results

This model is a fine-tuned version of facebook/mms-1b on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.3647
• Wer: 0.5377
• Cer: 0.2651

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 500
• num_epochs: 13
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.4128	1.6495	40	3.2154	1.0	1.0
2.8944	3.2990	80	2.7463	0.9896	0.9891
1.5023	4.9485	120	1.4803	0.6971	0.3166
1.1458	6.5979	160	1.2789	0.5580	0.2638
0.9619	8.2474	200	1.2553	0.5639	0.2702
0.8777	9.8969	240	1.2722	0.5215	0.2633
0.7732	11.5464	280	1.3647	0.5377	0.2651

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1