TExAS-SQuAD-da
This model is a fine-tuned version of xlm-roberta-base on the TExAS-SQuAD-da dataset. It achieves the following results on the evaluation set:
- Exact match: 63.96%
- F1-score: 68.40%
In comparison, the jacobshein/danish-bert-botxo-qa-squad model achieves 30.37% EM and 37.15% F1.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- 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 |
|---|---|---|---|
| 1.6438 | 1.0 | 4183 | 1.4711 |
| 1.4079 | 2.0 | 8366 | 1.4356 |
| 1.2532 | 3.0 | 12549 | 1.4509 |
Framework versions
- Transformers 4.12.2
- Pytorch 1.8.1+cu101
- Datasets 1.12.1
- Tokenizers 0.10.3