whisper-medium-afrispeech-general-v4
This model is a fine-tuned version of openai/whisper-medium on the afrispeech-200 dataset. It achieves the following results on the evaluation set:
- Loss: 0.5926
- Model Preparation Time: 0.0157
- Wer: 0.102
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: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.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: 1349
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Model Preparation Time | Wer |
|---|---|---|---|---|---|
| 0.3545 | 1.0 | 1350 | 0.4356 | 0.0157 | 0.118 |
| 0.2101 | 2.0 | 2700 | 0.4049 | 0.0157 | 0.106 |
| 0.0816 | 3.0 | 4050 | 0.4208 | 0.0157 | 0.107 |
| 0.0272 | 4.0 | 5400 | 0.4701 | 0.0157 | 0.106 |
| 0.0116 | 5.0 | 6750 | 0.4834 | 0.0157 | 0.109 |
| 0.0054 | 6.0 | 8100 | 0.5294 | 0.0157 | 0.107 |
| 0.0026 | 7.0 | 9450 | 0.5427 | 0.0157 | 0.106 |
| 0.0031 | 8.0 | 10800 | 0.5698 | 0.0157 | 0.103 |
| 0.0004 | 9.0 | 12150 | 0.5787 | 0.0157 | 0.103 |
| 0.0003 | 10.0 | 13500 | 0.5926 | 0.0157 | 0.102 |
Framework versions
- Transformers 5.3.0
- Pytorch 2.4.1+cu121
- Datasets 2.19.0
- Tokenizers 0.22.2