<!-- 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. -->
REPEAT622_2wangchanberta-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: 1.4995
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: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.3712 | 1.0 | 9531 | 0.7093 |
| 0.1678 | 2.0 | 19062 | 1.0832 |
| 0.1362 | 3.0 | 28593 | 1.0609 |
| 0.1211 | 4.0 | 38124 | 1.3014 |
| 0.1076 | 5.0 | 47655 | 1.4995 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.13.0+cu116
- Datasets 1.17.0
- Tokenizers 0.10.3
