w2v2-base-pretrained_lr5e-5_at0.0_da1

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2935
• Wer: 0.1709

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
22.479	3.91	250	4.5097	1.0
3.5142	7.81	500	3.2115	1.0
3.144	11.72	750	3.1091	1.0
2.5815	15.62	1000	1.2241	0.9991
0.6479	19.53	1250	0.5868	0.3328
0.3255	23.44	1500	0.7123	0.2050
0.2136	27.34	1750	0.8753	0.1854
0.1561	31.25	2000	0.9095	0.1892
0.1195	35.16	2250	1.0824	0.1828
0.0966	39.06	2500	1.0976	0.1756
0.0829	42.97	2750	1.1946	0.1734
0.0724	46.88	3000	1.2161	0.1713
0.0611	50.78	3250	1.2877	0.1739
0.0555	54.69	3500	1.3169	0.1687
0.0537	58.59	3750	1.2744	0.1764
0.0481	62.5	4000	1.2935	0.1709

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.0
• Datasets 2.14.6
• Tokenizers 0.14.1