Wav2vec2-afr-Hausa

This model is a fine-tuned version of asr-africa/wav2vec2-asr-africa-base-finetuned-naijavoices-hausa-v0.1 on the LAFRICAMOBILE/HAUSA - DEFAULT dataset. It achieves the following results on the evaluation set:

• Loss: 0.1778
• Wer: 0.2533
• Cer: 0.0659

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: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch 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: 1000
• num_epochs: 60.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5103	1.6595	500	0.3760	0.4019	0.1110
0.2346	3.3215	1000	0.2330	0.2877	0.0805
0.1783	4.9809	1500	0.1912	0.2676	0.0729
0.1552	6.6429	2000	0.1912	0.2604	0.0703
0.143	8.3049	2500	0.1810	0.2539	0.0678
0.1127	9.9644	3000	0.1805	0.2443	0.0646
0.1139	11.6263	3500	0.1778	0.2529	0.0660
0.096	13.2883	4000	0.1805	0.2529	0.0655
0.0949	14.9478	4500	0.1825	0.2464	0.0652
0.09	16.6098	5000	0.1926	0.2372	0.0625

Framework versions

• Transformers 4.50.3
• Pytorch 2.7.0+cu126
• Datasets 3.5.0
• Tokenizers 0.21.1