<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
Whisper Small Sanskrit lr scheduler - Bidit Sadhukhan
This model is a fine-tuned version of openai/whisper-small on the load_data dataset. It achieves the following results on the evaluation set:
- Loss: 0.1088
- Wer: 26.0197
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: 2.5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 44
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 500
- training_steps: 10000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1168 | 0.12 | 500 | 0.2069 | 47.5067 |
| 0.0905 | 0.23 | 1000 | 0.1778 | 43.9377 |
| 0.0844 | 0.35 | 1500 | 0.1738 | 41.1699 |
| 0.0664 | 0.47 | 2000 | 0.1584 | 39.2313 |
| 0.0693 | 0.58 | 2500 | 0.1460 | 36.5027 |
| 0.079 | 0.7 | 3000 | 0.1503 | 36.4747 |
| 0.0538 | 0.82 | 3500 | 0.1312 | 35.1860 |
| 0.0429 | 0.93 | 4000 | 0.1279 | 33.3035 |
| 0.0312 | 1.05 | 4500 | 0.1335 | 33.1410 |
| 0.0276 | 1.17 | 5000 | 0.1238 | 30.5693 |
| 0.0235 | 1.28 | 5500 | 0.1300 | 31.4209 |
| 0.0249 | 1.4 | 6000 | 0.1269 | 30.0034 |
| 0.0195 | 1.52 | 6500 | 0.1208 | 29.7344 |
| 0.0264 | 1.63 | 7000 | 0.1081 | 27.3756 |
| 0.018 | 1.75 | 7500 | 0.1093 | 27.2524 |
| 0.0229 | 1.87 | 8000 | 0.1036 | 26.9386 |
| 0.0204 | 1.98 | 8500 | 0.1045 | 26.4231 |
| 0.0057 | 2.1 | 9000 | 0.1077 | 25.9693 |
| 0.0076 | 2.21 | 9500 | 0.1087 | 26.0141 |
| 0.0091 | 2.33 | 10000 | 0.1088 | 26.0197 |
Framework versions
- Transformers 4.35.0.dev0
- Pytorch 2.0.1+cu117
- Datasets 2.14.5
- Tokenizers 0.14.0
