<!-- 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. -->
kobert-lm-finetuned-klue-v2
This model is a fine-tuned version of monologg/kobert-lm on the None dataset. It achieves the following results on the evaluation set:
- Loss: 4.9679
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: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
5.7641 | 0.52 | 500 | 5.6079 |
5.5176 | 1.04 | 1000 | 5.4635 |
5.4401 | 1.57 | 1500 | 5.3370 |
5.2919 | 2.09 | 2000 | 5.2396 |
5.1756 | 2.61 | 2500 | 5.2675 |
5.0793 | 3.13 | 3000 | 5.1546 |
4.8933 | 3.66 | 3500 | 5.0728 |
4.7273 | 4.18 | 4000 | 5.0165 |
4.5473 | 4.7 | 4500 | 4.9679 |
Framework versions
- Transformers 4.22.2
- Pytorch 1.12.1+cu113
- Datasets 2.5.2
- Tokenizers 0.12.1