='cuda:3') +2022-11-15 22:01:27,713 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.682e+02 2.055e+02 2.418e+02 5.378e+02, threshold=4.109e+02, percent-clipped=3.0 +2022-11-15 22:01:28,819 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2022-11-15 22:01:36,449 INFO [train.py:876] (3/4) Epoch 7, batch 800, loss[loss=0.2114, simple_loss=0.1935, pruned_loss=0.1146, over 5465.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1699, pruned_loss=0.07258, over 1065127.81 frames. ], batch size: 64, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 22:01:43,086 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:01:47,977 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 +2022-11-15 22:01:52,603 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8079, 2.1342, 1.8315, 1.2412, 1.5167, 2.3673, 2.1831, 2.3974], + device='cuda:3'), covar=tensor([0.1585, 0.1242, 0.1342, 0.2250, 0.0853, 0.0626, 0.0476, 0.0778], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0189, 0.0148, 0.0188, 0.0162, 0.0167, 0.0139, 0.0177], + 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') +2022-11-15 22:02:15,858 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:02:23,716 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5093, 2.4995, 2.1733, 2.7391, 2.3003, 2.2090, 2.3402, 2.7926], + device='cuda:3'), covar=tensor([0.0941, 0.1642, 0.2657, 0.1472, 0.1595, 0.1374, 0.1855, 0.2317], + device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0076, 0.0091, 0.0066, 0.0073, 0.0072, 0.0084, 0.0057], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:02:35,512 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 1.866e+02 2.294e+02 2.881e+02 8.701e+02, threshold=4.588e+02, percent-clipped=3.0 +2022-11-15 22:02:40,765 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:02:44,438 INFO [train.py:876] (3/4) Epoch 7, batch 900, loss[loss=0.2405, simple_loss=0.2089, pruned_loss=0.136, over 3040.00 frames. ], tot_loss[loss=0.1566, simple_loss=0.169, pruned_loss=0.07206, over 1071028.88 frames. ], batch size: 284, lr: 1.19e-02, grad_scale: 8.0 +2022-11-15 22:02:48,996 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8796, 2.6936, 2.6769, 1.4590, 2.8818, 3.0496, 3.1532, 2.9161], + device='cuda:3'), covar=tensor([0.2521, 0.2073, 0.1287, 0.3206, 0.0506, 0.0971, 0.0399, 0.0923], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0190, 0.0148, 0.0190, 0.0163, 0.0168, 0.0140, 0.0177], + 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') +2022-11-15 22:02:49,505 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2944, 4.3473, 4.3610, 4.5904, 4.1032, 3.7384, 5.0329, 4.3714], + device='cuda:3'), covar=tensor([0.0491, 0.1118, 0.0324, 0.1071, 0.0408, 0.0338, 0.0698, 0.0666], + device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0092, 0.0078, 0.0100, 0.0075, 0.0064, 0.0123, 0.0083], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:02:56,139 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:06,234 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:22,309 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:37,752 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:39,299 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:40,646 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:03:44,048 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.116e+02 1.775e+02 2.262e+02 2.701e+02 4.427e+02, threshold=4.524e+02, percent-clipped=0.0 +2022-11-15 22:03:52,912 INFO [train.py:876] (3/4) Epoch 7, batch 1000, loss[loss=0.1238, simple_loss=0.1427, pruned_loss=0.05248, over 5170.00 frames. ], tot_loss[loss=0.1563, simple_loss=0.1692, pruned_loss=0.07166, over 1077267.63 frames. ], batch size: 8, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:04:10,234 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-15 22:04:13,017 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:04:24,209 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3832, 2.5454, 3.7947, 3.2718, 4.2482, 2.7313, 3.7522, 4.2402], + device='cuda:3'), covar=tensor([0.0576, 0.2052, 0.0630, 0.1643, 0.0240, 0.1456, 0.1091, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0193, 0.0193, 0.0210, 0.0196, 0.0185, 0.0225, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 22:04:27,445 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9221, 4.7493, 3.5900, 2.0475, 4.5898, 1.9516, 4.0818, 2.4284], + device='cuda:3'), covar=tensor([0.1279, 0.0107, 0.0448, 0.2185, 0.0112, 0.1874, 0.0218, 0.1756], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0103, 0.0113, 0.0119, 0.0104, 0.0128, 0.0098, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 22:04:52,369 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.670e+02 2.114e+02 2.520e+02 5.115e+02, threshold=4.229e+02, percent-clipped=2.0 +2022-11-15 22:05:01,298 INFO [train.py:876] (3/4) Epoch 7, batch 1100, loss[loss=0.1959, simple_loss=0.1923, pruned_loss=0.09969, over 5472.00 frames. ], tot_loss[loss=0.1557, simple_loss=0.169, pruned_loss=0.07121, over 1080845.45 frames. ], batch size: 53, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:05:01,582 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.86 vs. limit=5.0 +2022-11-15 22:05:26,457 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9384, 1.1867, 1.6065, 1.4928, 1.5192, 1.4819, 1.7538, 1.3795], + device='cuda:3'), covar=tensor([0.0035, 0.0070, 0.0048, 0.0024, 0.0036, 0.0060, 0.0022, 0.0026], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0018, 0.0022, 0.0021, 0.0018, 0.0021, 0.0022], + device='cuda:3'), out_proj_covar=tensor([1.7162e-05, 1.8203e-05, 1.7484e-05, 2.2282e-05, 1.9685e-05, 1.8273e-05, + 2.1022e-05, 2.2850e-05], device='cuda:3') +2022-11-15 22:05:51,546 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1079, 3.8619, 2.7218, 3.7001, 2.9138, 2.7433, 2.1714, 3.2092], + device='cuda:3'), covar=tensor([0.1569, 0.0239, 0.1087, 0.0271, 0.0843, 0.1056, 0.1985, 0.0352], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0133, 0.0165, 0.0135, 0.0173, 0.0180, 0.0179, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:06:00,676 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 1.756e+02 2.130e+02 2.754e+02 4.425e+02, threshold=4.261e+02, percent-clipped=1.0 +2022-11-15 22:06:08,896 INFO [train.py:876] (3/4) Epoch 7, batch 1200, loss[loss=0.1177, simple_loss=0.1333, pruned_loss=0.05101, over 4640.00 frames. ], tot_loss[loss=0.1541, simple_loss=0.1675, pruned_loss=0.07029, over 1075055.04 frames. ], batch size: 5, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:06:42,977 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:06:59,240 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:07:08,224 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.841e+02 2.145e+02 2.633e+02 4.840e+02, threshold=4.289e+02, percent-clipped=5.0 +2022-11-15 22:07:16,620 INFO [train.py:876] (3/4) Epoch 7, batch 1300, loss[loss=0.1024, simple_loss=0.1353, pruned_loss=0.03475, over 5157.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.1687, pruned_loss=0.07185, over 1075647.71 frames. ], batch size: 8, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:07:20,365 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.04 vs. limit=5.0 +2022-11-15 22:08:19,522 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.856e+01 1.627e+02 2.176e+02 2.698e+02 5.044e+02, threshold=4.352e+02, percent-clipped=2.0 +2022-11-15 22:08:28,064 INFO [train.py:876] (3/4) Epoch 7, batch 1400, loss[loss=0.2043, simple_loss=0.2018, pruned_loss=0.1034, over 5595.00 frames. ], tot_loss[loss=0.153, simple_loss=0.1663, pruned_loss=0.06985, over 1083123.73 frames. ], batch size: 50, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:08:30,835 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4456, 2.7060, 2.2665, 3.0353, 2.2448, 2.6411, 2.7285, 3.3803], + device='cuda:3'), covar=tensor([0.0872, 0.1206, 0.2761, 0.0681, 0.1602, 0.0756, 0.1721, 0.1740], + device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0078, 0.0092, 0.0068, 0.0076, 0.0073, 0.0088, 0.0058], + 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') +2022-11-15 22:08:43,541 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:08:43,933 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 +2022-11-15 22:09:00,841 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:09:10,482 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6784, 0.8823, 0.8628, 0.7417, 0.7799, 1.2758, 0.8963, 0.7869], + device='cuda:3'), covar=tensor([0.1943, 0.0412, 0.1498, 0.1883, 0.2068, 0.0338, 0.1688, 0.1742], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0049, 0.0053, 0.0067, 0.0056, 0.0044, 0.0048, 0.0055], + 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') +2022-11-15 22:09:24,835 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:09:26,571 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.653e+02 2.032e+02 2.682e+02 5.131e+02, threshold=4.064e+02, percent-clipped=2.0 +2022-11-15 22:09:30,026 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8628, 2.2839, 2.9514, 3.7972, 3.9475, 2.8968, 2.3341, 3.8627], + device='cuda:3'), covar=tensor([0.0399, 0.3620, 0.2328, 0.2306, 0.1134, 0.2745, 0.2295, 0.0606], + device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0213, 0.0205, 0.0332, 0.0226, 0.0222, 0.0194, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005], + device='cuda:3') +2022-11-15 22:09:35,808 INFO [train.py:876] (3/4) Epoch 7, batch 1500, loss[loss=0.153, simple_loss=0.1716, pruned_loss=0.06716, over 5642.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.1661, pruned_loss=0.06928, over 1084981.44 frames. ], batch size: 29, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:09:38,590 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8397, 2.1235, 2.0620, 1.4340, 2.0788, 2.4030, 2.1065, 2.3551], + device='cuda:3'), covar=tensor([0.1484, 0.1314, 0.0971, 0.2106, 0.0644, 0.0557, 0.0476, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0187, 0.0148, 0.0187, 0.0160, 0.0169, 0.0140, 0.0178], + 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') +2022-11-15 22:09:41,804 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:09:46,971 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7294, 4.3899, 4.3743, 4.1887, 4.8411, 4.7823, 4.3411, 4.7514], + device='cuda:3'), covar=tensor([0.0797, 0.0819, 0.1044, 0.1026, 0.0847, 0.0334, 0.0554, 0.0914], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0123, 0.0091, 0.0124, 0.0132, 0.0079, 0.0105, 0.0121], + 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') +2022-11-15 22:10:09,053 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:10:24,458 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 +2022-11-15 22:10:24,977 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:10:34,225 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.120e+01 1.667e+02 2.029e+02 2.581e+02 4.326e+02, threshold=4.059e+02, percent-clipped=3.0 +2022-11-15 22:10:41,638 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:10:42,909 INFO [train.py:876] (3/4) Epoch 7, batch 1600, loss[loss=0.1528, simple_loss=0.1694, pruned_loss=0.06812, over 5753.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.1655, pruned_loss=0.06816, over 1093770.91 frames. ], batch size: 14, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:10:57,952 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:11:42,141 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.688e+02 2.070e+02 2.514e+02 3.710e+02, threshold=4.140e+02, percent-clipped=0.0 +2022-11-15 22:11:45,666 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7552, 4.3185, 3.7637, 3.6910, 2.4034, 4.2914, 2.2296, 3.5095], + device='cuda:3'), covar=tensor([0.0393, 0.0106, 0.0172, 0.0312, 0.0459, 0.0106, 0.0412, 0.0116], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0142, 0.0160, 0.0177, 0.0177, 0.0160, 0.0169, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 22:11:51,082 INFO [train.py:876] (3/4) Epoch 7, batch 1700, loss[loss=0.1501, simple_loss=0.163, pruned_loss=0.06862, over 5582.00 frames. ], tot_loss[loss=0.152, simple_loss=0.166, pruned_loss=0.06895, over 1088255.82 frames. ], batch size: 22, lr: 1.18e-02, grad_scale: 8.0 +2022-11-15 22:12:00,539 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 +2022-11-15 22:12:01,852 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:12:23,371 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2022-11-15 22:12:43,171 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 22:12:43,723 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8906, 2.8924, 2.2848, 2.6544, 1.7343, 2.2369, 1.6128, 2.6756], + device='cuda:3'), covar=tensor([0.1150, 0.0243, 0.0717, 0.0375, 0.1128, 0.0894, 0.1601, 0.0336], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0132, 0.0165, 0.0138, 0.0171, 0.0178, 0.0177, 0.0144], + 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') +2022-11-15 22:12:45,013 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 22:12:50,413 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.783e+02 2.189e+02 2.743e+02 5.258e+02, threshold=4.378e+02, percent-clipped=6.0 +2022-11-15 22:12:58,913 INFO [train.py:876] (3/4) Epoch 7, batch 1800, loss[loss=0.1583, simple_loss=0.1694, pruned_loss=0.0736, over 5105.00 frames. ], tot_loss[loss=0.152, simple_loss=0.1664, pruned_loss=0.06883, over 1092614.11 frames. ], batch size: 91, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:13:01,576 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:13:57,537 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.508e+01 1.821e+02 2.184e+02 2.835e+02 7.738e+02, threshold=4.367e+02, percent-clipped=5.0 +2022-11-15 22:14:06,281 INFO [train.py:876] (3/4) Epoch 7, batch 1900, loss[loss=0.1324, simple_loss=0.1543, pruned_loss=0.05524, over 5796.00 frames. ], tot_loss[loss=0.1532, simple_loss=0.1663, pruned_loss=0.07003, over 1081230.05 frames. ], batch size: 22, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:14:34,368 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1906, 2.8603, 2.9947, 2.5917, 1.8700, 2.9449, 2.0521, 2.4436], + device='cuda:3'), covar=tensor([0.0260, 0.0091, 0.0086, 0.0207, 0.0312, 0.0095, 0.0286, 0.0101], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0142, 0.0158, 0.0175, 0.0175, 0.0159, 0.0168, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:15:04,860 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 1.823e+02 2.221e+02 2.683e+02 3.782e+02, threshold=4.442e+02, percent-clipped=0.0 +2022-11-15 22:15:11,656 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 +2022-11-15 22:15:13,880 INFO [train.py:876] (3/4) Epoch 7, batch 2000, loss[loss=0.2064, simple_loss=0.1907, pruned_loss=0.111, over 4761.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.1645, pruned_loss=0.06786, over 1084316.06 frames. ], batch size: 135, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:15:25,906 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 +2022-11-15 22:15:42,484 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6996, 0.8805, 1.9757, 1.3779, 1.2580, 1.6786, 1.3560, 1.3728], + device='cuda:3'), covar=tensor([0.0014, 0.0093, 0.0019, 0.0059, 0.0100, 0.0069, 0.0027, 0.0056], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0020, 0.0023, 0.0022, 0.0019, 0.0023, 0.0023], + device='cuda:3'), out_proj_covar=tensor([1.7774e-05, 1.8941e-05, 1.8725e-05, 2.3214e-05, 2.0730e-05, 1.9183e-05, + 2.2346e-05, 2.4754e-05], device='cuda:3') +2022-11-15 22:16:01,071 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:02,311 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:16:07,683 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 22:16:12,591 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.738e+02 2.020e+02 2.541e+02 5.348e+02, threshold=4.040e+02, percent-clipped=1.0 +2022-11-15 22:16:21,530 INFO [train.py:876] (3/4) Epoch 7, batch 2100, loss[loss=0.1063, simple_loss=0.1361, pruned_loss=0.03822, over 5535.00 frames. ], tot_loss[loss=0.148, simple_loss=0.1633, pruned_loss=0.06641, over 1087206.13 frames. ], batch size: 10, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:16:24,220 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:39,954 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:42,029 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:49,031 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:55,982 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:16:56,663 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1010, 4.8766, 3.6442, 2.3225, 4.6417, 1.9290, 4.5145, 2.8482], + device='cuda:3'), covar=tensor([0.1022, 0.0096, 0.0364, 0.1881, 0.0120, 0.1650, 0.0136, 0.1281], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0103, 0.0113, 0.0116, 0.0103, 0.0127, 0.0095, 0.0117], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 22:17:20,073 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.330e+01 1.725e+02 2.150e+02 2.571e+02 5.272e+02, threshold=4.300e+02, percent-clipped=3.0 +2022-11-15 22:17:25,103 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 +2022-11-15 22:17:28,681 INFO [train.py:876] (3/4) Epoch 7, batch 2200, loss[loss=0.1438, simple_loss=0.1688, pruned_loss=0.05945, over 5518.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.1636, pruned_loss=0.06687, over 1078882.35 frames. ], batch size: 17, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:17:30,174 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:17:46,383 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:18:27,932 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.659e+02 2.110e+02 2.622e+02 5.586e+02, threshold=4.221e+02, percent-clipped=2.0 +2022-11-15 22:18:28,121 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:18:36,860 INFO [train.py:876] (3/4) Epoch 7, batch 2300, loss[loss=0.1156, simple_loss=0.1341, pruned_loss=0.04853, over 5467.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.1655, pruned_loss=0.06902, over 1078758.75 frames. ], batch size: 11, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:19:25,063 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:19:35,541 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.800e+02 2.204e+02 2.753e+02 4.636e+02, threshold=4.408e+02, percent-clipped=3.0 +2022-11-15 22:19:44,586 INFO [train.py:876] (3/4) Epoch 7, batch 2400, loss[loss=0.2048, simple_loss=0.1942, pruned_loss=0.1077, over 5457.00 frames. ], tot_loss[loss=0.1544, simple_loss=0.1678, pruned_loss=0.07054, over 1087725.21 frames. ], batch size: 58, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:19:57,578 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:20:01,770 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:20:16,790 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0736, 1.4370, 1.3289, 0.9852, 0.9008, 1.5848, 1.5658, 1.5205], + device='cuda:3'), covar=tensor([0.1034, 0.0703, 0.1289, 0.2041, 0.0920, 0.0430, 0.0476, 0.0980], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0190, 0.0153, 0.0192, 0.0162, 0.0174, 0.0142, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 22:20:43,096 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.760e+02 2.143e+02 2.682e+02 4.652e+02, threshold=4.287e+02, percent-clipped=1.0 +2022-11-15 22:20:49,995 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:20:51,909 INFO [train.py:876] (3/4) Epoch 7, batch 2500, loss[loss=0.1215, simple_loss=0.1419, pruned_loss=0.05062, over 5197.00 frames. ], tot_loss[loss=0.1536, simple_loss=0.1677, pruned_loss=0.06971, over 1084903.59 frames. ], batch size: 8, lr: 1.17e-02, grad_scale: 8.0 +2022-11-15 22:21:19,353 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-15 22:21:22,818 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2022-11-15 22:21:43,575 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9527, 1.8295, 2.0136, 2.1432, 1.8650, 1.5872, 1.8797, 2.2680], + device='cuda:3'), covar=tensor([0.1390, 0.2055, 0.2253, 0.2274, 0.1520, 0.2208, 0.2084, 0.0638], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0093, 0.0070, 0.0078, 0.0076, 0.0089, 0.0061], + 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') +2022-11-15 22:21:47,943 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:21:51,094 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.657e+02 1.935e+02 2.390e+02 4.850e+02, threshold=3.869e+02, percent-clipped=1.0 +2022-11-15 22:21:54,539 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2239, 3.7820, 3.3464, 3.7629, 3.7797, 3.2673, 3.4229, 3.3132], + device='cuda:3'), covar=tensor([0.0985, 0.0398, 0.1219, 0.0409, 0.0371, 0.0460, 0.0597, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0155, 0.0251, 0.0159, 0.0193, 0.0158, 0.0169, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 22:21:59,980 INFO [train.py:876] (3/4) Epoch 7, batch 2600, loss[loss=0.1514, simple_loss=0.1675, pruned_loss=0.06762, over 5615.00 frames. ], tot_loss[loss=0.1548, simple_loss=0.1682, pruned_loss=0.0707, over 1079446.52 frames. ], batch size: 18, lr: 1.16e-02, grad_scale: 8.0 +2022-11-15 22:22:04,521 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.66 vs. limit=5.0 +2022-11-15 22:22:29,341 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:22:55,184 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7638, 1.4610, 1.2414, 0.8613, 1.2342, 1.0855, 1.0096, 1.1851], + device='cuda:3'), covar=tensor([0.0051, 0.0041, 0.0030, 0.0047, 0.0036, 0.0028, 0.0045, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0034, 0.0036, 0.0036, 0.0033, 0.0030, 0.0035, 0.0029], + device='cuda:3'), out_proj_covar=tensor([3.4118e-05, 3.2088e-05, 3.2239e-05, 3.3407e-05, 2.9557e-05, 2.5302e-05, + 3.3629e-05, 2.6234e-05], device='cuda:3') +2022-11-15 22:22:59,256 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.863e+02 2.232e+02 2.729e+02 4.275e+02, threshold=4.463e+02, percent-clipped=4.0 +2022-11-15 22:23:07,784 INFO [train.py:876] (3/4) Epoch 7, batch 2700, loss[loss=0.1187, simple_loss=0.1495, pruned_loss=0.04399, over 5429.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.1655, pruned_loss=0.06799, over 1083976.77 frames. ], batch size: 10, lr: 1.16e-02, grad_scale: 8.0 +2022-11-15 22:23:10,635 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:23:21,376 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2482, 4.1284, 4.0554, 4.2729, 3.7088, 3.3906, 4.6489, 4.0074], + device='cuda:3'), covar=tensor([0.0309, 0.0870, 0.0328, 0.1008, 0.0562, 0.0403, 0.0680, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0096, 0.0079, 0.0104, 0.0076, 0.0065, 0.0130, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:23:24,475 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4519, 1.7332, 1.8088, 1.2424, 1.0337, 2.2452, 1.6313, 1.5419], + device='cuda:3'), covar=tensor([0.1258, 0.0846, 0.0912, 0.2191, 0.2678, 0.0865, 0.1433, 0.1173], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0051, 0.0056, 0.0068, 0.0055, 0.0045, 0.0050, 0.0057], + 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') +2022-11-15 22:23:25,122 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:23:59,110 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:24:09,519 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.800e+02 2.089e+02 2.479e+02 4.263e+02, threshold=4.179e+02, percent-clipped=0.0 +2022-11-15 22:24:16,891 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:24:18,865 INFO [train.py:876] (3/4) Epoch 7, batch 2800, loss[loss=0.168, simple_loss=0.1781, pruned_loss=0.07898, over 5663.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.164, pruned_loss=0.06677, over 1087377.07 frames. ], batch size: 36, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:24:49,434 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:25:14,653 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:25:18,215 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.794e+02 2.220e+02 2.623e+02 4.968e+02, threshold=4.440e+02, percent-clipped=2.0 +2022-11-15 22:25:27,097 INFO [train.py:876] (3/4) Epoch 7, batch 2900, loss[loss=0.1988, simple_loss=0.1991, pruned_loss=0.09924, over 5352.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.1659, pruned_loss=0.06876, over 1086501.19 frames. ], batch size: 70, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:25:32,760 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-15 22:25:47,078 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:26:08,957 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:26:13,979 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:26:22,635 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1951, 4.8971, 4.3501, 5.0168, 4.9612, 3.9011, 4.4174, 4.1393], + device='cuda:3'), covar=tensor([0.0352, 0.0469, 0.1788, 0.0277, 0.0385, 0.0458, 0.0513, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0160, 0.0253, 0.0161, 0.0195, 0.0158, 0.0170, 0.0155], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 22:26:26,117 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.852e+02 2.208e+02 2.830e+02 6.179e+02, threshold=4.416e+02, percent-clipped=4.0 +2022-11-15 22:26:34,364 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:26:34,917 INFO [train.py:876] (3/4) Epoch 7, batch 3000, loss[loss=0.1989, simple_loss=0.1966, pruned_loss=0.1006, over 5461.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.1666, pruned_loss=0.06961, over 1087121.03 frames. ], batch size: 53, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:26:34,917 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 22:26:43,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2081, 4.1289, 2.9593, 4.0442, 3.3209, 3.1242, 2.2556, 3.5462], + device='cuda:3'), covar=tensor([0.1807, 0.0239, 0.1032, 0.0222, 0.0604, 0.0885, 0.2362, 0.0372], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0134, 0.0168, 0.0138, 0.0173, 0.0177, 0.0177, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:26:51,165 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7189, 2.2735, 1.9825, 1.5367, 1.4949, 2.4589, 2.3203, 2.5347], + device='cuda:3'), covar=tensor([0.0862, 0.0714, 0.1013, 0.1316, 0.0585, 0.0381, 0.0265, 0.0573], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0186, 0.0149, 0.0187, 0.0162, 0.0173, 0.0142, 0.0179], + 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') +2022-11-15 22:26:52,608 INFO [train.py:908] (3/4) Epoch 7, validation: loss=0.1596, simple_loss=0.1815, pruned_loss=0.06886, over 1530663.00 frames. +2022-11-15 22:26:52,609 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 22:27:08,190 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 22:27:12,103 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-15 22:27:12,632 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:27:40,169 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9074, 2.0825, 2.4168, 1.3115, 1.0494, 2.6197, 2.0023, 1.6420], + device='cuda:3'), covar=tensor([0.0676, 0.0595, 0.0387, 0.1700, 0.1980, 0.0365, 0.1085, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0050, 0.0054, 0.0067, 0.0053, 0.0044, 0.0049, 0.0055], + 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') +2022-11-15 22:27:50,896 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.581e+01 1.880e+02 2.186e+02 2.617e+02 4.736e+02, threshold=4.372e+02, percent-clipped=3.0 +2022-11-15 22:27:59,278 INFO [train.py:876] (3/4) Epoch 7, batch 3100, loss[loss=0.1567, simple_loss=0.1711, pruned_loss=0.07115, over 5713.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.1664, pruned_loss=0.06933, over 1090905.57 frames. ], batch size: 34, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:28:02,307 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0012, 3.2460, 3.0927, 2.9767, 3.2109, 3.0665, 1.0644, 3.2516], + device='cuda:3'), covar=tensor([0.0446, 0.0287, 0.0382, 0.0334, 0.0465, 0.0356, 0.3536, 0.0397], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0077, 0.0079, 0.0072, 0.0098, 0.0081, 0.0128, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:28:33,996 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9849, 1.4095, 1.0417, 0.7264, 1.2692, 1.1170, 0.7965, 1.1777], + device='cuda:3'), covar=tensor([0.0039, 0.0024, 0.0034, 0.0041, 0.0031, 0.0026, 0.0048, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0034, 0.0036, 0.0036, 0.0033, 0.0030, 0.0034, 0.0029], + device='cuda:3'), out_proj_covar=tensor([3.4387e-05, 3.2128e-05, 3.2679e-05, 3.3600e-05, 2.9692e-05, 2.5788e-05, + 3.2412e-05, 2.5762e-05], device='cuda:3') +2022-11-15 22:28:46,050 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6098, 1.2157, 1.4179, 1.0316, 1.4522, 1.2867, 1.2370, 1.3289], + device='cuda:3'), covar=tensor([0.1245, 0.0623, 0.0479, 0.0653, 0.1049, 0.0986, 0.0758, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0016, 0.0011, 0.0013, 0.0012, 0.0010, 0.0014, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.1997e-05, 7.0194e-05, 5.4102e-05, 6.2110e-05, 5.6864e-05, 5.2321e-05, + 6.5202e-05, 5.4617e-05], device='cuda:3') +2022-11-15 22:28:51,922 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:28:55,091 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 +2022-11-15 22:28:58,608 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.858e+01 1.618e+02 2.001e+02 2.391e+02 5.637e+02, threshold=4.002e+02, percent-clipped=1.0 +2022-11-15 22:29:07,480 INFO [train.py:876] (3/4) Epoch 7, batch 3200, loss[loss=0.2812, simple_loss=0.2356, pruned_loss=0.1634, over 3025.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.166, pruned_loss=0.06911, over 1086355.46 frames. ], batch size: 284, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:29:08,242 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1132, 1.5784, 1.7916, 1.1574, 1.2473, 2.0363, 1.7669, 1.2808], + device='cuda:3'), covar=tensor([0.1271, 0.0754, 0.0833, 0.2129, 0.2309, 0.0700, 0.1232, 0.1415], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0052, 0.0056, 0.0069, 0.0055, 0.0045, 0.0051, 0.0056], + 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') +2022-11-15 22:29:08,405 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2022-11-15 22:29:08,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 22:29:14,242 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9805, 1.4465, 1.3248, 1.1181, 1.0924, 1.6314, 1.3832, 1.1102], + device='cuda:3'), covar=tensor([0.1932, 0.0415, 0.1460, 0.1973, 0.2521, 0.0392, 0.1229, 0.2212], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0052, 0.0056, 0.0069, 0.0055, 0.0045, 0.0051, 0.0057], + 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') +2022-11-15 22:29:33,372 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:29:56,043 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2022-11-15 22:30:04,526 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9917, 2.8512, 2.2786, 2.7907, 2.8675, 2.6196, 2.6214, 2.5915], + device='cuda:3'), covar=tensor([0.0401, 0.0812, 0.2250, 0.0841, 0.0867, 0.0606, 0.0995, 0.0803], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0157, 0.0249, 0.0155, 0.0192, 0.0155, 0.0167, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 22:30:06,305 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.769e+02 2.073e+02 2.656e+02 5.224e+02, threshold=4.147e+02, percent-clipped=4.0 +2022-11-15 22:30:14,559 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:30:15,148 INFO [train.py:876] (3/4) Epoch 7, batch 3300, loss[loss=0.147, simple_loss=0.1676, pruned_loss=0.06319, over 5816.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.1654, pruned_loss=0.06923, over 1087712.38 frames. ], batch size: 18, lr: 1.16e-02, grad_scale: 16.0 +2022-11-15 22:30:27,238 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 22:30:32,491 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:30:47,192 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:31:14,482 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.793e+02 2.115e+02 2.507e+02 4.736e+02, threshold=4.229e+02, percent-clipped=3.0 +2022-11-15 22:31:22,156 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8618, 4.7274, 4.8674, 4.9084, 4.3333, 4.3253, 5.3683, 4.7903], + device='cuda:3'), covar=tensor([0.0318, 0.0800, 0.0299, 0.1193, 0.0440, 0.0250, 0.0694, 0.0372], + device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0094, 0.0081, 0.0102, 0.0076, 0.0065, 0.0129, 0.0085], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:31:23,349 INFO [train.py:876] (3/4) Epoch 7, batch 3400, loss[loss=0.133, simple_loss=0.1643, pruned_loss=0.05089, over 5774.00 frames. ], tot_loss[loss=0.1533, simple_loss=0.1663, pruned_loss=0.0702, over 1080203.78 frames. ], batch size: 20, lr: 1.15e-02, grad_scale: 16.0 +2022-11-15 22:32:12,903 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-15 22:32:22,898 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.842e+02 2.111e+02 2.660e+02 4.907e+02, threshold=4.221e+02, percent-clipped=3.0 +2022-11-15 22:32:31,440 INFO [train.py:876] (3/4) Epoch 7, batch 3500, loss[loss=0.1668, simple_loss=0.1788, pruned_loss=0.07744, over 5630.00 frames. ], tot_loss[loss=0.151, simple_loss=0.1652, pruned_loss=0.06844, over 1083239.30 frames. ], batch size: 50, lr: 1.15e-02, grad_scale: 16.0 +2022-11-15 22:32:54,459 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:33:09,102 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1745, 1.4097, 1.7942, 1.7251, 1.7268, 2.0625, 1.8301, 1.3424], + device='cuda:3'), covar=tensor([0.0014, 0.0119, 0.0047, 0.0029, 0.0030, 0.0096, 0.0022, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0018, 0.0022, 0.0020, 0.0018, 0.0022, 0.0022], + device='cuda:3'), out_proj_covar=tensor([1.6768e-05, 1.8192e-05, 1.7309e-05, 2.1823e-05, 1.9379e-05, 1.8233e-05, + 2.1728e-05, 2.2886e-05], device='cuda:3') +2022-11-15 22:33:30,929 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.714e+02 2.031e+02 2.581e+02 4.185e+02, threshold=4.062e+02, percent-clipped=0.0 +2022-11-15 22:33:39,504 INFO [train.py:876] (3/4) Epoch 7, batch 3600, loss[loss=0.1339, simple_loss=0.1595, pruned_loss=0.05417, over 5514.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.1667, pruned_loss=0.06938, over 1086985.93 frames. ], batch size: 13, lr: 1.15e-02, grad_scale: 16.0 +2022-11-15 22:33:42,334 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.11 vs. limit=5.0 +2022-11-15 22:33:44,758 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:33:52,090 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:33:56,654 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:34:23,972 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:34:26,054 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:34:28,907 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:34:38,257 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.693e+02 2.082e+02 2.562e+02 5.983e+02, threshold=4.163e+02, percent-clipped=3.0 +2022-11-15 22:34:47,572 INFO [train.py:876] (3/4) Epoch 7, batch 3700, loss[loss=0.1453, simple_loss=0.1724, pruned_loss=0.05911, over 5567.00 frames. ], tot_loss[loss=0.1536, simple_loss=0.1674, pruned_loss=0.06989, over 1089349.47 frames. ], batch size: 15, lr: 1.15e-02, grad_scale: 16.0 +2022-11-15 22:35:18,594 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:35:34,056 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0761, 2.8494, 2.9156, 2.7643, 3.1805, 3.0753, 3.0160, 3.1114], + device='cuda:3'), covar=tensor([0.0564, 0.0441, 0.0590, 0.0468, 0.0462, 0.0237, 0.0365, 0.0576], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0125, 0.0094, 0.0124, 0.0138, 0.0081, 0.0107, 0.0122], + 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') +2022-11-15 22:35:43,488 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8322, 3.1729, 2.2052, 2.9213, 2.0848, 2.4149, 1.7085, 2.7732], + device='cuda:3'), covar=tensor([0.1443, 0.0220, 0.0953, 0.0345, 0.1019, 0.0869, 0.1899, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0135, 0.0169, 0.0140, 0.0173, 0.0178, 0.0175, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:35:46,987 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 1.850e+02 2.309e+02 2.787e+02 6.530e+02, threshold=4.618e+02, percent-clipped=2.0 +2022-11-15 22:35:56,051 INFO [train.py:876] (3/4) Epoch 7, batch 3800, loss[loss=0.1537, simple_loss=0.1572, pruned_loss=0.0751, over 5555.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.1667, pruned_loss=0.06928, over 1087858.74 frames. ], batch size: 40, lr: 1.15e-02, grad_scale: 16.0 +2022-11-15 22:36:01,311 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:36:18,977 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:36:51,696 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:36:56,172 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.606e+02 1.959e+02 2.426e+02 3.388e+02, threshold=3.918e+02, percent-clipped=0.0 +2022-11-15 22:37:04,432 INFO [train.py:876] (3/4) Epoch 7, batch 3900, loss[loss=0.07796, simple_loss=0.1077, pruned_loss=0.02409, over 5230.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.1664, pruned_loss=0.0682, over 1090098.49 frames. ], batch size: 8, lr: 1.15e-02, grad_scale: 8.0 +2022-11-15 22:37:29,182 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2022-11-15 22:37:38,116 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.63 vs. limit=5.0 +2022-11-15 22:37:47,950 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:37:49,601 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2473, 4.8096, 2.8804, 4.4272, 3.3006, 3.1288, 2.5161, 4.0525], + device='cuda:3'), covar=tensor([0.1610, 0.0107, 0.1122, 0.0303, 0.0657, 0.0941, 0.1930, 0.0232], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0136, 0.0170, 0.0142, 0.0177, 0.0180, 0.0179, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:38:04,271 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.833e+02 2.156e+02 2.681e+02 5.878e+02, threshold=4.311e+02, percent-clipped=5.0 +2022-11-15 22:38:12,272 INFO [train.py:876] (3/4) Epoch 7, batch 4000, loss[loss=0.1772, simple_loss=0.1717, pruned_loss=0.09137, over 4763.00 frames. ], tot_loss[loss=0.1511, simple_loss=0.1664, pruned_loss=0.0679, over 1095731.11 frames. ], batch size: 135, lr: 1.15e-02, grad_scale: 8.0 +2022-11-15 22:38:44,525 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 +2022-11-15 22:39:12,714 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.745e+02 2.156e+02 2.532e+02 4.633e+02, threshold=4.312e+02, percent-clipped=4.0 +2022-11-15 22:39:20,695 INFO [train.py:876] (3/4) Epoch 7, batch 4100, loss[loss=0.1237, simple_loss=0.1478, pruned_loss=0.04974, over 5749.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.1654, pruned_loss=0.06804, over 1094065.98 frames. ], batch size: 27, lr: 1.15e-02, grad_scale: 8.0 +2022-11-15 22:39:22,000 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:39:24,131 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:39:33,491 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1091, 2.3661, 2.6587, 3.8931, 3.9070, 3.0390, 2.4299, 3.8712], + device='cuda:3'), covar=tensor([0.0568, 0.2775, 0.2895, 0.3455, 0.0971, 0.2913, 0.2410, 0.0836], + device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0207, 0.0204, 0.0329, 0.0226, 0.0218, 0.0198, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 22:39:57,073 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 +2022-11-15 22:39:58,257 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:40:05,392 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:40:11,490 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2678, 1.9645, 3.0060, 2.6990, 2.8531, 2.0064, 2.6730, 3.2991], + device='cuda:3'), covar=tensor([0.0710, 0.1569, 0.0757, 0.1417, 0.0735, 0.1376, 0.1114, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0194, 0.0199, 0.0210, 0.0204, 0.0184, 0.0224, 0.0218], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 22:40:12,065 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:40:18,025 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4283, 1.7143, 2.2481, 2.2113, 2.3078, 1.4857, 2.0053, 2.3392], + device='cuda:3'), covar=tensor([0.0249, 0.0706, 0.0311, 0.0360, 0.0289, 0.0806, 0.0428, 0.0277], + device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0195, 0.0199, 0.0211, 0.0205, 0.0185, 0.0225, 0.0218], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 22:40:20,466 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.771e+02 2.148e+02 2.600e+02 4.098e+02, threshold=4.296e+02, percent-clipped=0.0 +2022-11-15 22:40:29,021 INFO [train.py:876] (3/4) Epoch 7, batch 4200, loss[loss=0.1802, simple_loss=0.1891, pruned_loss=0.08561, over 5568.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.1662, pruned_loss=0.06839, over 1090946.17 frames. ], batch size: 46, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:40:39,603 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:40:45,264 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1261, 3.1748, 2.4159, 1.6506, 3.0199, 1.2961, 3.0240, 1.7621], + device='cuda:3'), covar=tensor([0.1127, 0.0163, 0.0802, 0.1734, 0.0228, 0.1901, 0.0238, 0.1483], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0101, 0.0111, 0.0115, 0.0103, 0.0125, 0.0094, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 22:40:53,493 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 22:41:11,959 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:41:27,580 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.804e+02 2.056e+02 2.615e+02 5.152e+02, threshold=4.112e+02, percent-clipped=3.0 +2022-11-15 22:41:36,123 INFO [train.py:876] (3/4) Epoch 7, batch 4300, loss[loss=0.117, simple_loss=0.1403, pruned_loss=0.04682, over 5466.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.1653, pruned_loss=0.06724, over 1092310.98 frames. ], batch size: 10, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:41:44,862 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:42:36,450 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.735e+02 2.013e+02 2.634e+02 5.022e+02, threshold=4.026e+02, percent-clipped=3.0 +2022-11-15 22:42:44,692 INFO [train.py:876] (3/4) Epoch 7, batch 4400, loss[loss=0.212, simple_loss=0.1911, pruned_loss=0.1164, over 4706.00 frames. ], tot_loss[loss=0.149, simple_loss=0.1643, pruned_loss=0.06686, over 1087630.35 frames. ], batch size: 135, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:42:46,465 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:08,276 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3468, 4.9567, 3.2730, 4.6115, 3.6199, 3.3819, 2.8166, 4.1503], + device='cuda:3'), covar=tensor([0.1528, 0.0150, 0.0929, 0.0266, 0.0497, 0.0749, 0.1691, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0135, 0.0168, 0.0142, 0.0172, 0.0177, 0.0175, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:43:11,627 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:14,179 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8563, 4.2734, 3.7854, 4.1881, 4.2650, 3.4270, 3.8096, 3.5462], + device='cuda:3'), covar=tensor([0.0461, 0.0426, 0.1468, 0.0503, 0.0445, 0.0512, 0.0505, 0.0609], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0159, 0.0256, 0.0159, 0.0198, 0.0159, 0.0170, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 22:43:18,749 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:26,716 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:44,369 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.659e+02 2.129e+02 2.741e+02 6.031e+02, threshold=4.258e+02, percent-clipped=4.0 +2022-11-15 22:43:52,256 INFO [train.py:876] (3/4) Epoch 7, batch 4500, loss[loss=0.134, simple_loss=0.1602, pruned_loss=0.05388, over 5580.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.1643, pruned_loss=0.06652, over 1082753.24 frames. ], batch size: 24, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:43:52,422 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:59,756 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:43:59,825 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8151, 1.6320, 1.8655, 1.6251, 1.2482, 1.7524, 1.4095, 1.3420], + device='cuda:3'), covar=tensor([0.0015, 0.0026, 0.0026, 0.0021, 0.0042, 0.0032, 0.0025, 0.0029], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0019, 0.0023, 0.0021, 0.0020, 0.0023, 0.0023], + device='cuda:3'), out_proj_covar=tensor([1.6462e-05, 1.8689e-05, 1.8102e-05, 2.2618e-05, 1.9902e-05, 1.9818e-05, + 2.2773e-05, 2.5029e-05], device='cuda:3') +2022-11-15 22:44:11,226 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2022-11-15 22:44:14,576 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:44:14,612 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7523, 3.0736, 2.1826, 2.7459, 2.0677, 2.3965, 1.7904, 2.7210], + device='cuda:3'), covar=tensor([0.1602, 0.0301, 0.1190, 0.0495, 0.1154, 0.1077, 0.1976, 0.0409], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0136, 0.0169, 0.0142, 0.0173, 0.0178, 0.0176, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:44:46,318 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:44:52,379 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.044e+02 1.660e+02 2.072e+02 2.700e+02 4.661e+02, threshold=4.143e+02, percent-clipped=2.0 +2022-11-15 22:45:00,425 INFO [train.py:876] (3/4) Epoch 7, batch 4600, loss[loss=0.1199, simple_loss=0.1464, pruned_loss=0.04666, over 5240.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.1637, pruned_loss=0.06602, over 1083231.18 frames. ], batch size: 9, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:45:03,089 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0827, 3.5632, 2.6110, 3.4088, 3.4187, 3.2313, 3.4545, 3.4234], + device='cuda:3'), covar=tensor([0.1168, 0.0860, 0.2940, 0.1237, 0.1016, 0.0698, 0.0788, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0161, 0.0259, 0.0162, 0.0200, 0.0161, 0.0173, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:45:13,073 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4350, 2.5196, 2.1661, 2.6404, 2.1437, 2.2479, 2.6564, 3.2699], + device='cuda:3'), covar=tensor([0.1143, 0.1925, 0.3461, 0.1215, 0.2155, 0.1435, 0.1910, 0.1357], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0083, 0.0095, 0.0073, 0.0079, 0.0076, 0.0087, 0.0059], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:45:18,388 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1917, 3.1991, 3.2476, 1.6390, 3.0489, 3.7542, 3.7574, 3.9605], + device='cuda:3'), covar=tensor([0.1871, 0.1328, 0.0593, 0.2625, 0.0437, 0.0323, 0.0234, 0.0530], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0193, 0.0154, 0.0193, 0.0170, 0.0174, 0.0144, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 22:45:27,988 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 22:45:33,768 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0409, 1.0319, 1.2823, 0.9919, 1.1040, 1.3066, 0.9823, 1.2311], + device='cuda:3'), covar=tensor([0.0054, 0.0041, 0.0029, 0.0034, 0.0030, 0.0021, 0.0037, 0.0071], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0035, 0.0036, 0.0037, 0.0035, 0.0032, 0.0035, 0.0029], + device='cuda:3'), out_proj_covar=tensor([3.4390e-05, 3.3016e-05, 3.2711e-05, 3.4499e-05, 3.1223e-05, 2.7369e-05, + 3.4015e-05, 2.6051e-05], device='cuda:3') +2022-11-15 22:45:43,517 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1564, 1.7889, 2.4719, 1.6580, 0.9034, 2.9097, 2.0961, 1.6041], + device='cuda:3'), covar=tensor([0.0779, 0.0873, 0.0645, 0.2328, 0.3078, 0.0774, 0.1133, 0.1442], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0052, 0.0055, 0.0069, 0.0055, 0.0046, 0.0052, 0.0057], + 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') +2022-11-15 22:46:00,361 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.286e+02 1.710e+02 2.127e+02 2.706e+02 4.557e+02, threshold=4.254e+02, percent-clipped=4.0 +2022-11-15 22:46:02,406 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:46:08,675 INFO [train.py:876] (3/4) Epoch 7, batch 4700, loss[loss=0.1156, simple_loss=0.1434, pruned_loss=0.04385, over 5569.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.1645, pruned_loss=0.06699, over 1084844.07 frames. ], batch size: 14, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:46:44,374 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:46:50,278 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:46:52,227 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:08,411 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.686e+02 2.046e+02 2.615e+02 4.137e+02, threshold=4.091e+02, percent-clipped=0.0 +2022-11-15 22:47:10,266 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:13,457 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:16,466 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1294, 3.9016, 3.8868, 3.5976, 4.0720, 4.0773, 1.2642, 4.1741], + device='cuda:3'), covar=tensor([0.0277, 0.0528, 0.0307, 0.0651, 0.0418, 0.0420, 0.3883, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0079, 0.0078, 0.0073, 0.0095, 0.0082, 0.0128, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:47:17,033 INFO [train.py:876] (3/4) Epoch 7, batch 4800, loss[loss=0.1195, simple_loss=0.1463, pruned_loss=0.04633, over 5685.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.1649, pruned_loss=0.06719, over 1086899.05 frames. ], batch size: 17, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:47:23,002 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:24,433 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:33,555 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:34,162 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8892, 2.0160, 2.9020, 1.7638, 1.8345, 3.7705, 2.6383, 1.8694], + device='cuda:3'), covar=tensor([0.0658, 0.0627, 0.0303, 0.2117, 0.3150, 0.0411, 0.0964, 0.0969], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0052, 0.0055, 0.0069, 0.0055, 0.0047, 0.0052, 0.0058], + 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') +2022-11-15 22:47:38,235 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 22:47:50,409 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7995, 2.0511, 2.5676, 1.6404, 1.7110, 3.1699, 2.6049, 1.8155], + device='cuda:3'), covar=tensor([0.0595, 0.0814, 0.0557, 0.3046, 0.2635, 0.2281, 0.1202, 0.1383], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0052, 0.0055, 0.0069, 0.0055, 0.0046, 0.0052, 0.0057], + 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') +2022-11-15 22:47:51,736 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:47:57,027 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:48:10,993 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:48:16,161 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.802e+02 2.155e+02 2.691e+02 6.184e+02, threshold=4.310e+02, percent-clipped=3.0 +2022-11-15 22:48:25,048 INFO [train.py:876] (3/4) Epoch 7, batch 4900, loss[loss=0.115, simple_loss=0.1334, pruned_loss=0.04833, over 5521.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.1639, pruned_loss=0.0669, over 1083765.06 frames. ], batch size: 13, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:48:48,183 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 22:48:49,150 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.94 vs. limit=5.0 +2022-11-15 22:49:08,361 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6166, 3.5270, 3.5382, 3.3864, 3.7504, 3.4646, 1.3377, 3.7665], + device='cuda:3'), covar=tensor([0.0272, 0.0393, 0.0319, 0.0295, 0.0354, 0.0498, 0.3318, 0.0336], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0080, 0.0079, 0.0074, 0.0097, 0.0083, 0.0131, 0.0102], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 22:49:24,581 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.657e+02 1.917e+02 2.351e+02 4.435e+02, threshold=3.835e+02, percent-clipped=1.0 +2022-11-15 22:49:32,509 INFO [train.py:876] (3/4) Epoch 7, batch 5000, loss[loss=0.1191, simple_loss=0.1428, pruned_loss=0.04775, over 5452.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.163, pruned_loss=0.0659, over 1088074.38 frames. ], batch size: 12, lr: 1.14e-02, grad_scale: 8.0 +2022-11-15 22:49:48,726 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2022-11-15 22:50:04,381 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:50:11,749 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7872, 1.3811, 1.0028, 0.9199, 1.0401, 1.0052, 0.4674, 1.0624], + device='cuda:3'), covar=tensor([0.0032, 0.0018, 0.0029, 0.0022, 0.0028, 0.0023, 0.0052, 0.0032], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0035, 0.0037, 0.0038, 0.0036, 0.0032, 0.0037, 0.0030], + device='cuda:3'), out_proj_covar=tensor([3.5270e-05, 3.3003e-05, 3.3805e-05, 3.5146e-05, 3.1837e-05, 2.7623e-05, + 3.5383e-05, 2.6432e-05], device='cuda:3') +2022-11-15 22:50:27,949 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8937, 2.3158, 2.7900, 3.6097, 3.8978, 2.9637, 2.5508, 3.8441], + device='cuda:3'), covar=tensor([0.0491, 0.3634, 0.2570, 0.3653, 0.0917, 0.2907, 0.2335, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0204, 0.0201, 0.0323, 0.0224, 0.0215, 0.0195, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 22:50:32,599 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.591e+01 1.796e+02 2.203e+02 2.625e+02 4.237e+02, threshold=4.405e+02, percent-clipped=2.0 +2022-11-15 22:50:37,387 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:50:40,537 INFO [train.py:876] (3/4) Epoch 7, batch 5100, loss[loss=0.185, simple_loss=0.1852, pruned_loss=0.09235, over 5301.00 frames. ], tot_loss[loss=0.1465, simple_loss=0.1627, pruned_loss=0.06516, over 1091929.85 frames. ], batch size: 79, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:50:53,608 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:51:09,842 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:51:12,135 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:51:22,054 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9000, 2.2210, 3.4569, 2.9746, 3.5567, 2.4590, 3.1453, 3.8543], + device='cuda:3'), covar=tensor([0.0538, 0.1523, 0.0685, 0.1390, 0.0485, 0.1306, 0.0999, 0.0566], + device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0191, 0.0199, 0.0210, 0.0207, 0.0187, 0.0224, 0.0217], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 22:51:39,998 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.794e+02 2.055e+02 2.515e+02 6.538e+02, threshold=4.110e+02, percent-clipped=4.0 +2022-11-15 22:51:40,521 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:51:48,597 INFO [train.py:876] (3/4) Epoch 7, batch 5200, loss[loss=0.1694, simple_loss=0.1857, pruned_loss=0.07655, over 5768.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.1628, pruned_loss=0.06497, over 1090263.76 frames. ], batch size: 16, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:51:51,932 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:52:11,481 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:52:21,489 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:52:25,045 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 22:52:32,607 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:52:33,190 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0036, 2.8665, 2.2887, 1.5707, 2.8506, 1.1823, 2.7822, 1.6482], + device='cuda:3'), covar=tensor([0.1111, 0.0185, 0.0746, 0.1631, 0.0213, 0.1868, 0.0232, 0.1492], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0103, 0.0113, 0.0116, 0.0104, 0.0127, 0.0095, 0.0116], + 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') +2022-11-15 22:52:34,733 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-15 22:52:38,839 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-15 22:52:43,691 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 22:52:46,877 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.737e+02 2.203e+02 2.681e+02 5.321e+02, threshold=4.406e+02, percent-clipped=3.0 +2022-11-15 22:52:48,323 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3906, 1.4803, 1.9889, 1.2778, 0.9546, 2.2770, 1.7531, 1.5224], + device='cuda:3'), covar=tensor([0.0997, 0.0904, 0.0637, 0.2551, 0.2733, 0.0811, 0.1188, 0.1192], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0054, 0.0056, 0.0071, 0.0056, 0.0046, 0.0052, 0.0059], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 22:52:55,127 INFO [train.py:876] (3/4) Epoch 7, batch 5300, loss[loss=0.2674, simple_loss=0.23, pruned_loss=0.1524, over 3060.00 frames. ], tot_loss[loss=0.1463, simple_loss=0.163, pruned_loss=0.06478, over 1092352.40 frames. ], batch size: 284, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:52:57,208 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-15 22:52:59,621 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.84 vs. limit=5.0 +2022-11-15 22:53:23,419 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8368, 1.2118, 1.4416, 0.8116, 1.0739, 1.3201, 0.6345, 1.1661], + device='cuda:3'), covar=tensor([0.0034, 0.0024, 0.0031, 0.0034, 0.0033, 0.0022, 0.0047, 0.0049], + device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0036, 0.0038, 0.0039, 0.0037, 0.0033, 0.0038, 0.0031], + device='cuda:3'), out_proj_covar=tensor([3.6260e-05, 3.3392e-05, 3.4039e-05, 3.5350e-05, 3.2544e-05, 2.8855e-05, + 3.6641e-05, 2.7559e-05], device='cuda:3') +2022-11-15 22:53:26,620 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:53:54,574 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.623e+02 1.921e+02 2.407e+02 5.715e+02, threshold=3.842e+02, percent-clipped=1.0 +2022-11-15 22:53:59,201 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:54:02,407 INFO [train.py:876] (3/4) Epoch 7, batch 5400, loss[loss=0.1437, simple_loss=0.164, pruned_loss=0.0617, over 5548.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.1631, pruned_loss=0.06586, over 1088011.22 frames. ], batch size: 30, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:54:03,843 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6457, 0.9069, 0.8122, 0.6007, 0.7951, 0.9187, 0.3769, 0.9619], + device='cuda:3'), covar=tensor([0.0053, 0.0021, 0.0048, 0.0028, 0.0026, 0.0027, 0.0075, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0038, 0.0039, 0.0040, 0.0038, 0.0035, 0.0040, 0.0032], + device='cuda:3'), out_proj_covar=tensor([3.7650e-05, 3.5110e-05, 3.5359e-05, 3.7046e-05, 3.3833e-05, 3.0087e-05, + 3.8524e-05, 2.8943e-05], device='cuda:3') +2022-11-15 22:54:16,524 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:54:20,152 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-15 22:54:34,073 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:54:41,796 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7457, 3.0523, 2.2784, 2.7263, 1.9860, 2.2855, 1.7674, 2.6509], + device='cuda:3'), covar=tensor([0.1444, 0.0233, 0.0962, 0.0367, 0.1172, 0.0963, 0.1719, 0.0370], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0135, 0.0168, 0.0139, 0.0172, 0.0177, 0.0174, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 22:54:46,122 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3182, 3.7985, 3.5253, 3.2346, 2.1300, 3.5934, 2.1176, 3.0676], + device='cuda:3'), covar=tensor([0.0384, 0.0138, 0.0141, 0.0328, 0.0444, 0.0123, 0.0401, 0.0164], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0143, 0.0159, 0.0176, 0.0171, 0.0158, 0.0168, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 22:54:48,497 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:55:02,427 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.256e+02 1.756e+02 2.023e+02 2.468e+02 4.836e+02, threshold=4.046e+02, percent-clipped=5.0 +2022-11-15 22:55:06,392 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:55:10,243 INFO [train.py:876] (3/4) Epoch 7, batch 5500, loss[loss=0.1052, simple_loss=0.1289, pruned_loss=0.04081, over 5737.00 frames. ], tot_loss[loss=0.1478, simple_loss=0.1631, pruned_loss=0.06625, over 1086289.27 frames. ], batch size: 14, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:55:17,055 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8576, 1.0785, 2.1536, 1.5024, 1.2961, 1.6909, 1.4959, 1.5645], + device='cuda:3'), covar=tensor([0.0030, 0.0084, 0.0021, 0.0038, 0.0109, 0.0071, 0.0041, 0.0034], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0020, 0.0024, 0.0021, 0.0020, 0.0023, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.6984e-05, 1.9651e-05, 1.8688e-05, 2.3752e-05, 2.0116e-05, 1.9727e-05, + 2.3133e-05, 2.5980e-05], device='cuda:3') +2022-11-15 22:55:40,397 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:55:51,695 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:56:02,347 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6969, 4.2100, 4.5564, 4.2586, 4.7591, 4.6497, 4.3341, 4.7042], + device='cuda:3'), covar=tensor([0.0334, 0.0269, 0.0384, 0.0274, 0.0300, 0.0161, 0.0216, 0.0262], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0124, 0.0092, 0.0124, 0.0138, 0.0082, 0.0105, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 22:56:10,382 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.636e+02 2.022e+02 2.644e+02 5.189e+02, threshold=4.044e+02, percent-clipped=1.0 +2022-11-15 22:56:18,734 INFO [train.py:876] (3/4) Epoch 7, batch 5600, loss[loss=0.1163, simple_loss=0.1439, pruned_loss=0.04436, over 5790.00 frames. ], tot_loss[loss=0.1513, simple_loss=0.1653, pruned_loss=0.06867, over 1079833.88 frames. ], batch size: 14, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:56:20,832 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:56:32,948 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2022-11-15 22:57:02,614 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:57:09,812 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6998, 2.1868, 2.6004, 3.4925, 3.5258, 2.7823, 2.0165, 3.5422], + device='cuda:3'), covar=tensor([0.0518, 0.3525, 0.2395, 0.2290, 0.1020, 0.2513, 0.2614, 0.0883], + device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0207, 0.0202, 0.0325, 0.0224, 0.0213, 0.0196, 0.0215], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 22:57:18,850 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.878e+01 1.572e+02 2.070e+02 2.645e+02 4.455e+02, threshold=4.140e+02, percent-clipped=3.0 +2022-11-15 22:57:27,066 INFO [train.py:876] (3/4) Epoch 7, batch 5700, loss[loss=0.1116, simple_loss=0.1364, pruned_loss=0.04337, over 5467.00 frames. ], tot_loss[loss=0.1511, simple_loss=0.1653, pruned_loss=0.06842, over 1080418.77 frames. ], batch size: 11, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:57:49,104 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4734, 1.5846, 2.1886, 1.2618, 0.8441, 2.3321, 1.6356, 1.4692], + device='cuda:3'), covar=tensor([0.1111, 0.1025, 0.0766, 0.2432, 0.2946, 0.0857, 0.0778, 0.1553], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0053, 0.0056, 0.0072, 0.0055, 0.0046, 0.0049, 0.0058], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 22:58:27,011 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.691e+02 1.989e+02 2.573e+02 4.583e+02, threshold=3.978e+02, percent-clipped=3.0 +2022-11-15 22:58:27,817 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:58:34,813 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 22:58:35,267 INFO [train.py:876] (3/4) Epoch 7, batch 5800, loss[loss=0.1578, simple_loss=0.1715, pruned_loss=0.07207, over 5718.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.165, pruned_loss=0.06785, over 1081317.16 frames. ], batch size: 19, lr: 1.13e-02, grad_scale: 8.0 +2022-11-15 22:58:39,400 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-15 22:59:04,666 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:59:08,979 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:59:09,189 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-15 22:59:15,729 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 22:59:16,299 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:59:33,817 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 1.742e+02 2.165e+02 2.707e+02 6.167e+02, threshold=4.330e+02, percent-clipped=6.0 +2022-11-15 22:59:36,565 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:59:42,439 INFO [train.py:876] (3/4) Epoch 7, batch 5900, loss[loss=0.1192, simple_loss=0.1459, pruned_loss=0.04625, over 5675.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.164, pruned_loss=0.06676, over 1085780.90 frames. ], batch size: 28, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 22:59:48,543 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 22:59:58,131 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3626, 3.3402, 3.3465, 3.1972, 2.0580, 3.4234, 2.1167, 3.0279], + device='cuda:3'), covar=tensor([0.0322, 0.0138, 0.0122, 0.0215, 0.0375, 0.0111, 0.0351, 0.0097], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0145, 0.0161, 0.0178, 0.0173, 0.0159, 0.0171, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 23:00:21,900 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:00:28,608 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4681, 2.0980, 3.1696, 2.8033, 3.1443, 2.2789, 2.8536, 3.4939], + device='cuda:3'), covar=tensor([0.0537, 0.1512, 0.0627, 0.1240, 0.0519, 0.1386, 0.0957, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0185, 0.0193, 0.0207, 0.0205, 0.0184, 0.0219, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 23:00:42,046 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.707e+02 2.021e+02 2.580e+02 5.267e+02, threshold=4.043e+02, percent-clipped=3.0 +2022-11-15 23:00:50,008 INFO [train.py:876] (3/4) Epoch 7, batch 6000, loss[loss=0.115, simple_loss=0.148, pruned_loss=0.04102, over 5547.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.164, pruned_loss=0.06712, over 1087517.38 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:00:50,008 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 23:01:05,897 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6860, 1.3569, 1.7222, 1.0665, 1.2225, 1.6808, 1.2973, 0.9907], + device='cuda:3'), covar=tensor([0.0019, 0.0047, 0.0044, 0.0047, 0.0074, 0.0090, 0.0034, 0.0044], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0021, 0.0024, 0.0022, 0.0020, 0.0024, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.7134e-05, 1.9721e-05, 1.9342e-05, 2.4018e-05, 2.0715e-05, 1.9725e-05, + 2.3422e-05, 2.6503e-05], device='cuda:3') +2022-11-15 23:01:07,886 INFO [train.py:908] (3/4) Epoch 7, validation: loss=0.1616, simple_loss=0.1829, pruned_loss=0.07014, over 1530663.00 frames. +2022-11-15 23:01:07,886 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 23:02:07,547 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.774e+02 2.178e+02 2.539e+02 4.839e+02, threshold=4.355e+02, percent-clipped=2.0 +2022-11-15 23:02:15,504 INFO [train.py:876] (3/4) Epoch 7, batch 6100, loss[loss=0.182, simple_loss=0.186, pruned_loss=0.08894, over 5663.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.1629, pruned_loss=0.06597, over 1084294.89 frames. ], batch size: 32, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:02:38,521 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:02:46,269 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:02:48,352 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7566, 2.2502, 3.2070, 1.9822, 1.5040, 2.9990, 2.5011, 1.8451], + device='cuda:3'), covar=tensor([0.0641, 0.1105, 0.0321, 0.1715, 0.1598, 0.3677, 0.1065, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0055, 0.0058, 0.0072, 0.0056, 0.0047, 0.0051, 0.0059], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 23:02:52,919 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:02:57,337 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2022-11-15 23:03:16,328 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.740e+01 1.709e+02 2.118e+02 2.760e+02 4.478e+02, threshold=4.236e+02, percent-clipped=1.0 +2022-11-15 23:03:20,410 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:03:24,174 INFO [train.py:876] (3/4) Epoch 7, batch 6200, loss[loss=0.1263, simple_loss=0.1456, pruned_loss=0.05356, over 5561.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.1626, pruned_loss=0.06602, over 1081643.11 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:03:27,522 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3688, 1.7102, 1.9438, 1.4176, 0.9155, 2.3893, 1.5941, 1.4480], + device='cuda:3'), covar=tensor([0.1053, 0.1216, 0.0954, 0.1893, 0.3770, 0.0699, 0.1450, 0.1560], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0056, 0.0058, 0.0071, 0.0056, 0.0047, 0.0051, 0.0059], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 23:03:42,361 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7414, 1.3291, 1.7752, 1.4817, 1.1563, 1.7919, 1.6356, 1.5469], + device='cuda:3'), covar=tensor([0.0018, 0.0054, 0.0032, 0.0030, 0.0055, 0.0055, 0.0028, 0.0026], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0021, 0.0024, 0.0022, 0.0020, 0.0024, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.7492e-05, 1.9675e-05, 1.9372e-05, 2.3905e-05, 2.0668e-05, 1.9969e-05, + 2.3257e-05, 2.6078e-05], device='cuda:3') +2022-11-15 23:04:03,709 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:04:23,026 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 1.621e+02 1.970e+02 2.358e+02 3.613e+02, threshold=3.939e+02, percent-clipped=0.0 +2022-11-15 23:04:25,132 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2022-11-15 23:04:31,725 INFO [train.py:876] (3/4) Epoch 7, batch 6300, loss[loss=0.1762, simple_loss=0.1858, pruned_loss=0.08328, over 5562.00 frames. ], tot_loss[loss=0.149, simple_loss=0.1638, pruned_loss=0.06705, over 1086271.25 frames. ], batch size: 40, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:04:34,334 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9804, 3.8596, 3.8510, 3.6520, 3.9426, 3.8052, 1.4719, 4.1882], + device='cuda:3'), covar=tensor([0.0249, 0.0382, 0.0300, 0.0315, 0.0326, 0.0430, 0.3641, 0.0295], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0079, 0.0078, 0.0074, 0.0097, 0.0083, 0.0130, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:04:34,971 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1968, 3.4958, 3.3390, 3.1654, 3.2919, 3.1101, 1.3156, 3.4695], + device='cuda:3'), covar=tensor([0.0497, 0.0293, 0.0459, 0.0461, 0.0566, 0.0687, 0.4131, 0.0531], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0079, 0.0078, 0.0074, 0.0097, 0.0083, 0.0130, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:04:36,246 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:04:47,258 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2920, 3.8828, 4.1193, 3.8729, 4.3295, 4.1040, 3.9242, 4.2911], + device='cuda:3'), covar=tensor([0.0362, 0.0348, 0.0408, 0.0372, 0.0364, 0.0355, 0.0344, 0.0365], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0128, 0.0095, 0.0127, 0.0140, 0.0083, 0.0108, 0.0126], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:04:51,782 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6924, 4.5463, 3.4786, 1.9194, 4.4042, 1.6928, 4.0453, 2.6099], + device='cuda:3'), covar=tensor([0.1478, 0.0149, 0.0571, 0.2533, 0.0152, 0.2113, 0.0205, 0.1650], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0104, 0.0114, 0.0116, 0.0105, 0.0126, 0.0097, 0.0116], + 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') +2022-11-15 23:05:26,603 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8465, 2.0484, 2.6215, 1.5408, 1.0467, 2.8327, 2.0382, 1.5508], + device='cuda:3'), covar=tensor([0.0778, 0.1140, 0.0400, 0.1948, 0.2773, 0.1390, 0.0927, 0.1262], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0056, 0.0058, 0.0072, 0.0056, 0.0047, 0.0052, 0.0060], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 23:05:34,292 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.742e+02 2.037e+02 2.563e+02 6.362e+02, threshold=4.074e+02, percent-clipped=2.0 +2022-11-15 23:05:41,081 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 +2022-11-15 23:05:42,857 INFO [train.py:876] (3/4) Epoch 7, batch 6400, loss[loss=0.1211, simple_loss=0.1586, pruned_loss=0.04181, over 5802.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.1649, pruned_loss=0.06737, over 1088561.25 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:05:55,925 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7496, 0.5064, 0.6412, 0.5204, 0.7149, 0.6982, 0.4147, 0.7774], + device='cuda:3'), covar=tensor([0.0193, 0.0332, 0.0201, 0.0207, 0.0214, 0.0186, 0.0513, 0.0207], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0013, 0.0013, 0.0010, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.5154e-05, 7.3881e-05, 5.5258e-05, 6.3843e-05, 6.0373e-05, 5.3629e-05, + 6.8494e-05, 5.5715e-05], device='cuda:3') +2022-11-15 23:06:13,105 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:06:20,289 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:06:29,339 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3018, 3.8615, 2.9329, 1.8155, 3.7604, 1.3986, 3.2786, 1.9991], + device='cuda:3'), covar=tensor([0.1565, 0.0160, 0.0592, 0.2130, 0.0166, 0.2248, 0.0358, 0.1969], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0103, 0.0113, 0.0114, 0.0104, 0.0125, 0.0096, 0.0115], + 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') +2022-11-15 23:06:41,517 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.674e+02 2.100e+02 2.789e+02 6.462e+02, threshold=4.200e+02, percent-clipped=5.0 +2022-11-15 23:06:42,287 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:06:45,244 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:06:47,025 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:06:50,153 INFO [train.py:876] (3/4) Epoch 7, batch 6500, loss[loss=0.1179, simple_loss=0.1452, pruned_loss=0.04524, over 5625.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.1648, pruned_loss=0.06766, over 1081646.47 frames. ], batch size: 23, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:06:52,925 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:07:07,358 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0232, 4.1194, 2.5409, 3.9480, 3.2667, 2.5023, 1.9699, 3.5067], + device='cuda:3'), covar=tensor([0.2080, 0.0255, 0.1507, 0.0360, 0.0794, 0.1482, 0.2433, 0.0407], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0133, 0.0165, 0.0135, 0.0171, 0.0173, 0.0172, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 23:07:28,582 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:07:34,062 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:07:49,845 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.678e+02 2.090e+02 2.492e+02 4.594e+02, threshold=4.179e+02, percent-clipped=1.0 +2022-11-15 23:07:58,107 INFO [train.py:876] (3/4) Epoch 7, batch 6600, loss[loss=0.1631, simple_loss=0.1755, pruned_loss=0.07534, over 5504.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.1644, pruned_loss=0.06663, over 1081136.37 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:08:05,708 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-15 23:08:15,538 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:08:50,896 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6756, 2.6254, 2.2976, 2.8038, 2.1517, 2.2440, 2.2304, 3.1687], + device='cuda:3'), covar=tensor([0.0872, 0.1399, 0.2529, 0.1720, 0.1589, 0.1925, 0.1960, 0.0471], + device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0084, 0.0094, 0.0073, 0.0077, 0.0077, 0.0085, 0.0061], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:08:57,891 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.709e+02 2.078e+02 2.678e+02 6.792e+02, threshold=4.156e+02, percent-clipped=5.0 +2022-11-15 23:09:05,714 INFO [train.py:876] (3/4) Epoch 7, batch 6700, loss[loss=0.1761, simple_loss=0.1787, pruned_loss=0.08672, over 5150.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.1642, pruned_loss=0.06738, over 1077180.02 frames. ], batch size: 91, lr: 1.12e-02, grad_scale: 16.0 +2022-11-15 23:09:49,150 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1263, 2.3928, 3.8784, 3.1924, 4.4389, 2.7010, 3.9865, 4.4193], + device='cuda:3'), covar=tensor([0.0522, 0.1667, 0.0712, 0.1423, 0.0286, 0.1462, 0.1033, 0.0499], + device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0187, 0.0194, 0.0202, 0.0204, 0.0186, 0.0218, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 23:10:05,742 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.792e+02 2.397e+02 3.058e+02 5.863e+02, threshold=4.794e+02, percent-clipped=9.0 +2022-11-15 23:10:06,561 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:10:13,767 INFO [train.py:876] (3/4) Epoch 7, batch 6800, loss[loss=0.1117, simple_loss=0.1343, pruned_loss=0.0446, over 5494.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.1631, pruned_loss=0.06665, over 1086071.24 frames. ], batch size: 10, lr: 1.11e-02, grad_scale: 16.0 +2022-11-15 23:10:20,056 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.65 vs. limit=5.0 +2022-11-15 23:10:22,588 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2022-11-15 23:10:39,129 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:10:48,162 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:11:12,537 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.731e+02 2.163e+02 2.798e+02 6.704e+02, threshold=4.325e+02, percent-clipped=4.0 +2022-11-15 23:11:20,771 INFO [train.py:876] (3/4) Epoch 7, batch 6900, loss[loss=0.1279, simple_loss=0.151, pruned_loss=0.05238, over 5694.00 frames. ], tot_loss[loss=0.1494, simple_loss=0.164, pruned_loss=0.06746, over 1082422.64 frames. ], batch size: 36, lr: 1.11e-02, grad_scale: 16.0 +2022-11-15 23:11:29,494 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:11:34,022 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:11:44,507 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9590, 2.3026, 2.5244, 1.4279, 0.8838, 2.8162, 2.1154, 1.4076], + device='cuda:3'), covar=tensor([0.1121, 0.0694, 0.0526, 0.2302, 0.2365, 0.0721, 0.1376, 0.1206], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0053, 0.0057, 0.0070, 0.0055, 0.0046, 0.0051, 0.0059], + 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') +2022-11-15 23:11:44,702 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-15 23:11:49,254 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.77 vs. limit=5.0 +2022-11-15 23:12:10,969 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:12:20,424 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.773e+02 2.130e+02 2.487e+02 4.682e+02, threshold=4.260e+02, percent-clipped=1.0 +2022-11-15 23:12:22,635 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 +2022-11-15 23:12:28,770 INFO [train.py:876] (3/4) Epoch 7, batch 7000, loss[loss=0.142, simple_loss=0.1633, pruned_loss=0.06041, over 5745.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.1631, pruned_loss=0.06715, over 1083426.89 frames. ], batch size: 20, lr: 1.11e-02, grad_scale: 16.0 +2022-11-15 23:13:27,405 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.720e+02 2.093e+02 2.606e+02 4.257e+02, threshold=4.187e+02, percent-clipped=0.0 +2022-11-15 23:13:35,679 INFO [train.py:876] (3/4) Epoch 7, batch 7100, loss[loss=0.1678, simple_loss=0.1744, pruned_loss=0.08059, over 5295.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.1632, pruned_loss=0.06582, over 1087584.43 frames. ], batch size: 79, lr: 1.11e-02, grad_scale: 16.0 +2022-11-15 23:13:48,794 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3691, 0.9404, 1.4071, 0.7653, 1.4159, 1.3562, 0.6950, 1.0949], + device='cuda:3'), covar=tensor([0.0317, 0.0479, 0.0243, 0.1026, 0.0502, 0.0793, 0.0857, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0013, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.5079e-05, 7.4521e-05, 5.5912e-05, 6.4968e-05, 6.1107e-05, 5.5318e-05, + 6.8935e-05, 5.6088e-05], device='cuda:3') +2022-11-15 23:13:52,624 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-15 23:13:55,042 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2407, 4.4295, 2.9520, 4.0800, 3.2767, 2.8215, 2.2178, 3.7571], + device='cuda:3'), covar=tensor([0.1566, 0.0159, 0.0959, 0.0243, 0.0580, 0.1036, 0.1995, 0.0253], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0135, 0.0165, 0.0139, 0.0172, 0.0174, 0.0174, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 23:14:11,198 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:14:12,517 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:14:29,886 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:14:38,106 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.697e+02 1.983e+02 2.631e+02 5.249e+02, threshold=3.966e+02, percent-clipped=2.0 +2022-11-15 23:14:42,844 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3803, 1.5599, 1.5185, 1.0469, 1.3932, 1.2190, 1.2077, 1.4063], + device='cuda:3'), covar=tensor([0.0036, 0.0039, 0.0033, 0.0040, 0.0033, 0.0027, 0.0027, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0036, 0.0038, 0.0040, 0.0037, 0.0034, 0.0038, 0.0031], + device='cuda:3'), out_proj_covar=tensor([3.6902e-05, 3.2858e-05, 3.4710e-05, 3.6395e-05, 3.2541e-05, 2.9223e-05, + 3.5799e-05, 2.7703e-05], device='cuda:3') +2022-11-15 23:14:45,332 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:14:45,961 INFO [train.py:876] (3/4) Epoch 7, batch 7200, loss[loss=0.1082, simple_loss=0.1341, pruned_loss=0.04114, over 5693.00 frames. ], tot_loss[loss=0.1467, simple_loss=0.1623, pruned_loss=0.06554, over 1085049.87 frames. ], batch size: 11, lr: 1.11e-02, grad_scale: 16.0 +2022-11-15 23:14:53,262 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:14:59,650 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:15:10,562 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:15:28,579 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6285, 1.1473, 0.9636, 0.8258, 0.8155, 1.2237, 0.9557, 0.8503], + device='cuda:3'), covar=tensor([0.1884, 0.0408, 0.1780, 0.1626, 0.1732, 0.0352, 0.1965, 0.1521], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0053, 0.0057, 0.0070, 0.0055, 0.0045, 0.0052, 0.0059], + 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') +2022-11-15 23:15:31,058 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:15:31,701 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:15:31,761 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6697, 2.2779, 3.1070, 1.7752, 1.3707, 2.9289, 2.5116, 1.9578], + device='cuda:3'), covar=tensor([0.0562, 0.0809, 0.0344, 0.2107, 0.2059, 0.2670, 0.1025, 0.1194], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0053, 0.0057, 0.0070, 0.0055, 0.0045, 0.0051, 0.0059], + 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') +2022-11-15 23:16:19,012 INFO [train.py:876] (3/4) Epoch 8, batch 0, loss[loss=0.0931, simple_loss=0.1218, pruned_loss=0.03219, over 5489.00 frames. ], tot_loss[loss=0.0931, simple_loss=0.1218, pruned_loss=0.03219, over 5489.00 frames. ], batch size: 12, lr: 1.05e-02, grad_scale: 16.0 +2022-11-15 23:16:19,012 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 23:16:35,653 INFO [train.py:908] (3/4) Epoch 8, validation: loss=0.161, simple_loss=0.1821, pruned_loss=0.06991, over 1530663.00 frames. +2022-11-15 23:16:35,654 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 23:16:38,343 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8058, 2.7411, 2.3457, 3.0447, 2.3353, 2.4900, 2.8049, 3.5295], + device='cuda:3'), covar=tensor([0.0810, 0.1291, 0.2455, 0.1130, 0.1562, 0.0944, 0.1694, 0.0977], + device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0086, 0.0096, 0.0075, 0.0081, 0.0079, 0.0089, 0.0062], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:16:45,814 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.842e+02 2.230e+02 2.830e+02 5.263e+02, threshold=4.459e+02, percent-clipped=7.0 +2022-11-15 23:17:00,351 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0194, 1.4577, 1.3034, 1.0156, 0.9451, 1.2385, 0.6934, 1.1325], + device='cuda:3'), covar=tensor([0.0040, 0.0023, 0.0039, 0.0036, 0.0029, 0.0031, 0.0052, 0.0037], + device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0035, 0.0038, 0.0039, 0.0036, 0.0033, 0.0037, 0.0031], + device='cuda:3'), out_proj_covar=tensor([3.5973e-05, 3.1989e-05, 3.4514e-05, 3.5765e-05, 3.1863e-05, 2.8815e-05, + 3.5212e-05, 2.7106e-05], device='cuda:3') +2022-11-15 23:17:02,830 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6711, 4.2042, 4.3862, 4.2330, 4.7767, 4.6243, 4.2223, 4.6793], + device='cuda:3'), covar=tensor([0.0361, 0.0339, 0.0517, 0.0328, 0.0337, 0.0186, 0.0285, 0.0342], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0127, 0.0097, 0.0127, 0.0142, 0.0083, 0.0106, 0.0126], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:17:02,915 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:17:20,568 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 +2022-11-15 23:17:42,749 INFO [train.py:876] (3/4) Epoch 8, batch 100, loss[loss=0.1674, simple_loss=0.1588, pruned_loss=0.08806, over 4245.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.1635, pruned_loss=0.06688, over 432193.85 frames. ], batch size: 181, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:17:44,302 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:17:53,332 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.590e+02 1.934e+02 2.468e+02 5.065e+02, threshold=3.869e+02, percent-clipped=2.0 +2022-11-15 23:18:21,357 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:18:49,819 INFO [train.py:876] (3/4) Epoch 8, batch 200, loss[loss=0.1812, simple_loss=0.1724, pruned_loss=0.09502, over 5026.00 frames. ], tot_loss[loss=0.148, simple_loss=0.163, pruned_loss=0.06646, over 683258.67 frames. ], batch size: 110, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:19:00,022 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.568e+01 1.812e+02 2.179e+02 2.624e+02 4.566e+02, threshold=4.359e+02, percent-clipped=4.0 +2022-11-15 23:19:01,584 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:19:11,894 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:19:29,231 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 23:19:34,468 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:19:38,312 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7182, 1.1370, 1.2564, 0.9845, 1.6586, 1.1697, 0.9642, 1.1513], + device='cuda:3'), covar=tensor([0.0622, 0.0611, 0.0463, 0.0879, 0.0310, 0.0757, 0.0833, 0.0480], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.5287e-05, 7.3517e-05, 5.5280e-05, 6.4824e-05, 5.9116e-05, 5.5060e-05, + 6.8539e-05, 5.6143e-05], device='cuda:3') +2022-11-15 23:19:46,666 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8312, 4.4201, 4.5672, 4.4136, 4.9199, 4.7058, 4.3148, 4.8463], + device='cuda:3'), covar=tensor([0.0295, 0.0271, 0.0409, 0.0263, 0.0297, 0.0138, 0.0226, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0127, 0.0097, 0.0126, 0.0140, 0.0083, 0.0107, 0.0126], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:19:55,177 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:19:56,970 INFO [train.py:876] (3/4) Epoch 8, batch 300, loss[loss=0.1068, simple_loss=0.1363, pruned_loss=0.03863, over 5443.00 frames. ], tot_loss[loss=0.148, simple_loss=0.1638, pruned_loss=0.06608, over 844756.16 frames. ], batch size: 9, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:19:57,689 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-15 23:20:07,737 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.685e+02 2.012e+02 2.730e+02 5.121e+02, threshold=4.024e+02, percent-clipped=2.0 +2022-11-15 23:20:08,501 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7104, 1.9420, 1.6096, 1.8726, 1.9642, 1.8105, 1.5601, 1.8667], + device='cuda:3'), covar=tensor([0.0450, 0.0719, 0.1687, 0.0750, 0.0686, 0.0539, 0.1478, 0.0665], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0161, 0.0255, 0.0158, 0.0205, 0.0159, 0.0175, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:20:10,494 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5955, 1.9280, 1.8749, 1.6674, 1.8490, 1.9210, 0.9006, 1.9492], + device='cuda:3'), covar=tensor([0.0419, 0.0392, 0.0366, 0.0364, 0.0442, 0.0341, 0.2379, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0082, 0.0080, 0.0073, 0.0099, 0.0083, 0.0130, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:20:13,093 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3555, 4.6459, 3.2300, 4.5573, 3.5308, 3.1186, 2.9852, 4.0346], + device='cuda:3'), covar=tensor([0.1705, 0.0231, 0.0998, 0.0237, 0.0530, 0.0957, 0.1652, 0.0250], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0137, 0.0166, 0.0140, 0.0174, 0.0176, 0.0176, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 23:20:15,728 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:20:27,337 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:21:02,616 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2022-11-15 23:21:03,053 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:21:05,008 INFO [train.py:876] (3/4) Epoch 8, batch 400, loss[loss=0.1275, simple_loss=0.1509, pruned_loss=0.05207, over 5740.00 frames. ], tot_loss[loss=0.145, simple_loss=0.1613, pruned_loss=0.06437, over 934938.01 frames. ], batch size: 27, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:21:16,262 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.652e+01 1.583e+02 1.915e+02 2.557e+02 6.087e+02, threshold=3.830e+02, percent-clipped=2.0 +2022-11-15 23:21:31,489 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-15 23:21:52,581 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7262, 4.0872, 3.0860, 3.8491, 3.6606, 3.6232, 4.0187, 3.8742], + device='cuda:3'), covar=tensor([0.0594, 0.0763, 0.2401, 0.1027, 0.1391, 0.0535, 0.0596, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0161, 0.0257, 0.0158, 0.0204, 0.0159, 0.0174, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 23:21:55,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0047, 2.3576, 2.9976, 3.8790, 3.7568, 3.2576, 2.4710, 3.9598], + device='cuda:3'), covar=tensor([0.0471, 0.3054, 0.2030, 0.3024, 0.1058, 0.2714, 0.2230, 0.0476], + device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0208, 0.0205, 0.0326, 0.0229, 0.0219, 0.0199, 0.0220], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 23:21:57,203 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:21:58,519 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6450, 2.5281, 2.0950, 2.6719, 2.0124, 2.2507, 2.3455, 2.9001], + device='cuda:3'), covar=tensor([0.1740, 0.2069, 0.4132, 0.1910, 0.3620, 0.1780, 0.2857, 0.4542], + device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0087, 0.0098, 0.0077, 0.0082, 0.0080, 0.0090, 0.0064], + 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') +2022-11-15 23:22:03,772 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:22:08,548 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-15 23:22:12,920 INFO [train.py:876] (3/4) Epoch 8, batch 500, loss[loss=0.114, simple_loss=0.1506, pruned_loss=0.0387, over 5701.00 frames. ], tot_loss[loss=0.1457, simple_loss=0.1624, pruned_loss=0.06452, over 997528.36 frames. ], batch size: 17, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:22:21,688 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:22:23,653 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.464e+01 1.684e+02 2.091e+02 2.743e+02 4.142e+02, threshold=4.181e+02, percent-clipped=1.0 +2022-11-15 23:22:30,192 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6590, 4.3985, 3.2880, 1.9884, 4.1082, 1.9367, 4.1567, 2.3484], + device='cuda:3'), covar=tensor([0.1246, 0.0112, 0.0552, 0.2098, 0.0154, 0.1613, 0.0158, 0.1477], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0107, 0.0117, 0.0117, 0.0107, 0.0127, 0.0099, 0.0117], + 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') +2022-11-15 23:22:35,554 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:22:38,572 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 23:22:38,887 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:22:45,528 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:22:53,369 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:23:08,424 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:23:13,111 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5608, 2.6589, 2.1284, 2.9147, 1.9861, 2.4258, 2.4020, 3.2538], + device='cuda:3'), covar=tensor([0.1374, 0.1889, 0.4436, 0.1582, 0.3868, 0.2302, 0.3173, 0.2977], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0087, 0.0098, 0.0078, 0.0082, 0.0081, 0.0091, 0.0064], + 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') +2022-11-15 23:23:14,486 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:23:21,002 INFO [train.py:876] (3/4) Epoch 8, batch 600, loss[loss=0.1033, simple_loss=0.1397, pruned_loss=0.0335, over 5584.00 frames. ], tot_loss[loss=0.1446, simple_loss=0.1611, pruned_loss=0.06403, over 1032171.45 frames. ], batch size: 16, lr: 1.04e-02, grad_scale: 32.0 +2022-11-15 23:23:26,056 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:23:32,193 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.319e+01 1.675e+02 2.028e+02 2.576e+02 4.109e+02, threshold=4.056e+02, percent-clipped=0.0 +2022-11-15 23:23:34,389 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3230, 0.9148, 1.0434, 0.6900, 1.1050, 1.0251, 0.7680, 1.0470], + device='cuda:3'), covar=tensor([0.0433, 0.0589, 0.0271, 0.0935, 0.0259, 0.0589, 0.0663, 0.0362], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0013, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6152e-05, 7.4768e-05, 5.5745e-05, 6.6265e-05, 6.0960e-05, 5.5457e-05, + 6.9342e-05, 5.6124e-05], device='cuda:3') +2022-11-15 23:23:37,229 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 23:23:56,195 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:24:28,212 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:24:30,078 INFO [train.py:876] (3/4) Epoch 8, batch 700, loss[loss=0.1608, simple_loss=0.1626, pruned_loss=0.07949, over 5517.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.1604, pruned_loss=0.06348, over 1054227.17 frames. ], batch size: 49, lr: 1.04e-02, grad_scale: 32.0 +2022-11-15 23:24:37,649 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6766, 4.1800, 3.6499, 3.5907, 2.1361, 3.9483, 2.0989, 3.2470], + device='cuda:3'), covar=tensor([0.0371, 0.0139, 0.0165, 0.0261, 0.0503, 0.0125, 0.0500, 0.0171], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0148, 0.0162, 0.0183, 0.0177, 0.0161, 0.0175, 0.0157], + 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') +2022-11-15 23:24:39,231 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 +2022-11-15 23:24:40,706 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.048e+01 1.604e+02 2.114e+02 2.490e+02 4.177e+02, threshold=4.229e+02, percent-clipped=3.0 +2022-11-15 23:24:46,028 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5719, 3.6472, 3.4346, 3.1533, 2.0031, 3.5528, 2.0648, 3.0112], + device='cuda:3'), covar=tensor([0.0343, 0.0158, 0.0151, 0.0313, 0.0450, 0.0137, 0.0446, 0.0127], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0147, 0.0162, 0.0183, 0.0176, 0.0160, 0.0174, 0.0157], + 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') +2022-11-15 23:25:01,592 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:25:22,654 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3316, 1.1390, 1.2562, 0.8288, 1.3359, 1.3833, 0.8802, 0.9200], + device='cuda:3'), covar=tensor([0.0645, 0.0583, 0.1869, 0.1097, 0.0778, 0.0413, 0.0937, 0.1589], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0013, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.7036e-05, 7.5557e-05, 5.6434e-05, 6.7417e-05, 6.2130e-05, 5.6000e-05, + 7.0244e-05, 5.7500e-05], device='cuda:3') +2022-11-15 23:25:32,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2022-11-15 23:25:39,226 INFO [train.py:876] (3/4) Epoch 8, batch 800, loss[loss=0.08723, simple_loss=0.1174, pruned_loss=0.02854, over 5290.