wav2vec2bert-bengali-noisy-5db-1e5
This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.4485
- Wer: 0.5249
- Cer: 0.1546
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: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 10.4861 | 1.2675 | 100 | 4.9583 | 1.0125 | 0.6409 |
| 4.1577 | 2.5350 | 200 | 2.1901 | 0.7279 | 0.2387 |
| 3.2423 | 3.8025 | 300 | 1.8035 | 0.6420 | 0.1972 |
| 2.5448 | 5.0637 | 400 | 1.5855 | 0.5878 | 0.1783 |
| 2.0697 | 6.3312 | 500 | 1.4940 | 0.5455 | 0.1644 |
| 1.9701 | 7.5987 | 600 | 1.4620 | 0.5392 | 0.1597 |
| 1.5833 | 8.8662 | 700 | 1.4485 | 0.5249 | 0.1546 |
Framework versions
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.8.3
- Tokenizers 0.22.2