Whisper Small GA-EN Speech Translation
This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia dataset. It achieves the following results on the evaluation set:
- Loss: 1.4631
- Bleu: 32.04
- Chrf: 48.69
- Wer: 63.3949
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: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.03
- training_steps: 3000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Chrf | Wer |
|---|---|---|---|---|---|---|
| 2.3783 | 0.1312 | 100 | 1.8852 | 7.56 | 22.6 | 113.2823 |
| 1.92 | 0.2625 | 200 | 1.5276 | 16.93 | 32.19 | 81.4498 |
| 1.6661 | 0.3937 | 300 | 1.3907 | 16.26 | 35.75 | 99.1896 |
| 1.4712 | 0.5249 | 400 | 1.3126 | 24.55 | 42.56 | 77.8478 |
| 1.3404 | 0.6562 | 500 | 1.2960 | 23.94 | 42.25 | 77.3976 |
| 1.2106 | 0.7874 | 600 | 1.2556 | 23.82 | 43.46 | 73.5705 |
| 1.0312 | 0.9186 | 700 | 1.3002 | 23.73 | 43.09 | 74.6060 |
| 0.5265 | 1.0499 | 800 | 1.2993 | 28.09 | 45.57 | 69.1580 |
| 0.4802 | 1.1811 | 900 | 1.3466 | 25.21 | 43.38 | 75.7767 |
| 0.4415 | 1.3123 | 1000 | 1.3456 | 29.77 | 47.56 | 66.9968 |
| 0.4164 | 1.4436 | 1100 | 1.3373 | 27.92 | 45.54 | 70.9140 |
| 0.3937 | 1.5748 | 1200 | 1.3162 | 30.09 | 46.51 | 64.2053 |
| 0.3391 | 1.7060 | 1300 | 1.3424 | 24.82 | 45.35 | 72.9401 |
| 0.2969 | 1.8373 | 1400 | 1.3271 | 31.78 | 48.51 | 62.5394 |
| 0.2755 | 1.9685 | 1500 | 1.3523 | 31.6 | 48.33 | 61.3237 |
| 0.1059 | 2.0997 | 1600 | 1.3910 | 30.26 | 45.88 | 65.3309 |
| 0.0975 | 2.2310 | 1700 | 1.4255 | 30.28 | 46.1 | 64.1603 |
| 0.1047 | 2.3622 | 1800 | 1.3923 | 29.99 | 46.44 | 64.9257 |
| 0.0874 | 2.4934 | 1900 | 1.4111 | 30.14 | 47.09 | 65.1058 |
| 0.0838 | 2.6247 | 2000 | 1.4378 | 25.63 | 45.79 | 77.4426 |
| 0.0757 | 2.7559 | 2100 | 1.4356 | 29.28 | 47.5 | 65.0608 |
| 0.0749 | 2.8871 | 2200 | 1.4532 | 30.56 | 46.58 | 64.3854 |
| 0.0463 | 3.0184 | 2300 | 1.4324 | 32.69 | 49.04 | 62.6294 |
| 0.0265 | 3.1496 | 2400 | 1.4311 | 31.24 | 48.58 | 62.9896 |
| 0.0266 | 3.2808 | 2500 | 1.4409 | 31.97 | 47.99 | 62.4944 |
| 0.0237 | 3.4121 | 2600 | 1.4310 | 32.44 | 48.86 | 62.2692 |
| 0.0208 | 3.5433 | 2700 | 1.4483 | 31.3 | 47.49 | 63.5299 |
| 0.0185 | 3.6745 | 2800 | 1.4513 | 32.86 | 48.98 | 62.6294 |
| 0.0178 | 3.8058 | 2900 | 1.4583 | 31.77 | 48.91 | 63.0797 |
| 0.0194 | 3.9370 | 3000 | 1.4631 | 32.04 | 48.69 | 63.3949 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.2.0+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1