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.161, pruned_loss=0.06318, over 1072539.46 frames. ], batch size: 9, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:25:47,918 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:25:50,398 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.606e+01 1.543e+02 1.960e+02 2.359e+02 4.121e+02, threshold=3.919e+02, percent-clipped=0.0 +2022-11-15 23:25:51,834 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1553, 4.6152, 4.0620, 4.5526, 4.5617, 3.8127, 4.3628, 4.0990], + device='cuda:3'), covar=tensor([0.0414, 0.0430, 0.1481, 0.0454, 0.0429, 0.0437, 0.0389, 0.0482], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0165, 0.0258, 0.0157, 0.0203, 0.0159, 0.0173, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 23:26:01,940 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:26:08,360 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:26:08,937 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:26:20,930 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:26:47,892 INFO [train.py:876] (3/4) Epoch 8, batch 900, loss[loss=0.1485, simple_loss=0.1645, pruned_loss=0.06621, over 5605.00 frames. ], tot_loss[loss=0.1449, simple_loss=0.1615, pruned_loss=0.06419, over 1074721.24 frames. ], batch size: 24, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:26:50,119 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:26:59,515 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 1.796e+02 2.172e+02 2.751e+02 5.616e+02, threshold=4.345e+02, percent-clipped=4.0 +2022-11-15 23:27:03,650 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={2} +2022-11-15 23:27:19,694 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:27:37,041 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:27:38,396 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2435, 2.0198, 1.9174, 2.3102, 1.8724, 1.5391, 2.0082, 2.3191], + device='cuda:3'), covar=tensor([0.0997, 0.1820, 0.2791, 0.1146, 0.1878, 0.2087, 0.2094, 0.2117], + device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0085, 0.0096, 0.0076, 0.0079, 0.0079, 0.0087, 0.0063], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:27:49,033 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.68 vs. limit=2.0 +2022-11-15 23:27:56,256 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 +2022-11-15 23:27:57,326 INFO [train.py:876] (3/4) Epoch 8, batch 1000, loss[loss=0.1227, simple_loss=0.1474, pruned_loss=0.04904, over 5771.00 frames. ], tot_loss[loss=0.1455, simple_loss=0.1625, pruned_loss=0.0642, over 1078823.47 frames. ], batch size: 14, lr: 1.04e-02, grad_scale: 16.0 +2022-11-15 23:28:08,738 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.780e+02 2.132e+02 2.745e+02 5.068e+02, threshold=4.264e+02, percent-clipped=2.0 +2022-11-15 23:28:19,512 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4439, 4.5225, 4.5291, 4.2315, 4.3788, 4.5421, 2.1802, 4.6798], + device='cuda:3'), covar=tensor([0.0250, 0.0358, 0.0279, 0.0389, 0.0414, 0.0279, 0.2521, 0.0319], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0081, 0.0080, 0.0073, 0.0099, 0.0082, 0.0130, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:28:35,273 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7572, 2.9234, 2.8861, 2.6576, 2.8623, 2.9478, 1.0347, 2.9775], + device='cuda:3'), covar=tensor([0.0438, 0.0296, 0.0321, 0.0357, 0.0484, 0.0332, 0.3302, 0.0390], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0081, 0.0080, 0.0073, 0.0099, 0.0082, 0.0130, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:28:35,997 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4465, 1.0794, 1.2830, 0.7890, 1.3883, 1.3302, 0.7710, 0.8898], + device='cuda:3'), covar=tensor([0.0347, 0.0875, 0.0263, 0.0969, 0.0758, 0.0343, 0.0578, 0.0489], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.5527e-05, 7.3409e-05, 5.4731e-05, 6.5660e-05, 6.0466e-05, 5.4620e-05, + 6.7607e-05, 5.6052e-05], device='cuda:3') +2022-11-15 23:28:39,668 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4101, 1.9990, 3.0390, 2.5959, 3.1115, 2.0403, 2.8219, 3.3874], + device='cuda:3'), covar=tensor([0.0749, 0.1708, 0.0816, 0.1435, 0.0621, 0.1637, 0.1027, 0.0708], + device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0198, 0.0202, 0.0213, 0.0217, 0.0192, 0.0224, 0.0219], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 23:29:04,722 INFO [train.py:876] (3/4) Epoch 8, batch 1100, loss[loss=0.1025, simple_loss=0.1318, pruned_loss=0.03663, over 5480.00 frames. ], tot_loss[loss=0.1449, simple_loss=0.1624, pruned_loss=0.06373, over 1081331.50 frames. ], batch size: 11, lr: 1.03e-02, grad_scale: 16.0 +2022-11-15 23:29:16,567 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.856e+01 1.738e+02 2.117e+02 2.536e+02 5.317e+02, threshold=4.235e+02, percent-clipped=1.0 +2022-11-15 23:29:19,700 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4096, 1.2778, 1.8425, 1.3069, 1.3766, 1.7607, 1.2069, 1.1369], + device='cuda:3'), covar=tensor([0.0025, 0.0071, 0.0041, 0.0039, 0.0044, 0.0032, 0.0027, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0025, 0.0023, 0.0021, 0.0024, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8885e-05, 1.9755e-05, 1.9082e-05, 2.5378e-05, 2.1624e-05, 2.0688e-05, + 2.3764e-05, 2.6098e-05], device='cuda:3') +2022-11-15 23:29:25,949 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2022-11-15 23:29:26,356 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7733, 4.0273, 3.7443, 3.3405, 2.2357, 3.9661, 2.4106, 3.2641], + device='cuda:3'), covar=tensor([0.0378, 0.0123, 0.0198, 0.0351, 0.0494, 0.0155, 0.0460, 0.0132], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0147, 0.0161, 0.0182, 0.0174, 0.0160, 0.0173, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-15 23:29:27,943 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:29:29,220 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:29:34,326 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:29:59,645 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:30:06,072 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:30:09,796 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:30:10,945 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:30:12,209 INFO [train.py:876] (3/4) Epoch 8, batch 1200, loss[loss=0.136, simple_loss=0.1553, pruned_loss=0.05831, over 5652.00 frames. ], tot_loss[loss=0.1432, simple_loss=0.1611, pruned_loss=0.0626, over 1084324.19 frames. ], batch size: 36, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:30:23,865 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 1.672e+02 2.083e+02 2.557e+02 4.587e+02, threshold=4.167e+02, percent-clipped=2.0 +2022-11-15 23:30:43,488 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:30:53,793 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.51 vs. limit=5.0 +2022-11-15 23:31:12,356 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:31:16,112 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:31:20,578 INFO [train.py:876] (3/4) Epoch 8, batch 1300, loss[loss=0.08646, simple_loss=0.1163, pruned_loss=0.02832, over 5483.00 frames. ], tot_loss[loss=0.1424, simple_loss=0.1605, pruned_loss=0.06214, over 1084873.87 frames. ], batch size: 10, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:31:32,442 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.693e+02 2.062e+02 2.550e+02 7.238e+02, threshold=4.125e+02, percent-clipped=3.0 +2022-11-15 23:31:54,158 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:32:08,005 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 23:32:27,890 INFO [train.py:876] (3/4) Epoch 8, batch 1400, loss[loss=0.1931, simple_loss=0.1959, pruned_loss=0.09508, over 5554.00 frames. ], tot_loss[loss=0.1427, simple_loss=0.1603, pruned_loss=0.06256, over 1083813.39 frames. ], batch size: 46, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:32:32,565 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:32:39,667 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.794e+02 2.190e+02 2.627e+02 5.142e+02, threshold=4.380e+02, percent-clipped=4.0 +2022-11-15 23:32:43,631 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9317, 4.4772, 4.7266, 4.4664, 5.0255, 4.8786, 4.4089, 4.9970], + device='cuda:3'), covar=tensor([0.0366, 0.0254, 0.0418, 0.0311, 0.0306, 0.0150, 0.0240, 0.0233], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0130, 0.0099, 0.0129, 0.0145, 0.0087, 0.0109, 0.0132], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:33:00,595 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9513, 2.5818, 2.8211, 2.5041, 1.6514, 2.7489, 1.7402, 2.1211], + device='cuda:3'), covar=tensor([0.0351, 0.0172, 0.0144, 0.0281, 0.0420, 0.0172, 0.0402, 0.0186], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0150, 0.0163, 0.0186, 0.0179, 0.0163, 0.0176, 0.0163], + 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') +2022-11-15 23:33:13,759 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:33:29,155 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:33:33,946 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:33:35,125 INFO [train.py:876] (3/4) Epoch 8, batch 1500, loss[loss=0.2194, simple_loss=0.2191, pruned_loss=0.1099, over 5484.00 frames. ], tot_loss[loss=0.1425, simple_loss=0.1601, pruned_loss=0.06245, over 1087328.25 frames. ], batch size: 64, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:33:47,237 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.630e+02 1.908e+02 2.524e+02 5.804e+02, threshold=3.816e+02, percent-clipped=2.0 +2022-11-15 23:34:06,269 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:34:14,594 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0207, 4.5548, 4.8504, 4.5470, 5.1111, 5.0614, 4.4279, 5.1235], + device='cuda:3'), covar=tensor([0.0403, 0.0277, 0.0420, 0.0286, 0.0344, 0.0117, 0.0269, 0.0232], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0133, 0.0101, 0.0131, 0.0148, 0.0088, 0.0112, 0.0133], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:34:38,433 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2022-11-15 23:34:39,310 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2808, 3.6192, 2.8141, 1.6021, 3.4669, 1.3545, 3.4555, 1.7817], + device='cuda:3'), covar=tensor([0.1755, 0.0221, 0.1019, 0.2569, 0.0309, 0.2692, 0.0316, 0.2331], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0106, 0.0117, 0.0116, 0.0106, 0.0128, 0.0100, 0.0117], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 23:34:42,715 INFO [train.py:876] (3/4) Epoch 8, batch 1600, loss[loss=0.1164, simple_loss=0.1418, pruned_loss=0.0455, over 5534.00 frames. ], tot_loss[loss=0.1399, simple_loss=0.1587, pruned_loss=0.06059, over 1087291.22 frames. ], batch size: 14, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:34:47,240 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3022, 4.1497, 4.2266, 3.9595, 4.1379, 4.1934, 1.7704, 4.4900], + device='cuda:3'), covar=tensor([0.0303, 0.0670, 0.0387, 0.0445, 0.0475, 0.0419, 0.3316, 0.0418], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0082, 0.0079, 0.0073, 0.0097, 0.0082, 0.0127, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:34:55,355 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.681e+02 2.048e+02 2.311e+02 7.167e+02, threshold=4.097e+02, percent-clipped=4.0 +2022-11-15 23:35:01,280 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1841, 3.9693, 2.6485, 3.7633, 3.0152, 2.7986, 2.1479, 3.2952], + device='cuda:3'), covar=tensor([0.1699, 0.0217, 0.1159, 0.0342, 0.0815, 0.1055, 0.2175, 0.0359], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0141, 0.0169, 0.0143, 0.0179, 0.0180, 0.0176, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 23:35:13,816 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:35:16,729 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3535, 0.9770, 1.1422, 0.8665, 1.1557, 1.1592, 0.9179, 0.9760], + device='cuda:3'), covar=tensor([0.1023, 0.0626, 0.0353, 0.0728, 0.0523, 0.0578, 0.0665, 0.0676], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6398e-05, 7.3494e-05, 5.5089e-05, 6.4990e-05, 6.0673e-05, 5.4948e-05, + 6.9024e-05, 5.6322e-05], device='cuda:3') +2022-11-15 23:35:18,163 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={1} +2022-11-15 23:35:44,636 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3301, 1.5615, 1.7376, 1.6512, 1.5359, 1.4458, 1.5557, 1.5893], + device='cuda:3'), covar=tensor([0.2634, 0.2465, 0.2090, 0.1489, 0.2206, 0.2678, 0.2195, 0.0814], + device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0086, 0.0096, 0.0080, 0.0082, 0.0081, 0.0089, 0.0064], + 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') +2022-11-15 23:35:52,210 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9929, 2.6203, 3.0598, 3.9118, 4.0052, 3.2699, 2.5420, 3.9079], + device='cuda:3'), covar=tensor([0.0575, 0.3083, 0.2490, 0.4732, 0.1148, 0.3174, 0.2391, 0.0748], + device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0210, 0.0202, 0.0328, 0.0227, 0.0218, 0.0198, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 23:35:54,042 INFO [train.py:876] (3/4) Epoch 8, batch 1700, loss[loss=0.1748, simple_loss=0.1754, pruned_loss=0.0871, over 5337.00 frames. ], tot_loss[loss=0.1398, simple_loss=0.1585, pruned_loss=0.06056, over 1090621.43 frames. ], batch size: 70, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:36:02,209 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 23:36:02,855 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0591, 1.3412, 1.2661, 0.9992, 1.5527, 1.3955, 1.5099, 1.0813], + device='cuda:3'), covar=tensor([0.2261, 0.1038, 0.1062, 0.1196, 0.0646, 0.1121, 0.0610, 0.1470], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6274e-05, 7.3544e-05, 5.5335e-05, 6.5252e-05, 6.0871e-05, 5.5158e-05, + 6.9044e-05, 5.6558e-05], device='cuda:3') +2022-11-15 23:36:06,801 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.756e+02 2.105e+02 2.573e+02 4.026e+02, threshold=4.210e+02, percent-clipped=0.0 +2022-11-15 23:36:12,089 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2738, 1.1256, 1.6285, 1.0715, 1.6020, 1.5454, 1.2518, 1.3990], + device='cuda:3'), covar=tensor([0.0763, 0.0675, 0.0338, 0.0806, 0.0444, 0.0458, 0.0441, 0.0374], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6230e-05, 7.3315e-05, 5.5093e-05, 6.5030e-05, 6.0859e-05, 5.5046e-05, + 6.8826e-05, 5.6147e-05], device='cuda:3') +2022-11-15 23:36:38,979 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:36:59,018 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:37:03,717 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2022-11-15 23:37:05,450 INFO [train.py:876] (3/4) Epoch 8, batch 1800, loss[loss=0.1813, simple_loss=0.1812, pruned_loss=0.09067, over 5680.00 frames. ], tot_loss[loss=0.1386, simple_loss=0.1576, pruned_loss=0.05986, over 1090461.86 frames. ], batch size: 36, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:37:18,119 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.718e+02 1.975e+02 2.529e+02 4.433e+02, threshold=3.950e+02, percent-clipped=1.0 +2022-11-15 23:37:33,370 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:37:38,970 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:38:05,617 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4080, 2.3239, 2.2222, 2.3965, 2.4613, 2.2628, 2.6660, 2.4586], + device='cuda:3'), covar=tensor([0.0676, 0.1111, 0.0732, 0.1245, 0.0671, 0.0505, 0.1015, 0.0883], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0103, 0.0088, 0.0112, 0.0083, 0.0073, 0.0139, 0.0093], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:38:17,344 INFO [train.py:876] (3/4) Epoch 8, batch 1900, loss[loss=0.1548, simple_loss=0.1726, pruned_loss=0.06846, over 5691.00 frames. ], tot_loss[loss=0.137, simple_loss=0.1566, pruned_loss=0.0587, over 1088904.77 frames. ], batch size: 19, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:38:23,478 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:38:28,386 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8506, 4.6523, 3.5417, 2.0004, 4.3549, 2.0493, 4.3471, 2.5292], + device='cuda:3'), covar=tensor([0.1609, 0.0277, 0.0575, 0.2940, 0.0348, 0.2233, 0.0368, 0.2259], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0105, 0.0116, 0.0113, 0.0104, 0.0125, 0.0097, 0.0114], + 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') +2022-11-15 23:38:30,574 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.627e+02 1.958e+02 2.548e+02 4.819e+02, threshold=3.916e+02, percent-clipped=3.0 +2022-11-15 23:38:49,592 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:39:12,890 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2022-11-15 23:39:23,807 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:39:26,431 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0255, 1.3267, 1.1582, 1.0281, 1.3947, 1.2530, 0.9469, 1.4832], + device='cuda:3'), covar=tensor([0.0041, 0.0025, 0.0046, 0.0046, 0.0033, 0.0036, 0.0059, 0.0025], + device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0038, 0.0041, 0.0040, 0.0039, 0.0036, 0.0040, 0.0034], + device='cuda:3'), out_proj_covar=tensor([3.9622e-05, 3.4921e-05, 3.7240e-05, 3.6578e-05, 3.5083e-05, 3.0678e-05, + 3.8004e-05, 3.0082e-05], device='cuda:3') +2022-11-15 23:39:29,661 INFO [train.py:876] (3/4) Epoch 8, batch 2000, loss[loss=0.1549, simple_loss=0.1623, pruned_loss=0.0737, over 5270.00 frames. ], tot_loss[loss=0.1385, simple_loss=0.157, pruned_loss=0.06, over 1082428.98 frames. ], batch size: 79, lr: 1.03e-02, grad_scale: 8.0 +2022-11-15 23:39:33,830 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:39:42,130 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.867e+01 1.663e+02 2.013e+02 2.665e+02 5.051e+02, threshold=4.025e+02, percent-clipped=6.0 +2022-11-15 23:39:52,905 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5026, 3.3658, 3.4742, 3.3363, 2.2192, 3.4365, 2.1763, 2.8382], + device='cuda:3'), covar=tensor([0.0286, 0.0235, 0.0139, 0.0192, 0.0360, 0.0133, 0.0386, 0.0124], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0152, 0.0163, 0.0184, 0.0178, 0.0162, 0.0176, 0.0161], + 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') +2022-11-15 23:39:53,599 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5339, 1.2542, 1.9129, 1.3816, 1.0210, 1.8291, 1.5217, 1.4500], + device='cuda:3'), covar=tensor([0.0030, 0.0088, 0.0031, 0.0046, 0.0069, 0.0057, 0.0026, 0.0033], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0020, 0.0024, 0.0022, 0.0020, 0.0023, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.7451e-05, 1.8603e-05, 1.8425e-05, 2.4298e-05, 2.1232e-05, 1.9557e-05, + 2.2515e-05, 2.5543e-05], device='cuda:3') +2022-11-15 23:40:14,208 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:40:34,097 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 +2022-11-15 23:40:40,879 INFO [train.py:876] (3/4) Epoch 8, batch 2100, loss[loss=0.146, simple_loss=0.1783, pruned_loss=0.05689, over 5613.00 frames. ], tot_loss[loss=0.1389, simple_loss=0.1579, pruned_loss=0.05994, over 1081463.25 frames. ], batch size: 23, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:40:48,790 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:40:53,707 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 1.690e+02 2.165e+02 2.534e+02 4.185e+02, threshold=4.330e+02, percent-clipped=4.0 +2022-11-15 23:41:10,095 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6801, 4.5000, 4.8254, 4.8565, 4.4342, 4.1072, 5.2639, 4.6599], + device='cuda:3'), covar=tensor([0.0461, 0.1198, 0.0455, 0.1096, 0.0407, 0.0345, 0.0906, 0.0676], + device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0101, 0.0085, 0.0109, 0.0081, 0.0071, 0.0135, 0.0091], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:41:17,509 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:41:20,716 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 +2022-11-15 23:41:41,544 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1955, 4.7912, 4.9420, 4.7324, 5.3246, 5.2604, 4.5661, 5.2788], + device='cuda:3'), covar=tensor([0.0373, 0.0274, 0.0438, 0.0270, 0.0285, 0.0125, 0.0233, 0.0242], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0134, 0.0101, 0.0133, 0.0149, 0.0090, 0.0111, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-15 23:41:50,908 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5374, 1.2341, 1.7713, 1.1497, 1.0627, 1.8101, 1.4156, 1.1436], + device='cuda:3'), covar=tensor([0.0019, 0.0069, 0.0048, 0.0042, 0.0069, 0.0031, 0.0028, 0.0043], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0025, 0.0024, 0.0021, 0.0024, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8288e-05, 1.9748e-05, 1.9383e-05, 2.5449e-05, 2.2394e-05, 2.0376e-05, + 2.3785e-05, 2.6470e-05], device='cuda:3') +2022-11-15 23:41:52,187 INFO [train.py:876] (3/4) Epoch 8, batch 2200, loss[loss=0.196, simple_loss=0.1742, pruned_loss=0.1089, over 4137.00 frames. ], tot_loss[loss=0.1406, simple_loss=0.1591, pruned_loss=0.06108, over 1083023.66 frames. ], batch size: 181, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:41:54,663 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:42:00,747 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2022-11-15 23:42:01,091 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:42:05,401 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.642e+02 2.019e+02 2.545e+02 4.106e+02, threshold=4.038e+02, percent-clipped=0.0 +2022-11-15 23:42:18,252 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0 +2022-11-15 23:43:05,028 INFO [train.py:876] (3/4) Epoch 8, batch 2300, loss[loss=0.09697, simple_loss=0.1188, pruned_loss=0.03756, over 5281.00 frames. ], tot_loss[loss=0.141, simple_loss=0.1587, pruned_loss=0.06164, over 1085308.92 frames. ], batch size: 8, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:43:09,414 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:43:17,885 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.609e+02 1.989e+02 2.421e+02 4.681e+02, threshold=3.978e+02, percent-clipped=2.0 +2022-11-15 23:43:31,397 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:43:43,979 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:44:03,818 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3826, 3.6723, 3.0523, 1.6377, 3.4603, 1.4446, 3.5784, 1.9848], + device='cuda:3'), covar=tensor([0.1940, 0.0390, 0.0987, 0.3082, 0.0428, 0.2819, 0.0399, 0.2367], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0107, 0.0114, 0.0114, 0.0104, 0.0126, 0.0098, 0.0115], + 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') +2022-11-15 23:44:14,691 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:44:16,033 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:44:16,590 INFO [train.py:876] (3/4) Epoch 8, batch 2400, loss[loss=0.2433, simple_loss=0.2091, pruned_loss=0.1387, over 2923.00 frames. ], tot_loss[loss=0.1423, simple_loss=0.1594, pruned_loss=0.06257, over 1084334.34 frames. ], batch size: 284, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:44:24,083 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:44:29,616 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.667e+02 1.893e+02 2.315e+02 4.306e+02, threshold=3.787e+02, percent-clipped=3.0 +2022-11-15 23:44:31,580 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-15 23:44:36,220 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2022-11-15 23:45:00,229 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:45:01,570 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:45:07,913 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:45:28,841 INFO [train.py:876] (3/4) Epoch 8, batch 2500, loss[loss=0.1092, simple_loss=0.1331, pruned_loss=0.0426, over 5197.00 frames. ], tot_loss[loss=0.1409, simple_loss=0.1585, pruned_loss=0.06165, over 1083478.56 frames. ], batch size: 8, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:45:31,084 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:45:33,744 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:45:41,343 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.752e+02 2.233e+02 2.745e+02 4.955e+02, threshold=4.465e+02, percent-clipped=8.0 +2022-11-15 23:45:44,959 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={3} +2022-11-15 23:46:04,893 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:46:27,163 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-15 23:46:34,894 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.42 vs. limit=5.0 +2022-11-15 23:46:39,615 INFO [train.py:876] (3/4) Epoch 8, batch 2600, loss[loss=0.1342, simple_loss=0.149, pruned_loss=0.0597, over 5556.00 frames. ], tot_loss[loss=0.1407, simple_loss=0.1589, pruned_loss=0.06127, over 1086695.83 frames. ], batch size: 25, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:46:52,569 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.516e+01 1.584e+02 1.998e+02 2.446e+02 4.760e+02, threshold=3.997e+02, percent-clipped=2.0 +2022-11-15 23:46:58,649 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3730, 4.5314, 2.8747, 4.2712, 3.4658, 2.8928, 2.4322, 3.7774], + device='cuda:3'), covar=tensor([0.1539, 0.0175, 0.1077, 0.0357, 0.0645, 0.0980, 0.1893, 0.0337], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0142, 0.0167, 0.0142, 0.0177, 0.0179, 0.0177, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 23:47:28,262 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 +2022-11-15 23:47:45,390 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:47:51,091 INFO [train.py:876] (3/4) Epoch 8, batch 2700, loss[loss=0.1167, simple_loss=0.1401, pruned_loss=0.04663, over 5735.00 frames. ], tot_loss[loss=0.1402, simple_loss=0.1586, pruned_loss=0.06093, over 1087307.63 frames. ], batch size: 14, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:48:04,169 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.781e+02 2.176e+02 2.706e+02 9.486e+02, threshold=4.353e+02, percent-clipped=5.0 +2022-11-15 23:48:30,813 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:48:38,708 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:48:41,650 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-15 23:49:02,630 INFO [train.py:876] (3/4) Epoch 8, batch 2800, loss[loss=0.1134, simple_loss=0.1462, pruned_loss=0.04033, over 5751.00 frames. ], tot_loss[loss=0.1402, simple_loss=0.1583, pruned_loss=0.06104, over 1081703.68 frames. ], batch size: 16, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:49:07,933 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:49:15,686 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.500e+01 1.616e+02 2.009e+02 2.401e+02 5.865e+02, threshold=4.018e+02, percent-clipped=2.0 +2022-11-15 23:49:15,805 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:49:42,123 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:49:43,677 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2708, 2.5269, 3.0671, 4.0619, 4.0381, 3.3129, 2.4695, 4.0615], + device='cuda:3'), covar=tensor([0.0377, 0.2574, 0.2821, 0.3440, 0.1004, 0.2796, 0.2293, 0.0649], + device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0204, 0.0201, 0.0323, 0.0225, 0.0214, 0.0194, 0.0219], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 23:50:15,010 INFO [train.py:876] (3/4) Epoch 8, batch 2900, loss[loss=0.2114, simple_loss=0.1873, pruned_loss=0.1177, over 3042.00 frames. ], tot_loss[loss=0.1419, simple_loss=0.1594, pruned_loss=0.0622, over 1078211.64 frames. ], batch size: 284, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:50:21,373 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6614, 3.7587, 3.8222, 2.0378, 3.3737, 3.7813, 3.7618, 4.3438], + device='cuda:3'), covar=tensor([0.1632, 0.1031, 0.0486, 0.2260, 0.0340, 0.0485, 0.0258, 0.0405], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0186, 0.0160, 0.0193, 0.0171, 0.0186, 0.0152, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-15 23:50:27,692 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.192e+01 1.630e+02 2.037e+02 2.446e+02 6.104e+02, threshold=4.074e+02, percent-clipped=4.0 +2022-11-15 23:50:33,172 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8083, 1.4066, 1.2605, 1.0684, 0.9093, 1.6910, 1.3611, 0.9595], + device='cuda:3'), covar=tensor([0.1764, 0.0704, 0.1517, 0.2447, 0.2263, 0.0532, 0.1267, 0.1957], + device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0061, 0.0061, 0.0076, 0.0057, 0.0047, 0.0054, 0.0062], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 23:50:51,463 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2857, 4.3241, 4.3202, 4.5755, 3.8932, 3.5411, 4.9244, 4.3156], + device='cuda:3'), covar=tensor([0.0397, 0.0956, 0.0516, 0.0890, 0.0673, 0.0472, 0.0657, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0099, 0.0083, 0.0108, 0.0079, 0.0071, 0.0134, 0.0088], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:51:08,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0552, 3.5485, 3.1319, 3.4768, 3.5575, 3.0978, 3.0793, 3.0341], + device='cuda:3'), covar=tensor([0.1530, 0.0549, 0.1642, 0.0511, 0.0533, 0.0574, 0.0702, 0.0784], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0163, 0.0255, 0.0156, 0.0203, 0.0160, 0.0171, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-15 23:51:20,656 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:51:26,049 INFO [train.py:876] (3/4) Epoch 8, batch 3000, loss[loss=0.1164, simple_loss=0.146, pruned_loss=0.04335, over 5734.00 frames. ], tot_loss[loss=0.1421, simple_loss=0.1594, pruned_loss=0.06241, over 1081094.26 frames. ], batch size: 15, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:51:26,049 INFO [train.py:899] (3/4) Computing validation loss +2022-11-15 23:51:37,095 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4045, 2.5845, 3.2530, 4.1361, 4.6741, 3.4847, 2.7841, 4.1278], + device='cuda:3'), covar=tensor([0.0413, 0.3933, 0.2356, 0.3258, 0.0639, 0.2663, 0.2191, 0.0506], + device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0202, 0.0198, 0.0322, 0.0222, 0.0213, 0.0194, 0.0219], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-15 23:51:44,990 INFO [train.py:908] (3/4) Epoch 8, validation: loss=0.1608, simple_loss=0.1816, pruned_loss=0.06996, over 1530663.00 frames. +2022-11-15 23:51:44,991 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-15 23:51:57,592 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.245e+01 1.684e+02 1.979e+02 2.404e+02 5.002e+02, threshold=3.957e+02, percent-clipped=2.0 +2022-11-15 23:52:11,968 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-15 23:52:13,792 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:52:24,747 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:52:32,413 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:52:57,152 INFO [train.py:876] (3/4) Epoch 8, batch 3100, loss[loss=0.1544, simple_loss=0.165, pruned_loss=0.07187, over 5690.00 frames. ], tot_loss[loss=0.1409, simple_loss=0.1587, pruned_loss=0.06159, over 1085354.76 frames. ], batch size: 36, lr: 1.02e-02, grad_scale: 8.0 +2022-11-15 23:52:59,669 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:53:07,362 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:53:09,937 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.766e+01 1.773e+02 2.219e+02 2.737e+02 4.389e+02, threshold=4.437e+02, percent-clipped=4.0 +2022-11-15 23:53:10,107 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={0} +2022-11-15 23:53:44,200 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set() +2022-11-15 23:53:45,927 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-15 23:53:54,969 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-15 23:54:08,038 INFO [train.py:876] (3/4) Epoch 8, batch 3200, loss[loss=0.169, simple_loss=0.1653, pruned_loss=0.08637, over 4199.00 frames. ], tot_loss[loss=0.1421, simple_loss=0.1595, pruned_loss=0.06232, over 1085762.01 frames. ], batch size: 181, lr: 1.01e-02, grad_scale: 16.0 +2022-11-15 23:54:20,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7130, 2.7407, 2.1626, 2.4009, 1.5031, 2.2555, 1.5920, 2.4034], + device='cuda:3'), covar=tensor([0.1076, 0.0262, 0.0785, 0.0470, 0.1750, 0.0706, 0.1527, 0.0405], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0141, 0.0166, 0.0140, 0.0178, 0.0177, 0.0172, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-15 23:54:21,049 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.665e+02 2.003e+02 2.661e+02 5.081e+02, threshold=4.007e+02, percent-clipped=1.0 +2022-11-15 23:55:05,398 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3520, 2.3995, 2.6695, 1.6383, 1.0837, 3.0652, 2.3295, 2.0882], + device='cuda:3'), covar=tensor([0.0817, 0.0973, 0.0594, 0.2368, 0.2687, 0.0830, 0.1375, 0.0926], + device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0064, 0.0063, 0.0078, 0.0059, 0.0048, 0.0057, 0.0064], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-15 23:55:20,029 INFO [train.py:876] (3/4) Epoch 8, batch 3300, loss[loss=0.1554, simple_loss=0.1752, pruned_loss=0.0678, over 5569.00 frames. ], tot_loss[loss=0.1424, simple_loss=0.1599, pruned_loss=0.06243, over 1083680.66 frames. ], batch size: 25, lr: 1.01e-02, grad_scale: 16.0 +2022-11-15 23:55:22,381 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-15 23:55:32,990 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.566e+02 1.855e+02 2.366e+02 3.545e+02, threshold=3.710e+02, percent-clipped=0.0 +2022-11-15 23:55:58,026 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5135, 1.0352, 1.6606, 1.3750, 1.4528, 1.6560, 1.6987, 1.3592], + device='cuda:3'), covar=tensor([0.0040, 0.0119, 0.0062, 0.0050, 0.0059, 0.0081, 0.0034, 0.0042], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0026, 0.0023, 0.0022, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8173e-05, 2.0061e-05, 1.8934e-05, 2.6125e-05, 2.1920e-05, 2.1389e-05, + 2.4080e-05, 2.5830e-05], device='cuda:3') +2022-11-15 23:56:12,217 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6668, 2.3447, 3.3171, 3.0004, 3.3303, 2.4790, 3.1850, 3.6693], + device='cuda:3'), covar=tensor([0.0504, 0.1473, 0.0683, 0.1270, 0.0498, 0.1291, 0.1146, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0197, 0.0202, 0.0210, 0.0219, 0.0192, 0.0223, 0.0219], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-15 23:56:31,635 INFO [train.py:876] (3/4) Epoch 8, batch 3400, loss[loss=0.154, simple_loss=0.1682, pruned_loss=0.06994, over 5600.00 frames. ], tot_loss[loss=0.1379, simple_loss=0.157, pruned_loss=0.05943, over 1089108.99 frames. ], batch size: 18, lr: 1.01e-02, grad_scale: 16.0 +2022-11-15 23:56:43,983 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.624e+02 2.119e+02 2.818e+02 4.148e+02, threshold=4.237e+02, percent-clipped=5.0 +2022-11-15 23:56:58,769 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2022-11-15 23:56:59,825 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0208, 3.8575, 4.0045, 4.1100, 3.7603, 3.2568, 4.4965, 3.9543], + device='cuda:3'), covar=tensor([0.0455, 0.0826, 0.0392, 0.0938, 0.0509, 0.0522, 0.0711, 0.0617], + device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0099, 0.0085, 0.0108, 0.0079, 0.0070, 0.0134, 0.0089], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-15 23:57:26,429 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8025, 4.5990, 3.4905, 2.1781, 4.3522, 1.8889, 4.2778, 2.3776], + device='cuda:3'), covar=tensor([0.1403, 0.0125, 0.0493, 0.2035, 0.0163, 0.2014, 0.0157, 0.1716], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0107, 0.0117, 0.0116, 0.0104, 0.0129, 0.0101, 0.0117], + 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') +2022-11-15 23:57:44,047 INFO [train.py:876] (3/4) Epoch 8, batch 3500, loss[loss=0.08684, simple_loss=0.1271, pruned_loss=0.02327, over 5552.00 frames. ], tot_loss[loss=0.1375, simple_loss=0.1565, pruned_loss=0.0593, over 1084574.95 frames. ], batch size: 10, lr: 1.01e-02, grad_scale: 16.0 +2022-11-15 23:57:56,205 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.748e+02 2.123e+02 2.644e+02 4.958e+02, threshold=4.247e+02, percent-clipped=1.0 +2022-11-15 23:58:53,937 INFO [train.py:876] (3/4) Epoch 8, batch 3600, loss[loss=0.1325, simple_loss=0.1557, pruned_loss=0.05459, over 5627.00 frames. ], tot_loss[loss=0.1386, simple_loss=0.1569, pruned_loss=0.0602, over 1074318.64 frames. ], batch size: 29, lr: 1.01e-02, grad_scale: 16.0 +2022-11-15 23:59:05,647 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 1.765e+02 2.064e+02 2.542e+02 7.404e+02, threshold=4.127e+02, percent-clipped=4.0 +2022-11-15 23:59:45,437 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2022-11-16 00:00:01,686 INFO [train.py:876] (3/4) Epoch 8, batch 3700, loss[loss=0.161, simple_loss=0.1646, pruned_loss=0.07867, over 5129.00 frames. ], tot_loss[loss=0.1391, simple_loss=0.1578, pruned_loss=0.06022, over 1083061.24 frames. ], batch size: 91, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:00:14,193 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.627e+02 2.007e+02 2.375e+02 5.660e+02, threshold=4.014e+02, percent-clipped=3.0 +2022-11-16 00:00:34,508 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6262, 4.1284, 3.8520, 3.3743, 2.1082, 4.0789, 2.1909, 3.2534], + device='cuda:3'), covar=tensor([0.0447, 0.0143, 0.0181, 0.0418, 0.0589, 0.0147, 0.0551, 0.0152], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0155, 0.0163, 0.0184, 0.0177, 0.0165, 0.0175, 0.0160], + 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') +2022-11-16 00:00:42,266 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:01:09,319 INFO [train.py:876] (3/4) Epoch 8, batch 3800, loss[loss=0.1414, simple_loss=0.1687, pruned_loss=0.05705, over 5624.00 frames. ], tot_loss[loss=0.1374, simple_loss=0.1569, pruned_loss=0.05902, over 1092031.99 frames. ], batch size: 29, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:01:22,411 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.662e+02 2.074e+02 2.682e+02 3.562e+02, threshold=4.148e+02, percent-clipped=0.0 +2022-11-16 00:01:23,825 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:01:25,724 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6246, 1.9363, 1.8972, 1.7014, 1.8981, 1.8819, 0.9063, 1.9463], + device='cuda:3'), covar=tensor([0.0457, 0.0318, 0.0337, 0.0313, 0.0368, 0.0313, 0.2223, 0.0367], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0083, 0.0081, 0.0074, 0.0098, 0.0084, 0.0129, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:01:58,809 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2022-11-16 00:02:06,908 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5663, 2.3061, 2.8965, 1.6977, 1.7084, 3.1605, 2.2768, 2.0950], + device='cuda:3'), covar=tensor([0.0515, 0.0844, 0.0365, 0.2352, 0.2767, 0.1022, 0.2071, 0.1021], + device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0066, 0.0064, 0.0080, 0.0062, 0.0051, 0.0057, 0.0066], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-16 00:02:17,441 INFO [train.py:876] (3/4) Epoch 8, batch 3900, loss[loss=0.1206, simple_loss=0.1451, pruned_loss=0.04806, over 5589.00 frames. ], tot_loss[loss=0.1395, simple_loss=0.1584, pruned_loss=0.06031, over 1093106.82 frames. ], batch size: 22, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:02:29,731 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.833e+01 1.696e+02 2.085e+02 2.412e+02 7.560e+02, threshold=4.170e+02, percent-clipped=1.0 +2022-11-16 00:02:42,393 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1552, 0.8503, 1.0863, 0.7664, 1.0624, 1.3368, 0.7338, 1.0697], + device='cuda:3'), covar=tensor([0.0584, 0.0525, 0.0336, 0.0793, 0.1951, 0.0331, 0.2045, 0.0345], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6124e-05, 7.3392e-05, 5.6887e-05, 6.7910e-05, 6.1486e-05, 5.6108e-05, + 7.0097e-05, 5.7608e-05], device='cuda:3') +2022-11-16 00:02:56,167 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:03:25,451 INFO [train.py:876] (3/4) Epoch 8, batch 4000, loss[loss=0.1616, simple_loss=0.1569, pruned_loss=0.08321, over 4984.00 frames. ], tot_loss[loss=0.139, simple_loss=0.1582, pruned_loss=0.05994, over 1096865.77 frames. ], batch size: 109, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:03:32,720 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8413, 1.3892, 2.1501, 1.4713, 1.3496, 1.9267, 1.7246, 1.6255], + device='cuda:3'), covar=tensor([0.0035, 0.0082, 0.0025, 0.0052, 0.0094, 0.0099, 0.0033, 0.0029], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0026, 0.0023, 0.0022, 0.0024, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.8022e-05, 2.0107e-05, 1.8760e-05, 2.5900e-05, 2.1668e-05, 2.1726e-05, + 2.3731e-05, 2.4850e-05], device='cuda:3') +2022-11-16 00:03:37,028 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.640e+02 2.022e+02 2.606e+02 3.847e+02, threshold=4.045e+02, percent-clipped=0.0 +2022-11-16 00:03:37,223 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:04:39,882 INFO [train.py:876] (3/4) Epoch 8, batch 4100, loss[loss=0.1268, simple_loss=0.1446, pruned_loss=0.0545, over 5631.00 frames. ], tot_loss[loss=0.1395, simple_loss=0.1581, pruned_loss=0.06046, over 1095687.20 frames. ], batch size: 29, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:04:47,491 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 00:04:49,734 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:04:49,848 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7963, 2.1487, 3.3632, 2.8437, 3.6637, 2.4077, 3.3114, 3.8933], + device='cuda:3'), covar=tensor([0.0772, 0.2215, 0.1046, 0.2155, 0.0652, 0.1838, 0.1505, 0.0948], + device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0194, 0.0200, 0.0208, 0.0217, 0.0190, 0.0220, 0.0217], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:04:51,561 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.489e+01 1.632e+02 1.927e+02 2.505e+02 4.639e+02, threshold=3.854e+02, percent-clipped=4.0 +2022-11-16 00:05:47,254 INFO [train.py:876] (3/4) Epoch 8, batch 4200, loss[loss=0.1934, simple_loss=0.1982, pruned_loss=0.09428, over 5126.00 frames. ], tot_loss[loss=0.1419, simple_loss=0.1601, pruned_loss=0.06184, over 1092444.99 frames. ], batch size: 91, lr: 1.01e-02, grad_scale: 16.0 +2022-11-16 00:05:48,068 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7044, 1.1836, 1.7435, 1.2453, 1.2564, 1.1455, 1.2968, 1.4539], + device='cuda:3'), covar=tensor([0.1253, 0.0444, 0.0225, 0.0940, 0.1986, 0.1942, 0.0453, 0.1306], + device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0016, 0.0011, 0.0014, 0.0012, 0.0011, 0.0015, 0.0011], + device='cuda:3'), out_proj_covar=tensor([5.6992e-05, 7.4317e-05, 5.7636e-05, 6.7783e-05, 6.2109e-05, 5.6632e-05, + 7.0957e-05, 5.7774e-05], device='cuda:3') +2022-11-16 00:05:57,725 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2022-11-16 00:05:58,860 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5262, 3.8748, 3.7442, 3.4080, 2.0557, 3.8612, 2.1626, 3.0443], + device='cuda:3'), covar=tensor([0.0396, 0.0158, 0.0112, 0.0319, 0.0453, 0.0121, 0.0449, 0.0146], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0160, 0.0167, 0.0189, 0.0181, 0.0168, 0.0179, 0.0162], + 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') +2022-11-16 00:05:59,242 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.648e+02 1.989e+02 2.446e+02 4.173e+02, threshold=3.979e+02, percent-clipped=3.0 +2022-11-16 00:06:16,053 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:06:38,760 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2022-11-16 00:06:54,337 INFO [train.py:876] (3/4) Epoch 8, batch 4300, loss[loss=0.1233, simple_loss=0.153, pruned_loss=0.04675, over 5658.00 frames. ], tot_loss[loss=0.1403, simple_loss=0.1593, pruned_loss=0.06064, over 1088687.27 frames. ], batch size: 29, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:06:57,959 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:06:59,240 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:07:04,168 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 00:07:07,358 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.645e+01 1.651e+02 2.028e+02 2.598e+02 5.835e+02, threshold=4.056e+02, percent-clipped=3.0 +2022-11-16 00:07:32,215 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4499, 2.2488, 2.4134, 1.4210, 1.5243, 3.2033, 2.3114, 2.2573], + device='cuda:3'), covar=tensor([0.0837, 0.1021, 0.0587, 0.2713, 0.2919, 0.0845, 0.1330, 0.0793], + device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0066, 0.0066, 0.0082, 0.0062, 0.0050, 0.0058, 0.0065], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2022-11-16 00:07:40,472 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:07:51,169 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9638, 1.4178, 2.1555, 1.6064, 1.4562, 2.0595, 1.5470, 1.5322], + device='cuda:3'), covar=tensor([0.0028, 0.0087, 0.0024, 0.0041, 0.0059, 0.0038, 0.0031, 0.0029], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0021, 0.0027, 0.0024, 0.0023, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8795e-05, 2.0550e-05, 1.9302e-05, 2.6740e-05, 2.2330e-05, 2.2306e-05, + 2.4893e-05, 2.6396e-05], device='cuda:3') +2022-11-16 00:08:02,074 INFO [train.py:876] (3/4) Epoch 8, batch 4400, loss[loss=0.2589, simple_loss=0.2184, pruned_loss=0.1497, over 3033.00 frames. ], tot_loss[loss=0.1396, simple_loss=0.1588, pruned_loss=0.06022, over 1087175.74 frames. ], batch size: 284, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:08:12,094 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6199, 3.9058, 3.6707, 3.4257, 2.0355, 3.8485, 2.1302, 3.1303], + device='cuda:3'), covar=tensor([0.0362, 0.0108, 0.0146, 0.0299, 0.0465, 0.0109, 0.0448, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0154, 0.0163, 0.0183, 0.0175, 0.0162, 0.0174, 0.0157], + 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') +2022-11-16 00:08:12,963 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:08:14,768 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.701e+02 2.121e+02 2.893e+02 5.250e+02, threshold=4.241e+02, percent-clipped=3.0 +2022-11-16 00:08:43,196 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3090, 3.2661, 3.2717, 3.0355, 1.9322, 3.2549, 2.0540, 2.8949], + device='cuda:3'), covar=tensor([0.0296, 0.0123, 0.0135, 0.0232, 0.0364, 0.0111, 0.0356, 0.0128], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0154, 0.0163, 0.0183, 0.0176, 0.0163, 0.0174, 0.0158], + 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') +2022-11-16 00:08:44,939 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:09:06,113 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7035, 1.5995, 1.7310, 1.1815, 1.4973, 1.8513, 1.3168, 1.3254], + device='cuda:3'), covar=tensor([0.0027, 0.0048, 0.0057, 0.0051, 0.0041, 0.0038, 0.0033, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0021, 0.0027, 0.0024, 0.0023, 0.0025, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.8751e-05, 2.0504e-05, 1.9190e-05, 2.6924e-05, 2.2301e-05, 2.2031e-05, + 2.4856e-05, 2.6722e-05], device='cuda:3') +2022-11-16 00:09:10,910 INFO [train.py:876] (3/4) Epoch 8, batch 4500, loss[loss=0.1579, simple_loss=0.1657, pruned_loss=0.07503, over 5133.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.157, pruned_loss=0.05956, over 1087092.42 frames. ], batch size: 91, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:09:13,769 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-16 00:09:20,163 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1150, 1.2287, 1.3430, 0.7850, 0.9681, 1.2904, 0.9275, 0.7753], + device='cuda:3'), covar=tensor([0.0015, 0.0016, 0.0017, 0.0021, 0.0023, 0.0020, 0.0026, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0021, 0.0027, 0.0024, 0.0023, 0.0025, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.8890e-05, 2.0495e-05, 1.9191e-05, 2.7088e-05, 2.2322e-05, 2.2115e-05, + 2.4895e-05, 2.6870e-05], device='cuda:3') +2022-11-16 00:09:22,570 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.579e+01 1.575e+02 1.933e+02 2.382e+02 3.910e+02, threshold=3.866e+02, percent-clipped=0.0 +2022-11-16 00:10:18,063 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:10:18,620 INFO [train.py:876] (3/4) Epoch 8, batch 4600, loss[loss=0.1994, simple_loss=0.1899, pruned_loss=0.1045, over 5447.00 frames. ], tot_loss[loss=0.138, simple_loss=0.1573, pruned_loss=0.0593, over 1090295.37 frames. ], batch size: 58, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:10:27,251 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:10:30,361 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.826e+02 2.052e+02 2.615e+02 3.619e+02, threshold=4.103e+02, percent-clipped=0.0 +2022-11-16 00:10:35,977 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-16 00:10:59,633 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:11:00,331 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:11:25,532 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5417, 3.9270, 3.6811, 3.4868, 2.1012, 3.9157, 2.1154, 3.2759], + device='cuda:3'), covar=tensor([0.0443, 0.0375, 0.0258, 0.0369, 0.0556, 0.0166, 0.0572, 0.0142], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0156, 0.0167, 0.0185, 0.0180, 0.0165, 0.0177, 0.0160], + 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') +2022-11-16 00:11:26,709 INFO [train.py:876] (3/4) Epoch 8, batch 4700, loss[loss=0.1446, simple_loss=0.1629, pruned_loss=0.06317, over 5606.00 frames. ], tot_loss[loss=0.1375, simple_loss=0.1568, pruned_loss=0.05907, over 1090790.36 frames. ], batch size: 23, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:11:31,705 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-16 00:11:35,678 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 +2022-11-16 00:11:38,374 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.703e+02 2.017e+02 2.735e+02 4.468e+02, threshold=4.034e+02, percent-clipped=2.0 +2022-11-16 00:11:49,789 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7073, 2.4410, 3.0438, 3.7903, 3.8117, 3.0547, 2.6962, 3.8711], + device='cuda:3'), covar=tensor([0.0460, 0.3249, 0.2745, 0.3383, 0.1157, 0.2792, 0.2234, 0.0778], + device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0210, 0.0207, 0.0323, 0.0227, 0.0218, 0.0198, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:12:31,022 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4978, 3.9973, 2.9413, 1.7881, 3.7253, 1.4326, 3.7059, 2.0434], + device='cuda:3'), covar=tensor([0.1399, 0.0155, 0.0896, 0.2253, 0.0216, 0.2252, 0.0182, 0.1875], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0110, 0.0119, 0.0120, 0.0108, 0.0131, 0.0102, 0.0118], + 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') +2022-11-16 00:12:33,871 INFO [train.py:876] (3/4) Epoch 8, batch 4800, loss[loss=0.1243, simple_loss=0.1589, pruned_loss=0.04482, over 5715.00 frames. ], tot_loss[loss=0.1371, simple_loss=0.1571, pruned_loss=0.0586, over 1092866.50 frames. ], batch size: 17, lr: 1.00e-02, grad_scale: 16.0 +2022-11-16 00:12:46,303 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.529e+02 1.884e+02 2.231e+02 4.028e+02, threshold=3.767e+02, percent-clipped=0.0 +2022-11-16 00:13:02,520 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:13:40,606 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:13:41,117 INFO [train.py:876] (3/4) Epoch 8, batch 4900, loss[loss=0.1305, simple_loss=0.1567, pruned_loss=0.05218, over 5593.00 frames. ], tot_loss[loss=0.1374, simple_loss=0.157, pruned_loss=0.05891, over 1090539.10 frames. ], batch size: 24, lr: 9.99e-03, grad_scale: 16.0 +2022-11-16 00:13:43,892 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:13:53,468 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.007e+01 1.783e+02 2.085e+02 2.465e+02 4.573e+02, threshold=4.169e+02, percent-clipped=4.0 +2022-11-16 00:13:54,218 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 00:14:02,792 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4910, 4.0765, 3.1229, 1.9010, 3.8423, 1.4223, 3.6242, 2.0681], + device='cuda:3'), covar=tensor([0.1375, 0.0130, 0.0631, 0.2046, 0.0178, 0.2225, 0.0218, 0.1745], + device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0107, 0.0115, 0.0116, 0.0105, 0.0128, 0.0099, 0.0115], + 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') +2022-11-16 00:14:10,707 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:14:13,175 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:14:23,015 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:14:49,101 INFO [train.py:876] (3/4) Epoch 8, batch 5000, loss[loss=0.1692, simple_loss=0.1646, pruned_loss=0.08695, over 4151.00 frames. ], tot_loss[loss=0.1369, simple_loss=0.1569, pruned_loss=0.05849, over 1086645.63 frames. ], batch size: 181, lr: 9.98e-03, grad_scale: 16.0 +2022-11-16 00:14:51,915 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:14:53,212 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8809, 2.4241, 2.8351, 3.7232, 3.8542, 2.9546, 2.2518, 3.9465], + device='cuda:3'), covar=tensor([0.0993, 0.3618, 0.2570, 0.4742, 0.1197, 0.3459, 0.2829, 0.0578], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0211, 0.0203, 0.0323, 0.0229, 0.0219, 0.0199, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:14:55,678 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:15:00,898 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.639e+01 1.479e+02 1.807e+02 2.290e+02 3.768e+02, threshold=3.615e+02, percent-clipped=0.0 +2022-11-16 00:15:02,143 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6267, 2.0278, 3.1889, 2.7623, 3.4050, 2.2490, 2.8435, 3.5955], + device='cuda:3'), covar=tensor([0.0682, 0.1866, 0.0819, 0.1825, 0.0681, 0.1914, 0.1354, 0.0903], + device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0194, 0.0200, 0.0209, 0.0219, 0.0192, 0.0225, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:15:03,757 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2022-11-16 00:15:08,163 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-16 00:15:22,927 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8900, 4.4080, 4.6184, 4.4449, 4.9671, 4.8369, 4.3224, 4.8878], + device='cuda:3'), covar=tensor([0.0398, 0.0328, 0.0532, 0.0305, 0.0322, 0.0148, 0.0258, 0.0267], + device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0134, 0.0102, 0.0132, 0.0149, 0.0088, 0.0112, 0.0134], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 00:15:57,402 INFO [train.py:876] (3/4) Epoch 8, batch 5100, loss[loss=0.1068, simple_loss=0.1288, pruned_loss=0.04238, over 5300.00 frames. ], tot_loss[loss=0.1355, simple_loss=0.1558, pruned_loss=0.05758, over 1085903.58 frames. ], batch size: 9, lr: 9.97e-03, grad_scale: 16.0 +2022-11-16 00:15:57,525 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0485, 0.9293, 2.0859, 1.5632, 1.5192, 1.8918, 1.6647, 1.4788], + device='cuda:3'), covar=tensor([0.0033, 0.0129, 0.0019, 0.0042, 0.0075, 0.0044, 0.0031, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0026, 0.0023, 0.0022, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8250e-05, 1.9561e-05, 1.9300e-05, 2.6275e-05, 2.1773e-05, 2.1804e-05, + 2.4576e-05, 2.5829e-05], device='cuda:3') +2022-11-16 00:16:03,173 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2022-11-16 00:16:07,496 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-16 00:16:09,574 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.582e+02 1.987e+02 2.357e+02 3.737e+02, threshold=3.975e+02, percent-clipped=1.0 +2022-11-16 00:17:05,912 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:17:06,565 INFO [train.py:876] (3/4) Epoch 8, batch 5200, loss[loss=0.1777, simple_loss=0.1838, pruned_loss=0.08578, over 5343.00 frames. ], tot_loss[loss=0.1345, simple_loss=0.1551, pruned_loss=0.05697, over 1083131.30 frames. ], batch size: 79, lr: 9.96e-03, grad_scale: 32.0 +2022-11-16 00:17:18,355 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.676e+02 2.046e+02 2.590e+02 6.107e+02, threshold=4.093e+02, percent-clipped=5.0 +2022-11-16 00:17:24,831 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 +2022-11-16 00:17:32,923 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:18:02,485 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2022-11-16 00:18:13,847 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:18:14,419 INFO [train.py:876] (3/4) Epoch 8, batch 5300, loss[loss=0.1233, simple_loss=0.1631, pruned_loss=0.04174, over 5767.00 frames. ], tot_loss[loss=0.1374, simple_loss=0.1573, pruned_loss=0.05872, over 1088692.41 frames. ], batch size: 16, lr: 9.95e-03, grad_scale: 16.0 +2022-11-16 00:18:15,303 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:18:27,817 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.491e+02 1.984e+02 2.507e+02 5.516e+02, threshold=3.968e+02, percent-clipped=2.0 +2022-11-16 00:19:03,288 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2022-11-16 00:19:10,284 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3387, 2.3493, 2.0027, 2.3292, 2.3921, 2.2170, 2.1007, 2.2093], + device='cuda:3'), covar=tensor([0.0453, 0.0862, 0.1895, 0.0678, 0.0702, 0.0584, 0.1117, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0171, 0.0267, 0.0164, 0.0212, 0.0167, 0.0177, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:19:22,482 INFO [train.py:876] (3/4) Epoch 8, batch 5400, loss[loss=0.1755, simple_loss=0.177, pruned_loss=0.08704, over 5054.00 frames. ], tot_loss[loss=0.139, simple_loss=0.1587, pruned_loss=0.05961, over 1083550.15 frames. ], batch size: 110, lr: 9.94e-03, grad_scale: 16.0 +2022-11-16 00:19:26,546 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7069, 2.2218, 2.6376, 1.5378, 1.1636, 3.0231, 2.2417, 2.2482], + device='cuda:3'), covar=tensor([0.0791, 0.1127, 0.0480, 0.2989, 0.3309, 0.0569, 0.2286, 0.0840], + device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0067, 0.0065, 0.0082, 0.0061, 0.0051, 0.0059, 0.0067], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:19:35,868 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.666e+02 2.112e+02 2.749e+02 4.650e+02, threshold=4.223e+02, percent-clipped=6.0 +2022-11-16 00:19:52,085 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1657, 4.4905, 4.1370, 3.8843, 2.4519, 4.6606, 2.6413, 4.0542], + device='cuda:3'), covar=tensor([0.0391, 0.0204, 0.0235, 0.0340, 0.0616, 0.0127, 0.0476, 0.0105], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0155, 0.0166, 0.0185, 0.0182, 0.0166, 0.0178, 0.0161], + 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') +2022-11-16 00:20:29,847 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:20:30,393 INFO [train.py:876] (3/4) Epoch 8, batch 5500, loss[loss=0.1353, simple_loss=0.1518, pruned_loss=0.05939, over 5524.00 frames. ], tot_loss[loss=0.1394, simple_loss=0.1585, pruned_loss=0.06015, over 1081913.58 frames. ], batch size: 12, lr: 9.94e-03, grad_scale: 16.0 +2022-11-16 00:20:42,594 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.705e+02 2.205e+02 2.519e+02 5.507e+02, threshold=4.409e+02, percent-clipped=4.0 +2022-11-16 00:21:02,112 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:21:35,573 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:21:37,732 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:21:38,226 INFO [train.py:876] (3/4) Epoch 8, batch 5600, loss[loss=0.1759, simple_loss=0.1632, pruned_loss=0.0943, over 4234.00 frames. ], tot_loss[loss=0.1396, simple_loss=0.1588, pruned_loss=0.06023, over 1084808.64 frames. ], batch size: 181, lr: 9.93e-03, grad_scale: 16.0 +2022-11-16 00:21:50,130 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:21:50,546 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.658e+02 2.017e+02 2.598e+02 4.706e+02, threshold=4.034e+02, percent-clipped=2.0 +2022-11-16 00:22:09,938 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:22:24,401 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7839, 3.4259, 3.3208, 1.9510, 3.1668, 3.6619, 3.6471, 4.1396], + device='cuda:3'), covar=tensor([0.1924, 0.1420, 0.1039, 0.2977, 0.0417, 0.0779, 0.0373, 0.0515], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0187, 0.0161, 0.0191, 0.0172, 0.0183, 0.0157, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 00:22:29,154 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0559, 2.2896, 2.4338, 3.2997, 3.1185, 2.5055, 1.9607, 3.3971], + device='cuda:3'), covar=tensor([0.0814, 0.2813, 0.2395, 0.1953, 0.1338, 0.3018, 0.2212, 0.0819], + device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0206, 0.0198, 0.0322, 0.0222, 0.0212, 0.0196, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:22:31,089 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:22:46,105 INFO [train.py:876] (3/4) Epoch 8, batch 5700, loss[loss=0.105, simple_loss=0.1324, pruned_loss=0.03884, over 5733.00 frames. ], tot_loss[loss=0.138, simple_loss=0.1575, pruned_loss=0.0592, over 1086545.39 frames. ], batch size: 13, lr: 9.92e-03, grad_scale: 16.0 +2022-11-16 00:22:52,207 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 +2022-11-16 00:22:58,516 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.608e+02 1.896e+02 2.176e+02 4.174e+02, threshold=3.791e+02, percent-clipped=1.0 +2022-11-16 00:23:52,798 INFO [train.py:876] (3/4) Epoch 8, batch 5800, loss[loss=0.1339, simple_loss=0.1655, pruned_loss=0.05118, over 5574.00 frames. ], tot_loss[loss=0.1387, simple_loss=0.1577, pruned_loss=0.0598, over 1078483.34 frames. ], batch size: 25, lr: 9.91e-03, grad_scale: 16.0 +2022-11-16 00:23:55,902 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3880, 1.3124, 1.4983, 1.0932, 1.2321, 1.3071, 1.0056, 0.9016], + device='cuda:3'), covar=tensor([0.0020, 0.0031, 0.0030, 0.0039, 0.0034, 0.0030, 0.0034, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0027, 0.0024, 0.0023, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8070e-05, 1.9651e-05, 1.9500e-05, 2.6613e-05, 2.2002e-05, 2.1817e-05, + 2.4430e-05, 2.5477e-05], device='cuda:3') +2022-11-16 00:24:05,871 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.678e+02 1.927e+02 2.406e+02 3.937e+02, threshold=3.853e+02, percent-clipped=1.0 +2022-11-16 00:24:20,507 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 +2022-11-16 00:24:23,680 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 +2022-11-16 00:24:43,677 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2022-11-16 00:24:47,986 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9960, 2.3383, 3.4956, 2.9009, 3.7744, 2.4710, 3.3535, 3.8588], + device='cuda:3'), covar=tensor([0.0606, 0.1604, 0.0911, 0.1775, 0.0476, 0.1630, 0.1238, 0.0875], + device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0189, 0.0203, 0.0209, 0.0218, 0.0188, 0.0220, 0.0219], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:24:53,852 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9081, 1.2726, 1.5750, 1.5663, 1.1721, 1.6064, 1.5290, 1.2440], + device='cuda:3'), covar=tensor([0.0025, 0.0118, 0.0067, 0.0048, 0.0072, 0.0101, 0.0025, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0026, 0.0023, 0.0022, 0.0025, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.7958e-05, 1.9579e-05, 1.9425e-05, 2.6255e-05, 2.1936e-05, 2.1428e-05, + 2.3991e-05, 2.5351e-05], device='cuda:3') +2022-11-16 00:24:54,498 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4995, 4.3613, 3.4188, 1.9119, 4.0852, 1.4802, 3.9300, 2.2837], + device='cuda:3'), covar=tensor([0.1315, 0.0149, 0.0472, 0.1920, 0.0171, 0.2019, 0.0204, 0.1570], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0107, 0.0114, 0.0116, 0.0105, 0.0127, 0.0098, 0.0115], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:24:57,440 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:24:57,850 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.66 vs. limit=5.0 +2022-11-16 00:24:58,799 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:25:00,324 INFO [train.py:876] (3/4) Epoch 8, batch 5900, loss[loss=0.1061, simple_loss=0.1371, pruned_loss=0.03756, over 5425.00 frames. ], tot_loss[loss=0.1365, simple_loss=0.1562, pruned_loss=0.05844, over 1076631.36 frames. ], batch size: 11, lr: 9.90e-03, grad_scale: 16.0 +2022-11-16 00:25:13,652 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.513e+01 1.636e+02 2.086e+02 2.680e+02 4.178e+02, threshold=4.173e+02, percent-clipped=3.0 +2022-11-16 00:25:29,997 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:25:40,354 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:25:49,455 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1073, 1.8853, 2.3714, 1.6623, 0.8685, 2.6239, 2.2614, 1.8857], + device='cuda:3'), covar=tensor([0.0855, 0.1493, 0.0747, 0.2575, 0.3082, 0.2128, 0.1136, 0.1238], + device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0069, 0.0068, 0.0084, 0.0062, 0.0051, 0.0059, 0.0069], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:25:50,690 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 00:25:52,616 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4972, 3.3758, 3.4001, 3.4975, 3.1598, 3.0518, 3.9632, 3.2005], + device='cuda:3'), covar=tensor([0.0506, 0.0950, 0.0517, 0.1131, 0.0746, 0.0500, 0.0790, 0.1072], + device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0099, 0.0086, 0.0109, 0.0081, 0.0073, 0.0135, 0.0093], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:26:07,758 INFO [train.py:876] (3/4) Epoch 8, batch 6000, loss[loss=0.2406, simple_loss=0.2069, pruned_loss=0.1372, over 3051.00 frames. ], tot_loss[loss=0.1384, simple_loss=0.1577, pruned_loss=0.05952, over 1078841.28 frames. ], batch size: 284, lr: 9.89e-03, grad_scale: 16.0 +2022-11-16 00:26:07,759 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 00:26:25,619 INFO [train.py:908] (3/4) Epoch 8, validation: loss=0.1622, simple_loss=0.1823, pruned_loss=0.07105, over 1530663.00 frames. +2022-11-16 00:26:25,620 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 00:26:27,745 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:26:38,697 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.650e+01 1.622e+02 2.023e+02 2.494e+02 5.348e+02, threshold=4.047e+02, percent-clipped=2.0 +2022-11-16 00:26:49,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 +2022-11-16 00:26:51,959 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2022-11-16 00:27:02,250 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9111, 0.9974, 1.6046, 1.4367, 1.2470, 1.5977, 1.4195, 1.3530], + device='cuda:3'), covar=tensor([0.0022, 0.0091, 0.0043, 0.0046, 0.0078, 0.0076, 0.0029, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0021, 0.0026, 0.0023, 0.0022, 0.0024, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.7598e-05, 1.9538e-05, 1.9021e-05, 2.5936e-05, 2.1704e-05, 2.1333e-05, + 2.3658e-05, 2.5020e-05], device='cuda:3') +2022-11-16 00:27:09,203 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:27:25,017 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:27:26,250 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:27:32,824 INFO [train.py:876] (3/4) Epoch 8, batch 6100, loss[loss=0.1815, simple_loss=0.1832, pruned_loss=0.08993, over 5489.00 frames. ], tot_loss[loss=0.1383, simple_loss=0.1576, pruned_loss=0.05952, over 1085172.48 frames. ], batch size: 49, lr: 9.88e-03, grad_scale: 16.0 +2022-11-16 00:27:45,773 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.054e+01 1.623e+02 1.859e+02 2.466e+02 4.899e+02, threshold=3.719e+02, percent-clipped=4.0 +2022-11-16 00:27:58,088 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0359, 1.9473, 2.6826, 2.4382, 2.4630, 1.9905, 2.5810, 2.9935], + device='cuda:3'), covar=tensor([0.0539, 0.1206, 0.0713, 0.1081, 0.0693, 0.1113, 0.0878, 0.0741], + device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0190, 0.0208, 0.0212, 0.0220, 0.0191, 0.0223, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:28:07,747 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:28:09,146 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:28:39,198 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8656, 3.5482, 2.2837, 3.1588, 2.6106, 2.4419, 1.9419, 2.7805], + device='cuda:3'), covar=tensor([0.1611, 0.0220, 0.1121, 0.0392, 0.1064, 0.1052, 0.1941, 0.0424], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0138, 0.0166, 0.0142, 0.0176, 0.0178, 0.0174, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 00:28:43,871 INFO [train.py:876] (3/4) Epoch 8, batch 6200, loss[loss=0.09459, simple_loss=0.1261, pruned_loss=0.03155, over 5233.00 frames. ], tot_loss[loss=0.1384, simple_loss=0.1577, pruned_loss=0.05956, over 1083656.11 frames. ], batch size: 8, lr: 9.88e-03, grad_scale: 16.0 +2022-11-16 00:28:56,821 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.699e+02 2.012e+02 2.315e+02 3.529e+02, threshold=4.025e+02, percent-clipped=0.0 +2022-11-16 00:28:59,791 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0666, 2.7775, 2.9212, 2.6742, 1.6720, 2.8126, 1.7747, 2.3213], + device='cuda:3'), covar=tensor([0.0347, 0.0126, 0.0126, 0.0280, 0.0451, 0.0160, 0.0418, 0.0167], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0154, 0.0164, 0.0184, 0.0178, 0.0162, 0.0174, 0.0160], + 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') +2022-11-16 00:29:06,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5971, 1.1451, 1.6840, 1.2413, 1.8591, 1.6681, 1.0952, 1.4695], + device='cuda:3'), covar=tensor([0.1626, 0.0813, 0.0300, 0.0760, 0.0497, 0.0887, 0.0482, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0017, 0.0013, 0.0015, 0.0013, 0.0011, 0.0016, 0.0012], + device='cuda:3'), out_proj_covar=tensor([6.1102e-05, 8.1124e-05, 6.4194e-05, 7.4066e-05, 6.6552e-05, 6.0613e-05, + 7.5573e-05, 6.0975e-05], device='cuda:3') +2022-11-16 00:29:22,007 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:29:22,188 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2022-11-16 00:29:28,489 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 00:29:35,666 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:29:40,817 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:29:41,855 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2357, 4.1552, 2.7109, 3.9012, 3.2242, 2.7389, 2.2208, 3.3768], + device='cuda:3'), covar=tensor([0.1535, 0.0230, 0.1195, 0.0436, 0.0719, 0.1149, 0.2024, 0.0407], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0138, 0.0164, 0.0141, 0.0175, 0.0175, 0.0172, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 00:29:45,272 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0851, 4.3278, 3.9232, 4.2872, 4.2986, 4.0185, 1.3382, 4.3496], + device='cuda:3'), covar=tensor([0.0295, 0.0338, 0.0388, 0.0242, 0.0260, 0.0399, 0.3608, 0.0290], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0086, 0.0085, 0.0075, 0.0101, 0.0086, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:29:49,419 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:29:55,083 INFO [train.py:876] (3/4) Epoch 8, batch 6300, loss[loss=0.1588, simple_loss=0.1733, pruned_loss=0.07211, over 5559.00 frames. ], tot_loss[loss=0.1387, simple_loss=0.1578, pruned_loss=0.05978, over 1081134.00 frames. ], batch size: 21, lr: 9.87e-03, grad_scale: 16.0 +2022-11-16 00:30:08,025 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-16 00:30:08,131 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 1.749e+02 2.073e+02 2.619e+02 6.715e+02, threshold=4.147e+02, percent-clipped=6.0 +2022-11-16 00:30:10,296 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 00:30:23,467 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:30:25,503 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:30:31,766 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0510, 1.3736, 1.1950, 0.7782, 0.9836, 1.2963, 0.8712, 1.4231], + device='cuda:3'), covar=tensor([0.0041, 0.0028, 0.0041, 0.0041, 0.0041, 0.0029, 0.0049, 0.0059], + device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0041, 0.0044, 0.0043, 0.0042, 0.0038, 0.0043, 0.0037], + device='cuda:3'), out_proj_covar=tensor([4.1620e-05, 3.6811e-05, 3.9598e-05, 3.8601e-05, 3.6987e-05, 3.2832e-05, + 3.9808e-05, 3.2696e-05], device='cuda:3') +2022-11-16 00:30:33,127 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 00:30:37,152 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:30:50,245 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3634, 3.5163, 3.3856, 3.1847, 3.4813, 3.1644, 1.3180, 3.6276], + device='cuda:3'), covar=tensor([0.0334, 0.0353, 0.0411, 0.0376, 0.0405, 0.0523, 0.3381, 0.0336], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0085, 0.0084, 0.0074, 0.0100, 0.0085, 0.0128, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:31:06,477 INFO [train.py:876] (3/4) Epoch 8, batch 6400, loss[loss=0.08622, simple_loss=0.1212, pruned_loss=0.02561, over 5500.00 frames. ], tot_loss[loss=0.1375, simple_loss=0.1566, pruned_loss=0.05916, over 1082930.72 frames. ], batch size: 12, lr: 9.86e-03, grad_scale: 16.0 +2022-11-16 00:31:06,656 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:31:19,419 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.947e+01 1.601e+02 2.011e+02 2.621e+02 5.168e+02, threshold=4.022e+02, percent-clipped=2.0 +2022-11-16 00:31:23,504 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2022-11-16 00:31:36,916 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:31:38,745 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:31:41,279 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2022-11-16 00:32:08,468 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2337, 4.2539, 4.0171, 4.0897, 4.1058, 3.8846, 1.7659, 4.2551], + device='cuda:3'), covar=tensor([0.0260, 0.0405, 0.0315, 0.0297, 0.0420, 0.0494, 0.3115, 0.0320], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0084, 0.0083, 0.0073, 0.0099, 0.0085, 0.0127, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:32:11,879 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:32:17,278 INFO [train.py:876] (3/4) Epoch 8, batch 6500, loss[loss=0.1244, simple_loss=0.1463, pruned_loss=0.05127, over 5658.00 frames. ], tot_loss[loss=0.1349, simple_loss=0.1552, pruned_loss=0.05733, over 1081811.02 frames. ], batch size: 29, lr: 9.85e-03, grad_scale: 16.0 +2022-11-16 00:32:30,736 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.615e+02 1.950e+02 2.299e+02 5.072e+02, threshold=3.899e+02, percent-clipped=1.0 +2022-11-16 00:32:39,203 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2753, 2.1147, 2.8629, 2.5074, 2.7839, 2.0701, 2.7719, 3.2291], + device='cuda:3'), covar=tensor([0.0686, 0.1474, 0.0789, 0.1458, 0.1106, 0.1485, 0.1042, 0.0793], + device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0191, 0.0202, 0.0208, 0.0218, 0.0190, 0.0219, 0.0218], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:32:41,516 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2022-11-16 00:32:55,157 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:32:55,215 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:32:57,589 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:33:20,311 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:33:26,884 INFO [train.py:876] (3/4) Epoch 8, batch 6600, loss[loss=0.1208, simple_loss=0.1529, pruned_loss=0.04437, over 5495.00 frames. ], tot_loss[loss=0.1367, simple_loss=0.1563, pruned_loss=0.0585, over 1082181.97 frames. ], batch size: 17, lr: 9.84e-03, grad_scale: 16.0 +2022-11-16 00:33:28,257 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:33:39,760 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:33:40,181 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.548e+02 1.936e+02 2.529e+02 5.371e+02, threshold=3.872e+02, percent-clipped=1.0 +2022-11-16 00:33:45,223 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-16 00:33:52,096 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:33:59,269 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:34:01,291 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:34:06,417 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:34:16,539 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 +2022-11-16 00:34:30,738 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:34:33,916 INFO [train.py:876] (3/4) Epoch 8, batch 6700, loss[loss=0.1621, simple_loss=0.176, pruned_loss=0.07406, over 5592.00 frames. ], tot_loss[loss=0.1364, simple_loss=0.156, pruned_loss=0.05844, over 1076275.62 frames. ], batch size: 43, lr: 9.83e-03, grad_scale: 16.0 +2022-11-16 00:34:38,597 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:34:46,346 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.668e+02 2.078e+02 2.572e+02 7.235e+02, threshold=4.155e+02, percent-clipped=5.0 +2022-11-16 00:34:51,089 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2022-11-16 00:35:04,280 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:35:05,564 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:35:36,533 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:35:37,907 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:35:41,755 INFO [train.py:876] (3/4) Epoch 8, batch 6800, loss[loss=0.1177, simple_loss=0.1455, pruned_loss=0.04493, over 5745.00 frames. ], tot_loss[loss=0.1379, simple_loss=0.1568, pruned_loss=0.05948, over 1080860.12 frames. ], batch size: 14, lr: 9.82e-03, grad_scale: 16.0 +2022-11-16 00:35:53,963 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.669e+02 2.102e+02 2.619e+02 4.252e+02, threshold=4.204e+02, percent-clipped=2.0 +2022-11-16 00:35:56,888 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.09 vs. limit=5.0 +2022-11-16 00:36:00,322 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:36:14,622 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:36:24,030 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2300, 2.3306, 2.4447, 3.4283, 3.4119, 2.6538, 2.2053, 3.4240], + device='cuda:3'), covar=tensor([0.0896, 0.2646, 0.2050, 0.1775, 0.1124, 0.2295, 0.1923, 0.0526], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0206, 0.0195, 0.0323, 0.0224, 0.0213, 0.0192, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:36:29,294 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9014, 2.0044, 2.1825, 1.4598, 0.8784, 2.8719, 2.3059, 1.9551], + device='cuda:3'), covar=tensor([0.1240, 0.1244, 0.0744, 0.2608, 0.4387, 0.0701, 0.1021, 0.1253], + device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0067, 0.0067, 0.0080, 0.0060, 0.0051, 0.0058, 0.0066], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:36:32,310 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2022-11-16 00:36:41,876 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8211, 1.4074, 1.5534, 1.3674, 2.1472, 1.5530, 1.2757, 1.4626], + device='cuda:3'), covar=tensor([0.1843, 0.0454, 0.0548, 0.1005, 0.0872, 0.1584, 0.1213, 0.0681], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0017, 0.0012, 0.0015, 0.0013, 0.0011, 0.0016, 0.0011], + device='cuda:3'), out_proj_covar=tensor([6.0252e-05, 7.8945e-05, 6.1126e-05, 7.3277e-05, 6.5588e-05, 5.8999e-05, + 7.3850e-05, 5.8719e-05], device='cuda:3') +2022-11-16 00:36:41,890 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:36:49,274 INFO [train.py:876] (3/4) Epoch 8, batch 6900, loss[loss=0.1885, simple_loss=0.1978, pruned_loss=0.08954, over 5577.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.1567, pruned_loss=0.05979, over 1084614.71 frames. ], batch size: 43, lr: 9.82e-03, grad_scale: 16.0 +2022-11-16 00:36:57,822 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:01,645 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.691e+02 2.166e+02 2.756e+02 5.042e+02, threshold=4.332e+02, percent-clipped=3.0 +2022-11-16 00:37:13,691 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:20,212 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:21,630 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 00:37:45,649 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:53,399 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:54,292 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:37:56,852 INFO [train.py:876] (3/4) Epoch 8, batch 7000, loss[loss=0.1307, simple_loss=0.1501, pruned_loss=0.05567, over 5777.00 frames. ], tot_loss[loss=0.1365, simple_loss=0.1556, pruned_loss=0.05865, over 1088319.98 frames. ], batch size: 21, lr: 9.81e-03, grad_scale: 16.0 +2022-11-16 00:38:09,168 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.116e+02 1.725e+02 2.099e+02 2.587e+02 4.633e+02, threshold=4.198e+02, percent-clipped=1.0 +2022-11-16 00:38:23,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8957, 5.3719, 3.4287, 4.9894, 3.8809, 3.6898, 3.4787, 4.6722], + device='cuda:3'), covar=tensor([0.1284, 0.0144, 0.0804, 0.0267, 0.0435, 0.0693, 0.1247, 0.0214], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0140, 0.0166, 0.0143, 0.0176, 0.0177, 0.0173, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 00:38:25,214 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:38:45,854 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8261, 4.3811, 4.6892, 4.4014, 4.8858, 4.7290, 4.3764, 4.9065], + device='cuda:3'), covar=tensor([0.0357, 0.0301, 0.0440, 0.0306, 0.0372, 0.0189, 0.0242, 0.0253], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0135, 0.0102, 0.0133, 0.0152, 0.0089, 0.0114, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 00:39:03,878 INFO [train.py:876] (3/4) Epoch 8, batch 7100, loss[loss=0.1206, simple_loss=0.1521, pruned_loss=0.04454, over 5565.00 frames. ], tot_loss[loss=0.1378, simple_loss=0.1564, pruned_loss=0.05958, over 1079012.06 frames. ], batch size: 15, lr: 9.80e-03, grad_scale: 16.0 +2022-11-16 00:39:16,950 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.751e+02 2.181e+02 2.576e+02 4.312e+02, threshold=4.361e+02, percent-clipped=1.0 +2022-11-16 00:39:36,879 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:39:58,527 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2022-11-16 00:40:00,893 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:04,832 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4807, 2.1534, 2.3631, 1.5258, 1.1672, 3.1206, 2.5681, 2.2642], + device='cuda:3'), covar=tensor([0.0700, 0.0796, 0.0870, 0.2364, 0.1935, 0.1518, 0.0837, 0.0821], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0069, 0.0068, 0.0082, 0.0060, 0.0052, 0.0059, 0.0067], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:40:09,399 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:11,333 INFO [train.py:876] (3/4) Epoch 8, batch 7200, loss[loss=0.1796, simple_loss=0.1707, pruned_loss=0.09421, over 4155.00 frames. ], tot_loss[loss=0.1337, simple_loss=0.1536, pruned_loss=0.05685, over 1078073.66 frames. ], batch size: 181, lr: 9.79e-03, grad_scale: 16.0 +2022-11-16 00:40:17,034 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0168, 1.4053, 1.1702, 0.7783, 1.1464, 0.8234, 0.8034, 1.2418], + device='cuda:3'), covar=tensor([0.0038, 0.0030, 0.0045, 0.0048, 0.0041, 0.0035, 0.0056, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0042, 0.0044, 0.0044, 0.0043, 0.0039, 0.0043, 0.0038], + device='cuda:3'), out_proj_covar=tensor([4.2803e-05, 3.8112e-05, 3.9482e-05, 4.0213e-05, 3.7673e-05, 3.3568e-05, + 3.9976e-05, 3.3715e-05], device='cuda:3') +2022-11-16 00:40:20,270 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:24,025 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.930e+01 1.590e+02 1.880e+02 2.458e+02 4.390e+02, threshold=3.761e+02, percent-clipped=1.0 +2022-11-16 00:40:36,107 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:38,011 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:42,416 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:51,880 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:40:56,357 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7233, 0.6542, 0.5813, 0.4093, 0.7310, 0.7327, 0.3399, 0.7107], + device='cuda:3'), covar=tensor([0.0177, 0.0296, 0.0211, 0.0215, 0.0198, 0.0217, 0.0500, 0.0204], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0017, 0.0012, 0.0015, 0.0013, 0.0011, 0.0016, 0.0012], + device='cuda:3'), out_proj_covar=tensor([6.1491e-05, 8.1281e-05, 6.3012e-05, 7.4236e-05, 6.6306e-05, 6.0218e-05, + 7.5845e-05, 6.0162e-05], device='cuda:3') +2022-11-16 00:41:42,120 INFO [train.py:876] (3/4) Epoch 9, batch 0, loss[loss=0.185, simple_loss=0.1817, pruned_loss=0.09416, over 5428.00 frames. ], tot_loss[loss=0.185, simple_loss=0.1817, pruned_loss=0.09416, over 5428.00 frames. ], batch size: 58, lr: 9.26e-03, grad_scale: 16.0 +2022-11-16 00:41:42,120 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 00:41:49,352 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8622, 3.9943, 3.4423, 3.9788, 3.9546, 3.4566, 3.7709, 3.6357], + device='cuda:3'), covar=tensor([0.0195, 0.0432, 0.1337, 0.0376, 0.0545, 0.0439, 0.0389, 0.0432], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0165, 0.0260, 0.0159, 0.0207, 0.0164, 0.0174, 0.0162], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 00:41:58,752 INFO [train.py:908] (3/4) Epoch 9, validation: loss=0.1631, simple_loss=0.1836, pruned_loss=0.0713, over 1530663.00 frames. +2022-11-16 00:41:58,752 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 00:42:12,829 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 +2022-11-16 00:42:13,715 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:42:16,528 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:42:18,487 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:42:31,154 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.772e+02 2.208e+02 2.571e+02 4.464e+02, threshold=4.417e+02, percent-clipped=3.0 +2022-11-16 00:43:06,380 INFO [train.py:876] (3/4) Epoch 9, batch 100, loss[loss=0.1501, simple_loss=0.1805, pruned_loss=0.05992, over 5590.00 frames. ], tot_loss[loss=0.1339, simple_loss=0.1551, pruned_loss=0.05636, over 437404.30 frames. ], batch size: 30, lr: 9.26e-03, grad_scale: 16.0 +2022-11-16 00:43:08,622 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2022-11-16 00:43:25,414 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2263, 3.1621, 2.4555, 1.6212, 2.9947, 1.3880, 2.9940, 1.7599], + device='cuda:3'), covar=tensor([0.1183, 0.0213, 0.0897, 0.2059, 0.0296, 0.1968, 0.0312, 0.1537], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0109, 0.0118, 0.0119, 0.0110, 0.0130, 0.0100, 0.0119], + 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') +2022-11-16 00:43:38,872 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.840e+01 1.567e+02 1.869e+02 2.319e+02 4.319e+02, threshold=3.738e+02, percent-clipped=0.0 +2022-11-16 00:43:40,564 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2022-11-16 00:44:13,597 INFO [train.py:876] (3/4) Epoch 9, batch 200, loss[loss=0.102, simple_loss=0.1354, pruned_loss=0.03427, over 5715.00 frames. ], tot_loss[loss=0.1284, simple_loss=0.1515, pruned_loss=0.0527, over 694419.09 frames. ], batch size: 15, lr: 9.25e-03, grad_scale: 16.0 +2022-11-16 00:44:22,161 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:44:42,358 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4224, 4.4740, 3.0711, 4.2361, 3.3453, 2.9751, 2.4730, 3.8891], + device='cuda:3'), covar=tensor([0.1687, 0.0227, 0.0882, 0.0361, 0.0652, 0.1027, 0.1893, 0.0273], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0140, 0.0164, 0.0141, 0.0174, 0.0177, 0.0172, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 00:44:46,847 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.538e+01 1.502e+02 1.721e+02 2.221e+02 4.054e+02, threshold=3.441e+02, percent-clipped=2.0 +2022-11-16 00:44:55,448 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:45:18,171 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0287, 5.0356, 5.4222, 5.4098, 5.1584, 4.8813, 6.0460, 5.0893], + device='cuda:3'), covar=tensor([0.0461, 0.1178, 0.0367, 0.0840, 0.0559, 0.0326, 0.0566, 0.0524], + device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0100, 0.0088, 0.0112, 0.0082, 0.0074, 0.0139, 0.0094], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:45:21,749 INFO [train.py:876] (3/4) Epoch 9, batch 300, loss[loss=0.1248, simple_loss=0.1516, pruned_loss=0.04905, over 5759.00 frames. ], tot_loss[loss=0.1286, simple_loss=0.1508, pruned_loss=0.0532, over 850072.62 frames. ], batch size: 20, lr: 9.24e-03, grad_scale: 16.0 +2022-11-16 00:45:22,921 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6926, 2.6639, 2.3496, 2.8713, 2.0670, 2.4892, 2.4458, 3.4434], + device='cuda:3'), covar=tensor([0.1305, 0.1650, 0.2936, 0.1242, 0.2092, 0.1119, 0.2107, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0091, 0.0100, 0.0086, 0.0084, 0.0088, 0.0093, 0.0067], + 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') +2022-11-16 00:45:35,859 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:45:37,824 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:45:53,983 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.573e+02 2.012e+02 2.567e+02 4.757e+02, threshold=4.023e+02, percent-clipped=7.0 +2022-11-16 00:46:02,009 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6762, 1.1288, 1.0634, 1.3002, 1.1093, 1.1878, 1.0582, 1.1020], + device='cuda:3'), covar=tensor([0.2216, 0.1220, 0.1180, 0.0690, 0.1126, 0.1458, 0.1164, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0091, 0.0098, 0.0085, 0.0083, 0.0087, 0.0092, 0.0067], + 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') +2022-11-16 00:46:29,075 INFO [train.py:876] (3/4) Epoch 9, batch 400, loss[loss=0.1518, simple_loss=0.1572, pruned_loss=0.07324, over 5746.00 frames. ], tot_loss[loss=0.134, simple_loss=0.1545, pruned_loss=0.05671, over 937676.76 frames. ], batch size: 31, lr: 9.23e-03, grad_scale: 16.0 +2022-11-16 00:46:31,899 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6008, 2.2509, 3.1652, 2.8633, 3.3223, 2.1271, 2.9869, 3.6185], + device='cuda:3'), covar=tensor([0.0867, 0.2209, 0.1148, 0.1916, 0.0800, 0.2219, 0.1552, 0.1093], + device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0193, 0.0206, 0.0209, 0.0224, 0.0192, 0.0220, 0.0223], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:46:32,450 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4561, 3.4518, 3.2975, 3.7377, 3.1671, 3.1805, 4.0012, 3.2417], + device='cuda:3'), covar=tensor([0.0586, 0.0994, 0.0705, 0.0929, 0.0756, 0.0488, 0.0801, 0.1047], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0101, 0.0089, 0.0112, 0.0082, 0.0074, 0.0139, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:46:38,896 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7519, 3.7083, 3.7922, 4.2048, 3.6020, 3.4292, 4.4076, 3.6627], + device='cuda:3'), covar=tensor([0.0520, 0.0832, 0.0498, 0.0722, 0.0535, 0.0405, 0.0635, 0.0633], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0101, 0.0088, 0.0111, 0.0082, 0.0074, 0.0139, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:47:00,085 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1060, 3.4043, 3.3993, 3.3256, 3.4090, 3.3005, 1.2746, 3.3719], + device='cuda:3'), covar=tensor([0.0523, 0.0419, 0.0341, 0.0320, 0.0435, 0.0470, 0.3875, 0.0499], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0084, 0.0083, 0.0075, 0.0099, 0.0086, 0.0128, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:47:01,901 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.664e+02 1.953e+02 2.587e+02 4.932e+02, threshold=3.905e+02, percent-clipped=2.0 +2022-11-16 00:47:28,797 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6132, 2.6463, 2.3242, 2.7245, 2.0844, 2.0709, 2.4420, 3.0687], + device='cuda:3'), covar=tensor([0.1482, 0.1407, 0.2255, 0.1116, 0.1785, 0.1240, 0.1746, 0.0882], + device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0092, 0.0100, 0.0086, 0.0085, 0.0089, 0.0094, 0.0068], + 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') +2022-11-16 00:47:37,285 INFO [train.py:876] (3/4) Epoch 9, batch 500, loss[loss=0.1227, simple_loss=0.1424, pruned_loss=0.05151, over 5559.00 frames. ], tot_loss[loss=0.1355, simple_loss=0.1556, pruned_loss=0.05766, over 996910.42 frames. ], batch size: 13, lr: 9.22e-03, grad_scale: 16.0 +2022-11-16 00:47:40,005 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:47:44,736 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-16 00:48:09,720 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.707e+02 2.003e+02 2.573e+02 4.379e+02, threshold=4.006e+02, percent-clipped=1.0 +2022-11-16 00:48:17,717 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7717, 1.2184, 1.5721, 0.9159, 1.6510, 1.4805, 1.1529, 1.5850], + device='cuda:3'), covar=tensor([0.0627, 0.0751, 0.0545, 0.1143, 0.1881, 0.0587, 0.0756, 0.0504], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0018, 0.0013, 0.0015, 0.0014, 0.0011, 0.0016, 0.0012], + device='cuda:3'), out_proj_covar=tensor([6.2696e-05, 8.4118e-05, 6.3588e-05, 7.4588e-05, 6.7645e-05, 6.1147e-05, + 7.7497e-05, 6.1063e-05], device='cuda:3') +2022-11-16 00:48:20,972 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:48:36,112 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6172, 2.4209, 2.8131, 3.7007, 3.6751, 2.9809, 2.4088, 3.6216], + device='cuda:3'), covar=tensor([0.0622, 0.2829, 0.2196, 0.2662, 0.0947, 0.2645, 0.2088, 0.0535], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0204, 0.0195, 0.0320, 0.0225, 0.0209, 0.0195, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:48:43,683 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:48:44,825 INFO [train.py:876] (3/4) Epoch 9, batch 600, loss[loss=0.151, simple_loss=0.1757, pruned_loss=0.06312, over 5561.00 frames. ], tot_loss[loss=0.135, simple_loss=0.1555, pruned_loss=0.05728, over 1033704.26 frames. ], batch size: 30, lr: 9.22e-03, grad_scale: 16.0 +2022-11-16 00:48:58,075 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:49:00,352 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:49:17,736 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 1.543e+02 1.852e+02 2.303e+02 3.719e+02, threshold=3.705e+02, percent-clipped=0.0 +2022-11-16 00:49:25,152 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:49:30,911 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:49:32,859 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:49:52,480 INFO [train.py:876] (3/4) Epoch 9, batch 700, loss[loss=0.1357, simple_loss=0.1527, pruned_loss=0.05933, over 5730.00 frames. ], tot_loss[loss=0.1354, simple_loss=0.1561, pruned_loss=0.05731, over 1058685.45 frames. ], batch size: 31, lr: 9.21e-03, grad_scale: 8.0 +2022-11-16 00:50:11,762 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:50:25,240 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.741e+02 2.186e+02 2.790e+02 4.131e+02, threshold=4.372e+02, percent-clipped=5.0 +2022-11-16 00:50:34,381 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1753, 3.1725, 3.1106, 3.3433, 3.0481, 2.8430, 3.6056, 3.0469], + device='cuda:3'), covar=tensor([0.0505, 0.0862, 0.0559, 0.1091, 0.0666, 0.0518, 0.0869, 0.0865], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0100, 0.0089, 0.0111, 0.0083, 0.0073, 0.0139, 0.0094], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 00:50:52,786 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:50:59,888 INFO [train.py:876] (3/4) Epoch 9, batch 800, loss[loss=0.1509, simple_loss=0.1611, pruned_loss=0.07035, over 5559.00 frames. ], tot_loss[loss=0.1365, simple_loss=0.1568, pruned_loss=0.05813, over 1068562.43 frames. ], batch size: 40, lr: 9.20e-03, grad_scale: 8.0 +2022-11-16 00:51:08,758 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 +2022-11-16 00:51:33,802 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.601e+01 1.692e+02 2.170e+02 2.841e+02 6.263e+02, threshold=4.339e+02, percent-clipped=3.0 +2022-11-16 00:51:41,840 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:51:41,908 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5529, 1.4161, 1.4630, 0.9940, 1.1949, 1.4113, 1.0007, 0.9385], + device='cuda:3'), covar=tensor([0.0018, 0.0034, 0.0025, 0.0046, 0.0036, 0.0025, 0.0038, 0.0045], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0027, 0.0023, 0.0022, 0.0026, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8373e-05, 1.9801e-05, 1.9216e-05, 2.6197e-05, 2.2171e-05, 2.1253e-05, + 2.5000e-05, 2.5805e-05], device='cuda:3') +2022-11-16 00:52:08,557 INFO [train.py:876] (3/4) Epoch 9, batch 900, loss[loss=0.1259, simple_loss=0.1613, pruned_loss=0.04525, over 5531.00 frames. ], tot_loss[loss=0.137, simple_loss=0.1571, pruned_loss=0.05847, over 1068695.81 frames. ], batch size: 17, lr: 9.19e-03, grad_scale: 8.0 +2022-11-16 00:52:42,260 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.620e+01 1.634e+02 1.980e+02 2.467e+02 5.009e+02, threshold=3.960e+02, percent-clipped=1.0 +2022-11-16 00:52:42,445 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8473, 1.3053, 1.4688, 1.1481, 0.9535, 1.7922, 1.2156, 1.4804], + device='cuda:3'), covar=tensor([0.0018, 0.0052, 0.0047, 0.0070, 0.0159, 0.0039, 0.0032, 0.0035], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0022, 0.0027, 0.0024, 0.0022, 0.0026, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.8906e-05, 2.0386e-05, 1.9814e-05, 2.6730e-05, 2.2469e-05, 2.1543e-05, + 2.5362e-05, 2.6115e-05], device='cuda:3') +2022-11-16 00:52:46,083 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:53:08,448 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:53:16,811 INFO [train.py:876] (3/4) Epoch 9, batch 1000, loss[loss=0.1331, simple_loss=0.157, pruned_loss=0.05462, over 5546.00 frames. ], tot_loss[loss=0.1366, simple_loss=0.1567, pruned_loss=0.05832, over 1074532.14 frames. ], batch size: 40, lr: 9.19e-03, grad_scale: 8.0 +2022-11-16 00:53:18,484 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-16 00:53:50,612 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:53:51,081 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.814e+01 1.686e+02 1.957e+02 2.486e+02 6.337e+02, threshold=3.914e+02, percent-clipped=1.0 +2022-11-16 00:53:56,742 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-16 00:54:18,046 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:54:28,526 INFO [train.py:876] (3/4) Epoch 9, batch 1100, loss[loss=0.09247, simple_loss=0.1256, pruned_loss=0.02967, over 5728.00 frames. ], tot_loss[loss=0.1365, simple_loss=0.1565, pruned_loss=0.05823, over 1075979.58 frames. ], batch size: 15, lr: 9.18e-03, grad_scale: 8.0 +2022-11-16 00:54:50,125 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8305, 2.3207, 1.9515, 1.4023, 2.3098, 0.9674, 2.3457, 1.4865], + device='cuda:3'), covar=tensor([0.0986, 0.0269, 0.0904, 0.1382, 0.0310, 0.2000, 0.0311, 0.1214], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0106, 0.0115, 0.0114, 0.0105, 0.0124, 0.0098, 0.0115], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:55:01,717 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.767e+01 1.618e+02 1.962e+02 2.354e+02 6.461e+02, threshold=3.925e+02, percent-clipped=2.0 +2022-11-16 00:55:08,995 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:55:16,395 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.49 vs. limit=5.0 +2022-11-16 00:55:36,025 INFO [train.py:876] (3/4) Epoch 9, batch 1200, loss[loss=0.09264, simple_loss=0.1263, pruned_loss=0.0295, over 5724.00 frames. ], tot_loss[loss=0.1352, simple_loss=0.1559, pruned_loss=0.05728, over 1079917.29 frames. ], batch size: 13, lr: 9.17e-03, grad_scale: 8.0 +2022-11-16 00:55:41,360 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:56:09,405 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 1.585e+02 1.981e+02 2.415e+02 4.268e+02, threshold=3.961e+02, percent-clipped=1.0 +2022-11-16 00:56:12,833 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:56:14,203 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6940, 2.0191, 1.6883, 1.2422, 1.4642, 2.3138, 1.9300, 2.4544], + device='cuda:3'), covar=tensor([0.1847, 0.1469, 0.1749, 0.2720, 0.1341, 0.0975, 0.0614, 0.1136], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0181, 0.0161, 0.0189, 0.0171, 0.0186, 0.0155, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 00:56:43,597 INFO [train.py:876] (3/4) Epoch 9, batch 1300, loss[loss=0.08476, simple_loss=0.1149, pruned_loss=0.02731, over 5684.00 frames. ], tot_loss[loss=0.1352, simple_loss=0.1558, pruned_loss=0.05735, over 1082997.19 frames. ], batch size: 11, lr: 9.16e-03, grad_scale: 8.0 +2022-11-16 00:56:45,396 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:57:00,738 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:57:13,470 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:57:17,255 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.599e+01 1.572e+02 1.990e+02 2.382e+02 3.719e+02, threshold=3.980e+02, percent-clipped=0.0 +2022-11-16 00:57:22,272 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2287, 1.9282, 2.4220, 1.7207, 1.2574, 2.8041, 2.4678, 1.9774], + device='cuda:3'), covar=tensor([0.1006, 0.1223, 0.0844, 0.2935, 0.3168, 0.1594, 0.0845, 0.1128], + device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0072, 0.0071, 0.0084, 0.0063, 0.0053, 0.0059, 0.0070], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:57:26,356 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2022-11-16 00:57:38,565 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.51 vs. limit=5.0 +2022-11-16 00:57:40,991 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:57:41,699 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:57:42,909 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4877, 5.0913, 4.5825, 4.9728, 5.0839, 4.1828, 4.6391, 4.1659], + device='cuda:3'), covar=tensor([0.0296, 0.0449, 0.1594, 0.0433, 0.0365, 0.0480, 0.0352, 0.0715], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0167, 0.0264, 0.0163, 0.0207, 0.0168, 0.0175, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 00:57:43,017 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3690, 2.7471, 3.1499, 4.0948, 3.8689, 3.3442, 2.7800, 4.2129], + device='cuda:3'), covar=tensor([0.0476, 0.2965, 0.2019, 0.4374, 0.1249, 0.3114, 0.2416, 0.0595], + device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0204, 0.0194, 0.0317, 0.0221, 0.0210, 0.0196, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 00:57:48,627 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-16 00:57:51,211 INFO [train.py:876] (3/4) Epoch 9, batch 1400, loss[loss=0.1508, simple_loss=0.1733, pruned_loss=0.06415, over 5559.00 frames. ], tot_loss[loss=0.1334, simple_loss=0.1542, pruned_loss=0.0563, over 1080019.87 frames. ], batch size: 54, lr: 9.15e-03, grad_scale: 8.0 +2022-11-16 00:58:09,430 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1336, 2.0257, 2.4301, 1.7580, 1.2908, 2.6579, 2.4124, 2.1145], + device='cuda:3'), covar=tensor([0.0894, 0.1069, 0.0679, 0.2239, 0.1855, 0.2405, 0.1113, 0.1017], + device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0070, 0.0069, 0.0081, 0.0060, 0.0051, 0.0057, 0.0067], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 00:58:13,886 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 00:58:16,094 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2022-11-16 00:58:21,326 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4524, 2.5765, 3.8990, 3.3798, 4.3059, 2.7622, 3.8231, 4.3699], + device='cuda:3'), covar=tensor([0.0521, 0.1957, 0.0856, 0.1527, 0.0499, 0.1554, 0.1293, 0.0769], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0195, 0.0204, 0.0208, 0.0225, 0.0192, 0.0222, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 00:58:24,722 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.626e+02 1.961e+02 2.375e+02 3.725e+02, threshold=3.922e+02, percent-clipped=0.0 +2022-11-16 00:58:58,827 INFO [train.py:876] (3/4) Epoch 9, batch 1500, loss[loss=0.1588, simple_loss=0.1706, pruned_loss=0.07351, over 5740.00 frames. ], tot_loss[loss=0.1338, simple_loss=0.1546, pruned_loss=0.05645, over 1084255.98 frames. ], batch size: 27, lr: 9.15e-03, grad_scale: 8.0 +2022-11-16 00:59:18,813 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2022-11-16 00:59:31,911 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.993e+01 1.447e+02 1.801e+02 2.298e+02 3.676e+02, threshold=3.602e+02, percent-clipped=0.0 +2022-11-16 00:59:46,069 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0929, 4.2475, 4.0941, 4.4045, 3.8442, 3.5602, 4.8137, 4.0629], + device='cuda:3'), covar=tensor([0.0488, 0.0694, 0.0395, 0.0990, 0.0633, 0.0394, 0.0701, 0.0594], + device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0099, 0.0088, 0.0111, 0.0083, 0.0073, 0.0136, 0.0093], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:00:06,513 INFO [train.py:876] (3/4) Epoch 9, batch 1600, loss[loss=0.1808, simple_loss=0.1685, pruned_loss=0.09652, over 4594.00 frames. ], tot_loss[loss=0.1326, simple_loss=0.1539, pruned_loss=0.05567, over 1085947.58 frames. ], batch size: 135, lr: 9.14e-03, grad_scale: 8.0 +2022-11-16 01:00:36,436 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:00:40,151 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.161e+01 1.571e+02 1.942e+02 2.410e+02 4.577e+02, threshold=3.885e+02, percent-clipped=5.0 +2022-11-16 01:00:55,345 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:01:01,638 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:01:08,520 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:01:14,345 INFO [train.py:876] (3/4) Epoch 9, batch 1700, loss[loss=0.08841, simple_loss=0.1219, pruned_loss=0.02744, over 5466.00 frames. ], tot_loss[loss=0.1323, simple_loss=0.1537, pruned_loss=0.05544, over 1086069.15 frames. ], batch size: 11, lr: 9.13e-03, grad_scale: 8.0 +2022-11-16 01:01:22,411 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:01:35,853 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:01:47,042 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.735e+01 1.514e+02 1.939e+02 2.316e+02 5.054e+02, threshold=3.877e+02, percent-clipped=3.0 +2022-11-16 01:01:56,926 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9365, 1.4841, 1.4290, 1.3161, 1.0729, 1.9633, 1.5033, 1.2046], + device='cuda:3'), covar=tensor([0.2334, 0.1036, 0.2376, 0.2493, 0.2799, 0.0688, 0.1305, 0.2563], + device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0069, 0.0068, 0.0081, 0.0061, 0.0050, 0.0058, 0.0067], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 01:02:02,978 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:02:09,920 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1129, 3.7386, 4.0185, 3.7100, 4.1954, 3.9859, 3.8107, 4.2009], + device='cuda:3'), covar=tensor([0.0416, 0.0336, 0.0492, 0.0390, 0.0407, 0.0307, 0.0334, 0.0348], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0140, 0.0106, 0.0137, 0.0160, 0.0091, 0.0116, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 01:02:21,361 INFO [train.py:876] (3/4) Epoch 9, batch 1800, loss[loss=0.1184, simple_loss=0.1465, pruned_loss=0.04513, over 5276.00 frames. ], tot_loss[loss=0.1335, simple_loss=0.1547, pruned_loss=0.05613, over 1080665.75 frames. ], batch size: 79, lr: 9.12e-03, grad_scale: 8.0 +2022-11-16 01:02:23,210 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 +2022-11-16 01:02:57,436 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6971, 1.5501, 1.6069, 1.1866, 1.4316, 1.8349, 1.3122, 1.5504], + device='cuda:3'), covar=tensor([0.0036, 0.0036, 0.0022, 0.0047, 0.0047, 0.0027, 0.0033, 0.0035], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0022, 0.0028, 0.0024, 0.0024, 0.0026, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.9454e-05, 2.0337e-05, 1.9750e-05, 2.7663e-05, 2.3071e-05, 2.2827e-05, + 2.5551e-05, 2.6466e-05], device='cuda:3') +2022-11-16 01:02:59,238 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.503e+02 1.882e+02 2.236e+02 4.811e+02, threshold=3.764e+02, percent-clipped=2.0 +2022-11-16 01:03:08,920 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:03:16,837 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:03:25,135 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:03:33,729 INFO [train.py:876] (3/4) Epoch 9, batch 1900, loss[loss=0.1778, simple_loss=0.1776, pruned_loss=0.089, over 4981.00 frames. ], tot_loss[loss=0.1334, simple_loss=0.1543, pruned_loss=0.0563, over 1084121.21 frames. ], batch size: 109, lr: 9.12e-03, grad_scale: 8.0 +2022-11-16 01:03:34,465 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:03:49,824 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 01:03:57,586 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 01:03:57,810 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-16 01:04:06,950 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.557e+01 1.713e+02 2.167e+02 2.668e+02 5.052e+02, threshold=4.333e+02, percent-clipped=7.0 +2022-11-16 01:04:07,152 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:04:08,366 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6469, 3.6985, 3.6723, 3.5787, 3.6963, 3.7167, 1.4824, 3.8169], + device='cuda:3'), covar=tensor([0.0307, 0.0251, 0.0252, 0.0324, 0.0318, 0.0301, 0.2985, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0081, 0.0082, 0.0073, 0.0096, 0.0084, 0.0126, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:04:15,567 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:04:22,474 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2022-11-16 01:04:28,126 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:04:34,187 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 01:04:40,598 INFO [train.py:876] (3/4) Epoch 9, batch 2000, loss[loss=0.16, simple_loss=0.176, pruned_loss=0.07202, over 5563.00 frames. ], tot_loss[loss=0.1344, simple_loss=0.155, pruned_loss=0.05695, over 1083957.28 frames. ], batch size: 25, lr: 9.11e-03, grad_scale: 8.0 +2022-11-16 01:04:55,800 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:04:56,772 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 01:04:59,907 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:05:01,216 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:05:11,769 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9738, 2.1221, 2.7798, 2.4231, 2.5451, 1.9038, 2.6061, 2.9338], + device='cuda:3'), covar=tensor([0.0550, 0.1291, 0.0632, 0.1240, 0.0771, 0.1340, 0.0948, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0192, 0.0204, 0.0209, 0.0226, 0.0192, 0.0222, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:05:14,563 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.629e+02 2.002e+02 2.495e+02 4.506e+02, threshold=4.003e+02, percent-clipped=3.0 +2022-11-16 01:05:15,019 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4553, 3.9706, 4.2479, 3.9413, 4.4968, 4.2695, 4.0495, 4.5063], + device='cuda:3'), covar=tensor([0.0364, 0.0330, 0.0460, 0.0360, 0.0359, 0.0232, 0.0286, 0.0274], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0140, 0.0105, 0.0137, 0.0159, 0.0091, 0.0116, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 01:05:27,712 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:05:29,046 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8166, 3.6600, 3.8212, 3.6379, 3.8849, 3.8446, 1.4830, 4.0593], + device='cuda:3'), covar=tensor([0.0294, 0.0405, 0.0389, 0.0337, 0.0405, 0.0481, 0.3459, 0.0362], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0081, 0.0082, 0.0074, 0.0097, 0.0084, 0.0126, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:05:37,277 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:05:49,240 INFO [train.py:876] (3/4) Epoch 9, batch 2100, loss[loss=0.1166, simple_loss=0.1567, pruned_loss=0.03827, over 5581.00 frames. ], tot_loss[loss=0.1351, simple_loss=0.1553, pruned_loss=0.05741, over 1087244.69 frames. ], batch size: 24, lr: 9.10e-03, grad_scale: 8.0 +2022-11-16 01:05:56,130 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2022-11-16 01:06:22,290 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.667e+02 2.049e+02 2.499e+02 3.972e+02, threshold=4.098e+02, percent-clipped=0.0 +2022-11-16 01:06:56,810 INFO [train.py:876] (3/4) Epoch 9, batch 2200, loss[loss=0.1408, simple_loss=0.1565, pruned_loss=0.06258, over 5546.00 frames. ], tot_loss[loss=0.1361, simple_loss=0.156, pruned_loss=0.05808, over 1086266.47 frames. ], batch size: 40, lr: 9.09e-03, grad_scale: 8.0 +2022-11-16 01:07:09,682 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 01:07:18,005 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 01:07:26,522 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:07:30,026 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.625e+02 1.909e+02 2.506e+02 7.347e+02, threshold=3.819e+02, percent-clipped=1.0 +2022-11-16 01:07:35,207 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:07:36,622 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8495, 1.7633, 2.0400, 2.1198, 1.7417, 1.5863, 1.7187, 2.1996], + device='cuda:3'), covar=tensor([0.1578, 0.2184, 0.2243, 0.1312, 0.1711, 0.2406, 0.1823, 0.0862], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0091, 0.0101, 0.0085, 0.0085, 0.0088, 0.0092, 0.0068], + 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') +2022-11-16 01:07:36,869 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2022-11-16 01:08:04,142 INFO [train.py:876] (3/4) Epoch 9, batch 2300, loss[loss=0.09351, simple_loss=0.1228, pruned_loss=0.03208, over 5765.00 frames. ], tot_loss[loss=0.1318, simple_loss=0.1532, pruned_loss=0.05518, over 1088460.74 frames. ], batch size: 20, lr: 9.09e-03, grad_scale: 8.0 +2022-11-16 01:08:16,632 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2022-11-16 01:08:22,438 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:08:37,684 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.337e+01 1.638e+02 1.971e+02 2.511e+02 5.541e+02, threshold=3.943e+02, percent-clipped=0.0 +2022-11-16 01:08:45,376 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6150, 1.3658, 1.8332, 1.5389, 1.2963, 1.4125, 1.5968, 1.4811], + device='cuda:3'), covar=tensor([0.0063, 0.0069, 0.0024, 0.0035, 0.0124, 0.0111, 0.0026, 0.0037], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0022, 0.0029, 0.0025, 0.0024, 0.0026, 0.0026], + device='cuda:3'), out_proj_covar=tensor([1.9555e-05, 2.0441e-05, 1.9844e-05, 2.7812e-05, 2.3276e-05, 2.2903e-05, + 2.5555e-05, 2.5733e-05], device='cuda:3') +2022-11-16 01:08:50,485 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:08:54,915 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:08:56,296 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:09:08,596 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2022-11-16 01:09:12,097 INFO [train.py:876] (3/4) Epoch 9, batch 2400, loss[loss=0.1219, simple_loss=0.1503, pruned_loss=0.0468, over 5595.00 frames. ], tot_loss[loss=0.1329, simple_loss=0.1541, pruned_loss=0.05585, over 1090897.86 frames. ], batch size: 23, lr: 9.08e-03, grad_scale: 8.0 +2022-11-16 01:09:22,867 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:09:44,906 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.802e+02 2.307e+02 2.858e+02 9.175e+02, threshold=4.615e+02, percent-clipped=3.0 +2022-11-16 01:09:57,560 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4180, 2.5172, 2.1813, 2.6029, 2.0563, 2.1138, 2.2109, 2.9289], + device='cuda:3'), covar=tensor([0.1181, 0.1809, 0.2765, 0.1299, 0.2138, 0.1656, 0.1836, 0.3145], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0092, 0.0100, 0.0086, 0.0086, 0.0088, 0.0092, 0.0069], + 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') +2022-11-16 01:10:18,994 INFO [train.py:876] (3/4) Epoch 9, batch 2500, loss[loss=0.1484, simple_loss=0.1637, pruned_loss=0.06652, over 5683.00 frames. ], tot_loss[loss=0.1298, simple_loss=0.1522, pruned_loss=0.05367, over 1092032.39 frames. ], batch size: 17, lr: 9.07e-03, grad_scale: 8.0 +2022-11-16 01:10:31,965 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:10:40,130 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:10:46,714 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:10:48,607 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:10:51,736 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.647e+01 1.621e+02 1.973e+02 2.538e+02 4.790e+02, threshold=3.945e+02, percent-clipped=1.0 +2022-11-16 01:10:57,735 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:04,185 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:11,859 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:20,530 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:25,649 INFO [train.py:876] (3/4) Epoch 9, batch 2600, loss[loss=0.1137, simple_loss=0.1495, pruned_loss=0.03896, over 5804.00 frames. ], tot_loss[loss=0.1292, simple_loss=0.1522, pruned_loss=0.05311, over 1090308.13 frames. ], batch size: 21, lr: 9.06e-03, grad_scale: 8.0 +2022-11-16 01:11:27,139 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:28,890 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:11:58,598 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.669e+02 1.987e+02 2.479e+02 5.022e+02, threshold=3.974e+02, percent-clipped=3.0 +2022-11-16 01:12:06,034 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1883, 2.3567, 2.7089, 2.4505, 1.5761, 2.4783, 1.7194, 1.8888], + device='cuda:3'), covar=tensor([0.0284, 0.0121, 0.0130, 0.0197, 0.0350, 0.0152, 0.0346, 0.0201], + device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0159, 0.0169, 0.0192, 0.0179, 0.0168, 0.0179, 0.0168], + 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') +2022-11-16 01:12:18,172 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:12:33,334 INFO [train.py:876] (3/4) Epoch 9, batch 2700, loss[loss=0.1504, simple_loss=0.1722, pruned_loss=0.06427, over 5692.00 frames. ], tot_loss[loss=0.1305, simple_loss=0.1523, pruned_loss=0.05439, over 1084408.63 frames. ], batch size: 36, lr: 9.06e-03, grad_scale: 16.0 +2022-11-16 01:12:35,751 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 +2022-11-16 01:12:37,398 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7907, 3.0721, 2.5215, 3.1124, 2.4341, 3.0571, 2.7800, 3.4657], + device='cuda:3'), covar=tensor([0.1083, 0.0984, 0.2461, 0.1024, 0.1683, 0.0880, 0.1678, 0.1870], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0090, 0.0098, 0.0085, 0.0084, 0.0089, 0.0090, 0.0068], + 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') +2022-11-16 01:12:50,327 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:13:02,065 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6081, 4.5368, 4.5928, 4.7092, 4.0911, 3.6384, 5.0893, 4.4918], + device='cuda:3'), covar=tensor([0.0343, 0.0690, 0.0322, 0.0825, 0.0479, 0.0407, 0.0565, 0.0477], + device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0100, 0.0087, 0.0110, 0.0082, 0.0072, 0.0138, 0.0092], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:13:06,004 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.571e+02 1.865e+02 2.553e+02 5.202e+02, threshold=3.729e+02, percent-clipped=2.0 +2022-11-16 01:13:40,750 INFO [train.py:876] (3/4) Epoch 9, batch 2800, loss[loss=0.1473, simple_loss=0.1721, pruned_loss=0.06124, over 5587.00 frames. ], tot_loss[loss=0.13, simple_loss=0.1523, pruned_loss=0.05387, over 1088291.67 frames. ], batch size: 24, lr: 9.05e-03, grad_scale: 16.0 +2022-11-16 01:14:13,879 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.565e+02 1.869e+02 2.208e+02 4.201e+02, threshold=3.737e+02, percent-clipped=2.0 +2022-11-16 01:14:37,284 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:14:46,630 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:14:48,545 INFO [train.py:876] (3/4) Epoch 9, batch 2900, loss[loss=0.09106, simple_loss=0.1213, pruned_loss=0.03043, over 5726.00 frames. ], tot_loss[loss=0.1287, simple_loss=0.1512, pruned_loss=0.05304, over 1090681.30 frames. ], batch size: 9, lr: 9.04e-03, grad_scale: 16.0 +2022-11-16 01:15:18,593 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:15:21,362 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.585e+02 1.881e+02 2.261e+02 3.985e+02, threshold=3.762e+02, percent-clipped=2.0 +2022-11-16 01:15:52,807 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1324, 2.0566, 2.8172, 1.6289, 1.3146, 2.9974, 2.3017, 2.2572], + device='cuda:3'), covar=tensor([0.1092, 0.1287, 0.0537, 0.2652, 0.4169, 0.1661, 0.1056, 0.1166], + device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0071, 0.0071, 0.0084, 0.0063, 0.0052, 0.0059, 0.0070], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 01:15:54,246 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 01:15:55,636 INFO [train.py:876] (3/4) Epoch 9, batch 3000, loss[loss=0.09689, simple_loss=0.1282, pruned_loss=0.03277, over 5456.00 frames. ], tot_loss[loss=0.1319, simple_loss=0.1532, pruned_loss=0.05527, over 1090162.52 frames. ], batch size: 10, lr: 9.03e-03, grad_scale: 16.0 +2022-11-16 01:15:55,636 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 01:16:05,816 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7971, 1.6542, 1.2334, 1.0652, 1.5266, 1.4878, 1.3633, 0.7506], + device='cuda:3'), covar=tensor([0.0022, 0.0053, 0.0038, 0.0064, 0.0029, 0.0030, 0.0032, 0.0054], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0022, 0.0028, 0.0024, 0.0022, 0.0026, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.9213e-05, 1.9829e-05, 1.9593e-05, 2.6915e-05, 2.2621e-05, 2.1729e-05, + 2.5159e-05, 2.5526e-05], device='cuda:3') +2022-11-16 01:16:16,566 INFO [train.py:908] (3/4) Epoch 9, validation: loss=0.1637, simple_loss=0.1831, pruned_loss=0.07219, over 1530663.00 frames. +2022-11-16 01:16:16,567 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 01:16:44,214 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6917, 1.5101, 1.4399, 1.1601, 1.1922, 1.5259, 1.3924, 1.3623], + device='cuda:3'), covar=tensor([0.0044, 0.0043, 0.0052, 0.0052, 0.0071, 0.0057, 0.0031, 0.0031], + device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0020, 0.0021, 0.0027, 0.0023, 0.0022, 0.0026, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8931e-05, 1.9528e-05, 1.9389e-05, 2.6364e-05, 2.2142e-05, 2.1214e-05, + 2.4628e-05, 2.4944e-05], device='cuda:3') +2022-11-16 01:16:49,301 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.614e+02 2.000e+02 2.380e+02 4.865e+02, threshold=4.000e+02, percent-clipped=3.0 +2022-11-16 01:17:23,848 INFO [train.py:876] (3/4) Epoch 9, batch 3100, loss[loss=0.1331, simple_loss=0.1667, pruned_loss=0.04974, over 5650.00 frames. ], tot_loss[loss=0.1319, simple_loss=0.1535, pruned_loss=0.05517, over 1094679.79 frames. ], batch size: 32, lr: 9.03e-03, grad_scale: 16.0 +2022-11-16 01:17:57,137 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.659e+02 1.981e+02 2.507e+02 4.168e+02, threshold=3.962e+02, percent-clipped=1.0 +2022-11-16 01:18:03,845 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 +2022-11-16 01:18:08,519 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4826, 5.2173, 3.6841, 2.2363, 4.8112, 2.5060, 4.6450, 2.6769], + device='cuda:3'), covar=tensor([0.0973, 0.0112, 0.0530, 0.2033, 0.0135, 0.1414, 0.0170, 0.1478], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0105, 0.0113, 0.0114, 0.0102, 0.0124, 0.0098, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:18:30,145 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:18:31,951 INFO [train.py:876] (3/4) Epoch 9, batch 3200, loss[loss=0.1345, simple_loss=0.1626, pruned_loss=0.05317, over 5570.00 frames. ], tot_loss[loss=0.1319, simple_loss=0.1537, pruned_loss=0.05506, over 1094777.31 frames. ], batch size: 21, lr: 9.02e-03, grad_scale: 16.0 +2022-11-16 01:18:59,219 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:19:02,493 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:19:04,945 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.596e+02 1.869e+02 2.254e+02 4.160e+02, threshold=3.737e+02, percent-clipped=1.0 +2022-11-16 01:19:28,627 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5733, 2.8525, 2.3481, 2.9419, 2.1869, 2.7374, 2.3410, 3.2829], + device='cuda:3'), covar=tensor([0.1355, 0.1630, 0.2360, 0.1685, 0.2169, 0.1197, 0.1566, 0.1789], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0091, 0.0098, 0.0085, 0.0084, 0.0088, 0.0088, 0.0068], + 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') +2022-11-16 01:19:39,535 INFO [train.py:876] (3/4) Epoch 9, batch 3300, loss[loss=0.1074, simple_loss=0.1301, pruned_loss=0.0424, over 5734.00 frames. ], tot_loss[loss=0.1323, simple_loss=0.1535, pruned_loss=0.05558, over 1091100.75 frames. ], batch size: 11, lr: 9.01e-03, grad_scale: 16.0 +2022-11-16 01:19:41,540 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9150, 1.5359, 1.8136, 1.3955, 0.9610, 2.2508, 1.7379, 1.2860], + device='cuda:3'), covar=tensor([0.1622, 0.1118, 0.1240, 0.2350, 0.2677, 0.1013, 0.1002, 0.2120], + device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0070, 0.0071, 0.0084, 0.0063, 0.0053, 0.0059, 0.0070], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2022-11-16 01:19:50,006 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7686, 4.2739, 3.8096, 4.2039, 4.2422, 3.6706, 3.7184, 3.6805], + device='cuda:3'), covar=tensor([0.0636, 0.0379, 0.1438, 0.0437, 0.0391, 0.0409, 0.0638, 0.0492], + device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0169, 0.0265, 0.0166, 0.0209, 0.0168, 0.0178, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 01:20:10,388 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9117, 3.9083, 4.1420, 3.6427, 3.9585, 3.8458, 1.6615, 4.1312], + device='cuda:3'), covar=tensor([0.0446, 0.0492, 0.0365, 0.0523, 0.0548, 0.0658, 0.3409, 0.0416], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0083, 0.0085, 0.0076, 0.0101, 0.0087, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:20:12,867 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.659e+02 1.971e+02 2.388e+02 4.331e+02, threshold=3.941e+02, percent-clipped=1.0 +2022-11-16 01:20:47,469 INFO [train.py:876] (3/4) Epoch 9, batch 3400, loss[loss=0.1739, simple_loss=0.1635, pruned_loss=0.0922, over 4203.00 frames. ], tot_loss[loss=0.1301, simple_loss=0.1517, pruned_loss=0.05421, over 1087296.91 frames. ], batch size: 181, lr: 9.01e-03, grad_scale: 16.0 +2022-11-16 01:21:14,481 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0666, 1.2653, 1.3127, 0.8142, 0.8432, 1.1303, 0.9077, 0.8898], + device='cuda:3'), covar=tensor([0.0018, 0.0013, 0.0017, 0.0023, 0.0024, 0.0022, 0.0027, 0.0034], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0019, 0.0021, 0.0027, 0.0023, 0.0021, 0.0025, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.8178e-05, 1.8479e-05, 1.8909e-05, 2.5978e-05, 2.1436e-05, 2.0198e-05, + 2.4146e-05, 2.4188e-05], device='cuda:3') +2022-11-16 01:21:20,961 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.197e+01 1.625e+02 1.885e+02 2.405e+02 4.262e+02, threshold=3.771e+02, percent-clipped=2.0 +2022-11-16 01:21:36,199 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0727, 3.3010, 2.7824, 3.2319, 2.5939, 3.6640, 3.3793, 3.7606], + device='cuda:3'), covar=tensor([0.0954, 0.1367, 0.2334, 0.1275, 0.1774, 0.0647, 0.1278, 0.3345], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0093, 0.0099, 0.0086, 0.0087, 0.0089, 0.0090, 0.0069], + 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') +2022-11-16 01:21:55,064 INFO [train.py:876] (3/4) Epoch 9, batch 3500, loss[loss=0.1906, simple_loss=0.1929, pruned_loss=0.09411, over 5681.00 frames. ], tot_loss[loss=0.1289, simple_loss=0.1507, pruned_loss=0.0535, over 1082516.29 frames. ], batch size: 19, lr: 9.00e-03, grad_scale: 16.0 +2022-11-16 01:21:59,896 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9383, 1.4208, 1.4887, 1.4430, 1.3412, 1.3417, 1.2994, 1.4050], + device='cuda:3'), covar=tensor([0.3703, 0.2401, 0.2153, 0.1620, 0.2293, 0.2528, 0.2131, 0.1005], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0095, 0.0100, 0.0087, 0.0088, 0.0091, 0.0091, 0.0070], + 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') +2022-11-16 01:22:22,288 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:22:24,906 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9354, 1.1559, 1.6945, 1.3429, 1.4581, 1.4426, 1.2261, 1.3503], + device='cuda:3'), covar=tensor([0.0023, 0.0056, 0.0033, 0.0036, 0.0041, 0.0048, 0.0035, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0027, 0.0023, 0.0021, 0.0025, 0.0024], + device='cuda:3'), out_proj_covar=tensor([1.8368e-05, 1.8799e-05, 1.9131e-05, 2.6076e-05, 2.1591e-05, 2.0532e-05, + 2.4644e-05, 2.4179e-05], device='cuda:3') +2022-11-16 01:22:28,020 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.632e+02 2.023e+02 2.591e+02 5.866e+02, threshold=4.047e+02, percent-clipped=6.0 +2022-11-16 01:22:32,172 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:22:39,373 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:22:42,917 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7654, 4.4408, 3.2639, 1.9076, 4.1548, 1.7497, 4.1198, 2.4515], + device='cuda:3'), covar=tensor([0.1246, 0.0130, 0.0597, 0.2299, 0.0170, 0.1900, 0.0188, 0.1585], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0105, 0.0114, 0.0115, 0.0103, 0.0123, 0.0099, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:22:52,102 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-16 01:22:54,373 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:23:02,299 INFO [train.py:876] (3/4) Epoch 9, batch 3600, loss[loss=0.1439, simple_loss=0.1702, pruned_loss=0.05879, over 5508.