DistilBERT-POWO_Climber_Scratch
This model is a fine-tuned version of ViktorDo/DistilBERT-POWO_Scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1255
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
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.1177 | 1.0 | 2133 | 0.1311 |
| 0.0952 | 2.0 | 4266 | 0.1098 |
| 0.0792 | 3.0 | 6399 | 0.1255 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.1+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2