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: 1.3690
- Bleu: 32.44
- Chrf: 48.06
- Wer: 63.2598
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: 0.01
- training_steps: 3000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Chrf | Wer |
|---|---|---|---|---|---|---|
| 2.3655 | 0.0438 | 100 | 1.7709 | 8.84 | 26.21 | 127.7803 |
| 1.8998 | 0.0876 | 200 | 1.5198 | 14.9 | 32.89 | 99.2796 |
| 1.5421 | 0.1313 | 300 | 1.3972 | 16.15 | 35.77 | 86.8077 |
| 1.3154 | 0.1751 | 400 | 1.3412 | 20.46 | 39.48 | 83.0707 |
| 1.1138 | 0.2189 | 500 | 1.3126 | 23.16 | 41.28 | 74.1108 |
| 0.9814 | 0.2627 | 600 | 1.3217 | 25.56 | 41.67 | 68.7528 |
| 0.8897 | 0.3065 | 700 | 1.2859 | 27.0 | 43.54 | 66.3215 |
| 0.7495 | 0.3503 | 800 | 1.2668 | 21.71 | 43.03 | 75.7767 |
| 0.7068 | 0.3940 | 900 | 1.2852 | 17.86 | 40.88 | 106.0333 |
| 0.6002 | 0.4378 | 1000 | 1.2476 | 24.0 | 44.26 | 78.4331 |
| 0.4989 | 0.4816 | 1100 | 1.2756 | 28.88 | 45.57 | 67.2670 |
| 0.4464 | 0.5254 | 1200 | 1.2756 | 27.81 | 45.53 | 66.8618 |
| 0.3883 | 0.5692 | 1300 | 1.2799 | 29.84 | 46.03 | 64.0702 |
| 0.341 | 0.6130 | 1400 | 1.2693 | 26.51 | 43.97 | 75.3715 |
| 0.2853 | 0.6567 | 1500 | 1.3310 | 26.99 | 45.58 | 74.0207 |
| 0.2611 | 0.7005 | 1600 | 1.3022 | 25.83 | 44.79 | 73.4354 |
| 0.2013 | 0.7443 | 1700 | 1.3266 | 30.78 | 46.61 | 63.6650 |
| 0.1886 | 0.7881 | 1800 | 1.2943 | 25.56 | 45.46 | 73.7055 |
| 0.1517 | 0.8319 | 1900 | 1.3193 | 28.93 | 45.09 | 64.3854 |
| 0.1288 | 0.8757 | 2000 | 1.3567 | 28.22 | 44.75 | 67.6722 |
| 0.1129 | 0.9194 | 2100 | 1.3431 | 29.55 | 46.22 | 66.2314 |
| 0.1 | 0.9632 | 2200 | 1.3365 | 31.46 | 48.14 | 64.9257 |
| 0.0505 | 1.0070 | 2300 | 1.3557 | 30.37 | 47.16 | 64.1153 |
| 0.0468 | 1.0508 | 2400 | 1.3648 | 31.57 | 48.17 | 62.0891 |
| 0.0373 | 1.0946 | 2500 | 1.3661 | 31.56 | 47.76 | 64.7456 |
| 0.0297 | 1.1384 | 2600 | 1.3638 | 31.13 | 47.74 | 64.3854 |
| 0.0283 | 1.1821 | 2700 | 1.3847 | 29.98 | 47.54 | 65.9613 |
| 0.0302 | 1.2259 | 2800 | 1.3730 | 32.32 | 48.28 | 64.0252 |
| 0.0229 | 1.2697 | 2900 | 1.3702 | 31.47 | 47.55 | 65.1508 |
| 0.0262 | 1.3135 | 3000 | 1.3690 | 32.44 | 48.06 | 63.2598 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.2.0+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1