whisper-tiny-ft-marathi-numbers
This model is a fine-tuned version of openai/whisper-tiny on the sanchit-gandhi/marathi-numbers-test dataset. It achieves the following results on the evaluation set:
- Loss: 0.0270
- Accuracy: 1.0
You can reproduce this run by calling the script run.sh. Note that we use the same split for training and evaluation for demonstration purposes.
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: 3e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 0
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 1.0002 | 1.0 | 128 | 0.7066 | 0.9990 |
| 0.0895 | 2.0 | 256 | 0.0764 | 0.9990 |
| 0.0281 | 3.0 | 384 | 0.0270 | 1.0 |
| 0.0191 | 4.0 | 512 | 0.0186 | 1.0 |
| 0.017 | 5.0 | 640 | 0.0166 | 1.0 |
Framework versions
- Transformers 4.28.0.dev0
- Pytorch 2.1.0.dev20230308+cu118
- Datasets 2.10.2.dev0
- Tokenizers 0.13.2