indic-whisper-vulnerable
This model is a fine-tuned version of Vignesh-M/Indic-whisper on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.9019
- Wer: 73.2647
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
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- training_steps: 2000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.6317 | 0.8811 | 200 | 0.5725 | 73.7685 |
| 0.5098 | 1.7621 | 400 | 0.5493 | 73.3542 |
| 0.2528 | 2.6432 | 600 | 0.5969 | 74.4066 |
| 0.172 | 3.5242 | 800 | 0.6543 | 73.1639 |
| 0.1007 | 4.4053 | 1000 | 0.7227 | 75.2239 |
| 0.0737 | 5.2863 | 1200 | 0.7665 | 76.2763 |
| 0.0313 | 6.1674 | 1400 | 0.8143 | 74.5634 |
| 0.0242 | 7.0485 | 1600 | 0.8305 | 75.1679 |
| 0.0158 | 7.9295 | 1800 | 0.8710 | 74.6305 |
| 0.0085 | 8.8106 | 2000 | 0.9019 | 73.2647 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.3.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1