legal-roberta-base-filtered-cuad
This model is a fine-tuned version of saibo/legal-roberta-base on the cuad dataset. It achieves the following results on the evaluation set:
- Loss: 0.0428
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: 22
- eval_batch_size: 22
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.0556 | 1.0 | 12279 | 0.0517 |
| 0.0406 | 2.0 | 24558 | 0.0425 |
| 0.0332 | 3.0 | 36837 | 0.0428 |
Framework versions
- Transformers 4.21.1
- Pytorch 1.12.0+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1