xlsr-clp-nmcpc

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6379
• Wer: 0.3255

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: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 132
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.3195	3.125	200	3.1230	1.0
3.1989	6.25	400	2.9782	0.9915
3.0595	9.375	600	2.7710	1.0
2.7451	12.5	800	2.2290	0.9957
2.1191	15.625	1000	1.3855	0.9723
1.3034	18.75	1200	0.8246	0.6830
0.7783	21.875	1400	0.5309	0.5319
0.4741	25.0	1600	0.5412	0.4723
0.331	28.125	1800	0.4894	0.4234
0.2486	31.25	2000	0.5627	0.4213
0.2043	34.375	2200	0.5591	0.3830
0.16	37.5	2400	0.5430	0.3894
0.1479	40.625	2600	0.5401	0.3681
0.1154	43.75	2800	0.5947	0.3830
0.1125	46.875	3000	0.5841	0.3617
0.0929	50.0	3200	0.5909	0.3574
0.0801	53.125	3400	0.5169	0.3532
0.077	56.25	3600	0.5810	0.3532
0.0655	59.375	3800	0.6263	0.3553
0.064	62.5	4000	0.6379	0.3553
0.0533	65.625	4200	0.5857	0.3553
0.0481	68.75	4400	0.6265	0.3383
0.0417	71.875	4600	0.6352	0.3489
0.0354	75.0	4800	0.6084	0.3340
0.0355	78.125	5000	0.6516	0.3468
0.0326	81.25	5200	0.6279	0.3362
0.0261	84.375	5400	0.6127	0.3298
0.0267	87.5	5600	0.6371	0.3340
0.0225	90.625	5800	0.6300	0.3298
0.0241	93.75	6000	0.6390	0.3277
0.0196	96.875	6200	0.6432	0.3255
0.0171	100.0	6400	0.6379	0.3255

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.0
• Datasets 3.0.0
• Tokenizers 0.19.1