wav2vec2-1b-E30_fs2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5468
• Cer: 14.2798

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
10.8739	0.2580	200	2.7757	55.2690
1.817	0.5160	400	1.5611	35.3031
1.2242	0.7741	600	1.2254	29.7404
1.0296	1.0321	800	1.0612	26.5508
0.8143	1.2901	1000	1.0154	24.8179
0.7274	1.5481	1200	0.9467	24.0895
0.6812	1.8062	1400	0.8809	23.0204
0.5998	2.0642	1600	1.0420	26.1337
0.495	2.3222	1800	0.8705	22.3978
0.4552	2.5802	2000	0.8161	20.8412
0.4141	2.8383	2200	0.7798	20.8999
0.3507	3.0963	2400	0.7629	19.4901
0.2985	3.3543	2600	0.6205	16.3651
0.2554	3.6123	2800	0.5922	15.8952
0.2546	3.8703	3000	0.5993	16.0479
0.2114	4.1284	3200	0.5720	14.8849
0.1834	4.3864	3400	0.5807	15.2491
0.1723	4.6444	3600	0.5454	14.1800
0.1631	4.9024	3800	0.5468	14.2798

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1