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1514, pruned_loss=0.05423, over 1083036.33 frames. ], batch size: 17, lr: 8.99e-03, grad_scale: 16.0 +2022-11-16 01:23:12,905 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:23:20,619 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:23:35,484 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 1.617e+02 2.067e+02 2.692e+02 5.357e+02, threshold=4.135e+02, percent-clipped=2.0 +2022-11-16 01:23:40,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1601, 3.1746, 3.1905, 1.9939, 2.7343, 3.7243, 3.3638, 3.8593], + device='cuda:3'), covar=tensor([0.1866, 0.1431, 0.0765, 0.2392, 0.0698, 0.0597, 0.0509, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0182, 0.0165, 0.0186, 0.0169, 0.0188, 0.0158, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 01:23:54,574 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7299, 4.6130, 4.7568, 4.7859, 4.3585, 4.0262, 5.3124, 4.6748], + device='cuda:3'), covar=tensor([0.0379, 0.0791, 0.0398, 0.0926, 0.0472, 0.0325, 0.0622, 0.0606], + device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0099, 0.0087, 0.0109, 0.0081, 0.0072, 0.0138, 0.0092], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:24:05,497 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1960, 1.4356, 1.3222, 0.8584, 1.1436, 1.4187, 1.1105, 1.2180], + device='cuda:3'), covar=tensor([0.0054, 0.0043, 0.0052, 0.0046, 0.0041, 0.0033, 0.0048, 0.0060], + device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0045, 0.0047, 0.0047, 0.0046, 0.0041, 0.0045, 0.0040], + device='cuda:3'), out_proj_covar=tensor([4.5627e-05, 4.1081e-05, 4.2153e-05, 4.2526e-05, 4.1077e-05, 3.5967e-05, + 4.1363e-05, 3.5589e-05], device='cuda:3') +2022-11-16 01:24:10,273 INFO [train.py:876] (3/4) Epoch 9, batch 3700, loss[loss=0.1301, simple_loss=0.1643, pruned_loss=0.04791, over 5554.00 frames. ], tot_loss[loss=0.1305, simple_loss=0.1523, pruned_loss=0.05434, over 1088354.45 frames. ], batch size: 40, lr: 8.98e-03, grad_scale: 16.0 +2022-11-16 01:24:34,710 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:24:42,850 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.652e+01 1.593e+02 2.000e+02 2.466e+02 4.913e+02, threshold=3.999e+02, percent-clipped=2.0 +2022-11-16 01:24:46,035 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4763, 2.2467, 2.5874, 3.5020, 3.3140, 2.4312, 2.1610, 3.3884], + device='cuda:3'), covar=tensor([0.0752, 0.2780, 0.1889, 0.2359, 0.1334, 0.3135, 0.2183, 0.0736], + device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0202, 0.0194, 0.0315, 0.0220, 0.0206, 0.0192, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 01:25:15,833 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:25:16,949 INFO [train.py:876] (3/4) Epoch 9, batch 3800, loss[loss=0.1375, simple_loss=0.1669, pruned_loss=0.05411, over 5565.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1522, pruned_loss=0.05383, over 1089880.07 frames. ], batch size: 25, lr: 8.98e-03, grad_scale: 16.0 +2022-11-16 01:25:43,429 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 01:25:50,159 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.003e+01 1.597e+02 1.892e+02 2.479e+02 4.563e+02, threshold=3.783e+02, percent-clipped=2.0 +2022-11-16 01:25:55,126 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0243, 3.2585, 2.4459, 2.9605, 2.2071, 2.4910, 1.8117, 2.8639], + device='cuda:3'), covar=tensor([0.1437, 0.0320, 0.1121, 0.0428, 0.1289, 0.1017, 0.2019, 0.0420], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0138, 0.0163, 0.0141, 0.0174, 0.0174, 0.0171, 0.0155], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 01:26:21,601 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.87 vs. limit=5.0 +2022-11-16 01:26:24,516 INFO [train.py:876] (3/4) Epoch 9, batch 3900, loss[loss=0.1344, simple_loss=0.1553, pruned_loss=0.05675, over 5560.00 frames. ], tot_loss[loss=0.1306, simple_loss=0.1526, pruned_loss=0.05434, over 1084190.19 frames. ], batch size: 25, lr: 8.97e-03, grad_scale: 16.0 +2022-11-16 01:26:31,653 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:26:37,234 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4461, 2.5056, 2.4398, 2.7396, 2.0880, 2.3581, 2.1808, 3.2451], + device='cuda:3'), covar=tensor([0.1146, 0.1606, 0.2332, 0.1185, 0.2128, 0.1130, 0.1876, 0.1104], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0095, 0.0099, 0.0087, 0.0088, 0.0090, 0.0092, 0.0069], + 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') +2022-11-16 01:26:39,491 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:26:55,160 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:26:57,545 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.676e+02 2.020e+02 2.333e+02 7.144e+02, threshold=4.039e+02, percent-clipped=2.0 +2022-11-16 01:27:07,754 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-16 01:27:32,456 INFO [train.py:876] (3/4) Epoch 9, batch 4000, loss[loss=0.1233, simple_loss=0.1366, pruned_loss=0.05501, over 5589.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1521, pruned_loss=0.05385, over 1085002.37 frames. ], batch size: 43, lr: 8.96e-03, grad_scale: 16.0 +2022-11-16 01:27:36,458 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:28:05,699 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.381e+01 1.660e+02 2.083e+02 2.529e+02 5.015e+02, threshold=4.165e+02, percent-clipped=1.0 +2022-11-16 01:28:07,963 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3110, 2.1660, 2.7919, 3.5201, 3.4590, 2.5607, 2.3711, 3.4462], + device='cuda:3'), covar=tensor([0.0803, 0.2606, 0.1765, 0.2180, 0.0842, 0.2835, 0.1904, 0.0691], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0204, 0.0194, 0.0316, 0.0220, 0.0208, 0.0193, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 01:28:08,554 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0109, 1.5425, 1.9235, 1.1602, 1.6055, 1.6219, 1.2694, 1.4002], + device='cuda:3'), covar=tensor([0.0017, 0.0038, 0.0025, 0.0039, 0.0032, 0.0044, 0.0034, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0021, 0.0027, 0.0024, 0.0021, 0.0026, 0.0025], + device='cuda:3'), out_proj_covar=tensor([1.8507e-05, 1.8923e-05, 1.9053e-05, 2.6707e-05, 2.2281e-05, 2.0772e-05, + 2.4974e-05, 2.4613e-05], device='cuda:3') +2022-11-16 01:28:35,677 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:28:37,738 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:28:40,189 INFO [train.py:876] (3/4) Epoch 9, batch 4100, loss[loss=0.1201, simple_loss=0.1568, pruned_loss=0.0417, over 5750.00 frames. ], tot_loss[loss=0.1297, simple_loss=0.1521, pruned_loss=0.05368, over 1084176.95 frames. ], batch size: 13, lr: 8.96e-03, grad_scale: 16.0 +2022-11-16 01:28:40,294 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0105, 4.3115, 4.9012, 4.7694, 5.0077, 4.9669, 2.4100, 5.1318], + device='cuda:3'), covar=tensor([0.0267, 0.0589, 0.0275, 0.0198, 0.0369, 0.0275, 0.2673, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0084, 0.0087, 0.0078, 0.0104, 0.0089, 0.0133, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:28:45,336 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2022-11-16 01:29:00,968 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9218, 3.5695, 2.4466, 3.2250, 2.5921, 2.5821, 1.9024, 3.0605], + device='cuda:3'), covar=tensor([0.1563, 0.0224, 0.0994, 0.0385, 0.0855, 0.0998, 0.1826, 0.0349], + device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0138, 0.0161, 0.0141, 0.0173, 0.0172, 0.0169, 0.0154], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 01:29:13,573 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.483e+02 1.988e+02 2.459e+02 5.062e+02, threshold=3.976e+02, percent-clipped=4.0 +2022-11-16 01:29:14,502 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6857, 3.9214, 3.8332, 3.3152, 2.0927, 3.9091, 2.1823, 3.2469], + device='cuda:3'), covar=tensor([0.0433, 0.0187, 0.0146, 0.0403, 0.0574, 0.0178, 0.0528, 0.0288], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0161, 0.0169, 0.0192, 0.0182, 0.0172, 0.0181, 0.0171], + 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') +2022-11-16 01:29:19,053 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:29:47,829 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6244, 5.7422, 4.0199, 2.4576, 5.2120, 2.7241, 5.2376, 3.4409], + device='cuda:3'), covar=tensor([0.0927, 0.0069, 0.0495, 0.1848, 0.0138, 0.1417, 0.0122, 0.1176], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0105, 0.0115, 0.0115, 0.0105, 0.0124, 0.0099, 0.0114], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:29:48,369 INFO [train.py:876] (3/4) Epoch 9, batch 4200, loss[loss=0.0771, simple_loss=0.1056, pruned_loss=0.02428, over 4540.00 frames. ], tot_loss[loss=0.1286, simple_loss=0.151, pruned_loss=0.0531, over 1077782.15 frames. ], batch size: 5, lr: 8.95e-03, grad_scale: 16.0 +2022-11-16 01:29:55,571 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:29:57,573 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:30:02,844 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:30:21,289 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.948e+01 1.584e+02 2.002e+02 2.529e+02 4.819e+02, threshold=4.003e+02, percent-clipped=3.0 +2022-11-16 01:30:28,437 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:30:35,829 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:30:39,204 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:30:55,988 INFO [train.py:876] (3/4) Epoch 9, batch 4300, loss[loss=0.1016, simple_loss=0.1355, pruned_loss=0.03388, over 5584.00 frames. ], tot_loss[loss=0.1297, simple_loss=0.1518, pruned_loss=0.05383, over 1077171.90 frames. ], batch size: 24, lr: 8.94e-03, grad_scale: 16.0 +2022-11-16 01:30:57,017 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:31:07,153 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3419, 3.2328, 3.2319, 3.0517, 3.2228, 3.0790, 1.2599, 3.3575], + device='cuda:3'), covar=tensor([0.0522, 0.0704, 0.0612, 0.0498, 0.0632, 0.0780, 0.4165, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0084, 0.0086, 0.0077, 0.0101, 0.0086, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:31:25,064 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 +2022-11-16 01:31:28,849 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.728e+02 2.092e+02 2.577e+02 4.668e+02, threshold=4.183e+02, percent-clipped=3.0 +2022-11-16 01:31:42,541 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:31:58,791 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:03,250 INFO [train.py:876] (3/4) Epoch 9, batch 4400, loss[loss=0.1316, simple_loss=0.1563, pruned_loss=0.05339, over 5820.00 frames. ], tot_loss[loss=0.1283, simple_loss=0.1511, pruned_loss=0.05281, over 1077323.84 frames. ], batch size: 18, lr: 8.93e-03, grad_scale: 16.0 +2022-11-16 01:32:08,962 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:23,686 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:31,375 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:36,536 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 1.591e+02 1.968e+02 2.413e+02 5.184e+02, threshold=3.935e+02, percent-clipped=2.0 +2022-11-16 01:32:38,751 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:51,045 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:32:56,453 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0618, 4.5711, 4.8373, 4.6836, 5.1423, 5.0163, 4.5935, 5.1301], + device='cuda:3'), covar=tensor([0.0406, 0.0280, 0.0401, 0.0280, 0.0354, 0.0134, 0.0229, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0141, 0.0104, 0.0139, 0.0161, 0.0093, 0.0116, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 01:33:04,330 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:33:11,148 INFO [train.py:876] (3/4) Epoch 9, batch 4500, loss[loss=0.08627, simple_loss=0.1245, pruned_loss=0.02402, over 5356.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.151, pruned_loss=0.05256, over 1080823.77 frames. ], batch size: 6, lr: 8.93e-03, grad_scale: 16.0 +2022-11-16 01:33:44,304 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.986e+01 1.583e+02 2.006e+02 2.339e+02 4.798e+02, threshold=4.012e+02, percent-clipped=3.0 +2022-11-16 01:33:45,179 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:33:58,105 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:34:16,524 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:34:18,997 INFO [train.py:876] (3/4) Epoch 9, batch 4600, loss[loss=0.1406, simple_loss=0.1488, pruned_loss=0.06623, over 4178.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1506, pruned_loss=0.05274, over 1073119.43 frames. ], batch size: 181, lr: 8.92e-03, grad_scale: 16.0 +2022-11-16 01:34:19,709 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:34:52,258 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.537e+01 1.546e+02 1.870e+02 2.337e+02 4.153e+02, threshold=3.740e+02, percent-clipped=1.0 +2022-11-16 01:34:52,348 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:34:57,801 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:35:24,018 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1836, 3.1474, 3.0542, 1.6660, 3.1378, 3.4264, 3.4313, 3.8660], + device='cuda:3'), covar=tensor([0.2219, 0.1298, 0.0750, 0.3083, 0.0565, 0.0719, 0.0541, 0.0580], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0181, 0.0163, 0.0191, 0.0173, 0.0191, 0.0161, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 01:35:24,192 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 +2022-11-16 01:35:27,132 INFO [train.py:876] (3/4) Epoch 9, batch 4700, loss[loss=0.176, simple_loss=0.1837, pruned_loss=0.08416, over 5017.00 frames. ], tot_loss[loss=0.1279, simple_loss=0.1506, pruned_loss=0.05264, over 1079263.38 frames. ], batch size: 109, lr: 8.91e-03, grad_scale: 32.0 +2022-11-16 01:35:42,714 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:36:00,631 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.617e+02 2.054e+02 2.509e+02 4.948e+02, threshold=4.108e+02, percent-clipped=3.0 +2022-11-16 01:36:02,092 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:36:09,995 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:36:34,657 INFO [train.py:876] (3/4) Epoch 9, batch 4800, loss[loss=0.1138, simple_loss=0.1445, pruned_loss=0.04154, over 5709.00 frames. ], tot_loss[loss=0.1267, simple_loss=0.1495, pruned_loss=0.05191, over 1080399.47 frames. ], batch size: 12, lr: 8.91e-03, grad_scale: 16.0 +2022-11-16 01:36:34,704 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:36:54,058 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9901, 1.9792, 1.9281, 2.0937, 1.7216, 1.5137, 1.8028, 2.3123], + device='cuda:3'), covar=tensor([0.1520, 0.1977, 0.2604, 0.1380, 0.2264, 0.2767, 0.2180, 0.1750], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0094, 0.0098, 0.0086, 0.0086, 0.0089, 0.0091, 0.0068], + 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') +2022-11-16 01:37:05,676 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:37:09,060 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.223e+01 1.584e+02 1.833e+02 2.198e+02 5.399e+02, threshold=3.666e+02, percent-clipped=2.0 +2022-11-16 01:37:22,501 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:37:34,613 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6797, 4.8864, 3.3034, 4.8445, 3.6871, 3.4331, 2.7735, 4.3318], + device='cuda:3'), covar=tensor([0.1378, 0.0232, 0.0980, 0.0206, 0.0518, 0.0807, 0.1716, 0.0223], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0139, 0.0163, 0.0142, 0.0177, 0.0174, 0.0171, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 01:37:42,690 INFO [train.py:876] (3/4) Epoch 9, batch 4900, loss[loss=0.08761, simple_loss=0.1265, pruned_loss=0.02435, over 5411.00 frames. ], tot_loss[loss=0.1282, simple_loss=0.1506, pruned_loss=0.05286, over 1078389.99 frames. ], batch size: 11, lr: 8.90e-03, grad_scale: 16.0 +2022-11-16 01:37:54,809 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:38:17,345 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.788e+02 2.171e+02 2.737e+02 6.659e+02, threshold=4.342e+02, percent-clipped=3.0 +2022-11-16 01:38:18,118 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:38:18,827 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7749, 1.9287, 1.7538, 1.2544, 1.8563, 2.3538, 2.0484, 2.4129], + device='cuda:3'), covar=tensor([0.1647, 0.1814, 0.1723, 0.2770, 0.1095, 0.0921, 0.0791, 0.0975], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0183, 0.0166, 0.0190, 0.0173, 0.0189, 0.0162, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 01:38:22,371 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0550, 2.5826, 2.4202, 1.5162, 2.5555, 2.8142, 2.7086, 3.0399], + device='cuda:3'), covar=tensor([0.1648, 0.1501, 0.1259, 0.2670, 0.0762, 0.0985, 0.0738, 0.0812], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0183, 0.0166, 0.0189, 0.0173, 0.0189, 0.0162, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 01:38:49,665 INFO [train.py:876] (3/4) Epoch 9, batch 5000, loss[loss=0.095, simple_loss=0.1163, pruned_loss=0.03686, over 5535.00 frames. ], tot_loss[loss=0.1296, simple_loss=0.1515, pruned_loss=0.05384, over 1075101.39 frames. ], batch size: 10, lr: 8.89e-03, grad_scale: 8.0 +2022-11-16 01:39:06,647 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:39:19,714 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:39:24,000 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.228e+01 1.555e+02 1.927e+02 2.355e+02 4.069e+02, threshold=3.855e+02, percent-clipped=0.0 +2022-11-16 01:39:32,909 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:39:38,709 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:39:41,354 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7351, 4.7180, 4.7604, 4.9280, 4.4155, 4.2777, 5.4564, 4.7918], + device='cuda:3'), covar=tensor([0.0424, 0.0871, 0.0269, 0.1135, 0.0351, 0.0269, 0.0524, 0.0469], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0103, 0.0090, 0.0115, 0.0084, 0.0075, 0.0141, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:39:57,148 INFO [train.py:876] (3/4) Epoch 9, batch 5100, loss[loss=0.1182, simple_loss=0.1445, pruned_loss=0.04599, over 5737.00 frames. ], tot_loss[loss=0.1322, simple_loss=0.1532, pruned_loss=0.05565, over 1075774.00 frames. ], batch size: 13, lr: 8.88e-03, grad_scale: 8.0 +2022-11-16 01:40:00,538 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:40:05,005 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:40:24,385 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1163, 4.3310, 3.9924, 3.8388, 4.3807, 4.2752, 1.8132, 4.5995], + device='cuda:3'), covar=tensor([0.0457, 0.0397, 0.0346, 0.0418, 0.0316, 0.0384, 0.3058, 0.0329], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0083, 0.0086, 0.0077, 0.0102, 0.0087, 0.0131, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:40:27,688 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:40:31,502 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.064e+01 1.589e+02 1.945e+02 2.471e+02 3.864e+02, threshold=3.889e+02, percent-clipped=1.0 +2022-11-16 01:40:35,937 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-16 01:40:56,920 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9436, 3.2479, 2.4140, 2.8810, 2.1871, 2.4515, 1.8610, 2.8280], + device='cuda:3'), covar=tensor([0.1357, 0.0246, 0.0881, 0.0439, 0.1169, 0.0959, 0.1740, 0.0406], + device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0137, 0.0162, 0.0140, 0.0175, 0.0171, 0.0168, 0.0155], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 01:40:59,525 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:41:04,331 INFO [train.py:876] (3/4) Epoch 9, batch 5200, loss[loss=0.1218, simple_loss=0.1524, pruned_loss=0.04565, over 5630.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1521, pruned_loss=0.05384, over 1087668.36 frames. ], batch size: 38, lr: 8.88e-03, grad_scale: 8.0 +2022-11-16 01:41:11,938 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2022-11-16 01:41:33,504 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:41:34,115 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5475, 4.2413, 3.2614, 1.9549, 4.0045, 1.5027, 4.0278, 2.3730], + device='cuda:3'), covar=tensor([0.1434, 0.0156, 0.0691, 0.2177, 0.0228, 0.2089, 0.0189, 0.1638], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0105, 0.0115, 0.0115, 0.0105, 0.0123, 0.0099, 0.0113], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:41:34,791 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:41:38,978 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.457e+02 1.778e+02 2.243e+02 4.193e+02, threshold=3.556e+02, percent-clipped=1.0 +2022-11-16 01:41:39,820 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:41:57,218 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 01:42:11,810 INFO [train.py:876] (3/4) Epoch 9, batch 5300, loss[loss=0.0973, simple_loss=0.1267, pruned_loss=0.03394, over 5739.00 frames. ], tot_loss[loss=0.1272, simple_loss=0.1505, pruned_loss=0.05199, over 1089903.14 frames. ], batch size: 14, lr: 8.87e-03, grad_scale: 8.0 +2022-11-16 01:42:11,849 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:42:14,934 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:42:16,192 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:42:18,105 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4522, 4.0007, 3.6753, 3.3601, 1.9509, 3.8004, 2.0734, 3.3180], + device='cuda:3'), covar=tensor([0.0538, 0.0153, 0.0170, 0.0463, 0.0717, 0.0176, 0.0594, 0.0164], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0163, 0.0172, 0.0192, 0.0185, 0.0172, 0.0184, 0.0173], + 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') +2022-11-16 01:42:37,341 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6136, 2.4038, 2.6327, 3.6504, 3.4413, 2.6163, 2.4103, 3.6924], + device='cuda:3'), covar=tensor([0.0723, 0.2682, 0.2463, 0.1690, 0.1052, 0.2868, 0.1959, 0.0472], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0201, 0.0196, 0.0316, 0.0223, 0.0210, 0.0193, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 01:42:42,703 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2022-11-16 01:42:46,192 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.109e+01 1.553e+02 1.965e+02 2.267e+02 4.174e+02, threshold=3.929e+02, percent-clipped=2.0 +2022-11-16 01:43:13,466 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:43:19,364 INFO [train.py:876] (3/4) Epoch 9, batch 5400, loss[loss=0.08699, simple_loss=0.1258, pruned_loss=0.02408, over 5546.00 frames. ], tot_loss[loss=0.1269, simple_loss=0.1507, pruned_loss=0.05162, over 1086517.78 frames. ], batch size: 10, lr: 8.86e-03, grad_scale: 8.0 +2022-11-16 01:43:19,446 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:43:25,752 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:43:55,257 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 1.654e+02 2.116e+02 2.560e+02 5.660e+02, threshold=4.233e+02, percent-clipped=4.0 +2022-11-16 01:43:56,880 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:44:09,230 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:44:22,097 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5561, 0.9965, 1.6794, 1.0670, 1.3463, 1.5807, 1.2802, 1.1788], + device='cuda:3'), covar=tensor([0.0041, 0.0052, 0.0024, 0.0046, 0.0115, 0.0067, 0.0038, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0021, 0.0022, 0.0028, 0.0025, 0.0023, 0.0027, 0.0027], + device='cuda:3'), out_proj_covar=tensor([1.9798e-05, 2.0384e-05, 2.0239e-05, 2.7904e-05, 2.3205e-05, 2.2550e-05, + 2.5947e-05, 2.6792e-05], device='cuda:3') +2022-11-16 01:44:28,194 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3651, 2.3236, 2.1032, 2.3500, 2.1651, 1.7292, 2.2488, 2.5669], + device='cuda:3'), covar=tensor([0.1368, 0.2424, 0.2471, 0.1087, 0.1757, 0.3148, 0.1769, 0.1094], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0094, 0.0098, 0.0087, 0.0084, 0.0090, 0.0091, 0.0068], + 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') +2022-11-16 01:44:30,044 INFO [train.py:876] (3/4) Epoch 9, batch 5500, loss[loss=0.1295, simple_loss=0.1508, pruned_loss=0.05412, over 5236.00 frames. ], tot_loss[loss=0.1291, simple_loss=0.1521, pruned_loss=0.05302, over 1084298.60 frames. ], batch size: 79, lr: 8.86e-03, grad_scale: 8.0 +2022-11-16 01:45:04,281 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.831e+01 1.633e+02 1.870e+02 2.382e+02 5.170e+02, threshold=3.741e+02, percent-clipped=1.0 +2022-11-16 01:45:20,201 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6731, 2.3094, 2.7916, 2.0699, 1.2946, 3.2787, 2.6965, 2.4517], + device='cuda:3'), covar=tensor([0.0749, 0.0980, 0.0714, 0.2464, 0.4583, 0.1299, 0.1045, 0.1084], + device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0074, 0.0073, 0.0085, 0.0064, 0.0054, 0.0063, 0.0073], + 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') +2022-11-16 01:45:31,870 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2636, 3.6939, 3.2651, 3.6533, 3.6835, 3.1472, 3.3734, 3.0460], + device='cuda:3'), covar=tensor([0.0986, 0.0495, 0.1455, 0.0502, 0.0484, 0.0524, 0.0639, 0.0869], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0167, 0.0263, 0.0166, 0.0208, 0.0167, 0.0179, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 01:45:36,947 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:45:37,521 INFO [train.py:876] (3/4) Epoch 9, batch 5600, loss[loss=0.09077, simple_loss=0.1254, pruned_loss=0.02804, over 5721.00 frames. ], tot_loss[loss=0.1266, simple_loss=0.1502, pruned_loss=0.05146, over 1091016.77 frames. ], batch size: 15, lr: 8.85e-03, grad_scale: 8.0 +2022-11-16 01:45:38,273 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:45:46,085 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6680, 4.1290, 3.7536, 3.4634, 2.0370, 3.8355, 2.1066, 3.2079], + device='cuda:3'), covar=tensor([0.0451, 0.0162, 0.0198, 0.0380, 0.0633, 0.0184, 0.0591, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0162, 0.0171, 0.0192, 0.0184, 0.0171, 0.0182, 0.0172], + 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') +2022-11-16 01:46:00,503 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:46:12,214 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.664e+02 2.053e+02 2.570e+02 4.903e+02, threshold=4.106e+02, percent-clipped=3.0 +2022-11-16 01:46:23,806 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:46:41,788 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:46:45,509 INFO [train.py:876] (3/4) Epoch 9, batch 5700, loss[loss=0.08537, simple_loss=0.1232, pruned_loss=0.02376, over 5551.00 frames. ], tot_loss[loss=0.1263, simple_loss=0.1499, pruned_loss=0.05135, over 1088683.42 frames. ], batch size: 13, lr: 8.84e-03, grad_scale: 8.0 +2022-11-16 01:46:45,633 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:05,282 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:12,710 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:18,380 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:18,413 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:20,208 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.040e+01 1.586e+02 1.908e+02 2.453e+02 4.306e+02, threshold=3.816e+02, percent-clipped=1.0 +2022-11-16 01:47:29,395 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:47:53,503 INFO [train.py:876] (3/4) Epoch 9, batch 5800, loss[loss=0.0842, simple_loss=0.1177, pruned_loss=0.02535, over 5481.00 frames. ], tot_loss[loss=0.1274, simple_loss=0.1505, pruned_loss=0.05214, over 1088336.93 frames. ], batch size: 10, lr: 8.84e-03, grad_scale: 8.0 +2022-11-16 01:47:53,668 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:48:23,619 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-16 01:48:25,311 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:48:28,336 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.198e+01 1.501e+02 1.757e+02 2.288e+02 3.552e+02, threshold=3.514e+02, percent-clipped=0.0 +2022-11-16 01:48:38,124 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0763, 3.6605, 3.8753, 3.6500, 4.1208, 3.7704, 3.8020, 4.1136], + device='cuda:3'), covar=tensor([0.0350, 0.0361, 0.0462, 0.0338, 0.0397, 0.0508, 0.0348, 0.0324], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0136, 0.0101, 0.0134, 0.0159, 0.0090, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 01:49:00,846 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:49:01,351 INFO [train.py:876] (3/4) Epoch 9, batch 5900, loss[loss=0.1127, simple_loss=0.1452, pruned_loss=0.04013, over 5598.00 frames. ], tot_loss[loss=0.1262, simple_loss=0.1494, pruned_loss=0.05146, over 1086901.72 frames. ], batch size: 18, lr: 8.83e-03, grad_scale: 8.0 +2022-11-16 01:49:02,112 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:49:06,066 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:49:32,765 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:49:34,399 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:49:35,640 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.966e+01 1.656e+02 2.045e+02 2.527e+02 4.457e+02, threshold=4.090e+02, percent-clipped=4.0 +2022-11-16 01:49:35,766 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3418, 2.4084, 2.0926, 2.3917, 2.4198, 2.2098, 2.1252, 2.2558], + device='cuda:3'), covar=tensor([0.0461, 0.0706, 0.1815, 0.0615, 0.0657, 0.0608, 0.1092, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0170, 0.0264, 0.0165, 0.0207, 0.0166, 0.0179, 0.0165], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 01:49:56,342 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7912, 3.1412, 2.7799, 3.3165, 2.7364, 3.2196, 3.4741, 3.3959], + device='cuda:3'), covar=tensor([0.1379, 0.1505, 0.2141, 0.1270, 0.1739, 0.1074, 0.1152, 0.3858], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0095, 0.0100, 0.0088, 0.0086, 0.0090, 0.0092, 0.0070], + 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') +2022-11-16 01:50:00,881 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:50:08,147 INFO [train.py:876] (3/4) Epoch 9, batch 6000, loss[loss=0.121, simple_loss=0.1492, pruned_loss=0.04636, over 5762.00 frames. ], tot_loss[loss=0.1275, simple_loss=0.1502, pruned_loss=0.05235, over 1084662.06 frames. ], batch size: 16, lr: 8.82e-03, grad_scale: 8.0 +2022-11-16 01:50:08,147 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 01:50:16,261 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1752, 1.4769, 1.3130, 1.0346, 1.1844, 1.3479, 0.8469, 1.2769], + device='cuda:3'), covar=tensor([0.0054, 0.0027, 0.0038, 0.0059, 0.0046, 0.0035, 0.0087, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0045, 0.0047, 0.0047, 0.0046, 0.0041, 0.0045, 0.0040], + device='cuda:3'), out_proj_covar=tensor([4.5314e-05, 4.0484e-05, 4.2065e-05, 4.1928e-05, 4.0713e-05, 3.5636e-05, + 4.1294e-05, 3.5069e-05], device='cuda:3') +2022-11-16 01:50:25,850 INFO [train.py:908] (3/4) Epoch 9, validation: loss=0.1648, simple_loss=0.1829, pruned_loss=0.07333, over 1530663.00 frames. +2022-11-16 01:50:25,851 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 01:50:42,278 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:50:57,926 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:50:59,713 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.067e+01 1.626e+02 1.990e+02 2.309e+02 5.533e+02, threshold=3.980e+02, percent-clipped=3.0 +2022-11-16 01:51:02,873 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:06,690 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1127, 4.1506, 4.0104, 3.7633, 4.1919, 4.0655, 1.4751, 4.3072], + device='cuda:3'), covar=tensor([0.0316, 0.0323, 0.0305, 0.0394, 0.0354, 0.0420, 0.3521, 0.0337], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0081, 0.0085, 0.0076, 0.0100, 0.0085, 0.0129, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:51:10,134 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:15,907 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2022-11-16 01:51:29,854 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:30,476 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:33,035 INFO [train.py:876] (3/4) Epoch 9, batch 6100, loss[loss=0.1258, simple_loss=0.1525, pruned_loss=0.04954, over 5788.00 frames. ], tot_loss[loss=0.1261, simple_loss=0.1496, pruned_loss=0.05133, over 1087204.75 frames. ], batch size: 21, lr: 8.82e-03, grad_scale: 8.0 +2022-11-16 01:51:41,915 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:44,146 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:51:52,923 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 01:51:57,146 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4336, 1.1206, 1.3779, 0.9668, 1.5752, 1.3127, 1.0296, 1.1637], + device='cuda:3'), covar=tensor([0.0732, 0.0915, 0.0317, 0.0967, 0.0391, 0.0899, 0.0895, 0.0754], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0019, 0.0013, 0.0016, 0.0014, 0.0012, 0.0017, 0.0012], + device='cuda:3'), out_proj_covar=tensor([6.5137e-05, 8.8713e-05, 6.6620e-05, 7.9079e-05, 7.1674e-05, 6.5394e-05, + 8.1681e-05, 6.5081e-05], device='cuda:3') +2022-11-16 01:52:07,147 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.572e+02 1.862e+02 2.390e+02 5.405e+02, threshold=3.724e+02, percent-clipped=2.0 +2022-11-16 01:52:13,915 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5031, 1.3473, 1.3999, 1.1182, 1.3938, 1.3815, 1.2508, 0.8010], + device='cuda:3'), covar=tensor([0.0019, 0.0030, 0.0024, 0.0037, 0.0042, 0.0035, 0.0033, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0023, 0.0030, 0.0025, 0.0024, 0.0028, 0.0028], + device='cuda:3'), out_proj_covar=tensor([2.0344e-05, 2.0987e-05, 2.1230e-05, 2.9011e-05, 2.3960e-05, 2.3277e-05, + 2.7641e-05, 2.7701e-05], device='cuda:3') +2022-11-16 01:52:18,851 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:52:40,442 INFO [train.py:876] (3/4) Epoch 9, batch 6200, loss[loss=0.1359, simple_loss=0.1582, pruned_loss=0.05679, over 5545.00 frames. ], tot_loss[loss=0.1275, simple_loss=0.1507, pruned_loss=0.05214, over 1092489.69 frames. ], batch size: 40, lr: 8.81e-03, grad_scale: 8.0 +2022-11-16 01:52:41,778 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:53:00,598 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:53:14,907 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.648e+02 1.944e+02 2.421e+02 4.973e+02, threshold=3.888e+02, percent-clipped=5.0 +2022-11-16 01:53:40,225 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:53:40,816 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:53:48,093 INFO [train.py:876] (3/4) Epoch 9, batch 6300, loss[loss=0.1132, simple_loss=0.1427, pruned_loss=0.04183, over 5654.00 frames. ], tot_loss[loss=0.1294, simple_loss=0.1519, pruned_loss=0.05344, over 1091159.59 frames. ], batch size: 32, lr: 8.80e-03, grad_scale: 8.0 +2022-11-16 01:53:53,485 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1556, 2.5541, 2.7222, 3.7235, 3.8325, 2.9606, 2.5851, 4.0117], + device='cuda:3'), covar=tensor([0.0413, 0.3134, 0.2554, 0.3351, 0.1054, 0.2896, 0.2055, 0.0548], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0199, 0.0191, 0.0311, 0.0218, 0.0203, 0.0190, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 01:54:03,892 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:54:12,491 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:54:21,619 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:54:22,699 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 1.573e+02 1.938e+02 2.415e+02 6.064e+02, threshold=3.877e+02, percent-clipped=4.0 +2022-11-16 01:54:35,349 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.88 vs. limit=5.0 +2022-11-16 01:54:36,387 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:54:52,163 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:54:55,720 INFO [train.py:876] (3/4) Epoch 9, batch 6400, loss[loss=0.1713, simple_loss=0.1751, pruned_loss=0.08373, over 5375.00 frames. ], tot_loss[loss=0.1284, simple_loss=0.1511, pruned_loss=0.05285, over 1085245.12 frames. ], batch size: 70, lr: 8.80e-03, grad_scale: 8.0 +2022-11-16 01:55:03,012 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:55:03,718 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:55:24,425 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:55:30,096 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.639e+02 1.887e+02 2.519e+02 5.774e+02, threshold=3.775e+02, percent-clipped=3.0 +2022-11-16 01:55:39,833 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4148, 2.4137, 2.6578, 3.5230, 3.4333, 2.6936, 2.4538, 3.4158], + device='cuda:3'), covar=tensor([0.0767, 0.3208, 0.2079, 0.2829, 0.1173, 0.3377, 0.2003, 0.0670], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0198, 0.0191, 0.0311, 0.0219, 0.0205, 0.0191, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 01:55:43,726 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:55:44,483 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 +2022-11-16 01:55:45,011 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:01,745 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8427, 1.0750, 1.9434, 1.5243, 1.6499, 2.0728, 1.6821, 1.5814], + device='cuda:3'), covar=tensor([0.0027, 0.0056, 0.0035, 0.0087, 0.0035, 0.0043, 0.0033, 0.0033], + device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0023, 0.0030, 0.0025, 0.0024, 0.0028, 0.0028], + device='cuda:3'), out_proj_covar=tensor([2.0579e-05, 2.0807e-05, 2.1068e-05, 2.9178e-05, 2.3624e-05, 2.3010e-05, + 2.7510e-05, 2.7335e-05], device='cuda:3') +2022-11-16 01:56:02,881 INFO [train.py:876] (3/4) Epoch 9, batch 6500, loss[loss=0.1098, simple_loss=0.141, pruned_loss=0.03929, over 5589.00 frames. ], tot_loss[loss=0.1294, simple_loss=0.1519, pruned_loss=0.05346, over 1080623.56 frames. ], batch size: 22, lr: 8.79e-03, grad_scale: 8.0 +2022-11-16 01:56:04,264 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:19,250 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-16 01:56:19,469 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:20,838 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:24,834 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:36,724 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:37,276 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.652e+01 1.582e+02 1.894e+02 2.392e+02 5.037e+02, threshold=3.789e+02, percent-clipped=1.0 +2022-11-16 01:56:49,591 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:56:54,291 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:57:02,060 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:57:10,222 INFO [train.py:876] (3/4) Epoch 9, batch 6600, loss[loss=0.07687, simple_loss=0.123, pruned_loss=0.01537, over 5480.00 frames. ], tot_loss[loss=0.1289, simple_loss=0.1515, pruned_loss=0.05311, over 1081590.30 frames. ], batch size: 11, lr: 8.78e-03, grad_scale: 8.0 +2022-11-16 01:57:22,011 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.23 vs. limit=5.0 +2022-11-16 01:57:30,810 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:57:32,170 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5591, 3.2540, 3.3242, 3.1395, 2.0216, 3.4433, 2.2418, 2.9064], + device='cuda:3'), covar=tensor([0.0338, 0.0399, 0.0185, 0.0341, 0.0479, 0.0172, 0.0443, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0162, 0.0170, 0.0192, 0.0181, 0.0170, 0.0181, 0.0172], + 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') +2022-11-16 01:57:35,336 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:57:35,381 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0383, 4.1034, 4.0029, 3.9565, 2.3857, 4.3581, 2.5498, 3.8280], + device='cuda:3'), covar=tensor([0.0352, 0.0173, 0.0154, 0.0232, 0.0513, 0.0123, 0.0433, 0.0247], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0162, 0.0170, 0.0192, 0.0181, 0.0170, 0.0181, 0.0172], + 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') +2022-11-16 01:57:40,465 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:57:44,978 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.581e+02 1.869e+02 2.492e+02 4.006e+02, threshold=3.739e+02, percent-clipped=2.0 +2022-11-16 01:58:18,045 INFO [train.py:876] (3/4) Epoch 9, batch 6700, loss[loss=0.1142, simple_loss=0.1405, pruned_loss=0.04393, over 5740.00 frames. ], tot_loss[loss=0.1274, simple_loss=0.1502, pruned_loss=0.05226, over 1086608.14 frames. ], batch size: 15, lr: 8.77e-03, grad_scale: 8.0 +2022-11-16 01:58:25,385 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5287, 4.5401, 4.7153, 4.8658, 4.4452, 4.1155, 5.1676, 4.5933], + device='cuda:3'), covar=tensor([0.0425, 0.0703, 0.0316, 0.0859, 0.0367, 0.0307, 0.0525, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0105, 0.0090, 0.0115, 0.0085, 0.0076, 0.0142, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 01:58:25,467 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:58:52,429 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.525e+01 1.574e+02 2.010e+02 2.472e+02 4.884e+02, threshold=4.021e+02, percent-clipped=4.0 +2022-11-16 01:58:57,705 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:59:03,714 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:59:20,619 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3123, 1.7482, 1.5106, 1.1935, 1.4777, 1.5056, 1.1432, 1.3701], + device='cuda:3'), covar=tensor([0.0045, 0.0047, 0.0045, 0.0050, 0.0038, 0.0032, 0.0047, 0.0084], + device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0047, 0.0049, 0.0049, 0.0049, 0.0044, 0.0046, 0.0041], + device='cuda:3'), out_proj_covar=tensor([4.7223e-05, 4.2140e-05, 4.3485e-05, 4.4103e-05, 4.3212e-05, 3.8271e-05, + 4.2178e-05, 3.6417e-05], device='cuda:3') +2022-11-16 01:59:25,577 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:59:26,077 INFO [train.py:876] (3/4) Epoch 9, batch 6800, loss[loss=0.07921, simple_loss=0.1225, pruned_loss=0.01794, over 5524.00 frames. ], tot_loss[loss=0.1263, simple_loss=0.1497, pruned_loss=0.05144, over 1088053.47 frames. ], batch size: 13, lr: 8.77e-03, grad_scale: 8.0 +2022-11-16 01:59:45,809 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:59:46,799 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5907, 1.9106, 1.6179, 1.1744, 1.8284, 0.8573, 1.9570, 1.2693], + device='cuda:3'), covar=tensor([0.0953, 0.0313, 0.1038, 0.1335, 0.0362, 0.2087, 0.0341, 0.1340], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0102, 0.0113, 0.0111, 0.0101, 0.0121, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 01:59:48,108 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 01:59:58,318 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5381, 1.2171, 1.4947, 1.1579, 1.2736, 1.5222, 1.2442, 0.9338], + device='cuda:3'), covar=tensor([0.0029, 0.0046, 0.0051, 0.0054, 0.0044, 0.0041, 0.0036, 0.0050], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0024, 0.0031, 0.0026, 0.0025, 0.0029, 0.0029], + device='cuda:3'), out_proj_covar=tensor([2.1486e-05, 2.1814e-05, 2.2218e-05, 3.0765e-05, 2.4904e-05, 2.4107e-05, + 2.8489e-05, 2.8532e-05], device='cuda:3') +2022-11-16 02:00:04,567 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.642e+02 2.032e+02 2.678e+02 4.129e+02, threshold=4.063e+02, percent-clipped=1.0 +2022-11-16 02:00:09,662 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 +2022-11-16 02:00:10,606 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:00:18,371 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:00:22,724 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:00:25,196 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:00:37,934 INFO [train.py:876] (3/4) Epoch 9, batch 6900, loss[loss=0.1309, simple_loss=0.1469, pruned_loss=0.05745, over 5114.00 frames. ], tot_loss[loss=0.127, simple_loss=0.15, pruned_loss=0.05196, over 1088775.22 frames. ], batch size: 91, lr: 8.76e-03, grad_scale: 8.0 +2022-11-16 02:00:43,534 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2022-11-16 02:00:53,903 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:00:58,440 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:01:03,493 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5161, 1.7705, 2.0646, 1.4515, 1.3683, 2.4825, 1.9829, 1.7510], + device='cuda:3'), covar=tensor([0.1188, 0.1319, 0.1113, 0.2558, 0.2819, 0.0729, 0.1120, 0.1712], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0077, 0.0075, 0.0090, 0.0065, 0.0055, 0.0064, 0.0075], + 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') +2022-11-16 02:01:04,192 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:01:07,809 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:01:12,980 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.555e+02 1.818e+02 2.213e+02 4.720e+02, threshold=3.636e+02, percent-clipped=2.0 +2022-11-16 02:01:15,698 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4147, 3.4222, 3.5769, 3.4713, 3.5020, 3.3785, 1.2705, 3.5747], + device='cuda:3'), covar=tensor([0.0247, 0.0316, 0.0269, 0.0256, 0.0347, 0.0346, 0.3135, 0.0324], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0082, 0.0084, 0.0074, 0.0100, 0.0085, 0.0127, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:01:40,199 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:01:45,807 INFO [train.py:876] (3/4) Epoch 9, batch 7000, loss[loss=0.1354, simple_loss=0.1579, pruned_loss=0.05643, over 5536.00 frames. ], tot_loss[loss=0.126, simple_loss=0.1491, pruned_loss=0.05141, over 1085309.41 frames. ], batch size: 15, lr: 8.75e-03, grad_scale: 16.0 +2022-11-16 02:01:58,600 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7130, 4.7016, 3.5947, 1.9644, 4.2503, 1.8488, 4.3243, 2.5001], + device='cuda:3'), covar=tensor([0.1253, 0.0092, 0.0464, 0.2043, 0.0166, 0.1829, 0.0186, 0.1530], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0104, 0.0114, 0.0112, 0.0103, 0.0123, 0.0099, 0.0113], + 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') +2022-11-16 02:02:06,683 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6634, 2.8890, 2.6050, 3.0063, 2.4566, 2.5009, 2.7554, 3.4057], + device='cuda:3'), covar=tensor([0.1011, 0.1342, 0.2039, 0.1069, 0.1407, 0.1430, 0.1557, 0.0573], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0094, 0.0097, 0.0088, 0.0084, 0.0090, 0.0092, 0.0070], + 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') +2022-11-16 02:02:19,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.702e+02 2.050e+02 2.485e+02 3.887e+02, threshold=4.100e+02, percent-clipped=2.0 +2022-11-16 02:02:32,224 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:02:53,101 INFO [train.py:876] (3/4) Epoch 9, batch 7100, loss[loss=0.0984, simple_loss=0.1385, pruned_loss=0.02917, over 5576.00 frames. ], tot_loss[loss=0.1255, simple_loss=0.1494, pruned_loss=0.05079, over 1088574.24 frames. ], batch size: 16, lr: 8.75e-03, grad_scale: 16.0 +2022-11-16 02:02:58,818 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:03:04,614 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:03:12,357 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:03:16,362 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1843, 2.7901, 2.5882, 1.7207, 2.5792, 2.9549, 2.7635, 3.4454], + device='cuda:3'), covar=tensor([0.1702, 0.1575, 0.1122, 0.2608, 0.0745, 0.0938, 0.0500, 0.0642], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0178, 0.0163, 0.0187, 0.0175, 0.0192, 0.0157, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], + device='cuda:3') +2022-11-16 02:03:28,151 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.907e+01 1.625e+02 1.964e+02 2.489e+02 4.704e+02, threshold=3.927e+02, percent-clipped=1.0 +2022-11-16 02:03:31,219 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:03:36,720 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2022-11-16 02:03:41,034 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:03:45,050 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:03:49,612 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:04:01,553 INFO [train.py:876] (3/4) Epoch 9, batch 7200, loss[loss=0.1236, simple_loss=0.1488, pruned_loss=0.04919, over 5574.00 frames. ], tot_loss[loss=0.1262, simple_loss=0.15, pruned_loss=0.05119, over 1091766.19 frames. ], batch size: 22, lr: 8.74e-03, grad_scale: 16.0 +2022-11-16 02:04:04,219 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6410, 4.0788, 4.4361, 4.1700, 4.7206, 4.5597, 4.2408, 4.6610], + device='cuda:3'), covar=tensor([0.0356, 0.0393, 0.0428, 0.0401, 0.0344, 0.0209, 0.0269, 0.0317], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0141, 0.0104, 0.0139, 0.0162, 0.0093, 0.0117, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 02:04:18,240 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:04:20,595 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 +2022-11-16 02:04:22,062 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:04:22,791 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:04:24,637 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:04:28,760 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 +2022-11-16 02:04:35,328 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.623e+01 1.548e+02 1.861e+02 2.163e+02 4.412e+02, threshold=3.722e+02, percent-clipped=1.0 +2022-11-16 02:04:43,769 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6576, 5.1158, 3.3887, 4.8566, 3.7967, 3.4526, 3.0486, 4.3700], + device='cuda:3'), covar=tensor([0.1272, 0.0174, 0.0977, 0.0257, 0.0580, 0.0876, 0.1624, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0139, 0.0161, 0.0143, 0.0176, 0.0174, 0.0170, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 02:05:33,820 INFO [train.py:876] (3/4) Epoch 10, batch 0, loss[loss=0.1627, simple_loss=0.18, pruned_loss=0.07268, over 5615.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.18, pruned_loss=0.07268, over 5615.00 frames. ], batch size: 32, lr: 8.31e-03, grad_scale: 16.0 +2022-11-16 02:05:33,821 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 02:05:40,326 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7155, 1.7233, 1.5350, 1.1688, 1.5697, 1.9490, 1.5058, 0.8667], + device='cuda:3'), covar=tensor([0.0023, 0.0048, 0.0029, 0.0059, 0.0035, 0.0015, 0.0029, 0.0063], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0024, 0.0030, 0.0026, 0.0024, 0.0029, 0.0028], + device='cuda:3'), out_proj_covar=tensor([2.1427e-05, 2.1643e-05, 2.1729e-05, 2.9859e-05, 2.4173e-05, 2.3394e-05, + 2.7580e-05, 2.8184e-05], device='cuda:3') +2022-11-16 02:05:42,374 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7023, 2.6555, 2.6147, 2.2843, 2.7654, 2.7276, 2.6931, 2.7397], + device='cuda:3'), covar=tensor([0.0401, 0.0437, 0.0434, 0.0752, 0.0435, 0.0215, 0.0319, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0139, 0.0104, 0.0137, 0.0161, 0.0092, 0.0116, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 02:05:50,439 INFO [train.py:908] (3/4) Epoch 10, validation: loss=0.1665, simple_loss=0.1839, pruned_loss=0.07458, over 1530663.00 frames. +2022-11-16 02:05:50,440 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 02:05:50,498 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:05:55,449 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:06:18,046 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 +2022-11-16 02:06:43,805 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.371e+01 1.569e+02 1.978e+02 2.490e+02 6.089e+02, threshold=3.956e+02, percent-clipped=4.0 +2022-11-16 02:06:46,663 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:06:57,654 INFO [train.py:876] (3/4) Epoch 10, batch 100, loss[loss=0.06908, simple_loss=0.1072, pruned_loss=0.01547, over 5515.00 frames. ], tot_loss[loss=0.127, simple_loss=0.1493, pruned_loss=0.05231, over 426709.41 frames. ], batch size: 13, lr: 8.30e-03, grad_scale: 16.0 +2022-11-16 02:07:27,716 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:07:51,875 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.404e+01 1.672e+02 2.008e+02 2.472e+02 6.251e+02, threshold=4.017e+02, percent-clipped=3.0 +2022-11-16 02:07:54,630 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:08:00,533 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:08:05,593 INFO [train.py:876] (3/4) Epoch 10, batch 200, loss[loss=0.1324, simple_loss=0.1551, pruned_loss=0.05486, over 5550.00 frames. ], tot_loss[loss=0.1262, simple_loss=0.1496, pruned_loss=0.05143, over 690568.82 frames. ], batch size: 13, lr: 8.30e-03, grad_scale: 16.0 +2022-11-16 02:08:26,066 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:08:47,441 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:08:58,246 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.788e+01 1.559e+02 1.915e+02 2.411e+02 5.348e+02, threshold=3.830e+02, percent-clipped=2.0 +2022-11-16 02:09:08,268 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 02:09:12,969 INFO [train.py:876] (3/4) Epoch 10, batch 300, loss[loss=0.1321, simple_loss=0.158, pruned_loss=0.05314, over 5818.00 frames. ], tot_loss[loss=0.1283, simple_loss=0.151, pruned_loss=0.05284, over 846929.87 frames. ], batch size: 22, lr: 8.29e-03, grad_scale: 16.0 +2022-11-16 02:09:19,525 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:09:29,198 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-16 02:09:47,985 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7859, 1.3729, 1.5385, 1.0808, 2.2593, 1.5161, 1.5839, 1.8432], + device='cuda:3'), covar=tensor([0.0638, 0.0797, 0.0796, 0.1579, 0.0852, 0.0785, 0.0474, 0.0215], + device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0019, 0.0013, 0.0017, 0.0014, 0.0013, 0.0018, 0.0013], + device='cuda:3'), out_proj_covar=tensor([6.7162e-05, 9.1795e-05, 6.9735e-05, 8.2762e-05, 7.2521e-05, 6.8407e-05, + 8.5660e-05, 6.7800e-05], device='cuda:3') +2022-11-16 02:09:56,921 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7036, 2.6292, 2.2922, 2.8397, 2.2192, 2.4980, 2.6406, 3.2324], + device='cuda:3'), covar=tensor([0.1144, 0.1659, 0.3218, 0.1445, 0.2006, 0.1534, 0.2084, 0.2403], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0097, 0.0101, 0.0092, 0.0088, 0.0094, 0.0096, 0.0071], + 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') +2022-11-16 02:10:05,914 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.600e+01 1.581e+02 1.964e+02 2.494e+02 5.554e+02, threshold=3.929e+02, percent-clipped=0.0 +2022-11-16 02:10:12,712 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:10:20,645 INFO [train.py:876] (3/4) Epoch 10, batch 400, loss[loss=0.1172, simple_loss=0.153, pruned_loss=0.04069, over 5705.00 frames. ], tot_loss[loss=0.1249, simple_loss=0.1487, pruned_loss=0.05054, over 944456.44 frames. ], batch size: 13, lr: 8.28e-03, grad_scale: 16.0 +2022-11-16 02:10:46,451 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:10:53,797 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 02:11:13,839 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.863e+01 1.601e+02 1.999e+02 2.596e+02 7.493e+02, threshold=3.998e+02, percent-clipped=4.0 +2022-11-16 02:11:22,469 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:11:27,626 INFO [train.py:876] (3/4) Epoch 10, batch 500, loss[loss=0.1769, simple_loss=0.1641, pruned_loss=0.09485, over 4092.00 frames. ], tot_loss[loss=0.1253, simple_loss=0.149, pruned_loss=0.05081, over 1000158.90 frames. ], batch size: 181, lr: 8.28e-03, grad_scale: 16.0 +2022-11-16 02:11:52,024 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3221, 3.7255, 2.8101, 1.7879, 3.4354, 1.4412, 3.5551, 1.8407], + device='cuda:3'), covar=tensor([0.1243, 0.0166, 0.0718, 0.1944, 0.0242, 0.1872, 0.0239, 0.1601], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0103, 0.0113, 0.0113, 0.0102, 0.0122, 0.0099, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 02:11:55,177 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:11:57,929 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5311, 2.6515, 2.4542, 2.8070, 2.1497, 2.2860, 2.4488, 3.2198], + device='cuda:3'), covar=tensor([0.1170, 0.1625, 0.2122, 0.2811, 0.1758, 0.1800, 0.1873, 0.1351], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0096, 0.0100, 0.0092, 0.0088, 0.0093, 0.0095, 0.0071], + 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') +2022-11-16 02:12:02,632 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 02:12:21,979 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.783e+01 1.631e+02 2.001e+02 2.434e+02 4.617e+02, threshold=4.002e+02, percent-clipped=2.0 +2022-11-16 02:12:35,742 INFO [train.py:876] (3/4) Epoch 10, batch 600, loss[loss=0.08934, simple_loss=0.1167, pruned_loss=0.03099, over 5046.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1486, pruned_loss=0.05034, over 1035928.92 frames. ], batch size: 7, lr: 8.27e-03, grad_scale: 16.0 +2022-11-16 02:13:27,937 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.503e+01 1.627e+02 2.016e+02 2.685e+02 4.936e+02, threshold=4.031e+02, percent-clipped=2.0 +2022-11-16 02:13:39,959 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:13:43,118 INFO [train.py:876] (3/4) Epoch 10, batch 700, loss[loss=0.1433, simple_loss=0.1622, pruned_loss=0.06223, over 5741.00 frames. ], tot_loss[loss=0.1252, simple_loss=0.1487, pruned_loss=0.05088, over 1053242.64 frames. ], batch size: 31, lr: 8.26e-03, grad_scale: 16.0 +2022-11-16 02:13:50,240 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:14:04,239 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2022-11-16 02:14:08,607 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:14:12,740 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:14:15,732 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2308, 1.3815, 1.0187, 0.8895, 1.2325, 1.0075, 0.7508, 1.3966], + device='cuda:3'), covar=tensor([0.0054, 0.0032, 0.0054, 0.0046, 0.0043, 0.0043, 0.0077, 0.0028], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0048, 0.0049, 0.0050, 0.0049, 0.0044, 0.0046, 0.0041], + device='cuda:3'), out_proj_covar=tensor([4.7981e-05, 4.3205e-05, 4.4119e-05, 4.5483e-05, 4.3732e-05, 3.8734e-05, + 4.2322e-05, 3.6540e-05], device='cuda:3') +2022-11-16 02:14:21,175 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:14:22,849 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 02:14:31,680 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:14:35,978 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.700e+01 1.569e+02 1.911e+02 2.505e+02 4.164e+02, threshold=3.822e+02, percent-clipped=1.0 +2022-11-16 02:14:41,297 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:14:50,408 INFO [train.py:876] (3/4) Epoch 10, batch 800, loss[loss=0.1402, simple_loss=0.1426, pruned_loss=0.06893, over 4250.00 frames. ], tot_loss[loss=0.1256, simple_loss=0.1491, pruned_loss=0.05106, over 1067772.99 frames. ], batch size: 181, lr: 8.26e-03, grad_scale: 16.0 +2022-11-16 02:14:57,080 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:15:29,611 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8697, 2.3387, 3.5405, 3.0483, 3.7239, 2.4661, 3.4157, 3.8820], + device='cuda:3'), covar=tensor([0.0597, 0.1716, 0.0869, 0.1662, 0.0507, 0.1812, 0.1043, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0190, 0.0205, 0.0204, 0.0226, 0.0191, 0.0221, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 02:15:38,170 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:15:43,767 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.565e+02 1.875e+02 2.256e+02 4.217e+02, threshold=3.749e+02, percent-clipped=3.0 +2022-11-16 02:15:57,482 INFO [train.py:876] (3/4) Epoch 10, batch 900, loss[loss=0.1797, simple_loss=0.1762, pruned_loss=0.09154, over 5489.00 frames. ], tot_loss[loss=0.1257, simple_loss=0.1497, pruned_loss=0.05083, over 1079625.22 frames. ], batch size: 64, lr: 8.25e-03, grad_scale: 16.0 +2022-11-16 02:16:32,354 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1220, 3.0475, 2.7610, 3.0729, 3.0863, 2.7824, 2.6261, 2.8330], + device='cuda:3'), covar=tensor([0.0308, 0.0556, 0.1479, 0.0469, 0.0508, 0.0527, 0.0835, 0.0584], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0173, 0.0270, 0.0167, 0.0212, 0.0169, 0.0182, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:16:51,651 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 1.944e+02 2.274e+02 2.934e+02 5.796e+02, threshold=4.548e+02, percent-clipped=10.0 +2022-11-16 02:17:05,482 INFO [train.py:876] (3/4) Epoch 10, batch 1000, loss[loss=0.1098, simple_loss=0.1352, pruned_loss=0.04219, over 5592.00 frames. ], tot_loss[loss=0.1283, simple_loss=0.1511, pruned_loss=0.0528, over 1075295.15 frames. ], batch size: 18, lr: 8.25e-03, grad_scale: 16.0 +2022-11-16 02:17:11,579 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 +2022-11-16 02:17:35,898 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:17:39,656 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:17:50,084 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:17:53,489 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:17:58,496 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.268e+01 1.607e+02 1.990e+02 2.689e+02 6.236e+02, threshold=3.979e+02, percent-clipped=3.0 +2022-11-16 02:18:08,457 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:18:13,074 INFO [train.py:876] (3/4) Epoch 10, batch 1100, loss[loss=0.1418, simple_loss=0.1695, pruned_loss=0.05704, over 5707.00 frames. ], tot_loss[loss=0.127, simple_loss=0.1507, pruned_loss=0.05168, over 1082507.12 frames. ], batch size: 34, lr: 8.24e-03, grad_scale: 16.0 +2022-11-16 02:18:35,123 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:18:42,294 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1685, 4.6695, 4.2260, 4.6533, 4.6473, 3.9742, 4.3185, 4.0171], + device='cuda:3'), covar=tensor([0.0324, 0.0367, 0.1081, 0.0395, 0.0328, 0.0450, 0.0440, 0.0397], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0170, 0.0267, 0.0165, 0.0211, 0.0168, 0.0181, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:18:56,852 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:19:05,748 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.897e+01 1.663e+02 2.010e+02 2.428e+02 5.298e+02, threshold=4.020e+02, percent-clipped=1.0 +2022-11-16 02:19:17,629 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9998, 4.8321, 5.0389, 5.1912, 4.8549, 4.8048, 5.6245, 5.1354], + device='cuda:3'), covar=tensor([0.0443, 0.1132, 0.0411, 0.1014, 0.0535, 0.0231, 0.0769, 0.0487], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0106, 0.0090, 0.0115, 0.0084, 0.0074, 0.0141, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:19:20,778 INFO [train.py:876] (3/4) Epoch 10, batch 1200, loss[loss=0.07585, simple_loss=0.1155, pruned_loss=0.01811, over 5700.00 frames. ], tot_loss[loss=0.125, simple_loss=0.149, pruned_loss=0.05053, over 1085317.89 frames. ], batch size: 17, lr: 8.23e-03, grad_scale: 16.0 +2022-11-16 02:19:29,532 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 +2022-11-16 02:19:38,964 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1231, 5.0182, 3.5590, 2.2577, 4.6563, 1.8923, 4.6382, 2.5102], + device='cuda:3'), covar=tensor([0.1022, 0.0102, 0.0639, 0.1944, 0.0129, 0.1709, 0.0143, 0.1375], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0103, 0.0114, 0.0112, 0.0102, 0.0122, 0.0099, 0.0112], + 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') +2022-11-16 02:20:13,037 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.536e+02 1.899e+02 2.389e+02 5.504e+02, threshold=3.797e+02, percent-clipped=2.0 +2022-11-16 02:20:15,160 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:20:27,601 INFO [train.py:876] (3/4) Epoch 10, batch 1300, loss[loss=0.1148, simple_loss=0.1479, pruned_loss=0.04089, over 5572.00 frames. ], tot_loss[loss=0.1244, simple_loss=0.1486, pruned_loss=0.0501, over 1086161.42 frames. ], batch size: 15, lr: 8.23e-03, grad_scale: 16.0 +2022-11-16 02:20:52,176 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1084, 2.1084, 2.7776, 2.5585, 2.6335, 2.0693, 2.6971, 3.0564], + device='cuda:3'), covar=tensor([0.0768, 0.1427, 0.0877, 0.1286, 0.0816, 0.1454, 0.0981, 0.0937], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0187, 0.0206, 0.0205, 0.0225, 0.0190, 0.0220, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 02:20:56,020 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:21:01,747 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:21:04,992 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8845, 1.9158, 1.7236, 1.4438, 1.5307, 1.5902, 1.1727, 2.0866], + device='cuda:3'), covar=tensor([0.0042, 0.0054, 0.0036, 0.0047, 0.0040, 0.0027, 0.0038, 0.0030], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0047, 0.0050, 0.0050, 0.0049, 0.0044, 0.0047, 0.0042], + device='cuda:3'), out_proj_covar=tensor([4.7724e-05, 4.2891e-05, 4.4496e-05, 4.4707e-05, 4.3831e-05, 3.8077e-05, + 4.3249e-05, 3.6727e-05], device='cuda:3') +2022-11-16 02:21:13,054 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:21:20,618 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.546e+02 1.789e+02 2.414e+02 5.527e+02, threshold=3.579e+02, percent-clipped=2.0 +2022-11-16 02:21:33,580 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:21:34,166 INFO [train.py:876] (3/4) Epoch 10, batch 1400, loss[loss=0.1641, simple_loss=0.1661, pruned_loss=0.08104, over 5377.00 frames. ], tot_loss[loss=0.124, simple_loss=0.1486, pruned_loss=0.04975, over 1088763.44 frames. ], batch size: 70, lr: 8.22e-03, grad_scale: 16.0 +2022-11-16 02:21:44,991 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:21:53,526 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:21:56,511 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2022-11-16 02:22:11,116 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7513, 2.1617, 3.2652, 2.8188, 3.4322, 2.2774, 3.0766, 3.6844], + device='cuda:3'), covar=tensor([0.0588, 0.1677, 0.0852, 0.1556, 0.0675, 0.1598, 0.1102, 0.0787], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0189, 0.0207, 0.0207, 0.0228, 0.0192, 0.0222, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 02:22:18,750 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:22:27,873 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.435e+01 1.629e+02 1.952e+02 2.381e+02 3.716e+02, threshold=3.904e+02, percent-clipped=1.0 +2022-11-16 02:22:41,671 INFO [train.py:876] (3/4) Epoch 10, batch 1500, loss[loss=0.123, simple_loss=0.1598, pruned_loss=0.04312, over 5589.00 frames. ], tot_loss[loss=0.124, simple_loss=0.1485, pruned_loss=0.04976, over 1085346.43 frames. ], batch size: 22, lr: 8.21e-03, grad_scale: 16.0 +2022-11-16 02:22:46,961 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:22:50,118 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:23:10,736 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1204, 3.0211, 2.6733, 3.0141, 3.0347, 2.7041, 2.6958, 2.7067], + device='cuda:3'), covar=tensor([0.0304, 0.0619, 0.1613, 0.0528, 0.0568, 0.0542, 0.0896, 0.0666], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0169, 0.0267, 0.0167, 0.0211, 0.0169, 0.0182, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:23:25,985 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6949, 4.7877, 3.2084, 4.4715, 3.5482, 3.2392, 2.6126, 4.0082], + device='cuda:3'), covar=tensor([0.1199, 0.0144, 0.0876, 0.0263, 0.0540, 0.0841, 0.1628, 0.0244], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0138, 0.0160, 0.0144, 0.0174, 0.0171, 0.0167, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 02:23:27,988 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:23:34,295 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.428e+01 1.625e+02 1.914e+02 2.331e+02 6.825e+02, threshold=3.828e+02, percent-clipped=2.0 +2022-11-16 02:23:49,308 INFO [train.py:876] (3/4) Epoch 10, batch 1600, loss[loss=0.1583, simple_loss=0.1752, pruned_loss=0.07072, over 5493.00 frames. ], tot_loss[loss=0.1254, simple_loss=0.1497, pruned_loss=0.05059, over 1090215.61 frames. ], batch size: 53, lr: 8.21e-03, grad_scale: 16.0 +2022-11-16 02:24:11,999 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:24:15,230 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 02:24:41,917 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.648e+02 1.940e+02 2.386e+02 4.578e+02, threshold=3.880e+02, percent-clipped=4.0 +2022-11-16 02:24:53,237 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:24:56,728 INFO [train.py:876] (3/4) Epoch 10, batch 1700, loss[loss=0.1224, simple_loss=0.1568, pruned_loss=0.04394, over 5751.00 frames. ], tot_loss[loss=0.1251, simple_loss=0.1488, pruned_loss=0.05073, over 1080897.75 frames. ], batch size: 31, lr: 8.20e-03, grad_scale: 16.0 +2022-11-16 02:25:15,541 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:25:48,251 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:25:50,860 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.967e+01 1.444e+02 1.809e+02 2.360e+02 5.215e+02, threshold=3.618e+02, percent-clipped=3.0 +2022-11-16 02:26:04,152 INFO [train.py:876] (3/4) Epoch 10, batch 1800, loss[loss=0.1027, simple_loss=0.1385, pruned_loss=0.03344, over 5746.00 frames. ], tot_loss[loss=0.1249, simple_loss=0.1487, pruned_loss=0.05059, over 1085904.36 frames. ], batch size: 16, lr: 8.20e-03, grad_scale: 16.0 +2022-11-16 02:26:47,962 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:26:48,999 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2022-11-16 02:26:58,161 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.612e+02 1.959e+02 2.585e+02 8.694e+02, threshold=3.917e+02, percent-clipped=8.0 +2022-11-16 02:27:11,095 INFO [train.py:876] (3/4) Epoch 10, batch 1900, loss[loss=0.1825, simple_loss=0.1835, pruned_loss=0.09076, over 5449.00 frames. ], tot_loss[loss=0.1254, simple_loss=0.149, pruned_loss=0.05095, over 1082229.76 frames. ], batch size: 53, lr: 8.19e-03, grad_scale: 16.0 +2022-11-16 02:27:17,422 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-16 02:27:23,300 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7254, 4.6493, 3.5055, 2.0653, 4.3691, 1.7596, 4.3350, 2.4201], + device='cuda:3'), covar=tensor([0.1268, 0.0129, 0.0563, 0.2156, 0.0170, 0.1849, 0.0167, 0.1560], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0113, 0.0114, 0.0103, 0.0122, 0.0100, 0.0111], + 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') +2022-11-16 02:27:30,712 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6529, 3.4136, 3.6307, 3.4528, 3.6177, 3.3715, 1.3112, 3.7548], + device='cuda:3'), covar=tensor([0.0250, 0.0413, 0.0296, 0.0289, 0.0319, 0.0398, 0.3072, 0.0269], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0083, 0.0085, 0.0075, 0.0100, 0.0085, 0.0128, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:27:37,294 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:27:43,641 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5362, 4.4873, 4.4332, 4.1951, 4.5791, 4.2752, 1.8962, 4.7927], + device='cuda:3'), covar=tensor([0.0305, 0.0359, 0.0346, 0.0308, 0.0297, 0.0432, 0.3045, 0.0300], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0083, 0.0085, 0.0075, 0.0100, 0.0086, 0.0128, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:27:50,917 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6734, 4.4371, 4.7630, 4.6483, 4.3636, 4.2557, 5.0621, 4.6628], + device='cuda:3'), covar=tensor([0.0394, 0.0856, 0.0344, 0.1178, 0.0466, 0.0280, 0.0678, 0.0661], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0107, 0.0091, 0.0117, 0.0086, 0.0076, 0.0144, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:28:05,971 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.608e+02 1.927e+02 2.290e+02 4.521e+02, threshold=3.854e+02, percent-clipped=3.0 +2022-11-16 02:28:10,019 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:28:12,565 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:28:19,256 INFO [train.py:876] (3/4) Epoch 10, batch 2000, loss[loss=0.123, simple_loss=0.1521, pruned_loss=0.04694, over 5633.00 frames. ], tot_loss[loss=0.125, simple_loss=0.1486, pruned_loss=0.05075, over 1084027.43 frames. ], batch size: 29, lr: 8.18e-03, grad_scale: 16.0 +2022-11-16 02:28:36,389 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7106, 1.5263, 1.5666, 1.0640, 1.4176, 1.7109, 1.3023, 1.1462], + device='cuda:3'), covar=tensor([0.0040, 0.0049, 0.0029, 0.0058, 0.0042, 0.0068, 0.0034, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0023, 0.0031, 0.0026, 0.0025, 0.0028, 0.0028], + device='cuda:3'), out_proj_covar=tensor([2.1782e-05, 2.1724e-05, 2.1378e-05, 2.9872e-05, 2.4420e-05, 2.3589e-05, + 2.6831e-05, 2.8055e-05], device='cuda:3') +2022-11-16 02:29:02,518 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:29:09,005 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 +2022-11-16 02:29:14,258 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.999e+01 1.503e+02 1.762e+02 2.315e+02 5.487e+02, threshold=3.525e+02, percent-clipped=4.0 +2022-11-16 02:29:27,324 INFO [train.py:876] (3/4) Epoch 10, batch 2100, loss[loss=0.1448, simple_loss=0.17, pruned_loss=0.05983, over 5581.00 frames. ], tot_loss[loss=0.1236, simple_loss=0.1476, pruned_loss=0.04978, over 1084826.04 frames. ], batch size: 24, lr: 8.18e-03, grad_scale: 16.0 +2022-11-16 02:29:44,921 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:30:10,758 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:30:21,801 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.895e+01 1.533e+02 1.917e+02 2.461e+02 4.676e+02, threshold=3.833e+02, percent-clipped=3.0 +2022-11-16 02:30:35,269 INFO [train.py:876] (3/4) Epoch 10, batch 2200, loss[loss=0.08535, simple_loss=0.1256, pruned_loss=0.02255, over 5350.00 frames. ], tot_loss[loss=0.1231, simple_loss=0.1477, pruned_loss=0.0492, over 1086374.82 frames. ], batch size: 9, lr: 8.17e-03, grad_scale: 16.0 +2022-11-16 02:30:43,330 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:30:56,592 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:31:03,705 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4435, 4.4331, 3.4015, 1.9206, 4.2282, 1.7923, 4.0448, 2.3509], + device='cuda:3'), covar=tensor([0.1305, 0.0124, 0.0583, 0.1974, 0.0175, 0.1779, 0.0249, 0.1464], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0113, 0.0113, 0.0102, 0.0123, 0.0100, 0.0111], + 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') +2022-11-16 02:31:04,409 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:31:28,761 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.628e+02 1.848e+02 2.167e+02 3.547e+02, threshold=3.696e+02, percent-clipped=0.0 +2022-11-16 02:31:36,136 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:31:37,520 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:31:42,387 INFO [train.py:876] (3/4) Epoch 10, batch 2300, loss[loss=0.1587, simple_loss=0.1682, pruned_loss=0.07463, over 5710.00 frames. ], tot_loss[loss=0.1224, simple_loss=0.1473, pruned_loss=0.04875, over 1092334.52 frames. ], batch size: 31, lr: 8.17e-03, grad_scale: 16.0 +2022-11-16 02:31:45,496 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:31:59,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4757, 3.3626, 3.4251, 3.5514, 3.1959, 3.0894, 3.8474, 3.3979], + device='cuda:3'), covar=tensor([0.0510, 0.0850, 0.0520, 0.1015, 0.0677, 0.0406, 0.0798, 0.0729], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0105, 0.0089, 0.0116, 0.0085, 0.0076, 0.0141, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:32:08,292 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:32:10,491 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.62 vs. limit=5.0 +2022-11-16 02:32:33,335 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2022-11-16 02:32:36,227 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.663e+01 1.643e+02 1.965e+02 2.550e+02 4.357e+02, threshold=3.931e+02, percent-clipped=5.0 +2022-11-16 02:32:50,269 INFO [train.py:876] (3/4) Epoch 10, batch 2400, loss[loss=0.1558, simple_loss=0.173, pruned_loss=0.06927, over 5572.00 frames. ], tot_loss[loss=0.1243, simple_loss=0.1486, pruned_loss=0.05001, over 1084954.07 frames. ], batch size: 46, lr: 8.16e-03, grad_scale: 16.0 +2022-11-16 02:32:58,495 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.43 vs. limit=2.0 +2022-11-16 02:33:03,319 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:33:30,221 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 +2022-11-16 02:33:43,940 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.957e+01 1.533e+02 1.852e+02 2.262e+02 4.255e+02, threshold=3.703e+02, percent-clipped=1.0 +2022-11-16 02:33:55,846 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8029, 2.5933, 2.6760, 2.4262, 2.8358, 2.7072, 2.7195, 2.7853], + device='cuda:3'), covar=tensor([0.0457, 0.0483, 0.0534, 0.0497, 0.0476, 0.0259, 0.0392, 0.0600], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0141, 0.0104, 0.0137, 0.0163, 0.0096, 0.0117, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 02:33:58,901 INFO [train.py:876] (3/4) Epoch 10, batch 2500, loss[loss=0.117, simple_loss=0.1352, pruned_loss=0.04939, over 5700.00 frames. ], tot_loss[loss=0.1235, simple_loss=0.1481, pruned_loss=0.04941, over 1086354.34 frames. ], batch size: 36, lr: 8.15e-03, grad_scale: 16.0 +2022-11-16 02:34:22,091 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:34:29,599 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2022-11-16 02:34:54,543 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.495e+02 1.844e+02 2.229e+02 3.731e+02, threshold=3.687e+02, percent-clipped=1.0 +2022-11-16 02:34:59,156 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:35:03,046 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:35:06,808 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 02:35:07,302 INFO [train.py:876] (3/4) Epoch 10, batch 2600, loss[loss=0.1668, simple_loss=0.1547, pruned_loss=0.08945, over 4049.00 frames. ], tot_loss[loss=0.1239, simple_loss=0.1484, pruned_loss=0.04968, over 1081968.13 frames. ], batch size: 181, lr: 8.15e-03, grad_scale: 16.0 +2022-11-16 02:35:14,422 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 +2022-11-16 02:35:33,571 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:36:01,565 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.584e+02 1.812e+02 2.334e+02 4.676e+02, threshold=3.625e+02, percent-clipped=1.0 +2022-11-16 02:36:14,150 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:36:14,628 INFO [train.py:876] (3/4) Epoch 10, batch 2700, loss[loss=0.08874, simple_loss=0.1202, pruned_loss=0.02862, over 4807.00 frames. ], tot_loss[loss=0.1228, simple_loss=0.1476, pruned_loss=0.04897, over 1080156.86 frames. ], batch size: 5, lr: 8.14e-03, grad_scale: 16.0 +2022-11-16 02:36:28,850 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:37:00,605 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:37:09,358 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.044e+01 1.607e+02 1.920e+02 2.533e+02 3.834e+02, threshold=3.840e+02, percent-clipped=3.0 +2022-11-16 02:37:21,087 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9327, 2.4158, 3.4644, 2.7328, 3.8214, 2.4557, 3.4212, 3.9348], + device='cuda:3'), covar=tensor([0.0588, 0.1724, 0.1040, 0.1715, 0.0578, 0.1540, 0.1173, 0.0779], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0195, 0.0212, 0.0214, 0.0236, 0.0193, 0.0229, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 02:37:22,795 INFO [train.py:876] (3/4) Epoch 10, batch 2800, loss[loss=0.1112, simple_loss=0.1377, pruned_loss=0.04232, over 5495.00 frames. ], tot_loss[loss=0.1207, simple_loss=0.1457, pruned_loss=0.04784, over 1083861.55 frames. ], batch size: 12, lr: 8.14e-03, grad_scale: 16.0 +2022-11-16 02:38:01,739 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8177, 1.2404, 1.3224, 1.0545, 1.7293, 1.4660, 1.2061, 1.2736], + device='cuda:3'), covar=tensor([0.0436, 0.0850, 0.2262, 0.1367, 0.0905, 0.0884, 0.1158, 0.0998], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0020, 0.0013, 0.0017, 0.0014, 0.0013, 0.0018, 0.0013], + device='cuda:3'), out_proj_covar=tensor([6.9271e-05, 9.3956e-05, 7.0454e-05, 8.4114e-05, 7.4479e-05, 6.8921e-05, + 8.7529e-05, 6.8657e-05], device='cuda:3') +2022-11-16 02:38:05,030 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5283, 1.7633, 2.1788, 1.4915, 1.2192, 2.7038, 2.1891, 1.8951], + device='cuda:3'), covar=tensor([0.1191, 0.1571, 0.1048, 0.2671, 0.2697, 0.0628, 0.1665, 0.1726], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0083, 0.0081, 0.0092, 0.0070, 0.0061, 0.0068, 0.0079], + 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') +2022-11-16 02:38:16,586 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.672e+01 1.605e+02 1.834e+02 2.410e+02 3.703e+02, threshold=3.668e+02, percent-clipped=0.0 +2022-11-16 02:38:21,760 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:38:22,371 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:38:25,735 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6048, 2.2050, 2.4420, 3.3901, 3.3216, 2.5324, 2.1249, 3.4252], + device='cuda:3'), covar=tensor([0.0789, 0.2542, 0.2587, 0.2348, 0.1393, 0.3602, 0.2475, 0.0751], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0203, 0.0194, 0.0311, 0.0223, 0.0206, 0.0192, 0.0238], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:38:29,997 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:38:30,509 INFO [train.py:876] (3/4) Epoch 10, batch 2900, loss[loss=0.1562, simple_loss=0.1634, pruned_loss=0.07455, over 4726.00 frames. ], tot_loss[loss=0.1223, simple_loss=0.1468, pruned_loss=0.04892, over 1077006.83 frames. ], batch size: 135, lr: 8.13e-03, grad_scale: 16.0 +2022-11-16 02:38:53,227 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:39:02,044 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:39:04,136 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:39:23,258 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2022-11-16 02:39:23,979 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.585e+02 1.960e+02 2.484e+02 4.720e+02, threshold=3.919e+02, percent-clipped=5.0 +2022-11-16 02:39:33,259 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:39:38,224 INFO [train.py:876] (3/4) Epoch 10, batch 3000, loss[loss=0.09978, simple_loss=0.1228, pruned_loss=0.03838, over 5394.00 frames. ], tot_loss[loss=0.1212, simple_loss=0.146, pruned_loss=0.04816, over 1086610.66 frames. ], batch size: 9, lr: 8.12e-03, grad_scale: 16.0 +2022-11-16 02:39:38,224 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 02:39:50,951 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0684, 4.7287, 4.2032, 3.9151, 2.2133, 4.6073, 2.6654, 3.9406], + device='cuda:3'), covar=tensor([0.0377, 0.0118, 0.0159, 0.0285, 0.0706, 0.0112, 0.0460, 0.0156], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0167, 0.0176, 0.0199, 0.0188, 0.0174, 0.0187, 0.0177], + 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') +2022-11-16 02:39:51,536 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7862, 2.4672, 2.9089, 3.7705, 3.7813, 2.7797, 2.4598, 3.6664], + device='cuda:3'), covar=tensor([0.0654, 0.2767, 0.2333, 0.1857, 0.1224, 0.3133, 0.2219, 0.0636], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0206, 0.0196, 0.0315, 0.0226, 0.0208, 0.0194, 0.0241], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:39:56,212 INFO [train.py:908] (3/4) Epoch 10, validation: loss=0.1681, simple_loss=0.1842, pruned_loss=0.07602, over 1530663.00 frames. +2022-11-16 02:39:56,212 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 02:40:01,220 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 +2022-11-16 02:40:03,625 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:40:14,317 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0774, 4.5226, 3.6792, 4.3734, 4.4220, 3.7242, 4.2745, 3.9817], + device='cuda:3'), covar=tensor([0.0388, 0.0594, 0.2322, 0.0950, 0.0739, 0.0613, 0.0852, 0.0807], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0172, 0.0266, 0.0168, 0.0212, 0.0170, 0.0182, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:40:49,559 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.217e+01 1.619e+02 2.006e+02 2.438e+02 5.141e+02, threshold=4.012e+02, percent-clipped=2.0 +2022-11-16 02:41:02,560 INFO [train.py:876] (3/4) Epoch 10, batch 3100, loss[loss=0.1199, simple_loss=0.1464, pruned_loss=0.04673, over 5595.00 frames. ], tot_loss[loss=0.1239, simple_loss=0.1484, pruned_loss=0.04974, over 1082964.10 frames. ], batch size: 22, lr: 8.12e-03, grad_scale: 16.0 +2022-11-16 02:41:04,742 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9269, 4.5168, 3.3733, 2.0810, 4.2558, 1.7609, 4.0108, 2.3717], + device='cuda:3'), covar=tensor([0.1097, 0.0125, 0.0601, 0.1813, 0.0199, 0.1690, 0.0217, 0.1346], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0113, 0.0113, 0.0101, 0.0122, 0.0099, 0.0111], + 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') +2022-11-16 02:41:22,517 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:41:52,804 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-16 02:41:57,045 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.658e+01 1.525e+02 1.984e+02 2.613e+02 4.758e+02, threshold=3.969e+02, percent-clipped=4.0 +2022-11-16 02:42:02,890 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:42:04,173 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:42:10,477 INFO [train.py:876] (3/4) Epoch 10, batch 3200, loss[loss=0.115, simple_loss=0.1505, pruned_loss=0.03972, over 5756.00 frames. ], tot_loss[loss=0.1243, simple_loss=0.1484, pruned_loss=0.05013, over 1083921.81 frames. ], batch size: 14, lr: 8.11e-03, grad_scale: 16.0 +2022-11-16 02:42:35,038 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:42:35,132 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:43:04,535 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.562e+02 1.856e+02 2.201e+02 3.564e+02, threshold=3.712e+02, percent-clipped=0.0 +2022-11-16 02:43:05,760 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7041, 1.4543, 1.0703, 0.6961, 1.2632, 1.0824, 0.8772, 1.4687], + device='cuda:3'), covar=tensor([0.0061, 0.0037, 0.0052, 0.0057, 0.0052, 0.0054, 0.0071, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0047, 0.0050, 0.0049, 0.0049, 0.0045, 0.0045, 0.0042], + device='cuda:3'), out_proj_covar=tensor([4.7673e-05, 4.2655e-05, 4.4423e-05, 4.4456e-05, 4.3697e-05, 3.8986e-05, + 4.1103e-05, 3.6679e-05], device='cuda:3') +2022-11-16 02:43:08,877 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6578, 4.5426, 3.3999, 2.0940, 4.2438, 1.8480, 4.2890, 2.4475], + device='cuda:3'), covar=tensor([0.1311, 0.0114, 0.0678, 0.2078, 0.0157, 0.1754, 0.0206, 0.1469], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0103, 0.0112, 0.0113, 0.0101, 0.0122, 0.0099, 0.0111], + 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') +2022-11-16 02:43:09,609 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2593, 1.4561, 1.3576, 1.0869, 1.5310, 1.3987, 1.1507, 1.3020], + device='cuda:3'), covar=tensor([0.0043, 0.0044, 0.0037, 0.0039, 0.0034, 0.0033, 0.0041, 0.0049], + device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0047, 0.0050, 0.0049, 0.0049, 0.0045, 0.0045, 0.0042], + device='cuda:3'), out_proj_covar=tensor([4.7725e-05, 4.2624e-05, 4.4401e-05, 4.4414e-05, 4.3671e-05, 3.8982e-05, + 4.1034e-05, 3.6666e-05], device='cuda:3') +2022-11-16 02:43:14,432 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:43:16,474 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:43:18,250 INFO [train.py:876] (3/4) Epoch 10, batch 3300, loss[loss=0.1113, simple_loss=0.1442, pruned_loss=0.03917, over 5596.