<!-- 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-finetuned-klue-v2
This model is a fine-tuned version of monologg/kobert on the None dataset. It achieves the following results on the evaluation set:
- Loss: 5.3234
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: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 20
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
5.5898 | 1.08 | 500 | 5.2618 |
5.217 | 2.16 | 1000 | 5.1505 |
5.1044 | 3.24 | 1500 | 5.0895 |
5.0048 | 4.32 | 2000 | 5.0649 |
4.8292 | 5.4 | 2500 | 4.9589 |
4.5451 | 6.48 | 3000 | 4.8549 |
4.2284 | 7.56 | 3500 | 4.8801 |
3.9195 | 8.64 | 4000 | 4.8797 |
3.6506 | 9.72 | 4500 | 4.8009 |
3.4175 | 10.8 | 5000 | 4.8996 |
3.1964 | 11.88 | 5500 | 4.9734 |
3.0401 | 12.96 | 6000 | 4.9378 |
2.8965 | 14.04 | 6500 | 5.3631 |
2.7672 | 15.12 | 7000 | 5.3234 |
Framework versions
- Transformers 4.22.2
- Pytorch 1.12.1+cu113
- Datasets 2.5.2
- Tokenizers 0.12.1