wav2vec2-large-mms-1b-aft-hch

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: 0.5270
• Wer: 0.6696

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
7.3706	0.2370	100	1.2824	0.9847
0.9732	0.4739	200	0.9079	0.9222
0.8624	0.7109	300	0.8058	0.8791
0.7599	0.9479	400	0.8033	0.8824
0.6971	1.1848	500	0.6960	0.8122
0.6642	1.4218	600	0.6869	0.8134
0.6401	1.6588	700	0.6817	0.7969
0.6022	1.8957	800	0.6189	0.7678
0.5619	2.1327	900	0.6786	0.7951
0.5339	2.3697	1000	0.6041	0.7354
0.5639	2.6066	1100	0.5733	0.7110
0.5242	2.8436	1200	0.6208	0.7495
0.499	3.0806	1300	0.5747	0.7106
0.4904	3.3175	1400	0.5403	0.6728
0.4751	3.5545	1500	0.5549	0.6902
0.4814	3.7915	1600	0.5485	0.6990
0.4518	4.0284	1700	0.5487	0.6847
0.4134	4.2654	1800	0.5363	0.6733
0.4227	4.5024	1900	0.5499	0.6939
0.4246	4.7393	2000	0.5248	0.6678
0.4069	4.9763	2100	0.5270	0.6696

Framework versions

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