Intel / post

HY-MT1.5 pairs a 1GB on-device translator with a 7B cloud model

Source: x.com
Date: Published Dec 30, 2025

Context

Tencent's HY-MT1.5 gives developers two sizes of translation model: a 1.8-billion-parameter version intended for deployment on devices after compression, and a larger 7-billion-parameter version. Both support translation among 33 languages. The project documentation also supplies prompts for preferred terminology, surrounding and formatted text.

Those controls let an application tell the translator which meaning or wording to preserve. Context can be supplied as background without asking the model to translate it, while terminology instructions specify how particular terms should appear. The choice therefore concerns both where translation runs and how an application guides its output.

Terms in this piece · Glossary

context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source x.com