whisper-medium-dora-afrispeech-clinical
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: 0.7005
- Model Preparation Time: 0.0429
- Wer: 0.145
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: 0.0001
- 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: 2239
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Model Preparation Time | Wer |
|---|---|---|---|---|---|
| 0.5424 | 1.0 | 2240 | 0.6163 | 0.0429 | 0.162 |
| 0.4159 | 2.0 | 4480 | 0.5482 | 0.0429 | 0.147 |
| 0.2911 | 3.0 | 6720 | 0.5412 | 0.0429 | 0.143 |
| 0.2295 | 4.0 | 8960 | 0.5727 | 0.0429 | 0.142 |
| 0.1172 | 5.0 | 11200 | 0.6056 | 0.0429 | 0.146 |
| 0.0667 | 6.0 | 13440 | 0.6478 | 0.0429 | 0.146 |
| 0.0439 | 7.0 | 15680 | 0.7005 | 0.0429 | 0.145 |
Framework versions
- PEFT 0.18.1
- Transformers 5.3.0
- Pytorch 2.6.0+cu124
- Datasets 2.19.0
- Tokenizers 0.22.2