EYEDOL/whisper-base-hausa
This model is a fine-tuned version of openai/whisper-base on the EYEDOL/naija-voices-hausa-split_0-0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7587
- Wer Ortho: 0.5628
- Wer: 0.5064
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: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer Ortho | Wer |
|---|---|---|---|---|---|
| 4.0572 | 1.0 | 665 | 1.1072 | 0.7943 | 0.7484 |
| 1.8708 | 2.0 | 1330 | 0.8796 | 0.6937 | 0.6386 |
| 1.4800 | 3.0 | 1995 | 0.7925 | 0.6231 | 0.5681 |
| 1.2536 | 4.0 | 2660 | 0.7508 | 0.6062 | 0.5510 |
| 1.0804 | 5.0 | 3325 | 0.7291 | 0.5889 | 0.5353 |
| 0.9379 | 6.0 | 3990 | 0.7218 | 0.5799 | 0.5243 |
| 0.8095 | 7.0 | 4655 | 0.7232 | 0.5652 | 0.5091 |
| 0.6922 | 8.0 | 5320 | 0.7285 | 0.5667 | 0.5107 |
| 0.5838 | 9.0 | 5985 | 0.7408 | 0.5633 | 0.5107 |
| 0.4834 | 10.0 | 6650 | 0.7587 | 0.5628 | 0.5064 |
Framework versions
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.8.3
- Tokenizers 0.22.2