Wave2Vec2-Bert2.0 - Kiran Pantha
This model is a fine-tuned version of kiranpantha/w2v-bert-2.0-nepali on the kiranpantha/OpenSLR54-Balanced-Nepali dataset. It achieves the following results on the evaluation set:
- Loss: 0.3414
- Wer: 0.3612
- Cer: 0.0805
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: 8
- 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: 500
- num_epochs: 2
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.4176 | 0.24 | 300 | 0.3260 | 0.3485 | 0.0772 |
| 0.4128 | 0.48 | 600 | 0.3514 | 0.3620 | 0.0810 |
| 0.4161 | 0.72 | 900 | 0.3460 | 0.3618 | 0.0810 |
| 0.3578 | 0.96 | 1200 | 0.3366 | 0.3528 | 0.0804 |
| 0.359 | 1.2 | 1500 | 0.3595 | 0.3577 | 0.0787 |
| 0.3371 | 1.44 | 1800 | 0.3446 | 0.3634 | 0.0808 |
| 0.3309 | 1.6800 | 2100 | 0.3399 | 0.3677 | 0.0818 |
| 0.3441 | 1.92 | 2400 | 0.3414 | 0.3612 | 0.0805 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.5.0+cu124
- Datasets 3.0.2
- Tokenizers 0.20.1