Wave2Vec2-Bert2.0 - Kiran Pantha
This model is a fine-tuned version of kiranpantha/w2v-bert-2.0-nepali on the OpenSLR54 dataset. It achieves the following results on the evaluation set:
- Loss: 0.4052
- Wer: 0.4302
- Cer: 0.1029
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: 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 |
|---|---|---|---|---|---|
| 0.7515 | 0.15 | 300 | 0.4814 | 0.4911 | 0.1183 |
| 0.6554 | 0.3 | 600 | 0.5699 | 0.5382 | 0.1385 |
| 0.6723 | 0.45 | 900 | 0.5463 | 0.5401 | 0.1395 |
| 0.6635 | 0.6 | 1200 | 0.5244 | 0.5043 | 0.1250 |
| 0.6132 | 0.75 | 1500 | 0.4725 | 0.4831 | 0.1184 |
| 0.5786 | 0.9 | 1800 | 0.4620 | 0.4702 | 0.1147 |
| 0.5639 | 1.05 | 2100 | 0.4810 | 0.4668 | 0.1140 |
| 0.4863 | 1.2 | 2400 | 0.4639 | 0.4766 | 0.1151 |
| 0.4784 | 1.35 | 2700 | 0.4527 | 0.4611 | 0.1108 |
| 0.456 | 1.5 | 3000 | 0.4229 | 0.4458 | 0.1089 |
| 0.4613 | 1.65 | 3300 | 0.4460 | 0.4478 | 0.1095 |
| 0.4506 | 1.8 | 3600 | 0.4166 | 0.4413 | 0.1047 |
| 0.4369 | 1.95 | 3900 | 0.4052 | 0.4302 | 0.1029 |
Framework versions
- Transformers 4.45.0.dev0
- Pytorch 2.4.1+cu121
- Datasets 2.21.0
- Tokenizers 0.19.1