w2v2-base-pretrained_lr1e-4_at0.8_da0.9

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: 2.2030
• Wer: 0.1756

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: 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: 3500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
15.3627	6.1	250	3.8053	1.0
3.1688	12.2	500	3.1948	1.0
2.7881	18.29	750	1.6319	1.0047
0.3364	24.39	1000	0.9451	0.2704
0.1193	30.49	1250	1.3362	0.2089
0.0741	36.59	1500	1.8263	0.1914
0.0511	42.68	1750	1.8832	0.1747
0.0369	48.78	2000	1.9675	0.1794
0.0256	54.88	2250	2.2933	0.1820
0.0194	60.98	2500	2.3546	0.1875
0.0151	67.07	2750	2.3144	0.1773
0.0126	73.17	3000	2.1852	0.1713
0.0104	79.27	3250	2.2503	0.1756
0.0095	85.37	3500	2.2030	0.1756

Framework versions

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