openai/whisper-small
This model is a fine-tuned version of openai/whisper-small on the pphuc25/VietMed-split-8-2 dataset. It achieves the following results on the evaluation set:
- Loss: 0.8783
- Wer: 84.0710
- Cer: 69.8088
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 |
---|---|---|---|---|---|
0.5701 | 1.0 | 569 | 0.5608 | 46.2200 | 38.1533 |
0.4525 | 2.0 | 1138 | 0.5395 | 65.0522 | 58.3991 |
0.3578 | 3.0 | 1707 | 0.5448 | 46.9632 | 42.0497 |
0.267 | 4.0 | 2276 | 0.5691 | 40.8786 | 36.0588 |
0.1655 | 5.0 | 2845 | 0.6143 | 74.4609 | 62.7468 |
0.0965 | 6.0 | 3414 | 0.6592 | 66.0443 | 49.6775 |
0.0465 | 7.0 | 3983 | 0.7106 | 64.5982 | 55.2552 |
0.0222 | 8.0 | 4552 | 0.7337 | 75.1455 | 63.4091 |
0.0162 | 9.0 | 5121 | 0.7609 | 57.3787 | 49.1906 |
0.0071 | 10.0 | 5690 | 0.7812 | 65.8429 | 53.4811 |
0.0048 | 11.0 | 6259 | 0.7982 | 49.3355 | 40.4222 |
0.0048 | 12.0 | 6828 | 0.8064 | 53.4249 | 43.6294 |
0.0023 | 13.0 | 7397 | 0.8291 | 55.4347 | 44.3160 |
0.0032 | 14.0 | 7966 | 0.8303 | 63.7196 | 56.2369 |
0.0013 | 15.0 | 8535 | 0.8443 | 83.7342 | 69.0423 |
0.0034 | 16.0 | 9104 | 0.8523 | 88.0871 | 73.9847 |
0.0009 | 17.0 | 9673 | 0.8752 | 85.7148 | 69.7958 |
0.0004 | 18.0 | 10242 | 0.8713 | 90.1885 | 72.8172 |
0.0004 | 19.0 | 10811 | 0.8762 | 82.5773 | 68.7490 |
0.0003 | 20.0 | 11380 | 0.8783 | 84.0710 | 69.8088 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.3.0
- Datasets 2.19.1
- Tokenizers 0.19.1
- Downloads last month
- 5
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.