T6-Whisper-FineTuned-DL-Twi
This model is a fine-tuned version of openai/whisper-small on the Twi-native Ghanaian language. dataset. It achieves the following results on the evaluation set:
- Loss: 0.0077
- Wer: 40.8733
- Cer: 34.8904
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 3000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.0308 | 0.6333 | 1000 | 0.0273 | 45.2786 | 35.2605 |
| 0.0068 | 1.2666 | 2000 | 0.0108 | 30.0399 | 23.6567 |
| 0.0038 | 1.8999 | 3000 | 0.0077 | 40.8733 | 34.8904 |
Framework versions
- Transformers 4.48.0.dev0
- Pytorch 2.5.1+cu121
- Datasets 3.2.0
- Tokenizers 0.21.0