w2v-bert-2.0-malayalam_mixeddataset_two.0
This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.1425
- Wer: 0.1451
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: 5e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.9341 | 0.24 | 300 | 0.4363 | 0.5138 |
| 0.228 | 0.47 | 600 | 0.3644 | 0.4847 |
| 0.1828 | 0.71 | 900 | 0.2752 | 0.3807 |
| 0.1479 | 0.95 | 1200 | 0.2671 | 0.3583 |
| 0.1213 | 1.19 | 1500 | 0.2291 | 0.2861 |
| 0.1114 | 1.42 | 1800 | 0.2098 | 0.2754 |
| 0.1049 | 1.66 | 2100 | 0.2088 | 0.2832 |
| 0.0962 | 1.9 | 2400 | 0.1789 | 0.2501 |
| 0.0777 | 2.14 | 2700 | 0.1945 | 0.2371 |
| 0.0685 | 2.37 | 3000 | 0.1788 | 0.2433 |
| 0.0663 | 2.61 | 3300 | 0.1707 | 0.2264 |
| 0.0652 | 2.85 | 3600 | 0.1834 | 0.2227 |
| 0.0573 | 3.08 | 3900 | 0.1663 | 0.2065 |
| 0.0445 | 3.32 | 4200 | 0.1479 | 0.1981 |
| 0.0417 | 3.56 | 4500 | 0.1477 | 0.1779 |
| 0.0415 | 3.8 | 4800 | 0.1504 | 0.1774 |
| 0.0368 | 4.03 | 5100 | 0.1407 | 0.1655 |
| 0.0248 | 4.27 | 5400 | 0.1568 | 0.1672 |
| 0.0258 | 4.51 | 5700 | 0.1495 | 0.1582 |
| 0.0227 | 4.74 | 6000 | 0.1460 | 0.1510 |
| 0.0225 | 4.98 | 6300 | 0.1425 | 0.1451 |
Framework versions
- Transformers 4.39.3
- Pytorch 2.1.1+cu121
- Datasets 2.16.1
- Tokenizers 0.15.1