<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
NEW_DAT_10_8wangchanberta-base-att-spm-uncased
This model is a fine-tuned version of airesearch/wangchanberta-base-att-spm-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.2286
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: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.7307 | 1.0 | 3953 | 1.1431 |
| 0.6442 | 2.0 | 7906 | 0.5108 |
| 0.402 | 3.0 | 11859 | 0.4406 |
| 0.2948 | 4.0 | 15812 | 0.3048 |
| 0.2758 | 5.0 | 19765 | 0.3484 |
| 0.2247 | 6.0 | 23718 | 0.2783 |
| 0.1418 | 7.0 | 27671 | 0.2137 |
| 0.1301 | 8.0 | 31624 | 0.2409 |
| 0.1257 | 9.0 | 35577 | 0.2431 |
| 0.1379 | 10.0 | 39530 | 0.2286 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.13.0+cu116
- Datasets 1.17.0
- Tokenizers 0.10.3
