jem-whisper-yi-2026-09-02-v3
This model is a fine-tuned version of ivrit-ai/yi-whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.3918
- Wer Ortho: 1.0265
- Wer: 0.5948
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: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_ratio: 0.1
- lr_scheduler_warmup_steps: 100
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer Ortho | Wer |
|---|---|---|---|---|---|
| 0.5207 | 0.2861 | 250 | 0.4941 | 1.0022 | 0.8149 |
| 0.3343 | 0.5722 | 500 | 0.4368 | 1.0 | 0.7733 |
| 0.2921 | 0.8584 | 750 | 0.4080 | 0.9978 | 0.6446 |
| 0.3128 | 1.1442 | 1000 | 0.3904 | 1.0 | 0.6040 |
| 0.298 | 1.4303 | 1250 | 0.3870 | 0.9978 | 0.6234 |
| 0.2771 | 1.7165 | 1500 | 0.3759 | 0.9956 | 0.6227 |
| 0.2479 | 2.0023 | 1750 | 0.3757 | 1.0088 | 0.5854 |
| 0.2037 | 2.2884 | 2000 | 0.3775 | 0.9978 | 0.5591 |
| 0.2161 | 2.5745 | 2250 | 0.3754 | 0.9934 | 0.6139 |
| 0.2843 | 2.8607 | 2500 | 0.3728 | 0.9978 | 0.5425 |
| 0.1498 | 3.1465 | 2750 | 0.3909 | 1.0177 | 0.6225 |
| 0.1867 | 3.4326 | 3000 | 0.3967 | 1.0353 | 0.6351 |
| 0.2026 | 3.7187 | 3250 | 0.3918 | 1.0265 | 0.5948 |
Framework versions
- Transformers 4.48.1
- Pytorch 2.4.0
- Datasets 3.6.0
- Tokenizers 0.21.4