openai/whisper-small
This model is a fine-tuned version of openai/whisper-small on the Hanhpt23/GermanMed-full dataset. It achieves the following results on the evaluation set:
- Loss: 0.5780
- Wer: 22.2257
- Cer: 13.8298
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: 8
- 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: 100
- num_epochs: 20
Training results
Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
---|---|---|---|---|---|
0.4708 | 1.0 | 194 | 0.4614 | 98.9304 | 70.2528 |
0.2235 | 2.0 | 388 | 0.4587 | 60.4135 | 40.8815 |
0.1189 | 3.0 | 582 | 0.4685 | 47.7116 | 28.7979 |
0.0683 | 4.0 | 776 | 0.5008 | 35.8531 | 23.1294 |
0.0525 | 5.0 | 970 | 0.4930 | 28.8388 | 18.5559 |
0.0383 | 6.0 | 1164 | 0.5284 | 28.0263 | 17.5719 |
0.0286 | 7.0 | 1358 | 0.5381 | 28.2012 | 17.5926 |
0.0223 | 8.0 | 1552 | 0.5445 | 25.1054 | 16.0976 |
0.0094 | 9.0 | 1746 | 0.5428 | 26.6276 | 17.1769 |
0.0057 | 10.0 | 1940 | 0.5409 | 25.1671 | 15.5778 |
0.0064 | 11.0 | 2134 | 0.5533 | 26.6996 | 17.6030 |
0.002 | 12.0 | 2328 | 0.5644 | 22.5033 | 13.9009 |
0.0026 | 13.0 | 2522 | 0.5626 | 22.9559 | 14.3305 |
0.0005 | 14.0 | 2716 | 0.5667 | 22.7810 | 14.1555 |
0.0004 | 15.0 | 2910 | 0.5679 | 22.5651 | 13.9840 |
0.0004 | 16.0 | 3104 | 0.5719 | 23.1307 | 14.5211 |
0.0003 | 17.0 | 3298 | 0.5753 | 22.4416 | 13.9199 |
0.0003 | 18.0 | 3492 | 0.5765 | 22.4108 | 13.9269 |
0.0003 | 19.0 | 3686 | 0.5776 | 22.2874 | 13.8853 |
0.0003 | 20.0 | 3880 | 0.5780 | 22.2257 | 13.8298 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.3.0
- Datasets 2.19.1
- Tokenizers 0.19.1
- Downloads last month
- 0
This model does not have enough activity to be deployed to Inference API (serverless) yet.
Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.