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1483, pruned_loss=0.05041, over 1082256.70 frames. ], batch size: 24, lr: 8.11e-03, grad_scale: 16.0 +2022-11-16 02:43:22,187 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:43:30,090 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8330, 1.2445, 1.6053, 1.0686, 1.8684, 1.9037, 1.2335, 1.6098], + device='cuda:3'), covar=tensor([0.1015, 0.0623, 0.0852, 0.1293, 0.0549, 0.0513, 0.0653, 0.0693], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0020, 0.0013, 0.0017, 0.0014, 0.0013, 0.0018, 0.0013], + device='cuda:3'), out_proj_covar=tensor([7.0584e-05, 9.5641e-05, 7.0497e-05, 8.5638e-05, 7.5285e-05, 6.8971e-05, + 8.9147e-05, 6.9543e-05], device='cuda:3') +2022-11-16 02:43:46,089 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5946, 4.4621, 4.5054, 4.6602, 4.3008, 3.9563, 5.1227, 4.5422], + device='cuda:3'), covar=tensor([0.0386, 0.0890, 0.0388, 0.1015, 0.0507, 0.0487, 0.0607, 0.0505], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0104, 0.0090, 0.0114, 0.0085, 0.0076, 0.0141, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:43:46,725 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:43:49,469 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3098, 2.1703, 2.5809, 3.3791, 3.3120, 2.7156, 2.2509, 3.3369], + device='cuda:3'), covar=tensor([0.1143, 0.2934, 0.2223, 0.3711, 0.1482, 0.3123, 0.2341, 0.0930], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0204, 0.0195, 0.0316, 0.0228, 0.0211, 0.0194, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:44:12,127 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.117e+01 1.492e+02 1.907e+02 2.389e+02 5.299e+02, threshold=3.813e+02, percent-clipped=1.0 +2022-11-16 02:44:22,871 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1777, 2.1553, 2.4383, 3.4597, 3.2558, 2.6224, 2.1019, 3.4315], + device='cuda:3'), covar=tensor([0.0992, 0.2508, 0.2412, 0.1697, 0.1259, 0.3048, 0.2323, 0.0963], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0205, 0.0195, 0.0317, 0.0230, 0.0211, 0.0195, 0.0244], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:44:22,969 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0 +2022-11-16 02:44:25,991 INFO [train.py:876] (3/4) Epoch 10, batch 3400, loss[loss=0.1133, simple_loss=0.1467, pruned_loss=0.03993, over 5734.00 frames. ], tot_loss[loss=0.123, simple_loss=0.148, pruned_loss=0.04901, over 1082947.72 frames. ], batch size: 14, lr: 8.10e-03, grad_scale: 16.0 +2022-11-16 02:44:57,549 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4663, 5.1969, 3.7043, 2.3463, 4.8268, 2.2469, 4.8457, 2.8622], + device='cuda:3'), covar=tensor([0.1091, 0.0143, 0.0638, 0.2169, 0.0179, 0.1839, 0.0230, 0.1495], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0105, 0.0113, 0.0114, 0.0101, 0.0124, 0.0100, 0.0112], + 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') +2022-11-16 02:45:20,578 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.200e+01 1.439e+02 1.799e+02 2.243e+02 3.372e+02, threshold=3.599e+02, percent-clipped=0.0 +2022-11-16 02:45:24,639 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:45:27,988 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5497, 3.2510, 3.3185, 3.2248, 2.0714, 3.3862, 2.1975, 2.9177], + device='cuda:3'), covar=tensor([0.0321, 0.0169, 0.0170, 0.0225, 0.0436, 0.0149, 0.0413, 0.0146], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0162, 0.0172, 0.0193, 0.0184, 0.0171, 0.0183, 0.0174], + 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') +2022-11-16 02:45:34,200 INFO [train.py:876] (3/4) Epoch 10, batch 3500, loss[loss=0.1247, simple_loss=0.158, pruned_loss=0.04574, over 5519.00 frames. ], tot_loss[loss=0.1232, simple_loss=0.1479, pruned_loss=0.04922, over 1083853.41 frames. ], batch size: 17, lr: 8.10e-03, grad_scale: 16.0 +2022-11-16 02:46:01,685 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 +2022-11-16 02:46:11,895 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 +2022-11-16 02:46:19,021 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3987, 4.4780, 4.3333, 4.1486, 4.4817, 4.4151, 1.6754, 4.5531], + device='cuda:3'), covar=tensor([0.0252, 0.0308, 0.0241, 0.0439, 0.0245, 0.0312, 0.3156, 0.0252], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0086, 0.0085, 0.0078, 0.0102, 0.0088, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:46:28,137 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.544e+01 1.634e+02 1.923e+02 2.372e+02 5.246e+02, threshold=3.846e+02, percent-clipped=2.0 +2022-11-16 02:46:36,716 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:46:38,115 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:46:41,872 INFO [train.py:876] (3/4) Epoch 10, batch 3600, loss[loss=0.08115, simple_loss=0.1069, pruned_loss=0.02768, over 5469.00 frames. ], tot_loss[loss=0.1234, simple_loss=0.1476, pruned_loss=0.04962, over 1080391.69 frames. ], batch size: 10, lr: 8.09e-03, grad_scale: 16.0 +2022-11-16 02:46:46,006 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:46:58,745 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2022-11-16 02:47:11,811 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9028, 3.8582, 3.7844, 3.6181, 3.8223, 3.7659, 1.5789, 3.9742], + device='cuda:3'), covar=tensor([0.0284, 0.0330, 0.0364, 0.0391, 0.0329, 0.0382, 0.3315, 0.0297], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0085, 0.0085, 0.0078, 0.0101, 0.0087, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:47:18,342 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:47:19,109 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:47:35,520 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.617e+01 1.458e+02 1.877e+02 2.187e+02 3.990e+02, threshold=3.754e+02, percent-clipped=1.0 +2022-11-16 02:47:49,300 INFO [train.py:876] (3/4) Epoch 10, batch 3700, loss[loss=0.1118, simple_loss=0.1442, pruned_loss=0.0397, over 5568.00 frames. ], tot_loss[loss=0.1242, simple_loss=0.1486, pruned_loss=0.04989, over 1083691.28 frames. ], batch size: 30, lr: 8.08e-03, grad_scale: 32.0 +2022-11-16 02:47:58,267 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:02,142 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:32,196 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6585, 1.5339, 1.5624, 1.2992, 1.4458, 1.5620, 1.2521, 1.0516], + device='cuda:3'), covar=tensor([0.0026, 0.0044, 0.0029, 0.0048, 0.0039, 0.0066, 0.0035, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0023, 0.0030, 0.0026, 0.0024, 0.0028, 0.0028], + device='cuda:3'), out_proj_covar=tensor([2.1435e-05, 2.0944e-05, 2.0643e-05, 2.9640e-05, 2.4425e-05, 2.2994e-05, + 2.7424e-05, 2.7463e-05], device='cuda:3') +2022-11-16 02:48:40,045 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:41,329 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:43,923 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:44,355 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.159e+01 1.529e+02 2.089e+02 2.315e+02 4.555e+02, threshold=4.177e+02, percent-clipped=1.0 +2022-11-16 02:48:47,073 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:48:54,973 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:48:57,844 INFO [train.py:876] (3/4) Epoch 10, batch 3800, loss[loss=0.08949, simple_loss=0.1092, pruned_loss=0.03491, over 5208.00 frames. ], tot_loss[loss=0.1229, simple_loss=0.1476, pruned_loss=0.04916, over 1083410.84 frames. ], batch size: 8, lr: 8.08e-03, grad_scale: 16.0 +2022-11-16 02:49:14,249 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:49:20,037 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:49:22,752 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:49:25,918 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3970, 3.8099, 2.9076, 1.9101, 3.6279, 1.5267, 3.5960, 1.9539], + device='cuda:3'), covar=tensor([0.1357, 0.0185, 0.0903, 0.2038, 0.0209, 0.1999, 0.0260, 0.1732], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0105, 0.0113, 0.0115, 0.0103, 0.0124, 0.0100, 0.0112], + 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') +2022-11-16 02:49:36,639 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:49:45,845 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:49:50,795 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8853, 1.9532, 2.3626, 3.1037, 3.0654, 2.3477, 1.9651, 3.1927], + device='cuda:3'), covar=tensor([0.1409, 0.3322, 0.2441, 0.1859, 0.1363, 0.3102, 0.2292, 0.0764], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0202, 0.0191, 0.0313, 0.0227, 0.0208, 0.0191, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:49:51,380 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1463, 3.6371, 3.2652, 3.5811, 3.6334, 3.1291, 3.2390, 3.2789], + device='cuda:3'), covar=tensor([0.1111, 0.0439, 0.1269, 0.0441, 0.0404, 0.0487, 0.0634, 0.0500], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0170, 0.0265, 0.0166, 0.0210, 0.0170, 0.0181, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 02:49:53,219 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.561e+02 1.843e+02 2.119e+02 2.947e+02, threshold=3.686e+02, percent-clipped=0.0 +2022-11-16 02:49:55,287 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 02:49:59,882 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:05,017 INFO [train.py:876] (3/4) Epoch 10, batch 3900, loss[loss=0.1074, simple_loss=0.1375, pruned_loss=0.03862, over 5736.00 frames. ], tot_loss[loss=0.1238, simple_loss=0.1483, pruned_loss=0.04968, over 1081651.64 frames. ], batch size: 20, lr: 8.07e-03, grad_scale: 8.0 +2022-11-16 02:50:18,588 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:19,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1132, 1.6781, 1.9930, 2.1212, 2.4413, 1.8499, 1.4777, 2.2084], + device='cuda:3'), covar=tensor([0.1864, 0.2401, 0.1905, 0.1197, 0.1189, 0.2680, 0.2372, 0.1826], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0202, 0.0192, 0.0314, 0.0227, 0.0208, 0.0191, 0.0241], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:50:27,389 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:32,358 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-16 02:50:32,636 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:36,067 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:39,286 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:50:46,665 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4840, 1.3388, 1.5513, 1.3769, 1.7612, 1.4386, 1.0912, 1.6413], + device='cuda:3'), covar=tensor([0.1207, 0.1335, 0.1527, 0.1012, 0.1391, 0.1429, 0.2367, 0.1799], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0201, 0.0190, 0.0312, 0.0226, 0.0207, 0.0190, 0.0240], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:51:00,416 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:51:01,538 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.631e+02 1.973e+02 2.599e+02 5.372e+02, threshold=3.946e+02, percent-clipped=3.0 +2022-11-16 02:51:13,626 INFO [train.py:876] (3/4) Epoch 10, batch 4000, loss[loss=0.1082, simple_loss=0.1424, pruned_loss=0.03694, over 5576.00 frames. ], tot_loss[loss=0.1234, simple_loss=0.1481, pruned_loss=0.04939, over 1084792.33 frames. ], batch size: 22, lr: 8.07e-03, grad_scale: 8.0 +2022-11-16 02:51:17,615 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:51:59,805 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:52:03,559 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:52:08,748 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.549e+01 1.604e+02 1.884e+02 2.288e+02 4.952e+02, threshold=3.768e+02, percent-clipped=2.0 +2022-11-16 02:52:20,881 INFO [train.py:876] (3/4) Epoch 10, batch 4100, loss[loss=0.126, simple_loss=0.1628, pruned_loss=0.04456, over 5527.00 frames. ], tot_loss[loss=0.122, simple_loss=0.1467, pruned_loss=0.04862, over 1087038.14 frames. ], batch size: 21, lr: 8.06e-03, grad_scale: 8.0 +2022-11-16 02:52:42,303 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:52:56,108 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:53:00,708 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9008, 4.9129, 3.7068, 2.1648, 4.5235, 2.0551, 4.2339, 2.6082], + device='cuda:3'), covar=tensor([0.1226, 0.0110, 0.0357, 0.2267, 0.0180, 0.1470, 0.0245, 0.1413], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0103, 0.0111, 0.0113, 0.0101, 0.0121, 0.0098, 0.0110], + 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') +2022-11-16 02:53:14,382 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:53:15,586 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 1.609e+02 2.045e+02 2.634e+02 5.309e+02, threshold=4.090e+02, percent-clipped=7.0 +2022-11-16 02:53:28,115 INFO [train.py:876] (3/4) Epoch 10, batch 4200, loss[loss=0.1072, simple_loss=0.1358, pruned_loss=0.03927, over 5697.00 frames. ], tot_loss[loss=0.1223, simple_loss=0.147, pruned_loss=0.04879, over 1082267.93 frames. ], batch size: 19, lr: 8.05e-03, grad_scale: 8.0 +2022-11-16 02:53:37,298 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3019, 1.3855, 1.0962, 0.6027, 1.5233, 1.1838, 0.6224, 1.4611], + device='cuda:3'), covar=tensor([0.0054, 0.0033, 0.0045, 0.0050, 0.0041, 0.0043, 0.0088, 0.0037], + device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0050, 0.0052, 0.0052, 0.0052, 0.0046, 0.0048, 0.0044], + device='cuda:3'), out_proj_covar=tensor([4.9246e-05, 4.4874e-05, 4.5942e-05, 4.6597e-05, 4.6099e-05, 4.0695e-05, + 4.3034e-05, 3.9051e-05], device='cuda:3') +2022-11-16 02:53:45,417 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:53:46,617 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:54:01,870 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:54:12,615 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0043, 4.3146, 3.9742, 3.7339, 2.2908, 4.2494, 2.4816, 3.5572], + device='cuda:3'), covar=tensor([0.0407, 0.0118, 0.0166, 0.0298, 0.0579, 0.0143, 0.0489, 0.0145], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0166, 0.0175, 0.0195, 0.0186, 0.0173, 0.0185, 0.0176], + 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') +2022-11-16 02:54:18,350 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:54:23,253 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.585e+02 1.904e+02 2.422e+02 5.734e+02, threshold=3.808e+02, percent-clipped=2.0 +2022-11-16 02:54:26,136 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 02:54:29,287 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8927, 2.8940, 2.5050, 2.8574, 2.9025, 2.5994, 2.4687, 2.6052], + device='cuda:3'), covar=tensor([0.0309, 0.0562, 0.1717, 0.0545, 0.0611, 0.0549, 0.1027, 0.0725], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0167, 0.0262, 0.0165, 0.0209, 0.0168, 0.0178, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 02:54:34,708 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:54:36,044 INFO [train.py:876] (3/4) Epoch 10, batch 4300, loss[loss=0.123, simple_loss=0.1481, pruned_loss=0.04897, over 5608.00 frames. ], tot_loss[loss=0.1233, simple_loss=0.1478, pruned_loss=0.04943, over 1078945.43 frames. ], batch size: 18, lr: 8.05e-03, grad_scale: 8.0 +2022-11-16 02:54:37,125 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:55:23,539 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:55:27,402 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:55:31,844 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.497e+01 1.489e+02 1.786e+02 2.151e+02 5.165e+02, threshold=3.571e+02, percent-clipped=4.0 +2022-11-16 02:55:44,013 INFO [train.py:876] (3/4) Epoch 10, batch 4400, loss[loss=0.1243, simple_loss=0.1534, pruned_loss=0.04761, over 5641.00 frames. ], tot_loss[loss=0.1227, simple_loss=0.1472, pruned_loss=0.04913, over 1080404.76 frames. ], batch size: 38, lr: 8.04e-03, grad_scale: 8.0 +2022-11-16 02:55:56,380 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:55:57,375 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-16 02:56:00,381 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:56:06,507 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:56:19,536 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 02:56:29,994 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7552, 2.4649, 2.7863, 3.7407, 3.8007, 2.8707, 2.4208, 3.6781], + device='cuda:3'), covar=tensor([0.0506, 0.2937, 0.1808, 0.2732, 0.0983, 0.2730, 0.1929, 0.0653], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0198, 0.0188, 0.0311, 0.0224, 0.0208, 0.0188, 0.0240], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 02:56:38,888 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:56:38,968 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:56:40,107 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.783e+01 1.456e+02 1.859e+02 2.401e+02 3.905e+02, threshold=3.718e+02, percent-clipped=1.0 +2022-11-16 02:56:52,007 INFO [train.py:876] (3/4) Epoch 10, batch 4500, loss[loss=0.09246, simple_loss=0.1284, pruned_loss=0.02824, over 5463.00 frames. ], tot_loss[loss=0.1241, simple_loss=0.1485, pruned_loss=0.04992, over 1082544.33 frames. ], batch size: 11, lr: 8.04e-03, grad_scale: 8.0 +2022-11-16 02:56:52,042 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 02:57:10,771 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:57:11,363 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:57:11,720 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2022-11-16 02:57:32,770 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.99 vs. limit=5.0 +2022-11-16 02:57:47,880 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:57:47,958 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:57:52,471 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 02:57:52,995 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.972e+01 1.578e+02 1.852e+02 2.258e+02 4.002e+02, threshold=3.705e+02, percent-clipped=2.0 +2022-11-16 02:57:55,235 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7138, 4.1168, 3.6650, 3.4908, 2.1292, 3.8344, 2.2818, 3.1622], + device='cuda:3'), covar=tensor([0.0396, 0.0132, 0.0184, 0.0362, 0.0603, 0.0178, 0.0489, 0.0192], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0165, 0.0171, 0.0193, 0.0183, 0.0170, 0.0183, 0.0174], + 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') +2022-11-16 02:58:03,834 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 02:58:05,288 INFO [train.py:876] (3/4) Epoch 10, batch 4600, loss[loss=0.1346, simple_loss=0.1564, pruned_loss=0.05638, over 5556.00 frames. ], tot_loss[loss=0.1213, simple_loss=0.1469, pruned_loss=0.04784, over 1089625.80 frames. ], batch size: 43, lr: 8.03e-03, grad_scale: 8.0 +2022-11-16 02:58:06,027 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:58:20,691 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:58:29,050 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2022-11-16 02:58:38,728 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:58:46,571 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:58:52,769 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8056, 3.8751, 3.7369, 3.6494, 3.8139, 3.6550, 1.5865, 3.9535], + device='cuda:3'), covar=tensor([0.0299, 0.0330, 0.0334, 0.0343, 0.0346, 0.0419, 0.2874, 0.0318], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0084, 0.0084, 0.0077, 0.0101, 0.0087, 0.0129, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 02:59:00,300 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 1.693e+02 2.084e+02 2.528e+02 4.221e+02, threshold=4.168e+02, percent-clipped=4.0 +2022-11-16 02:59:13,004 INFO [train.py:876] (3/4) Epoch 10, batch 4700, loss[loss=0.1184, simple_loss=0.1441, pruned_loss=0.04637, over 5575.00 frames. ], tot_loss[loss=0.1191, simple_loss=0.1451, pruned_loss=0.04655, over 1091777.16 frames. ], batch size: 43, lr: 8.03e-03, grad_scale: 8.0 +2022-11-16 02:59:14,508 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9417, 1.2343, 1.5100, 1.1255, 1.6086, 1.8960, 1.2705, 1.5762], + device='cuda:3'), covar=tensor([0.0817, 0.0571, 0.0312, 0.0787, 0.1621, 0.0536, 0.0713, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0020, 0.0014, 0.0018, 0.0014, 0.0013, 0.0019, 0.0013], + device='cuda:3'), out_proj_covar=tensor([7.0757e-05, 9.6815e-05, 7.2856e-05, 8.6594e-05, 7.4792e-05, 6.9999e-05, + 9.1748e-05, 7.0149e-05], device='cuda:3') +2022-11-16 02:59:25,095 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 02:59:27,628 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 02:59:35,450 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 +2022-11-16 02:59:36,524 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2022-11-16 02:59:45,542 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9619, 2.3570, 3.4605, 2.8139, 3.6503, 2.3765, 3.2326, 3.8000], + device='cuda:3'), covar=tensor([0.0616, 0.1652, 0.0810, 0.1686, 0.0701, 0.1597, 0.1253, 0.0923], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0197, 0.0214, 0.0212, 0.0235, 0.0195, 0.0229, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:00:08,313 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.549e+01 1.608e+02 1.952e+02 2.269e+02 5.385e+02, threshold=3.904e+02, percent-clipped=1.0 +2022-11-16 03:00:20,902 INFO [train.py:876] (3/4) Epoch 10, batch 4800, loss[loss=0.1282, simple_loss=0.1431, pruned_loss=0.05666, over 5666.00 frames. ], tot_loss[loss=0.1185, simple_loss=0.1447, pruned_loss=0.04618, over 1088435.21 frames. ], batch size: 34, lr: 8.02e-03, grad_scale: 8.0 +2022-11-16 03:00:26,866 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1900, 1.3461, 1.2492, 1.2429, 1.4648, 1.4349, 1.1782, 1.4278], + device='cuda:3'), covar=tensor([0.0051, 0.0041, 0.0042, 0.0053, 0.0052, 0.0043, 0.0050, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0050, 0.0052, 0.0053, 0.0053, 0.0047, 0.0049, 0.0045], + device='cuda:3'), out_proj_covar=tensor([4.9845e-05, 4.5256e-05, 4.6618e-05, 4.7426e-05, 4.7234e-05, 4.0844e-05, + 4.4300e-05, 3.9667e-05], device='cuda:3') +2022-11-16 03:00:28,834 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:01:10,000 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:01:11,994 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2022-11-16 03:01:15,766 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 03:01:16,224 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 1.701e+02 2.058e+02 2.547e+02 4.780e+02, threshold=4.115e+02, percent-clipped=1.0 +2022-11-16 03:01:20,961 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6705, 5.0188, 3.3471, 4.8330, 3.6734, 3.2901, 2.7812, 4.3016], + device='cuda:3'), covar=tensor([0.1461, 0.0185, 0.0947, 0.0215, 0.0596, 0.0869, 0.1777, 0.0259], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0140, 0.0162, 0.0147, 0.0175, 0.0173, 0.0167, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:01:23,646 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 03:01:28,223 INFO [train.py:876] (3/4) Epoch 10, batch 4900, loss[loss=0.207, simple_loss=0.1986, pruned_loss=0.1077, over 5350.00 frames. ], tot_loss[loss=0.1206, simple_loss=0.1466, pruned_loss=0.04725, over 1094852.44 frames. ], batch size: 70, lr: 8.01e-03, grad_scale: 8.0 +2022-11-16 03:01:47,957 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 03:02:14,123 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1145, 1.5053, 0.9969, 1.0623, 1.2904, 1.5327, 1.6450, 1.4574], + device='cuda:3'), covar=tensor([0.1036, 0.0661, 0.1656, 0.1808, 0.1048, 0.0839, 0.0637, 0.1191], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0184, 0.0162, 0.0187, 0.0178, 0.0198, 0.0167, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:02:24,079 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.613e+02 1.862e+02 2.297e+02 4.157e+02, threshold=3.724e+02, percent-clipped=1.0 +2022-11-16 03:02:24,959 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0059, 3.0919, 3.1957, 1.7188, 3.1433, 3.5416, 3.3153, 3.6331], + device='cuda:3'), covar=tensor([0.2082, 0.1416, 0.0858, 0.2714, 0.0594, 0.0772, 0.0506, 0.0788], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0185, 0.0162, 0.0187, 0.0178, 0.0199, 0.0168, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:02:35,808 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8944, 1.4811, 1.5621, 1.3036, 1.2819, 1.3192, 1.3544, 1.3781], + device='cuda:3'), covar=tensor([0.4457, 0.2157, 0.1725, 0.1830, 0.2573, 0.3279, 0.2402, 0.1172], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0099, 0.0100, 0.0093, 0.0087, 0.0095, 0.0094, 0.0072], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 03:02:36,299 INFO [train.py:876] (3/4) Epoch 10, batch 5000, loss[loss=0.1624, simple_loss=0.1604, pruned_loss=0.08218, over 4151.00 frames. ], tot_loss[loss=0.1212, simple_loss=0.147, pruned_loss=0.04769, over 1084312.41 frames. ], batch size: 181, lr: 8.01e-03, grad_scale: 8.0 +2022-11-16 03:02:48,243 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:03:03,222 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8706, 2.2839, 2.4126, 3.1165, 3.0411, 2.3313, 1.9763, 3.1139], + device='cuda:3'), covar=tensor([0.1512, 0.2885, 0.2023, 0.2138, 0.1085, 0.2839, 0.2149, 0.1232], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0202, 0.0189, 0.0314, 0.0228, 0.0208, 0.0191, 0.0243], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 03:03:11,311 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9265, 2.6478, 2.1009, 1.5607, 2.6192, 1.1528, 2.6406, 1.6198], + device='cuda:3'), covar=tensor([0.1222, 0.0282, 0.0999, 0.1523, 0.0306, 0.1986, 0.0325, 0.1366], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0113, 0.0114, 0.0101, 0.0122, 0.0099, 0.0111], + 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') +2022-11-16 03:03:32,300 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.646e+01 1.626e+02 2.052e+02 2.453e+02 4.423e+02, threshold=4.104e+02, percent-clipped=3.0 +2022-11-16 03:03:44,016 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 03:03:44,158 INFO [train.py:876] (3/4) Epoch 10, batch 5100, loss[loss=0.1212, simple_loss=0.1419, pruned_loss=0.05026, over 5012.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1459, pruned_loss=0.04686, over 1082879.21 frames. ], batch size: 109, lr: 8.00e-03, grad_scale: 8.0 +2022-11-16 03:03:52,161 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 +2022-11-16 03:04:04,479 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6632, 4.0856, 3.8288, 3.3933, 2.3075, 4.0338, 2.2487, 3.3449], + device='cuda:3'), covar=tensor([0.0458, 0.0221, 0.0224, 0.0459, 0.0644, 0.0165, 0.0640, 0.0169], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0168, 0.0175, 0.0197, 0.0186, 0.0172, 0.0185, 0.0177], + 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') +2022-11-16 03:04:30,173 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:04:40,395 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.739e+01 1.539e+02 1.803e+02 2.270e+02 4.392e+02, threshold=3.606e+02, percent-clipped=1.0 +2022-11-16 03:04:52,544 INFO [train.py:876] (3/4) Epoch 10, batch 5200, loss[loss=0.118, simple_loss=0.1463, pruned_loss=0.04488, over 5590.00 frames. ], tot_loss[loss=0.1205, simple_loss=0.1462, pruned_loss=0.0474, over 1081499.42 frames. ], batch size: 18, lr: 8.00e-03, grad_scale: 8.0 +2022-11-16 03:05:05,280 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 +2022-11-16 03:05:19,721 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8649, 3.9773, 3.8507, 3.6600, 3.9646, 3.9675, 1.4240, 4.1236], + device='cuda:3'), covar=tensor([0.0315, 0.0376, 0.0418, 0.0465, 0.0371, 0.0406, 0.3666, 0.0359], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0085, 0.0086, 0.0080, 0.0102, 0.0088, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:05:19,781 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5534, 2.6694, 2.2747, 2.6620, 2.3127, 2.1821, 2.2963, 3.1029], + device='cuda:3'), covar=tensor([0.1190, 0.1137, 0.1883, 0.0970, 0.1496, 0.1655, 0.1525, 0.0998], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0098, 0.0098, 0.0093, 0.0087, 0.0094, 0.0094, 0.0072], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 03:05:42,203 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 +2022-11-16 03:05:47,144 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.458e+01 1.499e+02 1.855e+02 2.184e+02 5.918e+02, threshold=3.711e+02, percent-clipped=4.0 +2022-11-16 03:05:59,634 INFO [train.py:876] (3/4) Epoch 10, batch 5300, loss[loss=0.06691, simple_loss=0.1041, pruned_loss=0.01485, over 5348.00 frames. ], tot_loss[loss=0.1196, simple_loss=0.1455, pruned_loss=0.04688, over 1084509.41 frames. ], batch size: 9, lr: 7.99e-03, grad_scale: 8.0 +2022-11-16 03:06:00,493 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5962, 1.5067, 1.6770, 1.3631, 1.4029, 1.4872, 1.0909, 0.8945], + device='cuda:3'), covar=tensor([0.0023, 0.0030, 0.0026, 0.0039, 0.0052, 0.0040, 0.0039, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0022, 0.0023, 0.0030, 0.0026, 0.0024, 0.0029, 0.0029], + device='cuda:3'), out_proj_covar=tensor([2.1711e-05, 2.1250e-05, 2.0760e-05, 2.9410e-05, 2.4780e-05, 2.2535e-05, + 2.7310e-05, 2.8542e-05], device='cuda:3') +2022-11-16 03:06:11,224 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:06:43,995 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:06:55,260 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.500e+01 1.530e+02 1.938e+02 2.261e+02 4.134e+02, threshold=3.876e+02, percent-clipped=2.0 +2022-11-16 03:07:00,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2022-11-16 03:07:04,537 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2022-11-16 03:07:07,455 INFO [train.py:876] (3/4) Epoch 10, batch 5400, loss[loss=0.08218, simple_loss=0.1271, pruned_loss=0.01864, over 5576.00 frames. ], tot_loss[loss=0.1189, simple_loss=0.1449, pruned_loss=0.04647, over 1082131.47 frames. ], batch size: 16, lr: 7.99e-03, grad_scale: 8.0 +2022-11-16 03:07:46,717 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 +2022-11-16 03:07:52,770 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:08:02,763 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.679e+02 2.107e+02 2.563e+02 5.005e+02, threshold=4.215e+02, percent-clipped=8.0 +2022-11-16 03:08:14,642 INFO [train.py:876] (3/4) Epoch 10, batch 5500, loss[loss=0.1253, simple_loss=0.1528, pruned_loss=0.04889, over 5725.00 frames. ], tot_loss[loss=0.117, simple_loss=0.1436, pruned_loss=0.04523, over 1085787.90 frames. ], batch size: 12, lr: 7.98e-03, grad_scale: 8.0 +2022-11-16 03:08:24,878 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:08:41,950 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1333, 5.3302, 5.1430, 4.8913, 5.2239, 4.9697, 2.4650, 5.4112], + device='cuda:3'), covar=tensor([0.0235, 0.0195, 0.0234, 0.0227, 0.0259, 0.0362, 0.2257, 0.0200], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0086, 0.0086, 0.0079, 0.0102, 0.0088, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:08:43,996 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:09:04,767 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4830, 2.1811, 3.1796, 2.7945, 2.9468, 2.1418, 2.9424, 3.4035], + device='cuda:3'), covar=tensor([0.0694, 0.1489, 0.0845, 0.1424, 0.0910, 0.1621, 0.1035, 0.0864], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0190, 0.0207, 0.0206, 0.0231, 0.0190, 0.0222, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:09:10,363 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:09:10,842 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.597e+02 2.081e+02 2.610e+02 5.785e+02, threshold=4.161e+02, percent-clipped=1.0 +2022-11-16 03:09:11,034 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7048, 2.1842, 2.8894, 1.5280, 1.0045, 3.2745, 2.5470, 2.1618], + device='cuda:3'), covar=tensor([0.0744, 0.1303, 0.0511, 0.2928, 0.2995, 0.1666, 0.1303, 0.1223], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0082, 0.0083, 0.0094, 0.0069, 0.0060, 0.0068, 0.0080], + 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') +2022-11-16 03:09:22,360 INFO [train.py:876] (3/4) Epoch 10, batch 5600, loss[loss=0.1239, simple_loss=0.157, pruned_loss=0.04538, over 5755.00 frames. ], tot_loss[loss=0.1185, simple_loss=0.1451, pruned_loss=0.04595, over 1086839.39 frames. ], batch size: 21, lr: 7.98e-03, grad_scale: 8.0 +2022-11-16 03:09:25,142 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:09:51,322 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:09:55,850 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1399, 3.9604, 2.7217, 3.6626, 2.9558, 2.7637, 2.1481, 3.3429], + device='cuda:3'), covar=tensor([0.1573, 0.0255, 0.1106, 0.0387, 0.0954, 0.1184, 0.2026, 0.0445], + device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0141, 0.0163, 0.0148, 0.0177, 0.0173, 0.0167, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:09:57,910 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6228, 3.9877, 3.7612, 3.5092, 2.1008, 3.9280, 2.2268, 3.2410], + device='cuda:3'), covar=tensor([0.0428, 0.0175, 0.0175, 0.0291, 0.0607, 0.0160, 0.0546, 0.0181], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0167, 0.0175, 0.0197, 0.0188, 0.0171, 0.0185, 0.0177], + 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') +2022-11-16 03:10:02,094 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.98 vs. limit=5.0 +2022-11-16 03:10:18,057 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.341e+01 1.536e+02 1.905e+02 2.360e+02 4.065e+02, threshold=3.810e+02, percent-clipped=0.0 +2022-11-16 03:10:30,666 INFO [train.py:876] (3/4) Epoch 10, batch 5700, loss[loss=0.09712, simple_loss=0.128, pruned_loss=0.03313, over 5569.00 frames. ], tot_loss[loss=0.1188, simple_loss=0.1448, pruned_loss=0.04639, over 1080028.32 frames. ], batch size: 15, lr: 7.97e-03, grad_scale: 8.0 +2022-11-16 03:10:39,002 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-16 03:10:42,074 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5992, 1.0687, 1.5737, 0.8685, 1.5292, 1.4877, 1.2478, 1.5814], + device='cuda:3'), covar=tensor([0.0732, 0.0812, 0.0562, 0.1306, 0.1242, 0.0799, 0.0648, 0.0440], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0021, 0.0014, 0.0018, 0.0014, 0.0013, 0.0019, 0.0014], + device='cuda:3'), out_proj_covar=tensor([7.1701e-05, 9.8928e-05, 7.5209e-05, 8.8589e-05, 7.6659e-05, 7.0844e-05, + 9.3479e-05, 7.2458e-05], device='cuda:3') +2022-11-16 03:11:26,879 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.887e+01 1.504e+02 1.909e+02 2.394e+02 3.838e+02, threshold=3.819e+02, percent-clipped=2.0 +2022-11-16 03:11:38,466 INFO [train.py:876] (3/4) Epoch 10, batch 5800, loss[loss=0.1524, simple_loss=0.1668, pruned_loss=0.06905, over 5687.00 frames. ], tot_loss[loss=0.1188, simple_loss=0.1445, pruned_loss=0.0465, over 1083573.00 frames. ], batch size: 36, lr: 7.96e-03, grad_scale: 4.0 +2022-11-16 03:12:34,045 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.935e+01 1.517e+02 1.924e+02 2.485e+02 4.438e+02, threshold=3.847e+02, percent-clipped=4.0 +2022-11-16 03:12:44,747 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:12:45,333 INFO [train.py:876] (3/4) Epoch 10, batch 5900, loss[loss=0.08261, simple_loss=0.1211, pruned_loss=0.02206, over 5716.00 frames. ], tot_loss[loss=0.1174, simple_loss=0.1435, pruned_loss=0.04561, over 1087980.22 frames. ], batch size: 17, lr: 7.96e-03, grad_scale: 4.0 +2022-11-16 03:13:02,482 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 03:13:10,619 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:13:30,822 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1080, 1.3315, 1.8117, 1.6957, 1.4647, 1.9534, 1.7734, 1.6685], + device='cuda:3'), covar=tensor([0.0049, 0.0085, 0.0065, 0.0046, 0.0072, 0.0128, 0.0035, 0.0046], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0023, 0.0025, 0.0032, 0.0028, 0.0026, 0.0030, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.2612e-05, 2.2117e-05, 2.2369e-05, 3.1023e-05, 2.6157e-05, 2.4259e-05, + 2.9153e-05, 3.0424e-05], device='cuda:3') +2022-11-16 03:13:42,134 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.550e+02 1.857e+02 2.377e+02 5.014e+02, threshold=3.713e+02, percent-clipped=7.0 +2022-11-16 03:13:53,302 INFO [train.py:876] (3/4) Epoch 10, batch 6000, loss[loss=0.1016, simple_loss=0.1278, pruned_loss=0.03773, over 5482.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1428, pruned_loss=0.04516, over 1090752.40 frames. ], batch size: 12, lr: 7.95e-03, grad_scale: 8.0 +2022-11-16 03:13:53,302 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 03:14:00,845 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5052, 0.8435, 1.2765, 0.9327, 1.4558, 1.2072, 0.4729, 1.1166], + device='cuda:3'), covar=tensor([0.0289, 0.0102, 0.0141, 0.0399, 0.0119, 0.0114, 0.1157, 0.0170], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0021, 0.0014, 0.0018, 0.0015, 0.0013, 0.0020, 0.0014], + device='cuda:3'), out_proj_covar=tensor([7.3372e-05, 1.0030e-04, 7.6640e-05, 8.9689e-05, 7.7935e-05, 7.2331e-05, + 9.5427e-05, 7.3835e-05], device='cuda:3') +2022-11-16 03:14:03,324 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3118, 3.9054, 2.9361, 1.8397, 3.6230, 1.5927, 3.3106, 2.4594], + device='cuda:3'), covar=tensor([0.1519, 0.0147, 0.0834, 0.2162, 0.0187, 0.1768, 0.0295, 0.1425], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0113, 0.0114, 0.0101, 0.0122, 0.0099, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:14:08,845 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1743, 1.7066, 1.8601, 1.4300, 1.3683, 1.9353, 1.2269, 1.3945], + device='cuda:3'), covar=tensor([0.0019, 0.0050, 0.0029, 0.0053, 0.0120, 0.0053, 0.0043, 0.0050], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0025, 0.0032, 0.0028, 0.0026, 0.0031, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.2826e-05, 2.2276e-05, 2.2625e-05, 3.1359e-05, 2.6302e-05, 2.4512e-05, + 2.9451e-05, 3.0802e-05], device='cuda:3') +2022-11-16 03:14:11,198 INFO [train.py:908] (3/4) Epoch 10, validation: loss=0.1673, simple_loss=0.1835, pruned_loss=0.0755, over 1530663.00 frames. +2022-11-16 03:14:11,199 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 03:15:07,799 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.614e+01 1.652e+02 1.842e+02 2.318e+02 4.092e+02, threshold=3.683e+02, percent-clipped=3.0 +2022-11-16 03:15:18,741 INFO [train.py:876] (3/4) Epoch 10, batch 6100, loss[loss=0.1425, simple_loss=0.1533, pruned_loss=0.06587, over 5474.00 frames. ], tot_loss[loss=0.1201, simple_loss=0.145, pruned_loss=0.04754, over 1084447.89 frames. ], batch size: 64, lr: 7.95e-03, grad_scale: 8.0 +2022-11-16 03:16:04,748 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:16:16,026 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.003e+01 1.497e+02 1.833e+02 2.234e+02 4.359e+02, threshold=3.667e+02, percent-clipped=3.0 +2022-11-16 03:16:26,690 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:16:27,197 INFO [train.py:876] (3/4) Epoch 10, batch 6200, loss[loss=0.1047, simple_loss=0.1345, pruned_loss=0.03745, over 5562.00 frames. ], tot_loss[loss=0.1187, simple_loss=0.1439, pruned_loss=0.04673, over 1083172.53 frames. ], batch size: 21, lr: 7.94e-03, grad_scale: 8.0 +2022-11-16 03:16:33,789 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5322, 5.1262, 4.6254, 5.0575, 5.0841, 4.2506, 4.6008, 4.3564], + device='cuda:3'), covar=tensor([0.0287, 0.0452, 0.1565, 0.0329, 0.0397, 0.0444, 0.0322, 0.0443], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0173, 0.0269, 0.0173, 0.0214, 0.0173, 0.0185, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:16:43,907 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9406, 2.9486, 3.0824, 2.8119, 3.0580, 2.9033, 1.1826, 3.1261], + device='cuda:3'), covar=tensor([0.0320, 0.0321, 0.0269, 0.0312, 0.0329, 0.0397, 0.2808, 0.0314], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0088, 0.0087, 0.0080, 0.0103, 0.0089, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:16:45,869 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:16:52,351 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:16:53,142 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.75 vs. limit=5.0 +2022-11-16 03:16:58,955 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:17:06,337 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2022-11-16 03:17:23,599 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.569e+02 1.914e+02 2.269e+02 4.591e+02, threshold=3.828e+02, percent-clipped=5.0 +2022-11-16 03:17:24,937 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:17:33,693 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.22 vs. limit=2.0 +2022-11-16 03:17:34,640 INFO [train.py:876] (3/4) Epoch 10, batch 6300, loss[loss=0.1494, simple_loss=0.1707, pruned_loss=0.06408, over 5586.00 frames. ], tot_loss[loss=0.1185, simple_loss=0.1441, pruned_loss=0.04648, over 1082369.96 frames. ], batch size: 25, lr: 7.94e-03, grad_scale: 8.0 +2022-11-16 03:17:41,051 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2022-11-16 03:18:07,825 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1874, 4.2454, 2.8928, 3.9478, 3.3048, 2.7594, 2.1551, 3.6634], + device='cuda:3'), covar=tensor([0.1543, 0.0211, 0.1062, 0.0355, 0.0654, 0.1047, 0.2106, 0.0362], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0137, 0.0161, 0.0144, 0.0173, 0.0169, 0.0164, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 03:18:30,420 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.579e+02 1.935e+02 2.633e+02 4.632e+02, threshold=3.870e+02, percent-clipped=2.0 +2022-11-16 03:18:42,503 INFO [train.py:876] (3/4) Epoch 10, batch 6400, loss[loss=0.1453, simple_loss=0.1661, pruned_loss=0.06232, over 5259.00 frames. ], tot_loss[loss=0.1191, simple_loss=0.145, pruned_loss=0.04659, over 1087166.10 frames. ], batch size: 79, lr: 7.93e-03, grad_scale: 8.0 +2022-11-16 03:19:09,846 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 +2022-11-16 03:19:29,515 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:19:37,764 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.943e+01 1.534e+02 1.967e+02 2.415e+02 6.448e+02, threshold=3.935e+02, percent-clipped=1.0 +2022-11-16 03:19:43,799 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8289, 2.1641, 3.2899, 2.9266, 3.5204, 2.1200, 2.8812, 3.7008], + device='cuda:3'), covar=tensor([0.0617, 0.1880, 0.0786, 0.1378, 0.0636, 0.1764, 0.1431, 0.0781], + device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0196, 0.0213, 0.0211, 0.0235, 0.0193, 0.0227, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:19:50,173 INFO [train.py:876] (3/4) Epoch 10, batch 6500, loss[loss=0.1197, simple_loss=0.142, pruned_loss=0.04863, over 5542.00 frames. ], tot_loss[loss=0.1222, simple_loss=0.1471, pruned_loss=0.04864, over 1083830.03 frames. ], batch size: 40, lr: 7.93e-03, grad_scale: 8.0 +2022-11-16 03:20:05,479 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:20:10,752 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3087, 3.9229, 2.6452, 3.7153, 2.9977, 2.8373, 2.2329, 3.3542], + device='cuda:3'), covar=tensor([0.1238, 0.0250, 0.1040, 0.0371, 0.0835, 0.0950, 0.1720, 0.0380], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0138, 0.0162, 0.0145, 0.0175, 0.0169, 0.0164, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:20:10,808 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:20:46,046 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.311e+01 1.515e+02 1.926e+02 2.349e+02 4.433e+02, threshold=3.852e+02, percent-clipped=3.0 +2022-11-16 03:20:46,901 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9622, 1.4433, 1.7091, 1.4704, 1.5322, 1.6477, 1.2203, 1.5540], + device='cuda:3'), covar=tensor([0.0024, 0.0058, 0.0070, 0.0054, 0.0090, 0.0053, 0.0045, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0025, 0.0032, 0.0028, 0.0026, 0.0031, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.3299e-05, 2.2322e-05, 2.2488e-05, 3.1120e-05, 2.5817e-05, 2.4570e-05, + 2.9792e-05, 3.0409e-05], device='cuda:3') +2022-11-16 03:20:57,525 INFO [train.py:876] (3/4) Epoch 10, batch 6600, loss[loss=0.1208, simple_loss=0.1535, pruned_loss=0.04406, over 5756.00 frames. ], tot_loss[loss=0.1206, simple_loss=0.1462, pruned_loss=0.04746, over 1086089.40 frames. ], batch size: 20, lr: 7.92e-03, grad_scale: 8.0 +2022-11-16 03:21:12,066 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7889, 2.6152, 2.6642, 2.4245, 2.8235, 2.6723, 2.7099, 2.7734], + device='cuda:3'), covar=tensor([0.0484, 0.0501, 0.0610, 0.0595, 0.0573, 0.0328, 0.0422, 0.0707], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0148, 0.0109, 0.0143, 0.0171, 0.0101, 0.0121, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:21:22,922 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 03:21:31,190 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2022-11-16 03:21:51,552 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:21:53,317 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.616e+01 1.605e+02 1.991e+02 2.572e+02 4.942e+02, threshold=3.982e+02, percent-clipped=4.0 +2022-11-16 03:22:04,581 INFO [train.py:876] (3/4) Epoch 10, batch 6700, loss[loss=0.2185, simple_loss=0.2033, pruned_loss=0.1169, over 5464.00 frames. ], tot_loss[loss=0.1221, simple_loss=0.147, pruned_loss=0.04864, over 1079278.38 frames. ], batch size: 64, lr: 7.91e-03, grad_scale: 8.0 +2022-11-16 03:22:11,327 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7622, 1.2365, 2.0929, 0.8218, 1.6928, 1.7166, 1.1267, 1.5515], + device='cuda:3'), covar=tensor([0.1097, 0.0733, 0.0167, 0.1644, 0.0790, 0.0678, 0.0983, 0.0873], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0021, 0.0015, 0.0018, 0.0015, 0.0013, 0.0020, 0.0014], + device='cuda:3'), out_proj_covar=tensor([7.4983e-05, 1.0195e-04, 7.7561e-05, 9.1532e-05, 8.0101e-05, 7.3898e-05, + 9.7160e-05, 7.5367e-05], device='cuda:3') +2022-11-16 03:22:33,100 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:23:01,932 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.623e+02 1.994e+02 2.476e+02 5.420e+02, threshold=3.989e+02, percent-clipped=4.0 +2022-11-16 03:23:13,011 INFO [train.py:876] (3/4) Epoch 10, batch 6800, loss[loss=0.1124, simple_loss=0.142, pruned_loss=0.04139, over 5742.00 frames. ], tot_loss[loss=0.1213, simple_loss=0.1464, pruned_loss=0.04806, over 1084691.78 frames. ], batch size: 27, lr: 7.91e-03, grad_scale: 8.0 +2022-11-16 03:23:13,717 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5191, 4.3894, 3.5464, 2.0128, 4.1814, 1.9107, 4.2100, 2.5037], + device='cuda:3'), covar=tensor([0.1969, 0.0481, 0.0696, 0.3526, 0.0391, 0.2573, 0.0268, 0.2542], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0104, 0.0112, 0.0112, 0.0100, 0.0120, 0.0097, 0.0110], + 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') +2022-11-16 03:23:23,737 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2022-11-16 03:23:28,137 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:23:30,341 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:23:55,166 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2022-11-16 03:23:55,273 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 +2022-11-16 03:24:02,089 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:24:05,850 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2022-11-16 03:24:06,388 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:24:11,065 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.569e+02 1.889e+02 2.290e+02 4.536e+02, threshold=3.778e+02, percent-clipped=2.0 +2022-11-16 03:24:23,846 INFO [train.py:876] (3/4) Epoch 10, batch 6900, loss[loss=0.1348, simple_loss=0.138, pruned_loss=0.06578, over 4097.00 frames. ], tot_loss[loss=0.1186, simple_loss=0.1448, pruned_loss=0.04618, over 1087007.18 frames. ], batch size: 183, lr: 7.90e-03, grad_scale: 8.0 +2022-11-16 03:24:49,670 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:25:19,931 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.314e+01 1.574e+02 1.843e+02 2.228e+02 3.077e+02, threshold=3.686e+02, percent-clipped=0.0 +2022-11-16 03:25:32,128 INFO [train.py:876] (3/4) Epoch 10, batch 7000, loss[loss=0.1003, simple_loss=0.1289, pruned_loss=0.03589, over 5609.00 frames. ], tot_loss[loss=0.1188, simple_loss=0.1451, pruned_loss=0.04627, over 1083754.57 frames. ], batch size: 23, lr: 7.90e-03, grad_scale: 8.0 +2022-11-16 03:25:37,083 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8031, 2.3133, 2.0015, 1.4716, 2.2860, 1.0834, 2.2881, 1.5389], + device='cuda:3'), covar=tensor([0.1164, 0.0298, 0.1032, 0.1579, 0.0371, 0.2233, 0.0359, 0.1467], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0104, 0.0113, 0.0112, 0.0100, 0.0120, 0.0098, 0.0111], + 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') +2022-11-16 03:25:56,126 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:26:01,067 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2022-11-16 03:26:02,215 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3349, 2.3691, 2.8173, 3.5181, 3.5644, 2.7833, 2.3209, 3.5169], + device='cuda:3'), covar=tensor([0.1196, 0.3400, 0.1820, 0.2134, 0.1000, 0.2442, 0.2239, 0.1739], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0199, 0.0186, 0.0309, 0.0225, 0.0204, 0.0187, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 03:26:12,519 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8163, 3.0183, 4.0316, 2.9618, 2.8751, 4.1885, 3.1920, 3.2647], + device='cuda:3'), covar=tensor([0.0462, 0.1013, 0.0297, 0.1868, 0.1310, 0.2075, 0.1252, 0.0609], + device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0083, 0.0084, 0.0095, 0.0068, 0.0061, 0.0069, 0.0081], + 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') +2022-11-16 03:26:27,976 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.555e+02 1.806e+02 2.267e+02 5.483e+02, threshold=3.612e+02, percent-clipped=3.0 +2022-11-16 03:26:39,475 INFO [train.py:876] (3/4) Epoch 10, batch 7100, loss[loss=0.1005, simple_loss=0.1399, pruned_loss=0.03059, over 5560.00 frames. ], tot_loss[loss=0.1182, simple_loss=0.1446, pruned_loss=0.04592, over 1087563.47 frames. ], batch size: 15, lr: 7.89e-03, grad_scale: 8.0 +2022-11-16 03:26:57,716 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:27:30,439 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:27:36,241 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.573e+02 1.962e+02 2.591e+02 6.270e+02, threshold=3.924e+02, percent-clipped=1.0 +2022-11-16 03:27:47,354 INFO [train.py:876] (3/4) Epoch 10, batch 7200, loss[loss=0.09597, simple_loss=0.1351, pruned_loss=0.0284, over 5729.00 frames. ], tot_loss[loss=0.1172, simple_loss=0.1435, pruned_loss=0.0454, over 1083312.89 frames. ], batch size: 15, lr: 7.89e-03, grad_scale: 8.0 +2022-11-16 03:27:47,470 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0299, 4.0249, 4.2251, 3.6944, 4.0221, 3.9506, 1.6516, 4.1649], + device='cuda:3'), covar=tensor([0.0362, 0.0400, 0.0253, 0.0413, 0.0332, 0.0391, 0.3326, 0.0334], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0084, 0.0085, 0.0078, 0.0101, 0.0088, 0.0128, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:28:10,213 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:28:33,040 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4770, 1.8841, 1.5508, 1.1725, 1.8139, 0.8405, 1.8898, 1.1799], + device='cuda:3'), covar=tensor([0.1056, 0.0320, 0.1039, 0.1268, 0.0397, 0.2109, 0.0384, 0.1442], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0104, 0.0113, 0.0112, 0.0101, 0.0120, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:29:18,982 INFO [train.py:876] (3/4) Epoch 11, batch 0, loss[loss=0.1057, simple_loss=0.1417, pruned_loss=0.03483, over 5599.00 frames. ], tot_loss[loss=0.1057, simple_loss=0.1417, pruned_loss=0.03483, over 5599.00 frames. ], batch size: 18, lr: 7.53e-03, grad_scale: 8.0 +2022-11-16 03:29:18,983 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 03:29:24,144 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4430, 3.0414, 3.3729, 3.0669, 3.2288, 3.1805, 3.5011, 3.4529], + device='cuda:3'), covar=tensor([0.0608, 0.1258, 0.0582, 0.1236, 0.0733, 0.0505, 0.0950, 0.0655], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0106, 0.0090, 0.0117, 0.0087, 0.0077, 0.0143, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:29:28,506 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5079, 4.6101, 3.5456, 2.2268, 4.2638, 1.9197, 4.0636, 2.7933], + device='cuda:3'), covar=tensor([0.1457, 0.0100, 0.0460, 0.2150, 0.0151, 0.1694, 0.0208, 0.1385], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0104, 0.0113, 0.0112, 0.0101, 0.0119, 0.0097, 0.0111], + 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') +2022-11-16 03:29:35,597 INFO [train.py:908] (3/4) Epoch 11, validation: loss=0.1663, simple_loss=0.1831, pruned_loss=0.07475, over 1530663.00 frames. +2022-11-16 03:29:35,598 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 03:29:43,099 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.965e+01 1.625e+02 2.129e+02 2.463e+02 4.242e+02, threshold=4.258e+02, percent-clipped=1.0 +2022-11-16 03:30:00,435 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2022-11-16 03:30:05,547 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6864, 2.1586, 2.3147, 3.0145, 2.9434, 2.2840, 2.0056, 3.0303], + device='cuda:3'), covar=tensor([0.1209, 0.2030, 0.2034, 0.1606, 0.1161, 0.3036, 0.2113, 0.0986], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0197, 0.0187, 0.0306, 0.0224, 0.0203, 0.0189, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 03:30:17,737 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:30:34,424 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7500, 3.8156, 3.6221, 3.4739, 2.0576, 3.8197, 2.2192, 3.1468], + device='cuda:3'), covar=tensor([0.0380, 0.0199, 0.0195, 0.0354, 0.0584, 0.0151, 0.0457, 0.0142], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0172, 0.0178, 0.0199, 0.0189, 0.0174, 0.0186, 0.0179], + 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') +2022-11-16 03:30:36,603 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4941, 4.5036, 3.1056, 4.4400, 3.5219, 3.2138, 2.4352, 4.0303], + device='cuda:3'), covar=tensor([0.1522, 0.0318, 0.1234, 0.0295, 0.0799, 0.0965, 0.2121, 0.0386], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0139, 0.0160, 0.0144, 0.0176, 0.0169, 0.0164, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:30:42,330 INFO [train.py:876] (3/4) Epoch 11, batch 100, loss[loss=0.1371, simple_loss=0.1531, pruned_loss=0.06054, over 4767.00 frames. ], tot_loss[loss=0.1223, simple_loss=0.1484, pruned_loss=0.04805, over 433567.64 frames. ], batch size: 135, lr: 7.52e-03, grad_scale: 8.0 +2022-11-16 03:30:49,527 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 1.571e+02 1.949e+02 2.153e+02 3.381e+02, threshold=3.898e+02, percent-clipped=0.0 +2022-11-16 03:30:50,281 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:30:59,978 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2022-11-16 03:31:31,482 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:31:40,353 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3253, 4.7529, 4.2998, 4.6791, 4.8091, 4.0700, 4.3237, 3.9942], + device='cuda:3'), covar=tensor([0.0291, 0.0375, 0.1352, 0.0457, 0.0319, 0.0446, 0.0786, 0.0648], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0173, 0.0268, 0.0171, 0.0216, 0.0172, 0.0184, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:31:50,364 INFO [train.py:876] (3/4) Epoch 11, batch 200, loss[loss=0.1447, simple_loss=0.1532, pruned_loss=0.06814, over 4129.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1472, pruned_loss=0.04753, over 688236.09 frames. ], batch size: 183, lr: 7.52e-03, grad_scale: 8.0 +2022-11-16 03:31:57,298 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.513e+01 1.538e+02 1.800e+02 2.272e+02 4.125e+02, threshold=3.600e+02, percent-clipped=3.0 +2022-11-16 03:32:10,243 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1045, 3.7909, 3.9813, 3.7747, 4.1882, 3.8847, 3.7444, 4.1149], + device='cuda:3'), covar=tensor([0.0338, 0.0378, 0.0411, 0.0323, 0.0330, 0.0306, 0.0349, 0.0394], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0145, 0.0106, 0.0140, 0.0168, 0.0098, 0.0119, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:32:12,296 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:32:18,627 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 03:32:24,035 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2909, 3.7931, 2.8485, 1.7207, 3.5621, 1.5297, 3.4617, 1.9880], + device='cuda:3'), covar=tensor([0.1410, 0.0152, 0.0963, 0.2110, 0.0242, 0.2023, 0.0283, 0.1618], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0105, 0.0114, 0.0113, 0.0101, 0.0121, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:32:27,365 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3415, 4.1376, 3.0861, 1.7485, 3.8939, 1.5022, 3.7249, 2.1380], + device='cuda:3'), covar=tensor([0.1435, 0.0163, 0.0842, 0.2184, 0.0207, 0.1974, 0.0263, 0.1639], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0105, 0.0114, 0.0113, 0.0101, 0.0121, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:32:30,973 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:32:36,412 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.61 vs. limit=5.0 +2022-11-16 03:32:37,517 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5802, 4.7960, 4.3429, 4.4801, 4.6249, 4.3058, 1.6828, 4.7934], + device='cuda:3'), covar=tensor([0.0220, 0.0191, 0.0257, 0.0264, 0.0293, 0.0429, 0.2846, 0.0226], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0085, 0.0086, 0.0079, 0.0102, 0.0089, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:32:57,222 INFO [train.py:876] (3/4) Epoch 11, batch 300, loss[loss=0.0765, simple_loss=0.1118, pruned_loss=0.02059, over 5474.00 frames. ], tot_loss[loss=0.1205, simple_loss=0.1462, pruned_loss=0.04745, over 846055.60 frames. ], batch size: 10, lr: 7.51e-03, grad_scale: 8.0 +2022-11-16 03:33:00,034 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 03:33:03,471 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:33:04,730 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.624e+02 1.928e+02 2.465e+02 5.255e+02, threshold=3.856e+02, percent-clipped=4.0 +2022-11-16 03:33:28,486 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9086, 1.1790, 1.5755, 0.8439, 1.5633, 1.6488, 1.1052, 1.6396], + device='cuda:3'), covar=tensor([0.0559, 0.0918, 0.0873, 0.1785, 0.1219, 0.0360, 0.1550, 0.0647], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0021, 0.0015, 0.0018, 0.0015, 0.0014, 0.0020, 0.0014], + device='cuda:3'), out_proj_covar=tensor([7.6414e-05, 1.0324e-04, 7.8932e-05, 9.2784e-05, 8.1180e-05, 7.4857e-05, + 9.9579e-05, 7.6368e-05], device='cuda:3') +2022-11-16 03:34:03,258 INFO [train.py:876] (3/4) Epoch 11, batch 400, loss[loss=0.105, simple_loss=0.1393, pruned_loss=0.03534, over 5563.00 frames. ], tot_loss[loss=0.1185, simple_loss=0.1447, pruned_loss=0.04615, over 938990.36 frames. ], batch size: 16, lr: 7.51e-03, grad_scale: 8.0 +2022-11-16 03:34:11,192 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.143e+01 1.547e+02 1.866e+02 2.274e+02 4.703e+02, threshold=3.733e+02, percent-clipped=2.0 +2022-11-16 03:34:25,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8195, 2.9170, 2.8943, 2.7096, 2.9100, 2.8608, 1.2609, 3.0698], + device='cuda:3'), covar=tensor([0.0303, 0.0305, 0.0332, 0.0285, 0.0364, 0.0297, 0.2667, 0.0307], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0084, 0.0085, 0.0078, 0.0100, 0.0088, 0.0129, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:35:10,797 INFO [train.py:876] (3/4) Epoch 11, batch 500, loss[loss=0.08314, simple_loss=0.1224, pruned_loss=0.02194, over 5763.00 frames. ], tot_loss[loss=0.1148, simple_loss=0.142, pruned_loss=0.04385, over 992486.78 frames. ], batch size: 16, lr: 7.50e-03, grad_scale: 8.0 +2022-11-16 03:35:17,968 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.259e+01 1.458e+02 1.748e+02 2.225e+02 4.920e+02, threshold=3.496e+02, percent-clipped=3.0 +2022-11-16 03:35:18,765 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0947, 1.9272, 2.3986, 1.5985, 1.2400, 3.0118, 2.5583, 1.8997], + device='cuda:3'), covar=tensor([0.1053, 0.1332, 0.0814, 0.2572, 0.3620, 0.0573, 0.0926, 0.1482], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0083, 0.0086, 0.0094, 0.0070, 0.0062, 0.0070, 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') +2022-11-16 03:35:23,730 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1501, 1.5052, 1.1576, 1.2542, 1.4178, 1.2805, 1.0059, 1.5415], + device='cuda:3'), covar=tensor([0.0067, 0.0047, 0.0069, 0.0058, 0.0055, 0.0056, 0.0068, 0.0041], + device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0051, 0.0052, 0.0054, 0.0054, 0.0048, 0.0049, 0.0046], + device='cuda:3'), out_proj_covar=tensor([5.0293e-05, 4.6080e-05, 4.6706e-05, 4.7988e-05, 4.7935e-05, 4.2239e-05, + 4.4417e-05, 4.0746e-05], device='cuda:3') +2022-11-16 03:35:30,817 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:35:47,405 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0644, 1.5453, 1.7210, 1.5983, 1.4453, 1.8782, 1.6288, 1.6036], + device='cuda:3'), covar=tensor([0.0025, 0.0079, 0.0061, 0.0049, 0.0146, 0.0293, 0.0049, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0023, 0.0024, 0.0031, 0.0026, 0.0026, 0.0031, 0.0030], + device='cuda:3'), out_proj_covar=tensor([2.2788e-05, 2.1444e-05, 2.1729e-05, 3.0723e-05, 2.4676e-05, 2.4416e-05, + 2.9608e-05, 2.9568e-05], device='cuda:3') +2022-11-16 03:35:57,775 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 +2022-11-16 03:35:59,474 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:36:11,743 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2022-11-16 03:36:18,449 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 03:36:19,023 INFO [train.py:876] (3/4) Epoch 11, batch 600, loss[loss=0.1161, simple_loss=0.1448, pruned_loss=0.04369, over 5730.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1429, pruned_loss=0.04459, over 1026287.09 frames. ], batch size: 16, lr: 7.50e-03, grad_scale: 16.0 +2022-11-16 03:36:23,330 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2022-11-16 03:36:25,840 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2022-11-16 03:36:26,010 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.407e+01 1.498e+02 1.818e+02 2.192e+02 5.468e+02, threshold=3.637e+02, percent-clipped=3.0 +2022-11-16 03:36:41,250 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:37:26,415 INFO [train.py:876] (3/4) Epoch 11, batch 700, loss[loss=0.2099, simple_loss=0.1928, pruned_loss=0.1135, over 2968.00 frames. ], tot_loss[loss=0.1174, simple_loss=0.1441, pruned_loss=0.04533, over 1048904.92 frames. ], batch size: 284, lr: 7.49e-03, grad_scale: 16.0 +2022-11-16 03:37:33,752 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.643e+01 1.493e+02 1.779e+02 2.171e+02 7.161e+02, threshold=3.558e+02, percent-clipped=3.0 +2022-11-16 03:38:33,545 INFO [train.py:876] (3/4) Epoch 11, batch 800, loss[loss=0.08935, simple_loss=0.1235, pruned_loss=0.02762, over 5703.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1434, pruned_loss=0.04489, over 1068429.32 frames. ], batch size: 12, lr: 7.49e-03, grad_scale: 8.0 +2022-11-16 03:38:34,621 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.6542, 4.9776, 5.3921, 4.9368, 5.6784, 5.4458, 4.7244, 5.5507], + device='cuda:3'), covar=tensor([0.0263, 0.0331, 0.0446, 0.0322, 0.0247, 0.0192, 0.0269, 0.0244], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0149, 0.0110, 0.0145, 0.0175, 0.0101, 0.0124, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:38:41,646 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.679e+01 1.505e+02 1.889e+02 2.408e+02 4.187e+02, threshold=3.778e+02, percent-clipped=1.0 +2022-11-16 03:38:41,806 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6342, 4.8560, 3.3065, 4.7932, 3.5969, 3.4290, 3.0815, 4.3478], + device='cuda:3'), covar=tensor([0.1543, 0.0251, 0.0974, 0.0259, 0.0647, 0.0942, 0.1728, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0140, 0.0160, 0.0145, 0.0177, 0.0170, 0.0165, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:38:48,736 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4582, 2.1608, 3.0849, 2.6866, 3.0633, 2.1831, 2.8757, 3.5215], + device='cuda:3'), covar=tensor([0.0810, 0.1677, 0.0902, 0.1460, 0.0727, 0.1493, 0.1157, 0.0814], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0194, 0.0213, 0.0211, 0.0237, 0.0195, 0.0226, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:38:53,246 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:39:20,193 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:39:25,555 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:39:40,080 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 03:39:40,651 INFO [train.py:876] (3/4) Epoch 11, batch 900, loss[loss=0.1457, simple_loss=0.1755, pruned_loss=0.05794, over 5770.00 frames. ], tot_loss[loss=0.117, simple_loss=0.1442, pruned_loss=0.04493, over 1078137.04 frames. ], batch size: 21, lr: 7.48e-03, grad_scale: 8.0 +2022-11-16 03:39:49,605 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.416e+01 1.675e+02 2.016e+02 2.471e+02 4.865e+02, threshold=4.032e+02, percent-clipped=2.0 +2022-11-16 03:40:00,148 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:40:01,551 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:40:12,986 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 03:40:32,345 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 +2022-11-16 03:40:39,049 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-16 03:40:47,969 INFO [train.py:876] (3/4) Epoch 11, batch 1000, loss[loss=0.07601, simple_loss=0.108, pruned_loss=0.02203, over 5217.00 frames. ], tot_loss[loss=0.1176, simple_loss=0.1447, pruned_loss=0.04527, over 1081917.78 frames. ], batch size: 8, lr: 7.48e-03, grad_scale: 8.0 +2022-11-16 03:40:55,186 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:40:55,646 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.576e+01 1.693e+02 2.139e+02 2.600e+02 5.774e+02, threshold=4.279e+02, percent-clipped=7.0 +2022-11-16 03:41:00,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 03:41:36,504 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:41:55,285 INFO [train.py:876] (3/4) Epoch 11, batch 1100, loss[loss=0.2157, simple_loss=0.1971, pruned_loss=0.1171, over 3128.00 frames. ], tot_loss[loss=0.1192, simple_loss=0.1455, pruned_loss=0.04645, over 1075074.13 frames. ], batch size: 284, lr: 7.47e-03, grad_scale: 8.0 +2022-11-16 03:42:02,966 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.524e+02 1.845e+02 2.203e+02 3.683e+02, threshold=3.689e+02, percent-clipped=0.0 +2022-11-16 03:42:11,058 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1156, 2.9833, 3.1016, 3.0717, 3.0581, 2.7981, 3.4801, 3.1265], + device='cuda:3'), covar=tensor([0.0561, 0.0965, 0.0471, 0.1168, 0.0666, 0.0527, 0.0881, 0.0865], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0104, 0.0090, 0.0116, 0.0087, 0.0076, 0.0142, 0.0099], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:42:19,515 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6912, 2.7760, 2.5497, 2.7862, 2.2703, 2.3459, 2.5063, 3.2844], + device='cuda:3'), covar=tensor([0.1115, 0.1902, 0.2272, 0.1365, 0.1921, 0.1733, 0.1750, 0.0956], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0101, 0.0102, 0.0096, 0.0089, 0.0097, 0.0093, 0.0075], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 03:42:45,502 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2022-11-16 03:43:01,960 INFO [train.py:876] (3/4) Epoch 11, batch 1200, loss[loss=0.1336, simple_loss=0.1527, pruned_loss=0.05731, over 5116.00 frames. ], tot_loss[loss=0.1184, simple_loss=0.1447, pruned_loss=0.04608, over 1080659.62 frames. ], batch size: 91, lr: 7.47e-03, grad_scale: 8.0 +2022-11-16 03:43:02,114 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7841, 1.1678, 1.6120, 1.0894, 1.4498, 1.5185, 0.9718, 1.3989], + device='cuda:3'), covar=tensor([0.0487, 0.0525, 0.0591, 0.0758, 0.1227, 0.0455, 0.1393, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0015, 0.0014, 0.0020, 0.0015], + device='cuda:3'), out_proj_covar=tensor([7.6675e-05, 1.0411e-04, 7.9489e-05, 9.4802e-05, 8.0868e-05, 7.4939e-05, + 9.9939e-05, 7.7974e-05], device='cuda:3') +2022-11-16 03:43:04,746 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6963, 0.9879, 1.4789, 0.9344, 1.5069, 1.4742, 0.9551, 1.2453], + device='cuda:3'), covar=tensor([0.0447, 0.0458, 0.0313, 0.0909, 0.0367, 0.0483, 0.0498, 0.0303], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0015, 0.0014, 0.0020, 0.0015], + device='cuda:3'), out_proj_covar=tensor([7.6676e-05, 1.0409e-04, 7.9507e-05, 9.4846e-05, 8.0878e-05, 7.4974e-05, + 9.9978e-05, 7.7962e-05], device='cuda:3') +2022-11-16 03:43:10,219 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.261e+01 1.559e+02 1.976e+02 2.426e+02 6.394e+02, threshold=3.952e+02, percent-clipped=4.0 +2022-11-16 03:43:15,028 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4397, 4.8854, 5.3043, 4.8967, 5.5110, 5.3589, 4.6785, 5.4919], + device='cuda:3'), covar=tensor([0.0345, 0.0269, 0.0370, 0.0278, 0.0316, 0.0157, 0.0227, 0.0224], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0150, 0.0110, 0.0145, 0.0175, 0.0101, 0.0123, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:43:18,854 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:43:20,840 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:43:22,434 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 03:43:29,788 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6243, 3.8614, 3.5762, 3.3209, 1.8109, 3.6973, 2.2114, 3.2097], + device='cuda:3'), covar=tensor([0.0439, 0.0190, 0.0231, 0.0344, 0.0648, 0.0180, 0.0538, 0.0196], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0172, 0.0178, 0.0198, 0.0187, 0.0175, 0.0187, 0.0178], + 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') +2022-11-16 03:43:33,079 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9518, 2.4793, 3.6684, 3.3196, 3.8968, 2.6313, 3.4943, 4.0901], + device='cuda:3'), covar=tensor([0.0812, 0.1584, 0.0744, 0.1483, 0.0545, 0.1493, 0.1150, 0.0678], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0190, 0.0208, 0.0207, 0.0232, 0.0190, 0.0221, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:43:53,689 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:44:10,602 INFO [train.py:876] (3/4) Epoch 11, batch 1300, loss[loss=0.09784, simple_loss=0.1324, pruned_loss=0.03166, over 5670.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1441, pruned_loss=0.04504, over 1084268.43 frames. ], batch size: 19, lr: 7.46e-03, grad_scale: 8.0 +2022-11-16 03:44:11,404 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5041, 4.9243, 5.4230, 4.9318, 5.5299, 5.3629, 4.7870, 5.4984], + device='cuda:3'), covar=tensor([0.0267, 0.0253, 0.0287, 0.0240, 0.0291, 0.0162, 0.0205, 0.0192], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0150, 0.0110, 0.0145, 0.0175, 0.0102, 0.0122, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:44:18,270 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.904e+01 1.577e+02 1.830e+02 2.359e+02 4.082e+02, threshold=3.660e+02, percent-clipped=1.0 +2022-11-16 03:44:18,501 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6626, 1.8418, 2.4761, 2.2900, 2.4181, 1.7391, 2.2261, 2.6881], + device='cuda:3'), covar=tensor([0.0668, 0.1436, 0.0861, 0.1108, 0.0827, 0.1276, 0.1021, 0.0735], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0190, 0.0208, 0.0208, 0.0232, 0.0190, 0.0222, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:44:38,548 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:44:43,024 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-16 03:44:55,162 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:45:15,728 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.93 vs. limit=5.0 +2022-11-16 03:45:16,721 INFO [train.py:876] (3/4) Epoch 11, batch 1400, loss[loss=0.144, simple_loss=0.1628, pruned_loss=0.06264, over 5455.00 frames. ], tot_loss[loss=0.1191, simple_loss=0.1452, pruned_loss=0.04647, over 1084480.02 frames. ], batch size: 58, lr: 7.46e-03, grad_scale: 8.0 +2022-11-16 03:45:19,895 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:45:25,523 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.235e+01 1.555e+02 1.864e+02 2.328e+02 5.952e+02, threshold=3.728e+02, percent-clipped=5.0 +2022-11-16 03:46:24,540 INFO [train.py:876] (3/4) Epoch 11, batch 1500, loss[loss=0.07504, simple_loss=0.1168, pruned_loss=0.01665, over 5562.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1442, pruned_loss=0.04504, over 1088165.89 frames. ], batch size: 15, lr: 7.45e-03, grad_scale: 8.0 +2022-11-16 03:46:32,814 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.551e+02 1.772e+02 2.146e+02 3.863e+02, threshold=3.544e+02, percent-clipped=1.0 +2022-11-16 03:46:37,965 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:46:42,227 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:47:15,034 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:47:19,873 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:47:25,026 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8153, 4.7779, 4.8100, 4.8180, 4.5161, 4.2499, 5.4393, 4.7422], + device='cuda:3'), covar=tensor([0.0360, 0.0962, 0.0287, 0.1028, 0.0510, 0.0333, 0.0528, 0.0474], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0107, 0.0092, 0.0117, 0.0088, 0.0077, 0.0143, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:47:25,526 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2022-11-16 03:47:33,156 INFO [train.py:876] (3/4) Epoch 11, batch 1600, loss[loss=0.1132, simple_loss=0.1358, pruned_loss=0.04531, over 5649.00 frames. ], tot_loss[loss=0.1161, simple_loss=0.1433, pruned_loss=0.04447, over 1083998.05 frames. ], batch size: 38, lr: 7.45e-03, grad_scale: 8.0 +2022-11-16 03:47:36,802 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2022-11-16 03:47:40,118 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 03:47:40,982 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.188e+01 1.509e+02 1.863e+02 2.484e+02 5.200e+02, threshold=3.726e+02, percent-clipped=6.0 +2022-11-16 03:47:53,049 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 03:48:16,920 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3175, 4.1059, 2.7802, 3.8076, 3.1128, 2.8871, 2.4623, 3.3575], + device='cuda:3'), covar=tensor([0.1373, 0.0280, 0.1126, 0.0458, 0.0955, 0.0987, 0.1674, 0.0471], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0138, 0.0158, 0.0144, 0.0177, 0.0168, 0.0163, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:48:18,285 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:48:22,803 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7311, 1.6279, 2.1231, 1.5559, 2.0485, 1.6558, 1.4400, 1.6136], + device='cuda:3'), covar=tensor([0.0184, 0.0656, 0.0504, 0.0782, 0.0965, 0.0742, 0.1034, 0.1133], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0016, 0.0014, 0.0021, 0.0015], + device='cuda:3'), out_proj_covar=tensor([7.8356e-05, 1.0652e-04, 8.1406e-05, 9.5879e-05, 8.2482e-05, 7.6554e-05, + 1.0078e-04, 7.8455e-05], device='cuda:3') +2022-11-16 03:48:39,919 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:48:40,466 INFO [train.py:876] (3/4) Epoch 11, batch 1700, loss[loss=0.1266, simple_loss=0.1551, pruned_loss=0.04901, over 5632.00 frames. ], tot_loss[loss=0.1156, simple_loss=0.1426, pruned_loss=0.04433, over 1085892.53 frames. ], batch size: 29, lr: 7.44e-03, grad_scale: 8.0 +2022-11-16 03:48:48,599 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 1.679e+02 2.068e+02 2.361e+02 5.198e+02, threshold=4.137e+02, percent-clipped=4.0 +2022-11-16 03:48:50,645 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:49:43,798 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3563, 3.8551, 3.4760, 3.8712, 3.8360, 3.2191, 3.4269, 3.2784], + device='cuda:3'), covar=tensor([0.1073, 0.0571, 0.1363, 0.0445, 0.0519, 0.0502, 0.0743, 0.0796], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0170, 0.0266, 0.0170, 0.0215, 0.0169, 0.0182, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 03:49:45,123 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3756, 4.2780, 2.8256, 4.1508, 3.2899, 2.7897, 2.1391, 3.6756], + device='cuda:3'), covar=tensor([0.1286, 0.0217, 0.1075, 0.0277, 0.0670, 0.1038, 0.1990, 0.0361], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0139, 0.0160, 0.0146, 0.0178, 0.0170, 0.0165, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:49:48,194 INFO [train.py:876] (3/4) Epoch 11, batch 1800, loss[loss=0.1156, simple_loss=0.1507, pruned_loss=0.04028, over 5748.00 frames. ], tot_loss[loss=0.1154, simple_loss=0.1428, pruned_loss=0.04399, over 1087298.98 frames. ], batch size: 14, lr: 7.44e-03, grad_scale: 8.0 +2022-11-16 03:49:55,852 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.990e+01 1.615e+02 2.043e+02 2.453e+02 6.860e+02, threshold=4.086e+02, percent-clipped=1.0 +2022-11-16 03:50:38,144 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:50:38,728 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:50:50,582 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7520, 4.6963, 3.5849, 2.1675, 4.3708, 1.9135, 4.4738, 2.5896], + device='cuda:3'), covar=tensor([0.1373, 0.0134, 0.0563, 0.2130, 0.0182, 0.1942, 0.0172, 0.1665], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0105, 0.0115, 0.0114, 0.0101, 0.0124, 0.0100, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:50:56,052 INFO [train.py:876] (3/4) Epoch 11, batch 1900, loss[loss=0.1074, simple_loss=0.1399, pruned_loss=0.03747, over 5604.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.1433, pruned_loss=0.04389, over 1089368.30 frames. ], batch size: 24, lr: 7.43e-03, grad_scale: 8.0 +2022-11-16 03:51:04,167 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.647e+01 1.532e+02 1.879e+02 2.235e+02 4.032e+02, threshold=3.759e+02, percent-clipped=0.0 +2022-11-16 03:51:19,907 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:52:03,035 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:52:03,553 INFO [train.py:876] (3/4) Epoch 11, batch 2000, loss[loss=0.1463, simple_loss=0.1604, pruned_loss=0.06611, over 5573.00 frames. ], tot_loss[loss=0.1149, simple_loss=0.1421, pruned_loss=0.0439, over 1089370.29 frames. ], batch size: 50, lr: 7.43e-03, grad_scale: 8.0 +2022-11-16 03:52:12,066 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.989e+01 1.482e+02 1.897e+02 2.343e+02 3.956e+02, threshold=3.795e+02, percent-clipped=2.0 +2022-11-16 03:52:24,029 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8995, 3.9120, 3.9344, 3.4647, 2.1636, 4.2476, 2.4107, 3.5135], + device='cuda:3'), covar=tensor([0.0379, 0.0232, 0.0178, 0.0445, 0.0637, 0.0137, 0.0522, 0.0137], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0173, 0.0176, 0.0198, 0.0189, 0.0176, 0.0187, 0.0180], + 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') +2022-11-16 03:52:26,916 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1890, 4.6150, 4.2082, 4.6586, 4.6409, 3.8856, 4.1629, 4.1461], + device='cuda:3'), covar=tensor([0.0397, 0.0504, 0.1439, 0.0424, 0.0479, 0.0573, 0.0730, 0.0665], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0172, 0.0269, 0.0171, 0.0215, 0.0171, 0.0184, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:52:35,421 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:53:11,057 INFO [train.py:876] (3/4) Epoch 11, batch 2100, loss[loss=0.1129, simple_loss=0.1423, pruned_loss=0.04178, over 5617.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1417, pruned_loss=0.04378, over 1085003.50 frames. ], batch size: 32, lr: 7.42e-03, grad_scale: 8.0 +2022-11-16 03:53:19,079 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.996e+01 1.479e+02 1.848e+02 2.338e+02 4.200e+02, threshold=3.697e+02, percent-clipped=1.0 +2022-11-16 03:53:25,042 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8376, 1.0601, 0.8070, 0.7875, 1.1207, 1.1005, 0.6015, 1.2166], + device='cuda:3'), covar=tensor([0.0066, 0.0036, 0.0057, 0.0037, 0.0044, 0.0046, 0.0078, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0050, 0.0052, 0.0054, 0.0054, 0.0048, 0.0049, 0.0046], + device='cuda:3'), out_proj_covar=tensor([4.9590e-05, 4.4929e-05, 4.5814e-05, 4.8115e-05, 4.7888e-05, 4.2050e-05, + 4.3677e-05, 4.0941e-05], device='cuda:3') +2022-11-16 03:53:27,209 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2022-11-16 03:53:40,146 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 03:53:45,570 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0789, 4.4704, 4.0943, 4.4744, 4.4868, 3.8359, 3.9974, 3.7672], + device='cuda:3'), covar=tensor([0.0411, 0.0521, 0.1392, 0.0417, 0.0437, 0.0536, 0.0584, 0.0988], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0172, 0.0268, 0.0170, 0.0213, 0.0170, 0.0185, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:53:46,273 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5002, 1.1213, 1.3228, 0.9905, 1.3843, 1.3595, 1.0000, 1.2602], + device='cuda:3'), covar=tensor([0.0478, 0.0440, 0.0363, 0.0782, 0.0887, 0.0548, 0.0565, 0.0421], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0015, 0.0014, 0.0020, 0.0014], + device='cuda:3'), out_proj_covar=tensor([7.7731e-05, 1.0606e-04, 8.0221e-05, 9.5838e-05, 8.2069e-05, 7.6136e-05, + 9.9361e-05, 7.7092e-05], device='cuda:3') +2022-11-16 03:53:52,165 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2022-11-16 03:54:02,267 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:54:07,983 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 +2022-11-16 03:54:10,812 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:54:15,682 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9399, 1.3826, 1.4506, 1.3165, 1.2940, 1.3901, 1.3109, 1.2967], + device='cuda:3'), covar=tensor([0.4025, 0.2514, 0.1922, 0.1659, 0.2248, 0.3173, 0.2150, 0.1088], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0101, 0.0102, 0.0097, 0.0090, 0.0098, 0.0093, 0.0077], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 03:54:18,890 INFO [train.py:876] (3/4) Epoch 11, batch 2200, loss[loss=0.1163, simple_loss=0.1518, pruned_loss=0.0404, over 5715.00 frames. ], tot_loss[loss=0.1162, simple_loss=0.1429, pruned_loss=0.04477, over 1076699.83 frames. ], batch size: 31, lr: 7.42e-03, grad_scale: 8.0 +2022-11-16 03:54:26,978 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.575e+02 1.874e+02 2.286e+02 4.723e+02, threshold=3.748e+02, percent-clipped=3.0 +2022-11-16 03:54:31,460 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 03:54:34,898 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:54:36,636 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2022-11-16 03:54:39,524 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:54:48,156 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:54:52,442 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:55:11,185 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 +2022-11-16 03:55:25,268 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8962, 1.8602, 1.9370, 1.9844, 1.8189, 1.5042, 1.8048, 2.1630], + device='cuda:3'), covar=tensor([0.1649, 0.2122, 0.2005, 0.1485, 0.2139, 0.2887, 0.1908, 0.1178], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0101, 0.0103, 0.0098, 0.0090, 0.0099, 0.0094, 0.0077], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 03:55:30,346 INFO [train.py:876] (3/4) Epoch 11, batch 2300, loss[loss=0.1202, simple_loss=0.1542, pruned_loss=0.0431, over 5761.00 frames. ], tot_loss[loss=0.1165, simple_loss=0.1432, pruned_loss=0.04485, over 1081806.21 frames. ], batch size: 21, lr: 7.41e-03, grad_scale: 8.0 +2022-11-16 03:55:33,372 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:55:38,336 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.756e+01 1.570e+02 1.933e+02 2.290e+02 4.748e+02, threshold=3.866e+02, percent-clipped=2.0 +2022-11-16 03:55:45,356 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5338, 3.7383, 3.5707, 3.2911, 1.9760, 3.6663, 2.1661, 3.1711], + device='cuda:3'), covar=tensor([0.0410, 0.0209, 0.0150, 0.0404, 0.0587, 0.0173, 0.0527, 0.0171], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0169, 0.0173, 0.0194, 0.0185, 0.0172, 0.0184, 0.0176], + 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') +2022-11-16 03:56:37,692 INFO [train.py:876] (3/4) Epoch 11, batch 2400, loss[loss=0.09043, simple_loss=0.1289, pruned_loss=0.02599, over 5518.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.143, pruned_loss=0.0448, over 1087263.50 frames. ], batch size: 12, lr: 7.41e-03, grad_scale: 8.0 +2022-11-16 03:56:45,362 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.612e+02 2.010e+02 2.396e+02 4.325e+02, threshold=4.021e+02, percent-clipped=4.0 +2022-11-16 03:56:54,212 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5242, 4.5282, 4.4857, 4.7816, 4.4263, 3.9625, 5.2702, 4.6532], + device='cuda:3'), covar=tensor([0.0447, 0.0858, 0.0313, 0.1113, 0.0541, 0.0422, 0.0586, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0108, 0.0093, 0.0118, 0.0088, 0.0077, 0.0144, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:57:14,541 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2022-11-16 03:57:43,227 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4353, 4.9613, 5.2506, 4.9354, 5.5677, 5.3815, 4.6674, 5.4773], + device='cuda:3'), covar=tensor([0.0293, 0.0263, 0.0337, 0.0312, 0.0222, 0.0142, 0.0217, 0.0190], + device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0147, 0.0108, 0.0144, 0.0172, 0.0102, 0.0122, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 03:57:45,187 INFO [train.py:876] (3/4) Epoch 11, batch 2500, loss[loss=0.1269, simple_loss=0.1562, pruned_loss=0.04878, over 5632.00 frames. ], tot_loss[loss=0.1168, simple_loss=0.1439, pruned_loss=0.04486, over 1087245.80 frames. ], batch size: 32, lr: 7.40e-03, grad_scale: 8.0 +2022-11-16 03:57:50,340 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:57:53,348 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.180e+01 1.567e+02 1.927e+02 2.439e+02 5.845e+02, threshold=3.854e+02, percent-clipped=5.0 +2022-11-16 03:58:05,556 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:15,032 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:18,682 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:31,510 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:38,037 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:40,775 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-16 03:58:52,012 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7588, 3.6712, 3.6819, 3.8616, 3.4781, 3.2538, 4.1151, 3.6556], + device='cuda:3'), covar=tensor([0.0489, 0.0998, 0.0581, 0.1320, 0.0621, 0.0493, 0.0865, 0.0805], + device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0108, 0.0093, 0.0117, 0.0088, 0.0078, 0.0143, 0.0101], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:58:52,034 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 03:58:52,581 INFO [train.py:876] (3/4) Epoch 11, batch 2600, loss[loss=0.1766, simple_loss=0.1891, pruned_loss=0.08208, over 5551.00 frames. ], tot_loss[loss=0.1152, simple_loss=0.1421, pruned_loss=0.04411, over 1088865.09 frames. ], batch size: 46, lr: 7.40e-03, grad_scale: 8.0 +2022-11-16 03:59:00,308 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 03:59:01,316 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.425e+02 1.750e+02 2.205e+02 4.754e+02, threshold=3.499e+02, percent-clipped=3.0 +2022-11-16 03:59:13,881 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0275, 4.1472, 3.9730, 3.5222, 2.2536, 4.3165, 2.3937, 3.4920], + device='cuda:3'), covar=tensor([0.0379, 0.0239, 0.0208, 0.0501, 0.0620, 0.0180, 0.0514, 0.0298], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0169, 0.0173, 0.0193, 0.0183, 0.0171, 0.0183, 0.0176], + 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') +2022-11-16 03:59:23,200 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6931, 1.9428, 1.9157, 1.6644, 1.9260, 1.9114, 0.9807, 1.9744], + device='cuda:3'), covar=tensor([0.0455, 0.0397, 0.0356, 0.0379, 0.0429, 0.0376, 0.2093, 0.0410], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0089, 0.0087, 0.0082, 0.0105, 0.0090, 0.0134, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:59:27,142 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0019, 2.2124, 3.5799, 2.9445, 3.7910, 2.5097, 3.2979, 4.0984], + device='cuda:3'), covar=tensor([0.0570, 0.1652, 0.0716, 0.1382, 0.0590, 0.1425, 0.1198, 0.0675], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0196, 0.0213, 0.0214, 0.0240, 0.0196, 0.0226, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 03:59:32,654 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4822, 4.0536, 3.7106, 4.0620, 4.0406, 3.4150, 3.7386, 3.5710], + device='cuda:3'), covar=tensor([0.0815, 0.0389, 0.1227, 0.0436, 0.0474, 0.0559, 0.0559, 0.0644], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0174, 0.0272, 0.0170, 0.0215, 0.0171, 0.0185, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 03:59:53,821 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8878, 5.0084, 5.1714, 4.8807, 4.6897, 4.4743, 5.6849, 5.0058], + device='cuda:3'), covar=tensor([0.0324, 0.0860, 0.0272, 0.1222, 0.0366, 0.0288, 0.0428, 0.0454], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0107, 0.0092, 0.0116, 0.0087, 0.0077, 0.0142, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:00:00,557 INFO [train.py:876] (3/4) Epoch 11, batch 2700, loss[loss=0.1159, simple_loss=0.1493, pruned_loss=0.04124, over 5683.00 frames. ], tot_loss[loss=0.1152, simple_loss=0.1421, pruned_loss=0.04411, over 1085770.30 frames. ], batch size: 19, lr: 7.39e-03, grad_scale: 8.0 +2022-11-16 04:00:08,241 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.727e+01 1.476e+02 1.842e+02 2.376e+02 5.290e+02, threshold=3.683e+02, percent-clipped=4.0 +2022-11-16 04:00:21,726 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1409, 4.2842, 4.0682, 3.9791, 4.2173, 3.9426, 1.7660, 4.2740], + device='cuda:3'), covar=tensor([0.0311, 0.0442, 0.0367, 0.0364, 0.0435, 0.0465, 0.2953, 0.0397], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0088, 0.0086, 0.0080, 0.0103, 0.0088, 0.0132, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:01:07,891 INFO [train.py:876] (3/4) Epoch 11, batch 2800, loss[loss=0.1401, simple_loss=0.164, pruned_loss=0.0581, over 5628.00 frames. ], tot_loss[loss=0.1151, simple_loss=0.1423, pruned_loss=0.044, over 1087680.81 frames. ], batch size: 38, lr: 7.39e-03, grad_scale: 16.0 +2022-11-16 04:01:15,808 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.621e+01 1.514e+02 1.754e+02 2.242e+02 3.721e+02, threshold=3.509e+02, percent-clipped=2.0 +2022-11-16 04:01:32,015 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2022-11-16 04:01:37,954 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:01:45,923 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0343, 1.8755, 2.4296, 1.8138, 1.4506, 2.8944, 2.3386, 1.9748], + device='cuda:3'), covar=tensor([0.1330, 0.1605, 0.0912, 0.3167, 0.3667, 0.0878, 0.1415, 0.1784], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0087, 0.0088, 0.0097, 0.0071, 0.0065, 0.0074, 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') +2022-11-16 04:01:50,659 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:02:02,140 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5161, 2.7533, 3.9270, 3.5278, 4.6407, 3.1249, 3.9801, 4.5121], + device='cuda:3'), covar=tensor([0.0503, 0.1619, 0.0819, 0.1379, 0.0310, 0.1333, 0.1028, 0.0550], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0190, 0.0208, 0.0208, 0.0234, 0.0192, 0.0222, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:02:03,387 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2856, 3.9734, 2.7371, 3.6382, 3.0428, 2.7477, 2.0116, 3.3366], + device='cuda:3'), covar=tensor([0.1346, 0.0215, 0.1071, 0.0382, 0.0815, 0.0962, 0.1907, 0.0350], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0139, 0.0159, 0.0144, 0.0174, 0.0168, 0.0161, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:02:10,830 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:02:15,429 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:02:15,978 INFO [train.py:876] (3/4) Epoch 11, batch 2900, loss[loss=0.1189, simple_loss=0.1472, pruned_loss=0.04532, over 5595.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1436, pruned_loss=0.04421, over 1090908.65 frames. ], batch size: 43, lr: 7.38e-03, grad_scale: 16.0 +2022-11-16 04:02:19,377 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:02:23,739 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.526e+01 1.563e+02 1.912e+02 2.291e+02 3.744e+02, threshold=3.824e+02, percent-clipped=2.0 +2022-11-16 04:02:47,901 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:03:12,272 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9020, 1.5029, 1.6234, 1.2623, 1.3675, 1.6213, 1.3191, 1.2168], + device='cuda:3'), covar=tensor([0.0027, 0.0048, 0.0031, 0.0074, 0.0065, 0.0049, 0.0048, 0.0057], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0025, 0.0025, 0.0034, 0.0028, 0.0027, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.4463e-05, 2.3140e-05, 2.2908e-05, 3.3152e-05, 2.6224e-05, 2.6069e-05, + 3.1608e-05, 3.1349e-05], device='cuda:3') +2022-11-16 04:03:23,300 INFO [train.py:876] (3/4) Epoch 11, batch 3000, loss[loss=0.1344, simple_loss=0.161, pruned_loss=0.05387, over 5709.00 frames. ], tot_loss[loss=0.1172, simple_loss=0.1441, pruned_loss=0.04514, over 1089020.78 frames. ], batch size: 36, lr: 7.38e-03, grad_scale: 16.0 +2022-11-16 04:03:23,300 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 04:03:29,105 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9283, 4.1109, 3.9542, 3.4637, 2.0074, 3.9549, 2.3381, 3.4379], + device='cuda:3'), covar=tensor([0.0379, 0.0173, 0.0157, 0.0375, 0.0649, 0.0141, 0.0562, 0.0167], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0171, 0.0175, 0.0197, 0.0187, 0.0173, 0.0187, 0.0178], + 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') +2022-11-16 04:03:40,544 INFO [train.py:908] (3/4) Epoch 11, validation: loss=0.1699, simple_loss=0.1855, pruned_loss=0.07718, over 1530663.00 frames. +2022-11-16 04:03:40,545 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 04:03:48,302 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 1.514e+02 1.845e+02 2.226e+02 5.649e+02, threshold=3.690e+02, percent-clipped=5.0 +2022-11-16 04:04:49,201 INFO [train.py:876] (3/4) Epoch 11, batch 3100, loss[loss=0.1693, simple_loss=0.1753, pruned_loss=0.08158, over 5285.00 frames. ], tot_loss[loss=0.1182, simple_loss=0.1446, pruned_loss=0.04593, over 1084077.70 frames. ], batch size: 79, lr: 7.37e-03, grad_scale: 16.0 +2022-11-16 04:04:56,948 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.177e+01 1.516e+02 1.803e+02 2.135e+02 3.632e+02, threshold=3.607e+02, percent-clipped=0.0 +2022-11-16 04:05:31,465 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:05:39,363 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9694, 2.3805, 3.1201, 2.1910, 2.1897, 3.4804, 2.7736, 2.2888], + device='cuda:3'), covar=tensor([0.0770, 0.1306, 0.0552, 0.2470, 0.2956, 0.1877, 0.1320, 0.1300], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0086, 0.0086, 0.0095, 0.0069, 0.0064, 0.0072, 0.0084], + 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') +2022-11-16 04:05:47,884 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8192, 2.4610, 3.0773, 1.3995, 2.8946, 3.3394, 3.2132, 3.1955], + device='cuda:3'), covar=tensor([0.2603, 0.2036, 0.0972, 0.3569, 0.0723, 0.0950, 0.0626, 0.1034], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0182, 0.0164, 0.0182, 0.0180, 0.0198, 0.0167, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:05:55,896 INFO [train.py:876] (3/4) Epoch 11, batch 3200, loss[loss=0.08601, simple_loss=0.1144, pruned_loss=0.0288, over 5717.00 frames. ], tot_loss[loss=0.118, simple_loss=0.1446, pruned_loss=0.04572, over 1082987.89 frames. ], batch size: 11, lr: 7.37e-03, grad_scale: 16.0 +2022-11-16 04:05:59,615 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:06:04,154 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:06:04,812 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.709e+01 1.702e+02 2.039e+02 2.411e+02 4.513e+02, threshold=4.077e+02, percent-clipped=5.0 +2022-11-16 04:06:31,890 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:06:33,249 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8141, 1.5816, 1.6795, 1.2557, 1.6060, 1.5716, 1.4128, 1.0882], + device='cuda:3'), covar=tensor([0.0027, 0.0050, 0.0034, 0.0055, 0.0054, 0.0069, 0.0046, 0.0063], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0025, 0.0025, 0.0033, 0.0028, 0.0027, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.4325e-05, 2.3034e-05, 2.2540e-05, 3.2689e-05, 2.6225e-05, 2.5783e-05, + 3.1609e-05, 3.1257e-05], device='cuda:3') +2022-11-16 04:07:03,628 INFO [train.py:876] (3/4) Epoch 11, batch 3300, loss[loss=0.1774, simple_loss=0.1677, pruned_loss=0.09354, over 3087.00 frames. ], tot_loss[loss=0.1173, simple_loss=0.144, pruned_loss=0.04524, over 1085566.51 frames. ], batch size: 284, lr: 7.36e-03, grad_scale: 16.0 +2022-11-16 04:07:06,285 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2234, 4.2750, 4.2367, 4.3759, 4.0369, 3.6692, 4.8394, 4.2768], + device='cuda:3'), covar=tensor([0.0532, 0.0690, 0.0474, 0.1162, 0.0598, 0.0376, 0.0623, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0110, 0.0094, 0.0121, 0.0090, 0.0079, 0.0146, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:07:11,809 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.536e+01 1.447e+02 1.827e+02 2.353e+02 6.584e+02, threshold=3.655e+02, percent-clipped=2.0 +2022-11-16 04:07:12,003 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4830, 1.2299, 0.9650, 1.0220, 1.2781, 1.4452, 0.7871, 1.1606], + device='cuda:3'), covar=tensor([0.0350, 0.0346, 0.0646, 0.0499, 0.0358, 0.0618, 0.0930, 0.0350], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0015, 0.0014, 0.0020, 0.0015], + device='cuda:3'), out_proj_covar=tensor([7.7729e-05, 1.0663e-04, 8.0606e-05, 9.4916e-05, 8.1418e-05, 7.6923e-05, + 1.0045e-04, 7.7835e-05], device='cuda:3') +2022-11-16 04:07:37,870 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.98 vs. limit=5.0 +2022-11-16 04:08:11,730 INFO [train.py:876] (3/4) Epoch 11, batch 3400, loss[loss=0.06951, simple_loss=0.1092, pruned_loss=0.0149, over 5653.00 frames. ], tot_loss[loss=0.1181, simple_loss=0.1442, pruned_loss=0.04597, over 1080672.34 frames. ], batch size: 11, lr: 7.36e-03, grad_scale: 8.0 +2022-11-16 04:08:20,082 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.517e+02 1.870e+02 2.344e+02 4.526e+02, threshold=3.741e+02, percent-clipped=4.0 +2022-11-16 04:08:29,554 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2022-11-16 04:08:44,823 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7718, 2.2052, 1.9057, 1.3755, 2.0855, 2.4140, 2.2442, 2.5142], + device='cuda:3'), covar=tensor([0.1777, 0.1601, 0.1670, 0.2728, 0.1095, 0.0900, 0.0721, 0.1096], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0185, 0.0166, 0.0184, 0.0181, 0.0198, 0.0168, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:08:46,834 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 +2022-11-16 04:09:19,518 INFO [train.py:876] (3/4) Epoch 11, batch 3500, loss[loss=0.1213, simple_loss=0.1454, pruned_loss=0.04856, over 5603.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1429, pruned_loss=0.04461, over 1087256.49 frames. ], batch size: 18, lr: 7.35e-03, grad_scale: 8.0 +2022-11-16 04:09:27,971 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.997e+01 1.638e+02 2.033e+02 2.357e+02 4.512e+02, threshold=4.066e+02, percent-clipped=3.0 +2022-11-16 04:09:41,886 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9356, 1.8904, 1.9402, 2.0314, 1.8078, 1.5134, 1.8011, 2.3199], + device='cuda:3'), covar=tensor([0.1848, 0.2339, 0.2412, 0.1429, 0.1641, 0.1854, 0.1809, 0.0947], + device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0104, 0.0105, 0.0101, 0.0091, 0.0100, 0.0096, 0.0077], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 04:09:52,296 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:10:08,548 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:10:27,147 INFO [train.py:876] (3/4) Epoch 11, batch 3600, loss[loss=0.2493, simple_loss=0.2127, pruned_loss=0.1429, over 3025.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1432, pruned_loss=0.04467, over 1084535.74 frames. ], batch size: 284, lr: 7.35e-03, grad_scale: 8.0 +2022-11-16 04:10:28,583 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:10:33,281 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:10:35,632 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.560e+02 1.906e+02 2.408e+02 5.224e+02, threshold=3.812e+02, percent-clipped=4.0 +2022-11-16 04:10:49,309 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9797, 2.4877, 2.1160, 1.4184, 2.2902, 2.6105, 2.4727, 2.7515], + device='cuda:3'), covar=tensor([0.1618, 0.1311, 0.1595, 0.2561, 0.0851, 0.0904, 0.0516, 0.0924], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0182, 0.0165, 0.0182, 0.0181, 0.0198, 0.0168, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:10:50,277 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:11:09,727 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:11:14,449 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2022-11-16 04:11:35,222 INFO [train.py:876] (3/4) Epoch 11, batch 3700, loss[loss=0.08395, simple_loss=0.1201, pruned_loss=0.02389, over 5551.00 frames. ], tot_loss[loss=0.1168, simple_loss=0.144, pruned_loss=0.04482, over 1086168.09 frames. ], batch size: 16, lr: 7.34e-03, grad_scale: 8.0 +2022-11-16 04:11:41,116 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4994, 5.0833, 5.3705, 4.9967, 5.5568, 5.4524, 4.6668, 5.6079], + device='cuda:3'), covar=tensor([0.0334, 0.0285, 0.0389, 0.0313, 0.0344, 0.0181, 0.0219, 0.0192], + device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0146, 0.0108, 0.0142, 0.0172, 0.0102, 0.0122, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:11:43,591 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.959e+01 1.528e+02 1.916e+02 2.228e+02 3.767e+02, threshold=3.832e+02, percent-clipped=0.0 +2022-11-16 04:12:41,655 INFO [train.py:876] (3/4) Epoch 11, batch 3800, loss[loss=0.1014, simple_loss=0.1322, pruned_loss=0.03527, over 5523.00 frames. ], tot_loss[loss=0.1176, simple_loss=0.1447, pruned_loss=0.04522, over 1085690.85 frames. ], batch size: 17, lr: 7.34e-03, grad_scale: 8.0 +2022-11-16 04:12:50,478 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.425e+01 1.575e+02 2.020e+02 2.543e+02 6.057e+02, threshold=4.040e+02, percent-clipped=8.0 +2022-11-16 04:13:36,015 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8809, 1.9309, 1.6746, 1.9168, 1.9669, 1.8799, 1.6559, 1.8596], + device='cuda:3'), covar=tensor([0.0481, 0.0841, 0.1763, 0.0796, 0.0718, 0.0604, 0.1508, 0.0717], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0178, 0.0273, 0.0173, 0.0218, 0.0173, 0.0185, 0.0175], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:13:49,822 INFO [train.py:876] (3/4) Epoch 11, batch 3900, loss[loss=0.1568, simple_loss=0.1699, pruned_loss=0.07191, over 5116.00 frames. ], tot_loss[loss=0.1178, simple_loss=0.1447, pruned_loss=0.04548, over 1088838.99 frames. ], batch size: 91, lr: 7.33e-03, grad_scale: 8.0 +2022-11-16 04:13:53,085 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:13:59,641 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.483e+01 1.484e+02 1.748e+02 2.175e+02 4.162e+02, threshold=3.496e+02, percent-clipped=1.0 +2022-11-16 04:14:11,304 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:14:31,200 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:14:58,293 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2022-11-16 04:14:58,607 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9687, 3.0414, 3.0767, 2.8316, 2.9965, 2.9222, 1.2246, 3.1338], + device='cuda:3'), covar=tensor([0.0327, 0.0272, 0.0313, 0.0306, 0.0382, 0.0427, 0.2888, 0.0308], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0086, 0.0087, 0.0080, 0.0102, 0.0088, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:14:59,830 INFO [train.py:876] (3/4) Epoch 11, batch 4000, loss[loss=0.1176, simple_loss=0.1447, pruned_loss=0.04522, over 5605.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1432, pruned_loss=0.04438, over 1089641.96 frames. ], batch size: 40, lr: 7.33e-03, grad_scale: 8.0 +2022-11-16 04:15:08,072 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.537e+02 1.803e+02 2.088e+02 3.858e+02, threshold=3.606e+02, percent-clipped=2.0 +2022-11-16 04:15:28,267 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4521, 2.9666, 3.1354, 2.8288, 1.8673, 2.9768, 2.0675, 2.6438], + device='cuda:3'), covar=tensor([0.0286, 0.0139, 0.0121, 0.0232, 0.0386, 0.0144, 0.0360, 0.0140], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0172, 0.0178, 0.0200, 0.0189, 0.0177, 0.0188, 0.0179], + 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') +2022-11-16 04:15:59,086 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1308, 3.5721, 2.5114, 3.3318, 2.6995, 2.6480, 1.8499, 2.9788], + device='cuda:3'), covar=tensor([0.1302, 0.0291, 0.0998, 0.0359, 0.1040, 0.0968, 0.1870, 0.0504], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0141, 0.0159, 0.0145, 0.0176, 0.0167, 0.0162, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:16:07,302 INFO [train.py:876] (3/4) Epoch 11, batch 4100, loss[loss=0.09722, simple_loss=0.1425, pruned_loss=0.02599, over 5683.00 frames. ], tot_loss[loss=0.1156, simple_loss=0.1427, pruned_loss=0.04429, over 1084652.09 frames. ], batch size: 19, lr: 7.32e-03, grad_scale: 8.0 +2022-11-16 04:16:15,768 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.454e+02 1.745e+02 2.235e+02 4.051e+02, threshold=3.490e+02, percent-clipped=2.0 +2022-11-16 04:16:19,238 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:16:19,295 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6810, 2.1327, 3.0990, 2.6700, 3.4314, 2.2455, 3.0244, 3.6330], + device='cuda:3'), covar=tensor([0.0713, 0.2128, 0.1039, 0.2061, 0.0776, 0.1920, 0.1406, 0.0944], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0193, 0.0216, 0.0212, 0.0240, 0.0196, 0.0228, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:16:47,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2820, 2.1644, 2.5356, 3.3915, 3.3892, 2.6608, 2.1716, 3.3969], + device='cuda:3'), covar=tensor([0.1024, 0.2988, 0.2081, 0.2412, 0.1322, 0.2730, 0.2255, 0.0927], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0201, 0.0190, 0.0307, 0.0222, 0.0203, 0.0189, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:17:00,349 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:17:05,726 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9032, 2.4721, 3.3855, 3.0332, 3.7452, 2.3918, 3.1897, 3.8714], + device='cuda:3'), covar=tensor([0.0671, 0.1730, 0.1026, 0.1795, 0.0626, 0.1889, 0.1508, 0.0906], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0193, 0.0215, 0.0211, 0.0239, 0.0196, 0.0226, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:17:11,630 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2022-11-16 04:17:14,931 INFO [train.py:876] (3/4) Epoch 11, batch 4200, loss[loss=0.1026, simple_loss=0.1331, pruned_loss=0.03608, over 5741.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1421, pruned_loss=0.0435, over 1088550.54 frames. ], batch size: 13, lr: 7.32e-03, grad_scale: 8.0 +2022-11-16 04:17:17,657 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:17:23,250 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.852e+01 1.372e+02 1.800e+02 2.122e+02 4.143e+02, threshold=3.599e+02, percent-clipped=4.0 +2022-11-16 04:17:34,014 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:17:49,769 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:17:53,750 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:17:54,425 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7506, 1.0823, 1.2594, 0.9941, 1.3854, 1.4919, 0.8645, 1.2093], + device='cuda:3'), covar=tensor([0.0202, 0.0654, 0.0351, 0.0576, 0.0964, 0.0651, 0.0576, 0.0519], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0022, 0.0015, 0.0019, 0.0015, 0.0014, 0.0021, 0.0015], + device='cuda:3'), out_proj_covar=tensor([7.7620e-05, 1.0814e-04, 8.1528e-05, 9.5489e-05, 8.2004e-05, 7.8293e-05, + 1.0199e-04, 7.9094e-05], device='cuda:3') +2022-11-16 04:18:06,028 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:18:16,727 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6101, 1.3752, 1.5089, 1.2242, 1.6703, 1.4435, 1.1255, 0.8085], + device='cuda:3'), covar=tensor([0.0029, 0.0044, 0.0030, 0.0053, 0.0036, 0.0055, 0.0050, 0.0065], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0025, 0.0025, 0.0034, 0.0028, 0.0027, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.4917e-05, 2.3152e-05, 2.2326e-05, 3.2994e-05, 2.6076e-05, 2.5679e-05, + 3.2019e-05, 3.0734e-05], device='cuda:3') +2022-11-16 04:18:21,656 INFO [train.py:876] (3/4) Epoch 11, batch 4300, loss[loss=0.08962, simple_loss=0.1312, pruned_loss=0.02402, over 5528.00 frames. ], tot_loss[loss=0.1164, simple_loss=0.1436, pruned_loss=0.0446, over 1088761.08 frames. ], batch size: 17, lr: 7.31e-03, grad_scale: 8.0 +2022-11-16 04:18:25,914 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:18:30,433 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.745e+01 1.515e+02 1.890e+02 2.345e+02 3.579e+02, threshold=3.779e+02, percent-clipped=0.0 +2022-11-16 04:19:10,234 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5227, 1.9572, 1.8680, 1.2788, 1.8046, 2.3217, 1.9694, 2.1670], + device='cuda:3'), covar=tensor([0.2063, 0.1564, 0.1918, 0.2887, 0.1289, 0.0998, 0.1103, 0.1404], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0182, 0.0166, 0.0182, 0.0181, 0.0197, 0.0167, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:19:11,521 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:19:28,273 INFO [train.py:876] (3/4) Epoch 11, batch 4400, loss[loss=0.07865, simple_loss=0.1049, pruned_loss=0.02618, over 5330.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1438, pruned_loss=0.04468, over 1085536.02 frames. ], batch size: 9, lr: 7.31e-03, grad_scale: 8.0 +2022-11-16 04:19:37,948 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.506e+01 1.521e+02 1.875e+02 2.343e+02 5.225e+02, threshold=3.749e+02, percent-clipped=3.0 +2022-11-16 04:19:52,383 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:20:19,050 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:20:19,142 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8769, 1.0174, 0.9703, 1.0901, 1.2235, 1.0195, 0.6384, 1.2934], + device='cuda:3'), covar=tensor([0.0057, 0.0039, 0.0058, 0.0042, 0.0046, 0.0053, 0.0092, 0.0045], + device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0053, 0.0053, 0.0056, 0.0055, 0.0050, 0.0050, 0.0048], + device='cuda:3'), out_proj_covar=tensor([5.1444e-05, 4.6917e-05, 4.6967e-05, 5.0301e-05, 4.8918e-05, 4.3901e-05, + 4.4908e-05, 4.2297e-05], device='cuda:3') +2022-11-16 04:20:29,850 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:20:36,411 INFO [train.py:876] (3/4) Epoch 11, batch 4500, loss[loss=0.09967, simple_loss=0.1387, pruned_loss=0.03032, over 5571.00 frames. ], tot_loss[loss=0.1142, simple_loss=0.1416, pruned_loss=0.04337, over 1083781.11 frames. ], batch size: 16, lr: 7.31e-03, grad_scale: 8.0 +2022-11-16 04:20:37,166 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.8861, 5.2476, 5.7347, 5.2337, 5.9351, 5.6143, 4.8597, 5.8067], + device='cuda:3'), covar=tensor([0.0299, 0.0266, 0.0311, 0.0274, 0.0223, 0.0181, 0.0212, 0.0211], + device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0149, 0.0108, 0.0141, 0.0171, 0.0103, 0.0122, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:20:45,512 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.537e+01 1.435e+02 1.838e+02 2.199e+02 4.502e+02, threshold=3.675e+02, percent-clipped=1.0 +2022-11-16 04:21:07,685 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2022-11-16 04:21:11,333 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:21:11,934 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:21:14,575 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:21:44,388 INFO [train.py:876] (3/4) Epoch 11, batch 4600, loss[loss=0.09526, simple_loss=0.1235, pruned_loss=0.0335, over 5193.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1427, pruned_loss=0.0449, over 1078673.32 frames. ], batch size: 7, lr: 7.30e-03, grad_scale: 8.0 +2022-11-16 04:21:52,864 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.709e+01 1.569e+02 2.032e+02 2.456e+02 5.240e+02, threshold=4.063e+02, percent-clipped=2.0 +2022-11-16 04:21:53,354 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:21:56,318 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:22:04,279 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0710, 1.2156, 1.0966, 1.0898, 1.3822, 1.1534, 0.7474, 1.3028], + device='cuda:3'), covar=tensor([0.0061, 0.0042, 0.0048, 0.0055, 0.0048, 0.0048, 0.0080, 0.0054], + device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0053, 0.0053, 0.0057, 0.0056, 0.0051, 0.0050, 0.0048], + device='cuda:3'), out_proj_covar=tensor([5.2230e-05, 4.7291e-05, 4.6783e-05, 5.1037e-05, 4.9494e-05, 4.4265e-05, + 4.5236e-05, 4.2680e-05], device='cuda:3') +2022-11-16 04:22:13,949 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1527, 4.0163, 2.6217, 3.7981, 3.1380, 2.7191, 2.2025, 3.3745], + device='cuda:3'), covar=tensor([0.1504, 0.0324, 0.1257, 0.0392, 0.0961, 0.0994, 0.1866, 0.0521], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0142, 0.0158, 0.0145, 0.0176, 0.0166, 0.0163, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:22:50,875 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 04:22:52,445 INFO [train.py:876] (3/4) Epoch 11, batch 4700, loss[loss=0.1091, simple_loss=0.1351, pruned_loss=0.04153, over 5709.