odexine/umamusume-translator-hy-mt2-7b-gguf

umamusume-translator-hy-mt2-7b-gguf

GGUF quantisations of Mario51/umamusume-translator-hy-mt2-7b, a finetune of tencent/Hy-MT2-7B

Model Description

This model was fine-tuned on a dataset created from the official English localization of the game, allowing it to capture the specific character voices, names, and tone of the Umamusume universe. It is intended to be used for translating untranslated story content.

Provided Quants

Recommended inference parameters

Tencent recommends the following generation parameters for Hy-MT2 7B:

temperature: 0.7
top_p: 0.6
top_k: 20
repetition_penalty: 1.05
max_tokens: 4096

Hy-MT2 does not use a default system prompt.

Training Data

The model was trained on a dataset generated by aligning the Japanese and official English story files. Thank you to the official localization team for their high-quality work which made this possible.

Limitations

This model is a specialist. It may not perform well on general-purpose Japanese translation outside of the Umamusume context. It has been trained primarily on story dialogue and may not know all specific racing or training terminology.