Whisper Small Serbian
This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 sr dataset. It achieves the following results on the evaluation set:
- Loss: 0.5670
- Wer: 22.9057
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
The model was trained on the training + validation splits of the Serbian split of the Common Voice dataset and evaluated on the test split of the Serbian split from the Common Voice dataset.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 64
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 800
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.4257 | 5.0 | 100 | 0.4377 | 32.4160 |
| 0.0779 | 10.0 | 200 | 0.3928 | 23.7556 |
| 0.0108 | 15.0 | 300 | 0.4856 | 23.4318 |
| 0.0104 | 20.0 | 400 | 0.5637 | 25.4958 |
| 0.0069 | 25.0 | 500 | 0.5289 | 23.1485 |
| 0.0022 | 30.0 | 600 | 0.5670 | 22.9057 |
| 0.0012 | 35.0 | 700 | 0.5746 | 23.0271 |
| 0.0006 | 40.0 | 800 | 0.5810 | 23.1890 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.1.dev0
- Tokenizers 0.13.2
- Downloads last month
- 3
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
the model is not deployed on the HF Inference API.
Dataset used to train asusevski/whisper-small-sr
Evaluation results
- Wer on mozilla-foundation/common_voice_11_0 srtest set self-reported22.906