EYEDOL/whisper-tiny-hausa1
This model is a fine-tuned version of EYEDOL/whisper-tiny-hausa on the EYEDOL/naija-voices-hausa-split_0-4 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7789
- Wer Ortho: 0.5659
- Wer: 0.5024
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 |
|---|---|---|---|---|---|
| 1.3554 | 1.0 | 665 | 0.6892 | 0.5658 | 0.5012 |
| 1.2612 | 2.0 | 1330 | 0.6763 | 0.5469 | 0.4888 |
| 1.1278 | 3.0 | 1995 | 0.6709 | 0.5400 | 0.4834 |
| 1.0170 | 4.0 | 2660 | 0.6717 | 0.5386 | 0.4779 |
| 0.9213 | 5.0 | 3325 | 0.6772 | 0.5329 | 0.4771 |
| 0.8332 | 6.0 | 3990 | 0.6860 | 0.5527 | 0.4855 |
| 0.7524 | 7.0 | 4655 | 0.6932 | 0.5433 | 0.4790 |
| 0.6750 | 8.0 | 5320 | 0.7081 | 0.5478 | 0.4916 |
| 0.6030 | 9.0 | 5985 | 0.7204 | 0.5631 | 0.4990 |
| 0.5348 | 10.0 | 6650 | 0.7364 | 0.5463 | 0.4798 |
| 0.4693 | 11.0 | 7315 | 0.7564 | 0.5753 | 0.5104 |
| 0.4085 | 12.0 | 7980 | 0.7789 | 0.5659 | 0.5024 |
Framework versions
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.8.3
- Tokenizers 0.22.2