wav2vec2-large-mms-1b-aft-meh

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

• Loss: 0.5176
• Wer: 0.4064

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: 0.001
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 100
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
29.4973	0.3195	100	3.4384	0.9990
1.4758	0.6390	200	0.6976	0.5184
1.7661	0.9585	300	0.6074	0.4742
0.9346	1.2780	400	0.5946	0.4596
0.6044	1.5974	500	0.5625	0.4750
0.7533	1.9169	600	0.5578	0.4500
0.7023	2.2364	700	0.5483	0.4298
0.4969	2.5559	800	0.5361	0.4193
0.5412	2.8754	900	0.5401	0.4236
0.5239	3.1949	1000	0.5485	0.4202
0.5133	3.5144	1100	0.5390	0.4164
0.4787	3.8339	1200	0.5371	0.4159
0.4596	4.1534	1300	0.5187	0.4050
0.4482	4.4728	1400	0.5175	0.4076
0.4293	4.7923	1500	0.5176	0.4064

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.1.1
• Tokenizers 0.22.1