wav2vec2-E30_sp

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: 1.4665
• Cer: 30.6287

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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• 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
32.3037	0.1289	200	5.0215	100.0
4.9642	0.2579	400	4.6861	100.0
4.8422	0.3868	600	4.7308	100.0
4.7474	0.5158	800	4.6660	97.7203
4.6975	0.6447	1000	4.5628	97.9436
4.6688	0.7737	1200	4.5418	97.7262
4.6225	0.9026	1400	4.5220	97.9377
4.5627	1.0316	1600	4.5568	97.2092
4.5233	1.1605	1800	4.4908	96.2926
4.4088	1.2895	2000	4.4416	95.4465
4.0188	1.4184	2200	3.7750	77.5206
3.2832	1.5474	2400	3.1100	61.0576
2.7716	1.6763	2600	2.6530	51.1222
2.4346	1.8053	2800	2.3400	46.3220
2.1907	1.9342	3000	2.0907	40.3290
2.0173	2.0632	3200	1.9195	39.0541
1.8574	2.1921	3400	1.8219	37.3796
1.6992	2.3211	3600	1.7006	34.4947
1.623	2.4500	3800	1.6380	34.0129
1.5586	2.5790	4000	1.5646	32.3913
1.5001	2.7079	4200	1.4950	31.5570
1.4357	2.8369	4400	1.4842	31.3337
1.3935	2.9658	4600	1.4665	30.6287

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0