Whisper Small Basque
This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_13_0 eu dataset. It achieves the following results on the evaluation set:
- Loss: 0.2287
- Wer: 12.8397
If you need to use this model with whisper.cpp, you can download the ggml file: ggml-medium-eu.bin
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: 4
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 8000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.4415 | 0.06 | 500 | 0.5092 | 36.9699 |
| 0.4206 | 0.12 | 1000 | 0.4144 | 28.3365 |
| 0.272 | 0.19 | 1500 | 0.3554 | 24.7438 |
| 0.2681 | 0.25 | 2000 | 0.3271 | 22.1414 |
| 0.2099 | 0.31 | 2500 | 0.2973 | 19.5350 |
| 0.2283 | 0.38 | 3000 | 0.2760 | 18.5042 |
| 0.1477 | 1.03 | 3500 | 0.2637 | 17.1493 |
| 0.1008 | 1.09 | 4000 | 0.2592 | 16.3939 |
| 0.0866 | 1.15 | 4500 | 0.2561 | 15.8066 |
| 0.0915 | 1.21 | 5000 | 0.2411 | 15.0310 |
| 0.0803 | 1.28 | 5500 | 0.2330 | 14.7616 |
| 0.0674 | 1.34 | 6000 | 0.2325 | 13.8462 |
| 0.0679 | 1.4 | 6500 | 0.2299 | 13.5809 |
| 0.027 | 2.05 | 7000 | 0.2304 | 13.3805 |
| 0.0231 | 2.11 | 7500 | 0.2287 | 12.8397 |
| 0.0285 | 2.18 | 8000 | 0.2304 | 12.8883 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.1.dev0
- Tokenizers 0.13.2
