<!-- 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. -->
Bundestag-gpt2-large
This model is a fine-tuned version of benjamin/gerpt2-large on the None dataset. It achieves the following results on the evaluation set:
- Loss: 2.8236
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: 2
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.7899 | 1.0 | 32852 | 2.6879 |
| 2.4421 | 2.0 | 65704 | 2.6749 |
| 2.206 | 3.0 | 98556 | 2.7354 |
| 1.9544 | 4.0 | 131408 | 2.8236 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2
