w2v2-base-pretrained_lr5e-5_at0.8_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: 2.3273
• Wer: 0.1747

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
18.3928	5.43	250	4.5566	1.0
3.4003	10.87	500	3.2320	1.0
2.9641	16.3	750	2.5134	1.0
0.9965	21.74	1000	1.2335	0.4165
0.2103	27.17	1250	1.5883	0.2012
0.1252	32.61	1500	1.5291	0.1837
0.0911	38.04	1750	1.8433	0.1901
0.0717	43.48	2000	1.9624	0.1858
0.0609	48.91	2250	1.9417	0.1781
0.0488	54.35	2500	2.0794	0.1751
0.041	59.78	2750	2.2785	0.1837
0.0357	65.22	3000	2.1884	0.1709
0.0325	70.65	3250	2.2440	0.1764
0.0286	76.09	3500	2.2743	0.1764
0.0263	81.52	3750	2.2614	0.1730
0.0256	86.96	4000	2.3273	0.1747

Framework versions

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