cmb-20s_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: 0.2887
• Per: 0.1279
• Pcc: 0.6501
• Ctc Loss: 0.3968
• Mse Loss: 1.0164

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: 2222
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 8928
• training_steps: 89280
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
11.516	3.0	8928	4.4204	0.9956	0.6302	3.7718	0.8695
2.6821	6.0	17856	1.4363	0.1800	0.6754	0.6151	0.8033
1.0205	9.0	26784	1.4608	0.1457	0.6734	0.4819	0.9911
0.5238	12.0	35712	1.1546	0.1375	0.6624	0.4382	0.8751
0.0849	15.0	44640	1.2248	0.1338	0.6576	0.4181	1.0163
-0.3644	18.0	53568	1.4907	0.1314	0.6513	0.4077	1.2464
-0.8166	21.0	62496	0.5897	0.1302	0.6518	0.4054	0.8912
-1.2642	24.0	71424	0.5638	0.1286	0.6511	0.3985	0.9995
-1.6765	27.0	80352	0.4494	0.1282	0.6490	0.3980	1.0442
-1.9395	30.0	89280	0.2887	0.1279	0.6501	0.3968	1.0164

Framework versions

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