metadata
license: mit
base_model: gpt2
tags:
- generated_from_trainer
model-index:
- name: gpt2-wikitext2-full
results: []
gpt2-wikitext2-full
This model is a fine-tuned version of gpt2 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 3.9528
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: 3
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 4.0418 | 1.0 | 549 | 3.9434 |
| 3.6533 | 2.0 | 1098 | 3.9445 |
| 3.5037 | 3.0 | 1647 | 3.9528 |
Framework versions
- Transformers 4.43.3
- Pytorch 2.4.0+cu121
- Datasets 3.1.0
- Tokenizers 0.19.1