w2v2-base-pretrained_lr5e-5_at0.2_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: 1.3038
• 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
19.5617	3.91	250	4.1984	1.0
3.3686	7.81	500	3.2319	1.0
3.1228	11.72	750	3.1341	1.0
2.9603	15.62	1000	2.3654	1.0
1.0738	19.53	1250	0.7578	0.5485
0.3549	23.44	1500	0.6579	0.2337
0.2219	27.34	1750	0.8304	0.1999
0.1579	31.25	2000	0.9428	0.1828
0.1216	35.16	2250	1.0046	0.1747
0.0958	39.06	2500	1.0114	0.1751
0.076	42.97	2750	1.2645	0.1768
0.0662	46.88	3000	1.2588	0.1739
0.0569	50.78	3250	1.3057	0.1730
0.0514	54.69	3500	1.2869	0.1696
0.0479	58.59	3750	1.2697	0.1704
0.0451	62.5	4000	1.3038	0.1709

Framework versions

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