<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
deberta-large-finetuned-cola
This model is a fine-tuned version of microsoft/deberta-large on the glue dataset. It achieves the following results on the evaluation set:
- Loss: 0.5893
- Matthews Correlation: 0.6406
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
Training results
| Training Loss | Epoch | Step | Validation Loss | Matthews Correlation |
|---|---|---|---|---|
| 0.4838 | 1.0 | 535 | 0.5167 | 0.5613 |
| 0.2803 | 2.0 | 1070 | 0.4892 | 0.6232 |
| 0.1597 | 3.0 | 1605 | 0.5893 | 0.6406 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.1+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2
