|
--- |
|
license: mit |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: tolkien-mythopoeic-gen |
|
results: |
|
- task: |
|
name: Causal Language Modeling |
|
type: text-generation |
|
--- |
|
|
|
<!-- 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. --> |
|
|
|
# tolkien-mythopoeic-gen |
|
|
|
This model is a fine-tuned version of [gpt2](https://huggingface.co/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 |
|
|
