wav2vec2-E30_pause

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0562
• Cer: 22.1393

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: 8
• 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: 50
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
30.6015	0.1289	200	5.2403	100.0
4.9516	0.2579	400	4.7381	100.0
4.8148	0.3868	600	4.6941	100.0
4.7459	0.5158	800	4.5804	100.0
4.7075	0.6447	1000	4.5734	100.0
4.6607	0.7737	1200	4.5106	100.0
4.3823	0.9026	1400	4.0717	96.7340
3.3809	1.0316	1600	3.0736	55.4746
2.7624	1.1605	1800	2.5641	45.3654
2.4	1.2895	2000	2.3126	43.4269
2.1103	1.4184	2200	2.0157	38.2343
1.9096	1.5474	2400	1.8777	35.6086
1.7178	1.6763	2600	1.7423	34.2164
1.5487	1.8053	2800	1.5431	30.9504
1.4451	1.9342	3000	1.4347	29.0002
1.3031	2.0632	3200	1.3301	26.7975
1.1895	2.1921	3400	1.2335	25.5110
1.1376	2.3211	3600	1.2340	25.0352
1.071	2.4500	3800	1.1303	23.8957
1.0488	2.5790	4000	1.1066	22.7855
1.0068	2.7079	4200	1.0825	22.4624
0.9812	2.8369	4400	1.0719	22.3449
0.9545	2.9658	4600	1.0562	22.1393

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1