wav2vec2-4

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: 3.1442
• Wer: 1.0

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: 400
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.1303	1.37	200	3.2783	1.0
2.8798	2.74	400	3.1233	1.0
2.8586	4.11	600	3.1612	1.0
2.8613	5.48	800	3.1354	1.0
2.8588	6.85	1000	3.2634	1.0
2.8572	8.22	1200	3.0905	1.0
2.8573	9.59	1400	3.2315	1.0
2.8532	10.96	1600	3.0999	1.0
2.8567	12.33	1800	3.1496	1.0
2.8556	13.7	2000	3.1081	1.0
2.8551	15.07	2200	3.1139	1.0
2.8545	16.44	2400	3.1621	1.0
2.8547	17.81	2600	3.1124	1.0
2.8551	19.18	2800	3.1612	1.0
2.854	20.55	3000	3.1052	1.0
2.8542	21.92	3200	3.1558	1.0
2.8544	23.29	3400	3.1370	1.0
2.8546	24.66	3600	3.1616	1.0
2.8563	26.03	3800	3.1366	1.0
2.8514	27.4	4000	3.1434	1.0
2.8543	28.77	4200	3.1442	1.0

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1