Whisper Small Fr - Radiologie2.0 Encoder-Layer[0:3]+ LoRa(QKVO; FFN)
This model is a fine-tuned version of StephaneBah/whisper-small-rad-fr1.1 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.8256
- Wer: 34.6130
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: 3e-05
- train_batch_size: 8
- eval_batch_size: 6
- seed: 3407
- optimizer: Use OptimizerNames.ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- training_steps: 1000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| No log | 6.25 | 100 | 0.8140 | 34.4912 |
| No log | 12.5 | 200 | 0.8138 | 34.4912 |
| No log | 18.75 | 300 | 0.8165 | 34.4912 |
| No log | 25.0 | 400 | 0.8190 | 34.4302 |
| 0.0001 | 31.25 | 500 | 0.8202 | 34.5521 |
| 0.0001 | 37.5 | 600 | 0.8234 | 34.4302 |
| 0.0001 | 43.75 | 700 | 0.8248 | 34.4302 |
| 0.0001 | 50.0 | 800 | 0.8261 | 34.2474 |
| 0.0001 | 56.25 | 900 | 0.8266 | 34.6740 |
| 0.0 | 62.5 | 1000 | 0.8256 | 34.6130 |
Framework versions
- Transformers 4.51.3
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.21.2