|
--- |
|
language: english |
|
tags: |
|
- t5 |
|
widget: |
|
- text: "dekingify: " |
|
example_title: "Translate 17th-century English to modern English" |
|
- text: "kingify: " |
|
example_title: "Translate modern English to 17th-century English" |
|
--- |
|
# Kingify 2Way |
|
This is a custom AI model that translates modern English into 17th-century English or "King James" English. |
|
|
|
## Details of the model |
|
|
|
This model is a fine-tuned version of [google/t5-v1_1-large] on a dataset of a modern Bible translation with matching King James Bible verses. |
|
|
|
## Intended uses & limitations |
|
|
|
At times, despite sharing the same language and general grammatical rules, English from previous centuries can be easily misunderstood. The purpose of this was to explore ways to understand texts from the 17th-century more clearly. |
|
|
|
#### How to use |
|
|
|
```python |
|
from transformers import AutoTokenizer, AutoModelWithLMHead |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("swcrazyfan/Kingify-2Way") |
|
|
|
model = AutoModelWithLMHead.from_pretrained("swcrazyfan/Kingify-2Way") |
|
``` |
|
|
|
#### Limitations and bias |
|
|
|
- The model is trained on the King James Version of the Bible, so it will work best with Christian-style language (or even clichés). |
|
- Before the 18th and 19th centuries, English spelling was inconsistent. Because of this, the model often does not recognize spellings different from those in the KJV. |
|
- The model was trained on a relatively small amount of data, so it will not be as accurate as a model trained on a larger data set. |
|
|
|
## Training data |
|
|
|
The data used to train this model is from the New English Translation and the King James Version of the Bible. |
|
|
|
## Training procedure |
|
|
|
The model was trained on Kaggle using the Hugging Face Transformers library. |
|
|
|
### Training hyperparameters |
|
The following hyperparameters were used during training: |
|
- num_train_epochs: 4 |
|
- learning_rate: 5e-04 |
|
- train_batch_size: 2 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
|
|
## Eval results |
|
|
|
The model was evaluated using a human test. A human was asked to evaluate the translation quality of the model. The human was not told which sentences were translated by the model and which sentences were written by a human. |
|
|
|
## BibTeX entry and citation info |
|
|
|
```bibtex |
|
@inproceedings{, |
|
title={Kingify 2Way}, |
|
author={Joshua Kaufmann}, |
|
year={2022}, |
|
url={https://huggingface.co/swcrazyfan/Kingify-2Way-T5-Large-v1_1} |
|
} |
|
``` |
