wav2vec2-large-mms-1b-aft-led

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: 0.4923
• Wer: 0.3307

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
7.8608	0.1361	100	1.8473	0.9437
1.3703	0.2721	200	1.0214	0.6846
0.9658	0.4082	300	0.8563	0.5411
0.8803	0.5442	400	0.8306	0.5752
0.9234	0.6803	500	0.7605	0.5431
0.8064	0.8163	600	0.7326	0.4793
0.7492	0.9524	700	0.7121	0.4692
0.8054	1.0884	800	0.6845	0.4151
0.7078	1.2245	900	0.6491	0.4141
0.6766	1.3605	1000	0.6145	0.4033
0.665	1.4966	1100	0.6198	0.4052
0.6308	1.6327	1200	0.6079	0.3905
0.6869	1.7687	1300	0.5892	0.4030
0.6207	1.9048	1400	0.5757	0.3786
0.5922	2.0408	1500	0.5647	0.3802
0.6242	2.1769	1600	0.5670	0.3717
0.5857	2.3129	1700	0.5658	0.3941
0.5197	2.4490	1800	0.5544	0.3684
0.5493	2.5850	1900	0.5420	0.3554
0.5614	2.7211	2000	0.5271	0.3633
0.5574	2.8571	2100	0.5271	0.3552
0.5387	2.9932	2200	0.5182	0.3572
0.4993	3.1293	2300	0.5225	0.3494
0.4719	3.2653	2400	0.5178	0.3593
0.4438	3.4014	2500	0.5182	0.3505
0.5668	3.5374	2600	0.5071	0.3486
0.4859	3.6735	2700	0.5064	0.3371
0.4871	3.8095	2800	0.5077	0.3448
0.4706	3.9456	2900	0.5042	0.3447
0.4661	4.0816	3000	0.4995	0.3490
0.4369	4.2177	3100	0.5021	0.3452
0.4899	4.3537	3200	0.4987	0.3371
0.4404	4.4898	3300	0.4964	0.3395
0.4347	4.6259	3400	0.4929	0.3369
0.4327	4.7619	3500	0.4930	0.3340
0.4328	4.8980	3600	0.4923	0.3307

Framework versions

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