Whisper Large-v3 Shona - S2S-M Curriculum 4500 steps
This model is a fine-tuned version of openai/whisper-large-v3 on the Cleaned Google WAXAL Shona dataset. It achieves the following results on the evaluation set:
- Loss: 0.4129
- Wer: 36.5071
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: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 16
- optimizer: Use adamw_torch_fused 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: 450
- training_steps: 4500
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 13.4173 | 0.5996 | 500 | 0.6003 | 49.4012 |
| 5.2073 | 1.1991 | 1000 | 0.4735 | 39.8392 |
| 7.1774 | 1.7986 | 1500 | 0.4383 | 38.8371 |
| 4.4790 | 2.3981 | 2000 | 0.4283 | 37.0747 |
| 7.1915 | 2.9977 | 2500 | 0.4226 | 37.4413 |
| 4.8270 | 3.5972 | 3000 | 0.4157 | 36.6130 |
| 7.9870 | 4.1967 | 3500 | 0.4197 | 36.1432 |
| 6.2143 | 4.7962 | 4000 | 0.4134 | 36.4745 |
| 3.6004 | 5.3957 | 4500 | 0.4129 | 36.5071 |
Framework versions
- Transformers 5.14.1
- Pytorch 2.13.0+cu130
- Datasets 5.0.1
- Tokenizers 0.22.2