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1416, pruned_loss=0.04385, over 1079144.52 frames. ], batch size: 15, lr: 7.30e-03, grad_scale: 8.0 +2022-11-16 04:23:00,908 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.694e+01 1.424e+02 1.701e+02 2.094e+02 3.279e+02, threshold=3.401e+02, percent-clipped=0.0 +2022-11-16 04:23:12,370 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:23:42,072 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:23:45,064 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3202, 4.4058, 2.9091, 4.2602, 3.4355, 2.9484, 2.4717, 3.7148], + device='cuda:3'), covar=tensor([0.1580, 0.0230, 0.1095, 0.0319, 0.0757, 0.0997, 0.1922, 0.0456], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0141, 0.0158, 0.0144, 0.0175, 0.0166, 0.0163, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:24:00,657 INFO [train.py:876] (3/4) Epoch 11, batch 4800, loss[loss=0.09814, simple_loss=0.1255, pruned_loss=0.0354, over 5553.00 frames. ], tot_loss[loss=0.1144, simple_loss=0.1413, pruned_loss=0.04373, over 1079310.32 frames. ], batch size: 13, lr: 7.29e-03, grad_scale: 8.0 +2022-11-16 04:24:08,692 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2853, 4.3024, 2.8069, 4.1718, 3.3947, 2.8463, 2.1512, 3.7453], + device='cuda:3'), covar=tensor([0.1529, 0.0232, 0.1191, 0.0314, 0.0649, 0.1044, 0.2048, 0.0319], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0141, 0.0158, 0.0144, 0.0176, 0.0166, 0.0164, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:24:09,189 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.251e+01 1.590e+02 1.859e+02 2.447e+02 5.021e+02, threshold=3.719e+02, percent-clipped=6.0 +2022-11-16 04:24:15,207 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:24:32,378 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:24:58,692 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7445, 3.6111, 3.6373, 3.7720, 3.5439, 3.2053, 4.1353, 3.5867], + device='cuda:3'), covar=tensor([0.0505, 0.0930, 0.0570, 0.1223, 0.0562, 0.0488, 0.0780, 0.0812], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0109, 0.0094, 0.0119, 0.0089, 0.0080, 0.0144, 0.0102], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:25:05,986 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:25:09,013 INFO [train.py:876] (3/4) Epoch 11, batch 4900, loss[loss=0.104, simple_loss=0.147, pruned_loss=0.03053, over 5728.00 frames. ], tot_loss[loss=0.1139, simple_loss=0.1408, pruned_loss=0.04349, over 1076274.41 frames. ], batch size: 15, lr: 7.29e-03, grad_scale: 8.0 +2022-11-16 04:25:14,424 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:25:17,037 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:25:17,557 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.666e+01 1.544e+02 1.877e+02 2.554e+02 4.573e+02, threshold=3.753e+02, percent-clipped=4.0 +2022-11-16 04:25:37,952 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4240, 4.1722, 2.8089, 3.9639, 3.2430, 2.9323, 2.3622, 3.5944], + device='cuda:3'), covar=tensor([0.1281, 0.0190, 0.0999, 0.0359, 0.0731, 0.0891, 0.1778, 0.0371], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0141, 0.0158, 0.0145, 0.0177, 0.0166, 0.0164, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:25:46,985 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:25:47,007 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:26:15,139 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4716, 4.9263, 5.2843, 4.9451, 5.5656, 5.4530, 4.7218, 5.4800], + device='cuda:3'), covar=tensor([0.0307, 0.0331, 0.0378, 0.0271, 0.0281, 0.0143, 0.0237, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0151, 0.0111, 0.0143, 0.0176, 0.0105, 0.0125, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:26:16,784 INFO [train.py:876] (3/4) Epoch 11, batch 5000, loss[loss=0.1112, simple_loss=0.1484, pruned_loss=0.03705, over 5692.00 frames. ], tot_loss[loss=0.1135, simple_loss=0.141, pruned_loss=0.04298, over 1081932.07 frames. ], batch size: 28, lr: 7.28e-03, grad_scale: 8.0 +2022-11-16 04:26:25,204 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.970e+01 1.490e+02 1.911e+02 2.304e+02 5.675e+02, threshold=3.822e+02, percent-clipped=3.0 +2022-11-16 04:26:28,007 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:26:36,345 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 04:26:41,519 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2700, 4.2873, 3.2506, 1.8442, 3.9671, 1.7995, 3.9184, 1.9378], + device='cuda:3'), covar=tensor([0.1619, 0.0156, 0.0762, 0.2085, 0.0218, 0.1780, 0.0249, 0.1739], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0102, 0.0111, 0.0112, 0.0100, 0.0120, 0.0098, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:27:09,052 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:27:23,476 INFO [train.py:876] (3/4) Epoch 11, batch 5100, loss[loss=0.09523, simple_loss=0.1269, pruned_loss=0.03178, over 5580.00 frames. ], tot_loss[loss=0.1147, simple_loss=0.1418, pruned_loss=0.04374, over 1084538.86 frames. ], batch size: 23, lr: 7.28e-03, grad_scale: 8.0 +2022-11-16 04:27:30,760 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2634, 4.1887, 2.8154, 3.9335, 3.2953, 3.0234, 2.2039, 3.4952], + device='cuda:3'), covar=tensor([0.1555, 0.0193, 0.0948, 0.0320, 0.0708, 0.0877, 0.1924, 0.0494], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0143, 0.0160, 0.0147, 0.0179, 0.0168, 0.0165, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:27:32,555 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.765e+01 1.557e+02 2.003e+02 2.576e+02 4.677e+02, threshold=4.007e+02, percent-clipped=1.0 +2022-11-16 04:27:54,809 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:28:03,828 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7965, 1.7071, 1.7852, 1.4865, 1.4498, 1.4599, 1.4644, 1.6925], + device='cuda:3'), covar=tensor([0.0057, 0.0065, 0.0042, 0.0053, 0.0058, 0.0043, 0.0044, 0.0051], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0055, 0.0054, 0.0058, 0.0057, 0.0052, 0.0051, 0.0049], + device='cuda:3'), out_proj_covar=tensor([5.3302e-05, 4.8631e-05, 4.7736e-05, 5.2171e-05, 5.0837e-05, 4.5890e-05, + 4.5878e-05, 4.3131e-05], device='cuda:3') +2022-11-16 04:28:07,234 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-16 04:28:07,694 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1233, 2.9350, 2.8073, 1.4783, 2.7733, 3.0963, 2.9341, 3.2825], + device='cuda:3'), covar=tensor([0.1959, 0.1416, 0.1223, 0.3096, 0.0698, 0.0964, 0.0475, 0.0884], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0179, 0.0163, 0.0182, 0.0182, 0.0196, 0.0165, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:28:20,291 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2804, 5.2563, 5.2658, 5.0626, 4.7529, 4.8506, 5.7314, 5.4566], + device='cuda:3'), covar=tensor([0.0329, 0.0795, 0.0337, 0.1107, 0.0425, 0.0278, 0.0681, 0.0362], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0109, 0.0094, 0.0119, 0.0089, 0.0080, 0.0145, 0.0102], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:28:26,617 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:28:30,551 INFO [train.py:876] (3/4) Epoch 11, batch 5200, loss[loss=0.0813, simple_loss=0.1145, pruned_loss=0.02407, over 5350.00 frames. ], tot_loss[loss=0.1137, simple_loss=0.1418, pruned_loss=0.04279, over 1089922.56 frames. ], batch size: 9, lr: 7.27e-03, grad_scale: 8.0 +2022-11-16 04:28:30,719 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5037, 1.2275, 1.1824, 0.7967, 1.1816, 1.3873, 0.6484, 1.0189], + device='cuda:3'), covar=tensor([0.0275, 0.0433, 0.0367, 0.0738, 0.0351, 0.0237, 0.0874, 0.0440], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0023, 0.0016, 0.0020, 0.0016, 0.0014, 0.0022, 0.0015], + device='cuda:3'), out_proj_covar=tensor([8.1310e-05, 1.1013e-04, 8.4073e-05, 9.8358e-05, 8.4888e-05, 7.9580e-05, + 1.0597e-04, 8.1142e-05], device='cuda:3') +2022-11-16 04:28:35,984 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:28:38,576 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:28:39,079 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.507e+02 1.817e+02 2.273e+02 5.327e+02, threshold=3.634e+02, percent-clipped=3.0 +2022-11-16 04:28:52,007 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2482, 2.2076, 2.5264, 3.4878, 3.3051, 2.6886, 2.3001, 3.4010], + device='cuda:3'), covar=tensor([0.1104, 0.2495, 0.2229, 0.1900, 0.1245, 0.2422, 0.1989, 0.1462], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0201, 0.0191, 0.0304, 0.0226, 0.0205, 0.0193, 0.0247], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:29:02,646 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 +2022-11-16 04:29:04,827 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:29:08,139 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:29:08,472 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-16 04:29:10,794 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:29:12,240 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8380, 1.3961, 2.0092, 1.6478, 1.7911, 1.7720, 1.2567, 1.6713], + device='cuda:3'), covar=tensor([0.0074, 0.0100, 0.0035, 0.0066, 0.0141, 0.0063, 0.0054, 0.0045], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0026, 0.0026, 0.0034, 0.0029, 0.0028, 0.0034, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.5288e-05, 2.3937e-05, 2.3135e-05, 3.3684e-05, 2.6989e-05, 2.6260e-05, + 3.2774e-05, 3.1001e-05], device='cuda:3') +2022-11-16 04:29:13,565 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7595, 1.9437, 2.3072, 3.1255, 3.0167, 2.4681, 2.0871, 3.1031], + device='cuda:3'), covar=tensor([0.2021, 0.3242, 0.2203, 0.1983, 0.1266, 0.2767, 0.2285, 0.1105], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0203, 0.0191, 0.0306, 0.0226, 0.0206, 0.0193, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:29:38,674 INFO [train.py:876] (3/4) Epoch 11, batch 5300, loss[loss=0.07377, simple_loss=0.09704, pruned_loss=0.02525, over 5047.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1411, pruned_loss=0.0427, over 1082569.66 frames. ], batch size: 5, lr: 7.27e-03, grad_scale: 8.0 +2022-11-16 04:29:46,506 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:29:47,076 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.555e+01 1.542e+02 1.854e+02 2.251e+02 5.839e+02, threshold=3.709e+02, percent-clipped=3.0 +2022-11-16 04:30:35,923 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2919, 4.7756, 5.1373, 4.7361, 5.3609, 5.2107, 4.5702, 5.3482], + device='cuda:3'), covar=tensor([0.0349, 0.0372, 0.0351, 0.0320, 0.0324, 0.0253, 0.0289, 0.0256], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0151, 0.0110, 0.0143, 0.0176, 0.0105, 0.0124, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:30:46,414 INFO [train.py:876] (3/4) Epoch 11, batch 5400, loss[loss=0.06191, simple_loss=0.09574, pruned_loss=0.01404, over 5129.00 frames. ], tot_loss[loss=0.1118, simple_loss=0.1404, pruned_loss=0.04159, over 1084997.32 frames. ], batch size: 7, lr: 7.26e-03, grad_scale: 16.0 +2022-11-16 04:30:47,686 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 04:30:55,255 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.532e+02 1.889e+02 2.326e+02 4.779e+02, threshold=3.778e+02, percent-clipped=5.0 +2022-11-16 04:31:33,439 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 04:31:33,778 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6689, 3.6153, 3.8734, 2.0709, 3.5251, 4.1004, 3.7464, 4.3558], + device='cuda:3'), covar=tensor([0.1966, 0.1136, 0.0609, 0.2606, 0.0744, 0.0412, 0.0612, 0.0456], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0181, 0.0165, 0.0184, 0.0184, 0.0197, 0.0167, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:31:46,105 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8453, 3.1434, 2.2387, 2.8462, 2.3211, 2.3339, 1.8868, 2.7232], + device='cuda:3'), covar=tensor([0.1658, 0.0374, 0.1340, 0.0632, 0.1491, 0.1242, 0.2073, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0141, 0.0158, 0.0145, 0.0177, 0.0166, 0.0163, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:31:55,269 INFO [train.py:876] (3/4) Epoch 11, batch 5500, loss[loss=0.1288, simple_loss=0.1473, pruned_loss=0.05515, over 4678.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1434, pruned_loss=0.04463, over 1083296.92 frames. ], batch size: 135, lr: 7.26e-03, grad_scale: 16.0 +2022-11-16 04:32:02,850 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:32:04,000 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.487e+02 1.903e+02 2.328e+02 5.113e+02, threshold=3.806e+02, percent-clipped=2.0 +2022-11-16 04:32:08,646 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0337, 3.6211, 2.4327, 3.3609, 2.7930, 2.5151, 1.9766, 3.0473], + device='cuda:3'), covar=tensor([0.1468, 0.0310, 0.1168, 0.0423, 0.1062, 0.1048, 0.1994, 0.0489], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0140, 0.0157, 0.0144, 0.0177, 0.0165, 0.0162, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:32:29,974 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:32:37,474 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5827, 1.2138, 0.9208, 0.8079, 1.1256, 0.8922, 0.6943, 1.2047], + device='cuda:3'), covar=tensor([0.0081, 0.0037, 0.0061, 0.0050, 0.0050, 0.0056, 0.0087, 0.0053], + device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0053, 0.0052, 0.0057, 0.0055, 0.0051, 0.0049, 0.0047], + device='cuda:3'), out_proj_covar=tensor([5.1562e-05, 4.7067e-05, 4.5974e-05, 5.1075e-05, 4.8628e-05, 4.4377e-05, + 4.3850e-05, 4.1506e-05], device='cuda:3') +2022-11-16 04:32:43,995 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 04:32:50,513 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 +2022-11-16 04:32:52,381 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7506, 1.5340, 1.7545, 1.8066, 1.8758, 1.2961, 1.7102, 1.8243], + device='cuda:3'), covar=tensor([0.0401, 0.0780, 0.0466, 0.0361, 0.0415, 0.0905, 0.0507, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0192, 0.0211, 0.0209, 0.0236, 0.0192, 0.0223, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:33:01,993 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:33:02,617 INFO [train.py:876] (3/4) Epoch 11, batch 5600, loss[loss=0.1013, simple_loss=0.135, pruned_loss=0.03387, over 5646.00 frames. ], tot_loss[loss=0.1156, simple_loss=0.1428, pruned_loss=0.04418, over 1080743.31 frames. ], batch size: 32, lr: 7.25e-03, grad_scale: 16.0 +2022-11-16 04:33:11,304 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:33:11,813 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.428e+01 1.405e+02 1.623e+02 2.102e+02 3.893e+02, threshold=3.245e+02, percent-clipped=1.0 +2022-11-16 04:33:20,975 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2022-11-16 04:33:43,985 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:33:57,461 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2022-11-16 04:34:11,589 INFO [train.py:876] (3/4) Epoch 11, batch 5700, loss[loss=0.1153, simple_loss=0.1392, pruned_loss=0.04565, over 4707.00 frames. ], tot_loss[loss=0.1136, simple_loss=0.1413, pruned_loss=0.04293, over 1080725.01 frames. ], batch size: 135, lr: 7.25e-03, grad_scale: 16.0 +2022-11-16 04:34:20,539 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.114e+01 1.453e+02 1.885e+02 2.462e+02 5.318e+02, threshold=3.770e+02, percent-clipped=5.0 +2022-11-16 04:34:30,137 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6814, 0.6639, 0.6999, 0.4818, 0.5791, 0.5990, 0.3123, 0.6342], + device='cuda:3'), covar=tensor([0.0269, 0.0393, 0.0350, 0.0370, 0.0270, 0.0282, 0.0693, 0.0276], + device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0023, 0.0016, 0.0019, 0.0016, 0.0014, 0.0021, 0.0015], + device='cuda:3'), out_proj_covar=tensor([8.0000e-05, 1.1017e-04, 8.3790e-05, 9.8180e-05, 8.4740e-05, 7.9339e-05, + 1.0486e-04, 8.0020e-05], device='cuda:3') +2022-11-16 04:34:37,880 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:35:18,649 INFO [train.py:876] (3/4) Epoch 11, batch 5800, loss[loss=0.1633, simple_loss=0.1751, pruned_loss=0.07568, over 5561.00 frames. ], tot_loss[loss=0.1127, simple_loss=0.1404, pruned_loss=0.04255, over 1078131.09 frames. ], batch size: 43, lr: 7.24e-03, grad_scale: 16.0 +2022-11-16 04:35:18,837 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:35:27,563 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.244e+01 1.530e+02 1.946e+02 2.399e+02 7.039e+02, threshold=3.892e+02, percent-clipped=3.0 +2022-11-16 04:35:42,016 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2022-11-16 04:35:42,330 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:35:51,981 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 +2022-11-16 04:36:03,412 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 04:36:12,329 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6463, 2.0782, 1.6701, 1.2732, 1.5358, 1.9563, 1.9363, 2.1794], + device='cuda:3'), covar=tensor([0.1732, 0.1364, 0.1957, 0.2633, 0.1495, 0.1065, 0.0861, 0.1288], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0182, 0.0164, 0.0185, 0.0183, 0.0199, 0.0169, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:36:23,760 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:36:26,217 INFO [train.py:876] (3/4) Epoch 11, batch 5900, loss[loss=0.1075, simple_loss=0.1461, pruned_loss=0.03447, over 5599.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1406, pruned_loss=0.04295, over 1078513.05 frames. ], batch size: 24, lr: 7.24e-03, grad_scale: 16.0 +2022-11-16 04:36:34,706 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.483e+02 1.879e+02 2.262e+02 4.846e+02, threshold=3.758e+02, percent-clipped=3.0 +2022-11-16 04:37:33,703 INFO [train.py:876] (3/4) Epoch 11, batch 6000, loss[loss=0.05682, simple_loss=0.08333, pruned_loss=0.01516, over 5167.00 frames. ], tot_loss[loss=0.1121, simple_loss=0.1399, pruned_loss=0.0421, over 1079962.18 frames. ], batch size: 7, lr: 7.24e-03, grad_scale: 16.0 +2022-11-16 04:37:33,703 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 04:37:45,797 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8134, 4.5113, 3.3220, 2.2420, 4.1440, 2.3531, 4.1180, 2.6415], + device='cuda:3'), covar=tensor([0.1167, 0.0102, 0.0564, 0.1921, 0.0163, 0.1376, 0.0152, 0.1595], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0114, 0.0114, 0.0101, 0.0122, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:37:51,351 INFO [train.py:908] (3/4) Epoch 11, validation: loss=0.1691, simple_loss=0.1834, pruned_loss=0.07744, over 1530663.00 frames. +2022-11-16 04:37:51,352 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 04:37:59,341 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9818, 1.6684, 1.7506, 1.2828, 1.7960, 1.7917, 1.4144, 1.2172], + device='cuda:3'), covar=tensor([0.0031, 0.0037, 0.0066, 0.0054, 0.0050, 0.0041, 0.0039, 0.0050], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0025, 0.0025, 0.0034, 0.0028, 0.0026, 0.0032, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.4541e-05, 2.2961e-05, 2.2658e-05, 3.2832e-05, 2.6323e-05, 2.5077e-05, + 3.1001e-05, 3.0053e-05], device='cuda:3') +2022-11-16 04:37:59,809 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.487e+02 1.823e+02 2.133e+02 3.868e+02, threshold=3.646e+02, percent-clipped=1.0 +2022-11-16 04:38:19,453 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4930, 5.4563, 3.9209, 2.5521, 5.0090, 2.5122, 4.7616, 2.6182], + device='cuda:3'), covar=tensor([0.0886, 0.0101, 0.0419, 0.1612, 0.0139, 0.1379, 0.0252, 0.1342], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0104, 0.0113, 0.0113, 0.0100, 0.0121, 0.0098, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:38:56,300 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:38:59,574 INFO [train.py:876] (3/4) Epoch 11, batch 6100, loss[loss=0.1258, simple_loss=0.1494, pruned_loss=0.05109, over 5717.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1392, pruned_loss=0.042, over 1078263.62 frames. ], batch size: 27, lr: 7.23e-03, grad_scale: 16.0 +2022-11-16 04:39:08,232 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.645e+01 1.456e+02 1.777e+02 2.158e+02 4.181e+02, threshold=3.555e+02, percent-clipped=3.0 +2022-11-16 04:39:34,836 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2896, 1.1424, 1.9537, 1.6293, 1.8863, 1.8645, 1.7721, 1.6812], + device='cuda:3'), covar=tensor([0.0041, 0.0076, 0.0027, 0.0060, 0.0063, 0.0146, 0.0042, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0024, 0.0025, 0.0033, 0.0028, 0.0026, 0.0032, 0.0031], + device='cuda:3'), out_proj_covar=tensor([2.4090e-05, 2.2784e-05, 2.2299e-05, 3.2583e-05, 2.6079e-05, 2.4837e-05, + 3.0664e-05, 2.9628e-05], device='cuda:3') +2022-11-16 04:39:44,979 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:39:51,847 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6439, 4.4877, 4.4548, 4.3459, 4.7462, 4.5131, 4.2567, 4.7969], + device='cuda:3'), covar=tensor([0.0750, 0.0610, 0.0787, 0.0862, 0.0798, 0.0716, 0.0580, 0.0753], + device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0148, 0.0107, 0.0141, 0.0173, 0.0104, 0.0122, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:40:01,682 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:40:07,448 INFO [train.py:876] (3/4) Epoch 11, batch 6200, loss[loss=0.09024, simple_loss=0.1324, pruned_loss=0.02405, over 5511.00 frames. ], tot_loss[loss=0.1138, simple_loss=0.1413, pruned_loss=0.04313, over 1081021.67 frames. ], batch size: 17, lr: 7.23e-03, grad_scale: 16.0 +2022-11-16 04:40:15,897 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0822, 2.6148, 2.7307, 1.5487, 2.7350, 3.0716, 2.9340, 3.2599], + device='cuda:3'), covar=tensor([0.1905, 0.1997, 0.1122, 0.2958, 0.0761, 0.0889, 0.0608, 0.1034], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0180, 0.0161, 0.0181, 0.0180, 0.0196, 0.0167, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:40:16,300 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.628e+01 1.452e+02 1.821e+02 2.109e+02 4.865e+02, threshold=3.642e+02, percent-clipped=3.0 +2022-11-16 04:40:17,681 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:40:37,743 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:41:15,713 INFO [train.py:876] (3/4) Epoch 11, batch 6300, loss[loss=0.1541, simple_loss=0.1676, pruned_loss=0.07027, over 5457.00 frames. ], tot_loss[loss=0.1135, simple_loss=0.1413, pruned_loss=0.04289, over 1082135.84 frames. ], batch size: 64, lr: 7.22e-03, grad_scale: 16.0 +2022-11-16 04:41:19,125 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:41:22,392 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9467, 2.2463, 2.5503, 2.2410, 1.4361, 2.3034, 1.6830, 1.9543], + device='cuda:3'), covar=tensor([0.0280, 0.0146, 0.0120, 0.0213, 0.0414, 0.0169, 0.0386, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0175, 0.0179, 0.0201, 0.0191, 0.0178, 0.0190, 0.0181], + 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') +2022-11-16 04:41:24,115 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.477e+02 1.821e+02 2.187e+02 5.336e+02, threshold=3.643e+02, percent-clipped=2.0 +2022-11-16 04:41:33,970 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5069, 4.8954, 5.3472, 4.8700, 5.6114, 5.4248, 4.7510, 5.5734], + device='cuda:3'), covar=tensor([0.0328, 0.0369, 0.0357, 0.0315, 0.0276, 0.0208, 0.0238, 0.0282], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0151, 0.0109, 0.0143, 0.0176, 0.0105, 0.0123, 0.0152], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:41:59,021 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2691, 3.0907, 3.0916, 2.8763, 1.8917, 3.0174, 2.0538, 2.7880], + device='cuda:3'), covar=tensor([0.0403, 0.0172, 0.0208, 0.0293, 0.0525, 0.0226, 0.0545, 0.0193], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0173, 0.0178, 0.0199, 0.0189, 0.0176, 0.0188, 0.0179], + 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') +2022-11-16 04:42:19,628 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:42:23,417 INFO [train.py:876] (3/4) Epoch 11, batch 6400, loss[loss=0.1008, simple_loss=0.1349, pruned_loss=0.03333, over 5734.00 frames. ], tot_loss[loss=0.1145, simple_loss=0.1423, pruned_loss=0.04335, over 1084125.94 frames. ], batch size: 17, lr: 7.22e-03, grad_scale: 16.0 +2022-11-16 04:42:24,302 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.66 vs. limit=5.0 +2022-11-16 04:42:32,273 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.453e+01 1.576e+02 1.936e+02 2.244e+02 4.119e+02, threshold=3.873e+02, percent-clipped=1.0 +2022-11-16 04:42:52,527 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:43:02,138 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.06 vs. limit=5.0 +2022-11-16 04:43:25,625 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:43:31,337 INFO [train.py:876] (3/4) Epoch 11, batch 6500, loss[loss=0.1163, simple_loss=0.1509, pruned_loss=0.04087, over 5612.00 frames. ], tot_loss[loss=0.1135, simple_loss=0.1414, pruned_loss=0.04281, over 1084570.67 frames. ], batch size: 23, lr: 7.21e-03, grad_scale: 16.0 +2022-11-16 04:43:40,081 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.904e+01 1.514e+02 1.877e+02 2.297e+02 4.296e+02, threshold=3.754e+02, percent-clipped=3.0 +2022-11-16 04:43:57,781 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:44:14,389 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:44:39,321 INFO [train.py:876] (3/4) Epoch 11, batch 6600, loss[loss=0.07936, simple_loss=0.1089, pruned_loss=0.02491, over 5541.00 frames. ], tot_loss[loss=0.1134, simple_loss=0.1413, pruned_loss=0.04275, over 1084815.64 frames. ], batch size: 10, lr: 7.21e-03, grad_scale: 16.0 +2022-11-16 04:44:39,404 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:44:47,233 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4139, 1.4243, 2.3108, 1.9678, 1.5175, 2.0181, 2.1961, 1.9481], + device='cuda:3'), covar=tensor([0.0036, 0.0153, 0.0029, 0.0040, 0.0137, 0.0048, 0.0042, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0026, 0.0026, 0.0034, 0.0029, 0.0027, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.4630e-05, 2.3861e-05, 2.3070e-05, 3.3353e-05, 2.7148e-05, 2.5643e-05, + 3.1689e-05, 3.0796e-05], device='cuda:3') +2022-11-16 04:44:47,249 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:44:47,746 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.843e+01 1.552e+02 2.036e+02 2.359e+02 4.515e+02, threshold=4.072e+02, percent-clipped=1.0 +2022-11-16 04:44:56,159 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:45:02,026 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-16 04:45:08,445 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:45:15,645 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:45:28,439 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:45:46,988 INFO [train.py:876] (3/4) Epoch 11, batch 6700, loss[loss=0.1226, simple_loss=0.1499, pruned_loss=0.04761, over 5119.00 frames. ], tot_loss[loss=0.1114, simple_loss=0.14, pruned_loss=0.0414, over 1091310.47 frames. ], batch size: 91, lr: 7.20e-03, grad_scale: 16.0 +2022-11-16 04:45:49,768 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:45:55,337 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.416e+01 1.570e+02 1.880e+02 2.430e+02 4.197e+02, threshold=3.759e+02, percent-clipped=3.0 +2022-11-16 04:45:56,833 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 04:45:56,974 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2022-11-16 04:46:05,014 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5588, 2.2747, 2.5876, 3.4545, 3.4491, 2.6442, 2.2604, 3.5960], + device='cuda:3'), covar=tensor([0.0853, 0.2977, 0.2700, 0.3126, 0.1130, 0.3571, 0.2495, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0201, 0.0191, 0.0306, 0.0226, 0.0206, 0.0192, 0.0247], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:46:09,675 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2022-11-16 04:46:14,707 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4630, 2.2888, 2.5857, 3.2194, 3.3142, 2.4442, 2.1223, 3.5011], + device='cuda:3'), covar=tensor([0.1133, 0.2595, 0.2190, 0.3498, 0.1505, 0.3664, 0.2459, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0200, 0.0190, 0.0305, 0.0224, 0.0205, 0.0191, 0.0246], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:46:54,263 INFO [train.py:876] (3/4) Epoch 11, batch 6800, loss[loss=0.07621, simple_loss=0.1, pruned_loss=0.02619, over 5072.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1416, pruned_loss=0.04227, over 1091937.06 frames. ], batch size: 7, lr: 7.20e-03, grad_scale: 16.0 +2022-11-16 04:47:03,442 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.898e+01 1.544e+02 1.830e+02 2.349e+02 4.429e+02, threshold=3.660e+02, percent-clipped=3.0 +2022-11-16 04:47:05,941 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2022-11-16 04:47:28,937 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6769, 2.5769, 2.3885, 2.6875, 2.2236, 1.9917, 2.4384, 3.0237], + device='cuda:3'), covar=tensor([0.1176, 0.1461, 0.1724, 0.1376, 0.1485, 0.2726, 0.1500, 0.1906], + device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0103, 0.0104, 0.0100, 0.0091, 0.0099, 0.0095, 0.0079], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 04:47:38,078 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:47:39,366 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6377, 1.6288, 1.6112, 1.3039, 1.6496, 1.6282, 1.2944, 0.9046], + device='cuda:3'), covar=tensor([0.0023, 0.0025, 0.0026, 0.0047, 0.0028, 0.0063, 0.0038, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0025, 0.0025, 0.0034, 0.0029, 0.0027, 0.0033, 0.0032], + device='cuda:3'), out_proj_covar=tensor([2.4488e-05, 2.3269e-05, 2.2995e-05, 3.3301e-05, 2.7027e-05, 2.5247e-05, + 3.1403e-05, 3.0957e-05], device='cuda:3') +2022-11-16 04:47:43,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8581, 2.4929, 3.0190, 3.7768, 3.7800, 2.8587, 2.4261, 3.6689], + device='cuda:3'), covar=tensor([0.0560, 0.3423, 0.1953, 0.2465, 0.1195, 0.3106, 0.2414, 0.0740], + device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0198, 0.0188, 0.0302, 0.0224, 0.0204, 0.0188, 0.0244], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 04:47:45,436 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8006, 4.8337, 3.4599, 2.1115, 4.5452, 2.1187, 4.2310, 2.3653], + device='cuda:3'), covar=tensor([0.1308, 0.0124, 0.0548, 0.1872, 0.0181, 0.1532, 0.0238, 0.1502], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0102, 0.0114, 0.0111, 0.0099, 0.0119, 0.0097, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:48:02,160 INFO [train.py:876] (3/4) Epoch 11, batch 6900, loss[loss=0.08049, simple_loss=0.1113, pruned_loss=0.02485, over 5494.00 frames. ], tot_loss[loss=0.113, simple_loss=0.1413, pruned_loss=0.04233, over 1085082.29 frames. ], batch size: 10, lr: 7.19e-03, grad_scale: 16.0 +2022-11-16 04:48:02,287 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:48:10,567 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.132e+01 1.500e+02 1.797e+02 2.147e+02 3.952e+02, threshold=3.594e+02, percent-clipped=1.0 +2022-11-16 04:48:15,840 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 04:48:19,217 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:48:25,125 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2022-11-16 04:48:26,773 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:48:34,581 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:48:48,424 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:49:08,149 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:49:09,343 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:49:09,912 INFO [train.py:876] (3/4) Epoch 11, batch 7000, loss[loss=0.08278, simple_loss=0.1277, pruned_loss=0.01895, over 5731.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1411, pruned_loss=0.04258, over 1081074.67 frames. ], batch size: 17, lr: 7.19e-03, grad_scale: 16.0 +2022-11-16 04:49:16,823 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 04:49:18,677 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.908e+01 1.674e+02 1.897e+02 2.324e+02 4.828e+02, threshold=3.794e+02, percent-clipped=2.0 +2022-11-16 04:49:37,067 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9296, 2.4506, 2.3129, 1.7022, 2.6796, 2.8106, 2.7224, 2.9173], + device='cuda:3'), covar=tensor([0.1675, 0.1454, 0.1341, 0.2190, 0.0767, 0.0998, 0.0665, 0.0898], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0181, 0.0165, 0.0183, 0.0182, 0.0199, 0.0167, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:49:38,282 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:49:46,518 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2022-11-16 04:49:50,777 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4825, 1.1399, 1.3683, 0.8723, 1.3781, 1.3022, 1.1798, 0.8816], + device='cuda:3'), covar=tensor([0.0054, 0.0079, 0.0052, 0.0129, 0.0131, 0.0092, 0.0073, 0.0115], + device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0026, 0.0026, 0.0035, 0.0030, 0.0027, 0.0033, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.4949e-05, 2.3995e-05, 2.3335e-05, 3.3860e-05, 2.7815e-05, 2.5647e-05, + 3.1958e-05, 3.1634e-05], device='cuda:3') +2022-11-16 04:50:18,625 INFO [train.py:876] (3/4) Epoch 11, batch 7100, loss[loss=0.1082, simple_loss=0.1493, pruned_loss=0.03359, over 5702.00 frames. ], tot_loss[loss=0.1141, simple_loss=0.1416, pruned_loss=0.04327, over 1079823.70 frames. ], batch size: 19, lr: 7.19e-03, grad_scale: 16.0 +2022-11-16 04:50:18,829 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5527, 2.3335, 3.2793, 2.9064, 3.1867, 2.2759, 2.9981, 3.5827], + device='cuda:3'), covar=tensor([0.0867, 0.1399, 0.0834, 0.1381, 0.0810, 0.1559, 0.1087, 0.0695], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0194, 0.0217, 0.0214, 0.0243, 0.0197, 0.0225, 0.0232], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:50:20,097 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:50:21,296 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6586, 3.7216, 3.5584, 3.7758, 3.6046, 3.3913, 4.1943, 3.5642], + device='cuda:3'), covar=tensor([0.0488, 0.0672, 0.0451, 0.1146, 0.0567, 0.0364, 0.0642, 0.0665], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0113, 0.0097, 0.0123, 0.0091, 0.0082, 0.0150, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:50:27,110 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.428e+02 1.794e+02 2.274e+02 4.053e+02, threshold=3.587e+02, percent-clipped=1.0 +2022-11-16 04:50:37,759 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3541, 2.7149, 3.8900, 3.5532, 4.4748, 3.0177, 3.9390, 4.3093], + device='cuda:3'), covar=tensor([0.0578, 0.1408, 0.0982, 0.1312, 0.0462, 0.1346, 0.0924, 0.0585], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0190, 0.0213, 0.0210, 0.0238, 0.0194, 0.0221, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:50:40,994 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3520, 3.2104, 3.2219, 3.3379, 3.2623, 3.1367, 3.6611, 3.3115], + device='cuda:3'), covar=tensor([0.0514, 0.0973, 0.0536, 0.1337, 0.0652, 0.0469, 0.0842, 0.0780], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0112, 0.0096, 0.0122, 0.0091, 0.0081, 0.0148, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:51:27,327 INFO [train.py:876] (3/4) Epoch 11, batch 7200, loss[loss=0.1833, simple_loss=0.1716, pruned_loss=0.09752, over 3104.00 frames. ], tot_loss[loss=0.1147, simple_loss=0.1422, pruned_loss=0.04357, over 1076029.65 frames. ], batch size: 284, lr: 7.18e-03, grad_scale: 16.0 +2022-11-16 04:51:35,800 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.528e+01 1.486e+02 1.793e+02 2.179e+02 3.743e+02, threshold=3.587e+02, percent-clipped=3.0 +2022-11-16 04:51:40,436 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:51:40,483 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 04:52:11,196 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 04:52:11,248 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:52:57,788 INFO [train.py:876] (3/4) Epoch 12, batch 0, loss[loss=0.07902, simple_loss=0.1251, pruned_loss=0.0165, over 5694.00 frames. ], tot_loss[loss=0.07902, simple_loss=0.1251, pruned_loss=0.0165, over 5694.00 frames. ], batch size: 19, lr: 6.88e-03, grad_scale: 16.0 +2022-11-16 04:52:57,788 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 04:53:01,721 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2925, 3.4681, 2.6703, 3.2088, 2.5251, 2.7510, 2.1764, 3.0270], + device='cuda:3'), covar=tensor([0.1090, 0.0300, 0.0882, 0.0439, 0.1322, 0.0810, 0.1659, 0.0427], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0141, 0.0155, 0.0144, 0.0170, 0.0163, 0.0159, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:53:03,475 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9245, 3.6662, 3.7621, 3.4558, 3.9204, 3.7259, 3.7474, 3.8884], + device='cuda:3'), covar=tensor([0.0273, 0.0449, 0.0419, 0.0465, 0.0373, 0.0279, 0.0327, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0152, 0.0110, 0.0143, 0.0177, 0.0105, 0.0124, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 04:53:14,373 INFO [train.py:908] (3/4) Epoch 12, validation: loss=0.1725, simple_loss=0.1858, pruned_loss=0.07956, over 1530663.00 frames. +2022-11-16 04:53:14,374 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 04:53:15,916 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1493, 2.8759, 2.9405, 1.7265, 3.0136, 3.3401, 3.1218, 3.5300], + device='cuda:3'), covar=tensor([0.1792, 0.1315, 0.1077, 0.2717, 0.0613, 0.0722, 0.0759, 0.0689], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0183, 0.0166, 0.0187, 0.0184, 0.0203, 0.0168, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:53:31,369 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:53:36,410 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:53:43,646 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 04:53:45,398 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.136e+01 1.488e+02 1.843e+02 2.324e+02 4.228e+02, threshold=3.685e+02, percent-clipped=3.0 +2022-11-16 04:53:47,754 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:54:04,476 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2022-11-16 04:54:05,351 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9365, 3.8725, 3.8680, 3.5591, 3.9035, 3.6899, 1.6626, 4.0713], + device='cuda:3'), covar=tensor([0.0246, 0.0241, 0.0295, 0.0311, 0.0295, 0.0461, 0.2793, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0086, 0.0088, 0.0081, 0.0101, 0.0090, 0.0132, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:54:08,554 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:54:15,971 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 04:54:25,347 INFO [train.py:876] (3/4) Epoch 12, batch 100, loss[loss=0.09873, simple_loss=0.1364, pruned_loss=0.03051, over 5643.00 frames. ], tot_loss[loss=0.1115, simple_loss=0.1394, pruned_loss=0.04175, over 428196.42 frames. ], batch size: 32, lr: 6.87e-03, grad_scale: 16.0 +2022-11-16 04:54:42,316 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:54:43,309 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2022-11-16 04:54:52,891 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.598e+01 1.606e+02 2.099e+02 2.573e+02 5.035e+02, threshold=4.198e+02, percent-clipped=4.0 +2022-11-16 04:55:07,825 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9826, 3.0676, 2.5792, 3.0368, 2.5325, 2.7459, 3.2825, 3.7529], + device='cuda:3'), covar=tensor([0.1097, 0.1185, 0.1959, 0.1107, 0.1720, 0.2382, 0.1036, 0.0837], + device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0102, 0.0104, 0.0100, 0.0090, 0.0100, 0.0096, 0.0078], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 04:55:16,214 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9945, 3.9544, 3.8452, 3.8212, 4.0644, 3.9399, 1.5568, 4.1174], + device='cuda:3'), covar=tensor([0.0303, 0.0352, 0.0426, 0.0303, 0.0332, 0.0374, 0.3335, 0.0370], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0085, 0.0086, 0.0079, 0.0100, 0.0089, 0.0130, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:55:32,801 INFO [train.py:876] (3/4) Epoch 12, batch 200, loss[loss=0.167, simple_loss=0.1716, pruned_loss=0.0812, over 5107.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1407, pruned_loss=0.0429, over 689351.37 frames. ], batch size: 91, lr: 6.87e-03, grad_scale: 16.0 +2022-11-16 04:56:01,418 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.921e+01 1.528e+02 1.725e+02 2.144e+02 5.994e+02, threshold=3.450e+02, percent-clipped=2.0 +2022-11-16 04:56:05,490 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:56:28,181 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1441, 1.4837, 1.0234, 1.0072, 1.4842, 1.0649, 0.7675, 1.3387], + device='cuda:3'), covar=tensor([0.0066, 0.0047, 0.0062, 0.0063, 0.0050, 0.0059, 0.0093, 0.0055], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0056, 0.0061, 0.0059, 0.0054, 0.0053, 0.0051], + device='cuda:3'), out_proj_covar=tensor([5.4953e-05, 5.0772e-05, 4.9526e-05, 5.5186e-05, 5.2399e-05, 4.7537e-05, + 4.7411e-05, 4.4800e-05], device='cuda:3') +2022-11-16 04:56:34,006 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:56:37,805 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:56:40,372 INFO [train.py:876] (3/4) Epoch 12, batch 300, loss[loss=0.1116, simple_loss=0.1392, pruned_loss=0.04203, over 5547.00 frames. ], tot_loss[loss=0.1138, simple_loss=0.1413, pruned_loss=0.04315, over 845473.74 frames. ], batch size: 13, lr: 6.86e-03, grad_scale: 16.0 +2022-11-16 04:56:53,751 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:56:58,976 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0117, 1.8832, 1.9549, 1.9959, 1.8590, 1.3216, 1.9158, 2.1800], + device='cuda:3'), covar=tensor([0.1906, 0.1699, 0.2202, 0.1289, 0.1762, 0.2427, 0.1764, 0.0895], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0103, 0.0105, 0.0101, 0.0091, 0.0101, 0.0097, 0.0079], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 04:57:08,168 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.446e+01 1.444e+02 1.739e+02 2.202e+02 4.917e+02, threshold=3.479e+02, percent-clipped=4.0 +2022-11-16 04:57:14,550 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:57:25,857 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:57:38,840 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2290, 4.7225, 4.2652, 4.6804, 4.6389, 3.9342, 4.2943, 4.0922], + device='cuda:3'), covar=tensor([0.0379, 0.0345, 0.1224, 0.0419, 0.0464, 0.0431, 0.0434, 0.0525], + device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0181, 0.0278, 0.0175, 0.0222, 0.0174, 0.0190, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 04:57:40,563 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3550, 2.3359, 2.5541, 3.4380, 3.3264, 2.5406, 2.3305, 3.5077], + device='cuda:3'), covar=tensor([0.1159, 0.2691, 0.2574, 0.2450, 0.1237, 0.3325, 0.2081, 0.1359], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0196, 0.0189, 0.0299, 0.0220, 0.0203, 0.0189, 0.0243], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 04:57:42,650 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2022-11-16 04:57:47,431 INFO [train.py:876] (3/4) Epoch 12, batch 400, loss[loss=0.1344, simple_loss=0.1399, pruned_loss=0.06445, over 5018.00 frames. ], tot_loss[loss=0.1127, simple_loss=0.1408, pruned_loss=0.04228, over 944373.83 frames. ], batch size: 109, lr: 6.86e-03, grad_scale: 16.0 +2022-11-16 04:57:47,575 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:57:54,799 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7645, 4.6425, 3.6171, 1.9205, 4.3662, 1.8531, 4.4678, 2.3699], + device='cuda:3'), covar=tensor([0.1326, 0.0165, 0.0615, 0.2446, 0.0209, 0.2035, 0.0165, 0.1756], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0115, 0.0113, 0.0101, 0.0121, 0.0100, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:58:04,470 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:58:15,553 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.295e+01 1.482e+02 1.820e+02 2.321e+02 3.687e+02, threshold=3.641e+02, percent-clipped=1.0 +2022-11-16 04:58:21,500 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2022-11-16 04:58:21,829 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8563, 2.9723, 2.4501, 3.0485, 2.3340, 2.6345, 2.8099, 3.5652], + device='cuda:3'), covar=tensor([0.0876, 0.0947, 0.1860, 0.0927, 0.1346, 0.0916, 0.1128, 0.0812], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0103, 0.0104, 0.0101, 0.0091, 0.0101, 0.0097, 0.0078], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], + device='cuda:3') +2022-11-16 04:58:28,298 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:58:30,107 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3460, 4.1938, 2.7073, 3.8458, 3.0621, 2.8077, 2.1798, 3.3250], + device='cuda:3'), covar=tensor([0.1396, 0.0234, 0.1121, 0.0474, 0.0945, 0.1008, 0.1863, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0142, 0.0157, 0.0146, 0.0172, 0.0166, 0.0161, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:58:34,740 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:58:36,493 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:58:39,725 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7373, 3.4649, 4.4850, 4.2323, 5.0757, 4.1129, 4.5392, 4.7381], + device='cuda:3'), covar=tensor([0.0995, 0.1461, 0.0797, 0.1172, 0.0321, 0.1074, 0.1149, 0.0902], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0192, 0.0213, 0.0212, 0.0238, 0.0197, 0.0225, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 04:58:54,181 INFO [train.py:876] (3/4) Epoch 12, batch 500, loss[loss=0.09073, simple_loss=0.131, pruned_loss=0.02524, over 5722.00 frames. ], tot_loss[loss=0.1117, simple_loss=0.1402, pruned_loss=0.04158, over 1007298.18 frames. ], batch size: 15, lr: 6.86e-03, grad_scale: 16.0 +2022-11-16 04:58:54,308 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8610, 1.5777, 1.7890, 1.5634, 1.6323, 1.5541, 1.3303, 1.5106], + device='cuda:3'), covar=tensor([0.0050, 0.0061, 0.0029, 0.0064, 0.0111, 0.0294, 0.0085, 0.0059], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0027, 0.0026, 0.0035, 0.0031, 0.0028, 0.0034, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.5803e-05, 2.4913e-05, 2.3965e-05, 3.4573e-05, 2.8449e-05, 2.6190e-05, + 3.2792e-05, 3.2053e-05], device='cuda:3') +2022-11-16 04:58:54,994 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:58:55,672 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0900, 2.5232, 3.0988, 3.8216, 3.7093, 3.1022, 2.5629, 3.9008], + device='cuda:3'), covar=tensor([0.0645, 0.3325, 0.2061, 0.3240, 0.1113, 0.3101, 0.2461, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0197, 0.0189, 0.0299, 0.0222, 0.0202, 0.0189, 0.0243], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 04:59:05,062 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7757, 1.5039, 1.5892, 1.2538, 1.6637, 1.6291, 1.4289, 1.1508], + device='cuda:3'), covar=tensor([0.0035, 0.0054, 0.0037, 0.0072, 0.0074, 0.0060, 0.0048, 0.0072], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0027, 0.0026, 0.0035, 0.0030, 0.0028, 0.0034, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.5781e-05, 2.4871e-05, 2.3968e-05, 3.4547e-05, 2.8346e-05, 2.6104e-05, + 3.2739e-05, 3.2013e-05], device='cuda:3') +2022-11-16 04:59:15,517 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:59:18,478 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 +2022-11-16 04:59:22,650 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.271e+01 1.520e+02 1.906e+02 2.365e+02 4.910e+02, threshold=3.812e+02, percent-clipped=3.0 +2022-11-16 04:59:36,596 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 04:59:54,901 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1965, 1.5017, 1.8225, 1.5382, 1.8592, 1.7734, 1.7705, 1.6391], + device='cuda:3'), covar=tensor([0.0035, 0.0080, 0.0040, 0.0067, 0.0114, 0.0072, 0.0046, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0027, 0.0027, 0.0036, 0.0031, 0.0028, 0.0034, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.6104e-05, 2.5073e-05, 2.4328e-05, 3.4885e-05, 2.8560e-05, 2.6388e-05, + 3.3096e-05, 3.2364e-05], device='cuda:3') +2022-11-16 05:00:01,973 INFO [train.py:876] (3/4) Epoch 12, batch 600, loss[loss=0.1154, simple_loss=0.1542, pruned_loss=0.03825, over 5701.00 frames. ], tot_loss[loss=0.11, simple_loss=0.1394, pruned_loss=0.04029, over 1041100.79 frames. ], batch size: 19, lr: 6.85e-03, grad_scale: 16.0 +2022-11-16 05:00:08,836 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.56 vs. limit=5.0 +2022-11-16 05:00:14,741 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2281, 3.7018, 3.3115, 3.7315, 3.7166, 3.1094, 3.3123, 3.3487], + device='cuda:3'), covar=tensor([0.1113, 0.0577, 0.1524, 0.0479, 0.0517, 0.0659, 0.0913, 0.0633], + device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0180, 0.0278, 0.0175, 0.0221, 0.0174, 0.0190, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:00:30,595 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.813e+01 1.490e+02 1.792e+02 2.361e+02 3.754e+02, threshold=3.583e+02, percent-clipped=0.0 +2022-11-16 05:00:33,650 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:00:42,347 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3774, 3.5168, 3.6404, 1.9319, 3.5418, 3.9331, 3.7828, 4.0340], + device='cuda:3'), covar=tensor([0.1823, 0.1001, 0.0553, 0.2638, 0.0398, 0.0637, 0.0415, 0.0630], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0183, 0.0170, 0.0189, 0.0187, 0.0203, 0.0169, 0.0190], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:01:10,385 INFO [train.py:876] (3/4) Epoch 12, batch 700, loss[loss=0.09306, simple_loss=0.1307, pruned_loss=0.02771, over 5721.00 frames. ], tot_loss[loss=0.1114, simple_loss=0.1399, pruned_loss=0.04141, over 1055595.23 frames. ], batch size: 15, lr: 6.85e-03, grad_scale: 16.0 +2022-11-16 05:01:19,101 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9723, 1.5268, 1.4746, 1.4663, 1.3028, 1.3601, 1.2953, 1.4008], + device='cuda:3'), covar=tensor([0.3197, 0.1915, 0.1852, 0.1533, 0.1854, 0.2392, 0.2023, 0.0909], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0105, 0.0106, 0.0103, 0.0092, 0.0102, 0.0098, 0.0080], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 05:01:31,984 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 05:01:38,617 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.475e+02 1.756e+02 2.175e+02 4.412e+02, threshold=3.511e+02, percent-clipped=5.0 +2022-11-16 05:01:39,448 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:01:48,504 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:01:56,950 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1603, 2.6014, 3.6828, 3.1141, 4.0683, 2.7069, 3.5306, 4.1340], + device='cuda:3'), covar=tensor([0.0560, 0.1350, 0.1019, 0.1713, 0.0451, 0.1460, 0.1237, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0193, 0.0215, 0.0213, 0.0238, 0.0196, 0.0223, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:02:04,776 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:02:17,705 INFO [train.py:876] (3/4) Epoch 12, batch 800, loss[loss=0.1608, simple_loss=0.1756, pruned_loss=0.07296, over 5444.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1399, pruned_loss=0.04194, over 1066030.36 frames. ], batch size: 53, lr: 6.84e-03, grad_scale: 16.0 +2022-11-16 05:02:20,486 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:02:35,212 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:02:45,865 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:02:46,324 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.454e+01 1.512e+02 1.839e+02 2.382e+02 3.696e+02, threshold=3.678e+02, percent-clipped=2.0 +2022-11-16 05:02:56,121 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:02:56,805 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4455, 5.2168, 4.6743, 5.2363, 5.1528, 4.3447, 4.7728, 4.4283], + device='cuda:3'), covar=tensor([0.0343, 0.0405, 0.1427, 0.0279, 0.0386, 0.0535, 0.0630, 0.0740], + device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0180, 0.0277, 0.0174, 0.0220, 0.0173, 0.0190, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:03:20,690 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.68 vs. limit=5.0 +2022-11-16 05:03:25,608 INFO [train.py:876] (3/4) Epoch 12, batch 900, loss[loss=0.09046, simple_loss=0.1187, pruned_loss=0.03111, over 5714.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.1403, pruned_loss=0.04277, over 1070712.99 frames. ], batch size: 12, lr: 6.84e-03, grad_scale: 8.0 +2022-11-16 05:03:25,804 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7659, 2.5675, 2.8390, 3.7230, 3.6092, 2.8935, 2.3180, 3.6428], + device='cuda:3'), covar=tensor([0.0835, 0.2708, 0.2290, 0.2710, 0.1269, 0.2684, 0.2518, 0.0918], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0195, 0.0190, 0.0296, 0.0221, 0.0201, 0.0189, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2022-11-16 05:03:56,033 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.546e+02 1.867e+02 2.262e+02 5.374e+02, threshold=3.734e+02, percent-clipped=6.0 +2022-11-16 05:03:58,288 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:04:25,587 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3611, 1.5806, 1.2547, 1.2643, 1.6131, 1.7044, 1.4322, 1.6141], + device='cuda:3'), covar=tensor([0.0063, 0.0062, 0.0059, 0.0053, 0.0056, 0.0043, 0.0055, 0.0052], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0056, 0.0055, 0.0059, 0.0057, 0.0052, 0.0051, 0.0049], + device='cuda:3'), out_proj_covar=tensor([5.2843e-05, 4.9556e-05, 4.8349e-05, 5.2969e-05, 5.0548e-05, 4.5597e-05, + 4.5928e-05, 4.3246e-05], device='cuda:3') +2022-11-16 05:04:33,107 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80987.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:04:37,285 INFO [train.py:876] (3/4) Epoch 12, batch 1000, loss[loss=0.09693, simple_loss=0.1338, pruned_loss=0.03002, over 5546.00 frames. ], tot_loss[loss=0.1121, simple_loss=0.1405, pruned_loss=0.04184, over 1084539.63 frames. ], batch size: 16, lr: 6.83e-03, grad_scale: 8.0 +2022-11-16 05:05:06,152 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.560e+01 1.507e+02 1.766e+02 2.175e+02 5.874e+02, threshold=3.531e+02, percent-clipped=3.0 +2022-11-16 05:05:15,381 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81049.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:05:43,843 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81092.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:05:44,403 INFO [train.py:876] (3/4) Epoch 12, batch 1100, loss[loss=0.146, simple_loss=0.1466, pruned_loss=0.0727, over 4205.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1397, pruned_loss=0.04098, over 1087156.34 frames. ], batch size: 181, lr: 6.83e-03, grad_scale: 8.0 +2022-11-16 05:05:47,052 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81097.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:05:49,647 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2022-11-16 05:06:00,242 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81116.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:02,124 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81119.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:09,123 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81129.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:13,616 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.779e+01 1.469e+02 1.860e+02 2.387e+02 4.762e+02, threshold=3.720e+02, percent-clipped=3.0 +2022-11-16 05:06:23,034 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81150.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:34,436 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81167.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:41,055 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81177.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:06:52,016 INFO [train.py:876] (3/4) Epoch 12, batch 1200, loss[loss=0.09862, simple_loss=0.1348, pruned_loss=0.03123, over 5683.00 frames. ], tot_loss[loss=0.1104, simple_loss=0.1396, pruned_loss=0.04065, over 1092613.46 frames. ], batch size: 36, lr: 6.83e-03, grad_scale: 8.0 +2022-11-16 05:06:55,258 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81198.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:07:18,302 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9248, 3.1436, 2.3704, 2.8059, 2.0110, 2.4141, 1.6747, 2.5413], + device='cuda:3'), covar=tensor([0.1277, 0.0265, 0.0892, 0.0488, 0.1486, 0.0928, 0.1829, 0.0544], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0141, 0.0156, 0.0146, 0.0174, 0.0166, 0.0160, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:07:20,807 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.439e+02 1.852e+02 2.463e+02 6.772e+02, threshold=3.705e+02, percent-clipped=5.0 +2022-11-16 05:07:59,530 INFO [train.py:876] (3/4) Epoch 12, batch 1300, loss[loss=0.08746, simple_loss=0.1359, pruned_loss=0.01953, over 5488.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1379, pruned_loss=0.03917, over 1089244.92 frames. ], batch size: 17, lr: 6.82e-03, grad_scale: 8.0 +2022-11-16 05:08:03,919 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81299.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:08:05,226 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81301.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:08:09,018 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0645, 3.4688, 2.6193, 1.7025, 3.2311, 1.2289, 3.1953, 1.7132], + device='cuda:3'), covar=tensor([0.1538, 0.0185, 0.0813, 0.1840, 0.0297, 0.2321, 0.0305, 0.1654], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0103, 0.0114, 0.0111, 0.0100, 0.0120, 0.0099, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:08:15,137 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 +2022-11-16 05:08:28,629 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.965e+01 1.370e+02 1.672e+02 2.098e+02 3.683e+02, threshold=3.343e+02, percent-clipped=0.0 +2022-11-16 05:08:44,930 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81360.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:08:46,212 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81362.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:09:06,316 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81392.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:09:06,895 INFO [train.py:876] (3/4) Epoch 12, batch 1400, loss[loss=0.1162, simple_loss=0.1513, pruned_loss=0.04054, over 5622.00 frames. ], tot_loss[loss=0.1102, simple_loss=0.1394, pruned_loss=0.04052, over 1087872.69 frames. ], batch size: 32, lr: 6.82e-03, grad_scale: 8.0 +2022-11-16 05:09:31,107 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81429.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:09:35,795 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.508e+01 1.575e+02 1.879e+02 2.348e+02 6.067e+02, threshold=3.757e+02, percent-clipped=7.0 +2022-11-16 05:09:38,802 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81440.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:09:46,793 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3532, 1.8911, 1.4455, 1.3346, 0.8918, 1.5139, 1.2124, 1.5646], + device='cuda:3'), covar=tensor([0.0971, 0.0421, 0.0949, 0.1000, 0.2238, 0.0902, 0.1505, 0.0720], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0141, 0.0155, 0.0147, 0.0173, 0.0166, 0.0159, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:10:00,101 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81472.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:10:03,340 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81477.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:10:14,234 INFO [train.py:876] (3/4) Epoch 12, batch 1500, loss[loss=0.0972, simple_loss=0.1373, pruned_loss=0.02857, over 5734.00 frames. ], tot_loss[loss=0.1097, simple_loss=0.1392, pruned_loss=0.04012, over 1089324.70 frames. ], batch size: 13, lr: 6.81e-03, grad_scale: 8.0 +2022-11-16 05:10:21,372 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9318, 3.9767, 3.9669, 3.6824, 4.0572, 3.9002, 1.5895, 4.1215], + device='cuda:3'), covar=tensor([0.0305, 0.0314, 0.0282, 0.0397, 0.0248, 0.0442, 0.3072, 0.0299], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0086, 0.0087, 0.0080, 0.0099, 0.0089, 0.0130, 0.0105], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:10:35,151 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2022-11-16 05:10:42,646 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.569e+01 1.444e+02 1.773e+02 2.242e+02 5.218e+02, threshold=3.547e+02, percent-clipped=2.0 +2022-11-16 05:11:09,827 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2022-11-16 05:11:14,350 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.49 vs. limit=5.0 +2022-11-16 05:11:21,230 INFO [train.py:876] (3/4) Epoch 12, batch 1600, loss[loss=0.09085, simple_loss=0.1279, pruned_loss=0.0269, over 5757.00 frames. ], tot_loss[loss=0.1091, simple_loss=0.139, pruned_loss=0.03966, over 1089522.63 frames. ], batch size: 14, lr: 6.81e-03, grad_scale: 8.0 +2022-11-16 05:11:44,608 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9643, 3.1147, 2.1574, 2.7867, 2.0663, 2.4076, 1.6858, 2.5787], + device='cuda:3'), covar=tensor([0.1293, 0.0323, 0.1085, 0.0580, 0.1513, 0.1048, 0.1857, 0.0593], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0140, 0.0157, 0.0147, 0.0174, 0.0166, 0.0160, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:11:51,016 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.578e+01 1.477e+02 1.862e+02 2.272e+02 5.995e+02, threshold=3.723e+02, percent-clipped=4.0 +2022-11-16 05:11:54,101 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 05:11:57,309 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4062, 3.4536, 3.5964, 3.2621, 3.5115, 3.5092, 1.4159, 3.6694], + device='cuda:3'), covar=tensor([0.0327, 0.0352, 0.0351, 0.0394, 0.0371, 0.0399, 0.3306, 0.0312], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0087, 0.0088, 0.0081, 0.0100, 0.0090, 0.0132, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:12:04,154 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81655.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:12:05,448 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81657.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:12:14,072 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81670.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:12:29,715 INFO [train.py:876] (3/4) Epoch 12, batch 1700, loss[loss=0.1631, simple_loss=0.1793, pruned_loss=0.07343, over 5369.00 frames. ], tot_loss[loss=0.1084, simple_loss=0.1385, pruned_loss=0.03916, over 1093352.38 frames. ], batch size: 70, lr: 6.80e-03, grad_scale: 8.0 +2022-11-16 05:12:55,335 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81731.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 05:12:59,064 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.982e+01 1.420e+02 1.796e+02 2.143e+02 4.079e+02, threshold=3.592e+02, percent-clipped=2.0 +2022-11-16 05:13:14,491 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1149, 2.5436, 2.8429, 2.5852, 1.5932, 2.6700, 1.8247, 2.3490], + device='cuda:3'), covar=tensor([0.0350, 0.0197, 0.0155, 0.0253, 0.0481, 0.0198, 0.0459, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0176, 0.0180, 0.0200, 0.0191, 0.0179, 0.0189, 0.0183], + 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') +2022-11-16 05:13:23,671 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81772.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:13:37,840 INFO [train.py:876] (3/4) Epoch 12, batch 1800, loss[loss=0.09286, simple_loss=0.1318, pruned_loss=0.02696, over 5509.00 frames. ], tot_loss[loss=0.1094, simple_loss=0.1386, pruned_loss=0.04009, over 1081509.45 frames. ], batch size: 17, lr: 6.80e-03, grad_scale: 8.0 +2022-11-16 05:13:56,462 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81820.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:14:06,568 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.901e+01 1.448e+02 1.812e+02 2.296e+02 5.003e+02, threshold=3.624e+02, percent-clipped=1.0 +2022-11-16 05:14:45,099 INFO [train.py:876] (3/4) Epoch 12, batch 1900, loss[loss=0.09539, simple_loss=0.129, pruned_loss=0.0309, over 5590.00 frames. ], tot_loss[loss=0.1094, simple_loss=0.1388, pruned_loss=0.03994, over 1087434.57 frames. ], batch size: 24, lr: 6.80e-03, grad_scale: 8.0 +2022-11-16 05:15:04,520 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9775, 5.6540, 4.9979, 5.6778, 5.6457, 4.7807, 5.3799, 5.0411], + device='cuda:3'), covar=tensor([0.0230, 0.0364, 0.1428, 0.0336, 0.0333, 0.0497, 0.0315, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0178, 0.0274, 0.0173, 0.0221, 0.0172, 0.0190, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:15:04,611 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9214, 2.4846, 2.3282, 1.8614, 1.9234, 1.8974, 1.6307, 2.3893], + device='cuda:3'), covar=tensor([0.0045, 0.0039, 0.0036, 0.0043, 0.0037, 0.0033, 0.0037, 0.0044], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0055, 0.0055, 0.0059, 0.0057, 0.0053, 0.0051, 0.0049], + device='cuda:3'), out_proj_covar=tensor([5.3465e-05, 4.8952e-05, 4.7982e-05, 5.3036e-05, 5.0097e-05, 4.6150e-05, + 4.5814e-05, 4.3507e-05], device='cuda:3') +2022-11-16 05:15:14,249 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.303e+01 1.364e+02 1.744e+02 2.090e+02 4.733e+02, threshold=3.488e+02, percent-clipped=3.0 +2022-11-16 05:15:27,307 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81955.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:15:28,634 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81957.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:15:32,942 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2022-11-16 05:15:41,988 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3174, 3.9107, 4.1279, 3.8829, 4.3933, 4.1645, 3.9759, 4.4031], + device='cuda:3'), covar=tensor([0.0462, 0.0418, 0.0555, 0.0386, 0.0404, 0.0351, 0.0342, 0.0322], + device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0150, 0.0109, 0.0140, 0.0175, 0.0104, 0.0122, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 05:15:47,943 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1774, 3.7150, 2.7622, 1.6489, 3.4349, 1.4667, 3.3546, 1.7161], + device='cuda:3'), covar=tensor([0.1470, 0.0180, 0.0768, 0.1920, 0.0260, 0.1921, 0.0383, 0.1659], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0104, 0.0115, 0.0111, 0.0100, 0.0121, 0.0100, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:15:48,679 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81987.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:15:52,892 INFO [train.py:876] (3/4) Epoch 12, batch 2000, loss[loss=0.1184, simple_loss=0.1478, pruned_loss=0.04443, over 5567.00 frames. ], tot_loss[loss=0.1114, simple_loss=0.1401, pruned_loss=0.04133, over 1082753.54 frames. ], batch size: 50, lr: 6.79e-03, grad_scale: 8.0 +2022-11-16 05:15:59,646 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82003.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:16:01,283 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:16:15,346 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82026.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 05:16:16,169 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3789, 2.7858, 4.0872, 3.5515, 4.2700, 2.8790, 3.8619, 4.5238], + device='cuda:3'), covar=tensor([0.0570, 0.1498, 0.0734, 0.1322, 0.0505, 0.1667, 0.1224, 0.0733], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0193, 0.0214, 0.0211, 0.0237, 0.0196, 0.0223, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:16:19,181 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2022-11-16 05:16:22,027 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.158e+01 1.417e+02 1.705e+02 2.224e+02 2.971e+02, threshold=3.410e+02, percent-clipped=0.0 +2022-11-16 05:16:30,580 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82048.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:16:35,238 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82055.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:17:00,878 INFO [train.py:876] (3/4) Epoch 12, batch 2100, loss[loss=0.1087, simple_loss=0.1392, pruned_loss=0.0391, over 5621.00 frames. ], tot_loss[loss=0.1092, simple_loss=0.1388, pruned_loss=0.03979, over 1088946.54 frames. ], batch size: 23, lr: 6.79e-03, grad_scale: 8.0 +2022-11-16 05:17:08,558 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6991, 3.7328, 3.8231, 2.1962, 3.8550, 3.9440, 3.9816, 4.5355], + device='cuda:3'), covar=tensor([0.1715, 0.0982, 0.0964, 0.2320, 0.0450, 0.1794, 0.0397, 0.0418], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0179, 0.0164, 0.0184, 0.0182, 0.0199, 0.0166, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:17:16,453 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82116.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:17:26,643 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5083, 3.6868, 3.5439, 3.3989, 1.8882, 3.6394, 2.1241, 3.2076], + device='cuda:3'), covar=tensor([0.0475, 0.0332, 0.0228, 0.0367, 0.0701, 0.0231, 0.0596, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0176, 0.0180, 0.0200, 0.0192, 0.0179, 0.0188, 0.0184], + 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') +2022-11-16 05:17:29,604 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.760e+01 1.480e+02 1.837e+02 2.085e+02 4.685e+02, threshold=3.673e+02, percent-clipped=4.0 +2022-11-16 05:18:07,794 INFO [train.py:876] (3/4) Epoch 12, batch 2200, loss[loss=0.1144, simple_loss=0.1365, pruned_loss=0.04618, over 5679.00 frames. ], tot_loss[loss=0.109, simple_loss=0.1384, pruned_loss=0.03974, over 1089890.77 frames. ], batch size: 19, lr: 6.78e-03, grad_scale: 8.0 +2022-11-16 05:18:29,109 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1803, 1.5407, 1.2609, 1.2324, 1.5245, 1.5776, 1.3594, 1.7071], + device='cuda:3'), covar=tensor([0.0070, 0.0063, 0.0057, 0.0055, 0.0051, 0.0040, 0.0061, 0.0038], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0055, 0.0060, 0.0058, 0.0054, 0.0053, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.4303e-05, 4.9830e-05, 4.8728e-05, 5.4247e-05, 5.1264e-05, 4.7090e-05, + 4.6884e-05, 4.4282e-05], device='cuda:3') +2022-11-16 05:18:37,330 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.617e+01 1.469e+02 1.927e+02 2.568e+02 3.955e+02, threshold=3.853e+02, percent-clipped=1.0 +2022-11-16 05:18:48,111 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4222, 1.2734, 1.2063, 0.8477, 1.4524, 1.3676, 0.7541, 1.1436], + device='cuda:3'), covar=tensor([0.0470, 0.0385, 0.0378, 0.0795, 0.0292, 0.0405, 0.0916, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0023, 0.0016, 0.0020, 0.0016, 0.0015, 0.0022, 0.0015], + device='cuda:3'), out_proj_covar=tensor([8.1551e-05, 1.1335e-04, 8.5902e-05, 9.9667e-05, 8.7747e-05, 8.2265e-05, + 1.0703e-04, 8.1403e-05], device='cuda:3') +2022-11-16 05:19:09,729 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1995, 4.2446, 2.7297, 3.9392, 3.2330, 2.8790, 2.2707, 3.4090], + device='cuda:3'), covar=tensor([0.1480, 0.0191, 0.1022, 0.0444, 0.0784, 0.1033, 0.1780, 0.0410], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0141, 0.0159, 0.0149, 0.0178, 0.0168, 0.0161, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:19:15,470 INFO [train.py:876] (3/4) Epoch 12, batch 2300, loss[loss=0.06474, simple_loss=0.1022, pruned_loss=0.01363, over 5477.00 frames. ], tot_loss[loss=0.1092, simple_loss=0.1383, pruned_loss=0.03999, over 1079536.52 frames. ], batch size: 10, lr: 6.78e-03, grad_scale: 8.0 +2022-11-16 05:19:37,396 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82326.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:19:44,122 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.401e+02 1.761e+02 2.241e+02 4.168e+02, threshold=3.523e+02, percent-clipped=1.0 +2022-11-16 05:19:49,491 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82343.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:19:51,474 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82346.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:20:09,897 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82374.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:20:22,508 INFO [train.py:876] (3/4) Epoch 12, batch 2400, loss[loss=0.1402, simple_loss=0.1663, pruned_loss=0.05711, over 5340.00 frames. ], tot_loss[loss=0.1086, simple_loss=0.1379, pruned_loss=0.03968, over 1086860.05 frames. ], batch size: 79, lr: 6.78e-03, grad_scale: 8.0 +2022-11-16 05:20:29,474 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-16 05:20:32,559 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82407.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:20:35,162 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82411.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:20:51,657 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.786e+01 1.416e+02 1.815e+02 2.094e+02 3.842e+02, threshold=3.631e+02, percent-clipped=3.0 +2022-11-16 05:21:29,660 INFO [train.py:876] (3/4) Epoch 12, batch 2500, loss[loss=0.1077, simple_loss=0.1392, pruned_loss=0.03807, over 5631.00 frames. ], tot_loss[loss=0.1079, simple_loss=0.1375, pruned_loss=0.03909, over 1087895.90 frames. ], batch size: 18, lr: 6.77e-03, grad_scale: 8.0 +2022-11-16 05:21:29,829 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82493.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 05:21:31,121 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9086, 1.7249, 1.6221, 1.4641, 1.5254, 1.8064, 1.3991, 1.3903], + device='cuda:3'), covar=tensor([0.0040, 0.0055, 0.0042, 0.0061, 0.0071, 0.0070, 0.0052, 0.0056], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0027, 0.0026, 0.0035, 0.0031, 0.0028, 0.0035, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.6313e-05, 2.4720e-05, 2.3700e-05, 3.4488e-05, 2.8736e-05, 2.6645e-05, + 3.4170e-05, 3.2937e-05], device='cuda:3') +2022-11-16 05:21:58,346 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.955e+01 1.483e+02 1.924e+02 2.282e+02 6.804e+02, threshold=3.849e+02, percent-clipped=1.0 +2022-11-16 05:22:10,926 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82554.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 05:22:21,431 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82569.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:22:37,473 INFO [train.py:876] (3/4) Epoch 12, batch 2600, loss[loss=0.1239, simple_loss=0.1525, pruned_loss=0.04767, over 5679.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1366, pruned_loss=0.0384, over 1090439.21 frames. ], batch size: 36, lr: 6.77e-03, grad_scale: 8.0 +2022-11-16 05:22:45,398 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1281, 1.4931, 1.0010, 0.9298, 1.3643, 1.2572, 0.7856, 1.4218], + device='cuda:3'), covar=tensor([0.0055, 0.0043, 0.0060, 0.0057, 0.0055, 0.0047, 0.0082, 0.0047], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0057, 0.0055, 0.0060, 0.0058, 0.0053, 0.0052, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.4347e-05, 5.0394e-05, 4.8546e-05, 5.3625e-05, 5.1338e-05, 4.6714e-05, + 4.6533e-05, 4.4281e-05], device='cuda:3') +2022-11-16 05:22:52,988 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9145, 2.8870, 2.8129, 2.9997, 2.6784, 2.6720, 3.1364, 3.3269], + device='cuda:3'), covar=tensor([0.0995, 0.1520, 0.1518, 0.1637, 0.1251, 0.1160, 0.0919, 0.1970], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0107, 0.0106, 0.0106, 0.0094, 0.0103, 0.0100, 0.0081], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 05:22:53,632 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4671, 1.5340, 1.5245, 1.2954, 1.3804, 1.3469, 1.1427, 0.8367], + device='cuda:3'), covar=tensor([0.0036, 0.0037, 0.0032, 0.0047, 0.0050, 0.0054, 0.0053, 0.0083], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0027, 0.0027, 0.0036, 0.0031, 0.0028, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.6581e-05, 2.5232e-05, 2.4008e-05, 3.4611e-05, 2.8951e-05, 2.6786e-05, + 3.4496e-05, 3.3176e-05], device='cuda:3') +2022-11-16 05:23:03,082 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82630.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:23:06,796 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.089e+01 1.435e+02 1.814e+02 2.327e+02 4.653e+02, threshold=3.628e+02, percent-clipped=2.0 +2022-11-16 05:23:11,767 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82643.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:23:16,998 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5555, 1.5305, 1.5373, 1.4283, 1.5684, 1.6098, 1.5660, 1.4621], + device='cuda:3'), covar=tensor([0.0067, 0.0091, 0.0055, 0.0056, 0.0059, 0.0042, 0.0053, 0.0094], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0055, 0.0060, 0.0058, 0.0054, 0.0052, 0.0051], + device='cuda:3'), out_proj_covar=tensor([5.4608e-05, 5.0617e-05, 4.8660e-05, 5.3880e-05, 5.1473e-05, 4.6909e-05, + 4.6638e-05, 4.4495e-05], device='cuda:3') +2022-11-16 05:23:18,676 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2022-11-16 05:23:43,823 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82691.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:23:45,099 INFO [train.py:876] (3/4) Epoch 12, batch 2700, loss[loss=0.05508, simple_loss=0.08302, pruned_loss=0.01357, over 5157.00 frames. ], tot_loss[loss=0.1076, simple_loss=0.1372, pruned_loss=0.03901, over 1085951.16 frames. ], batch size: 8, lr: 6.76e-03, grad_scale: 8.0 +2022-11-16 05:23:47,226 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 05:23:49,258 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2022-11-16 05:23:51,439 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82702.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:23:57,156 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82711.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:24:14,436 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.412e+02 1.711e+02 2.254e+02 4.662e+02, threshold=3.423e+02, percent-clipped=2.0 +2022-11-16 05:24:29,687 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82759.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:24:49,426 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-16 05:24:52,719 INFO [train.py:876] (3/4) Epoch 12, batch 2800, loss[loss=0.1395, simple_loss=0.1503, pruned_loss=0.06435, over 5310.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1373, pruned_loss=0.03902, over 1086449.11 frames. ], batch size: 79, lr: 6.76e-03, grad_scale: 8.0 +2022-11-16 05:25:00,722 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2022-11-16 05:25:21,516 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.398e+02 1.709e+02 2.150e+02 5.035e+02, threshold=3.419e+02, percent-clipped=1.0 +2022-11-16 05:25:30,445 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82849.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 05:25:34,406 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6298, 1.7366, 2.2068, 1.6847, 1.5150, 2.5878, 2.1602, 1.7032], + device='cuda:3'), covar=tensor([0.1411, 0.1637, 0.1432, 0.2473, 0.2646, 0.0463, 0.1457, 0.2095], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0092, 0.0091, 0.0097, 0.0072, 0.0065, 0.0074, 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') +2022-11-16 05:25:59,791 INFO [train.py:876] (3/4) Epoch 12, batch 2900, loss[loss=0.07678, simple_loss=0.1134, pruned_loss=0.02007, over 5713.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1382, pruned_loss=0.03973, over 1083091.01 frames. ], batch size: 19, lr: 6.76e-03, grad_scale: 16.0 +2022-11-16 05:26:15,197 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3675, 3.9487, 4.1282, 3.9401, 4.4317, 4.2222, 4.0474, 4.4188], + device='cuda:3'), covar=tensor([0.0406, 0.0360, 0.0485, 0.0357, 0.0369, 0.0265, 0.0286, 0.0322], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0154, 0.0112, 0.0144, 0.0181, 0.0107, 0.0126, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 05:26:21,468 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82925.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:26:28,505 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.244e+01 1.420e+02 1.709e+02 2.033e+02 4.998e+02, threshold=3.418e+02, percent-clipped=2.0 +2022-11-16 05:26:30,418 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2022-11-16 05:26:47,408 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82963.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:26:52,665 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2022-11-16 05:27:07,644 INFO [train.py:876] (3/4) Epoch 12, batch 3000, loss[loss=0.1263, simple_loss=0.1518, pruned_loss=0.05038, over 5746.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.1378, pruned_loss=0.04004, over 1079037.15 frames. ], batch size: 20, lr: 6.75e-03, grad_scale: 16.0 +2022-11-16 05:27:07,644 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 05:27:15,856 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0860, 2.9636, 2.8051, 3.3220, 2.6156, 3.1816, 2.9227, 3.7367], + device='cuda:3'), covar=tensor([0.0714, 0.1888, 0.1658, 0.0790, 0.1161, 0.0620, 0.1248, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0107, 0.0106, 0.0106, 0.0093, 0.0103, 0.0099, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 05:27:17,454 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1180, 4.8610, 4.7673, 4.9873, 5.2038, 5.0336, 4.7593, 5.2246], + device='cuda:3'), covar=tensor([0.0300, 0.0210, 0.0443, 0.0285, 0.0334, 0.0161, 0.0179, 0.0211], + device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0154, 0.0113, 0.0145, 0.0181, 0.0108, 0.0127, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 05:27:25,141 INFO [train.py:908] (3/4) Epoch 12, validation: loss=0.1722, simple_loss=0.1854, pruned_loss=0.07947, over 1530663.00 frames. +2022-11-16 05:27:25,142 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 05:27:31,383 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83002.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:27:33,818 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5571, 4.3601, 4.5597, 4.7677, 4.3264, 4.0939, 5.1148, 4.4743], + device='cuda:3'), covar=tensor([0.0382, 0.0794, 0.0496, 0.1038, 0.0573, 0.0386, 0.0519, 0.0635], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0109, 0.0096, 0.0121, 0.0090, 0.0080, 0.0145, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:27:45,970 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83024.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:27:46,814 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2022-11-16 05:27:53,838 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.307e+01 1.447e+02 1.727e+02 2.150e+02 3.702e+02, threshold=3.454e+02, percent-clipped=1.0 +2022-11-16 05:28:03,082 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83050.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:28:22,569 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.5618, 1.1684, 0.7929, 0.8379, 0.9658, 0.7766, 0.5532, 1.0175], + device='cuda:3'), covar=tensor([0.0091, 0.0052, 0.0086, 0.0058, 0.0059, 0.0074, 0.0106, 0.0055], + device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0055, 0.0060, 0.0058, 0.0054, 0.0053, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.4783e-05, 5.0842e-05, 4.8628e-05, 5.3536e-05, 5.1748e-05, 4.7265e-05, + 4.6750e-05, 4.4334e-05], device='cuda:3') +2022-11-16 05:28:32,006 INFO [train.py:876] (3/4) Epoch 12, batch 3100, loss[loss=0.09752, simple_loss=0.1306, pruned_loss=0.03223, over 5758.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1384, pruned_loss=0.0401, over 1081614.07 frames. ], batch size: 13, lr: 6.75e-03, grad_scale: 16.0 +2022-11-16 05:28:43,702 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3097, 3.9902, 3.0685, 1.8629, 3.6021, 1.5441, 3.7351, 2.1391], + device='cuda:3'), covar=tensor([0.1669, 0.0234, 0.0967, 0.2222, 0.0290, 0.2354, 0.0288, 0.1766], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0106, 0.0116, 0.0113, 0.0102, 0.0123, 0.0101, 0.0112], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:28:54,555 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7091, 1.2665, 1.1492, 1.2165, 0.9554, 1.6459, 1.3513, 1.1622], + device='cuda:3'), covar=tensor([0.3908, 0.1293, 0.3663, 0.3002, 0.2830, 0.0579, 0.1980, 0.2932], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0093, 0.0092, 0.0098, 0.0073, 0.0065, 0.0075, 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') +2022-11-16 05:29:01,264 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.527e+02 1.920e+02 2.292e+02 4.866e+02, threshold=3.839e+02, percent-clipped=2.0 +2022-11-16 05:29:09,695 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83149.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 05:29:17,637 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 +2022-11-16 05:29:39,573 INFO [train.py:876] (3/4) Epoch 12, batch 3200, loss[loss=0.09497, simple_loss=0.1377, pruned_loss=0.02613, over 5560.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1379, pruned_loss=0.03915, over 1089056.09 frames. ], batch size: 15, lr: 6.74e-03, grad_scale: 16.0 +2022-11-16 05:29:41,737 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83196.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:29:42,244 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83197.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 05:30:01,769 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83225.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:30:09,030 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.526e+02 1.871e+02 2.239e+02 5.623e+02, threshold=3.742e+02, percent-clipped=1.0 +2022-11-16 05:30:15,928 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2022-11-16 05:30:23,060 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:30:33,811 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:30:41,816 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83284.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:30:47,949 INFO [train.py:876] (3/4) Epoch 12, batch 3300, loss[loss=0.1189, simple_loss=0.1428, pruned_loss=0.04748, over 5559.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1376, pruned_loss=0.0397, over 1080171.30 frames. ], batch size: 43, lr: 6.74e-03, grad_scale: 16.0 +2022-11-16 05:31:05,238 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83319.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:31:17,549 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.733e+01 1.415e+02 1.663e+02 2.154e+02 3.672e+02, threshold=3.327e+02, percent-clipped=0.0 +2022-11-16 05:31:24,260 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83345.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:31:31,751 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 +2022-11-16 05:31:46,239 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2931, 4.9408, 4.4519, 4.9409, 4.8984, 4.1775, 4.5960, 4.3541], + device='cuda:3'), covar=tensor([0.0353, 0.0424, 0.1280, 0.0405, 0.0409, 0.0473, 0.0427, 0.0494], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0179, 0.0274, 0.0174, 0.0222, 0.0172, 0.0189, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:31:56,030 INFO [train.py:876] (3/4) Epoch 12, batch 3400, loss[loss=0.09526, simple_loss=0.1312, pruned_loss=0.02966, over 5529.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1371, pruned_loss=0.03913, over 1083032.96 frames. ], batch size: 14, lr: 6.74e-03, grad_scale: 16.0 +2022-11-16 05:32:21,702 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5866, 1.6601, 1.6313, 1.3362, 1.4934, 1.6248, 1.1541, 0.9243], + device='cuda:3'), covar=tensor([0.0029, 0.0026, 0.0026, 0.0052, 0.0040, 0.0040, 0.0042, 0.0055], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0027, 0.0027, 0.0036, 0.0031, 0.0028, 0.0035, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.6523e-05, 2.4976e-05, 2.4515e-05, 3.5201e-05, 2.8879e-05, 2.6512e-05, + 3.4053e-05, 3.2925e-05], device='cuda:3') +2022-11-16 05:32:25,092 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.935e+01 1.460e+02 1.839e+02 2.206e+02 3.947e+02, threshold=3.678e+02, percent-clipped=4.0 +2022-11-16 05:32:38,134 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9536, 2.9752, 2.5322, 3.0176, 2.5022, 2.9259, 3.0582, 3.4704], + device='cuda:3'), covar=tensor([0.0881, 0.1123, 0.1860, 0.1082, 0.1438, 0.1032, 0.1083, 0.1706], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0106, 0.0105, 0.0104, 0.0092, 0.0101, 0.0098, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 05:32:44,267 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.8562, 5.3274, 5.6415, 5.1629, 5.8932, 5.6920, 4.8110, 5.8566], + device='cuda:3'), covar=tensor([0.0334, 0.0298, 0.0412, 0.0338, 0.0278, 0.0158, 0.0295, 0.0206], + device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0154, 0.0111, 0.0144, 0.0180, 0.0108, 0.0127, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 05:33:03,158 INFO [train.py:876] (3/4) Epoch 12, batch 3500, loss[loss=0.08586, simple_loss=0.1238, pruned_loss=0.02394, over 5605.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1375, pruned_loss=0.03938, over 1079938.24 frames. ], batch size: 24, lr: 6.73e-03, grad_scale: 16.0 +2022-11-16 05:33:32,640 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.791e+01 1.416e+02 1.765e+02 2.301e+02 4.941e+02, threshold=3.529e+02, percent-clipped=1.0 +2022-11-16 05:33:43,841 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83552.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:33:47,140 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0572, 4.2073, 4.0273, 4.0175, 4.1297, 4.0222, 1.6022, 4.2605], + device='cuda:3'), covar=tensor([0.0272, 0.0308, 0.0298, 0.0305, 0.0364, 0.0394, 0.3413, 0.0301], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0090, 0.0082, 0.0103, 0.0092, 0.0133, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:33:54,987 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 +2022-11-16 05:34:08,751 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1880, 1.4883, 1.2233, 1.2567, 1.3610, 1.2895, 1.0776, 1.4339], + device='cuda:3'), covar=tensor([0.0070, 0.0057, 0.0068, 0.0058, 0.0059, 0.0055, 0.0089, 0.0055], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0054, 0.0059, 0.0058, 0.0053, 0.0052, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.3895e-05, 4.9724e-05, 4.7419e-05, 5.2423e-05, 5.1039e-05, 4.6544e-05, + 4.6499e-05, 4.3927e-05], device='cuda:3') +2022-11-16 05:34:11,111 INFO [train.py:876] (3/4) Epoch 12, batch 3600, loss[loss=0.1251, simple_loss=0.1414, pruned_loss=0.05441, over 5697.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1376, pruned_loss=0.03966, over 1084941.72 frames. ], batch size: 28, lr: 6.73e-03, grad_scale: 16.0 +2022-11-16 05:34:27,252 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 +2022-11-16 05:34:27,760 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 +2022-11-16 05:34:29,882 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83619.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:34:40,744 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.534e+01 1.623e+02 1.893e+02 2.295e+02 4.939e+02, threshold=3.787e+02, percent-clipped=6.0 +2022-11-16 05:34:43,448 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83640.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:35:02,006 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83667.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:35:19,166 INFO [train.py:876] (3/4) Epoch 12, batch 3700, loss[loss=0.07509, simple_loss=0.1117, pruned_loss=0.01923, over 5517.00 frames. ], tot_loss[loss=0.1083, simple_loss=0.1374, pruned_loss=0.03957, over 1084992.16 frames. ], batch size: 10, lr: 6.72e-03, grad_scale: 16.0 +2022-11-16 05:35:43,321 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9476, 4.2739, 3.9750, 3.8749, 2.0906, 4.1594, 2.4120, 3.6154], + device='cuda:3'), covar=tensor([0.0399, 0.0146, 0.0175, 0.0338, 0.0631, 0.0175, 0.0540, 0.0183], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0178, 0.0182, 0.0202, 0.0192, 0.0180, 0.0190, 0.0185], + 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') +2022-11-16 05:35:48,568 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.368e+01 1.468e+02 1.783e+02 2.131e+02 3.533e+02, threshold=3.566e+02, percent-clipped=0.0 +2022-11-16 05:36:02,231 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7070, 2.4959, 2.6589, 3.7912, 3.7426, 2.7975, 2.6395, 3.8016], + device='cuda:3'), covar=tensor([0.0684, 0.2386, 0.1979, 0.2529, 0.1031, 0.2987, 0.1808, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0197, 0.0186, 0.0300, 0.0223, 0.0204, 0.0189, 0.0246], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 05:36:04,195 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83760.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:36:21,949 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.30 vs. limit=5.0 +2022-11-16 05:36:27,058 INFO [train.py:876] (3/4) Epoch 12, batch 3800, loss[loss=0.07349, simple_loss=0.09627, pruned_loss=0.02535, over 5375.00 frames. ], tot_loss[loss=0.1095, simple_loss=0.1378, pruned_loss=0.04056, over 1077603.14 frames. ], batch size: 9, lr: 6.72e-03, grad_scale: 16.0 +2022-11-16 05:36:46,519 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83821.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:36:47,174 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2316, 1.9927, 2.8324, 1.7679, 1.4356, 3.0707, 2.4958, 2.1593], + device='cuda:3'), covar=tensor([0.1334, 0.1549, 0.0688, 0.3046, 0.4445, 0.0430, 0.1154, 0.1603], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0095, 0.0094, 0.0100, 0.0075, 0.0067, 0.0078, 0.0088], + 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') +2022-11-16 05:36:47,798 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0079, 4.8169, 3.6442, 2.2274, 4.4665, 2.0054, 4.6174, 2.6704], + device='cuda:3'), covar=tensor([0.1169, 0.0121, 0.0594, 0.2010, 0.0179, 0.1691, 0.0128, 0.1432], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0104, 0.0115, 0.0111, 0.0100, 0.0121, 0.0099, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:36:56,434 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.699e+01 1.465e+02 1.775e+02 2.276e+02 4.868e+02, threshold=3.550e+02, percent-clipped=4.0 +2022-11-16 05:36:58,233 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5539, 2.1823, 2.9044, 1.6971, 1.3177, 3.3847, 2.5508, 2.3917], + device='cuda:3'), covar=tensor([0.0990, 0.1266, 0.0736, 0.2790, 0.3177, 0.0601, 0.1258, 0.1296], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0095, 0.0094, 0.0100, 0.0075, 0.0067, 0.0078, 0.0088], + 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') +2022-11-16 05:37:07,267 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83852.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:37:32,289 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0240, 4.1580, 4.0580, 3.7073, 4.0243, 3.7590, 1.7410, 4.2443], + device='cuda:3'), covar=tensor([0.0266, 0.0256, 0.0270, 0.0335, 0.0274, 0.0405, 0.2989, 0.0284], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0088, 0.0089, 0.0082, 0.0102, 0.0090, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:37:35,200 INFO [train.py:876] (3/4) Epoch 12, batch 3900, loss[loss=0.1104, simple_loss=0.1462, pruned_loss=0.03728, over 5761.00 frames. ], tot_loss[loss=0.1104, simple_loss=0.1391, pruned_loss=0.04078, over 1080969.49 frames. ], batch size: 16, lr: 6.72e-03, grad_scale: 16.0 +2022-11-16 05:37:39,879 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83900.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:38:04,276 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.940e+01 1.504e+02 1.885e+02 2.376e+02 4.069e+02, threshold=3.770e+02, percent-clipped=4.0 +2022-11-16 05:38:07,077 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83940.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:38:13,233 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83949.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:38:15,195 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1261, 3.3005, 2.4621, 1.7178, 3.1347, 1.3413, 3.1144, 1.8300], + device='cuda:3'), covar=tensor([0.1486, 0.0209, 0.1219, 0.1875, 0.0275, 0.2224, 0.0342, 0.1463], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0105, 0.0116, 0.0112, 0.0101, 0.0122, 0.0100, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:38:37,789 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83985.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:38:39,631 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83988.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:38:43,117 INFO [train.py:876] (3/4) Epoch 12, batch 4000, loss[loss=0.1047, simple_loss=0.1361, pruned_loss=0.03666, over 5666.00 frames. ], tot_loss[loss=0.111, simple_loss=0.1396, pruned_loss=0.0412, over 1079454.27 frames. ], batch size: 36, lr: 6.71e-03, grad_scale: 16.0 +2022-11-16 05:38:54,484 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84010.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 05:39:11,844 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.002e+01 1.513e+02 1.878e+02 2.345e+02 5.861e+02, threshold=3.757e+02, percent-clipped=4.0 +2022-11-16 05:39:18,967 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84046.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:39:32,303 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1740, 2.5611, 2.8144, 2.5212, 1.6509, 2.7693, 1.9686, 2.1123], + device='cuda:3'), covar=tensor([0.0319, 0.0178, 0.0146, 0.0260, 0.0447, 0.0172, 0.0402, 0.0212], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0178, 0.0182, 0.0204, 0.0193, 0.0181, 0.0191, 0.0185], + 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') +2022-11-16 05:39:34,997 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 05:39:50,160 INFO [train.py:876] (3/4) Epoch 12, batch 4100, loss[loss=0.1588, simple_loss=0.1773, pruned_loss=0.07011, over 5469.00 frames. ], tot_loss[loss=0.1101, simple_loss=0.1391, pruned_loss=0.04058, over 1085035.01 frames. ], batch size: 58, lr: 6.71e-03, grad_scale: 16.0 +2022-11-16 05:39:53,225 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84097.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:40:06,007 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84116.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:40:18,963 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.530e+02 1.828e+02 2.227e+02 3.985e+02, threshold=3.657e+02, percent-clipped=1.0 +2022-11-16 05:40:34,671 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84158.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:40:46,781 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9507, 2.0176, 2.6538, 1.7054, 1.3489, 2.8548, 2.2059, 2.0361], + device='cuda:3'), covar=tensor([0.1176, 0.1288, 0.0669, 0.2557, 0.3409, 0.1210, 0.1235, 0.1726], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0095, 0.0094, 0.0100, 0.0076, 0.0067, 0.0078, 0.0089], + 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') +2022-11-16 05:40:49,286 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8696, 4.7745, 4.5660, 4.6948, 4.8628, 4.4752, 1.9505, 5.0006], + device='cuda:3'), covar=tensor([0.0194, 0.0445, 0.0343, 0.0274, 0.0274, 0.0365, 0.2843, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0089, 0.0082, 0.0103, 0.0091, 0.0132, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:40:56,962 INFO [train.py:876] (3/4) Epoch 12, batch 4200, loss[loss=0.08612, simple_loss=0.1235, pruned_loss=0.02439, over 5564.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1368, pruned_loss=0.03891, over 1082054.44 frames. ], batch size: 14, lr: 6.70e-03, grad_scale: 16.0 +2022-11-16 05:41:01,797 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84199.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:41:26,498 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.244e+01 1.460e+02 1.805e+02 2.127e+02 5.710e+02, threshold=3.610e+02, percent-clipped=1.0 +2022-11-16 05:41:26,697 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84236.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:41:28,835 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 +2022-11-16 05:41:42,994 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84260.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:42:04,886 INFO [train.py:876] (3/4) Epoch 12, batch 4300, loss[loss=0.1221, simple_loss=0.1491, pruned_loss=0.04751, over 5002.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1367, pruned_loss=0.0388, over 1085705.67 frames. ], batch size: 109, lr: 6.70e-03, grad_scale: 16.0 +2022-11-16 05:42:07,610 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84297.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:42:12,737 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84305.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 05:42:22,959 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0988, 2.7599, 3.7024, 3.3878, 4.2204, 2.7631, 3.5804, 4.1594], + device='cuda:3'), covar=tensor([0.0656, 0.1368, 0.1050, 0.1461, 0.0432, 0.1505, 0.1296, 0.0812], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0195, 0.0217, 0.0213, 0.0240, 0.0198, 0.0224, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:42:34,272 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.802e+01 1.531e+02 1.904e+02 2.409e+02 4.137e+02, threshold=3.809e+02, percent-clipped=6.0 +2022-11-16 05:42:34,459 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84336.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:42:37,626 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84341.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:43:12,340 INFO [train.py:876] (3/4) Epoch 12, batch 4400, loss[loss=0.1654, simple_loss=0.1719, pruned_loss=0.07944, over 5376.00 frames. ], tot_loss[loss=0.107, simple_loss=0.1363, pruned_loss=0.03883, over 1088565.01 frames. ], batch size: 70, lr: 6.70e-03, grad_scale: 16.0 +2022-11-16 05:43:12,591 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2022-11-16 05:43:15,065 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84397.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:43:27,613 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84416.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:43:34,744 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9361, 3.7621, 3.9034, 3.8868, 3.5916, 3.5307, 4.2309, 3.8514], + device='cuda:3'), covar=tensor([0.0375, 0.0969, 0.0358, 0.1128, 0.0619, 0.0373, 0.0705, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0108, 0.0096, 0.0121, 0.0089, 0.0079, 0.0145, 0.0102], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:43:41,383 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.927e+01 1.500e+02 1.777e+02 2.208e+02 3.922e+02, threshold=3.553e+02, percent-clipped=1.0 +2022-11-16 05:43:52,912 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84453.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:43:56,979 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.31 vs. limit=5.0 +2022-11-16 05:43:59,962 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84464.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:44:19,824 INFO [train.py:876] (3/4) Epoch 12, batch 4500, loss[loss=0.09282, simple_loss=0.1276, pruned_loss=0.02904, over 5555.00 frames. ], tot_loss[loss=0.1071, simple_loss=0.1365, pruned_loss=0.0389, over 1090065.85 frames. ], batch size: 13, lr: 6.69e-03, grad_scale: 16.0 +2022-11-16 05:44:41,053 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84525.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:44:48,042 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.515e+02 1.740e+02 2.350e+02 4.214e+02, threshold=3.480e+02, percent-clipped=2.0 +2022-11-16 05:44:52,849 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1054, 4.4733, 4.2525, 3.7405, 2.4591, 4.4781, 2.6236, 3.8731], + device='cuda:3'), covar=tensor([0.0392, 0.0121, 0.0168, 0.0353, 0.0621, 0.0131, 0.0487, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0179, 0.0182, 0.0204, 0.0194, 0.0181, 0.0190, 0.0184], + 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') +2022-11-16 05:44:57,240 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2022-11-16 05:45:01,602 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84555.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:45:22,424 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84586.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:45:26,273 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84592.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:45:26,815 INFO [train.py:876] (3/4) Epoch 12, batch 4600, loss[loss=0.08613, simple_loss=0.115, pruned_loss=0.02866, over 5284.00 frames. ], tot_loss[loss=0.1078, simple_loss=0.1378, pruned_loss=0.03893, over 1091239.55 frames. ], batch size: 9, lr: 6.69e-03, grad_scale: 16.0 +2022-11-16 05:45:29,722 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 +2022-11-16 05:45:36,018 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84605.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:45:53,540 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-16 05:45:54,577 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84633.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:45:56,354 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.464e+02 1.836e+02 2.237e+02 3.755e+02, threshold=3.672e+02, percent-clipped=3.0 +2022-11-16 05:45:59,684 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84641.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:46:08,142 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84653.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:46:32,303 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84689.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:46:34,328 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84692.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:46:34,898 INFO [train.py:876] (3/4) Epoch 12, batch 4700, loss[loss=0.09063, simple_loss=0.114, pruned_loss=0.03363, over 3865.