metadata
tags:
- generated_from_trainer
model-index:
- name: BertjeWDialDataALLQonly128
results: []
BertjeWDialDataALLQonly128
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: 2.0364
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: 5e-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: 12.0
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.2326 | 1.0 | 871 | 2.1509 |
| 2.1375 | 2.0 | 1742 | 2.1089 |
| 2.0442 | 3.0 | 2613 | 2.0655 |
| 2.0116 | 4.0 | 3484 | 2.0433 |
| 1.9346 | 5.0 | 4355 | 2.0134 |
| 1.9056 | 6.0 | 5226 | 1.9956 |
| 1.8295 | 7.0 | 6097 | 2.0287 |
| 1.8204 | 8.0 | 6968 | 2.0173 |
| 1.7928 | 9.0 | 7839 | 2.0251 |
| 1.7357 | 10.0 | 8710 | 2.0148 |
| 1.7318 | 11.0 | 9581 | 1.9274 |
| 1.7311 | 12.0 | 10452 | 1.9314 |
Framework versions
- Transformers 4.13.0.dev0
- Pytorch 1.10.0
- Datasets 1.16.1
- Tokenizers 0.10.3