roberta_checkpoint-finetuned-squad
This model is a fine-tuned version of WillHeld/roberta-base-coqa on the squad dataset. It achieves the following results on the evaluation set:
- Loss: 0.8934
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 |
|---|---|---|---|
| 0.8468 | 1.0 | 5536 | 0.8168 |
| 0.6239 | 2.0 | 11072 | 0.8237 |
| 0.4805 | 3.0 | 16608 | 0.8934 |
Framework versions
- Transformers 4.24.0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2