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1368, pruned_loss=0.03852, over 1084506.40 frames. ], batch size: 4, lr: 6.68e-03, grad_scale: 16.0 +2022-11-16 05:46:35,703 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84694.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:47:03,946 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.103e+01 1.427e+02 1.725e+02 2.096e+02 3.748e+02, threshold=3.451e+02, percent-clipped=1.0 +2022-11-16 05:47:13,236 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7675, 1.4307, 1.5906, 1.3141, 1.9190, 1.9592, 1.2356, 1.4277], + device='cuda:3'), covar=tensor([0.0299, 0.0411, 0.0397, 0.0398, 0.0645, 0.0927, 0.0688, 0.0944], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0024, 0.0017, 0.0021, 0.0017, 0.0015, 0.0023, 0.0016], + device='cuda:3'), out_proj_covar=tensor([8.4369e-05, 1.1624e-04, 8.9800e-05, 1.0459e-04, 9.1532e-05, 8.5432e-05, + 1.1286e-04, 8.6028e-05], device='cuda:3') +2022-11-16 05:47:15,128 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:47:24,459 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2022-11-16 05:47:42,122 INFO [train.py:876] (3/4) Epoch 12, batch 4800, loss[loss=0.1302, simple_loss=0.1615, pruned_loss=0.04942, over 5730.00 frames. ], tot_loss[loss=0.1074, simple_loss=0.1368, pruned_loss=0.03897, over 1079321.99 frames. ], batch size: 31, lr: 6.68e-03, grad_scale: 16.0 +2022-11-16 05:47:47,072 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84800.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:47:47,585 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84801.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:48:11,628 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.403e+01 1.631e+02 1.987e+02 2.475e+02 5.083e+02, threshold=3.974e+02, percent-clipped=5.0 +2022-11-16 05:48:23,426 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7167, 2.6587, 2.2450, 2.7643, 2.2047, 2.4859, 2.3145, 2.7998], + device='cuda:3'), covar=tensor([0.1198, 0.1400, 0.2141, 0.1358, 0.1745, 0.1122, 0.1594, 0.4345], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0105, 0.0105, 0.0103, 0.0092, 0.0101, 0.0098, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 05:48:24,060 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:48:24,763 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7852, 1.8090, 1.6379, 1.6037, 1.5886, 1.7058, 1.7361, 1.6187], + device='cuda:3'), covar=tensor([0.0071, 0.0070, 0.0060, 0.0052, 0.0056, 0.0051, 0.0043, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0055, 0.0060, 0.0058, 0.0053, 0.0053, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.4183e-05, 5.0136e-05, 4.8305e-05, 5.3323e-05, 5.1392e-05, 4.6169e-05, + 4.7151e-05, 4.4016e-05], device='cuda:3') +2022-11-16 05:48:27,989 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84861.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:48:41,878 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84881.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:48:42,273 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 +2022-11-16 05:48:49,148 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84892.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:48:49,704 INFO [train.py:876] (3/4) Epoch 12, batch 4900, loss[loss=0.08648, simple_loss=0.1296, pruned_loss=0.0217, over 5676.00 frames. ], tot_loss[loss=0.1083, simple_loss=0.1377, pruned_loss=0.03945, over 1082091.01 frames. ], batch size: 19, lr: 6.68e-03, grad_scale: 32.0 +2022-11-16 05:48:53,843 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.80 vs. limit=5.0 +2022-11-16 05:48:56,265 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84903.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:49:19,714 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.452e+01 1.404e+02 1.712e+02 2.121e+02 6.209e+02, threshold=3.423e+02, percent-clipped=1.0 +2022-11-16 05:49:21,810 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84940.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:49:28,971 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84951.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:49:54,405 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84989.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:49:56,368 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84992.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:49:56,852 INFO [train.py:876] (3/4) Epoch 12, batch 5000, loss[loss=0.08839, simple_loss=0.1294, pruned_loss=0.02366, over 5550.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1371, pruned_loss=0.03874, over 1082518.80 frames. ], batch size: 14, lr: 6.67e-03, grad_scale: 16.0 +2022-11-16 05:50:13,400 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85012.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:50:29,276 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.668e+01 1.461e+02 1.751e+02 2.205e+02 3.739e+02, threshold=3.502e+02, percent-clipped=4.0 +2022-11-16 05:50:31,270 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85040.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:50:47,543 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4585, 3.3421, 3.7238, 1.8647, 3.5279, 3.9943, 3.9473, 4.3949], + device='cuda:3'), covar=tensor([0.1879, 0.1428, 0.0627, 0.2773, 0.0390, 0.0584, 0.0358, 0.0508], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0181, 0.0167, 0.0183, 0.0181, 0.0198, 0.0166, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:51:06,696 INFO [train.py:876] (3/4) Epoch 12, batch 5100, loss[loss=0.1137, simple_loss=0.1513, pruned_loss=0.03804, over 5614.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1371, pruned_loss=0.03871, over 1085521.01 frames. ], batch size: 18, lr: 6.67e-03, grad_scale: 16.0 +2022-11-16 05:51:16,483 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85107.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:51:17,816 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7783, 1.5310, 1.8600, 1.6513, 1.6143, 1.6337, 1.5695, 1.5854], + device='cuda:3'), covar=tensor([0.0048, 0.0083, 0.0041, 0.0065, 0.0140, 0.0061, 0.0063, 0.0069], + device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0026, 0.0027, 0.0035, 0.0030, 0.0027, 0.0034, 0.0033], + device='cuda:3'), out_proj_covar=tensor([2.6392e-05, 2.4494e-05, 2.3896e-05, 3.3412e-05, 2.8353e-05, 2.6158e-05, + 3.3061e-05, 3.1983e-05], device='cuda:3') +2022-11-16 05:51:21,651 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85115.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:51:22,016 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 05:51:36,164 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.733e+01 1.524e+02 1.873e+02 2.260e+02 4.795e+02, threshold=3.745e+02, percent-clipped=3.0 +2022-11-16 05:51:49,420 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85156.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:51:52,545 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3546, 2.9397, 3.0174, 2.7792, 1.8308, 2.9505, 2.0510, 2.5895], + device='cuda:3'), covar=tensor([0.0267, 0.0144, 0.0137, 0.0204, 0.0399, 0.0155, 0.0365, 0.0151], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0178, 0.0181, 0.0204, 0.0193, 0.0181, 0.0190, 0.0183], + 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') +2022-11-16 05:51:57,591 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85168.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 05:52:03,257 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85176.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:52:06,507 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85181.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:52:14,134 INFO [train.py:876] (3/4) Epoch 12, batch 5200, loss[loss=0.07925, simple_loss=0.1198, pruned_loss=0.01935, over 5493.00 frames. ], tot_loss[loss=0.1099, simple_loss=0.139, pruned_loss=0.04039, over 1082513.75 frames. ], batch size: 10, lr: 6.66e-03, grad_scale: 16.0 +2022-11-16 05:52:39,251 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85229.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:52:45,062 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.605e+01 1.468e+02 1.779e+02 2.161e+02 4.129e+02, threshold=3.557e+02, percent-clipped=1.0 +2022-11-16 05:53:02,940 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6957, 5.0478, 3.7924, 2.0832, 4.6304, 2.0027, 4.7575, 2.6787], + device='cuda:3'), covar=tensor([0.1288, 0.0093, 0.0509, 0.2052, 0.0137, 0.1625, 0.0163, 0.1359], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0103, 0.0115, 0.0110, 0.0100, 0.0119, 0.0099, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:53:13,773 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.10 vs. limit=5.0 +2022-11-16 05:53:20,071 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85289.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:53:22,520 INFO [train.py:876] (3/4) Epoch 12, batch 5300, loss[loss=0.07807, simple_loss=0.1155, pruned_loss=0.02033, over 5523.00 frames. ], tot_loss[loss=0.1107, simple_loss=0.1391, pruned_loss=0.04111, over 1080731.43 frames. ], batch size: 10, lr: 6.66e-03, grad_scale: 8.0 +2022-11-16 05:53:30,325 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0962, 2.7142, 3.6232, 3.2562, 4.0056, 2.6468, 3.5364, 4.0404], + device='cuda:3'), covar=tensor([0.0538, 0.1420, 0.0751, 0.1243, 0.0430, 0.1386, 0.1106, 0.0657], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0194, 0.0214, 0.0210, 0.0240, 0.0197, 0.0224, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:53:31,528 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85307.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:53:44,881 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6830, 3.7985, 3.6230, 3.3528, 3.7693, 3.6018, 1.4507, 3.9023], + device='cuda:3'), covar=tensor([0.0254, 0.0364, 0.0353, 0.0381, 0.0362, 0.0397, 0.3150, 0.0293], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0088, 0.0088, 0.0081, 0.0102, 0.0089, 0.0130, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 05:53:50,783 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8935, 1.8881, 1.9063, 1.5820, 1.5778, 1.6680, 1.5476, 2.0407], + device='cuda:3'), covar=tensor([0.0045, 0.0055, 0.0044, 0.0051, 0.0049, 0.0040, 0.0043, 0.0036], + device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0055, 0.0054, 0.0059, 0.0057, 0.0053, 0.0053, 0.0050], + device='cuda:3'), out_proj_covar=tensor([5.2947e-05, 4.9124e-05, 4.7482e-05, 5.2388e-05, 5.0659e-05, 4.5852e-05, + 4.7010e-05, 4.3708e-05], device='cuda:3') +2022-11-16 05:53:52,739 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85337.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:53:53,357 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.591e+01 1.461e+02 1.746e+02 2.193e+02 3.892e+02, threshold=3.493e+02, percent-clipped=1.0 +2022-11-16 05:54:26,428 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7930, 2.4052, 2.7783, 3.6682, 3.8554, 2.7396, 2.2683, 3.7019], + device='cuda:3'), covar=tensor([0.1134, 0.2963, 0.2230, 0.2920, 0.0970, 0.3161, 0.2368, 0.0842], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0199, 0.0190, 0.0306, 0.0224, 0.0203, 0.0189, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 05:54:33,130 INFO [train.py:876] (3/4) Epoch 12, batch 5400, loss[loss=0.08179, simple_loss=0.1123, pruned_loss=0.02562, over 5485.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1397, pruned_loss=0.04093, over 1087526.78 frames. ], batch size: 11, lr: 6.66e-03, grad_scale: 8.0 +2022-11-16 05:54:45,889 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6470, 1.6214, 1.7915, 1.6530, 1.0053, 1.5437, 1.1722, 1.4295], + device='cuda:3'), covar=tensor([0.0133, 0.0081, 0.0078, 0.0094, 0.0256, 0.0114, 0.0199, 0.0130], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0181, 0.0182, 0.0206, 0.0194, 0.0183, 0.0191, 0.0184], + 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') +2022-11-16 05:54:57,347 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:04,100 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.454e+02 1.853e+02 2.296e+02 5.814e+02, threshold=3.706e+02, percent-clipped=5.0 +2022-11-16 05:55:10,869 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2022-11-16 05:55:15,946 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85456.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:20,368 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85463.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 05:55:24,984 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2661, 2.6359, 3.1360, 4.1350, 4.2518, 3.2988, 2.7618, 4.1849], + device='cuda:3'), covar=tensor([0.0566, 0.2817, 0.2102, 0.2212, 0.0900, 0.2688, 0.1993, 0.0472], + device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0200, 0.0191, 0.0307, 0.0225, 0.0205, 0.0191, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 05:55:25,889 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85471.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:34,536 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85483.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:38,418 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85489.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:41,211 INFO [train.py:876] (3/4) Epoch 12, batch 5500, loss[loss=0.06414, simple_loss=0.1017, pruned_loss=0.01331, over 5540.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1403, pruned_loss=0.04171, over 1079284.21 frames. ], batch size: 10, lr: 6.65e-03, grad_scale: 8.0 +2022-11-16 05:55:48,427 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85504.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:55:52,673 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2022-11-16 05:56:06,089 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85530.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:56:11,498 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.321e+01 1.546e+02 1.853e+02 2.385e+02 3.916e+02, threshold=3.707e+02, percent-clipped=1.0 +2022-11-16 05:56:16,101 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:56:34,735 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.0225, 4.5826, 4.8500, 4.6278, 5.0694, 4.9102, 4.4145, 5.0891], + device='cuda:3'), covar=tensor([0.0345, 0.0365, 0.0429, 0.0320, 0.0309, 0.0227, 0.0254, 0.0252], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0153, 0.0110, 0.0143, 0.0181, 0.0110, 0.0126, 0.0155], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 05:56:47,703 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85591.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:56:48,892 INFO [train.py:876] (3/4) Epoch 12, batch 5600, loss[loss=0.09802, simple_loss=0.1386, pruned_loss=0.02872, over 5815.00 frames. ], tot_loss[loss=0.1111, simple_loss=0.1399, pruned_loss=0.04112, over 1078668.09 frames. ], batch size: 18, lr: 6.65e-03, grad_scale: 8.0 +2022-11-16 05:56:58,728 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85607.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:57:10,792 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85625.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:57:10,826 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1495, 2.9846, 3.1775, 1.7880, 2.9832, 3.3679, 3.3729, 3.7851], + device='cuda:3'), covar=tensor([0.2013, 0.1542, 0.0926, 0.2712, 0.0548, 0.0791, 0.0507, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0182, 0.0167, 0.0184, 0.0183, 0.0200, 0.0166, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:57:20,010 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.221e+01 1.474e+02 1.888e+02 2.414e+02 5.206e+02, threshold=3.776e+02, percent-clipped=5.0 +2022-11-16 05:57:32,029 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85655.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:57:39,256 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1979, 2.7033, 3.0135, 3.9854, 4.0342, 3.3077, 2.6969, 3.9006], + device='cuda:3'), covar=tensor([0.0631, 0.2556, 0.2411, 0.2422, 0.0983, 0.2432, 0.2130, 0.1177], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0198, 0.0188, 0.0304, 0.0226, 0.0204, 0.0189, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 05:57:52,345 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85686.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 05:57:56,995 INFO [train.py:876] (3/4) Epoch 12, batch 5700, loss[loss=0.1258, simple_loss=0.1462, pruned_loss=0.05263, over 5322.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1384, pruned_loss=0.03964, over 1082887.46 frames. ], batch size: 9, lr: 6.64e-03, grad_scale: 8.0 +2022-11-16 05:58:26,981 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.513e+01 1.496e+02 1.877e+02 2.228e+02 5.709e+02, threshold=3.754e+02, percent-clipped=3.0 +2022-11-16 05:58:43,944 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85763.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:58:49,778 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85771.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:58:58,186 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85784.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:59:04,117 INFO [train.py:876] (3/4) Epoch 12, batch 5800, loss[loss=0.1338, simple_loss=0.1667, pruned_loss=0.05042, over 5560.00 frames. ], tot_loss[loss=0.1079, simple_loss=0.1375, pruned_loss=0.0391, over 1080987.88 frames. ], batch size: 46, lr: 6.64e-03, grad_scale: 8.0 +2022-11-16 05:59:04,246 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8929, 2.5077, 3.0073, 1.8887, 1.8517, 3.6192, 2.9580, 2.4618], + device='cuda:3'), covar=tensor([0.1171, 0.1256, 0.0647, 0.3076, 0.2251, 0.0563, 0.1184, 0.1146], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0095, 0.0094, 0.0100, 0.0074, 0.0067, 0.0078, 0.0089], + 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') +2022-11-16 05:59:11,985 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5267, 4.0722, 3.1867, 2.0367, 3.8257, 1.5991, 3.8701, 2.2399], + device='cuda:3'), covar=tensor([0.1390, 0.0195, 0.0792, 0.1861, 0.0230, 0.1859, 0.0294, 0.1467], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0104, 0.0115, 0.0112, 0.0102, 0.0120, 0.0101, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 05:59:16,826 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85811.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:59:22,788 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85819.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:59:35,105 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.803e+01 1.482e+02 1.820e+02 2.147e+02 4.590e+02, threshold=3.641e+02, percent-clipped=4.0 +2022-11-16 05:59:35,897 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85839.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 05:59:56,204 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.1255, 0.8019, 0.9650, 0.7859, 1.0595, 0.9407, 0.5136, 0.7677], + device='cuda:3'), covar=tensor([0.0281, 0.0351, 0.0415, 0.0472, 0.0434, 0.0317, 0.0818, 0.0332], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0024, 0.0017, 0.0020, 0.0017, 0.0016, 0.0023, 0.0016], + device='cuda:3'), out_proj_covar=tensor([8.5451e-05, 1.1778e-04, 9.1951e-05, 1.0450e-04, 9.2725e-05, 8.6402e-05, + 1.1407e-04, 8.7801e-05], device='cuda:3') +2022-11-16 06:00:00,457 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85875.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:00:07,909 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85886.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:00:12,394 INFO [train.py:876] (3/4) Epoch 12, batch 5900, loss[loss=0.1059, simple_loss=0.1393, pruned_loss=0.03624, over 5778.00 frames. ], tot_loss[loss=0.1065, simple_loss=0.1362, pruned_loss=0.03838, over 1084554.33 frames. ], batch size: 21, lr: 6.64e-03, grad_scale: 8.0 +2022-11-16 06:00:19,754 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2022-11-16 06:00:38,046 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85930.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:00:41,980 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85936.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:00:43,059 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.779e+01 1.458e+02 1.851e+02 2.281e+02 4.967e+02, threshold=3.703e+02, percent-clipped=4.0 +2022-11-16 06:00:55,692 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85957.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:01:12,072 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85981.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:01:16,394 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 +2022-11-16 06:01:18,822 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85991.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:01:19,973 INFO [train.py:876] (3/4) Epoch 12, batch 6000, loss[loss=0.09342, simple_loss=0.1322, pruned_loss=0.02733, over 5775.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1367, pruned_loss=0.03829, over 1088814.49 frames. ], batch size: 16, lr: 6.63e-03, grad_scale: 8.0 +2022-11-16 06:01:19,973 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 06:01:33,153 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2672, 2.0858, 2.8170, 1.8477, 1.4918, 3.1981, 2.6672, 2.1363], + device='cuda:3'), covar=tensor([0.1105, 0.1759, 0.0658, 0.2714, 0.3372, 0.0382, 0.0870, 0.1904], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0098, 0.0097, 0.0103, 0.0076, 0.0069, 0.0080, 0.0092], + 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') +2022-11-16 06:01:34,015 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8447, 3.0805, 2.9914, 3.1516, 3.1361, 2.9907, 2.5388, 2.7649], + device='cuda:3'), covar=tensor([0.0382, 0.0467, 0.1069, 0.0385, 0.0583, 0.0643, 0.0908, 0.0644], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0179, 0.0275, 0.0175, 0.0224, 0.0175, 0.0191, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:01:37,497 INFO [train.py:908] (3/4) Epoch 12, validation: loss=0.1738, simple_loss=0.1864, pruned_loss=0.08063, over 1530663.00 frames. +2022-11-16 06:01:37,497 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 06:01:43,089 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7322, 4.4401, 3.4515, 2.1696, 4.1949, 1.7943, 4.1924, 2.4611], + device='cuda:3'), covar=tensor([0.1298, 0.0141, 0.0679, 0.1749, 0.0185, 0.1811, 0.0158, 0.1558], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0105, 0.0115, 0.0112, 0.0102, 0.0120, 0.0101, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:01:43,765 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86002.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:01:54,415 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:02:08,196 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.233e+01 1.359e+02 1.723e+02 2.216e+02 5.600e+02, threshold=3.445e+02, percent-clipped=2.0 +2022-11-16 06:02:17,508 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86052.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:02:24,816 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86063.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:02:25,425 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7885, 1.2561, 1.0887, 1.2182, 1.1127, 1.2323, 0.9654, 1.2411], + device='cuda:3'), covar=tensor([0.2475, 0.1341, 0.1410, 0.0981, 0.1360, 0.1566, 0.1424, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0107, 0.0105, 0.0104, 0.0092, 0.0102, 0.0098, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:02:29,104 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86069.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:02:39,116 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86084.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:02:45,094 INFO [train.py:876] (3/4) Epoch 12, batch 6100, loss[loss=0.1309, simple_loss=0.1621, pruned_loss=0.04986, over 5546.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1371, pruned_loss=0.0384, over 1084064.71 frames. ], batch size: 46, lr: 6.63e-03, grad_scale: 8.0 +2022-11-16 06:02:50,604 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 06:02:58,474 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86113.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:10,251 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86130.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:11,383 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86132.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:15,121 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.336e+01 1.468e+02 1.787e+02 2.256e+02 5.479e+02, threshold=3.574e+02, percent-clipped=5.0 +2022-11-16 06:03:15,257 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6118, 5.1730, 4.8178, 5.1801, 5.2166, 4.2736, 4.7652, 4.5029], + device='cuda:3'), covar=tensor([0.0262, 0.0389, 0.1090, 0.0353, 0.0335, 0.0515, 0.0528, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0180, 0.0276, 0.0176, 0.0225, 0.0175, 0.0191, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:03:15,901 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86139.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:47,221 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:47,784 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86187.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:03:51,955 INFO [train.py:876] (3/4) Epoch 12, batch 6200, loss[loss=0.08562, simple_loss=0.1148, pruned_loss=0.02821, over 5483.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1371, pruned_loss=0.03872, over 1085590.01 frames. ], batch size: 17, lr: 6.63e-03, grad_scale: 8.0 +2022-11-16 06:04:11,410 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5249, 4.3148, 4.4268, 4.4933, 4.4030, 4.1275, 4.8579, 4.4656], + device='cuda:3'), covar=tensor([0.0376, 0.0941, 0.0396, 0.0892, 0.0416, 0.0317, 0.0638, 0.0542], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0109, 0.0097, 0.0122, 0.0090, 0.0081, 0.0146, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:04:17,347 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86231.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:04:19,259 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86234.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:04:22,081 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.231e+01 1.405e+02 1.749e+02 2.219e+02 4.004e+02, threshold=3.499e+02, percent-clipped=3.0 +2022-11-16 06:04:27,864 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86246.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:04:51,626 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86281.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:04:54,821 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86286.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 06:04:59,675 INFO [train.py:876] (3/4) Epoch 12, batch 6300, loss[loss=0.0633, simple_loss=0.09886, pruned_loss=0.01387, over 5489.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1372, pruned_loss=0.03817, over 1091535.67 frames. ], batch size: 10, lr: 6.62e-03, grad_scale: 8.0 +2022-11-16 06:05:01,721 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2022-11-16 06:05:09,333 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86307.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:05:10,416 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7037, 3.9586, 3.8497, 3.4830, 1.9375, 3.9699, 2.2646, 3.2342], + device='cuda:3'), covar=tensor([0.0469, 0.0209, 0.0190, 0.0469, 0.0769, 0.0161, 0.0669, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0180, 0.0181, 0.0204, 0.0193, 0.0181, 0.0189, 0.0184], + 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') +2022-11-16 06:05:13,606 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86313.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:05:22,825 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2967, 3.9746, 3.0960, 1.9116, 3.8546, 1.5688, 3.6101, 2.0596], + device='cuda:3'), covar=tensor([0.1387, 0.0147, 0.0733, 0.1779, 0.0153, 0.1767, 0.0290, 0.1404], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0115, 0.0111, 0.0101, 0.0119, 0.0101, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:05:24,117 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86329.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:05:29,914 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.920e+01 1.421e+02 1.647e+02 2.112e+02 5.317e+02, threshold=3.295e+02, percent-clipped=6.0 +2022-11-16 06:05:34,160 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5528, 3.5150, 3.4418, 3.1843, 1.8827, 3.5139, 2.2870, 3.0248], + device='cuda:3'), covar=tensor([0.0428, 0.0223, 0.0189, 0.0417, 0.0625, 0.0191, 0.0530, 0.0266], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0179, 0.0180, 0.0204, 0.0192, 0.0180, 0.0188, 0.0183], + 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') +2022-11-16 06:05:35,438 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0617, 3.8460, 3.8152, 3.5260, 4.0701, 3.8695, 1.4998, 4.1294], + device='cuda:3'), covar=tensor([0.0264, 0.0369, 0.0382, 0.0500, 0.0332, 0.0498, 0.3258, 0.0378], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0090, 0.0090, 0.0083, 0.0104, 0.0091, 0.0133, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:05:44,655 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:06:00,956 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9331, 1.6904, 2.2889, 2.0916, 2.1815, 1.9418, 1.7620, 2.1128], + device='cuda:3'), covar=tensor([0.1315, 0.0540, 0.0333, 0.0256, 0.0958, 0.0660, 0.0531, 0.0239], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0024, 0.0017, 0.0021, 0.0018, 0.0016, 0.0023, 0.0017], + device='cuda:3'), out_proj_covar=tensor([8.7241e-05, 1.2035e-04, 9.2797e-05, 1.0689e-04, 9.4950e-05, 8.7784e-05, + 1.1581e-04, 9.0221e-05], device='cuda:3') +2022-11-16 06:06:07,708 INFO [train.py:876] (3/4) Epoch 12, batch 6400, loss[loss=0.09273, simple_loss=0.1257, pruned_loss=0.0299, over 5730.00 frames. ], tot_loss[loss=0.1076, simple_loss=0.1371, pruned_loss=0.03904, over 1080328.40 frames. ], batch size: 14, lr: 6.62e-03, grad_scale: 8.0 +2022-11-16 06:06:09,562 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4910, 1.2852, 1.7465, 1.1540, 1.8682, 1.8761, 0.9997, 1.5470], + device='cuda:3'), covar=tensor([0.0774, 0.0547, 0.0532, 0.0755, 0.0429, 0.0611, 0.0684, 0.0635], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0024, 0.0017, 0.0021, 0.0018, 0.0016, 0.0023, 0.0017], + device='cuda:3'), out_proj_covar=tensor([8.7591e-05, 1.2078e-04, 9.3034e-05, 1.0723e-04, 9.5271e-05, 8.8069e-05, + 1.1613e-04, 9.0542e-05], device='cuda:3') +2022-11-16 06:06:18,205 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86408.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:06:30,012 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86425.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:06:38,420 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.070e+01 1.434e+02 1.773e+02 2.236e+02 3.206e+02, threshold=3.547e+02, percent-clipped=0.0 +2022-11-16 06:06:39,882 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86440.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:06:43,508 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:07:15,664 INFO [train.py:876] (3/4) Epoch 12, batch 6500, loss[loss=0.1262, simple_loss=0.1541, pruned_loss=0.04912, over 5425.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1376, pruned_loss=0.03926, over 1085262.33 frames. ], batch size: 58, lr: 6.61e-03, grad_scale: 8.0 +2022-11-16 06:07:21,494 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86501.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:07:25,134 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86506.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:07:39,814 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 06:07:42,123 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86531.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:07:46,543 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.116e+01 1.482e+02 1.807e+02 2.369e+02 3.734e+02, threshold=3.614e+02, percent-clipped=1.0 +2022-11-16 06:08:01,499 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4364, 4.4322, 4.3078, 3.6536, 2.3634, 4.5841, 2.6504, 3.8275], + device='cuda:3'), covar=tensor([0.0321, 0.0158, 0.0135, 0.0418, 0.0646, 0.0139, 0.0527, 0.0139], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0180, 0.0181, 0.0205, 0.0193, 0.0181, 0.0189, 0.0184], + 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') +2022-11-16 06:08:05,892 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2022-11-16 06:08:13,936 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86579.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:08:14,716 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86580.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:08:18,597 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86586.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 06:08:23,752 INFO [train.py:876] (3/4) Epoch 12, batch 6600, loss[loss=0.1011, simple_loss=0.1296, pruned_loss=0.0363, over 5579.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1358, pruned_loss=0.03764, over 1082671.20 frames. ], batch size: 22, lr: 6.61e-03, grad_scale: 8.0 +2022-11-16 06:08:26,475 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6185, 2.0067, 1.6500, 1.3896, 1.8037, 2.0395, 2.0400, 2.2334], + device='cuda:3'), covar=tensor([0.1973, 0.1354, 0.2100, 0.2579, 0.1365, 0.1231, 0.0933, 0.1250], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0184, 0.0170, 0.0186, 0.0185, 0.0202, 0.0169, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:08:30,007 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86602.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:08:37,141 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86613.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:08:51,384 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86634.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:08:54,694 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.903e+01 1.415e+02 1.832e+02 2.260e+02 3.608e+02, threshold=3.664e+02, percent-clipped=0.0 +2022-11-16 06:08:56,755 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9682, 4.4444, 4.7074, 4.4163, 5.0445, 4.8026, 4.4032, 5.0014], + device='cuda:3'), covar=tensor([0.0322, 0.0344, 0.0466, 0.0338, 0.0309, 0.0186, 0.0260, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0112, 0.0147, 0.0183, 0.0109, 0.0129, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 06:08:56,858 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86641.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:02,394 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2022-11-16 06:09:03,709 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2022-11-16 06:09:08,025 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86658.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:09,957 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86661.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:32,602 INFO [train.py:876] (3/4) Epoch 12, batch 6700, loss[loss=0.1113, simple_loss=0.139, pruned_loss=0.04181, over 5606.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.1354, pruned_loss=0.0377, over 1080943.46 frames. ], batch size: 38, lr: 6.61e-03, grad_scale: 8.0 +2022-11-16 06:09:39,591 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2022-11-16 06:09:41,155 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86706.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:42,538 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86708.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:53,711 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:09:59,099 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9308, 4.5839, 4.8056, 4.8943, 4.5251, 4.3489, 5.2544, 4.8517], + device='cuda:3'), covar=tensor([0.0326, 0.0962, 0.0486, 0.1083, 0.0391, 0.0316, 0.0624, 0.0554], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0109, 0.0097, 0.0122, 0.0091, 0.0081, 0.0147, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:10:02,549 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.559e+02 1.954e+02 2.479e+02 4.501e+02, threshold=3.908e+02, percent-clipped=4.0 +2022-11-16 06:10:13,508 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2022-11-16 06:10:15,006 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86756.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:10:25,967 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86773.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:10:29,668 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0890, 2.4097, 3.6823, 3.0822, 3.9414, 2.3919, 3.5094, 4.0290], + device='cuda:3'), covar=tensor([0.0578, 0.1687, 0.0892, 0.1562, 0.0631, 0.1801, 0.1080, 0.0802], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0194, 0.0216, 0.0214, 0.0243, 0.0198, 0.0226, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:10:39,694 INFO [train.py:876] (3/4) Epoch 12, batch 6800, loss[loss=0.1475, simple_loss=0.162, pruned_loss=0.06649, over 5033.00 frames. ], tot_loss[loss=0.107, simple_loss=0.1369, pruned_loss=0.03851, over 1082749.70 frames. ], batch size: 110, lr: 6.60e-03, grad_scale: 8.0 +2022-11-16 06:10:41,678 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86796.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:10:45,593 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86801.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:10:54,175 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86814.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:10:55,612 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 06:11:03,710 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7080, 0.9034, 0.9054, 0.7743, 0.8475, 0.9056, 0.7976, 0.6316], + device='cuda:3'), covar=tensor([0.0037, 0.0028, 0.0027, 0.0035, 0.0035, 0.0033, 0.0045, 0.0066], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0028, 0.0028, 0.0036, 0.0031, 0.0029, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.7722e-05, 2.5899e-05, 2.5118e-05, 3.4996e-05, 2.9212e-05, 2.7670e-05, + 3.4583e-05, 3.3056e-05], device='cuda:3') +2022-11-16 06:11:10,427 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 1.446e+02 1.789e+02 2.436e+02 4.053e+02, threshold=3.578e+02, percent-clipped=1.0 +2022-11-16 06:11:31,497 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0737, 4.5661, 4.1815, 4.6078, 4.5935, 3.8823, 4.1951, 4.0251], + device='cuda:3'), covar=tensor([0.0429, 0.0413, 0.1117, 0.0318, 0.0339, 0.0514, 0.0517, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0178, 0.0276, 0.0176, 0.0225, 0.0176, 0.0191, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:11:35,471 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86875.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:11:37,382 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86878.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:11:47,374 INFO [train.py:876] (3/4) Epoch 12, batch 6900, loss[loss=0.1379, simple_loss=0.1604, pruned_loss=0.05775, over 5283.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1367, pruned_loss=0.03834, over 1087179.36 frames. ], batch size: 79, lr: 6.60e-03, grad_scale: 8.0 +2022-11-16 06:11:53,840 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86902.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:12:17,110 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86936.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:12:18,367 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.928e+01 1.457e+02 1.817e+02 2.231e+02 4.523e+02, threshold=3.633e+02, percent-clipped=5.0 +2022-11-16 06:12:19,216 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86939.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:12:24,716 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5604, 3.4429, 3.5014, 3.0984, 2.0368, 3.5143, 2.1997, 3.1064], + device='cuda:3'), covar=tensor([0.0431, 0.0288, 0.0176, 0.0413, 0.0569, 0.0201, 0.0570, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0180, 0.0182, 0.0204, 0.0192, 0.0181, 0.0189, 0.0185], + 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') +2022-11-16 06:12:26,805 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86950.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:12:35,134 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 +2022-11-16 06:12:48,291 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.76 vs. limit=5.0 +2022-11-16 06:12:55,753 INFO [train.py:876] (3/4) Epoch 12, batch 7000, loss[loss=0.1082, simple_loss=0.1401, pruned_loss=0.03814, over 5781.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1358, pruned_loss=0.03852, over 1075993.26 frames. ], batch size: 21, lr: 6.60e-03, grad_scale: 8.0 +2022-11-16 06:13:02,852 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87002.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:13:26,384 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.516e+01 1.515e+02 1.846e+02 2.332e+02 4.129e+02, threshold=3.691e+02, percent-clipped=3.0 +2022-11-16 06:13:27,183 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8498, 4.0926, 4.0262, 3.7416, 3.9998, 3.7914, 1.4276, 4.0748], + device='cuda:3'), covar=tensor([0.0281, 0.0246, 0.0284, 0.0278, 0.0250, 0.0361, 0.3186, 0.0242], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0088, 0.0088, 0.0082, 0.0103, 0.0090, 0.0131, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:13:41,438 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1367, 3.9837, 2.6001, 3.7652, 3.1028, 2.7362, 2.0622, 3.3891], + device='cuda:3'), covar=tensor([0.1662, 0.0258, 0.1115, 0.0340, 0.0771, 0.1084, 0.1938, 0.0394], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0144, 0.0160, 0.0151, 0.0174, 0.0169, 0.0160, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:13:43,358 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87063.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:13:45,254 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.8431, 1.2293, 0.8643, 1.0592, 1.2023, 1.0979, 0.5375, 1.2423], + device='cuda:3'), covar=tensor([0.0088, 0.0048, 0.0073, 0.0054, 0.0071, 0.0064, 0.0117, 0.0061], + device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0057, 0.0057, 0.0060, 0.0059, 0.0055, 0.0054, 0.0052], + device='cuda:3'), out_proj_covar=tensor([5.5269e-05, 5.0956e-05, 4.9736e-05, 5.3763e-05, 5.1934e-05, 4.7634e-05, + 4.8241e-05, 4.5642e-05], device='cuda:3') +2022-11-16 06:14:03,309 INFO [train.py:876] (3/4) Epoch 12, batch 7100, loss[loss=0.1142, simple_loss=0.1461, pruned_loss=0.04118, over 5630.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1355, pruned_loss=0.03788, over 1084423.86 frames. ], batch size: 29, lr: 6.59e-03, grad_scale: 8.0 +2022-11-16 06:14:05,398 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87096.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:14:08,681 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87101.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 06:14:33,938 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.177e+01 1.551e+02 1.888e+02 2.451e+02 4.689e+02, threshold=3.775e+02, percent-clipped=4.0 +2022-11-16 06:14:37,911 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87144.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:14:41,186 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87149.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:14:55,687 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87170.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:15:06,549 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7316, 2.8373, 2.4635, 2.8179, 2.3629, 2.4108, 2.6519, 3.1019], + device='cuda:3'), covar=tensor([0.1212, 0.1109, 0.1916, 0.2886, 0.1500, 0.3527, 0.1312, 0.0882], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0105, 0.0103, 0.0104, 0.0092, 0.0102, 0.0097, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:15:08,766 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2784, 2.1919, 2.0992, 2.2854, 2.0514, 1.7354, 2.1275, 2.4302], + device='cuda:3'), covar=tensor([0.1504, 0.1682, 0.2115, 0.1459, 0.1452, 0.2385, 0.1540, 0.1189], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0105, 0.0103, 0.0104, 0.0092, 0.0102, 0.0097, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:15:11,218 INFO [train.py:876] (3/4) Epoch 12, batch 7200, loss[loss=0.07699, simple_loss=0.1102, pruned_loss=0.02192, over 5568.00 frames. ], tot_loss[loss=0.1059, simple_loss=0.1357, pruned_loss=0.03808, over 1088143.22 frames. ], batch size: 15, lr: 6.59e-03, grad_scale: 8.0 +2022-11-16 06:15:36,453 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87230.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:15:39,017 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87234.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:15:40,327 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87236.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:15:41,450 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.393e+01 1.550e+02 1.929e+02 2.381e+02 4.425e+02, threshold=3.859e+02, percent-clipped=3.0 +2022-11-16 06:16:42,439 INFO [train.py:876] (3/4) Epoch 13, batch 0, loss[loss=0.1038, simple_loss=0.1427, pruned_loss=0.03243, over 5594.00 frames. ], tot_loss[loss=0.1038, simple_loss=0.1427, pruned_loss=0.03243, over 5594.00 frames. ], batch size: 16, lr: 6.33e-03, grad_scale: 16.0 +2022-11-16 06:16:42,439 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 06:16:58,482 INFO [train.py:908] (3/4) Epoch 13, validation: loss=0.175, simple_loss=0.1891, pruned_loss=0.08049, over 1530663.00 frames. +2022-11-16 06:16:58,483 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 06:17:11,205 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87284.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:17:16,600 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:17:31,315 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8484, 4.0553, 3.8071, 3.5415, 2.0460, 3.9970, 2.3365, 3.3061], + device='cuda:3'), covar=tensor([0.0477, 0.0213, 0.0165, 0.0405, 0.0696, 0.0199, 0.0526, 0.0165], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0182, 0.0184, 0.0206, 0.0194, 0.0183, 0.0191, 0.0187], + 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') +2022-11-16 06:17:37,477 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 +2022-11-16 06:17:39,173 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3214, 2.6535, 3.1073, 4.1266, 4.1121, 3.3095, 3.0248, 4.2033], + device='cuda:3'), covar=tensor([0.0574, 0.3141, 0.2102, 0.2914, 0.1047, 0.2618, 0.1869, 0.0606], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0194, 0.0184, 0.0297, 0.0222, 0.0200, 0.0186, 0.0243], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 06:17:47,396 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.806e+01 1.427e+02 1.803e+02 2.265e+02 3.823e+02, threshold=3.607e+02, percent-clipped=0.0 +2022-11-16 06:18:01,506 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87358.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:18:04,458 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87362.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:18:06,212 INFO [train.py:876] (3/4) Epoch 13, batch 100, loss[loss=0.09633, simple_loss=0.1256, pruned_loss=0.03356, over 5587.00 frames. ], tot_loss[loss=0.1045, simple_loss=0.1352, pruned_loss=0.03688, over 429784.54 frames. ], batch size: 43, lr: 6.32e-03, grad_scale: 16.0 +2022-11-16 06:18:32,900 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7017, 3.7404, 3.7084, 3.5578, 3.7260, 3.4999, 1.3796, 3.8291], + device='cuda:3'), covar=tensor([0.0308, 0.0389, 0.0357, 0.0353, 0.0423, 0.0509, 0.3346, 0.0351], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0088, 0.0088, 0.0082, 0.0103, 0.0090, 0.0130, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:18:33,641 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5491, 1.8484, 2.3353, 2.2917, 2.3529, 1.6420, 2.2310, 2.4293], + device='cuda:3'), covar=tensor([0.0630, 0.1196, 0.0697, 0.0781, 0.0818, 0.1467, 0.1042, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0193, 0.0213, 0.0211, 0.0240, 0.0196, 0.0223, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:18:45,464 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87423.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:18:55,225 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.013e+01 1.498e+02 1.837e+02 2.189e+02 4.153e+02, threshold=3.674e+02, percent-clipped=6.0 +2022-11-16 06:19:10,507 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87461.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:19:12,363 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5426, 3.6277, 3.6244, 3.3029, 3.6037, 3.3988, 1.3945, 3.7302], + device='cuda:3'), covar=tensor([0.0318, 0.0230, 0.0263, 0.0333, 0.0315, 0.0397, 0.2866, 0.0281], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0088, 0.0082, 0.0103, 0.0090, 0.0131, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:19:12,928 INFO [train.py:876] (3/4) Epoch 13, batch 200, loss[loss=0.111, simple_loss=0.1438, pruned_loss=0.03911, over 5665.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1387, pruned_loss=0.03949, over 690736.23 frames. ], batch size: 36, lr: 6.32e-03, grad_scale: 16.0 +2022-11-16 06:19:17,087 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87470.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:19:40,379 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87505.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:19:49,629 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87518.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:19:52,424 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87522.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:19:53,686 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9328, 4.3455, 3.8991, 4.2651, 4.2712, 3.5652, 3.8840, 3.7643], + device='cuda:3'), covar=tensor([0.0529, 0.0475, 0.1556, 0.0463, 0.0509, 0.0651, 0.0940, 0.0836], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0176, 0.0273, 0.0174, 0.0221, 0.0172, 0.0189, 0.0177], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:20:00,745 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87534.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:03,213 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.494e+01 1.564e+02 1.812e+02 2.322e+02 4.189e+02, threshold=3.625e+02, percent-clipped=2.0 +2022-11-16 06:20:21,146 INFO [train.py:876] (3/4) Epoch 13, batch 300, loss[loss=0.08963, simple_loss=0.1345, pruned_loss=0.02238, over 5547.00 frames. ], tot_loss[loss=0.1079, simple_loss=0.1377, pruned_loss=0.03907, over 852054.90 frames. ], batch size: 13, lr: 6.32e-03, grad_scale: 16.0 +2022-11-16 06:20:21,978 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87566.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 06:20:24,787 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.64 vs. limit=5.0 +2022-11-16 06:20:33,024 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87582.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:34,424 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87584.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:35,589 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87586.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:37,647 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87589.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:48,690 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87605.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:20:52,703 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7196, 2.7635, 2.2708, 2.9538, 2.3070, 2.5910, 2.5164, 3.2836], + device='cuda:3'), covar=tensor([0.0860, 0.0960, 0.1613, 0.0919, 0.1275, 0.0728, 0.1171, 0.1132], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0105, 0.0103, 0.0103, 0.0092, 0.0101, 0.0096, 0.0081], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:21:00,236 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0786, 0.7870, 0.9593, 0.6847, 1.1758, 1.1153, 0.5055, 0.8648], + device='cuda:3'), covar=tensor([0.0327, 0.0439, 0.0359, 0.0563, 0.0365, 0.0319, 0.0980, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0025, 0.0017, 0.0021, 0.0018, 0.0016, 0.0024, 0.0017], + device='cuda:3'), out_proj_covar=tensor([8.8359e-05, 1.2151e-04, 9.2981e-05, 1.0827e-04, 9.5139e-05, 8.7841e-05, + 1.1732e-04, 8.9943e-05], device='cuda:3') +2022-11-16 06:21:11,330 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.900e+01 1.347e+02 1.607e+02 1.950e+02 4.005e+02, threshold=3.214e+02, percent-clipped=2.0 +2022-11-16 06:21:16,209 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87645.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:21:19,726 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87650.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:21:24,834 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87658.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:21:29,302 INFO [train.py:876] (3/4) Epoch 13, batch 400, loss[loss=0.08871, simple_loss=0.124, pruned_loss=0.02669, over 5547.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1366, pruned_loss=0.03797, over 943706.92 frames. ], batch size: 13, lr: 6.31e-03, grad_scale: 16.0 +2022-11-16 06:21:30,109 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87666.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:21:57,181 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87706.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:21:59,243 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9899, 4.7608, 3.6572, 2.1102, 4.4784, 2.1357, 4.3536, 2.6520], + device='cuda:3'), covar=tensor([0.1195, 0.0163, 0.0470, 0.2043, 0.0229, 0.1591, 0.0295, 0.1579], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0105, 0.0116, 0.0113, 0.0103, 0.0120, 0.0101, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:22:01,456 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2022-11-16 06:22:05,082 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87718.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:22:19,046 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.955e+01 1.567e+02 1.911e+02 2.428e+02 4.922e+02, threshold=3.823e+02, percent-clipped=4.0 +2022-11-16 06:22:37,357 INFO [train.py:876] (3/4) Epoch 13, batch 500, loss[loss=0.1039, simple_loss=0.1357, pruned_loss=0.03608, over 5737.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1344, pruned_loss=0.03669, over 1003017.25 frames. ], batch size: 27, lr: 6.31e-03, grad_scale: 16.0 +2022-11-16 06:23:12,963 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87817.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:23:17,608 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5256, 1.5727, 1.4005, 1.3310, 1.6861, 1.5406, 1.2784, 1.1932], + device='cuda:3'), covar=tensor([0.0038, 0.0061, 0.0049, 0.0081, 0.0066, 0.0054, 0.0052, 0.0061], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0029, 0.0028, 0.0037, 0.0032, 0.0029, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.8506e-05, 2.7092e-05, 2.5657e-05, 3.5341e-05, 2.9645e-05, 2.7761e-05, + 3.5070e-05, 3.2801e-05], device='cuda:3') +2022-11-16 06:23:24,722 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3693, 4.5537, 4.3396, 4.2253, 4.4184, 4.2446, 1.9400, 4.6893], + device='cuda:3'), covar=tensor([0.0260, 0.0269, 0.0298, 0.0263, 0.0251, 0.0405, 0.2604, 0.0235], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0087, 0.0087, 0.0081, 0.0102, 0.0089, 0.0129, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:23:26,585 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.493e+01 1.445e+02 1.920e+02 2.398e+02 4.024e+02, threshold=3.840e+02, percent-clipped=2.0 +2022-11-16 06:23:42,719 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87861.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:23:45,207 INFO [train.py:876] (3/4) Epoch 13, batch 600, loss[loss=0.1463, simple_loss=0.1741, pruned_loss=0.0593, over 5726.00 frames. ], tot_loss[loss=0.1051, simple_loss=0.1357, pruned_loss=0.03724, over 1035611.37 frames. ], batch size: 27, lr: 6.31e-03, grad_scale: 16.0 +2022-11-16 06:23:59,299 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87886.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:24:31,960 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87934.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:24:35,162 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.722e+01 1.443e+02 1.741e+02 2.053e+02 3.488e+02, threshold=3.481e+02, percent-clipped=0.0 +2022-11-16 06:24:35,933 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87940.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 06:24:39,196 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:24:49,806 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87961.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:24:52,674 INFO [train.py:876] (3/4) Epoch 13, batch 700, loss[loss=0.1424, simple_loss=0.1639, pruned_loss=0.06044, over 5547.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.1361, pruned_loss=0.03739, over 1051774.67 frames. ], batch size: 46, lr: 6.30e-03, grad_scale: 8.0 +2022-11-16 06:25:06,171 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7753, 3.8022, 3.8220, 3.9003, 3.6028, 3.6690, 4.2712, 3.8069], + device='cuda:3'), covar=tensor([0.0561, 0.0926, 0.0575, 0.1261, 0.0567, 0.0399, 0.0827, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0108, 0.0096, 0.0122, 0.0091, 0.0080, 0.0146, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:25:29,003 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88018.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:25:41,726 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9128, 2.7247, 2.1519, 1.5629, 2.5778, 1.1895, 2.6546, 1.7642], + device='cuda:3'), covar=tensor([0.1141, 0.0270, 0.1071, 0.1408, 0.0306, 0.1867, 0.0312, 0.1156], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0115, 0.0112, 0.0102, 0.0119, 0.0099, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:25:42,923 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.823e+01 1.432e+02 1.789e+02 2.095e+02 4.590e+02, threshold=3.577e+02, percent-clipped=1.0 +2022-11-16 06:26:00,229 INFO [train.py:876] (3/4) Epoch 13, batch 800, loss[loss=0.08969, simple_loss=0.133, pruned_loss=0.02317, over 5717.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.136, pruned_loss=0.03746, over 1067209.48 frames. ], batch size: 15, lr: 6.30e-03, grad_scale: 8.0 +2022-11-16 06:26:01,330 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:26:17,110 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2022-11-16 06:26:24,663 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88100.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:26:36,777 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88117.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:26:51,415 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.479e+01 1.431e+02 1.753e+02 2.206e+02 3.833e+02, threshold=3.505e+02, percent-clipped=1.0 +2022-11-16 06:27:05,811 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88161.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 06:27:05,853 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88161.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:27:08,635 INFO [train.py:876] (3/4) Epoch 13, batch 900, loss[loss=0.06103, simple_loss=0.105, pruned_loss=0.008536, over 5745.00 frames. ], tot_loss[loss=0.1057, simple_loss=0.1364, pruned_loss=0.03751, over 1069118.09 frames. ], batch size: 15, lr: 6.30e-03, grad_scale: 8.0 +2022-11-16 06:27:08,679 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88165.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:27:38,499 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88209.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:27:54,821 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1500, 3.1574, 2.8185, 3.2396, 2.6566, 3.4847, 3.3792, 3.6359], + device='cuda:3'), covar=tensor([0.0878, 0.1917, 0.1650, 0.1428, 0.1438, 0.0938, 0.1099, 0.1221], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0108, 0.0105, 0.0106, 0.0095, 0.0104, 0.0099, 0.0083], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:27:59,344 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.909e+01 1.508e+02 1.868e+02 2.272e+02 4.107e+02, threshold=3.735e+02, percent-clipped=5.0 +2022-11-16 06:28:00,102 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88240.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:28:03,439 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88245.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:28:08,090 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 +2022-11-16 06:28:14,167 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88261.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 06:28:16,634 INFO [train.py:876] (3/4) Epoch 13, batch 1000, loss[loss=0.1167, simple_loss=0.1408, pruned_loss=0.04624, over 5332.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1357, pruned_loss=0.03741, over 1079002.40 frames. ], batch size: 9, lr: 6.29e-03, grad_scale: 8.0 +2022-11-16 06:28:32,435 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88288.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:28:34,844 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88291.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:28:36,009 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:28:45,900 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9970, 3.2101, 2.8134, 3.2753, 2.6609, 3.1955, 3.3343, 3.8355], + device='cuda:3'), covar=tensor([0.0941, 0.1176, 0.1382, 0.0926, 0.1358, 0.0998, 0.0879, 0.0974], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0109, 0.0106, 0.0107, 0.0095, 0.0104, 0.0099, 0.0084], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:28:46,441 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88309.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:28:46,714 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 +2022-11-16 06:29:06,114 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9625, 2.1005, 1.9789, 1.8017, 2.0920, 1.7077, 1.7215, 1.6795], + device='cuda:3'), covar=tensor([0.0052, 0.0058, 0.0047, 0.0066, 0.0067, 0.0123, 0.0057, 0.0049], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0028, 0.0028, 0.0036, 0.0032, 0.0029, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.8047e-05, 2.6673e-05, 2.5311e-05, 3.4846e-05, 2.9366e-05, 2.7424e-05, + 3.5096e-05, 3.2652e-05], device='cuda:3') +2022-11-16 06:29:06,586 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.385e+01 1.405e+02 1.701e+02 2.123e+02 3.653e+02, threshold=3.402e+02, percent-clipped=0.0 +2022-11-16 06:29:16,007 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88352.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:29:24,127 INFO [train.py:876] (3/4) Epoch 13, batch 1100, loss[loss=0.07499, simple_loss=0.1117, pruned_loss=0.01915, over 5494.00 frames. ], tot_loss[loss=0.1049, simple_loss=0.1354, pruned_loss=0.03724, over 1080660.24 frames. ], batch size: 10, lr: 6.29e-03, grad_scale: 8.0 +2022-11-16 06:30:08,885 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5483, 1.9386, 2.4190, 2.3412, 2.3617, 1.7575, 2.2963, 2.4594], + device='cuda:3'), covar=tensor([0.0570, 0.1082, 0.0591, 0.0731, 0.0765, 0.1215, 0.0819, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0191, 0.0216, 0.0210, 0.0241, 0.0196, 0.0226, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:30:13,852 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.379e+01 1.472e+02 1.907e+02 2.402e+02 6.330e+02, threshold=3.813e+02, percent-clipped=8.0 +2022-11-16 06:30:15,326 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9766, 2.1017, 2.1442, 1.7984, 1.9882, 1.6318, 1.9866, 1.7441], + device='cuda:3'), covar=tensor([0.0053, 0.0091, 0.0035, 0.0111, 0.0069, 0.0097, 0.0046, 0.0050], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0028, 0.0027, 0.0036, 0.0031, 0.0028, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.7592e-05, 2.6079e-05, 2.4798e-05, 3.4221e-05, 2.8808e-05, 2.6937e-05, + 3.4461e-05, 3.2140e-05], device='cuda:3') +2022-11-16 06:30:18,485 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9366, 1.8563, 2.0177, 1.7186, 2.0938, 1.7596, 1.6483, 1.5781], + device='cuda:3'), covar=tensor([0.0040, 0.0058, 0.0028, 0.0064, 0.0061, 0.0061, 0.0051, 0.0062], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0028, 0.0027, 0.0036, 0.0031, 0.0028, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.7535e-05, 2.6051e-05, 2.4754e-05, 3.4153e-05, 2.8765e-05, 2.6897e-05, + 3.4412e-05, 3.2082e-05], device='cuda:3') +2022-11-16 06:30:20,884 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 06:30:25,732 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88456.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:30:31,424 INFO [train.py:876] (3/4) Epoch 13, batch 1200, loss[loss=0.1152, simple_loss=0.1492, pruned_loss=0.04063, over 5603.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1365, pruned_loss=0.03808, over 1083163.37 frames. ], batch size: 23, lr: 6.28e-03, grad_scale: 8.0 +2022-11-16 06:30:53,161 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 06:31:21,163 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.717e+01 1.518e+02 1.854e+02 2.184e+02 7.084e+02, threshold=3.708e+02, percent-clipped=2.0 +2022-11-16 06:31:38,390 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9739, 1.7709, 2.2102, 1.7361, 1.3711, 2.7322, 2.2211, 1.9214], + device='cuda:3'), covar=tensor([0.1267, 0.1567, 0.1155, 0.3086, 0.2982, 0.0466, 0.1151, 0.1678], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0100, 0.0099, 0.0102, 0.0076, 0.0069, 0.0081, 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') +2022-11-16 06:31:38,931 INFO [train.py:876] (3/4) Epoch 13, batch 1300, loss[loss=0.1132, simple_loss=0.1483, pruned_loss=0.03908, over 5684.00 frames. ], tot_loss[loss=0.1052, simple_loss=0.1358, pruned_loss=0.0373, over 1079977.70 frames. ], batch size: 36, lr: 6.28e-03, grad_scale: 8.0 +2022-11-16 06:32:28,430 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.235e+01 1.439e+02 1.728e+02 2.189e+02 4.268e+02, threshold=3.455e+02, percent-clipped=2.0 +2022-11-16 06:32:31,878 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2120, 2.8401, 3.9443, 3.5610, 4.3031, 3.3179, 3.8670, 4.3801], + device='cuda:3'), covar=tensor([0.0519, 0.1379, 0.0695, 0.1142, 0.0374, 0.1202, 0.0967, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0192, 0.0218, 0.0212, 0.0244, 0.0198, 0.0227, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:32:33,698 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88647.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:32:39,067 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2022-11-16 06:32:45,381 INFO [train.py:876] (3/4) Epoch 13, batch 1400, loss[loss=0.0641, simple_loss=0.09558, pruned_loss=0.01631, over 5503.00 frames. ], tot_loss[loss=0.1036, simple_loss=0.1343, pruned_loss=0.03642, over 1085772.17 frames. ], batch size: 10, lr: 6.28e-03, grad_scale: 8.0 +2022-11-16 06:32:48,675 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5478, 5.0204, 5.2498, 4.9145, 5.6142, 5.3905, 4.8020, 5.6012], + device='cuda:3'), covar=tensor([0.0272, 0.0337, 0.0423, 0.0399, 0.0271, 0.0221, 0.0311, 0.0276], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0152, 0.0109, 0.0142, 0.0180, 0.0107, 0.0126, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 06:33:21,730 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8663, 3.8815, 3.9613, 3.8073, 3.9660, 3.8002, 1.4009, 4.0523], + device='cuda:3'), covar=tensor([0.0260, 0.0326, 0.0298, 0.0360, 0.0311, 0.0340, 0.3400, 0.0321], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0090, 0.0090, 0.0085, 0.0105, 0.0092, 0.0133, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:33:34,912 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.374e+02 1.560e+02 2.014e+02 3.886e+02, threshold=3.121e+02, percent-clipped=4.0 +2022-11-16 06:33:42,315 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0007, 2.5728, 2.5174, 1.6711, 2.8042, 2.8241, 2.9377, 3.2242], + device='cuda:3'), covar=tensor([0.1892, 0.1730, 0.1464, 0.2635, 0.0772, 0.1385, 0.0594, 0.0925], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0184, 0.0170, 0.0183, 0.0183, 0.0203, 0.0169, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:33:46,724 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88756.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:33:52,517 INFO [train.py:876] (3/4) Epoch 13, batch 1500, loss[loss=0.1093, simple_loss=0.1368, pruned_loss=0.04087, over 5625.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1336, pruned_loss=0.03524, over 1089140.96 frames. ], batch size: 32, lr: 6.27e-03, grad_scale: 8.0 +2022-11-16 06:34:19,412 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88804.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:34:26,635 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9321, 2.8400, 2.4351, 3.0138, 2.3054, 2.5483, 2.8191, 3.6854], + device='cuda:3'), covar=tensor([0.0830, 0.1639, 0.2084, 0.0938, 0.1570, 0.1174, 0.1180, 0.0441], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0107, 0.0104, 0.0106, 0.0093, 0.0103, 0.0098, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:34:36,709 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5006, 2.1545, 3.1772, 2.8392, 3.1733, 2.2359, 2.9380, 3.5358], + device='cuda:3'), covar=tensor([0.0608, 0.1601, 0.0956, 0.1330, 0.0954, 0.1654, 0.1194, 0.0842], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0191, 0.0215, 0.0208, 0.0241, 0.0196, 0.0224, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:34:42,724 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.503e+02 1.931e+02 2.477e+02 5.840e+02, threshold=3.862e+02, percent-clipped=6.0 +2022-11-16 06:35:00,132 INFO [train.py:876] (3/4) Epoch 13, batch 1600, loss[loss=0.1147, simple_loss=0.139, pruned_loss=0.04523, over 5536.00 frames. ], tot_loss[loss=0.1003, simple_loss=0.1322, pruned_loss=0.03423, over 1090376.98 frames. ], batch size: 16, lr: 6.27e-03, grad_scale: 8.0 +2022-11-16 06:35:49,234 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.383e+02 1.790e+02 2.013e+02 5.184e+02, threshold=3.580e+02, percent-clipped=2.0 +2022-11-16 06:35:55,093 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88947.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:36:07,120 INFO [train.py:876] (3/4) Epoch 13, batch 1700, loss[loss=0.06885, simple_loss=0.1015, pruned_loss=0.0181, over 5722.00 frames. ], tot_loss[loss=0.1026, simple_loss=0.1339, pruned_loss=0.03569, over 1092005.06 frames. ], batch size: 11, lr: 6.27e-03, grad_scale: 8.0 +2022-11-16 06:36:09,409 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-16 06:36:26,873 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88995.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:36:42,014 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2912, 3.1917, 3.1658, 2.9826, 1.8132, 3.2084, 2.1024, 2.9204], + device='cuda:3'), covar=tensor([0.0430, 0.0191, 0.0181, 0.0292, 0.0591, 0.0198, 0.0484, 0.0231], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0181, 0.0186, 0.0207, 0.0196, 0.0183, 0.0193, 0.0187], + 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') +2022-11-16 06:36:49,154 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89027.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:36:56,973 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.337e+01 1.392e+02 1.730e+02 2.257e+02 5.092e+02, threshold=3.461e+02, percent-clipped=3.0 +2022-11-16 06:37:14,201 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2658, 2.7133, 2.9292, 2.7140, 1.6649, 2.8457, 1.9629, 2.6125], + device='cuda:3'), covar=tensor([0.0309, 0.0233, 0.0163, 0.0243, 0.0480, 0.0213, 0.0466, 0.0194], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0179, 0.0184, 0.0206, 0.0195, 0.0182, 0.0191, 0.0186], + 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') +2022-11-16 06:37:15,265 INFO [train.py:876] (3/4) Epoch 13, batch 1800, loss[loss=0.121, simple_loss=0.1503, pruned_loss=0.04585, over 5559.00 frames. ], tot_loss[loss=0.1052, simple_loss=0.1359, pruned_loss=0.03727, over 1088274.90 frames. ], batch size: 25, lr: 6.26e-03, grad_scale: 8.0 +2022-11-16 06:37:30,559 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89088.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:38:04,921 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.586e+01 1.379e+02 1.721e+02 2.183e+02 4.295e+02, threshold=3.442e+02, percent-clipped=5.0 +2022-11-16 06:38:13,628 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2022-11-16 06:38:23,032 INFO [train.py:876] (3/4) Epoch 13, batch 1900, loss[loss=0.0906, simple_loss=0.1362, pruned_loss=0.02248, over 5565.00 frames. ], tot_loss[loss=0.1057, simple_loss=0.1367, pruned_loss=0.03735, over 1089154.76 frames. ], batch size: 16, lr: 6.26e-03, grad_scale: 8.0 +2022-11-16 06:38:25,957 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89169.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:38:35,782 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0800, 2.7659, 2.7334, 1.7697, 2.9108, 3.1469, 2.9666, 3.3012], + device='cuda:3'), covar=tensor([0.1779, 0.1431, 0.1031, 0.2384, 0.0636, 0.0778, 0.0460, 0.0764], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0185, 0.0169, 0.0184, 0.0183, 0.0203, 0.0169, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:39:00,611 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5708, 4.5055, 3.2230, 4.2676, 3.4703, 3.1550, 2.5374, 3.6877], + device='cuda:3'), covar=tensor([0.1316, 0.0227, 0.0984, 0.0271, 0.0707, 0.1008, 0.1734, 0.0383], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0143, 0.0156, 0.0149, 0.0173, 0.0169, 0.0159, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:39:06,971 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89230.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:39:12,533 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.915e+01 1.392e+02 1.772e+02 2.206e+02 3.328e+02, threshold=3.543e+02, percent-clipped=0.0 +2022-11-16 06:39:19,002 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6932, 2.8934, 2.9678, 2.6977, 2.9331, 2.8472, 1.2569, 2.9341], + device='cuda:3'), covar=tensor([0.0530, 0.0500, 0.0447, 0.0446, 0.0471, 0.0458, 0.3615, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0088, 0.0087, 0.0082, 0.0102, 0.0089, 0.0129, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:39:26,218 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2173, 4.2273, 4.1648, 3.9533, 4.2153, 4.1256, 1.5834, 4.3299], + device='cuda:3'), covar=tensor([0.0264, 0.0368, 0.0377, 0.0472, 0.0423, 0.0447, 0.3369, 0.0361], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0088, 0.0087, 0.0082, 0.0102, 0.0089, 0.0129, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:39:29,336 INFO [train.py:876] (3/4) Epoch 13, batch 2000, loss[loss=0.1493, simple_loss=0.1531, pruned_loss=0.07277, over 4208.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1367, pruned_loss=0.03791, over 1088639.10 frames. ], batch size: 181, lr: 6.26e-03, grad_scale: 8.0 +2022-11-16 06:39:35,422 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 +2022-11-16 06:39:53,779 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 06:40:02,853 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9676, 3.2298, 4.2148, 3.8572, 5.0056, 3.8146, 4.4474, 4.8355], + device='cuda:3'), covar=tensor([0.0383, 0.1321, 0.0654, 0.1413, 0.0198, 0.1270, 0.1113, 0.0512], + device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0191, 0.0213, 0.0207, 0.0238, 0.0194, 0.0222, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:40:09,812 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9283, 2.6080, 3.2058, 1.6157, 2.9902, 3.5561, 3.3733, 3.6034], + device='cuda:3'), covar=tensor([0.2031, 0.1828, 0.1214, 0.2849, 0.0810, 0.0811, 0.0485, 0.0745], + device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0185, 0.0170, 0.0184, 0.0182, 0.0203, 0.0170, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:40:20,291 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.714e+01 1.486e+02 1.827e+02 2.274e+02 3.584e+02, threshold=3.655e+02, percent-clipped=1.0 +2022-11-16 06:40:29,720 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5308, 1.8804, 2.2819, 2.2495, 2.2986, 1.7423, 2.2803, 2.4745], + device='cuda:3'), covar=tensor([0.0602, 0.1079, 0.0792, 0.0886, 0.0793, 0.1160, 0.0871, 0.0677], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0192, 0.0215, 0.0209, 0.0240, 0.0195, 0.0223, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:40:33,200 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-16 06:40:37,211 INFO [train.py:876] (3/4) Epoch 13, batch 2100, loss[loss=0.08289, simple_loss=0.123, pruned_loss=0.0214, over 5762.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1359, pruned_loss=0.03766, over 1083783.71 frames. ], batch size: 15, lr: 6.25e-03, grad_scale: 8.0 +2022-11-16 06:40:38,602 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7340, 4.7207, 3.3889, 4.5529, 3.7333, 3.2716, 2.6660, 3.8767], + device='cuda:3'), covar=tensor([0.1369, 0.0207, 0.0928, 0.0337, 0.0519, 0.0850, 0.1748, 0.0360], + device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0142, 0.0156, 0.0147, 0.0172, 0.0168, 0.0156, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:40:43,520 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89374.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:40:49,286 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89383.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:41:17,734 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89425.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:41:24,440 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89435.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:41:26,874 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.665e+01 1.565e+02 1.857e+02 2.395e+02 6.396e+02, threshold=3.713e+02, percent-clipped=5.0 +2022-11-16 06:41:27,078 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89439.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 06:41:27,654 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8727, 4.7574, 3.5530, 2.5133, 4.4476, 2.1392, 4.3407, 2.7450], + device='cuda:3'), covar=tensor([0.1300, 0.0141, 0.0542, 0.1723, 0.0208, 0.1785, 0.0166, 0.1565], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0105, 0.0117, 0.0112, 0.0104, 0.0120, 0.0102, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:41:41,556 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5669, 2.3834, 2.7808, 1.9187, 1.7786, 3.3455, 2.6946, 2.4497], + device='cuda:3'), covar=tensor([0.1138, 0.1339, 0.1032, 0.2683, 0.3983, 0.0646, 0.1653, 0.1630], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0104, 0.0102, 0.0105, 0.0079, 0.0072, 0.0084, 0.0095], + 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') +2022-11-16 06:41:44,656 INFO [train.py:876] (3/4) Epoch 13, batch 2200, loss[loss=0.09627, simple_loss=0.1259, pruned_loss=0.03334, over 5510.00 frames. ], tot_loss[loss=0.104, simple_loss=0.1345, pruned_loss=0.03679, over 1087431.86 frames. ], batch size: 17, lr: 6.25e-03, grad_scale: 8.0 +2022-11-16 06:41:54,354 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4919, 4.2537, 3.3016, 2.1423, 3.9185, 1.8676, 3.9830, 2.4838], + device='cuda:3'), covar=tensor([0.1851, 0.0293, 0.0867, 0.2432, 0.0357, 0.2371, 0.0367, 0.2123], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0105, 0.0116, 0.0112, 0.0105, 0.0120, 0.0103, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:41:58,291 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89486.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:42:04,177 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2022-11-16 06:42:07,608 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89500.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:42:14,370 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89509.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:42:20,380 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.04 vs. limit=5.0 +2022-11-16 06:42:24,622 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89525.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:42:33,577 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.106e+01 1.371e+02 1.691e+02 2.068e+02 3.234e+02, threshold=3.383e+02, percent-clipped=0.0 +2022-11-16 06:42:51,730 INFO [train.py:876] (3/4) Epoch 13, batch 2300, loss[loss=0.07118, simple_loss=0.1087, pruned_loss=0.01681, over 5356.00 frames. ], tot_loss[loss=0.1026, simple_loss=0.1334, pruned_loss=0.03596, over 1084397.92 frames. ], batch size: 6, lr: 6.25e-03, grad_scale: 8.0 +2022-11-16 06:42:55,285 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89570.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:43:29,343 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7494, 2.2715, 3.4835, 2.9840, 3.4900, 2.4954, 3.2756, 3.8236], + device='cuda:3'), covar=tensor([0.0838, 0.1715, 0.0777, 0.1630, 0.0833, 0.1600, 0.1134, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0193, 0.0214, 0.0212, 0.0242, 0.0198, 0.0226, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:43:36,641 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 06:43:41,342 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.175e+01 1.502e+02 1.727e+02 2.123e+02 1.355e+03, threshold=3.453e+02, percent-clipped=6.0 +2022-11-16 06:44:00,255 INFO [train.py:876] (3/4) Epoch 13, batch 2400, loss[loss=0.121, simple_loss=0.1424, pruned_loss=0.04979, over 4985.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.133, pruned_loss=0.03583, over 1086580.12 frames. ], batch size: 109, lr: 6.24e-03, grad_scale: 8.0 +2022-11-16 06:44:12,333 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:44:14,485 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89683.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:44:27,908 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4249, 1.1307, 1.1085, 0.8122, 1.2448, 1.2191, 0.7063, 1.0257], + device='cuda:3'), covar=tensor([0.0373, 0.0403, 0.0384, 0.0631, 0.0404, 0.0375, 0.0956, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0026, 0.0018, 0.0022, 0.0019, 0.0017, 0.0025, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.2862e-05, 1.2952e-04, 9.8713e-05, 1.1327e-04, 1.0098e-04, 9.5001e-05, + 1.2302e-04, 9.3952e-05], device='cuda:3') +2022-11-16 06:44:32,465 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3285, 4.3850, 3.9398, 3.8265, 4.3419, 4.1038, 1.7380, 4.4733], + device='cuda:3'), covar=tensor([0.0227, 0.0292, 0.0311, 0.0381, 0.0295, 0.0373, 0.2947, 0.0288], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0089, 0.0089, 0.0084, 0.0103, 0.0090, 0.0131, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:44:32,891 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2022-11-16 06:44:35,109 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0614, 1.4964, 1.0255, 1.0635, 1.4163, 1.2580, 0.9033, 1.3423], + device='cuda:3'), covar=tensor([0.0067, 0.0048, 0.0074, 0.0063, 0.0049, 0.0071, 0.0104, 0.0054], + device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0058, 0.0058, 0.0063, 0.0060, 0.0056, 0.0055, 0.0053], + device='cuda:3'), out_proj_covar=tensor([5.5768e-05, 5.1490e-05, 5.0572e-05, 5.5867e-05, 5.3228e-05, 4.8920e-05, + 4.8866e-05, 4.6482e-05], device='cuda:3') +2022-11-16 06:44:35,222 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2022-11-16 06:44:47,043 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89730.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:44:47,659 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89731.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:44:52,830 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.552e+02 1.853e+02 2.424e+02 4.958e+02, threshold=3.705e+02, percent-clipped=7.0 +2022-11-16 06:44:54,429 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89741.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:45:09,830 INFO [train.py:876] (3/4) Epoch 13, batch 2500, loss[loss=0.1773, simple_loss=0.1725, pruned_loss=0.09108, over 3087.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1343, pruned_loss=0.03633, over 1084442.60 frames. ], batch size: 284, lr: 6.24e-03, grad_scale: 8.0 +2022-11-16 06:45:21,168 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89781.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:45:30,647 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89795.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:45:31,393 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0036, 1.9847, 2.6701, 2.4344, 2.4742, 2.0408, 2.5365, 2.9118], + device='cuda:3'), covar=tensor([0.0751, 0.1504, 0.0922, 0.1270, 0.0977, 0.1416, 0.1173, 0.0973], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0193, 0.0214, 0.0211, 0.0240, 0.0197, 0.0226, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:45:50,791 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89825.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:46:00,865 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.702e+01 1.449e+02 1.693e+02 2.128e+02 5.529e+02, threshold=3.385e+02, percent-clipped=3.0 +2022-11-16 06:46:10,382 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2022-11-16 06:46:17,694 INFO [train.py:876] (3/4) Epoch 13, batch 2600, loss[loss=0.1042, simple_loss=0.1354, pruned_loss=0.03646, over 5560.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1364, pruned_loss=0.03852, over 1078980.06 frames. ], batch size: 46, lr: 6.24e-03, grad_scale: 8.0 +2022-11-16 06:46:17,788 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:46:17,866 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:46:23,250 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89873.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:46:58,616 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89926.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 06:47:07,228 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.464e+01 1.385e+02 1.759e+02 2.199e+02 3.359e+02, threshold=3.518e+02, percent-clipped=0.0 +2022-11-16 06:47:24,941 INFO [train.py:876] (3/4) Epoch 13, batch 2700, loss[loss=0.05381, simple_loss=0.08333, pruned_loss=0.01215, over 3470.00 frames. ], tot_loss[loss=0.1043, simple_loss=0.1344, pruned_loss=0.03709, over 1082448.88 frames. ], batch size: 4, lr: 6.23e-03, grad_scale: 16.0 +2022-11-16 06:47:39,350 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2413, 3.6985, 3.4223, 3.7251, 3.7718, 3.1373, 3.3610, 3.4110], + device='cuda:3'), covar=tensor([0.1003, 0.0496, 0.1053, 0.0430, 0.0405, 0.0561, 0.0840, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0179, 0.0276, 0.0177, 0.0223, 0.0177, 0.0192, 0.0176], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:47:55,761 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2022-11-16 06:48:12,096 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90030.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:48:15,891 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90036.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:48:17,698 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.688e+01 1.446e+02 1.718e+02 2.130e+02 5.119e+02, threshold=3.437e+02, percent-clipped=5.0 +2022-11-16 06:48:35,917 INFO [train.py:876] (3/4) Epoch 13, batch 2800, loss[loss=0.1072, simple_loss=0.1412, pruned_loss=0.03658, over 5649.00 frames. ], tot_loss[loss=0.1033, simple_loss=0.1344, pruned_loss=0.03615, over 1085494.01 frames. ], batch size: 32, lr: 6.23e-03, grad_scale: 16.0 +2022-11-16 06:48:44,257 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90078.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:48:46,319 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90081.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:48:55,340 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90095.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:49:00,698 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3080, 5.1731, 3.7433, 2.3115, 4.8724, 2.2801, 4.8711, 2.8743], + device='cuda:3'), covar=tensor([0.1111, 0.0118, 0.0601, 0.1832, 0.0137, 0.1572, 0.0160, 0.1382], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0116, 0.0112, 0.0104, 0.0119, 0.0102, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:49:18,683 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90129.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:49:19,462 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90130.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:49:25,057 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.858e+01 1.324e+02 1.624e+02 2.114e+02 4.134e+02, threshold=3.247e+02, percent-clipped=3.0 +2022-11-16 06:49:27,799 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90143.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 06:49:43,061 INFO [train.py:876] (3/4) Epoch 13, batch 2900, loss[loss=0.1048, simple_loss=0.1342, pruned_loss=0.03774, over 5566.00 frames. ], tot_loss[loss=0.1036, simple_loss=0.1344, pruned_loss=0.03642, over 1086497.70 frames. ], batch size: 30, lr: 6.23e-03, grad_scale: 16.0 +2022-11-16 06:49:43,171 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90165.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:50:00,526 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90191.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:50:16,487 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90213.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:50:21,685 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90221.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 06:50:25,032 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0474, 1.6573, 2.1924, 1.9095, 1.8140, 2.0003, 2.0553, 1.5397], + device='cuda:3'), covar=tensor([0.0055, 0.0111, 0.0028, 0.0044, 0.0076, 0.0072, 0.0035, 0.0048], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0027, 0.0028, 0.0036, 0.0031, 0.0028, 0.0035, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.7937e-05, 2.5441e-05, 2.5150e-05, 3.4360e-05, 2.9100e-05, 2.7137e-05, + 3.3371e-05, 3.2257e-05], device='cuda:3') +2022-11-16 06:50:28,901 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90232.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:50:33,305 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.557e+01 1.381e+02 1.773e+02 2.128e+02 3.504e+02, threshold=3.546e+02, percent-clipped=3.0 +2022-11-16 06:50:37,667 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2022-11-16 06:50:51,297 INFO [train.py:876] (3/4) Epoch 13, batch 3000, loss[loss=0.09179, simple_loss=0.1348, pruned_loss=0.0244, over 5723.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1349, pruned_loss=0.03779, over 1082997.99 frames. ], batch size: 17, lr: 6.22e-03, grad_scale: 16.0 +2022-11-16 06:50:51,297 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 06:50:56,577 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0844, 3.9872, 3.8538, 3.8230, 4.1184, 3.8348, 1.8970, 4.1448], + device='cuda:3'), covar=tensor([0.0204, 0.0265, 0.0224, 0.0289, 0.0166, 0.0273, 0.2603, 0.0195], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0091, 0.0090, 0.0084, 0.0104, 0.0091, 0.0133, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:51:00,314 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0298, 1.9319, 2.5935, 1.8722, 1.8202, 3.0205, 2.2913, 2.1829], + device='cuda:3'), covar=tensor([0.0787, 0.1464, 0.0877, 0.2179, 0.2872, 0.0523, 0.1592, 0.1705], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0101, 0.0099, 0.0103, 0.0076, 0.0071, 0.0080, 0.0092], + 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') +2022-11-16 06:51:03,089 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4374, 4.2435, 4.5696, 4.1083, 4.1323, 4.3257, 4.7194, 4.6638], + device='cuda:3'), covar=tensor([0.0425, 0.1047, 0.0264, 0.1254, 0.0401, 0.0252, 0.0604, 0.0422], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0107, 0.0095, 0.0121, 0.0089, 0.0080, 0.0147, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:51:04,184 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9486, 1.2366, 1.9410, 1.8290, 1.8597, 1.6209, 1.6819, 1.3684], + device='cuda:3'), covar=tensor([0.0047, 0.0083, 0.0052, 0.0059, 0.0126, 0.0214, 0.0046, 0.0056], + device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0027, 0.0028, 0.0036, 0.0031, 0.0028, 0.0035, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.7871e-05, 2.5456e-05, 2.5137e-05, 3.4283e-05, 2.9077e-05, 2.7169e-05, + 3.3249e-05, 3.2123e-05], device='cuda:3') +2022-11-16 06:51:07,794 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.0351, 4.1974, 4.1319, 3.7579, 2.1299, 4.5339, 2.4913, 3.7878], + device='cuda:3'), covar=tensor([0.0350, 0.0155, 0.0135, 0.0335, 0.0649, 0.0112, 0.0708, 0.0129], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0178, 0.0181, 0.0204, 0.0192, 0.0178, 0.0189, 0.0183], + 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') +2022-11-16 06:51:08,998 INFO [train.py:908] (3/4) Epoch 13, validation: loss=0.1737, simple_loss=0.1855, pruned_loss=0.08091, over 1530663.00 frames. +2022-11-16 06:51:08,998 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 06:51:27,462 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:51:45,997 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4960, 3.4301, 3.5379, 3.3696, 3.5790, 3.5233, 1.3941, 3.7267], + device='cuda:3'), covar=tensor([0.0302, 0.0477, 0.0373, 0.0365, 0.0343, 0.0429, 0.3307, 0.0374], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0091, 0.0090, 0.0084, 0.0104, 0.0091, 0.0133, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 06:51:46,771 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7080, 2.0830, 2.2373, 2.8323, 2.9148, 2.2477, 1.9677, 2.9345], + device='cuda:3'), covar=tensor([0.1644, 0.2631, 0.2161, 0.1680, 0.1581, 0.3313, 0.2351, 0.1230], + device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0198, 0.0189, 0.0298, 0.0225, 0.0202, 0.0189, 0.0249], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 06:51:56,871 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90336.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:51:58,705 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.650e+01 1.477e+02 1.762e+02 2.223e+02 4.727e+02, threshold=3.524e+02, percent-clipped=4.0 +2022-11-16 06:52:06,690 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90351.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:52:16,335 INFO [train.py:876] (3/4) Epoch 13, batch 3100, loss[loss=0.1297, simple_loss=0.1341, pruned_loss=0.06262, over 4140.00 frames. ], tot_loss[loss=0.1042, simple_loss=0.1346, pruned_loss=0.03687, over 1090243.18 frames. ], batch size: 181, lr: 6.22e-03, grad_scale: 16.0 +2022-11-16 06:52:23,787 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2503, 3.1880, 3.1311, 1.7475, 2.8881, 3.3577, 3.3785, 3.6532], + device='cuda:3'), covar=tensor([0.1870, 0.1546, 0.0934, 0.2993, 0.0659, 0.0859, 0.0483, 0.0652], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0183, 0.0171, 0.0184, 0.0183, 0.0204, 0.0170, 0.0182], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:52:29,315 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90384.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:52:47,892 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90412.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:53:06,644 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.938e+01 1.382e+02 1.732e+02 2.119e+02 3.320e+02, threshold=3.464e+02, percent-clipped=0.0 +2022-11-16 06:53:19,991 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90459.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:53:23,666 INFO [train.py:876] (3/4) Epoch 13, batch 3200, loss[loss=0.07432, simple_loss=0.1057, pruned_loss=0.02148, over 4494.00 frames. ], tot_loss[loss=0.1029, simple_loss=0.1341, pruned_loss=0.03585, over 1089339.58 frames. ], batch size: 5, lr: 6.22e-03, grad_scale: 16.0 +2022-11-16 06:53:38,089 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90486.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:53:44,964 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90496.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:53:45,526 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6241, 3.3414, 3.4922, 3.2111, 3.6923, 3.5308, 3.4475, 3.6428], + device='cuda:3'), covar=tensor([0.0412, 0.0457, 0.0448, 0.0443, 0.0389, 0.0275, 0.0372, 0.0452], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0156, 0.0111, 0.0146, 0.0183, 0.0113, 0.0128, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 06:54:00,530 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90520.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:54:01,093 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90521.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 06:54:13,667 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.039e+02 1.437e+02 1.897e+02 2.279e+02 5.045e+02, threshold=3.794e+02, percent-clipped=5.0 +2022-11-16 06:54:14,511 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5361, 2.2372, 3.2735, 2.8451, 3.3076, 2.3624, 3.1029, 3.5569], + device='cuda:3'), covar=tensor([0.0943, 0.1678, 0.1053, 0.1680, 0.1005, 0.1718, 0.1335, 0.1061], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0196, 0.0218, 0.0213, 0.0244, 0.0199, 0.0228, 0.0233], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:54:17,007 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.15 vs. limit=5.0 +2022-11-16 06:54:23,956 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2546, 2.3380, 2.5370, 3.4060, 3.3830, 2.5877, 2.1935, 3.4824], + device='cuda:3'), covar=tensor([0.1354, 0.2197, 0.2205, 0.1865, 0.1237, 0.3153, 0.2297, 0.0706], + device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0200, 0.0192, 0.0304, 0.0228, 0.0205, 0.0192, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 06:54:25,902 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90557.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:54:30,905 INFO [train.py:876] (3/4) Epoch 13, batch 3300, loss[loss=0.104, simple_loss=0.1382, pruned_loss=0.03492, over 5632.00 frames. ], tot_loss[loss=0.1026, simple_loss=0.134, pruned_loss=0.03561, over 1088785.19 frames. ], batch size: 29, lr: 6.21e-03, grad_scale: 16.0 +2022-11-16 06:54:33,620 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90569.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:54:38,297 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3148, 1.1292, 1.1129, 1.0888, 1.3349, 1.3818, 0.7798, 0.9482], + device='cuda:3'), covar=tensor([0.0239, 0.0456, 0.0437, 0.0454, 0.0343, 0.0268, 0.0859, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0025, 0.0018, 0.0021, 0.0018, 0.0016, 0.0024, 0.0017], + device='cuda:3'), out_proj_covar=tensor([8.9544e-05, 1.2590e-04, 9.5725e-05, 1.0939e-04, 9.8074e-05, 9.1231e-05, + 1.1916e-04, 9.1685e-05], device='cuda:3') +2022-11-16 06:54:39,558 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1210, 4.7151, 4.9973, 4.5661, 5.2199, 5.0263, 4.4604, 5.1916], + device='cuda:3'), covar=tensor([0.0323, 0.0364, 0.0365, 0.0329, 0.0337, 0.0251, 0.0262, 0.0255], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0157, 0.0112, 0.0146, 0.0184, 0.0113, 0.0129, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 06:54:46,738 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90588.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:54:56,645 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8264, 2.8133, 2.4760, 2.9850, 2.2631, 2.4261, 2.8246, 3.4322], + device='cuda:3'), covar=tensor([0.1224, 0.1348, 0.1698, 0.1336, 0.1514, 0.1250, 0.1064, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0108, 0.0104, 0.0106, 0.0093, 0.0103, 0.0098, 0.0082], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 06:55:15,450 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7885, 4.8887, 3.2513, 4.5095, 3.6965, 3.2149, 2.6807, 4.2054], + device='cuda:3'), covar=tensor([0.1200, 0.0157, 0.0905, 0.0367, 0.0497, 0.0866, 0.1678, 0.0282], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0143, 0.0157, 0.0150, 0.0172, 0.0168, 0.0160, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:55:21,064 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.442e+01 1.383e+02 1.673e+02 2.134e+02 3.431e+02, threshold=3.345e+02, percent-clipped=0.0 +2022-11-16 06:55:38,732 INFO [train.py:876] (3/4) Epoch 13, batch 3400, loss[loss=0.1034, simple_loss=0.141, pruned_loss=0.03289, over 5683.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1345, pruned_loss=0.03546, over 1091882.79 frames. ], batch size: 19, lr: 6.21e-03, grad_scale: 16.0 +2022-11-16 06:55:44,158 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2054, 1.4306, 1.2213, 1.0151, 1.2052, 1.6122, 1.6205, 1.4601], + device='cuda:3'), covar=tensor([0.1097, 0.0884, 0.2197, 0.2398, 0.1479, 0.1142, 0.1042, 0.1317], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0181, 0.0170, 0.0184, 0.0182, 0.0201, 0.0170, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:55:44,195 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5506, 1.5104, 1.9028, 1.6524, 0.9948, 1.4270, 1.2164, 1.3373], + device='cuda:3'), covar=tensor([0.0144, 0.0081, 0.0076, 0.0099, 0.0231, 0.0102, 0.0180, 0.0136], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0181, 0.0183, 0.0205, 0.0194, 0.0181, 0.0190, 0.0186], + 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') +2022-11-16 06:56:07,884 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90707.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:56:29,388 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.201e+01 1.464e+02 1.802e+02 2.100e+02 5.077e+02, threshold=3.604e+02, percent-clipped=5.0 +2022-11-16 06:56:42,641 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8319, 2.9438, 2.2514, 2.5398, 1.8465, 2.3231, 1.7654, 2.4369], + device='cuda:3'), covar=tensor([0.1405, 0.0439, 0.1047, 0.0756, 0.2324, 0.1100, 0.1967, 0.0744], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0144, 0.0157, 0.0152, 0.0173, 0.0169, 0.0161, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:56:46,669 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3659, 1.1809, 1.1799, 0.9245, 1.2186, 1.3042, 0.7375, 0.9832], + device='cuda:3'), covar=tensor([0.0258, 0.0394, 0.0338, 0.0672, 0.0743, 0.0340, 0.0819, 0.0435], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0025, 0.0018, 0.0021, 0.0018, 0.0016, 0.0023, 0.0017], + device='cuda:3'), out_proj_covar=tensor([8.8923e-05, 1.2472e-04, 9.4792e-05, 1.0820e-04, 9.6481e-05, 9.0533e-05, + 1.1754e-04, 9.0449e-05], device='cuda:3') +2022-11-16 06:56:47,177 INFO [train.py:876] (3/4) Epoch 13, batch 3500, loss[loss=0.1378, simple_loss=0.1514, pruned_loss=0.06212, over 5357.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1344, pruned_loss=0.03599, over 1080601.46 frames. ], batch size: 70, lr: 6.21e-03, grad_scale: 16.0 +2022-11-16 06:57:00,244 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0805, 2.4060, 2.6553, 2.3352, 1.4726, 2.4357, 1.8007, 2.0645], + device='cuda:3'), covar=tensor([0.0309, 0.0177, 0.0159, 0.0245, 0.0463, 0.0184, 0.0439, 0.0226], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0183, 0.0185, 0.0207, 0.0195, 0.0183, 0.0192, 0.0188], + 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') +2022-11-16 06:57:00,826 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90786.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:57:21,073 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90815.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:57:33,473 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90834.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:57:36,721 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.181e+01 1.483e+02 1.758e+02 2.114e+02 3.884e+02, threshold=3.515e+02, percent-clipped=1.0 +2022-11-16 06:57:43,831 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90849.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:57:46,108 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90852.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:57:54,731 INFO [train.py:876] (3/4) Epoch 13, batch 3600, loss[loss=0.1116, simple_loss=0.1561, pruned_loss=0.03358, over 5655.00 frames. ], tot_loss[loss=0.103, simple_loss=0.1341, pruned_loss=0.03596, over 1082977.44 frames. ], batch size: 32, lr: 6.20e-03, grad_scale: 16.0 +2022-11-16 06:58:02,066 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3347, 4.2381, 2.9124, 4.0164, 3.2419, 2.7876, 2.2025, 3.5036], + device='cuda:3'), covar=tensor([0.1609, 0.0199, 0.1220, 0.0354, 0.0696, 0.1199, 0.2181, 0.0418], + device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0143, 0.0157, 0.0151, 0.0172, 0.0169, 0.0161, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:58:10,071 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:58:21,399 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9907, 2.6285, 2.6891, 1.4764, 2.6845, 2.8805, 2.7233, 3.0848], + device='cuda:3'), covar=tensor([0.1922, 0.1703, 0.1301, 0.3077, 0.0950, 0.1122, 0.0638, 0.0996], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0180, 0.0167, 0.0182, 0.0181, 0.0199, 0.0168, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:58:25,588 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90910.