README.md

metadata

language:
  - en
  - ja

Pythia 410M fine-tuned on Light Novels

This model was fine-tuned on light and web novels. This model was trained for translation, but can do generation too.

This model is a test of using monolingual data to improve translation as well as improving translation by adding similar sentence pairs to prompts.

English generation

To generate English text with this model, start your prompt with <|gen_en|>.

Japanese generation

To generate Japanese text with this model, start your prompt with <|gen_ja|>.

Japanese to English translation

To translate, format your prompt as such

<|tl_ja|>JAPANESE EXAMPLE SENTENCE 1<|tl_en|>ENGLISH EXAMPLE SENTENCE 1<|tl_end|>
<|tl_ja|>JAPANESE EXAMPLE SENTENCE 2<|tl_en|>ENGLISH EXAMPLE SENTENCE 2<|tl_end|>
<|tl_ja|>JAPANESE SENTENCE TO TRANSLATE<|tl_en|>