mms-1b-all-multilingual_baseline-1.0.0

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5623
• Wer: 0.4566
• Cer: 0.1522

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.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_ratio: 0.1
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.6028	1.0	247	0.8022	0.5728	0.1968
0.9617	2.0	494	0.9078	0.6380	0.2448
0.8932	3.0	741	0.8112	0.5736	0.2005
0.8212	4.0	988	0.7287	0.5554	0.2072
0.7525	5.0	1235	0.6847	0.5221	0.1837
0.7243	6.0	1482	0.6573	0.5079	0.1792
0.6728	7.0	1729	0.6458	0.5147	0.1764
0.6532	8.0	1976	0.6869	0.5262	0.1686
0.6202	9.0	2223	0.6305	0.4977	0.1703
0.5919	10.0	2470	0.6300	0.4953	0.1672
0.5646	11.0	2717	0.5994	0.4873	0.1692
0.5345	12.0	2964	0.6070	0.4876	0.1701
0.5035	13.0	3211	0.6027	0.4803	0.1610
0.4783	14.0	3458	0.5880	0.4792	0.1637
0.452	15.0	3705	0.5698	0.4726	0.1601
0.4271	16.0	3952	0.5780	0.4727	0.1556
0.3998	17.0	4199	0.5637	0.4681	0.1581
0.3761	18.0	4446	0.5732	0.4618	0.1516
0.3598	19.0	4693	0.5699	0.4588	0.1515
0.3475	20.0	4940	0.5623	0.4566	0.1522

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 4.4.0
• Tokenizers 0.22.1