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:58:42,751 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:58:42,921 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2986, 2.8268, 3.9120, 3.3032, 4.3633, 3.1647, 3.7971, 4.2021], + device='cuda:3'), covar=tensor([0.0636, 0.1435, 0.0921, 0.1867, 0.0482, 0.1650, 0.1220, 0.1039], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0195, 0.0218, 0.0214, 0.0244, 0.0197, 0.0228, 0.0233], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:58:44,663 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.811e+01 1.461e+02 1.838e+02 2.236e+02 5.014e+02, threshold=3.676e+02, percent-clipped=2.0 +2022-11-16 06:58:48,982 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2022-11-16 06:58:56,069 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4507, 3.3808, 3.2668, 3.1623, 2.0185, 3.3493, 2.1603, 2.8959], + device='cuda:3'), covar=tensor([0.0398, 0.0183, 0.0217, 0.0280, 0.0515, 0.0174, 0.0553, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0182, 0.0185, 0.0207, 0.0194, 0.0183, 0.0191, 0.0187], + 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') +2022-11-16 06:59:02,292 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3967, 1.7983, 1.5711, 1.2507, 1.3509, 1.9186, 1.8205, 1.8554], + device='cuda:3'), covar=tensor([0.1365, 0.1076, 0.1758, 0.2342, 0.1450, 0.1168, 0.0929, 0.1276], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0180, 0.0168, 0.0182, 0.0181, 0.0199, 0.0167, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:59:02,754 INFO [train.py:876] (3/4) Epoch 13, batch 3700, loss[loss=0.1341, simple_loss=0.1351, pruned_loss=0.06657, over 4118.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.135, pruned_loss=0.0364, over 1086357.26 frames. ], batch size: 183, lr: 6.20e-03, grad_scale: 16.0 +2022-11-16 06:59:18,359 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2022-11-16 06:59:27,342 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3157, 3.0376, 3.8781, 3.5149, 4.3822, 3.0906, 3.9397, 4.2729], + device='cuda:3'), covar=tensor([0.0634, 0.1215, 0.0815, 0.1373, 0.0338, 0.1170, 0.0948, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0192, 0.0215, 0.0210, 0.0241, 0.0195, 0.0224, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 06:59:30,956 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91007.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 06:59:52,311 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.488e+01 1.377e+02 1.670e+02 2.030e+02 3.964e+02, threshold=3.341e+02, percent-clipped=2.0 +2022-11-16 07:00:03,059 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91055.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:00:09,530 INFO [train.py:876] (3/4) Epoch 13, batch 3800, loss[loss=0.1031, simple_loss=0.141, pruned_loss=0.03259, over 5563.00 frames. ], tot_loss[loss=0.1043, simple_loss=0.1358, pruned_loss=0.03645, over 1086336.78 frames. ], batch size: 24, lr: 6.19e-03, grad_scale: 16.0 +2022-11-16 07:00:40,829 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6020, 3.8635, 3.8754, 3.6794, 3.9698, 3.8810, 1.3541, 3.9736], + device='cuda:3'), covar=tensor([0.0528, 0.0363, 0.0340, 0.0357, 0.0346, 0.0389, 0.3961, 0.0423], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0089, 0.0082, 0.0103, 0.0090, 0.0132, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:00:43,432 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91115.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:00:59,664 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.615e+01 1.416e+02 1.762e+02 2.192e+02 4.990e+02, threshold=3.525e+02, percent-clipped=3.0 +2022-11-16 07:01:08,644 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91152.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:01:15,770 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91163.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:01:17,058 INFO [train.py:876] (3/4) Epoch 13, batch 3900, loss[loss=0.07901, simple_loss=0.1195, pruned_loss=0.01928, over 5573.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.1343, pruned_loss=0.03495, over 1095192.46 frames. ], batch size: 16, lr: 6.19e-03, grad_scale: 16.0 +2022-11-16 07:01:28,323 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8768, 2.9973, 3.1161, 2.7624, 3.0468, 2.8836, 1.1924, 3.1392], + device='cuda:3'), covar=tensor([0.0366, 0.0357, 0.0285, 0.0375, 0.0371, 0.0478, 0.2970, 0.0381], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0089, 0.0082, 0.0103, 0.0090, 0.0132, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:01:41,634 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91200.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:01:44,967 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91205.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:02:07,454 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.326e+01 1.380e+02 1.738e+02 2.230e+02 3.262e+02, threshold=3.475e+02, percent-clipped=0.0 +2022-11-16 07:02:25,432 INFO [train.py:876] (3/4) Epoch 13, batch 4000, loss[loss=0.07227, simple_loss=0.1199, pruned_loss=0.01233, over 5694.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1334, pruned_loss=0.03529, over 1094069.30 frames. ], batch size: 15, lr: 6.19e-03, grad_scale: 16.0 +2022-11-16 07:02:38,649 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91285.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:03:10,993 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 +2022-11-16 07:03:15,061 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.467e+01 1.389e+02 1.729e+02 2.051e+02 4.497e+02, threshold=3.458e+02, percent-clipped=2.0 +2022-11-16 07:03:19,975 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91346.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:03:23,185 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3027, 3.8645, 2.9833, 1.8754, 3.6951, 1.5052, 3.7268, 2.1037], + device='cuda:3'), covar=tensor([0.1462, 0.0149, 0.0865, 0.1849, 0.0229, 0.1961, 0.0204, 0.1606], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0104, 0.0116, 0.0111, 0.0104, 0.0118, 0.0101, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:03:24,006 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2022-11-16 07:03:33,554 INFO [train.py:876] (3/4) Epoch 13, batch 4100, loss[loss=0.1478, simple_loss=0.1618, pruned_loss=0.06685, over 5470.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1337, pruned_loss=0.03582, over 1096355.97 frames. ], batch size: 58, lr: 6.18e-03, grad_scale: 16.0 +2022-11-16 07:03:43,326 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8072, 2.0272, 2.5345, 2.5240, 2.4922, 1.9053, 2.5191, 2.7982], + device='cuda:3'), covar=tensor([0.0862, 0.1124, 0.0953, 0.1068, 0.1062, 0.1440, 0.0983, 0.0856], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0192, 0.0216, 0.0211, 0.0240, 0.0195, 0.0224, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:04:05,782 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.4773, 1.5910, 1.5652, 1.6741, 1.7247, 1.6896, 1.3656, 1.6725], + device='cuda:3'), covar=tensor([0.0082, 0.0070, 0.0058, 0.0061, 0.0056, 0.0059, 0.0069, 0.0077], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0060, 0.0059, 0.0064, 0.0062, 0.0057, 0.0056, 0.0054], + device='cuda:3'), out_proj_covar=tensor([5.7924e-05, 5.3114e-05, 5.1437e-05, 5.6777e-05, 5.4410e-05, 4.9847e-05, + 5.0095e-05, 4.6828e-05], device='cuda:3') +2022-11-16 07:04:21,228 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 +2022-11-16 07:04:23,285 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.983e+01 1.379e+02 1.736e+02 2.115e+02 4.817e+02, threshold=3.473e+02, percent-clipped=3.0 +2022-11-16 07:04:40,816 INFO [train.py:876] (3/4) Epoch 13, batch 4200, loss[loss=0.05559, simple_loss=0.08715, pruned_loss=0.01201, over 5690.00 frames. ], tot_loss[loss=0.1028, simple_loss=0.1338, pruned_loss=0.03593, over 1096256.91 frames. ], batch size: 11, lr: 6.18e-03, grad_scale: 16.0 +2022-11-16 07:04:41,710 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5527, 4.0434, 3.6993, 3.3912, 2.0476, 3.8792, 2.2345, 3.3720], + device='cuda:3'), covar=tensor([0.0545, 0.0171, 0.0267, 0.0422, 0.0708, 0.0211, 0.0592, 0.0210], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0184, 0.0185, 0.0209, 0.0196, 0.0185, 0.0194, 0.0188], + 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') +2022-11-16 07:04:43,688 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 +2022-11-16 07:04:52,138 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5100, 4.5193, 4.5625, 4.5235, 4.1260, 4.1372, 5.0181, 4.4284], + device='cuda:3'), covar=tensor([0.0432, 0.0985, 0.0411, 0.1266, 0.0545, 0.0379, 0.0731, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0107, 0.0095, 0.0122, 0.0089, 0.0081, 0.0146, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:04:52,893 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9992, 1.9411, 1.7347, 1.9575, 1.6969, 1.3793, 1.7363, 2.1685], + device='cuda:3'), covar=tensor([0.1750, 0.1905, 0.2161, 0.1472, 0.1957, 0.2699, 0.1855, 0.0999], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0111, 0.0108, 0.0109, 0.0096, 0.0107, 0.0100, 0.0085], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 07:05:07,944 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91505.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:05:19,084 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4747, 3.2501, 3.2825, 3.0918, 1.9069, 3.2236, 2.0859, 3.0240], + device='cuda:3'), covar=tensor([0.0382, 0.0206, 0.0187, 0.0301, 0.0546, 0.0190, 0.0541, 0.0163], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0183, 0.0184, 0.0209, 0.0196, 0.0184, 0.0193, 0.0187], + 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') +2022-11-16 07:05:31,428 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.444e+01 1.459e+02 1.798e+02 2.244e+02 4.763e+02, threshold=3.595e+02, percent-clipped=2.0 +2022-11-16 07:05:33,637 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91542.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:05:37,504 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6933, 4.0947, 3.6594, 4.0687, 4.0984, 3.4963, 3.6907, 3.6737], + device='cuda:3'), covar=tensor([0.0621, 0.0553, 0.1730, 0.0530, 0.0513, 0.0587, 0.0918, 0.0755], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0181, 0.0275, 0.0176, 0.0224, 0.0174, 0.0190, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:05:40,835 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91553.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:05:48,486 INFO [train.py:876] (3/4) Epoch 13, batch 4300, loss[loss=0.1276, simple_loss=0.1501, pruned_loss=0.05253, over 5571.00 frames. ], tot_loss[loss=0.1042, simple_loss=0.1347, pruned_loss=0.03684, over 1094098.40 frames. ], batch size: 30, lr: 6.18e-03, grad_scale: 16.0 +2022-11-16 07:06:15,240 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91603.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 07:06:20,028 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 +2022-11-16 07:06:34,417 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2022-11-16 07:06:39,669 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.763e+01 1.435e+02 1.703e+02 2.070e+02 3.900e+02, threshold=3.406e+02, percent-clipped=1.0 +2022-11-16 07:06:41,081 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91641.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:06:56,954 INFO [train.py:876] (3/4) Epoch 13, batch 4400, loss[loss=0.08701, simple_loss=0.1324, pruned_loss=0.02079, over 5703.00 frames. ], tot_loss[loss=0.104, simple_loss=0.135, pruned_loss=0.03652, over 1097237.38 frames. ], batch size: 28, lr: 6.17e-03, grad_scale: 16.0 +2022-11-16 07:07:00,150 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 +2022-11-16 07:07:22,334 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2022-11-16 07:07:46,910 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.810e+01 1.451e+02 1.858e+02 2.310e+02 4.864e+02, threshold=3.715e+02, percent-clipped=3.0 +2022-11-16 07:08:04,751 INFO [train.py:876] (3/4) Epoch 13, batch 4500, loss[loss=0.06163, simple_loss=0.09538, pruned_loss=0.01394, over 5316.00 frames. ], tot_loss[loss=0.1026, simple_loss=0.1341, pruned_loss=0.03554, over 1091518.86 frames. ], batch size: 9, lr: 6.17e-03, grad_scale: 16.0 +2022-11-16 07:08:12,270 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2022-11-16 07:08:30,589 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 07:08:55,631 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.055e+01 1.330e+02 1.643e+02 2.153e+02 4.136e+02, threshold=3.287e+02, percent-clipped=1.0 +2022-11-16 07:09:13,846 INFO [train.py:876] (3/4) Epoch 13, batch 4600, loss[loss=0.08201, simple_loss=0.1068, pruned_loss=0.02862, over 5346.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1326, pruned_loss=0.03427, over 1085856.26 frames. ], batch size: 9, lr: 6.17e-03, grad_scale: 16.0 +2022-11-16 07:09:31,317 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91891.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:09:34,277 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.48 vs. limit=5.0 +2022-11-16 07:09:35,816 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91898.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 07:10:03,761 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.956e+01 1.427e+02 1.793e+02 2.306e+02 3.919e+02, threshold=3.587e+02, percent-clipped=3.0 +2022-11-16 07:10:05,209 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91941.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:10:12,676 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91952.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:10:14,974 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91955.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:10:21,953 INFO [train.py:876] (3/4) Epoch 13, batch 4700, loss[loss=0.118, simple_loss=0.1419, pruned_loss=0.04706, over 4959.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1322, pruned_loss=0.03482, over 1082261.86 frames. ], batch size: 109, lr: 6.16e-03, grad_scale: 32.0 +2022-11-16 07:10:30,884 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2022-11-16 07:10:32,698 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91981.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:10:36,983 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3049, 2.3002, 2.8396, 1.9717, 1.6534, 3.1558, 2.6075, 2.3916], + device='cuda:3'), covar=tensor([0.1293, 0.1228, 0.0981, 0.2425, 0.2895, 0.0639, 0.1234, 0.1093], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0102, 0.0101, 0.0105, 0.0077, 0.0073, 0.0083, 0.0095], + 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') +2022-11-16 07:10:38,258 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91989.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:10:57,225 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92016.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:11:03,722 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6284, 3.9790, 3.2290, 3.7453, 3.8735, 3.6177, 3.9435, 3.7095], + device='cuda:3'), covar=tensor([0.0710, 0.0636, 0.2046, 0.1007, 0.0851, 0.0563, 0.0608, 0.0680], + device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0182, 0.0275, 0.0178, 0.0222, 0.0175, 0.0190, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:11:13,168 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.635e+01 1.436e+02 1.731e+02 2.246e+02 5.128e+02, threshold=3.463e+02, percent-clipped=1.0 +2022-11-16 07:11:14,675 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92042.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:11:25,214 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.68 vs. limit=5.0 +2022-11-16 07:11:29,851 INFO [train.py:876] (3/4) Epoch 13, batch 4800, loss[loss=0.159, simple_loss=0.16, pruned_loss=0.07898, over 4165.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1324, pruned_loss=0.03546, over 1072368.67 frames. ], batch size: 181, lr: 6.16e-03, grad_scale: 16.0 +2022-11-16 07:11:35,562 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.60 vs. limit=5.0 +2022-11-16 07:12:10,041 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 07:12:21,125 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.456e+02 1.773e+02 2.176e+02 4.110e+02, threshold=3.546e+02, percent-clipped=4.0 +2022-11-16 07:12:37,615 INFO [train.py:876] (3/4) Epoch 13, batch 4900, loss[loss=0.08932, simple_loss=0.129, pruned_loss=0.02482, over 5725.00 frames. ], tot_loss[loss=0.09997, simple_loss=0.1309, pruned_loss=0.03452, over 1074007.55 frames. ], batch size: 27, lr: 6.16e-03, grad_scale: 8.0 +2022-11-16 07:12:43,193 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2022-11-16 07:13:00,133 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92198.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:13:17,105 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7918, 2.4756, 2.7710, 3.8138, 3.7306, 2.7619, 2.5116, 3.7353], + device='cuda:3'), covar=tensor([0.0797, 0.2639, 0.2542, 0.2192, 0.1072, 0.3017, 0.2241, 0.0766], + device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0198, 0.0188, 0.0300, 0.0225, 0.0203, 0.0190, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:13:23,336 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0787, 4.0950, 4.2214, 3.7622, 4.0930, 3.8881, 1.6473, 4.0454], + device='cuda:3'), covar=tensor([0.0231, 0.0328, 0.0237, 0.0417, 0.0285, 0.0418, 0.3268, 0.0349], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0090, 0.0089, 0.0083, 0.0104, 0.0091, 0.0133, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:13:26,989 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4250, 2.5445, 2.6779, 2.3537, 2.5833, 2.5544, 1.1835, 2.6643], + device='cuda:3'), covar=tensor([0.0366, 0.0372, 0.0327, 0.0415, 0.0406, 0.0435, 0.3022, 0.0450], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0091, 0.0090, 0.0083, 0.0104, 0.0091, 0.0133, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:13:29,749 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.574e+01 1.436e+02 1.818e+02 2.548e+02 4.452e+02, threshold=3.637e+02, percent-clipped=5.0 +2022-11-16 07:13:33,145 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92246.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:13:33,778 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92247.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:13:45,762 INFO [train.py:876] (3/4) Epoch 13, batch 5000, loss[loss=0.08754, simple_loss=0.1272, pruned_loss=0.02394, over 5718.00 frames. ], tot_loss[loss=0.1011, simple_loss=0.1322, pruned_loss=0.03498, over 1078978.36 frames. ], batch size: 17, lr: 6.15e-03, grad_scale: 8.0 +2022-11-16 07:13:54,897 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.3358, 5.5137, 5.8244, 5.6775, 5.4825, 4.8015, 6.2405, 5.5006], + device='cuda:3'), covar=tensor([0.0502, 0.0777, 0.0224, 0.1076, 0.0343, 0.0344, 0.0508, 0.0588], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0107, 0.0094, 0.0122, 0.0089, 0.0080, 0.0145, 0.0103], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:14:07,056 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8230, 1.8662, 1.7331, 1.6535, 1.7415, 1.5667, 1.6646, 1.7277], + device='cuda:3'), covar=tensor([0.0065, 0.0068, 0.0059, 0.0063, 0.0053, 0.0046, 0.0053, 0.0064], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0061, 0.0060, 0.0066, 0.0063, 0.0059, 0.0057, 0.0055], + device='cuda:3'), out_proj_covar=tensor([5.8765e-05, 5.3589e-05, 5.2755e-05, 5.8000e-05, 5.5419e-05, 5.1000e-05, + 5.0571e-05, 4.8138e-05], device='cuda:3') +2022-11-16 07:14:16,846 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92311.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:14:34,566 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92337.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:14:37,061 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.814e+01 1.442e+02 1.699e+02 2.118e+02 7.275e+02, threshold=3.398e+02, percent-clipped=6.0 +2022-11-16 07:14:51,616 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6607, 4.3644, 3.8373, 3.5718, 1.8701, 4.1125, 2.3527, 3.5555], + device='cuda:3'), covar=tensor([0.0536, 0.0164, 0.0205, 0.0401, 0.0817, 0.0197, 0.0672, 0.0197], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0183, 0.0184, 0.0208, 0.0196, 0.0184, 0.0193, 0.0187], + 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') +2022-11-16 07:14:54,056 INFO [train.py:876] (3/4) Epoch 13, batch 5100, loss[loss=0.1175, simple_loss=0.1386, pruned_loss=0.04816, over 5445.00 frames. ], tot_loss[loss=0.1008, simple_loss=0.1323, pruned_loss=0.03461, over 1079529.07 frames. ], batch size: 58, lr: 6.15e-03, grad_scale: 8.0 +2022-11-16 07:15:12,150 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2022-11-16 07:15:45,934 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.294e+01 1.423e+02 1.797e+02 2.276e+02 4.290e+02, threshold=3.595e+02, percent-clipped=2.0 +2022-11-16 07:16:02,830 INFO [train.py:876] (3/4) Epoch 13, batch 5200, loss[loss=0.1083, simple_loss=0.1404, pruned_loss=0.03812, over 5755.00 frames. ], tot_loss[loss=0.1028, simple_loss=0.1339, pruned_loss=0.0358, over 1083111.63 frames. ], batch size: 20, lr: 6.15e-03, grad_scale: 8.0 +2022-11-16 07:16:14,243 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9092, 4.7040, 5.0164, 4.9521, 4.7583, 4.5709, 5.5412, 4.9948], + device='cuda:3'), covar=tensor([0.0371, 0.0862, 0.0387, 0.0970, 0.0396, 0.0290, 0.0576, 0.0512], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0110, 0.0097, 0.0124, 0.0090, 0.0081, 0.0148, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:16:53,187 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1086, 2.8876, 3.2109, 1.6985, 3.0900, 3.4328, 3.1350, 3.6272], + device='cuda:3'), covar=tensor([0.2365, 0.2010, 0.0741, 0.3586, 0.0748, 0.1250, 0.0941, 0.1003], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0184, 0.0170, 0.0187, 0.0185, 0.0206, 0.0171, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:16:53,846 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8692, 2.4450, 2.4979, 1.4070, 2.7373, 2.8323, 2.5400, 2.8461], + device='cuda:3'), covar=tensor([0.2229, 0.2066, 0.1288, 0.3484, 0.0972, 0.1128, 0.0681, 0.1281], + device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0184, 0.0170, 0.0187, 0.0185, 0.0206, 0.0171, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:16:54,292 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.360e+01 1.391e+02 1.845e+02 2.323e+02 4.876e+02, threshold=3.690e+02, percent-clipped=4.0 +2022-11-16 07:16:58,345 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92547.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:17:07,557 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2022-11-16 07:17:10,185 INFO [train.py:876] (3/4) Epoch 13, batch 5300, loss[loss=0.1203, simple_loss=0.1492, pruned_loss=0.04564, over 5644.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1346, pruned_loss=0.03621, over 1085370.50 frames. ], batch size: 29, lr: 6.14e-03, grad_scale: 8.0 +2022-11-16 07:17:30,908 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92595.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:17:39,749 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.1495, 2.6587, 3.4189, 2.2206, 1.6953, 3.6745, 2.8110, 2.5151], + device='cuda:3'), covar=tensor([0.0646, 0.1185, 0.0517, 0.2275, 0.3676, 0.0874, 0.1618, 0.1308], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0102, 0.0101, 0.0104, 0.0078, 0.0072, 0.0082, 0.0095], + 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') +2022-11-16 07:17:41,730 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92611.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:17:42,373 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1204, 2.3193, 2.3994, 2.0588, 2.3234, 2.2566, 1.1582, 2.4456], + device='cuda:3'), covar=tensor([0.0391, 0.0368, 0.0342, 0.0442, 0.0439, 0.0515, 0.2803, 0.0432], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0091, 0.0090, 0.0082, 0.0104, 0.0091, 0.0134, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:17:42,475 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8075, 1.7241, 1.8639, 1.9028, 2.2413, 1.7821, 1.4753, 1.8250], + device='cuda:3'), covar=tensor([0.1911, 0.1444, 0.1673, 0.1002, 0.0957, 0.2110, 0.2298, 0.2703], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0198, 0.0188, 0.0301, 0.0227, 0.0204, 0.0190, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:17:51,963 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92625.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:17:59,650 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92637.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:18:02,110 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.870e+01 1.449e+02 1.761e+02 2.157e+02 4.829e+02, threshold=3.522e+02, percent-clipped=3.0 +2022-11-16 07:18:04,561 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 07:18:14,401 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92659.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:18:18,357 INFO [train.py:876] (3/4) Epoch 13, batch 5400, loss[loss=0.09864, simple_loss=0.1402, pruned_loss=0.02851, over 5758.00 frames. ], tot_loss[loss=0.1033, simple_loss=0.1344, pruned_loss=0.03609, over 1086057.02 frames. ], batch size: 21, lr: 6.14e-03, grad_scale: 8.0 +2022-11-16 07:18:32,618 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92685.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:18:33,354 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92686.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:18:42,765 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4872, 2.9944, 3.9896, 2.3865, 2.3027, 4.0654, 3.3649, 2.7755], + device='cuda:3'), covar=tensor([0.0601, 0.1135, 0.0296, 0.2306, 0.1428, 0.0597, 0.0616, 0.1055], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0100, 0.0099, 0.0102, 0.0075, 0.0071, 0.0079, 0.0092], + 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') +2022-11-16 07:18:54,620 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1658, 2.0552, 1.8923, 1.8576, 2.0067, 2.2690, 2.0848, 1.7283], + device='cuda:3'), covar=tensor([0.0051, 0.0055, 0.0059, 0.0045, 0.0069, 0.0058, 0.0043, 0.0054], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0028, 0.0028, 0.0037, 0.0032, 0.0029, 0.0036, 0.0035], + device='cuda:3'), out_proj_covar=tensor([2.8088e-05, 2.5845e-05, 2.5632e-05, 3.5627e-05, 2.9467e-05, 2.7523e-05, + 3.4352e-05, 3.3179e-05], device='cuda:3') +2022-11-16 07:18:55,991 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92720.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:18:59,276 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1450, 3.4455, 2.5141, 3.1907, 2.5711, 2.5570, 1.8177, 2.8671], + device='cuda:3'), covar=tensor([0.1331, 0.0338, 0.1098, 0.0532, 0.1376, 0.1120, 0.2117, 0.0663], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0144, 0.0157, 0.0150, 0.0176, 0.0168, 0.0160, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:19:03,229 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7547, 4.2262, 4.5597, 4.3276, 4.8432, 4.6162, 4.2602, 4.7860], + device='cuda:3'), covar=tensor([0.0386, 0.0391, 0.0443, 0.0313, 0.0343, 0.0274, 0.0330, 0.0299], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0158, 0.0113, 0.0147, 0.0186, 0.0115, 0.0132, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 07:19:10,733 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.348e+01 1.471e+02 1.783e+02 2.230e+02 3.628e+02, threshold=3.566e+02, percent-clipped=2.0 +2022-11-16 07:19:26,810 INFO [train.py:876] (3/4) Epoch 13, batch 5500, loss[loss=0.1611, simple_loss=0.1666, pruned_loss=0.07778, over 5467.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1329, pruned_loss=0.03478, over 1091067.21 frames. ], batch size: 64, lr: 6.14e-03, grad_scale: 8.0 +2022-11-16 07:19:27,008 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3969, 2.3913, 2.5321, 3.5312, 3.4909, 2.6212, 2.2628, 3.5569], + device='cuda:3'), covar=tensor([0.1053, 0.2556, 0.2208, 0.2605, 0.1068, 0.2787, 0.2109, 0.0816], + device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0201, 0.0193, 0.0307, 0.0232, 0.0209, 0.0194, 0.0257], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:19:38,277 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92781.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:19:48,209 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2022-11-16 07:20:04,799 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7365, 4.9935, 3.7872, 2.1932, 4.6581, 2.1514, 4.7766, 2.8261], + device='cuda:3'), covar=tensor([0.1389, 0.0081, 0.0467, 0.1938, 0.0133, 0.1542, 0.0112, 0.1337], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0105, 0.0115, 0.0112, 0.0103, 0.0119, 0.0101, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:20:05,570 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2022-11-16 07:20:12,844 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92831.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:20:19,501 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.251e+01 1.445e+02 1.860e+02 2.400e+02 4.307e+02, threshold=3.721e+02, percent-clipped=4.0 +2022-11-16 07:20:32,210 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2022-11-16 07:20:35,691 INFO [train.py:876] (3/4) Epoch 13, batch 5600, loss[loss=0.05777, simple_loss=0.09204, pruned_loss=0.01175, over 4856.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.133, pruned_loss=0.03471, over 1088227.34 frames. ], batch size: 5, lr: 6.13e-03, grad_scale: 8.0 +2022-11-16 07:20:54,525 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:20:59,082 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92899.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:21:20,278 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92931.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:21:27,403 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.557e+01 1.397e+02 1.621e+02 2.063e+02 4.647e+02, threshold=3.241e+02, percent-clipped=3.0 +2022-11-16 07:21:40,275 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92960.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:21:43,359 INFO [train.py:876] (3/4) Epoch 13, batch 5700, loss[loss=0.09473, simple_loss=0.1381, pruned_loss=0.02567, over 5492.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1336, pruned_loss=0.0349, over 1088826.67 frames. ], batch size: 17, lr: 6.13e-03, grad_scale: 8.0 +2022-11-16 07:21:50,432 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92975.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:21:54,338 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92981.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:01,406 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2022-11-16 07:22:02,215 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2594, 3.8206, 3.5098, 3.8250, 3.8784, 3.3762, 3.4809, 3.4115], + device='cuda:3'), covar=tensor([0.1222, 0.0526, 0.1414, 0.0545, 0.0533, 0.0524, 0.0908, 0.0741], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0180, 0.0275, 0.0179, 0.0223, 0.0175, 0.0191, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:22:02,290 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92992.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:14,758 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93010.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:32,135 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:32,159 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:36,329 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.192e+01 1.368e+02 1.655e+02 2.036e+02 3.681e+02, threshold=3.311e+02, percent-clipped=3.0 +2022-11-16 07:22:52,166 INFO [train.py:876] (3/4) Epoch 13, batch 5800, loss[loss=0.1027, simple_loss=0.1378, pruned_loss=0.03377, over 5739.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.1335, pruned_loss=0.0354, over 1079470.93 frames. ], batch size: 15, lr: 6.13e-03, grad_scale: 8.0 +2022-11-16 07:22:53,806 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 +2022-11-16 07:22:56,537 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93071.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:22:59,772 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93076.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:23:07,718 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4050, 2.3238, 2.1778, 2.3726, 2.1517, 1.7351, 2.3209, 2.6754], + device='cuda:3'), covar=tensor([0.1482, 0.1942, 0.2037, 0.1394, 0.1716, 0.2255, 0.1479, 0.1554], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0109, 0.0108, 0.0108, 0.0094, 0.0105, 0.0098, 0.0085], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 07:23:13,898 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93097.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:23:26,248 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93115.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:23:33,557 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3019, 1.6790, 1.2581, 1.1845, 1.4664, 1.2004, 1.0888, 1.5394], + device='cuda:3'), covar=tensor([0.0073, 0.0051, 0.0081, 0.0084, 0.0058, 0.0062, 0.0101, 0.0057], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0060, 0.0060, 0.0066, 0.0063, 0.0059, 0.0056, 0.0055], + device='cuda:3'), out_proj_covar=tensor([5.9149e-05, 5.3464e-05, 5.2081e-05, 5.8005e-05, 5.5301e-05, 5.1254e-05, + 4.9999e-05, 4.7907e-05], device='cuda:3') +2022-11-16 07:23:43,301 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.091e+01 1.350e+02 1.786e+02 2.182e+02 6.733e+02, threshold=3.572e+02, percent-clipped=3.0 +2022-11-16 07:24:00,085 INFO [train.py:876] (3/4) Epoch 13, batch 5900, loss[loss=0.04508, simple_loss=0.07614, pruned_loss=0.007005, over 5095.00 frames. ], tot_loss[loss=0.1029, simple_loss=0.1337, pruned_loss=0.03607, over 1086016.26 frames. ], batch size: 7, lr: 6.12e-03, grad_scale: 8.0 +2022-11-16 07:24:07,376 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93176.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:24:14,471 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93187.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:24:37,574 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5239, 1.8498, 1.5575, 1.2798, 1.6904, 1.9313, 2.0362, 1.9980], + device='cuda:3'), covar=tensor([0.1923, 0.1396, 0.2196, 0.2734, 0.1351, 0.1281, 0.0768, 0.1407], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0180, 0.0167, 0.0183, 0.0181, 0.0201, 0.0168, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:24:51,116 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.010e+02 1.444e+02 1.638e+02 1.961e+02 3.238e+02, threshold=3.277e+02, percent-clipped=0.0 +2022-11-16 07:25:00,829 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93255.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:25:07,941 INFO [train.py:876] (3/4) Epoch 13, batch 6000, loss[loss=0.09769, simple_loss=0.1357, pruned_loss=0.02984, over 5662.00 frames. ], tot_loss[loss=0.1011, simple_loss=0.1327, pruned_loss=0.03476, over 1087790.03 frames. ], batch size: 29, lr: 6.12e-03, grad_scale: 8.0 +2022-11-16 07:25:07,941 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 07:25:31,733 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7834, 2.5192, 3.4256, 3.3210, 3.3232, 2.5750, 3.2324, 3.6901], + device='cuda:3'), covar=tensor([0.0716, 0.1295, 0.0853, 0.1058, 0.0761, 0.1362, 0.1176, 0.0762], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0191, 0.0217, 0.0210, 0.0240, 0.0195, 0.0225, 0.0232], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:25:32,847 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9860, 3.6604, 4.0989, 3.5062, 3.5261, 3.9867, 4.2130, 3.9046], + device='cuda:3'), covar=tensor([0.0291, 0.0970, 0.0287, 0.1125, 0.0454, 0.0191, 0.0614, 0.0463], + device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0108, 0.0096, 0.0122, 0.0089, 0.0080, 0.0146, 0.0104], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:25:35,118 INFO [train.py:908] (3/4) Epoch 13, validation: loss=0.1768, simple_loss=0.1872, pruned_loss=0.08323, over 1530663.00 frames. +2022-11-16 07:25:35,119 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 07:25:40,547 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2022-11-16 07:25:45,589 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93281.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:25:49,438 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93287.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:25:52,090 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8577, 4.7137, 3.1341, 4.5092, 3.6462, 3.2902, 2.6731, 3.9595], + device='cuda:3'), covar=tensor([0.1262, 0.0192, 0.0980, 0.0285, 0.0654, 0.0846, 0.1710, 0.0381], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0140, 0.0153, 0.0145, 0.0172, 0.0164, 0.0155, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:26:18,408 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93329.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:26:19,781 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93331.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:26:26,166 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.791e+01 1.381e+02 1.698e+02 2.176e+02 5.974e+02, threshold=3.396e+02, percent-clipped=7.0 +2022-11-16 07:26:32,012 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 +2022-11-16 07:26:42,555 INFO [train.py:876] (3/4) Epoch 13, batch 6100, loss[loss=0.06798, simple_loss=0.1066, pruned_loss=0.0147, over 5695.00 frames. ], tot_loss[loss=0.0996, simple_loss=0.1314, pruned_loss=0.0339, over 1086650.54 frames. ], batch size: 12, lr: 6.12e-03, grad_scale: 8.0 +2022-11-16 07:26:43,309 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93366.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:26:46,740 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3751, 1.6288, 1.3201, 1.4440, 1.5092, 1.4886, 1.3341, 1.5118], + device='cuda:3'), covar=tensor([0.0095, 0.0058, 0.0075, 0.0074, 0.0069, 0.0063, 0.0075, 0.0072], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0060, 0.0059, 0.0066, 0.0062, 0.0059, 0.0056, 0.0055], + device='cuda:3'), out_proj_covar=tensor([5.9067e-05, 5.3038e-05, 5.1866e-05, 5.8166e-05, 5.5099e-05, 5.1104e-05, + 4.9544e-05, 4.7777e-05], device='cuda:3') +2022-11-16 07:26:50,789 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:27:01,545 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93392.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 07:27:23,295 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93424.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:27:35,453 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.736e+01 1.359e+02 1.712e+02 2.042e+02 3.975e+02, threshold=3.424e+02, percent-clipped=1.0 +2022-11-16 07:27:51,738 INFO [train.py:876] (3/4) Epoch 13, batch 6200, loss[loss=0.0994, simple_loss=0.1298, pruned_loss=0.03448, over 5456.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1324, pruned_loss=0.03452, over 1083848.19 frames. ], batch size: 11, lr: 6.12e-03, grad_scale: 8.0 +2022-11-16 07:27:53,798 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9088, 2.3622, 2.9366, 3.7694, 3.8192, 2.9941, 2.7508, 3.7644], + device='cuda:3'), covar=tensor([0.0740, 0.3115, 0.2263, 0.2243, 0.1051, 0.2708, 0.2125, 0.1185], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0197, 0.0187, 0.0299, 0.0228, 0.0202, 0.0190, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:27:55,648 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93471.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:28:06,764 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:28:08,154 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7164, 3.7359, 3.6051, 3.3809, 2.1043, 3.7060, 2.3957, 3.1705], + device='cuda:3'), covar=tensor([0.0425, 0.0206, 0.0206, 0.0341, 0.0628, 0.0202, 0.0557, 0.0205], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0183, 0.0180, 0.0208, 0.0194, 0.0184, 0.0193, 0.0186], + 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') +2022-11-16 07:28:38,381 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93535.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:28:42,534 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.266e+01 1.465e+02 1.713e+02 2.165e+02 4.081e+02, threshold=3.427e+02, percent-clipped=1.0 +2022-11-16 07:28:52,491 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93555.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:28:58,917 INFO [train.py:876] (3/4) Epoch 13, batch 6300, loss[loss=0.113, simple_loss=0.1414, pruned_loss=0.04235, over 5630.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1334, pruned_loss=0.03477, over 1085002.29 frames. ], batch size: 29, lr: 6.11e-03, grad_scale: 8.0 +2022-11-16 07:29:08,364 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7685, 4.5517, 4.9028, 4.9148, 4.6035, 4.3389, 5.4444, 4.9185], + device='cuda:3'), covar=tensor([0.0448, 0.1004, 0.0398, 0.1114, 0.0329, 0.0282, 0.0488, 0.0516], + device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0110, 0.0096, 0.0124, 0.0090, 0.0081, 0.0148, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:29:12,389 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8681, 2.6346, 3.2157, 2.0745, 1.5964, 3.1249, 2.8851, 2.3855], + device='cuda:3'), covar=tensor([0.0792, 0.0833, 0.0519, 0.2183, 0.2661, 0.2248, 0.0730, 0.1252], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0102, 0.0102, 0.0105, 0.0076, 0.0072, 0.0083, 0.0095], + 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') +2022-11-16 07:29:13,679 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93587.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:29:24,621 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93603.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:29:42,077 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6918, 3.7111, 3.8237, 2.1508, 3.8133, 3.9689, 3.9133, 4.6008], + device='cuda:3'), covar=tensor([0.1705, 0.1236, 0.0807, 0.2541, 0.0350, 0.1233, 0.0408, 0.0430], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0178, 0.0166, 0.0182, 0.0179, 0.0202, 0.0167, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:29:43,332 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93631.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:29:46,181 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93635.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:29:49,998 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.661e+01 1.430e+02 1.750e+02 2.354e+02 5.950e+02, threshold=3.500e+02, percent-clipped=2.0 +2022-11-16 07:30:07,004 INFO [train.py:876] (3/4) Epoch 13, batch 6400, loss[loss=0.1064, simple_loss=0.1347, pruned_loss=0.03911, over 5423.00 frames. ], tot_loss[loss=0.1001, simple_loss=0.1322, pruned_loss=0.03397, over 1088839.93 frames. ], batch size: 53, lr: 6.11e-03, grad_scale: 8.0 +2022-11-16 07:30:07,764 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93666.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:30:16,123 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93679.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:30:24,955 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93692.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:30:39,897 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93714.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:30:57,385 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93740.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:30:57,984 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.087e+01 1.406e+02 1.692e+02 2.048e+02 4.065e+02, threshold=3.385e+02, percent-clipped=2.0 +2022-11-16 07:31:13,824 INFO [train.py:876] (3/4) Epoch 13, batch 6500, loss[loss=0.09591, simple_loss=0.1343, pruned_loss=0.02875, over 5689.00 frames. ], tot_loss[loss=0.1005, simple_loss=0.1322, pruned_loss=0.03434, over 1088968.10 frames. ], batch size: 17, lr: 6.11e-03, grad_scale: 8.0 +2022-11-16 07:31:17,581 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93770.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:31:18,256 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93771.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:31:50,713 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93819.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:31:59,067 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93831.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:32:05,378 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.939e+01 1.358e+02 1.733e+02 2.139e+02 3.711e+02, threshold=3.467e+02, percent-clipped=1.0 +2022-11-16 07:32:21,362 INFO [train.py:876] (3/4) Epoch 13, batch 6600, loss[loss=0.0895, simple_loss=0.1311, pruned_loss=0.02397, over 5708.00 frames. ], tot_loss[loss=0.1019, simple_loss=0.1331, pruned_loss=0.0353, over 1092325.88 frames. ], batch size: 17, lr: 6.10e-03, grad_scale: 8.0 +2022-11-16 07:33:00,609 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2022-11-16 07:33:13,069 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.396e+02 1.723e+02 2.017e+02 3.846e+02, threshold=3.447e+02, percent-clipped=2.0 +2022-11-16 07:33:29,293 INFO [train.py:876] (3/4) Epoch 13, batch 6700, loss[loss=0.1844, simple_loss=0.1721, pruned_loss=0.09832, over 3137.00 frames. ], tot_loss[loss=0.103, simple_loss=0.134, pruned_loss=0.03599, over 1090111.21 frames. ], batch size: 284, lr: 6.10e-03, grad_scale: 8.0 +2022-11-16 07:33:40,304 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93981.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:34:23,829 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.687e+01 1.406e+02 1.734e+02 2.317e+02 5.958e+02, threshold=3.468e+02, percent-clipped=4.0 +2022-11-16 07:34:25,508 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94042.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:34:35,134 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.6352, 1.0212, 1.0322, 0.9088, 1.0100, 1.3038, 1.1182, 1.2229], + device='cuda:3'), covar=tensor([0.4341, 0.0966, 0.3994, 0.2978, 0.1846, 0.0830, 0.2722, 0.1659], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0102, 0.0101, 0.0105, 0.0076, 0.0072, 0.0084, 0.0093], + 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') +2022-11-16 07:34:40,853 INFO [train.py:876] (3/4) Epoch 13, batch 6800, loss[loss=0.05995, simple_loss=0.09394, pruned_loss=0.01298, over 5175.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1342, pruned_loss=0.03608, over 1087059.77 frames. ], batch size: 8, lr: 6.10e-03, grad_scale: 8.0 +2022-11-16 07:35:22,168 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94126.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:35:30,216 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2022-11-16 07:35:32,243 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.571e+01 1.399e+02 1.659e+02 1.999e+02 4.068e+02, threshold=3.319e+02, percent-clipped=3.0 +2022-11-16 07:35:36,029 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2022-11-16 07:35:48,960 INFO [train.py:876] (3/4) Epoch 13, batch 6900, loss[loss=0.09761, simple_loss=0.1387, pruned_loss=0.02827, over 5512.00 frames. ], tot_loss[loss=0.1024, simple_loss=0.1337, pruned_loss=0.03552, over 1091674.25 frames. ], batch size: 17, lr: 6.09e-03, grad_scale: 16.0 +2022-11-16 07:36:28,745 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94223.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:36:37,414 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2504, 1.0330, 1.1384, 0.9823, 1.2124, 1.1971, 0.6030, 0.9897], + device='cuda:3'), covar=tensor([0.0244, 0.0375, 0.0324, 0.0638, 0.0332, 0.0223, 0.0774, 0.0319], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0025, 0.0018, 0.0021, 0.0018, 0.0016, 0.0024, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.0844e-05, 1.2706e-04, 9.7785e-05, 1.1040e-04, 9.7599e-05, 9.1840e-05, + 1.2063e-04, 9.2648e-05], device='cuda:3') +2022-11-16 07:36:40,443 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.408e+01 1.493e+02 1.868e+02 2.271e+02 3.997e+02, threshold=3.737e+02, percent-clipped=4.0 +2022-11-16 07:36:56,959 INFO [train.py:876] (3/4) Epoch 13, batch 7000, loss[loss=0.1049, simple_loss=0.1201, pruned_loss=0.04481, over 5123.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1331, pruned_loss=0.03524, over 1085625.81 frames. ], batch size: 91, lr: 6.09e-03, grad_scale: 16.0 +2022-11-16 07:37:09,859 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94284.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:37:18,927 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.6210, 4.4942, 4.5814, 4.5467, 4.1592, 4.1278, 5.0082, 4.5362], + device='cuda:3'), covar=tensor([0.0334, 0.0716, 0.0394, 0.1131, 0.0486, 0.0304, 0.0589, 0.0566], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0111, 0.0098, 0.0125, 0.0091, 0.0082, 0.0148, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:37:35,874 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8883, 2.3237, 2.9660, 3.6182, 3.8355, 2.7888, 2.6674, 3.6739], + device='cuda:3'), covar=tensor([0.0785, 0.2792, 0.2221, 0.3378, 0.1001, 0.3285, 0.2080, 0.1114], + device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0192, 0.0183, 0.0296, 0.0223, 0.0199, 0.0187, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:37:45,796 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94337.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:37:48,222 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.793e+01 1.390e+02 1.725e+02 2.069e+02 3.894e+02, threshold=3.450e+02, percent-clipped=1.0 +2022-11-16 07:37:52,963 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94348.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:37:57,080 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2022-11-16 07:38:04,148 INFO [train.py:876] (3/4) Epoch 13, batch 7100, loss[loss=0.06487, simple_loss=0.1031, pruned_loss=0.01332, over 5401.00 frames. ], tot_loss[loss=0.1022, simple_loss=0.1335, pruned_loss=0.03551, over 1086466.21 frames. ], batch size: 11, lr: 6.09e-03, grad_scale: 16.0 +2022-11-16 07:38:28,801 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9729, 4.0319, 3.8731, 3.6817, 4.0293, 3.7450, 1.5355, 4.1628], + device='cuda:3'), covar=tensor([0.0244, 0.0283, 0.0374, 0.0303, 0.0277, 0.0512, 0.3067, 0.0331], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0087, 0.0082, 0.0103, 0.0090, 0.0131, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:38:29,483 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3645, 3.8555, 3.0497, 1.8805, 3.7192, 1.5810, 3.5320, 2.0852], + device='cuda:3'), covar=tensor([0.1535, 0.0155, 0.0657, 0.1967, 0.0213, 0.1824, 0.0286, 0.1411], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0104, 0.0114, 0.0110, 0.0102, 0.0117, 0.0100, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:38:34,089 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94409.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 07:38:45,552 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94426.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:38:56,308 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.683e+01 1.440e+02 1.714e+02 2.140e+02 3.966e+02, threshold=3.428e+02, percent-clipped=2.0 +2022-11-16 07:39:04,313 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9405, 2.5509, 2.8618, 1.5551, 2.9150, 3.1505, 2.9444, 3.3878], + device='cuda:3'), covar=tensor([0.1914, 0.1883, 0.1221, 0.3055, 0.0718, 0.1097, 0.0785, 0.0746], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0177, 0.0167, 0.0181, 0.0181, 0.0201, 0.0168, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:39:04,836 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5487, 4.4811, 4.6796, 4.4932, 4.4690, 4.0465, 5.0751, 4.5014], + device='cuda:3'), covar=tensor([0.0438, 0.1024, 0.0366, 0.1383, 0.0343, 0.0347, 0.0721, 0.0632], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0112, 0.0097, 0.0126, 0.0091, 0.0083, 0.0149, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:39:12,056 INFO [train.py:876] (3/4) Epoch 13, batch 7200, loss[loss=0.1136, simple_loss=0.1475, pruned_loss=0.03985, over 5576.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1344, pruned_loss=0.0362, over 1077386.68 frames. ], batch size: 23, lr: 6.08e-03, grad_scale: 16.0 +2022-11-16 07:39:16,638 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2022-11-16 07:39:18,294 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94474.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:40:43,967 INFO [train.py:876] (3/4) Epoch 14, batch 0, loss[loss=0.07185, simple_loss=0.11, pruned_loss=0.01687, over 5419.00 frames. ], tot_loss[loss=0.07185, simple_loss=0.11, pruned_loss=0.01687, over 5419.00 frames. ], batch size: 11, lr: 5.86e-03, grad_scale: 16.0 +2022-11-16 07:40:43,967 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 07:41:00,499 INFO [train.py:908] (3/4) Epoch 14, validation: loss=0.1755, simple_loss=0.1868, pruned_loss=0.08205, over 1530663.00 frames. +2022-11-16 07:41:00,499 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 07:41:03,038 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.829e+01 1.398e+02 1.682e+02 2.138e+02 4.621e+02, threshold=3.364e+02, percent-clipped=3.0 +2022-11-16 07:41:18,692 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.27 vs. limit=5.0 +2022-11-16 07:41:29,028 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94579.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 07:41:42,892 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2155, 3.4992, 2.6650, 1.7803, 3.3469, 1.3635, 3.3995, 1.8089], + device='cuda:3'), covar=tensor([0.1459, 0.0238, 0.0909, 0.1846, 0.0259, 0.2076, 0.0257, 0.1541], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0105, 0.0115, 0.0111, 0.0103, 0.0119, 0.0100, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:41:58,488 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8307, 4.6246, 4.9901, 4.7240, 4.4308, 4.3704, 5.2631, 4.7219], + device='cuda:3'), covar=tensor([0.0371, 0.1065, 0.0355, 0.1743, 0.0345, 0.0273, 0.0722, 0.0618], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0110, 0.0096, 0.0124, 0.0090, 0.0082, 0.0148, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:42:05,898 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2022-11-16 07:42:08,084 INFO [train.py:876] (3/4) Epoch 14, batch 100, loss[loss=0.08654, simple_loss=0.1238, pruned_loss=0.02463, over 5569.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1338, pruned_loss=0.03679, over 426365.49 frames. ], batch size: 13, lr: 5.86e-03, grad_scale: 16.0 +2022-11-16 07:42:08,194 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94637.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:42:10,675 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.173e+01 1.465e+02 1.762e+02 2.317e+02 5.551e+02, threshold=3.525e+02, percent-clipped=6.0 +2022-11-16 07:42:16,100 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8195, 2.6050, 2.8446, 1.3344, 2.8569, 3.1131, 3.0748, 3.2431], + device='cuda:3'), covar=tensor([0.1966, 0.1598, 0.1012, 0.3026, 0.0652, 0.1298, 0.0617, 0.0851], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0180, 0.0168, 0.0183, 0.0183, 0.0203, 0.0170, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:42:18,090 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.4129, 0.8473, 0.7766, 0.6759, 0.9468, 0.6890, 0.6026, 0.8484], + device='cuda:3'), covar=tensor([0.0104, 0.0043, 0.0074, 0.0055, 0.0056, 0.0072, 0.0098, 0.0045], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0059, 0.0059, 0.0064, 0.0061, 0.0057, 0.0056, 0.0053], + device='cuda:3'), out_proj_covar=tensor([5.7595e-05, 5.2403e-05, 5.1673e-05, 5.6852e-05, 5.3934e-05, 5.0057e-05, + 4.9146e-05, 4.6746e-05], device='cuda:3') +2022-11-16 07:42:32,943 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7998, 3.1736, 2.1005, 2.9093, 2.3180, 2.2288, 1.7661, 2.5952], + device='cuda:3'), covar=tensor([0.1959, 0.0444, 0.1764, 0.0655, 0.1690, 0.1714, 0.2620, 0.0770], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0142, 0.0157, 0.0149, 0.0174, 0.0168, 0.0158, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:42:40,596 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94685.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:42:50,327 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2189, 3.4132, 3.3448, 3.1881, 3.4468, 3.2011, 1.4108, 3.5037], + device='cuda:3'), covar=tensor([0.0316, 0.0301, 0.0338, 0.0295, 0.0319, 0.0356, 0.3083, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0088, 0.0086, 0.0082, 0.0101, 0.0089, 0.0129, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:42:52,949 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94704.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 07:43:06,817 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5292, 2.2573, 2.7214, 1.7604, 2.0811, 3.0692, 2.6580, 2.2993], + device='cuda:3'), covar=tensor([0.0875, 0.1200, 0.0835, 0.2801, 0.1409, 0.2386, 0.0812, 0.1181], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0102, 0.0102, 0.0104, 0.0076, 0.0072, 0.0083, 0.0094], + 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') +2022-11-16 07:43:15,765 INFO [train.py:876] (3/4) Epoch 14, batch 200, loss[loss=0.08868, simple_loss=0.1227, pruned_loss=0.02733, over 5755.00 frames. ], tot_loss[loss=0.1001, simple_loss=0.1318, pruned_loss=0.03418, over 689110.61 frames. ], batch size: 20, lr: 5.85e-03, grad_scale: 16.0 +2022-11-16 07:43:18,294 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.376e+02 1.680e+02 2.123e+02 3.782e+02, threshold=3.359e+02, percent-clipped=1.0 +2022-11-16 07:43:25,681 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3737, 1.5183, 1.5917, 1.2914, 1.3397, 1.2882, 1.1750, 0.8373], + device='cuda:3'), covar=tensor([0.0038, 0.0045, 0.0030, 0.0061, 0.0061, 0.0063, 0.0055, 0.0086], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0028, 0.0029, 0.0037, 0.0032, 0.0029, 0.0036, 0.0034], + device='cuda:3'), out_proj_covar=tensor([2.8523e-05, 2.6640e-05, 2.5550e-05, 3.5752e-05, 3.0156e-05, 2.7887e-05, + 3.4880e-05, 3.2585e-05], device='cuda:3') +2022-11-16 07:44:21,157 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0509, 2.1818, 2.5820, 2.2734, 1.3679, 2.3090, 1.6434, 1.8289], + device='cuda:3'), covar=tensor([0.0339, 0.0191, 0.0169, 0.0246, 0.0469, 0.0220, 0.0453, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0184, 0.0182, 0.0208, 0.0196, 0.0185, 0.0194, 0.0187], + 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') +2022-11-16 07:44:22,890 INFO [train.py:876] (3/4) Epoch 14, batch 300, loss[loss=0.132, simple_loss=0.1558, pruned_loss=0.05406, over 5465.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1324, pruned_loss=0.03564, over 843836.30 frames. ], batch size: 53, lr: 5.85e-03, grad_scale: 16.0 +2022-11-16 07:44:25,438 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.663e+01 1.544e+02 1.893e+02 2.592e+02 6.103e+02, threshold=3.786e+02, percent-clipped=6.0 +2022-11-16 07:44:34,433 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 +2022-11-16 07:44:50,716 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94879.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:45:05,214 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6160, 2.2039, 3.2294, 2.7844, 3.1926, 2.1618, 3.0339, 3.5173], + device='cuda:3'), covar=tensor([0.0814, 0.1587, 0.0846, 0.1478, 0.1007, 0.1554, 0.1112, 0.0958], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0193, 0.0215, 0.0212, 0.0242, 0.0195, 0.0225, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:45:22,539 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94927.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:45:24,590 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.2597, 1.5283, 1.2032, 1.0475, 1.4354, 1.3022, 1.0821, 1.4280], + device='cuda:3'), covar=tensor([0.0086, 0.0054, 0.0079, 0.0087, 0.0070, 0.0064, 0.0092, 0.0077], + device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0059, 0.0059, 0.0064, 0.0061, 0.0057, 0.0056, 0.0053], + device='cuda:3'), out_proj_covar=tensor([5.7096e-05, 5.2494e-05, 5.1447e-05, 5.6753e-05, 5.4070e-05, 4.9744e-05, + 4.9400e-05, 4.6592e-05], device='cuda:3') +2022-11-16 07:45:29,277 INFO [train.py:876] (3/4) Epoch 14, batch 400, loss[loss=0.1203, simple_loss=0.1502, pruned_loss=0.04521, over 5730.00 frames. ], tot_loss[loss=0.1024, simple_loss=0.1331, pruned_loss=0.03582, over 937908.82 frames. ], batch size: 31, lr: 5.85e-03, grad_scale: 16.0 +2022-11-16 07:45:32,605 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.318e+01 1.380e+02 1.703e+02 1.935e+02 3.356e+02, threshold=3.406e+02, percent-clipped=0.0 +2022-11-16 07:45:57,959 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2022-11-16 07:46:19,816 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95004.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:46:33,726 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5970, 3.4366, 3.4757, 3.5862, 3.2781, 3.2432, 3.9173, 3.5164], + device='cuda:3'), covar=tensor([0.0501, 0.0958, 0.0617, 0.1195, 0.0625, 0.0474, 0.0730, 0.0878], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0112, 0.0097, 0.0124, 0.0091, 0.0082, 0.0148, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 07:46:40,791 INFO [train.py:876] (3/4) Epoch 14, batch 500, loss[loss=0.1139, simple_loss=0.1403, pruned_loss=0.04378, over 5610.00 frames. ], tot_loss[loss=0.1036, simple_loss=0.134, pruned_loss=0.03661, over 995421.66 frames. ], batch size: 50, lr: 5.84e-03, grad_scale: 16.0 +2022-11-16 07:46:43,324 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.308e+01 1.428e+02 1.816e+02 2.349e+02 3.391e+02, threshold=3.632e+02, percent-clipped=0.0 +2022-11-16 07:46:51,602 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95052.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:47:26,150 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95105.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:47:48,215 INFO [train.py:876] (3/4) Epoch 14, batch 600, loss[loss=0.1207, simple_loss=0.1536, pruned_loss=0.04386, over 5551.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1347, pruned_loss=0.03639, over 1029243.32 frames. ], batch size: 40, lr: 5.84e-03, grad_scale: 16.0 +2022-11-16 07:47:50,760 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.356e+01 1.449e+02 1.769e+02 2.281e+02 4.546e+02, threshold=3.538e+02, percent-clipped=1.0 +2022-11-16 07:48:07,451 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:48:56,203 INFO [train.py:876] (3/4) Epoch 14, batch 700, loss[loss=0.1182, simple_loss=0.1492, pruned_loss=0.04364, over 5695.00 frames. ], tot_loss[loss=0.1036, simple_loss=0.1349, pruned_loss=0.03616, over 1051147.23 frames. ], batch size: 34, lr: 5.84e-03, grad_scale: 16.0 +2022-11-16 07:48:58,828 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.646e+01 1.509e+02 1.874e+02 2.495e+02 6.608e+02, threshold=3.748e+02, percent-clipped=12.0 +2022-11-16 07:49:11,309 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95260.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:49:16,509 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95268.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:49:16,540 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9005, 2.3739, 2.3385, 1.5765, 2.5593, 2.6748, 2.5032, 2.7543], + device='cuda:3'), covar=tensor([0.1641, 0.1445, 0.1538, 0.2522, 0.0980, 0.1302, 0.0780, 0.0955], + device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0177, 0.0167, 0.0180, 0.0183, 0.0200, 0.0169, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:49:30,085 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2022-11-16 07:49:34,768 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2022-11-16 07:49:44,200 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2022-11-16 07:49:52,354 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95321.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:49:58,203 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9098, 1.3327, 1.6223, 1.6671, 1.4351, 1.6111, 1.6306, 1.4786], + device='cuda:3'), covar=tensor([0.0053, 0.0102, 0.0083, 0.0068, 0.0141, 0.0145, 0.0058, 0.0058], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0029, 0.0030, 0.0039, 0.0034, 0.0030, 0.0037, 0.0035], + device='cuda:3'), out_proj_covar=tensor([2.9188e-05, 2.7645e-05, 2.6744e-05, 3.6963e-05, 3.1244e-05, 2.8944e-05, + 3.5588e-05, 3.3582e-05], device='cuda:3') +2022-11-16 07:49:58,242 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95329.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:50:03,503 INFO [train.py:876] (3/4) Epoch 14, batch 800, loss[loss=0.06778, simple_loss=0.09974, pruned_loss=0.0179, over 5447.00 frames. ], tot_loss[loss=0.1019, simple_loss=0.1337, pruned_loss=0.03508, over 1062503.64 frames. ], batch size: 10, lr: 5.83e-03, grad_scale: 16.0 +2022-11-16 07:50:04,887 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95339.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:50:06,024 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.478e+01 1.483e+02 1.769e+02 2.212e+02 4.574e+02, threshold=3.537e+02, percent-clipped=3.0 +2022-11-16 07:50:38,053 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-16 07:50:46,442 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95400.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:51:11,128 INFO [train.py:876] (3/4) Epoch 14, batch 900, loss[loss=0.07558, simple_loss=0.1208, pruned_loss=0.01518, over 5697.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1328, pruned_loss=0.03511, over 1062814.14 frames. ], batch size: 12, lr: 5.83e-03, grad_scale: 16.0 +2022-11-16 07:51:13,910 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.380e+01 1.449e+02 1.681e+02 2.078e+02 5.193e+02, threshold=3.361e+02, percent-clipped=2.0 +2022-11-16 07:51:26,943 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95461.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 07:52:09,644 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0906, 1.6489, 1.3617, 1.1752, 1.4818, 1.2751, 1.1335, 1.5606], + device='cuda:3'), covar=tensor([0.0074, 0.0041, 0.0059, 0.0074, 0.0063, 0.0058, 0.0077, 0.0067], + device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0059, 0.0059, 0.0064, 0.0061, 0.0057, 0.0056, 0.0054], + device='cuda:3'), out_proj_covar=tensor([5.7188e-05, 5.2475e-05, 5.1053e-05, 5.6306e-05, 5.4165e-05, 4.9811e-05, + 4.9329e-05, 4.7109e-05], device='cuda:3') +2022-11-16 07:52:17,320 INFO [train.py:876] (3/4) Epoch 14, batch 1000, loss[loss=0.1033, simple_loss=0.1385, pruned_loss=0.03411, over 5593.00 frames. ], tot_loss[loss=0.1024, simple_loss=0.134, pruned_loss=0.03545, over 1074264.39 frames. ], batch size: 23, lr: 5.83e-03, grad_scale: 16.0 +2022-11-16 07:52:19,862 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.883e+01 1.427e+02 1.771e+02 2.173e+02 4.557e+02, threshold=3.542e+02, percent-clipped=6.0 +2022-11-16 07:52:20,714 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8012, 1.9162, 2.1213, 1.6183, 1.8183, 1.7949, 1.8327, 1.9547], + device='cuda:3'), covar=tensor([0.0074, 0.0080, 0.0054, 0.0065, 0.0059, 0.0060, 0.0050, 0.0085], + device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0059, 0.0058, 0.0064, 0.0061, 0.0057, 0.0056, 0.0054], + device='cuda:3'), out_proj_covar=tensor([5.6986e-05, 5.2369e-05, 5.0891e-05, 5.6312e-05, 5.4133e-05, 4.9757e-05, + 4.9270e-05, 4.7088e-05], device='cuda:3') +2022-11-16 07:52:29,046 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2022-11-16 07:53:11,119 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:53:11,839 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95617.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:53:16,362 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95624.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:53:24,670 INFO [train.py:876] (3/4) Epoch 14, batch 1100, loss[loss=0.07469, simple_loss=0.1063, pruned_loss=0.02154, over 5728.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1336, pruned_loss=0.0349, over 1071614.77 frames. ], batch size: 13, lr: 5.83e-03, grad_scale: 16.0 +2022-11-16 07:53:27,206 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.869e+01 1.386e+02 1.674e+02 2.201e+02 3.601e+02, threshold=3.349e+02, percent-clipped=2.0 +2022-11-16 07:53:53,101 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95678.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:54:03,919 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95695.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:54:10,049 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.29 vs. limit=5.0 +2022-11-16 07:54:13,042 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6125, 4.8065, 3.0185, 4.6199, 3.7578, 3.0973, 2.6546, 4.0693], + device='cuda:3'), covar=tensor([0.1300, 0.0195, 0.1115, 0.0295, 0.0534, 0.1107, 0.1627, 0.0320], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0141, 0.0154, 0.0146, 0.0172, 0.0167, 0.0157, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:54:13,685 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6720, 4.8609, 3.6992, 2.2062, 4.6157, 2.1864, 4.3802, 2.7439], + device='cuda:3'), covar=tensor([0.1385, 0.0089, 0.0527, 0.1950, 0.0131, 0.1519, 0.0188, 0.1392], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0104, 0.0115, 0.0111, 0.0102, 0.0119, 0.0100, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:54:25,174 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4812, 2.2915, 3.1787, 2.9158, 3.1820, 2.2110, 2.9231, 3.5346], + device='cuda:3'), covar=tensor([0.0850, 0.1411, 0.1157, 0.1202, 0.1221, 0.1513, 0.1254, 0.0896], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0190, 0.0210, 0.0208, 0.0237, 0.0192, 0.0220, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:54:32,046 INFO [train.py:876] (3/4) Epoch 14, batch 1200, loss[loss=0.08789, simple_loss=0.1312, pruned_loss=0.02227, over 5754.00 frames. ], tot_loss[loss=0.1001, simple_loss=0.1322, pruned_loss=0.03399, over 1079044.80 frames. ], batch size: 14, lr: 5.82e-03, grad_scale: 16.0 +2022-11-16 07:54:34,544 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.521e+01 1.381e+02 1.760e+02 2.068e+02 4.246e+02, threshold=3.521e+02, percent-clipped=4.0 +2022-11-16 07:54:38,787 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9166, 2.4187, 3.5700, 3.2229, 3.6895, 2.4218, 3.2611, 3.8849], + device='cuda:3'), covar=tensor([0.0734, 0.1726, 0.0713, 0.1279, 0.0639, 0.1662, 0.1185, 0.0665], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0191, 0.0211, 0.0209, 0.0237, 0.0193, 0.0220, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:54:47,846 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95761.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:55:20,111 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95809.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:55:22,113 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95812.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:55:33,948 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 +2022-11-16 07:55:38,696 INFO [train.py:876] (3/4) Epoch 14, batch 1300, loss[loss=0.1758, simple_loss=0.1797, pruned_loss=0.08599, over 5490.00 frames. ], tot_loss[loss=0.1005, simple_loss=0.1326, pruned_loss=0.03421, over 1088469.93 frames. ], batch size: 64, lr: 5.82e-03, grad_scale: 16.0 +2022-11-16 07:55:41,903 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.562e+01 1.314e+02 1.675e+02 2.012e+02 3.727e+02, threshold=3.350e+02, percent-clipped=1.0 +2022-11-16 07:55:45,978 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95847.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:56:02,941 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95873.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:56:04,267 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7609, 2.2131, 2.4201, 3.0129, 2.9687, 2.3849, 2.0792, 3.0424], + device='cuda:3'), covar=tensor([0.2163, 0.2311, 0.2111, 0.1917, 0.1393, 0.2652, 0.2339, 0.1572], + device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0199, 0.0188, 0.0304, 0.0228, 0.0203, 0.0191, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:56:18,630 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1780, 1.8668, 2.1220, 2.2580, 2.6231, 1.9824, 1.7401, 2.4003], + device='cuda:3'), covar=tensor([0.2488, 0.2278, 0.1814, 0.0845, 0.1161, 0.2653, 0.2329, 0.1782], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0197, 0.0186, 0.0301, 0.0226, 0.0201, 0.0190, 0.0250], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 07:56:26,521 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8004, 4.1876, 3.8248, 3.5668, 2.1338, 4.0626, 2.3600, 3.3571], + device='cuda:3'), covar=tensor([0.0409, 0.0170, 0.0162, 0.0320, 0.0652, 0.0141, 0.0504, 0.0180], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0184, 0.0183, 0.0208, 0.0196, 0.0184, 0.0194, 0.0186], + 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') +2022-11-16 07:56:27,147 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95908.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 07:56:32,278 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95916.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:56:37,638 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95924.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:56:37,684 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.9957, 0.7331, 0.8755, 0.7607, 1.0592, 0.8114, 0.5594, 0.7464], + device='cuda:3'), covar=tensor([0.0241, 0.0402, 0.0331, 0.0417, 0.0312, 0.0309, 0.0776, 0.0296], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0026, 0.0019, 0.0022, 0.0018, 0.0017, 0.0025, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.2978e-05, 1.2983e-04, 9.9898e-05, 1.1243e-04, 1.0075e-04, 9.4303e-05, + 1.2414e-04, 9.3207e-05], device='cuda:3') +2022-11-16 07:56:46,581 INFO [train.py:876] (3/4) Epoch 14, batch 1400, loss[loss=0.09372, simple_loss=0.1234, pruned_loss=0.032, over 5601.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.133, pruned_loss=0.03506, over 1081365.26 frames. ], batch size: 23, lr: 5.82e-03, grad_scale: 16.0 +2022-11-16 07:56:49,505 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.691e+01 1.448e+02 1.696e+02 2.137e+02 4.589e+02, threshold=3.392e+02, percent-clipped=2.0 +2022-11-16 07:57:04,976 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95964.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:10,152 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95972.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:10,817 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95973.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:15,504 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95980.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:25,801 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95995.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:48,962 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5068, 1.3134, 1.2824, 0.8518, 1.3187, 1.5484, 0.8002, 1.0563], + device='cuda:3'), covar=tensor([0.0351, 0.0473, 0.0418, 0.0938, 0.0421, 0.0258, 0.0968, 0.0420], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0026, 0.0019, 0.0022, 0.0019, 0.0017, 0.0025, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.4635e-05, 1.3177e-04, 1.0164e-04, 1.1396e-04, 1.0226e-04, 9.5565e-05, + 1.2593e-04, 9.4567e-05], device='cuda:3') +2022-11-16 07:57:52,008 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2022-11-16 07:57:54,059 INFO [train.py:876] (3/4) Epoch 14, batch 1500, loss[loss=0.1115, simple_loss=0.1318, pruned_loss=0.04563, over 5355.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1327, pruned_loss=0.03486, over 1081910.27 frames. ], batch size: 70, lr: 5.81e-03, grad_scale: 16.0 +2022-11-16 07:57:56,679 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.465e+02 1.739e+02 2.113e+02 3.912e+02, threshold=3.478e+02, percent-clipped=1.0 +2022-11-16 07:57:56,878 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96041.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:57:58,097 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96043.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:58:20,665 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0309, 2.5268, 2.5901, 1.5612, 2.6329, 2.8991, 2.8289, 2.9110], + device='cuda:3'), covar=tensor([0.1894, 0.1892, 0.1634, 0.3129, 0.0908, 0.1429, 0.0791, 0.1148], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0179, 0.0169, 0.0182, 0.0185, 0.0202, 0.0170, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 07:58:24,065 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96080.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:58:52,334 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 +2022-11-16 07:58:55,897 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1149, 1.6680, 2.1787, 1.6686, 1.9074, 1.9729, 1.6891, 1.5809], + device='cuda:3'), covar=tensor([0.0052, 0.0073, 0.0026, 0.0073, 0.0093, 0.0121, 0.0050, 0.0074], + device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0029, 0.0029, 0.0038, 0.0033, 0.0030, 0.0037, 0.0036], + device='cuda:3'), out_proj_covar=tensor([2.8699e-05, 2.7520e-05, 2.6460e-05, 3.6547e-05, 3.0645e-05, 2.8430e-05, + 3.4843e-05, 3.3952e-05], device='cuda:3') +2022-11-16 07:59:02,197 INFO [train.py:876] (3/4) Epoch 14, batch 1600, loss[loss=0.1417, simple_loss=0.1576, pruned_loss=0.06292, over 5467.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.1328, pruned_loss=0.03487, over 1076669.94 frames. ], batch size: 53, lr: 5.81e-03, grad_scale: 16.0 +2022-11-16 07:59:04,692 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.966e+01 1.432e+02 1.726e+02 2.177e+02 5.569e+02, threshold=3.453e+02, percent-clipped=4.0 +2022-11-16 07:59:04,912 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96141.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:59:18,932 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2022-11-16 07:59:21,829 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96165.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 07:59:23,754 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96168.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 07:59:40,369 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2022-11-16 07:59:47,263 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 07:59:59,992 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9758, 4.7285, 4.8915, 5.0718, 4.4422, 4.3362, 5.3432, 4.8400], + device='cuda:3'), covar=tensor([0.0286, 0.0848, 0.0302, 0.1117, 0.0527, 0.0327, 0.0729, 0.0681], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0110, 0.0096, 0.0122, 0.0090, 0.0082, 0.0146, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:00:03,449 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96226.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:00:10,795 INFO [train.py:876] (3/4) Epoch 14, batch 1700, loss[loss=0.1925, simple_loss=0.1818, pruned_loss=0.1016, over 2941.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1326, pruned_loss=0.03466, over 1073573.76 frames. ], batch size: 284, lr: 5.81e-03, grad_scale: 16.0 +2022-11-16 08:00:14,147 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.983e+01 1.431e+02 1.731e+02 2.193e+02 6.139e+02, threshold=3.462e+02, percent-clipped=4.0 +2022-11-16 08:00:36,067 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96273.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:00:41,666 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 +2022-11-16 08:00:48,436 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6706, 2.3104, 3.2492, 2.9652, 3.3856, 2.1816, 2.9944, 3.5888], + device='cuda:3'), covar=tensor([0.0763, 0.1420, 0.0890, 0.1394, 0.0594, 0.1711, 0.1220, 0.0815], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0192, 0.0213, 0.0211, 0.0239, 0.0194, 0.0222, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:01:07,789 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96321.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:01:10,176 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:01:18,263 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96336.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:01:18,841 INFO [train.py:876] (3/4) Epoch 14, batch 1800, loss[loss=0.07742, simple_loss=0.1224, pruned_loss=0.01622, over 5527.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1323, pruned_loss=0.03475, over 1074485.01 frames. ], batch size: 13, lr: 5.80e-03, grad_scale: 16.0 +2022-11-16 08:01:19,772 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 +2022-11-16 08:01:22,026 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.395e+01 1.406e+02 1.732e+02 2.199e+02 6.902e+02, threshold=3.464e+02, percent-clipped=4.0 +2022-11-16 08:01:43,954 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6669, 1.6079, 1.7941, 1.3220, 1.5201, 1.6036, 1.3571, 0.9910], + device='cuda:3'), covar=tensor([0.0056, 0.0063, 0.0042, 0.0079, 0.0080, 0.0094, 0.0058, 0.0072], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0030, 0.0030, 0.0039, 0.0033, 0.0030, 0.0037, 0.0036], + device='cuda:3'), out_proj_covar=tensor([2.9261e-05, 2.7741e-05, 2.6730e-05, 3.6972e-05, 3.0714e-05, 2.8837e-05, + 3.5232e-05, 3.4275e-05], device='cuda:3') +2022-11-16 08:01:50,743 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96385.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:01:54,347 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96390.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:02:04,307 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8789, 2.1921, 1.9930, 1.4952, 2.1976, 2.3708, 2.1608, 2.5069], + device='cuda:3'), covar=tensor([0.1460, 0.1318, 0.1680, 0.2341, 0.0990, 0.1160, 0.0861, 0.1029], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0180, 0.0170, 0.0183, 0.0186, 0.0205, 0.0172, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:02:25,307 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96436.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:02:25,903 INFO [train.py:876] (3/4) Epoch 14, batch 1900, loss[loss=0.1485, simple_loss=0.1707, pruned_loss=0.06313, over 5455.00 frames. ], tot_loss[loss=0.1002, simple_loss=0.1317, pruned_loss=0.03431, over 1076707.54 frames. ], batch size: 58, lr: 5.80e-03, grad_scale: 16.0 +2022-11-16 08:02:29,415 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.308e+01 1.441e+02 1.720e+02 2.105e+02 7.193e+02, threshold=3.439e+02, percent-clipped=2.0 +2022-11-16 08:02:35,461 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96451.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:02:46,382 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96468.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:02:56,872 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 +2022-11-16 08:03:10,714 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96503.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:03:19,147 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96516.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:03:22,351 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96521.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:03:32,721 INFO [train.py:876] (3/4) Epoch 14, batch 2000, loss[loss=0.06534, simple_loss=0.1059, pruned_loss=0.01241, over 5690.00 frames. ], tot_loss[loss=0.0996, simple_loss=0.1315, pruned_loss=0.03386, over 1085165.25 frames. ], batch size: 12, lr: 5.80e-03, grad_scale: 16.0 +2022-11-16 08:03:36,667 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.185e+01 1.342e+02 1.750e+02 2.213e+02 4.524e+02, threshold=3.499e+02, percent-clipped=3.0 +2022-11-16 08:03:43,303 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96551.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:04:40,271 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96636.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:04:40,792 INFO [train.py:876] (3/4) Epoch 14, batch 2100, loss[loss=0.0717, simple_loss=0.1094, pruned_loss=0.01699, over 5543.00 frames. ], tot_loss[loss=0.09971, simple_loss=0.1321, pruned_loss=0.03368, over 1087332.38 frames. ], batch size: 13, lr: 5.80e-03, grad_scale: 8.0 +2022-11-16 08:04:45,309 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.473e+01 1.415e+02 1.802e+02 2.261e+02 4.449e+02, threshold=3.604e+02, percent-clipped=2.0 +2022-11-16 08:04:58,300 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2022-11-16 08:05:10,357 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:05:12,897 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96684.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:05:35,588 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7851, 2.2829, 2.2285, 1.4969, 2.6373, 2.6686, 2.5228, 2.4766], + device='cuda:3'), covar=tensor([0.2230, 0.1831, 0.1557, 0.3075, 0.1007, 0.1174, 0.0987, 0.1428], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0180, 0.0169, 0.0183, 0.0186, 0.0204, 0.0171, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:05:48,012 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96736.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:05:48,599 INFO [train.py:876] (3/4) Epoch 14, batch 2200, loss[loss=0.06865, simple_loss=0.09556, pruned_loss=0.02087, over 5159.00 frames. ], tot_loss[loss=0.1008, simple_loss=0.1327, pruned_loss=0.0344, over 1086718.65 frames. ], batch size: 8, lr: 5.79e-03, grad_scale: 8.0 +2022-11-16 08:05:52,487 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.562e+01 1.449e+02 1.748e+02 2.179e+02 3.480e+02, threshold=3.495e+02, percent-clipped=0.0 +2022-11-16 08:05:54,139 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 08:05:55,271 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96746.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:06:15,848 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96776.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:06:20,964 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96784.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:06:36,270 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6514, 5.1146, 3.4844, 4.8704, 3.8609, 3.4325, 2.9895, 4.3319], + device='cuda:3'), covar=tensor([0.1468, 0.0206, 0.0941, 0.0329, 0.0567, 0.0898, 0.1603, 0.0329], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0140, 0.0154, 0.0147, 0.0172, 0.0166, 0.0156, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:06:46,128 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9487, 1.9336, 2.1846, 1.7210, 1.8974, 1.9564, 2.1163, 1.7880], + device='cuda:3'), covar=tensor([0.0067, 0.0059, 0.0042, 0.0065, 0.0061, 0.0040, 0.0040, 0.0112], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0061, 0.0060, 0.0065, 0.0063, 0.0059, 0.0057, 0.0056], + device='cuda:3'), out_proj_covar=tensor([5.8275e-05, 5.3568e-05, 5.1942e-05, 5.7365e-05, 5.5654e-05, 5.0971e-05, + 5.0119e-05, 4.8723e-05], device='cuda:3') +2022-11-16 08:06:46,716 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96821.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:06:51,439 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2022-11-16 08:06:56,990 INFO [train.py:876] (3/4) Epoch 14, batch 2300, loss[loss=0.09052, simple_loss=0.1285, pruned_loss=0.02626, over 5711.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.1329, pruned_loss=0.03485, over 1086294.07 frames. ], batch size: 15, lr: 5.79e-03, grad_scale: 4.0 +2022-11-16 08:06:57,159 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96837.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:07:01,480 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.903e+01 1.452e+02 1.773e+02 2.447e+02 7.176e+02, threshold=3.545e+02, percent-clipped=7.0 +2022-11-16 08:07:17,353 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9332, 2.3217, 3.5102, 3.0513, 3.7025, 2.3924, 3.2587, 3.9000], + device='cuda:3'), covar=tensor([0.1196, 0.1683, 0.0919, 0.1512, 0.0679, 0.1628, 0.1226, 0.0771], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0194, 0.0215, 0.0213, 0.0241, 0.0194, 0.0225, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:07:18,438 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96869.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:07:57,883 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2947, 4.4783, 4.3165, 3.8034, 2.4666, 4.8433, 2.7126, 4.3050], + device='cuda:3'), covar=tensor([0.0380, 0.0165, 0.0209, 0.0354, 0.0620, 0.0125, 0.0591, 0.0126], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0184, 0.0183, 0.0210, 0.0197, 0.0185, 0.0196, 0.0189], + 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') +2022-11-16 08:08:04,678 INFO [train.py:876] (3/4) Epoch 14, batch 2400, loss[loss=0.0989, simple_loss=0.1314, pruned_loss=0.03318, over 5597.00 frames. ], tot_loss[loss=0.1, simple_loss=0.132, pruned_loss=0.03399, over 1090812.63 frames. ], batch size: 24, lr: 5.79e-03, grad_scale: 8.0 +2022-11-16 08:08:09,596 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.192e+01 1.391e+02 1.679e+02 2.056e+02 5.016e+02, threshold=3.358e+02, percent-clipped=6.0 +2022-11-16 08:08:34,222 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96980.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:08:48,040 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97000.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:09:06,580 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97028.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:09:12,706 INFO [train.py:876] (3/4) Epoch 14, batch 2500, loss[loss=0.08293, simple_loss=0.1271, pruned_loss=0.0194, over 5550.00 frames. ], tot_loss[loss=0.09864, simple_loss=0.1316, pruned_loss=0.03286, over 1093996.46 frames. ], batch size: 25, lr: 5.78e-03, grad_scale: 8.0 +2022-11-16 08:09:17,248 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.842e+01 1.491e+02 1.726e+02 2.085e+02 3.787e+02, threshold=3.452e+02, percent-clipped=1.0 +2022-11-16 08:09:18,704 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97046.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:09:29,240 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97061.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:09:36,615 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.46 vs. limit=5.0 +2022-11-16 08:09:51,073 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97094.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:10:16,530 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97132.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:10:19,664 INFO [train.py:876] (3/4) Epoch 14, batch 2600, loss[loss=0.1269, simple_loss=0.1541, pruned_loss=0.04983, over 5688.00 frames. ], tot_loss[loss=0.1008, simple_loss=0.1327, pruned_loss=0.03443, over 1087446.05 frames. ], batch size: 36, lr: 5.78e-03, grad_scale: 8.0 +2022-11-16 08:10:24,998 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.432e+01 1.490e+02 1.874e+02 2.362e+02 5.488e+02, threshold=3.748e+02, percent-clipped=3.0 +2022-11-16 08:10:48,696 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-16 08:11:23,860 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7458, 1.7063, 1.7492, 1.5018, 1.6489, 1.6246, 1.5586, 1.8415], + device='cuda:3'), covar=tensor([0.0081, 0.0073, 0.0069, 0.0063, 0.0064, 0.0063, 0.0069, 0.0054], + device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0062, 0.0061, 0.0067, 0.0065, 0.0060, 0.0059, 0.0057], + device='cuda:3'), out_proj_covar=tensor([6.0220e-05, 5.5043e-05, 5.3167e-05, 5.8846e-05, 5.7089e-05, 5.2373e-05, + 5.1969e-05, 4.9687e-05], device='cuda:3') +2022-11-16 08:11:27,612 INFO [train.py:876] (3/4) Epoch 14, batch 2700, loss[loss=0.09226, simple_loss=0.1335, pruned_loss=0.0255, over 5723.00 frames. ], tot_loss[loss=0.1001, simple_loss=0.1318, pruned_loss=0.03418, over 1083864.48 frames. ], batch size: 20, lr: 5.78e-03, grad_scale: 8.0 +2022-11-16 08:11:32,090 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.527e+01 1.417e+02 1.707e+02 2.038e+02 4.656e+02, threshold=3.414e+02, percent-clipped=3.0 +2022-11-16 08:11:43,166 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97260.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:11:56,797 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5123, 4.9425, 5.3030, 4.9519, 5.6001, 5.4875, 4.7298, 5.5514], + device='cuda:3'), covar=tensor([0.0352, 0.0352, 0.0472, 0.0312, 0.0336, 0.0197, 0.0272, 0.0267], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0159, 0.0111, 0.0148, 0.0189, 0.0116, 0.0132, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:12:05,078 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 +2022-11-16 08:12:14,967 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9573, 2.6015, 3.0077, 3.8830, 3.7990, 3.0382, 2.7284, 3.8766], + device='cuda:3'), covar=tensor([0.0820, 0.2916, 0.2409, 0.2362, 0.1314, 0.2906, 0.2058, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0198, 0.0188, 0.0297, 0.0228, 0.0201, 0.0188, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 08:12:24,367 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97321.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:12:34,940 INFO [train.py:876] (3/4) Epoch 14, batch 2800, loss[loss=0.08269, simple_loss=0.1201, pruned_loss=0.02264, over 5479.00 frames. ], tot_loss[loss=0.09881, simple_loss=0.1309, pruned_loss=0.03337, over 1086386.73 frames. ], batch size: 12, lr: 5.77e-03, grad_scale: 8.0 +2022-11-16 08:12:38,283 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97342.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:12:39,404 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.283e+01 1.377e+02 1.617e+02 1.956e+02 4.684e+02, threshold=3.233e+02, percent-clipped=2.0 +2022-11-16 08:12:41,615 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97347.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:12:47,488 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97356.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:12:57,872 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4696, 4.8857, 4.4833, 4.9782, 4.9496, 4.1431, 4.5231, 4.2665], + device='cuda:3'), covar=tensor([0.0274, 0.0322, 0.1166, 0.0300, 0.0310, 0.0457, 0.0474, 0.0447], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0179, 0.0271, 0.0175, 0.0222, 0.0171, 0.0188, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:12:59,855 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97374.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:13:16,318 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3398, 4.3799, 3.3982, 2.0037, 4.0456, 1.5548, 3.9686, 2.3577], + device='cuda:3'), covar=tensor([0.1474, 0.0137, 0.0587, 0.1776, 0.0206, 0.1764, 0.0231, 0.1365], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0103, 0.0115, 0.0110, 0.0102, 0.0118, 0.0100, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:13:19,933 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97403.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:13:21,883 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 +2022-11-16 08:13:23,118 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97408.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:13:39,493 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97432.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:13:41,859 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97435.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:13:43,017 INFO [train.py:876] (3/4) Epoch 14, batch 2900, loss[loss=0.1146, simple_loss=0.1381, pruned_loss=0.04554, over 5695.00 frames. ], tot_loss[loss=0.1, simple_loss=0.1317, pruned_loss=0.03413, over 1077707.17 frames. ], batch size: 34, lr: 5.77e-03, grad_scale: 8.0 +2022-11-16 08:13:47,951 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.266e+01 1.378e+02 1.705e+02 2.124e+02 3.777e+02, threshold=3.411e+02, percent-clipped=2.0 +2022-11-16 08:14:00,935 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8386, 2.5642, 2.8182, 3.6149, 3.6435, 2.7841, 2.3950, 3.6751], + device='cuda:3'), covar=tensor([0.0631, 0.2527, 0.2181, 0.2769, 0.1236, 0.2984, 0.2098, 0.1023], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0198, 0.0188, 0.0297, 0.0228, 0.0201, 0.0189, 0.0250], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 08:14:12,676 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97480.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:14:50,998 INFO [train.py:876] (3/4) Epoch 14, batch 3000, loss[loss=0.09062, simple_loss=0.1319, pruned_loss=0.02467, over 5595.00 frames. ], tot_loss[loss=0.1004, simple_loss=0.1322, pruned_loss=0.03435, over 1079029.27 frames. ], batch size: 23, lr: 5.77e-03, grad_scale: 8.0 +2022-11-16 08:14:50,998 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 08:14:59,695 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7802, 3.5601, 3.6702, 3.1593, 1.9960, 3.7031, 2.3389, 3.1687], + device='cuda:3'), covar=tensor([0.0416, 0.0184, 0.0169, 0.0368, 0.0676, 0.0171, 0.0620, 0.0206], + device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0187, 0.0185, 0.0211, 0.0200, 0.0188, 0.0196, 0.0190], + 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') +2022-11-16 08:15:01,806 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.1599, 4.9505, 4.9244, 4.8221, 5.2193, 4.9598, 4.7869, 5.3474], + device='cuda:3'), covar=tensor([0.0343, 0.0313, 0.0492, 0.0367, 0.0375, 0.0248, 0.0205, 0.0214], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0159, 0.0111, 0.0148, 0.0189, 0.0115, 0.0131, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:15:06,568 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5240, 3.7732, 3.1119, 3.6280, 2.8963, 2.8176, 2.3078, 3.2778], + device='cuda:3'), covar=tensor([0.1069, 0.0191, 0.0609, 0.0249, 0.1150, 0.0874, 0.1600, 0.0347], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0141, 0.0154, 0.0148, 0.0173, 0.0166, 0.0157, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:15:08,553 INFO [train.py:908] (3/4) Epoch 14, validation: loss=0.178, simple_loss=0.188, pruned_loss=0.08395, over 1530663.00 frames. +2022-11-16 08:15:08,554 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 08:15:12,972 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.365e+01 1.436e+02 1.776e+02 2.242e+02 5.969e+02, threshold=3.553e+02, percent-clipped=3.0 +2022-11-16 08:15:20,366 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97555.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:16:01,561 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:16:01,652 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:16:08,558 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6175, 1.6934, 1.7407, 1.6579, 1.7434, 1.7744, 1.6367, 1.6120], + device='cuda:3'), covar=tensor([0.0071, 0.0072, 0.0056, 0.0058, 0.0054, 0.0060, 0.0059, 0.0084], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0061, 0.0060, 0.0065, 0.0063, 0.0059, 0.0057, 0.0056], + device='cuda:3'), out_proj_covar=tensor([5.9248e-05, 5.4302e-05, 5.2646e-05, 5.7499e-05, 5.5907e-05, 5.1458e-05, + 5.0788e-05, 4.9250e-05], device='cuda:3') +2022-11-16 08:16:10,819 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2022-11-16 08:16:12,656 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2022-11-16 08:16:14,414 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.0608, 1.4465, 1.2926, 1.3256, 1.3922, 1.8082, 1.5551, 1.3192], + device='cuda:3'), covar=tensor([0.3668, 0.1271, 0.3565, 0.3233, 0.2213, 0.0869, 0.2235, 0.3160], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0107, 0.0108, 0.0108, 0.0078, 0.0074, 0.0088, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:16:15,010 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.3884, 3.2286, 3.3582, 3.3164, 3.1693, 2.9046, 3.6230, 3.3596], + device='cuda:3'), covar=tensor([0.0426, 0.0904, 0.0448, 0.1206, 0.0572, 0.0539, 0.0759, 0.0709], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0112, 0.0097, 0.0125, 0.0090, 0.0084, 0.0148, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:16:15,589 INFO [train.py:876] (3/4) Epoch 14, batch 3100, loss[loss=0.1101, simple_loss=0.1453, pruned_loss=0.0374, over 5311.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1331, pruned_loss=0.03509, over 1076691.22 frames. ], batch size: 79, lr: 5.77e-03, grad_scale: 8.0 +2022-11-16 08:16:16,391 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0005, 1.5060, 1.6583, 1.2727, 1.5028, 1.6120, 1.3583, 1.1472], + device='cuda:3'), covar=tensor([0.0033, 0.0089, 0.0057, 0.0082, 0.0083, 0.0095, 0.0059, 0.0075], + device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0029, 0.0029, 0.0038, 0.0034, 0.0030, 0.0037, 0.0036], + device='cuda:3'), out_proj_covar=tensor([2.9126e-05, 2.7479e-05, 2.6369e-05, 3.6547e-05, 3.1490e-05, 2.9179e-05, + 3.5078e-05, 3.4453e-05], device='cuda:3') +2022-11-16 08:16:20,446 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.977e+01 1.456e+02 1.784e+02 2.207e+02 3.883e+02, threshold=3.567e+02, percent-clipped=1.0 +2022-11-16 08:16:28,861 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97656.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:16:41,509 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97674.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:16:57,845 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97698.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:17:01,173 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97703.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:17:01,770 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:17:19,795 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97730.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:17:23,173 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97735.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:17:24,288 INFO [train.py:876] (3/4) Epoch 14, batch 3200, loss[loss=0.06707, simple_loss=0.108, pruned_loss=0.01306, over 5681.00 frames. ], tot_loss[loss=0.1019, simple_loss=0.1335, pruned_loss=0.03513, over 1078385.87 frames. ], batch size: 12, lr: 5.76e-03, grad_scale: 8.0 +2022-11-16 08:17:29,197 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.777e+01 1.455e+02 1.714e+02 2.116e+02 4.590e+02, threshold=3.428e+02, percent-clipped=2.0 +2022-11-16 08:18:32,093 INFO [train.py:876] (3/4) Epoch 14, batch 3300, loss[loss=0.1323, simple_loss=0.1447, pruned_loss=0.05996, over 4158.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1335, pruned_loss=0.03508, over 1078086.16 frames. ], batch size: 181, lr: 5.76e-03, grad_scale: 8.0 +2022-11-16 08:18:36,476 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.779e+01 1.482e+02 1.736e+02 2.155e+02 3.992e+02, threshold=3.473e+02, percent-clipped=3.0 +2022-11-16 08:18:37,339 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8807, 2.3282, 2.1403, 1.4907, 2.4004, 2.4822, 2.3935, 2.6368], + device='cuda:3'), covar=tensor([0.1839, 0.1500, 0.1630, 0.2681, 0.1024, 0.1437, 0.1001, 0.1116], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0178, 0.0167, 0.0181, 0.0186, 0.0204, 0.0171, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:18:39,689 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8741, 5.0841, 3.4094, 4.7016, 3.8846, 3.6984, 3.2605, 4.4820], + device='cuda:3'), covar=tensor([0.1309, 0.0275, 0.0896, 0.0360, 0.0554, 0.0776, 0.1657, 0.0248], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0141, 0.0153, 0.0147, 0.0173, 0.0164, 0.0156, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:19:06,975 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97888.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:19:14,769 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8937, 1.9531, 1.6856, 1.9037, 1.9753, 1.8519, 1.6939, 1.8559], + device='cuda:3'), covar=tensor([0.0470, 0.0862, 0.1717, 0.0838, 0.0784, 0.0549, 0.1484, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0182, 0.0273, 0.0176, 0.0225, 0.0173, 0.0189, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:19:18,621 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.5134, 4.9975, 5.2684, 4.9521, 5.5957, 5.3594, 4.7120, 5.5330], + device='cuda:3'), covar=tensor([0.0347, 0.0316, 0.0406, 0.0384, 0.0313, 0.0244, 0.0280, 0.0222], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0160, 0.0112, 0.0149, 0.0190, 0.0117, 0.0133, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:19:22,251 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:19:25,494 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97916.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:19:39,518 INFO [train.py:876] (3/4) Epoch 14, batch 3400, loss[loss=0.1138, simple_loss=0.1426, pruned_loss=0.04249, over 5552.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1328, pruned_loss=0.03423, over 1079406.51 frames. ], batch size: 40, lr: 5.76e-03, grad_scale: 8.0 +2022-11-16 08:19:44,334 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.552e+01 1.387e+02 1.696e+02 2.106e+02 3.635e+02, threshold=3.392e+02, percent-clipped=1.0 +2022-11-16 08:19:47,859 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97949.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:19:55,138 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 +2022-11-16 08:19:58,065 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97964.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:20:21,528 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97998.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:20:25,089 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98003.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:20:27,268 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.98 vs. limit=5.0 +2022-11-16 08:20:43,491 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98030.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:20:43,529 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98030.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:20:44,847 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6566, 1.5955, 2.2001, 1.7402, 2.2023, 2.1628, 1.6891, 1.8802], + device='cuda:3'), covar=tensor([0.1091, 0.0803, 0.0553, 0.0785, 0.0754, 0.0522, 0.0826, 0.0475], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0026, 0.0018, 0.0021, 0.0018, 0.0017, 0.0024, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.1351e-05, 1.2866e-04, 9.8330e-05, 1.1084e-04, 9.9307e-05, 9.2706e-05, + 1.2267e-04, 9.2135e-05], device='cuda:3') +2022-11-16 08:20:47,938 INFO [train.py:876] (3/4) Epoch 14, batch 3500, loss[loss=0.2315, simple_loss=0.2153, pruned_loss=0.1239, over 3008.00 frames. ], tot_loss[loss=0.09753, simple_loss=0.13, pruned_loss=0.03252, over 1080819.89 frames. ], batch size: 284, lr: 5.75e-03, grad_scale: 8.0 +2022-11-16 08:20:52,493 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.923e+01 1.343e+02 1.705e+02 2.352e+02 4.621e+02, threshold=3.411e+02, percent-clipped=6.0 +2022-11-16 08:20:53,902 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98046.