EYEDOL/whisper-tiny-yoruba

This model is a fine-tuned version of EYEDOL/whisper-tiny-yoruba on the EYEDOL/naija-voices-yoruba-split_0-1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8526
• Wer Ortho: 0.7773
• Wer: 0.7006

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
1.7353	1.0	583	0.8680	0.9448	0.8654
1.6436	2.0	1166	0.8494	0.8213	0.7490
1.5173	3.0	1749	0.8319	0.8237	0.7470
1.4143	4.0	2332	0.8215	0.7845	0.7128
1.3252	5.0	2915	0.8135	0.8788	0.7910
1.2425	6.0	3498	0.8106	0.7988	0.7224
1.1664	7.0	4081	0.8118	0.8508	0.7635
1.0950	8.0	4664	0.8156	0.7628	0.6813
1.0273	9.0	5247	0.8191	0.7867	0.7204
0.9611	10.0	5830	0.8292	0.7736	0.6948
0.8975	11.0	6413	0.8353	0.8007	0.7126
0.8363	12.0	6996	0.8526	0.7773	0.7006

Framework versions

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