albert-large-v2-subword-masking-domain-adapted-nbme
This model is a fine-tuned version of albert-large-v2 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.1807
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: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 1.7821 | 1.0 | 1790 | 1.3868 |
| 1.2825 | 2.0 | 3580 | 1.2248 |
| 1.1817 | 3.0 | 5370 | 1.1712 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.1.0
- Tokenizers 0.12.1