metadata
license: mit
tags:
- generated_from_trainer
model-index:
- name: tolkien-mythopoeic-gen
results:
- task:
name: Causal Language Modeling
type: text-generation
tolkien-mythopoeic-gen
This model is a fine-tuned version of gpt2 on Tolkien's mythopoeic works, namely The Silmarillion and Unfinished Tales of Numenor and Middle Earth.
It achieves the following results on the evaluation set:
- Loss: 3.5110
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: 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: 3.0
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 3.5732 | 1.0 | 145 | 3.5110 |
| 3.5713 | 2.0 | 290 | 3.5110 |
| 3.5718 | 3.0 | 435 | 3.5110 |
Framework versions
- Transformers 4.10.2
- Pytorch 1.9.0+cu102
- Tokenizers 0.10.3