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:20:57,584 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98051.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:21:15,968 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98078.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:21:35,049 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98106.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:21:55,983 INFO [train.py:876] (3/4) Epoch 14, batch 3600, loss[loss=0.101, simple_loss=0.1471, pruned_loss=0.02748, over 5677.00 frames. ], tot_loss[loss=0.09778, simple_loss=0.1305, pruned_loss=0.03252, over 1084299.17 frames. ], batch size: 36, lr: 5.75e-03, grad_scale: 8.0 +2022-11-16 08:21:58,800 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2022-11-16 08:22:00,947 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.265e+01 1.372e+02 1.706e+02 2.197e+02 4.106e+02, threshold=3.412e+02, percent-clipped=3.0 +2022-11-16 08:22:11,909 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2981, 4.7853, 5.0669, 4.7264, 5.3799, 5.1195, 4.6271, 5.2901], + device='cuda:3'), covar=tensor([0.0353, 0.0342, 0.0452, 0.0367, 0.0328, 0.0275, 0.0326, 0.0266], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0158, 0.0111, 0.0148, 0.0188, 0.0116, 0.0132, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:22:16,984 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98167.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:22:30,390 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8648, 4.0723, 3.7945, 3.4349, 1.9672, 3.9662, 2.3709, 3.3485], + device='cuda:3'), covar=tensor([0.0452, 0.0166, 0.0201, 0.0427, 0.0799, 0.0186, 0.0622, 0.0202], + device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0187, 0.0185, 0.0212, 0.0199, 0.0187, 0.0195, 0.0190], + 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') +2022-11-16 08:22:31,710 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.62 vs. limit=5.0 +2022-11-16 08:22:47,080 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98211.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:22:59,902 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 +2022-11-16 08:23:04,725 INFO [train.py:876] (3/4) Epoch 14, batch 3700, loss[loss=0.09248, simple_loss=0.1305, pruned_loss=0.02723, over 5494.00 frames. ], tot_loss[loss=0.09873, simple_loss=0.1317, pruned_loss=0.0329, over 1089670.95 frames. ], batch size: 12, lr: 5.75e-03, grad_scale: 8.0 +2022-11-16 08:23:09,245 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.991e+01 1.391e+02 1.713e+02 2.053e+02 4.916e+02, threshold=3.427e+02, percent-clipped=4.0 +2022-11-16 08:23:09,345 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98244.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:23:19,338 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98259.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:24:08,798 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98330.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:24:13,986 INFO [train.py:876] (3/4) Epoch 14, batch 3800, loss[loss=0.07971, simple_loss=0.1139, pruned_loss=0.02273, over 5689.00 frames. ], tot_loss[loss=0.0969, simple_loss=0.1302, pruned_loss=0.03181, over 1094688.15 frames. ], batch size: 15, lr: 5.74e-03, grad_scale: 4.0 +2022-11-16 08:24:19,594 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.532e+01 1.388e+02 1.687e+02 2.091e+02 4.683e+02, threshold=3.374e+02, percent-clipped=2.0 +2022-11-16 08:24:42,608 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98378.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:25:04,764 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98410.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:25:22,400 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7138, 1.2932, 1.7393, 1.2632, 1.7704, 1.7479, 1.2475, 1.4875], + device='cuda:3'), covar=tensor([0.0629, 0.0760, 0.0498, 0.0714, 0.0690, 0.0926, 0.0716, 0.0523], + device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0026, 0.0018, 0.0021, 0.0018, 0.0017, 0.0024, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.1846e-05, 1.2926e-04, 9.9229e-05, 1.1111e-04, 9.9514e-05, 9.2764e-05, + 1.2232e-04, 9.2444e-05], device='cuda:3') +2022-11-16 08:25:22,856 INFO [train.py:876] (3/4) Epoch 14, batch 3900, loss[loss=0.08596, simple_loss=0.1317, pruned_loss=0.0201, over 5488.00 frames. ], tot_loss[loss=0.09873, simple_loss=0.1313, pruned_loss=0.03306, over 1088483.75 frames. ], batch size: 17, lr: 5.74e-03, grad_scale: 4.0 +2022-11-16 08:25:27,986 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.450e+01 1.480e+02 1.725e+02 2.158e+02 4.236e+02, threshold=3.450e+02, percent-clipped=3.0 +2022-11-16 08:25:39,864 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98462.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:25:45,737 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98471.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:25:59,159 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4282, 4.4571, 3.4184, 1.9478, 4.1684, 1.6708, 4.1042, 2.3130], + device='cuda:3'), covar=tensor([0.1537, 0.0150, 0.0747, 0.2081, 0.0184, 0.1996, 0.0213, 0.1555], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0102, 0.0114, 0.0110, 0.0102, 0.0118, 0.0099, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:26:30,062 INFO [train.py:876] (3/4) Epoch 14, batch 4000, loss[loss=0.105, simple_loss=0.1393, pruned_loss=0.03533, over 5558.00 frames. ], tot_loss[loss=0.1002, simple_loss=0.1322, pruned_loss=0.03406, over 1083069.07 frames. ], batch size: 16, lr: 5.74e-03, grad_scale: 8.0 +2022-11-16 08:26:34,196 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98543.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:26:34,795 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98544.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:26:35,246 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.413e+02 1.702e+02 2.140e+02 3.638e+02, threshold=3.404e+02, percent-clipped=2.0 +2022-11-16 08:26:35,478 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0164, 1.8933, 1.9295, 1.6781, 1.8443, 2.0019, 1.9431, 1.8305], + device='cuda:3'), covar=tensor([0.0063, 0.0066, 0.0062, 0.0056, 0.0064, 0.0039, 0.0055, 0.0067], + device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0061, 0.0060, 0.0065, 0.0063, 0.0058, 0.0057, 0.0055], + device='cuda:3'), out_proj_covar=tensor([5.8265e-05, 5.3586e-05, 5.2038e-05, 5.7198e-05, 5.5984e-05, 5.0706e-05, + 5.0193e-05, 4.7818e-05], device='cuda:3') +2022-11-16 08:27:07,308 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98592.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:27:09,280 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2519, 3.1844, 2.8441, 3.1878, 3.1827, 2.8255, 2.7745, 2.9843], + device='cuda:3'), covar=tensor([0.0316, 0.0604, 0.1584, 0.0503, 0.0625, 0.0573, 0.1062, 0.0688], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0185, 0.0277, 0.0178, 0.0226, 0.0174, 0.0193, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:27:15,822 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98604.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:27:37,421 INFO [train.py:876] (3/4) Epoch 14, batch 4100, loss[loss=0.08417, simple_loss=0.1284, pruned_loss=0.01998, over 5748.00 frames. ], tot_loss[loss=0.09991, simple_loss=0.1318, pruned_loss=0.03401, over 1081117.48 frames. ], batch size: 20, lr: 5.74e-03, grad_scale: 8.0 +2022-11-16 08:27:42,931 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.009e+01 1.414e+02 1.742e+02 2.183e+02 4.032e+02, threshold=3.484e+02, percent-clipped=2.0 +2022-11-16 08:28:22,855 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2022-11-16 08:28:34,448 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2022-11-16 08:28:39,770 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4838, 3.6589, 3.5273, 3.3438, 1.8705, 3.5977, 2.0786, 3.0908], + device='cuda:3'), covar=tensor([0.0502, 0.0232, 0.0229, 0.0388, 0.0710, 0.0254, 0.0645, 0.0233], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0185, 0.0182, 0.0208, 0.0196, 0.0184, 0.0193, 0.0187], + 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') +2022-11-16 08:28:45,153 INFO [train.py:876] (3/4) Epoch 14, batch 4200, loss[loss=0.07739, simple_loss=0.1099, pruned_loss=0.02243, over 5502.00 frames. ], tot_loss[loss=0.09802, simple_loss=0.1309, pruned_loss=0.03257, over 1084742.02 frames. ], batch size: 12, lr: 5.73e-03, grad_scale: 8.0 +2022-11-16 08:28:50,363 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.341e+02 1.638e+02 2.138e+02 3.541e+02, threshold=3.276e+02, percent-clipped=2.0 +2022-11-16 08:29:01,488 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98762.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:29:04,384 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98766.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:29:13,489 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9505, 4.3343, 3.9051, 4.3384, 4.2723, 3.6780, 3.8093, 3.8478], + device='cuda:3'), covar=tensor([0.0551, 0.0437, 0.1319, 0.0357, 0.0481, 0.0498, 0.0840, 0.0630], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0183, 0.0275, 0.0176, 0.0224, 0.0174, 0.0191, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:29:34,479 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98810.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:29:40,191 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3491, 1.5459, 1.1404, 1.2753, 1.5481, 1.3921, 1.0416, 1.4632], + device='cuda:3'), covar=tensor([0.0079, 0.0054, 0.0078, 0.0081, 0.0066, 0.0055, 0.0097, 0.0074], + device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0062, 0.0061, 0.0066, 0.0065, 0.0060, 0.0058, 0.0056], + device='cuda:3'), out_proj_covar=tensor([5.9934e-05, 5.4559e-05, 5.3096e-05, 5.8450e-05, 5.7375e-05, 5.2124e-05, + 5.1453e-05, 4.9064e-05], device='cuda:3') +2022-11-16 08:29:53,227 INFO [train.py:876] (3/4) Epoch 14, batch 4300, loss[loss=0.1398, simple_loss=0.1513, pruned_loss=0.06415, over 5464.00 frames. ], tot_loss[loss=0.09893, simple_loss=0.1313, pruned_loss=0.03328, over 1078175.46 frames. ], batch size: 58, lr: 5.73e-03, grad_scale: 8.0 +2022-11-16 08:29:53,339 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3374, 4.4531, 2.7305, 4.3024, 3.6930, 2.9509, 2.6374, 3.7722], + device='cuda:3'), covar=tensor([0.2101, 0.0359, 0.1724, 0.0384, 0.0730, 0.1517, 0.2335, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0140, 0.0153, 0.0146, 0.0172, 0.0166, 0.0155, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:29:57,031 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.8748, 2.4297, 2.8859, 3.8166, 3.7393, 2.9140, 2.5279, 3.7516], + device='cuda:3'), covar=tensor([0.0770, 0.2444, 0.2163, 0.1736, 0.1496, 0.2658, 0.2164, 0.1083], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0199, 0.0187, 0.0295, 0.0224, 0.0200, 0.0187, 0.0249], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 08:29:58,787 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.664e+01 1.356e+02 1.673e+02 1.998e+02 3.650e+02, threshold=3.347e+02, percent-clipped=3.0 +2022-11-16 08:29:59,593 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4709, 4.3388, 3.3357, 1.9869, 4.0999, 1.7015, 4.0573, 2.2911], + device='cuda:3'), covar=tensor([0.1350, 0.0139, 0.0583, 0.1799, 0.0185, 0.1789, 0.0228, 0.1527], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0103, 0.0115, 0.0111, 0.0103, 0.0119, 0.0099, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:30:16,716 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7700, 3.4147, 3.6323, 3.3177, 3.7999, 3.6774, 3.5368, 3.7926], + device='cuda:3'), covar=tensor([0.0402, 0.0481, 0.0449, 0.0503, 0.0434, 0.0275, 0.0373, 0.0454], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0156, 0.0109, 0.0146, 0.0187, 0.0115, 0.0131, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:30:28,751 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98890.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:30:32,305 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98895.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:30:34,859 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98899.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:31:00,052 INFO [train.py:876] (3/4) Epoch 14, batch 4400, loss[loss=0.09776, simple_loss=0.128, pruned_loss=0.03373, over 5611.00 frames. ], tot_loss[loss=0.09922, simple_loss=0.1318, pruned_loss=0.0333, over 1081828.48 frames. ], batch size: 38, lr: 5.73e-03, grad_scale: 8.0 +2022-11-16 08:31:05,614 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.027e+01 1.520e+02 1.757e+02 2.071e+02 5.109e+02, threshold=3.514e+02, percent-clipped=4.0 +2022-11-16 08:31:09,745 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98951.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:31:13,487 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98956.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:31:23,418 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98971.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:32:04,699 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99032.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:32:07,916 INFO [train.py:876] (3/4) Epoch 14, batch 4500, loss[loss=0.09273, simple_loss=0.1326, pruned_loss=0.02643, over 5491.00 frames. ], tot_loss[loss=0.09921, simple_loss=0.1316, pruned_loss=0.03339, over 1085486.19 frames. ], batch size: 49, lr: 5.72e-03, grad_scale: 8.0 +2022-11-16 08:32:12,554 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3061, 4.7765, 4.3778, 4.7692, 4.7633, 4.0574, 4.2681, 4.2366], + device='cuda:3'), covar=tensor([0.0334, 0.0475, 0.1376, 0.0410, 0.0472, 0.0454, 0.0684, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0185, 0.0277, 0.0178, 0.0226, 0.0175, 0.0193, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:32:13,104 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.350e+01 1.443e+02 1.649e+02 2.186e+02 4.322e+02, threshold=3.298e+02, percent-clipped=3.0 +2022-11-16 08:32:17,535 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99051.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:32:27,870 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99066.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:32:58,836 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99112.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:33:00,365 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99114.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:33:16,011 INFO [train.py:876] (3/4) Epoch 14, batch 4600, loss[loss=0.08446, simple_loss=0.1256, pruned_loss=0.02166, over 5691.00 frames. ], tot_loss[loss=0.09846, simple_loss=0.1311, pruned_loss=0.03289, over 1091062.53 frames. ], batch size: 19, lr: 5.72e-03, grad_scale: 8.0 +2022-11-16 08:33:21,170 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.692e+01 1.360e+02 1.705e+02 2.390e+02 5.580e+02, threshold=3.409e+02, percent-clipped=5.0 +2022-11-16 08:33:57,238 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99199.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:00,118 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7456, 2.2254, 3.3380, 2.7895, 3.4123, 2.3403, 3.1109, 3.6146], + device='cuda:3'), covar=tensor([0.0815, 0.1847, 0.0910, 0.1682, 0.0729, 0.1711, 0.1432, 0.0999], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0193, 0.0216, 0.0214, 0.0242, 0.0196, 0.0224, 0.0233], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:34:10,360 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7498, 2.2830, 3.3654, 2.8719, 3.3808, 2.3471, 3.1083, 3.6232], + device='cuda:3'), covar=tensor([0.0669, 0.1626, 0.0958, 0.1560, 0.0809, 0.1655, 0.1361, 0.1002], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0194, 0.0216, 0.0214, 0.0243, 0.0196, 0.0225, 0.0234], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:34:12,291 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7036, 1.2729, 1.3754, 1.0482, 1.4155, 1.4755, 0.9538, 1.2354], + device='cuda:3'), covar=tensor([0.0382, 0.0461, 0.0327, 0.0571, 0.0361, 0.0668, 0.0663, 0.0390], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0027, 0.0019, 0.0022, 0.0018, 0.0017, 0.0025, 0.0017], + device='cuda:3'), out_proj_covar=tensor([9.4876e-05, 1.3312e-04, 1.0112e-04, 1.1399e-04, 1.0196e-04, 9.5957e-05, + 1.2494e-04, 9.5509e-05], device='cuda:3') +2022-11-16 08:34:17,416 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99229.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:22,858 INFO [train.py:876] (3/4) Epoch 14, batch 4700, loss[loss=0.0885, simple_loss=0.1206, pruned_loss=0.02818, over 5620.00 frames. ], tot_loss[loss=0.1005, simple_loss=0.1326, pruned_loss=0.03424, over 1086815.02 frames. ], batch size: 29, lr: 5.72e-03, grad_scale: 8.0 +2022-11-16 08:34:28,048 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.829e+01 1.395e+02 1.659e+02 2.125e+02 3.836e+02, threshold=3.317e+02, percent-clipped=3.0 +2022-11-16 08:34:28,803 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99246.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:29,432 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:32,128 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99251.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:43,504 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99268.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:34:58,194 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99290.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:35:15,431 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2022-11-16 08:35:21,411 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:35:23,243 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99327.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:35:24,671 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99329.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:35:29,929 INFO [train.py:876] (3/4) Epoch 14, batch 4800, loss[loss=0.1048, simple_loss=0.1417, pruned_loss=0.03391, over 5672.00 frames. ], tot_loss[loss=0.09857, simple_loss=0.131, pruned_loss=0.03309, over 1086272.93 frames. ], batch size: 19, lr: 5.72e-03, grad_scale: 8.0 +2022-11-16 08:35:31,395 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3036, 2.1045, 2.8568, 1.9612, 1.5859, 3.1194, 2.5449, 2.1597], + device='cuda:3'), covar=tensor([0.1090, 0.1520, 0.0650, 0.2425, 0.2377, 0.0898, 0.1039, 0.1509], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0108, 0.0107, 0.0107, 0.0081, 0.0074, 0.0088, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:35:35,130 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.609e+01 1.410e+02 1.721e+02 2.085e+02 4.264e+02, threshold=3.442e+02, percent-clipped=4.0 +2022-11-16 08:35:44,451 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2022-11-16 08:36:02,040 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99385.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:36:05,815 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([0.7018, 1.1220, 0.8918, 0.7838, 0.9121, 0.9061, 0.5321, 1.0778], + device='cuda:3'), covar=tensor([0.0133, 0.0059, 0.0104, 0.0068, 0.0093, 0.0091, 0.0136, 0.0068], + device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0063, 0.0062, 0.0068, 0.0066, 0.0061, 0.0059, 0.0057], + device='cuda:3'), out_proj_covar=tensor([6.1230e-05, 5.5552e-05, 5.4028e-05, 5.9826e-05, 5.8221e-05, 5.2961e-05, + 5.2376e-05, 5.0066e-05], device='cuda:3') +2022-11-16 08:36:17,344 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99407.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:36:25,862 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99420.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:36:37,295 INFO [train.py:876] (3/4) Epoch 14, batch 4900, loss[loss=0.07999, simple_loss=0.1189, pruned_loss=0.02054, over 5743.00 frames. ], tot_loss[loss=0.09765, simple_loss=0.13, pruned_loss=0.03264, over 1087907.75 frames. ], batch size: 14, lr: 5.71e-03, grad_scale: 8.0 +2022-11-16 08:36:43,021 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.258e+01 1.532e+02 1.838e+02 2.274e+02 5.384e+02, threshold=3.676e+02, percent-clipped=5.0 +2022-11-16 08:36:58,583 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99468.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:37:07,067 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99481.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:37:15,279 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-16 08:37:38,763 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2022-11-16 08:37:39,781 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99529.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:37:44,754 INFO [train.py:876] (3/4) Epoch 14, batch 5000, loss[loss=0.08373, simple_loss=0.1196, pruned_loss=0.02391, over 5492.00 frames. ], tot_loss[loss=0.09872, simple_loss=0.1309, pruned_loss=0.03324, over 1083905.07 frames. ], batch size: 17, lr: 5.71e-03, grad_scale: 8.0 +2022-11-16 08:37:50,173 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.179e+01 1.461e+02 1.855e+02 2.293e+02 4.970e+02, threshold=3.710e+02, percent-clipped=2.0 +2022-11-16 08:37:51,241 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99546.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:37:54,689 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99551.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:37:57,401 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5124, 1.7496, 1.5458, 1.4777, 1.6215, 1.8333, 1.5631, 1.5374], + device='cuda:3'), covar=tensor([0.0072, 0.0062, 0.0066, 0.0074, 0.0067, 0.0058, 0.0065, 0.0104], + device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0064, 0.0063, 0.0069, 0.0067, 0.0062, 0.0060, 0.0058], + device='cuda:3'), out_proj_covar=tensor([6.1766e-05, 5.6152e-05, 5.4709e-05, 6.0767e-05, 5.8723e-05, 5.3564e-05, + 5.2729e-05, 5.0600e-05], device='cuda:3') +2022-11-16 08:38:06,377 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99568.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:09,845 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2022-11-16 08:38:17,358 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99585.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:20,023 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99589.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:23,147 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99594.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:26,737 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99599.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:36,576 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2022-11-16 08:38:40,769 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5797, 1.7713, 2.7028, 2.3119, 1.8734, 2.2777, 2.9406, 2.1531], + device='cuda:3'), covar=tensor([0.0044, 0.0131, 0.0049, 0.0065, 0.0103, 0.0091, 0.0031, 0.0039], + device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0030, 0.0031, 0.0040, 0.0035, 0.0031, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.0741e-05, 2.8462e-05, 2.7604e-05, 3.7878e-05, 3.2378e-05, 2.9579e-05, + 3.6816e-05, 3.5853e-05], device='cuda:3') +2022-11-16 08:38:44,696 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99624.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:46,673 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99627.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:48,063 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99629.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:38:52,865 INFO [train.py:876] (3/4) Epoch 14, batch 5100, loss[loss=0.08453, simple_loss=0.1192, pruned_loss=0.02494, over 5308.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1324, pruned_loss=0.03466, over 1077164.96 frames. ], batch size: 9, lr: 5.71e-03, grad_scale: 8.0 +2022-11-16 08:38:53,062 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.5226, 2.3068, 3.1272, 2.8669, 3.0620, 2.3041, 3.0339, 3.4401], + device='cuda:3'), covar=tensor([0.0870, 0.1440, 0.1182, 0.1476, 0.0968, 0.1447, 0.1115, 0.1069], + device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0194, 0.0216, 0.0213, 0.0242, 0.0197, 0.0224, 0.0233], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:38:58,065 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.215e+01 1.466e+02 1.636e+02 2.038e+02 3.411e+02, threshold=3.271e+02, percent-clipped=0.0 +2022-11-16 08:39:01,521 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99650.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:39:19,180 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99675.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:39:22,554 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:39:22,621 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:39:24,243 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 +2022-11-16 08:39:40,064 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99707.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:40:00,861 INFO [train.py:876] (3/4) Epoch 14, batch 5200, loss[loss=0.1014, simple_loss=0.1363, pruned_loss=0.03327, over 5593.00 frames. ], tot_loss[loss=0.09917, simple_loss=0.1314, pruned_loss=0.03348, over 1081087.12 frames. ], batch size: 54, lr: 5.70e-03, grad_scale: 8.0 +2022-11-16 08:40:03,287 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2022-11-16 08:40:03,564 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99741.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:40:05,955 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.374e+02 1.799e+02 2.301e+02 6.123e+02, threshold=3.597e+02, percent-clipped=6.0 +2022-11-16 08:40:12,568 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99755.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:40:24,581 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0600, 2.6203, 3.6164, 3.1531, 3.7904, 2.6241, 3.3432, 3.9145], + device='cuda:3'), covar=tensor([0.0560, 0.1334, 0.0647, 0.1271, 0.0475, 0.1386, 0.1152, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0192, 0.0214, 0.0210, 0.0240, 0.0195, 0.0223, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:40:26,783 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99776.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:40:30,926 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2022-11-16 08:40:44,434 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5037, 1.4464, 1.6263, 1.4687, 1.7026, 1.8355, 1.6368, 1.7480], + device='cuda:3'), covar=tensor([0.0064, 0.0061, 0.0061, 0.0061, 0.0057, 0.0041, 0.0057, 0.0056], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0064, 0.0063, 0.0068, 0.0067, 0.0061, 0.0059, 0.0058], + device='cuda:3'), out_proj_covar=tensor([6.1897e-05, 5.6352e-05, 5.4748e-05, 6.0393e-05, 5.8705e-05, 5.3287e-05, + 5.2571e-05, 5.0202e-05], device='cuda:3') +2022-11-16 08:40:54,099 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9659, 4.0079, 4.1054, 4.1167, 3.7311, 3.6535, 4.6216, 4.1653], + device='cuda:3'), covar=tensor([0.0542, 0.0993, 0.0460, 0.1285, 0.0592, 0.0416, 0.0763, 0.0720], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0112, 0.0099, 0.0126, 0.0092, 0.0083, 0.0149, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:40:59,388 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99824.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:41:07,963 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2022-11-16 08:41:08,869 INFO [train.py:876] (3/4) Epoch 14, batch 5300, loss[loss=0.1083, simple_loss=0.1337, pruned_loss=0.04141, over 5712.00 frames. ], tot_loss[loss=0.09828, simple_loss=0.1306, pruned_loss=0.03296, over 1081671.94 frames. ], batch size: 19, lr: 5.70e-03, grad_scale: 8.0 +2022-11-16 08:41:14,412 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.304e+01 1.298e+02 1.618e+02 1.971e+02 5.007e+02, threshold=3.235e+02, percent-clipped=2.0 +2022-11-16 08:41:40,970 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99885.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:03,543 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0487, 2.6550, 3.7377, 3.2370, 3.8976, 2.4229, 3.4583, 4.0655], + device='cuda:3'), covar=tensor([0.0537, 0.1439, 0.0731, 0.1318, 0.0442, 0.1696, 0.1167, 0.0666], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0191, 0.0214, 0.0210, 0.0238, 0.0195, 0.0222, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:42:07,322 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:07,370 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:13,192 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99933.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:16,093 INFO [train.py:876] (3/4) Epoch 14, batch 5400, loss[loss=0.08995, simple_loss=0.1288, pruned_loss=0.02555, over 5462.00 frames. ], tot_loss[loss=0.09907, simple_loss=0.1315, pruned_loss=0.0333, over 1082865.90 frames. ], batch size: 10, lr: 5.70e-03, grad_scale: 8.0 +2022-11-16 08:42:21,976 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.513e+01 1.478e+02 1.709e+02 2.137e+02 3.244e+02, threshold=3.418e+02, percent-clipped=1.0 +2022-11-16 08:42:22,087 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99945.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:22,115 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1265, 4.5350, 4.2618, 4.2082, 4.2881, 4.0650, 1.6553, 4.5078], + device='cuda:3'), covar=tensor([0.0469, 0.0344, 0.0386, 0.0481, 0.0502, 0.0568, 0.3702, 0.0440], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0088, 0.0082, 0.0103, 0.0090, 0.0131, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:42:40,285 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99972.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:42:45,536 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99980.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:43:08,153 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100007.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:43:21,948 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100028.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:43:27,261 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100036.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:43:27,828 INFO [train.py:876] (3/4) Epoch 14, batch 5500, loss[loss=0.0948, simple_loss=0.1251, pruned_loss=0.03225, over 5537.00 frames. ], tot_loss[loss=0.1002, simple_loss=0.1321, pruned_loss=0.03409, over 1080282.20 frames. ], batch size: 40, lr: 5.70e-03, grad_scale: 8.0 +2022-11-16 08:43:32,932 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.703e+01 1.432e+02 1.744e+02 2.287e+02 4.720e+02, threshold=3.489e+02, percent-clipped=3.0 +2022-11-16 08:43:45,274 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3532, 1.3631, 1.3346, 1.1165, 1.0932, 1.1789, 1.0228, 0.7944], + device='cuda:3'), covar=tensor([0.0046, 0.0056, 0.0039, 0.0071, 0.0072, 0.0050, 0.0069, 0.0091], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0031, 0.0031, 0.0040, 0.0035, 0.0031, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.1404e-05, 2.8907e-05, 2.7669e-05, 3.8711e-05, 3.2745e-05, 2.9884e-05, + 3.7715e-05, 3.6826e-05], device='cuda:3') +2022-11-16 08:43:49,352 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100068.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:43:54,520 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100076.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:44:27,260 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100124.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:44:27,346 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100124.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:44:35,690 INFO [train.py:876] (3/4) Epoch 14, batch 5600, loss[loss=0.08135, simple_loss=0.1271, pruned_loss=0.01778, over 5743.00 frames. ], tot_loss[loss=0.09881, simple_loss=0.1313, pruned_loss=0.03318, over 1084073.10 frames. ], batch size: 16, lr: 5.69e-03, grad_scale: 8.0 +2022-11-16 08:44:40,973 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.553e+01 1.416e+02 1.691e+02 1.981e+02 4.410e+02, threshold=3.382e+02, percent-clipped=1.0 +2022-11-16 08:44:42,979 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7533, 4.2115, 4.6453, 4.2507, 4.7887, 4.5825, 4.2447, 4.7668], + device='cuda:3'), covar=tensor([0.0364, 0.0418, 0.0363, 0.0378, 0.0348, 0.0281, 0.0350, 0.0353], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0158, 0.0111, 0.0147, 0.0189, 0.0117, 0.0130, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:44:59,163 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4870, 3.1510, 3.6496, 1.9200, 3.3214, 3.6447, 3.6079, 3.9488], + device='cuda:3'), covar=tensor([0.1818, 0.1441, 0.0651, 0.2618, 0.0661, 0.0671, 0.0527, 0.0575], + device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0177, 0.0167, 0.0181, 0.0185, 0.0204, 0.0171, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:44:59,700 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100172.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:45:34,534 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100224.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:45:43,183 INFO [train.py:876] (3/4) Epoch 14, batch 5700, loss[loss=0.1129, simple_loss=0.1476, pruned_loss=0.03915, over 5694.00 frames. ], tot_loss[loss=0.09911, simple_loss=0.1313, pruned_loss=0.03346, over 1083965.56 frames. ], batch size: 36, lr: 5.69e-03, grad_scale: 8.0 +2022-11-16 08:45:43,984 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:45:48,356 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.517e+01 1.364e+02 1.751e+02 2.063e+02 4.628e+02, threshold=3.502e+02, percent-clipped=3.0 +2022-11-16 08:45:48,532 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100245.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:45:53,773 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6435, 2.3074, 3.0463, 1.8010, 1.7229, 3.3060, 2.7268, 2.4761], + device='cuda:3'), covar=tensor([0.1037, 0.1830, 0.0789, 0.2935, 0.4658, 0.0868, 0.1342, 0.1539], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0107, 0.0106, 0.0105, 0.0079, 0.0074, 0.0088, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:45:59,125 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2786, 2.5904, 3.5120, 2.0975, 1.8777, 3.9298, 3.0264, 2.6157], + device='cuda:3'), covar=tensor([0.0727, 0.1094, 0.0442, 0.2502, 0.2077, 0.0412, 0.0859, 0.1011], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0107, 0.0105, 0.0104, 0.0079, 0.0073, 0.0088, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:46:06,336 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100272.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:46:21,250 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100293.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:46:25,376 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100299.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:46:30,593 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9055, 2.2806, 2.1436, 1.4588, 2.3198, 2.2863, 2.2525, 2.5424], + device='cuda:3'), covar=tensor([0.1731, 0.1593, 0.1695, 0.2784, 0.1129, 0.1190, 0.0819, 0.1039], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0177, 0.0167, 0.0180, 0.0183, 0.0202, 0.0170, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:46:50,209 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:46:50,737 INFO [train.py:876] (3/4) Epoch 14, batch 5800, loss[loss=0.08285, simple_loss=0.1249, pruned_loss=0.02037, over 5567.00 frames. ], tot_loss[loss=0.09916, simple_loss=0.1314, pruned_loss=0.03346, over 1083502.71 frames. ], batch size: 15, lr: 5.69e-03, grad_scale: 16.0 +2022-11-16 08:46:51,650 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2022-11-16 08:46:56,198 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.798e+01 1.394e+02 1.721e+02 2.262e+02 4.141e+02, threshold=3.442e+02, percent-clipped=1.0 +2022-11-16 08:47:07,930 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:47:21,653 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100384.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:47:34,798 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100402.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:47:57,688 INFO [train.py:876] (3/4) Epoch 14, batch 5900, loss[loss=0.0654, simple_loss=0.1012, pruned_loss=0.01477, over 5509.00 frames. ], tot_loss[loss=0.09785, simple_loss=0.1304, pruned_loss=0.03266, over 1085983.15 frames. ], batch size: 12, lr: 5.68e-03, grad_scale: 16.0 +2022-11-16 08:48:03,426 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.325e+01 1.325e+02 1.657e+02 2.050e+02 5.165e+02, threshold=3.313e+02, percent-clipped=3.0 +2022-11-16 08:48:16,280 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100463.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:48:17,485 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5652, 4.7726, 3.6326, 2.1652, 4.4668, 1.9088, 4.1338, 2.6795], + device='cuda:3'), covar=tensor([0.1459, 0.0121, 0.0623, 0.2055, 0.0155, 0.1852, 0.0282, 0.1492], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0104, 0.0114, 0.0111, 0.0102, 0.0118, 0.0099, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:48:51,255 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2022-11-16 08:49:01,435 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.2287, 2.7632, 3.6240, 2.4177, 2.0376, 3.9657, 3.0402, 2.7599], + device='cuda:3'), covar=tensor([0.0764, 0.1525, 0.0456, 0.2358, 0.4161, 0.1223, 0.0968, 0.1420], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0105, 0.0105, 0.0104, 0.0079, 0.0072, 0.0087, 0.0095], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:49:05,459 INFO [train.py:876] (3/4) Epoch 14, batch 6000, loss[loss=0.1026, simple_loss=0.136, pruned_loss=0.03461, over 5304.00 frames. ], tot_loss[loss=0.09808, simple_loss=0.1304, pruned_loss=0.0329, over 1087634.03 frames. ], batch size: 79, lr: 5.68e-03, grad_scale: 16.0 +2022-11-16 08:49:05,460 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 08:49:16,992 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6244, 4.5911, 3.6112, 2.3752, 4.4248, 1.8832, 3.9981, 2.8271], + device='cuda:3'), covar=tensor([0.1304, 0.0121, 0.0499, 0.1678, 0.0127, 0.1536, 0.0169, 0.1211], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0103, 0.0113, 0.0110, 0.0101, 0.0117, 0.0098, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:49:23,825 INFO [train.py:908] (3/4) Epoch 14, validation: loss=0.1801, simple_loss=0.1888, pruned_loss=0.08568, over 1530663.00 frames. +2022-11-16 08:49:23,826 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 08:49:28,367 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2022-11-16 08:49:29,378 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.974e+01 1.386e+02 1.665e+02 1.958e+02 3.486e+02, threshold=3.330e+02, percent-clipped=1.0 +2022-11-16 08:50:01,486 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100594.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:50:05,057 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6515, 3.8193, 3.7296, 3.5749, 3.8071, 3.3761, 1.5037, 3.8953], + device='cuda:3'), covar=tensor([0.0622, 0.0565, 0.0560, 0.0611, 0.0547, 0.0944, 0.4175, 0.0541], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0091, 0.0088, 0.0081, 0.0103, 0.0090, 0.0130, 0.0109], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:50:13,569 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2022-11-16 08:50:30,876 INFO [train.py:876] (3/4) Epoch 14, batch 6100, loss[loss=0.1085, simple_loss=0.1446, pruned_loss=0.03615, over 5599.00 frames. ], tot_loss[loss=0.09846, simple_loss=0.1312, pruned_loss=0.03286, over 1087850.98 frames. ], batch size: 23, lr: 5.68e-03, grad_scale: 16.0 +2022-11-16 08:50:36,072 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.272e+01 1.427e+02 1.651e+02 1.977e+02 4.323e+02, threshold=3.302e+02, percent-clipped=4.0 +2022-11-16 08:50:48,648 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100663.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:51:19,448 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.3638, 1.2497, 1.2020, 1.0688, 1.4719, 1.5877, 0.9214, 1.1784], + device='cuda:3'), covar=tensor([0.0572, 0.0739, 0.0507, 0.0854, 0.0622, 0.0343, 0.0768, 0.0366], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0027, 0.0019, 0.0022, 0.0018, 0.0017, 0.0025, 0.0018], + device='cuda:3'), out_proj_covar=tensor([9.5002e-05, 1.3398e-04, 1.0072e-04, 1.1441e-04, 1.0168e-04, 9.6376e-05, + 1.2543e-04, 9.6978e-05], device='cuda:3') +2022-11-16 08:51:20,683 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100711.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:51:34,125 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100730.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:51:38,467 INFO [train.py:876] (3/4) Epoch 14, batch 6200, loss[loss=0.08656, simple_loss=0.1296, pruned_loss=0.02178, over 5717.00 frames. ], tot_loss[loss=0.09804, simple_loss=0.1306, pruned_loss=0.03275, over 1084702.14 frames. ], batch size: 12, lr: 5.68e-03, grad_scale: 16.0 +2022-11-16 08:51:42,598 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8149, 2.8279, 2.6170, 2.9190, 2.3237, 2.4692, 2.4782, 3.4180], + device='cuda:3'), covar=tensor([0.1129, 0.1639, 0.1576, 0.1203, 0.1399, 0.1261, 0.1324, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0111, 0.0108, 0.0111, 0.0097, 0.0107, 0.0100, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:51:43,691 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.597e+01 1.377e+02 1.629e+02 2.100e+02 5.485e+02, threshold=3.258e+02, percent-clipped=3.0 +2022-11-16 08:51:52,265 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100758.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 08:51:57,866 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9882, 2.8948, 2.4788, 2.9469, 2.3610, 2.6163, 2.7825, 3.3710], + device='cuda:3'), covar=tensor([0.0869, 0.1270, 0.1607, 0.0931, 0.1490, 0.1030, 0.1124, 0.1570], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0111, 0.0108, 0.0111, 0.0097, 0.0107, 0.0100, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:52:00,124 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2022-11-16 08:52:13,347 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9240, 2.3908, 2.2397, 1.5939, 2.5544, 2.5571, 2.4442, 2.7110], + device='cuda:3'), covar=tensor([0.1755, 0.1504, 0.1501, 0.2643, 0.0847, 0.1154, 0.0748, 0.1062], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0180, 0.0170, 0.0183, 0.0186, 0.0206, 0.0173, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:52:15,305 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100791.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:52:46,739 INFO [train.py:876] (3/4) Epoch 14, batch 6300, loss[loss=0.09304, simple_loss=0.1162, pruned_loss=0.03493, over 4715.00 frames. ], tot_loss[loss=0.099, simple_loss=0.1311, pruned_loss=0.03344, over 1081670.41 frames. ], batch size: 135, lr: 5.67e-03, grad_scale: 16.0 +2022-11-16 08:52:51,883 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.472e+01 1.427e+02 1.681e+02 2.021e+02 4.003e+02, threshold=3.363e+02, percent-clipped=4.0 +2022-11-16 08:53:02,734 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2022-11-16 08:53:24,998 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100894.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:53:34,813 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9842, 2.5679, 2.5322, 1.7338, 2.6582, 2.7206, 2.6037, 3.0022], + device='cuda:3'), covar=tensor([0.2051, 0.1801, 0.1803, 0.3000, 0.1078, 0.1380, 0.0989, 0.1178], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0181, 0.0170, 0.0184, 0.0187, 0.0207, 0.0174, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:53:50,235 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4398, 2.8971, 4.0392, 3.6940, 4.5068, 3.0010, 4.0389, 4.5179], + device='cuda:3'), covar=tensor([0.0622, 0.1323, 0.0785, 0.1170, 0.0366, 0.1449, 0.1065, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0193, 0.0214, 0.0210, 0.0241, 0.0196, 0.0222, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:53:50,822 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.8012, 2.4381, 3.0446, 2.1974, 1.8843, 3.5489, 2.7433, 2.4819], + device='cuda:3'), covar=tensor([0.0976, 0.1309, 0.0800, 0.2649, 0.1983, 0.1263, 0.1278, 0.1069], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0106, 0.0106, 0.0104, 0.0079, 0.0073, 0.0087, 0.0097], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 08:53:53,676 INFO [train.py:876] (3/4) Epoch 14, batch 6400, loss[loss=0.0853, simple_loss=0.1199, pruned_loss=0.02534, over 5355.00 frames. ], tot_loss[loss=0.09939, simple_loss=0.1321, pruned_loss=0.03333, over 1083608.57 frames. ], batch size: 9, lr: 5.67e-03, grad_scale: 16.0 +2022-11-16 08:53:57,518 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100942.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:53:59,455 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.356e+01 1.445e+02 1.710e+02 2.169e+02 3.491e+02, threshold=3.419e+02, percent-clipped=2.0 +2022-11-16 08:54:14,468 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2022-11-16 08:54:34,380 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2001, 4.0031, 3.0740, 1.8208, 3.7470, 1.6319, 3.6707, 2.0992], + device='cuda:3'), covar=tensor([0.1557, 0.0168, 0.0866, 0.2012, 0.0206, 0.1813, 0.0275, 0.1458], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0104, 0.0114, 0.0111, 0.0102, 0.0119, 0.0101, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:54:42,927 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2022-11-16 08:55:01,410 INFO [train.py:876] (3/4) Epoch 14, batch 6500, loss[loss=0.09517, simple_loss=0.1334, pruned_loss=0.02848, over 5578.00 frames. ], tot_loss[loss=0.09967, simple_loss=0.1322, pruned_loss=0.03355, over 1081313.69 frames. ], batch size: 16, lr: 5.67e-03, grad_scale: 16.0 +2022-11-16 08:55:06,942 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.989e+01 1.437e+02 1.725e+02 2.064e+02 3.698e+02, threshold=3.449e+02, percent-clipped=2.0 +2022-11-16 08:55:07,710 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0819, 3.5856, 2.7172, 1.7616, 3.3810, 1.4223, 3.2910, 1.9226], + device='cuda:3'), covar=tensor([0.1576, 0.0185, 0.0999, 0.1887, 0.0269, 0.2091, 0.0330, 0.1399], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0104, 0.0114, 0.0111, 0.0102, 0.0119, 0.0100, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:55:16,247 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101058.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:55:24,589 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7615, 1.6421, 1.8261, 1.3552, 1.7511, 1.5432, 1.2460, 1.3110], + device='cuda:3'), covar=tensor([0.0039, 0.0055, 0.0052, 0.0071, 0.0071, 0.0073, 0.0061, 0.0083], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0030, 0.0031, 0.0039, 0.0035, 0.0031, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.1250e-05, 2.8356e-05, 2.7713e-05, 3.7733e-05, 3.2661e-05, 2.9665e-05, + 3.7761e-05, 3.6587e-05], device='cuda:3') +2022-11-16 08:55:34,983 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101086.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 08:55:48,583 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101106.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:55:49,647 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101107.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:56:06,187 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.4592, 4.9043, 5.2733, 4.9474, 5.5554, 5.3580, 4.6756, 5.4716], + device='cuda:3'), covar=tensor([0.0358, 0.0362, 0.0450, 0.0293, 0.0286, 0.0250, 0.0305, 0.0269], + device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0161, 0.0113, 0.0150, 0.0192, 0.0118, 0.0133, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 08:56:09,261 INFO [train.py:876] (3/4) Epoch 14, batch 6600, loss[loss=0.1254, simple_loss=0.1468, pruned_loss=0.05198, over 4737.00 frames. ], tot_loss[loss=0.1004, simple_loss=0.1323, pruned_loss=0.03418, over 1081421.56 frames. ], batch size: 136, lr: 5.66e-03, grad_scale: 16.0 +2022-11-16 08:56:10,752 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1246, 1.4711, 1.9575, 1.6840, 1.6561, 1.5829, 1.6668, 1.7238], + device='cuda:3'), covar=tensor([0.0033, 0.0159, 0.0062, 0.0057, 0.0148, 0.0136, 0.0058, 0.0074], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0030, 0.0031, 0.0039, 0.0035, 0.0031, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.1311e-05, 2.8281e-05, 2.7583e-05, 3.7573e-05, 3.2456e-05, 2.9507e-05, + 3.7408e-05, 3.6380e-05], device='cuda:3') +2022-11-16 08:56:14,434 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.470e+01 1.360e+02 1.590e+02 2.159e+02 4.243e+02, threshold=3.180e+02, percent-clipped=1.0 +2022-11-16 08:56:18,129 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9701, 5.1763, 3.9228, 2.6121, 4.8368, 2.6551, 4.8399, 3.0995], + device='cuda:3'), covar=tensor([0.1103, 0.0116, 0.0391, 0.1725, 0.0138, 0.1226, 0.0141, 0.1219], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0104, 0.0114, 0.0110, 0.0102, 0.0118, 0.0100, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 08:56:27,678 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3246, 2.4341, 2.1511, 2.4032, 2.4616, 2.2935, 2.1274, 2.3224], + device='cuda:3'), covar=tensor([0.0486, 0.0783, 0.1777, 0.0706, 0.0708, 0.0577, 0.1370, 0.0700], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0184, 0.0276, 0.0179, 0.0224, 0.0177, 0.0190, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:56:31,082 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101168.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 08:56:54,251 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2022-11-16 08:57:17,779 INFO [train.py:876] (3/4) Epoch 14, batch 6700, loss[loss=0.09835, simple_loss=0.1322, pruned_loss=0.03226, over 5719.00 frames. ], tot_loss[loss=0.09924, simple_loss=0.1315, pruned_loss=0.0335, over 1081846.22 frames. ], batch size: 13, lr: 5.66e-03, grad_scale: 16.0 +2022-11-16 08:57:22,869 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.036e+01 1.360e+02 1.742e+02 2.134e+02 3.328e+02, threshold=3.484e+02, percent-clipped=2.0 +2022-11-16 08:58:25,804 INFO [train.py:876] (3/4) Epoch 14, batch 6800, loss[loss=0.1037, simple_loss=0.135, pruned_loss=0.03617, over 4678.00 frames. ], tot_loss[loss=0.09791, simple_loss=0.1307, pruned_loss=0.03256, over 1083413.08 frames. ], batch size: 135, lr: 5.66e-03, grad_scale: 16.0 +2022-11-16 08:58:27,791 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7926, 3.8140, 3.9977, 3.6676, 3.9112, 3.8452, 1.5117, 3.9361], + device='cuda:3'), covar=tensor([0.0442, 0.0588, 0.0382, 0.0382, 0.0390, 0.0447, 0.3816, 0.0446], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0088, 0.0081, 0.0101, 0.0090, 0.0128, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:58:28,428 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2160, 4.6249, 4.2522, 4.6164, 4.6256, 3.9397, 4.3131, 4.0632], + device='cuda:3'), covar=tensor([0.0392, 0.0423, 0.1103, 0.0413, 0.0383, 0.0497, 0.0752, 0.0649], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0184, 0.0276, 0.0178, 0.0225, 0.0177, 0.0192, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 08:58:30,941 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.618e+01 1.430e+02 1.640e+02 2.057e+02 3.965e+02, threshold=3.281e+02, percent-clipped=2.0 +2022-11-16 08:58:58,650 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101386.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 08:59:30,813 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101434.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 08:59:32,734 INFO [train.py:876] (3/4) Epoch 14, batch 6900, loss[loss=0.1181, simple_loss=0.151, pruned_loss=0.04262, over 5606.00 frames. ], tot_loss[loss=0.09969, simple_loss=0.1318, pruned_loss=0.03378, over 1079037.36 frames. ], batch size: 38, lr: 5.66e-03, grad_scale: 16.0 +2022-11-16 08:59:39,120 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.093e+01 1.364e+02 1.752e+02 2.207e+02 5.252e+02, threshold=3.504e+02, percent-clipped=3.0 +2022-11-16 08:59:50,950 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101463.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:00:40,655 INFO [train.py:876] (3/4) Epoch 14, batch 7000, loss[loss=0.09238, simple_loss=0.1349, pruned_loss=0.02494, over 5563.00 frames. ], tot_loss[loss=0.09921, simple_loss=0.1316, pruned_loss=0.03343, over 1080322.49 frames. ], batch size: 25, lr: 5.65e-03, grad_scale: 16.0 +2022-11-16 09:00:47,043 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.705e+01 1.319e+02 1.556e+02 2.148e+02 4.018e+02, threshold=3.112e+02, percent-clipped=2.0 +2022-11-16 09:00:56,752 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.2202, 4.2288, 3.8911, 3.8119, 4.0778, 4.0447, 1.7781, 4.3787], + device='cuda:3'), covar=tensor([0.0227, 0.0285, 0.0423, 0.0329, 0.0285, 0.0266, 0.3070, 0.0304], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0088, 0.0081, 0.0101, 0.0090, 0.0129, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:01:07,396 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7120, 4.0285, 3.8229, 3.5186, 2.0145, 3.9339, 2.2541, 3.3286], + device='cuda:3'), covar=tensor([0.0440, 0.0153, 0.0175, 0.0337, 0.0693, 0.0166, 0.0617, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0188, 0.0185, 0.0212, 0.0199, 0.0188, 0.0197, 0.0189], + 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') +2022-11-16 09:01:18,068 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9691, 2.6019, 3.6656, 3.3427, 3.7168, 2.3979, 3.2654, 3.9696], + device='cuda:3'), covar=tensor([0.0669, 0.1496, 0.0783, 0.1271, 0.0705, 0.1684, 0.1289, 0.0677], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0196, 0.0217, 0.0213, 0.0243, 0.0200, 0.0227, 0.0234], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:01:45,960 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0864, 2.0591, 2.5562, 1.8139, 1.3799, 2.6969, 2.2219, 2.0411], + device='cuda:3'), covar=tensor([0.1324, 0.1563, 0.1154, 0.2650, 0.2936, 0.1378, 0.1232, 0.1595], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0105, 0.0106, 0.0105, 0.0079, 0.0074, 0.0088, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 09:01:48,509 INFO [train.py:876] (3/4) Epoch 14, batch 7100, loss[loss=0.09551, simple_loss=0.1338, pruned_loss=0.02863, over 5766.00 frames. ], tot_loss[loss=0.09839, simple_loss=0.1309, pruned_loss=0.03296, over 1082791.16 frames. ], batch size: 21, lr: 5.65e-03, grad_scale: 8.0 +2022-11-16 09:01:53,533 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.4399, 4.3261, 4.2872, 3.9380, 4.4529, 4.4101, 2.2390, 4.7178], + device='cuda:3'), covar=tensor([0.0283, 0.0401, 0.0445, 0.0550, 0.0433, 0.0444, 0.2811, 0.0254], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0089, 0.0088, 0.0081, 0.0101, 0.0090, 0.0129, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:01:54,699 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.279e+01 1.412e+02 1.777e+02 2.341e+02 5.678e+02, threshold=3.553e+02, percent-clipped=7.0 +2022-11-16 09:02:01,834 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 +2022-11-16 09:02:12,359 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4912, 2.6117, 2.5240, 2.6174, 2.1923, 1.9905, 2.5149, 2.7961], + device='cuda:3'), covar=tensor([0.1291, 0.1384, 0.1599, 0.1222, 0.1485, 0.1816, 0.1402, 0.1782], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0108, 0.0107, 0.0109, 0.0094, 0.0106, 0.0098, 0.0086], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2022-11-16 09:02:21,015 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101684.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:02:42,813 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7181, 3.7186, 3.8017, 3.5980, 3.7306, 3.6522, 1.5022, 3.8749], + device='cuda:3'), covar=tensor([0.0253, 0.0292, 0.0285, 0.0315, 0.0350, 0.0356, 0.3090, 0.0356], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0089, 0.0088, 0.0081, 0.0101, 0.0090, 0.0128, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:02:49,460 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.7202, 1.6940, 1.7155, 1.2118, 1.4812, 1.6052, 1.2572, 1.0406], + device='cuda:3'), covar=tensor([0.0055, 0.0059, 0.0051, 0.0085, 0.0088, 0.0101, 0.0068, 0.0116], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0030, 0.0031, 0.0039, 0.0035, 0.0031, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.1584e-05, 2.8136e-05, 2.7437e-05, 3.7231e-05, 3.2137e-05, 2.9361e-05, + 3.7469e-05, 3.6717e-05], device='cuda:3') +2022-11-16 09:02:57,247 INFO [train.py:876] (3/4) Epoch 14, batch 7200, loss[loss=0.08512, simple_loss=0.1245, pruned_loss=0.02289, over 5773.00 frames. ], tot_loss[loss=0.1, simple_loss=0.1321, pruned_loss=0.03397, over 1080737.49 frames. ], batch size: 21, lr: 5.65e-03, grad_scale: 8.0 +2022-11-16 09:03:03,310 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101745.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:03:03,778 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.181e+01 1.351e+02 1.693e+02 2.156e+02 4.493e+02, threshold=3.386e+02, percent-clipped=3.0 +2022-11-16 09:03:15,178 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101763.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:03:23,770 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2022-11-16 09:04:28,272 INFO [train.py:876] (3/4) Epoch 15, batch 0, loss[loss=0.0781, simple_loss=0.114, pruned_loss=0.02112, over 4981.00 frames. ], tot_loss[loss=0.0781, simple_loss=0.114, pruned_loss=0.02112, over 4981.00 frames. ], batch size: 7, lr: 5.45e-03, grad_scale: 8.0 +2022-11-16 09:04:28,272 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 09:04:44,421 INFO [train.py:908] (3/4) Epoch 15, validation: loss=0.1798, simple_loss=0.1892, pruned_loss=0.08518, over 1530663.00 frames. +2022-11-16 09:04:44,422 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 09:04:45,740 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101811.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:05:09,640 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.429e+01 1.523e+02 1.865e+02 2.134e+02 5.248e+02, threshold=3.731e+02, percent-clipped=3.0 +2022-11-16 09:05:24,537 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6028, 3.5953, 3.7025, 3.3451, 3.6562, 3.4272, 1.4283, 3.8014], + device='cuda:3'), covar=tensor([0.0315, 0.0410, 0.0332, 0.0400, 0.0353, 0.0507, 0.3442, 0.0313], + device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0089, 0.0088, 0.0081, 0.0102, 0.0090, 0.0129, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:05:49,604 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1502, 2.3769, 2.7618, 2.3334, 1.5580, 2.4460, 1.8045, 2.1572], + device='cuda:3'), covar=tensor([0.0308, 0.0192, 0.0152, 0.0265, 0.0466, 0.0213, 0.0423, 0.0215], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0187, 0.0185, 0.0212, 0.0200, 0.0187, 0.0197, 0.0189], + 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') +2022-11-16 09:05:51,986 INFO [train.py:876] (3/4) Epoch 15, batch 100, loss[loss=0.07659, simple_loss=0.1064, pruned_loss=0.02341, over 5452.00 frames. ], tot_loss[loss=0.09704, simple_loss=0.1307, pruned_loss=0.03171, over 434415.20 frames. ], batch size: 11, lr: 5.45e-03, grad_scale: 8.0 +2022-11-16 09:05:59,281 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101920.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 09:06:16,934 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.051e+01 1.432e+02 1.657e+02 2.152e+02 4.167e+02, threshold=3.314e+02, percent-clipped=2.0 +2022-11-16 09:06:40,796 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101981.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 09:06:44,627 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101987.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:06:47,234 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2022-11-16 09:07:00,219 INFO [train.py:876] (3/4) Epoch 15, batch 200, loss[loss=0.07812, simple_loss=0.1116, pruned_loss=0.02231, over 5492.00 frames. ], tot_loss[loss=0.09568, simple_loss=0.1294, pruned_loss=0.03099, over 694744.84 frames. ], batch size: 10, lr: 5.45e-03, grad_scale: 8.0 +2022-11-16 09:07:20,906 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102040.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:07:24,293 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5512, 4.3846, 3.3933, 1.9613, 4.1542, 2.0013, 4.0700, 2.2953], + device='cuda:3'), covar=tensor([0.1282, 0.0134, 0.0678, 0.1817, 0.0174, 0.1472, 0.0211, 0.1390], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0104, 0.0113, 0.0109, 0.0102, 0.0118, 0.0100, 0.0108], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:07:24,774 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.824e+01 1.391e+02 1.728e+02 2.198e+02 6.457e+02, threshold=3.456e+02, percent-clipped=4.0 +2022-11-16 09:07:26,266 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102048.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:07:41,532 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.7259, 4.5032, 4.7683, 4.4913, 4.3155, 4.1112, 5.0793, 4.7089], + device='cuda:3'), covar=tensor([0.0319, 0.0761, 0.0305, 0.1237, 0.0375, 0.0311, 0.0522, 0.0384], + device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0110, 0.0096, 0.0124, 0.0090, 0.0082, 0.0146, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:07:42,956 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7090, 2.5013, 2.7071, 3.6592, 3.5791, 2.7718, 2.5970, 3.6453], + device='cuda:3'), covar=tensor([0.0906, 0.2628, 0.2310, 0.2499, 0.1534, 0.3069, 0.2067, 0.0986], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0193, 0.0186, 0.0293, 0.0227, 0.0196, 0.0186, 0.0249], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 09:07:44,822 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4651, 2.3533, 2.3138, 2.3403, 2.1342, 1.7854, 2.1908, 2.6849], + device='cuda:3'), covar=tensor([0.1135, 0.1188, 0.1572, 0.1081, 0.1218, 0.1731, 0.1266, 0.0899], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0110, 0.0108, 0.0111, 0.0095, 0.0107, 0.0099, 0.0087], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:08:07,555 INFO [train.py:876] (3/4) Epoch 15, batch 300, loss[loss=0.07868, simple_loss=0.1175, pruned_loss=0.01993, over 5752.00 frames. ], tot_loss[loss=0.09533, simple_loss=0.1282, pruned_loss=0.03125, over 840800.25 frames. ], batch size: 20, lr: 5.45e-03, grad_scale: 8.0 +2022-11-16 09:08:08,357 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.0923, 2.5177, 3.6777, 3.0757, 3.8967, 2.5558, 3.3106, 4.1134], + device='cuda:3'), covar=tensor([0.0473, 0.1656, 0.0835, 0.1708, 0.0526, 0.1715, 0.1397, 0.0685], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0195, 0.0219, 0.0214, 0.0245, 0.0200, 0.0228, 0.0235], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:08:32,100 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.719e+01 1.477e+02 1.712e+02 2.125e+02 3.835e+02, threshold=3.423e+02, percent-clipped=2.0 +2022-11-16 09:09:15,301 INFO [train.py:876] (3/4) Epoch 15, batch 400, loss[loss=0.08987, simple_loss=0.1322, pruned_loss=0.02379, over 5603.00 frames. ], tot_loss[loss=0.09717, simple_loss=0.1297, pruned_loss=0.03231, over 934058.23 frames. ], batch size: 24, lr: 5.44e-03, grad_scale: 8.0 +2022-11-16 09:09:40,375 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.002e+02 1.441e+02 1.640e+02 2.170e+02 4.348e+02, threshold=3.279e+02, percent-clipped=4.0 +2022-11-16 09:10:00,706 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102276.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 09:10:22,876 INFO [train.py:876] (3/4) Epoch 15, batch 500, loss[loss=0.09291, simple_loss=0.1319, pruned_loss=0.02693, over 5567.00 frames. ], tot_loss[loss=0.09681, simple_loss=0.13, pruned_loss=0.0318, over 995496.37 frames. ], batch size: 30, lr: 5.44e-03, grad_scale: 8.0 +2022-11-16 09:10:44,314 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102340.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:10:46,603 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102343.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:10:48,481 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.403e+01 1.330e+02 1.651e+02 2.070e+02 4.075e+02, threshold=3.302e+02, percent-clipped=1.0 +2022-11-16 09:10:50,603 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7655, 2.7282, 2.5636, 2.7342, 2.2693, 2.2110, 2.6012, 3.0816], + device='cuda:3'), covar=tensor([0.1195, 0.1211, 0.1668, 0.2104, 0.1455, 0.1238, 0.1490, 0.1519], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0111, 0.0110, 0.0112, 0.0097, 0.0108, 0.0100, 0.0088], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:11:04,943 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2022-11-16 09:11:10,113 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6132, 3.8091, 3.6608, 3.3930, 1.9864, 3.8105, 2.2757, 3.1952], + device='cuda:3'), covar=tensor([0.0483, 0.0300, 0.0191, 0.0333, 0.0663, 0.0167, 0.0575, 0.0212], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0185, 0.0183, 0.0208, 0.0196, 0.0185, 0.0194, 0.0187], + 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') +2022-11-16 09:11:13,341 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.7066, 4.6144, 3.5953, 2.1383, 4.3815, 2.1579, 4.3908, 2.6761], + device='cuda:3'), covar=tensor([0.1303, 0.0122, 0.0555, 0.1918, 0.0173, 0.1586, 0.0172, 0.1306], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0103, 0.0113, 0.0108, 0.0101, 0.0118, 0.0099, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:11:17,215 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102388.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:11:31,319 INFO [train.py:876] (3/4) Epoch 15, batch 600, loss[loss=0.1282, simple_loss=0.1548, pruned_loss=0.05078, over 5455.00 frames. ], tot_loss[loss=0.09822, simple_loss=0.131, pruned_loss=0.03272, over 1029887.11 frames. ], batch size: 58, lr: 5.44e-03, grad_scale: 8.0 +2022-11-16 09:11:53,797 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2022-11-16 09:11:56,869 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.396e+02 1.798e+02 2.195e+02 6.197e+02, threshold=3.596e+02, percent-clipped=7.0 +2022-11-16 09:12:10,723 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.5008, 1.7978, 1.6019, 1.4733, 1.5978, 1.7671, 1.4518, 1.6110], + device='cuda:3'), covar=tensor([0.0079, 0.0076, 0.0056, 0.0066, 0.0070, 0.0051, 0.0071, 0.0064], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0064, 0.0064, 0.0069, 0.0067, 0.0062, 0.0060, 0.0058], + device='cuda:3'), out_proj_covar=tensor([6.2282e-05, 5.6007e-05, 5.5621e-05, 6.0301e-05, 5.8778e-05, 5.4015e-05, + 5.3046e-05, 5.0600e-05], device='cuda:3') +2022-11-16 09:12:39,186 INFO [train.py:876] (3/4) Epoch 15, batch 700, loss[loss=0.08995, simple_loss=0.1321, pruned_loss=0.02392, over 5689.00 frames. ], tot_loss[loss=0.09651, simple_loss=0.1299, pruned_loss=0.03156, over 1057765.33 frames. ], batch size: 19, lr: 5.44e-03, grad_scale: 8.0 +2022-11-16 09:12:54,562 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6426, 4.5438, 3.5456, 2.0106, 4.2265, 2.0572, 4.2012, 2.5612], + device='cuda:3'), covar=tensor([0.1338, 0.0141, 0.0582, 0.1963, 0.0194, 0.1558, 0.0198, 0.1267], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0103, 0.0112, 0.0109, 0.0101, 0.0117, 0.0099, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:13:04,053 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.068e+01 1.444e+02 1.826e+02 2.249e+02 4.884e+02, threshold=3.652e+02, percent-clipped=1.0 +2022-11-16 09:13:10,954 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5543, 4.4036, 3.4390, 2.0141, 4.0737, 1.7748, 3.9915, 2.3215], + device='cuda:3'), covar=tensor([0.1329, 0.0118, 0.0564, 0.1884, 0.0167, 0.1681, 0.0215, 0.1387], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0103, 0.0112, 0.0109, 0.0101, 0.0118, 0.0099, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:13:24,170 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102576.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 09:13:46,902 INFO [train.py:876] (3/4) Epoch 15, batch 800, loss[loss=0.09552, simple_loss=0.1279, pruned_loss=0.03158, over 5466.00 frames. ], tot_loss[loss=0.09756, simple_loss=0.1303, pruned_loss=0.03241, over 1069839.72 frames. ], batch size: 12, lr: 5.43e-03, grad_scale: 8.0 +2022-11-16 09:13:57,852 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102624.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 09:14:11,163 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102643.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:14:13,462 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.304e+01 1.433e+02 1.767e+02 2.256e+02 4.472e+02, threshold=3.533e+02, percent-clipped=3.0 +2022-11-16 09:14:43,760 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1423, 2.3218, 2.3977, 2.1141, 2.3755, 2.2559, 1.1364, 2.4342], + device='cuda:3'), covar=tensor([0.0402, 0.0389, 0.0359, 0.0421, 0.0378, 0.0451, 0.2835, 0.0430], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0089, 0.0082, 0.0102, 0.0091, 0.0131, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:14:45,027 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102691.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:14:56,844 INFO [train.py:876] (3/4) Epoch 15, batch 900, loss[loss=0.1123, simple_loss=0.1425, pruned_loss=0.04107, over 5614.00 frames. ], tot_loss[loss=0.09857, simple_loss=0.1307, pruned_loss=0.03322, over 1076754.69 frames. ], batch size: 29, lr: 5.43e-03, grad_scale: 8.0 +2022-11-16 09:15:00,252 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2092, 1.5621, 1.8806, 1.3572, 1.8599, 2.0739, 1.4676, 1.8663], + device='cuda:3'), covar=tensor([0.0525, 0.0698, 0.0820, 0.0552, 0.1170, 0.0779, 0.0681, 0.0442], + device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0028, 0.0020, 0.0023, 0.0019, 0.0018, 0.0027, 0.0019], + 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') +2022-11-16 09:15:21,994 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.688e+01 1.409e+02 1.812e+02 2.361e+02 4.444e+02, threshold=3.625e+02, percent-clipped=1.0 +2022-11-16 09:15:42,238 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6783, 4.1462, 3.7729, 3.5589, 2.0858, 4.1057, 2.3413, 3.4189], + device='cuda:3'), covar=tensor([0.0535, 0.0162, 0.0231, 0.0403, 0.0769, 0.0202, 0.0670, 0.0257], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0187, 0.0185, 0.0209, 0.0199, 0.0187, 0.0196, 0.0188], + 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') +2022-11-16 09:16:00,658 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 +2022-11-16 09:16:04,878 INFO [train.py:876] (3/4) Epoch 15, batch 1000, loss[loss=0.0542, simple_loss=0.08806, pruned_loss=0.01017, over 5044.00 frames. ], tot_loss[loss=0.09646, simple_loss=0.1299, pruned_loss=0.03149, over 1082254.22 frames. ], batch size: 7, lr: 5.43e-03, grad_scale: 8.0 +2022-11-16 09:16:21,701 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6427, 2.2963, 3.3165, 2.8078, 3.4106, 2.1789, 3.1214, 3.7627], + device='cuda:3'), covar=tensor([0.0713, 0.1464, 0.0839, 0.1464, 0.0672, 0.1651, 0.1082, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0192, 0.0217, 0.0212, 0.0245, 0.0200, 0.0226, 0.0232], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:16:29,879 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.467e+01 1.525e+02 1.767e+02 2.154e+02 7.760e+02, threshold=3.535e+02, percent-clipped=4.0 +2022-11-16 09:16:33,638 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102851.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:16:43,137 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 +2022-11-16 09:16:57,600 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102887.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:17:12,899 INFO [train.py:876] (3/4) Epoch 15, batch 1100, loss[loss=0.1146, simple_loss=0.143, pruned_loss=0.04312, over 5452.00 frames. ], tot_loss[loss=0.09513, simple_loss=0.1285, pruned_loss=0.0309, over 1085310.00 frames. ], batch size: 58, lr: 5.42e-03, grad_scale: 8.0 +2022-11-16 09:17:14,972 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102912.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:17:26,655 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.71 vs. limit=5.0 +2022-11-16 09:17:29,189 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2022-11-16 09:17:37,835 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.870e+01 1.387e+02 1.709e+02 2.035e+02 6.034e+02, threshold=3.418e+02, percent-clipped=1.0 +2022-11-16 09:17:38,735 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102948.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:18:05,565 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2022-11-16 09:18:17,053 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.8007, 4.6022, 4.9183, 4.7904, 4.5703, 4.2072, 5.2800, 4.8757], + device='cuda:3'), covar=tensor([0.0328, 0.0935, 0.0345, 0.1078, 0.0354, 0.0376, 0.0627, 0.0423], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0114, 0.0100, 0.0128, 0.0093, 0.0085, 0.0151, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:18:19,903 INFO [train.py:876] (3/4) Epoch 15, batch 1200, loss[loss=0.1157, simple_loss=0.1351, pruned_loss=0.04814, over 5342.00 frames. ], tot_loss[loss=0.096, simple_loss=0.1291, pruned_loss=0.03147, over 1088163.67 frames. ], batch size: 70, lr: 5.42e-03, grad_scale: 8.0 +2022-11-16 09:18:45,657 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.196e+01 1.283e+02 1.545e+02 2.048e+02 3.817e+02, threshold=3.089e+02, percent-clipped=2.0 +2022-11-16 09:18:51,274 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2022-11-16 09:19:06,871 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.4623, 4.3896, 2.9526, 4.3323, 3.5146, 3.0452, 2.3419, 3.7425], + device='cuda:3'), covar=tensor([0.1355, 0.0334, 0.1051, 0.0374, 0.0644, 0.0963, 0.1966, 0.0474], + device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0142, 0.0149, 0.0144, 0.0170, 0.0161, 0.0155, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:19:20,296 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103098.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:19:27,356 INFO [train.py:876] (3/4) Epoch 15, batch 1300, loss[loss=0.06769, simple_loss=0.1049, pruned_loss=0.01522, over 5448.00 frames. ], tot_loss[loss=0.09577, simple_loss=0.1288, pruned_loss=0.03137, over 1085681.02 frames. ], batch size: 10, lr: 5.42e-03, grad_scale: 8.0 +2022-11-16 09:19:43,282 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103132.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:19:53,329 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.362e+01 1.342e+02 1.572e+02 1.992e+02 4.136e+02, threshold=3.143e+02, percent-clipped=6.0 +2022-11-16 09:20:01,539 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103159.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:20:13,656 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9768, 1.7358, 1.8539, 1.5380, 1.4183, 1.7898, 1.6750, 1.4949], + device='cuda:3'), covar=tensor([0.0043, 0.0069, 0.0065, 0.0082, 0.0100, 0.0177, 0.0054, 0.0071], + device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0031, 0.0031, 0.0040, 0.0035, 0.0032, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.2057e-05, 2.8444e-05, 2.8192e-05, 3.7751e-05, 3.2571e-05, 3.0305e-05, + 3.7525e-05, 3.6788e-05], device='cuda:3') +2022-11-16 09:20:24,864 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103193.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:20:34,616 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103207.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:20:35,872 INFO [train.py:876] (3/4) Epoch 15, batch 1400, loss[loss=0.0877, simple_loss=0.1285, pruned_loss=0.02343, over 5597.00 frames. ], tot_loss[loss=0.09527, simple_loss=0.1283, pruned_loss=0.03112, over 1080690.97 frames. ], batch size: 23, lr: 5.42e-03, grad_scale: 8.0 +2022-11-16 09:20:59,058 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103243.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:21:01,546 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.466e+01 1.279e+02 1.600e+02 2.005e+02 3.383e+02, threshold=3.199e+02, percent-clipped=1.0 +2022-11-16 09:21:04,710 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2022-11-16 09:21:14,605 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2022-11-16 09:21:35,379 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2022-11-16 09:21:42,722 INFO [train.py:876] (3/4) Epoch 15, batch 1500, loss[loss=0.07606, simple_loss=0.1108, pruned_loss=0.02065, over 5461.00 frames. ], tot_loss[loss=0.09732, simple_loss=0.1303, pruned_loss=0.03218, over 1080836.32 frames. ], batch size: 12, lr: 5.41e-03, grad_scale: 8.0 +2022-11-16 09:21:53,649 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7851, 3.7829, 3.7960, 3.6765, 3.8810, 3.6631, 1.4654, 3.9436], + device='cuda:3'), covar=tensor([0.0282, 0.0368, 0.0367, 0.0327, 0.0291, 0.0378, 0.3413, 0.0315], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0090, 0.0083, 0.0102, 0.0091, 0.0131, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:22:08,704 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.936e+01 1.307e+02 1.656e+02 2.050e+02 4.827e+02, threshold=3.313e+02, percent-clipped=2.0 +2022-11-16 09:22:21,966 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103366.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 09:22:50,187 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.3691, 4.6754, 4.3139, 4.7279, 4.6872, 3.9555, 4.4558, 4.2039], + device='cuda:3'), covar=tensor([0.0300, 0.0445, 0.1179, 0.0330, 0.0369, 0.0526, 0.0550, 0.0462], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0182, 0.0273, 0.0175, 0.0217, 0.0174, 0.0188, 0.0176], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 09:22:51,378 INFO [train.py:876] (3/4) Epoch 15, batch 1600, loss[loss=0.08685, simple_loss=0.1232, pruned_loss=0.02526, over 5808.00 frames. ], tot_loss[loss=0.09605, simple_loss=0.1293, pruned_loss=0.0314, over 1084906.83 frames. ], batch size: 21, lr: 5.41e-03, grad_scale: 8.0 +2022-11-16 09:23:03,574 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103427.0, num_to_drop=1, layers_to_drop={3} +2022-11-16 09:23:06,824 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5501, 2.7066, 3.1609, 4.3012, 4.1864, 3.2270, 2.9481, 4.1772], + device='cuda:3'), covar=tensor([0.0407, 0.2843, 0.2058, 0.1939, 0.0873, 0.2452, 0.1809, 0.0923], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0194, 0.0186, 0.0296, 0.0226, 0.0200, 0.0189, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 09:23:17,105 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.055e+01 1.366e+02 1.646e+02 2.008e+02 3.608e+02, threshold=3.293e+02, percent-clipped=2.0 +2022-11-16 09:23:21,833 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103454.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:23:21,877 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3956, 4.4503, 2.6862, 4.1343, 3.3920, 2.7997, 2.3750, 3.6397], + device='cuda:3'), covar=tensor([0.1454, 0.0279, 0.1363, 0.0367, 0.0879, 0.1266, 0.2032, 0.0494], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0143, 0.0152, 0.0146, 0.0173, 0.0165, 0.0157, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:23:44,576 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103488.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:23:56,688 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2022-11-16 09:23:57,692 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103507.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:23:58,830 INFO [train.py:876] (3/4) Epoch 15, batch 1700, loss[loss=0.1091, simple_loss=0.1307, pruned_loss=0.04377, over 5542.00 frames. ], tot_loss[loss=0.09513, simple_loss=0.1287, pruned_loss=0.03078, over 1089339.11 frames. ], batch size: 13, lr: 5.41e-03, grad_scale: 8.0 +2022-11-16 09:24:06,502 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.28 vs. limit=5.0 +2022-11-16 09:24:09,156 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.0356, 3.3769, 2.3944, 3.0956, 2.4892, 2.5067, 1.9115, 2.8243], + device='cuda:3'), covar=tensor([0.1465, 0.0379, 0.1228, 0.0593, 0.1568, 0.1136, 0.2175, 0.0731], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0143, 0.0151, 0.0145, 0.0172, 0.0165, 0.0157, 0.0157], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:24:21,638 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103543.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:24:24,446 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.609e+01 1.355e+02 1.661e+02 2.024e+02 3.979e+02, threshold=3.323e+02, percent-clipped=4.0 +2022-11-16 09:24:29,793 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2022-11-16 09:24:30,121 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103555.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:24:53,832 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103591.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:25:06,720 INFO [train.py:876] (3/4) Epoch 15, batch 1800, loss[loss=0.06642, simple_loss=0.1031, pruned_loss=0.01486, over 5451.00 frames. ], tot_loss[loss=0.09518, simple_loss=0.1285, pruned_loss=0.03094, over 1087587.23 frames. ], batch size: 10, lr: 5.41e-03, grad_scale: 8.0 +2022-11-16 09:25:31,583 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.004e+01 1.509e+02 1.805e+02 2.366e+02 4.290e+02, threshold=3.611e+02, percent-clipped=5.0 +2022-11-16 09:25:57,445 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9266, 1.2957, 1.6208, 1.2961, 1.8641, 1.9191, 1.1588, 1.5107], + device='cuda:3'), covar=tensor([0.0329, 0.0580, 0.0374, 0.0834, 0.0430, 0.0755, 0.0860, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0027, 0.0019, 0.0022, 0.0019, 0.0017, 0.0026, 0.0018], + device='cuda:3'), out_proj_covar=tensor([9.6566e-05, 1.3500e-04, 1.0270e-04, 1.1671e-04, 1.0424e-04, 9.8043e-05, + 1.2920e-04, 9.9795e-05], device='cuda:3') +2022-11-16 09:26:13,039 INFO [train.py:876] (3/4) Epoch 15, batch 1900, loss[loss=0.07158, simple_loss=0.1178, pruned_loss=0.01267, over 5498.00 frames. ], tot_loss[loss=0.09554, simple_loss=0.1287, pruned_loss=0.03119, over 1087274.13 frames. ], batch size: 17, lr: 5.40e-03, grad_scale: 8.0 +2022-11-16 09:26:22,466 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103722.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 09:26:24,325 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([5.2492, 4.7936, 5.1564, 4.7085, 5.3373, 5.1342, 4.5890, 5.3355], + device='cuda:3'), covar=tensor([0.0310, 0.0329, 0.0330, 0.0297, 0.0296, 0.0193, 0.0242, 0.0226], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0159, 0.0114, 0.0149, 0.0192, 0.0118, 0.0131, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 09:26:34,271 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103740.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:26:38,965 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.053e+01 1.311e+02 1.680e+02 2.035e+02 3.370e+02, threshold=3.360e+02, percent-clipped=0.0 +2022-11-16 09:26:43,680 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103754.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:26:50,731 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:26:54,717 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.9324, 3.0312, 3.1565, 2.8922, 3.1019, 2.8538, 1.3112, 3.1480], + device='cuda:3'), covar=tensor([0.0311, 0.0289, 0.0296, 0.0307, 0.0294, 0.0409, 0.2805, 0.0339], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0090, 0.0090, 0.0083, 0.0103, 0.0091, 0.0132, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:27:06,540 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103788.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:27:15,389 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103801.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:27:15,903 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103802.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:27:20,734 INFO [train.py:876] (3/4) Epoch 15, batch 2000, loss[loss=0.05013, simple_loss=0.08213, pruned_loss=0.009067, over 4610.00 frames. ], tot_loss[loss=0.09735, simple_loss=0.1305, pruned_loss=0.03209, over 1087076.18 frames. ], batch size: 5, lr: 5.40e-03, grad_scale: 8.0 +2022-11-16 09:27:27,460 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5172, 2.5247, 2.3172, 2.5301, 2.2127, 1.8150, 2.2642, 2.7377], + device='cuda:3'), covar=tensor([0.1396, 0.1619, 0.1631, 0.1190, 0.1445, 0.1718, 0.1633, 0.1160], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0110, 0.0108, 0.0111, 0.0096, 0.0107, 0.0100, 0.0087], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:27:32,514 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103826.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:27:39,664 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103836.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:27:47,518 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.204e+01 1.423e+02 1.683e+02 2.188e+02 4.061e+02, threshold=3.366e+02, percent-clipped=3.0 +2022-11-16 09:28:06,739 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103876.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:28:23,490 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.9691, 4.5846, 5.0413, 4.8613, 4.5041, 4.4475, 5.2621, 4.8501], + device='cuda:3'), covar=tensor([0.0323, 0.1077, 0.0322, 0.1231, 0.0459, 0.0279, 0.0576, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0115, 0.0100, 0.0128, 0.0093, 0.0084, 0.0151, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:28:29,290 INFO [train.py:876] (3/4) Epoch 15, batch 2100, loss[loss=0.07187, simple_loss=0.1096, pruned_loss=0.01708, over 5172.00 frames. ], tot_loss[loss=0.0972, simple_loss=0.1301, pruned_loss=0.03213, over 1080697.47 frames. ], batch size: 8, lr: 5.40e-03, grad_scale: 8.0 +2022-11-16 09:28:32,688 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5025, 2.1391, 2.2284, 2.6905, 2.8007, 2.0923, 1.9329, 2.6663], + device='cuda:3'), covar=tensor([0.1801, 0.1724, 0.1560, 0.0843, 0.1064, 0.2587, 0.1858, 0.1451], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0193, 0.0186, 0.0294, 0.0226, 0.0201, 0.0189, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006], + device='cuda:3') +2022-11-16 09:28:33,359 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.1354, 2.3604, 3.5429, 2.9204, 3.9623, 2.3160, 3.3137, 3.9599], + device='cuda:3'), covar=tensor([0.0500, 0.1760, 0.0799, 0.1542, 0.0495, 0.1733, 0.1265, 0.0695], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0190, 0.0214, 0.0208, 0.0239, 0.0196, 0.0224, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:28:43,906 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3315, 2.9671, 3.4735, 1.5043, 3.2002, 3.6036, 3.6250, 3.9518], + device='cuda:3'), covar=tensor([0.1922, 0.1618, 0.0785, 0.3010, 0.0675, 0.0649, 0.0480, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0179, 0.0169, 0.0183, 0.0188, 0.0207, 0.0174, 0.0185], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:28:48,745 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103937.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 09:28:53,561 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.63 vs. limit=5.0 +2022-11-16 09:28:55,839 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.938e+01 1.249e+02 1.692e+02 2.000e+02 3.700e+02, threshold=3.385e+02, percent-clipped=3.0 +2022-11-16 09:28:58,224 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2022-11-16 09:29:24,789 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103990.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:29:27,702 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2022-11-16 09:29:29,408 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103997.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:29:37,376 INFO [train.py:876] (3/4) Epoch 15, batch 2200, loss[loss=0.132, simple_loss=0.1637, pruned_loss=0.05016, over 5266.00 frames. ], tot_loss[loss=0.0963, simple_loss=0.1296, pruned_loss=0.03147, over 1081285.21 frames. ], batch size: 79, lr: 5.40e-03, grad_scale: 8.0 +2022-11-16 09:29:46,237 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104022.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 09:30:04,014 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.673e+01 1.367e+02 1.685e+02 2.131e+02 5.334e+02, threshold=3.371e+02, percent-clipped=2.0 +2022-11-16 09:30:04,869 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104049.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:06,195 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104051.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:10,747 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104058.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:18,833 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104070.0, num_to_drop=1, layers_to_drop={1} +2022-11-16 09:30:36,205 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104096.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:40,434 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.2308, 1.9144, 2.9749, 2.2889, 2.4736, 1.9137, 1.9821, 2.1940], + device='cuda:3'), covar=tensor([0.0562, 0.0468, 0.0198, 0.0340, 0.1253, 0.0743, 0.0376, 0.0198], + device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0026, 0.0019, 0.0022, 0.0018, 0.0017, 0.0025, 0.0018], + device='cuda:3'), out_proj_covar=tensor([9.5098e-05, 1.3328e-04, 1.0142e-04, 1.1436e-04, 1.0216e-04, 9.7095e-05, + 1.2717e-04, 9.7991e-05], device='cuda:3') +2022-11-16 09:30:45,072 INFO [train.py:876] (3/4) Epoch 15, batch 2300, loss[loss=0.07885, simple_loss=0.1233, pruned_loss=0.0172, over 5543.00 frames. ], tot_loss[loss=0.09566, simple_loss=0.1293, pruned_loss=0.03102, over 1086843.66 frames. ], batch size: 16, lr: 5.39e-03, grad_scale: 8.0 +2022-11-16 09:30:45,933 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104110.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:52,941 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104121.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:30:53,934 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2022-11-16 09:31:10,964 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.069e+01 1.436e+02 1.695e+02 2.146e+02 3.837e+02, threshold=3.391e+02, percent-clipped=3.0 +2022-11-16 09:31:19,501 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.87 vs. limit=5.0 +2022-11-16 09:31:52,450 INFO [train.py:876] (3/4) Epoch 15, batch 2400, loss[loss=0.07352, simple_loss=0.113, pruned_loss=0.01704, over 5761.00 frames. ], tot_loss[loss=0.09844, simple_loss=0.131, pruned_loss=0.03292, over 1080386.96 frames. ], batch size: 13, lr: 5.39e-03, grad_scale: 8.0 +2022-11-16 09:31:59,111 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.5055, 2.3399, 2.9465, 2.0716, 1.5272, 3.2741, 2.6939, 2.3667], + device='cuda:3'), covar=tensor([0.0958, 0.1511, 0.0650, 0.2547, 0.3306, 0.1137, 0.0827, 0.1500], + device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0106, 0.0108, 0.0106, 0.0080, 0.0074, 0.0088, 0.0100], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 09:32:08,477 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104232.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 09:32:19,521 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.767e+01 1.368e+02 1.673e+02 2.290e+02 7.384e+02, threshold=3.347e+02, percent-clipped=4.0 +2022-11-16 09:32:23,613 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.3643, 4.3701, 2.8309, 4.0740, 3.3415, 3.1062, 2.4788, 3.6161], + device='cuda:3'), covar=tensor([0.1473, 0.0239, 0.1180, 0.0538, 0.0781, 0.0996, 0.1959, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0142, 0.0151, 0.0145, 0.0171, 0.0163, 0.0156, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:33:00,205 INFO [train.py:876] (3/4) Epoch 15, batch 2500, loss[loss=0.08455, simple_loss=0.1189, pruned_loss=0.0251, over 5567.00 frames. ], tot_loss[loss=0.09698, simple_loss=0.1301, pruned_loss=0.03192, over 1081410.32 frames. ], batch size: 22, lr: 5.39e-03, grad_scale: 8.0 +2022-11-16 09:33:25,645 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104346.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:33:27,457 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.221e+01 1.438e+02 1.846e+02 2.292e+02 4.407e+02, threshold=3.693e+02, percent-clipped=4.0 +2022-11-16 09:33:30,201 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:33:43,986 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6162, 1.7528, 1.5748, 1.5231, 1.7090, 1.7694, 1.4379, 1.8731], + device='cuda:3'), covar=tensor([0.0087, 0.0071, 0.0070, 0.0077, 0.0067, 0.0071, 0.0077, 0.0085], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0064, 0.0064, 0.0068, 0.0067, 0.0062, 0.0060, 0.0059], + device='cuda:3'), out_proj_covar=tensor([6.2190e-05, 5.6400e-05, 5.5731e-05, 5.9778e-05, 5.9486e-05, 5.4118e-05, + 5.3400e-05, 5.1439e-05], device='cuda:3') +2022-11-16 09:33:59,808 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104396.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:34:06,227 INFO [zipformer.py:623] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104405.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:34:08,817 INFO [train.py:876] (3/4) Epoch 15, batch 2600, loss[loss=0.04878, simple_loss=0.07994, pruned_loss=0.008806, over 4077.00 frames. ], tot_loss[loss=0.09588, simple_loss=0.1294, pruned_loss=0.03117, over 1082552.20 frames. ], batch size: 4, lr: 5.39e-03, grad_scale: 8.0 +2022-11-16 09:34:12,852 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.6689, 1.7560, 1.9828, 1.9442, 1.7211, 1.4641, 1.7473, 1.7879], + device='cuda:3'), covar=tensor([0.2353, 0.2410, 0.1914, 0.1539, 0.1869, 0.3154, 0.1976, 0.1387], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0111, 0.0110, 0.0113, 0.0097, 0.0108, 0.0101, 0.0089], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:34:16,660 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104421.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:34:21,602 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.9327, 2.3239, 2.0740, 1.5345, 2.4539, 2.5777, 2.5924, 2.6074], + device='cuda:3'), covar=tensor([0.1655, 0.1542, 0.1566, 0.2588, 0.0875, 0.1215, 0.0740, 0.1089], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0178, 0.0166, 0.0182, 0.0188, 0.0205, 0.0175, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:34:32,573 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104444.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:34:35,747 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.383e+01 1.354e+02 1.630e+02 1.866e+02 3.463e+02, threshold=3.260e+02, percent-clipped=0.0 +2022-11-16 09:34:49,479 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104469.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:35:16,362 INFO [train.py:876] (3/4) Epoch 15, batch 2700, loss[loss=0.08429, simple_loss=0.1164, pruned_loss=0.02609, over 5286.00 frames. ], tot_loss[loss=0.09622, simple_loss=0.1294, pruned_loss=0.03152, over 1080901.25 frames. ], batch size: 79, lr: 5.38e-03, grad_scale: 8.0 +2022-11-16 09:35:21,656 INFO [scaling.py:664] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 +2022-11-16 09:35:31,693 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104532.0, num_to_drop=1, layers_to_drop={0} +2022-11-16 09:35:42,732 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.521e+01 1.320e+02 1.626e+02 1.993e+02 3.646e+02, threshold=3.252e+02, percent-clipped=1.0 +2022-11-16 09:36:04,050 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104580.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:36:24,511 INFO [train.py:876] (3/4) Epoch 15, batch 2800, loss[loss=0.09737, simple_loss=0.1244, pruned_loss=0.03516, over 5697.00 frames. ], tot_loss[loss=0.09593, simple_loss=0.1287, pruned_loss=0.03156, over 1078520.23 frames. ], batch size: 34, lr: 5.38e-03, grad_scale: 8.0 +2022-11-16 09:36:49,227 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104646.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:36:51,003 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.819e+01 1.524e+02 1.785e+02 2.265e+02 4.174e+02, threshold=3.570e+02, percent-clipped=2.0 +2022-11-16 09:36:53,539 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.4811, 2.8877, 3.6014, 2.2346, 2.3998, 3.5035, 3.0859, 2.8930], + device='cuda:3'), covar=tensor([0.0508, 0.1128, 0.0478, 0.2600, 0.1279, 0.1800, 0.0562, 0.0968], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0106, 0.0107, 0.0106, 0.0079, 0.0074, 0.0088, 0.0098], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2022-11-16 09:36:54,191 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104653.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:37:21,819 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104694.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:37:26,311 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104701.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:37:29,418 INFO [zipformer.py:623] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104705.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:37:32,267 INFO [train.py:876] (3/4) Epoch 15, batch 2900, loss[loss=0.08839, simple_loss=0.1262, pruned_loss=0.02529, over 5576.00 frames. ], tot_loss[loss=0.09508, simple_loss=0.1281, pruned_loss=0.03103, over 1080508.68 frames. ], batch size: 25, lr: 5.38e-03, grad_scale: 8.0 +2022-11-16 09:37:59,214 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.093e+01 1.400e+02 1.739e+02 2.219e+02 5.401e+02, threshold=3.478e+02, percent-clipped=7.0 +2022-11-16 09:38:02,045 INFO [zipformer.py:623] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104753.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:38:40,269 INFO [train.py:876] (3/4) Epoch 15, batch 3000, loss[loss=0.1098, simple_loss=0.1444, pruned_loss=0.03761, over 5643.00 frames. ], tot_loss[loss=0.09421, simple_loss=0.1279, pruned_loss=0.03028, over 1084848.33 frames. ], batch size: 29, lr: 5.38e-03, grad_scale: 8.0 +2022-11-16 09:38:40,269 INFO [train.py:899] (3/4) Computing validation loss +2022-11-16 09:38:58,034 INFO [train.py:908] (3/4) Epoch 15, validation: loss=0.1809, simple_loss=0.1888, pruned_loss=0.08654, over 1530663.00 frames. +2022-11-16 09:38:58,035 INFO [train.py:909] (3/4) Maximum memory allocated so far is 4742MB +2022-11-16 09:38:58,757 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.7769, 4.2531, 3.8233, 4.2488, 4.2248, 3.6781, 3.7786, 3.7023], + device='cuda:3'), covar=tensor([0.0624, 0.0423, 0.1354, 0.0411, 0.0431, 0.0536, 0.0881, 0.0664], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0185, 0.0278, 0.0180, 0.0218, 0.0178, 0.0191, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:39:25,316 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.210e+01 1.336e+02 1.693e+02 2.076e+02 4.663e+02, threshold=3.385e+02, percent-clipped=3.0 +2022-11-16 09:39:29,566 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2022-11-16 09:40:06,260 INFO [train.py:876] (3/4) Epoch 15, batch 3100, loss[loss=0.1105, simple_loss=0.1467, pruned_loss=0.03713, over 5555.00 frames. ], tot_loss[loss=0.09629, simple_loss=0.1289, pruned_loss=0.03182, over 1077537.26 frames. ], batch size: 43, lr: 5.37e-03, grad_scale: 8.0 +2022-11-16 09:40:31,320 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([4.5050, 5.0805, 4.5360, 5.2386, 5.0431, 4.3653, 4.7347, 4.4278], + device='cuda:3'), covar=tensor([0.0298, 0.0621, 0.1736, 0.0297, 0.0533, 0.0542, 0.0686, 0.0747], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0186, 0.0279, 0.0180, 0.0219, 0.0178, 0.0192, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:40:33,542 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.354e+01 1.349e+02 1.627e+02 2.090e+02 3.507e+02, threshold=3.255e+02, percent-clipped=1.0 +2022-11-16 09:40:49,174 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.9088, 2.6471, 3.5389, 3.0494, 3.6389, 2.4059, 3.2674, 3.8331], + device='cuda:3'), covar=tensor([0.0628, 0.1362, 0.0986, 0.1571, 0.1051, 0.1750, 0.1287, 0.0741], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0193, 0.0220, 0.0214, 0.0244, 0.0199, 0.0228, 0.0234], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2022-11-16 09:41:18,935 INFO [train.py:876] (3/4) Epoch 15, batch 3200, loss[loss=0.1196, simple_loss=0.1527, pruned_loss=0.04325, over 5578.00 frames. ], tot_loss[loss=0.09651, simple_loss=0.1294, pruned_loss=0.03183, over 1087063.45 frames. ], batch size: 43, lr: 5.37e-03, grad_scale: 8.0 +2022-11-16 09:41:46,947 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.649e+01 1.417e+02 1.645e+02 2.110e+02 3.664e+02, threshold=3.291e+02, percent-clipped=2.0 +2022-11-16 09:42:28,178 INFO [train.py:876] (3/4) Epoch 15, batch 3300, loss[loss=0.1001, simple_loss=0.1295, pruned_loss=0.03533, over 5605.00 frames. ], tot_loss[loss=0.09711, simple_loss=0.1299, pruned_loss=0.03217, over 1088034.56 frames. ], batch size: 18, lr: 5.37e-03, grad_scale: 8.0 +2022-11-16 09:42:56,051 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.723e+01 1.318e+02 1.706e+02 2.251e+02 4.750e+02, threshold=3.412e+02, percent-clipped=5.0 +2022-11-16 09:42:58,247 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([3.6473, 3.3210, 3.4453, 3.2383, 3.7054, 3.5158, 3.4382, 3.6650], + device='cuda:3'), covar=tensor([0.0418, 0.0468, 0.0575, 0.0447, 0.0415, 0.0283, 0.0409, 0.0471], + device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0161, 0.0116, 0.0151, 0.0196, 0.0120, 0.0133, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2022-11-16 09:43:02,575 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([1.8213, 1.7264, 1.7471, 1.7205, 1.8687, 1.7672, 1.9543, 1.9179], + device='cuda:3'), covar=tensor([0.0731, 0.1026, 0.0856, 0.1339, 0.0672, 0.0637, 0.1199, 0.0968], + device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0114, 0.0099, 0.0126, 0.0092, 0.0084, 0.0148, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2022-11-16 09:43:37,373 INFO [train.py:876] (3/4) Epoch 15, batch 3400, loss[loss=0.09587, simple_loss=0.1329, pruned_loss=0.02942, over 5757.00 frames. ], tot_loss[loss=0.09525, simple_loss=0.1283, pruned_loss=0.0311, over 1085975.23 frames. ], batch size: 20, lr: 5.37e-03, grad_scale: 8.0 +2022-11-16 09:44:05,013 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.327e+01 1.478e+02 1.801e+02 2.310e+02 1.078e+03, threshold=3.602e+02, percent-clipped=8.0 +2022-11-16 09:44:24,687 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2022-11-16 09:44:46,631 INFO [train.py:876] (3/4) Epoch 15, batch 3500, loss[loss=0.07064, simple_loss=0.0957, pruned_loss=0.02279, over 4813.00 frames. ], tot_loss[loss=0.09556, simple_loss=0.1285, pruned_loss=0.0313, over 1087833.97 frames. ], batch size: 5, lr: 5.36e-03, grad_scale: 8.0 +2022-11-16 09:45:15,055 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.540e+01 1.261e+02 1.523e+02 1.855e+02 3.568e+02, threshold=3.045e+02, percent-clipped=0.0 +2022-11-16 09:45:17,381 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.6988, 3.7493, 3.6889, 3.2965, 1.9429, 3.7194, 2.2787, 3.1844], + device='cuda:3'), covar=tensor([0.0467, 0.0221, 0.0183, 0.0486, 0.0783, 0.0197, 0.0642, 0.0209], + device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0191, 0.0185, 0.0213, 0.0202, 0.0190, 0.0199, 0.0192], + 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') +2022-11-16 09:45:42,582 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105388.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:45:57,219 INFO [train.py:876] (3/4) Epoch 15, batch 3600, loss[loss=0.08764, simple_loss=0.1341, pruned_loss=0.02057, over 5616.00 frames. ], tot_loss[loss=0.09588, simple_loss=0.1289, pruned_loss=0.03145, over 1088593.72 frames. ], batch size: 18, lr: 5.36e-03, grad_scale: 8.0 +2022-11-16 09:45:58,791 INFO [zipformer.py:1411] (3/4) attn_weights_entropy = tensor([2.1060, 1.6888, 2.1081, 1.9879, 1.4215, 2.2665, 2.0423, 1.5697], + device='cuda:3'), covar=tensor([0.0035, 0.0068, 0.0074, 0.0063, 0.0121, 0.0065, 0.0052, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0030, 0.0032, 0.0040, 0.0035, 0.0032, 0.0039, 0.0038], + device='cuda:3'), out_proj_covar=tensor([3.1568e-05, 2.8109e-05, 2.8267e-05, 3.7779e-05, 3.2686e-05, 3.0189e-05, + 3.7198e-05, 3.6194e-05], device='cuda:3') +2022-11-16 09:46:25,418 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.909e+01 1.343e+02 1.539e+02 1.882e+02 3.139e+02, threshold=3.078e+02, percent-clipped=1.0 +2022-11-16 09:46:26,016 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105449.0, num_to_drop=1, layers_to_drop={2} +2022-11-16 09:46:27,724 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2022-11-16 09:46:52,069 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105486.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:46:54,887 INFO [scaling.py:664] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2022-11-16 09:47:03,224 INFO [zipformer.py:623] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105502.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:47:07,845 INFO [train.py:876] (3/4) Epoch 15, batch 3700, loss[loss=0.1119, simple_loss=0.1371, pruned_loss=0.04332, over 5380.00 frames. ], tot_loss[loss=0.09755, simple_loss=0.1299, pruned_loss=0.03258, over 1080046.37 frames. ], batch size: 70, lr: 5.36e-03, grad_scale: 8.0 +2022-11-16 09:47:34,339 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105547.0, num_to_drop=0, layers_to_drop=set() +2022-11-16 09:47:35,430 INFO [optim.py:343] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.168e+01 1.329e+02 1.592e+02 1.987e+02 4.072e+02, threshold=3.183e+02, percent-clipped=1.0 +2022-11-16 09:47:45,379 INFO [zipformer.py:623] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105563.0, num_to_drop=0, layers_to_drop=set()