EYEDOL/whisper-tiny-yoruba2

This model is a fine-tuned version of EYEDOL/whisper-tiny-yoruba1 on the EYEDOL/naija-voices-yoruba-split_0-5 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8206
• Wer Ortho: 0.7328
• Wer: 0.6605

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 16
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.4647	1.0	583	0.7470	0.7172	0.6312
1.4031	2.0	1166	0.7409	0.7386	0.6564
1.3010	3.0	1749	0.7406	0.7229	0.6421
1.2136	4.0	2332	0.7384	0.7477	0.6644
1.1362	5.0	2915	0.7413	0.7877	0.7124
1.0626	6.0	3498	0.7462	0.7162	0.6332
0.9947	7.0	4081	0.7573	0.7547	0.6707
0.9314	8.0	4664	0.7624	0.7062	0.6266
0.8680	9.0	5247	0.7747	0.7211	0.6416
0.8080	10.0	5830	0.7881	0.7519	0.6641
0.7481	11.0	6413	0.8032	0.7310	0.6519
0.6922	12.0	6996	0.8206	0.7328	0.6605

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.8.3
• Tokenizers 0.22.2