swahili-distill_smoke-20260709-152928
This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:
- Loss: 2.2642
- Wer: 92.2175
- Cer: 35.3085
- Avg Pred Words: 32.51
- Avg Ref Words: 33.28
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: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- 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: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer | Avg Pred Words | Avg Ref Words |
|---|---|---|---|---|---|---|---|
| 2.997 | 1.5714 | 25 | 3.4215 | 102.3137 | 38.3373 | 34.21 | 33.28 |
| 2.7464 | 3.1270 | 50 | 3.0001 | 105.6190 | 39.4960 | 36.69 | 33.28 |
| 2.4057 | 4.6984 | 75 | 2.6360 | 101.9231 | 43.6317 | 34.85 | 33.28 |
| 2.0697 | 6.2540 | 100 | 2.2642 | 92.2175 | 35.3085 | 32.51 | 33.28 |
Framework versions
- Transformers 4.57.6
- Pytorch 2.10.0+cu128
- Datasets 3.6.0
- Tokenizers 0.22.2