Whisper Medium Basque
This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_17_0 eu dataset. It achieves the following results on the evaluation set:
- Loss: 0.3147
- Wer: 7.7586
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: 3.75e-05
- train_batch_size: 64
- eval_batch_size: 32
- 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: 500
- training_steps: 40000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0757 | 2.3474 | 1000 | 0.1944 | 11.3425 |
| 0.0341 | 4.6948 | 2000 | 0.2069 | 10.3402 |
| 0.0167 | 7.0423 | 3000 | 0.2218 | 9.6513 |
| 0.0109 | 9.3897 | 4000 | 0.2235 | 9.3508 |
| 0.0104 | 11.7371 | 5000 | 0.2353 | 9.7750 |
| 0.0077 | 14.0845 | 6000 | 0.2396 | 9.7118 |
| 0.0052 | 16.4319 | 7000 | 0.2522 | 9.3985 |
| 0.0059 | 18.7793 | 8000 | 0.2482 | 9.4883 |
| 0.0075 | 21.1268 | 9000 | 0.2527 | 9.8034 |
| 0.0048 | 23.4742 | 10000 | 0.2603 | 9.2785 |
| 0.0033 | 25.8216 | 11000 | 0.2584 | 9.0778 |
| 0.0039 | 28.1690 | 12000 | 0.2630 | 9.0962 |
| 0.0036 | 30.5164 | 13000 | 0.2666 | 9.3646 |
| 0.0031 | 32.8638 | 14000 | 0.2695 | 9.4342 |
| 0.0019 | 35.2113 | 15000 | 0.2687 | 9.0384 |
| 0.0025 | 37.5587 | 16000 | 0.2737 | 9.7768 |
| 0.0016 | 39.9061 | 17000 | 0.2789 | 8.8690 |
| 0.0011 | 42.2535 | 18000 | 0.2796 | 8.6280 |
| 0.0006 | 44.6009 | 19000 | 0.2774 | 8.9422 |
| 0.0025 | 46.9484 | 20000 | 0.2801 | 9.0458 |
| 0.0006 | 49.2958 | 21000 | 0.2815 | 8.6015 |
| 0.0016 | 51.6432 | 22000 | 0.2869 | 9.2134 |
| 0.0009 | 53.9906 | 23000 | 0.2873 | 8.9716 |
| 0.0007 | 56.3380 | 24000 | 0.2848 | 9.0183 |
| 0.0006 | 58.6854 | 25000 | 0.2849 | 8.7489 |
| 0.0004 | 61.0329 | 26000 | 0.2960 | 8.9514 |
| 0.0007 | 63.3803 | 27000 | 0.2874 | 8.7709 |
| 0.0005 | 65.7277 | 28000 | 0.2929 | 8.9193 |
| 0.0 | 68.0751 | 29000 | 0.2901 | 8.4988 |
| 0.0 | 70.4225 | 30000 | 0.2943 | 8.1296 |
| 0.0 | 72.7700 | 31000 | 0.2997 | 8.0307 |
| 0.0 | 75.1174 | 32000 | 0.3037 | 7.9812 |
| 0.0 | 77.4648 | 33000 | 0.3072 | 7.9483 |
| 0.0 | 79.8122 | 34000 | 0.3092 | 7.8667 |
| 0.0 | 82.1596 | 35000 | 0.3121 | 7.8264 |
| 0.0 | 84.5070 | 36000 | 0.3147 | 7.7586 |
| 0.0001 | 86.8545 | 37000 | 0.3071 | 8.1003 |
| 0.0 | 89.2019 | 38000 | 0.3069 | 8.0426 |
| 0.0 | 91.5493 | 39000 | 0.3081 | 8.0316 |
| 0.0 | 93.8967 | 40000 | 0.3088 | 8.0124 |
Framework versions
- Transformers 4.52.3
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.21.1