w2v2-base-pretrained_lr5e-5_at0.3_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: 1.4352
• Wer: 0.1704

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
19.6725	3.97	250	4.9409	1.0
3.5043	7.94	500	3.2414	1.0
3.1162	11.9	750	3.1351	1.0
2.2138	15.87	1000	0.9146	0.9184
0.5308	19.84	1250	0.6292	0.4357
0.2762	23.81	1500	0.8713	0.2384
0.1894	27.78	1750	0.9537	0.1905
0.1339	31.75	2000	1.2355	0.1824
0.1002	35.71	2250	1.2193	0.1739
0.0858	39.68	2500	1.1557	0.1709
0.0711	43.65	2750	1.3591	0.1692
0.0589	47.62	3000	1.3372	0.1683
0.0525	51.59	3250	1.4133	0.1683
0.0464	55.56	3500	1.4969	0.1679
0.0436	59.52	3750	1.4262	0.1674
0.0401	63.49	4000	1.4352	0.1704

Framework versions

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