las_asr-scr_w2v2-base_002

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: 1.4121
• Per: 0.1605
• Pcc: 0.7256
• Ctc Loss: 0.4686
• Mse Loss: 1.0317

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: 16
• eval_batch_size: 1
• seed: 2222
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 742
• training_steps: 7420
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
12.6219	1.0	742	5.8749	0.9897	0.5612	3.7480	2.3327
3.9885	2.0	1484	2.0871	0.3274	0.6986	1.2456	0.8803
1.7464	3.0	2226	1.5832	0.1972	0.7004	0.6288	0.9002
1.2583	4.0	2968	1.8791	0.1858	0.7250	0.5608	1.2261
0.9058	5.0	3710	1.9793	0.1736	0.7355	0.5207	1.3244
0.5912	6.0	4452	1.4470	0.1712	0.7297	0.4956	0.9576
0.2949	7.0	5194	1.3030	0.1670	0.7310	0.4880	0.8993
0.0454	8.0	5936	2.0771	0.1645	0.7234	0.4762	1.4066
-0.1676	9.0	6678	1.3317	0.1616	0.7278	0.4743	0.9749
-0.2955	10.0	7420	1.4121	0.1605	0.7256	0.4686	1.0317

Framework versions

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