wav2vec2-soninke

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.3270
• Wer: 0.1829

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.2113	1.1767	400	1.7410	0.9913
1.7439	2.3535	800	0.5347	0.5501
1.4710	3.5302	1200	0.4372	0.4206
1.2485	4.7069	1600	0.3799	0.3828
1.2148	5.8837	2000	0.3516	0.3523
1.0983	7.0589	2400	0.3203	0.3046
0.8259	8.2356	2800	0.3285	0.2942
0.6912	9.4124	3200	0.3193	0.2751
0.7422	10.5891	3600	0.3049	0.2551
0.6663	11.7658	4000	0.2963	0.2667
0.5808	12.9426	4400	0.2739	0.2437
0.5273	14.1178	4800	0.3011	0.2226
0.4595	15.2946	5200	0.2921	0.2336
0.4389	16.4713	5600	0.2870	0.2178
0.3840	17.6480	6000	0.3031	0.2202
0.4010	18.8247	6400	0.3058	0.2258
0.3712	20.0	6800	0.3151	0.2193
0.3377	21.1767	7200	0.3167	0.2008
0.2824	22.3535	7600	0.3275	0.2058
0.2756	23.5302	8000	0.3074	0.1981
0.2699	24.7069	8400	0.3133	0.1874
0.2307	25.8837	8800	0.3261	0.1912
0.2357	27.0589	9200	0.3304	0.1874
0.2249	28.2356	9600	0.3270	0.1829
0.2264	29.4124	10000	0.3339	0.1862

Framework versions

• Transformers 5.3.0
• Pytorch 2.10.0+cu128
• Datasets 4.8.4
• Tokenizers 0.22.2