88Z AI, 27 August — Tencent Hunyuan has released the on-device translation model Hy-MT2-1.8B, which uses proprietary compression technology to drastically reduce model size while maintaining high-quality translation that outperforms several commercial APIs. The model has been optimized for x86 operators and successfully deployed for real-time danmaku translation on Bilibili live streams, supporting 33 languages with a processing time of only 500-800 milliseconds per item.
At a glance
- Uses proprietary Sherry 1.25-bit technology for extreme model compression, with translation quality surpassing mainstream commercial APIs
- Collaborated with Intel on x86 operator optimization; deployed for real-time danmaku translation on Bilibili, supporting 33 languages
The update
Tencent Hunyuan has officially launched the on-device translation large model Hy-MT2-1.8B. By innovatively combining 2-bit quantization with its proprietary Sherry 1.25-bit technology, the model achieves extreme compression of the original 3.3GB size. Despite the significant reduction in volume, translation quality remains virtually lossless, outperforming mainstream commercial API services such as Microsoft and Doubao. In terms of engineering deployment, Tencent Hunyuan collaborated with Intel to complete operator optimization for the x86 architecture. The model is currently deployed in Bilibili live streaming scenarios for real-time danmaku translation. It supports up to 33 languages, with the translation time for a single danmaku controlled between 500 and 800 milliseconds, demonstrating efficient on-device inference capabilities.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 腾讯混元Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.