k2e-20s_asr-scr_w2v2-large_001

This model is a fine-tuned version of facebook/wav2vec2-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.7509
• Per: 0.1453
• Pcc: 0.6324
• Ctc Loss: 0.5111
• Mse Loss: 1.2263

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 1
• seed: 1111
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2235
• training_steps: 22350
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
14.3265	3.0	2235	4.6461	0.9890	0.6160	3.6176	1.0856
3.4446	6.01	4470	2.8757	0.3498	0.6516	1.0729	1.7895
1.5549	9.01	6705	2.0517	0.1846	0.6507	0.6672	1.3503
1.1194	12.02	8940	1.8625	0.1650	0.6443	0.5914	1.2388
0.9031	15.02	11175	1.8899	0.1554	0.6312	0.5473	1.3054
0.7614	18.02	13410	1.5491	0.1522	0.6307	0.5349	1.0200
0.6483	21.03	15645	1.8357	0.1481	0.6304	0.5215	1.2852
0.5561	24.03	17880	2.0576	0.1468	0.6263	0.5191	1.4774
0.5108	27.04	20115	1.8949	0.1452	0.6351	0.5090	1.3489
0.4778	30.04	22350	1.7509	0.1453	0.6324	0.5111	1.2263

Framework versions

• Transformers 4.38.1
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.2