BertjeWDialDataALLQonly
This model is a fine-tuned version of GroNLP/bert-base-dutch-cased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.9438
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: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 15.0
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.2122 | 1.0 | 871 | 2.0469 |
| 2.0961 | 2.0 | 1742 | 2.0117 |
| 2.0628 | 3.0 | 2613 | 2.0040 |
| 2.0173 | 4.0 | 3484 | 1.9901 |
| 1.9772 | 5.0 | 4355 | 1.9711 |
| 1.9455 | 6.0 | 5226 | 1.9785 |
| 1.917 | 7.0 | 6097 | 1.9380 |
| 1.8933 | 8.0 | 6968 | 1.9651 |
| 1.8708 | 9.0 | 7839 | 1.9915 |
| 1.862 | 10.0 | 8710 | 1.9310 |
| 1.8545 | 11.0 | 9581 | 1.9422 |
| 1.8231 | 12.0 | 10452 | 1.9310 |
| 1.8141 | 13.0 | 11323 | 1.9362 |
| 1.7939 | 14.0 | 12194 | 1.9334 |
| 1.8035 | 15.0 | 13065 | 1.9197 |
Framework versions
- Transformers 4.13.0.dev0
- Pytorch 1.10.0
- Datasets 1.16.1
- Tokenizers 0.10.3