wav2vec2-large-mms-1b-aft-bew

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: 0.6521
• Wer: 0.4987

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
5.0585	0.2469	100	0.8727	0.6269
0.8154	0.4938	200	0.8308	0.5949
0.8045	0.7407	300	0.7554	0.5535
0.7555	0.9877	400	0.7224	0.5515
0.6954	1.2346	500	0.7175	0.5414
0.7207	1.4815	600	0.7198	0.5393
0.7121	1.7284	700	0.7117	0.5420
0.6841	1.9753	800	0.6982	0.5284
0.6241	2.2222	900	0.6955	0.5327
0.6435	2.4691	1000	0.6822	0.5336
0.6824	2.7160	1100	0.6768	0.5269
0.6562	2.9630	1200	0.6706	0.5265
0.6151	3.2099	1300	0.6838	0.5118
0.6691	3.4568	1400	0.6691	0.5096
0.6004	3.7037	1500	0.6625	0.5067
0.5886	3.9506	1600	0.6611	0.5077
0.6093	4.1975	1700	0.6592	0.5019
0.5688	4.4444	1800	0.6599	0.4999
0.5745	4.6914	1900	0.6533	0.5010
0.6027	4.9383	2000	0.6521	0.4987

Framework versions

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