wav2vec2-large-mms-1b-aft-aln

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: 1.3110
• Wer: 0.5480

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
9.4884	0.2070	100	2.9621	0.9885
1.9509	0.4141	200	1.6554	0.6431
1.548	0.6211	300	1.6815	0.6276
1.4225	0.8282	400	1.5680	0.6218
1.5012	1.0352	500	1.3856	0.6255
1.5559	1.2422	600	1.5377	0.6050
1.2943	1.4493	700	1.4902	0.5942
1.2103	1.6563	800	1.5580	0.6001
1.3807	1.8634	900	1.4564	0.5954
1.4473	2.0704	1000	1.3519	0.5918
1.2579	2.2774	1100	1.4598	0.5799
1.2094	2.4845	1200	1.4669	0.5760
1.1881	2.6915	1300	1.4301	0.5673
1.3112	2.8986	1400	1.2790	0.5641
1.1826	3.1056	1500	1.3800	0.5652
1.2704	3.3126	1600	1.3190	0.5761
1.063	3.5197	1700	1.2034	0.5587
1.0273	3.7267	1800	1.2164	0.5601
1.1572	3.9337	1900	1.2023	0.5539
1.0812	4.1408	2000	1.2923	0.5538
1.1074	4.3478	2100	1.3394	0.5460
1.0665	4.5549	2200	1.3474	0.5456
1.0396	4.7619	2300	1.3025	0.5480
1.0812	4.9689	2400	1.3110	0.5480

Framework versions

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