HY-MT 1.5

Copyright Notice

© 2025 Tencent-Hunyuan. All intellectual property rights of Tencent Hunyuan HY-MT1.5 Series Models are owned by Tencent. Tampering with or removing copyright notices is strictly prohibited.

Compliance Notes

• Comply with the Tencent Hunyuan Community License Agreement and relevant laws; use only for legitimate purposes.
• Users in restricted regions (e.g., EU, UK, South Korea) are prohibited from downloading/using the models. Downloading constitutes confirmation that your region complies with the requirements, and you shall bear sole responsibility for any non-compliant use.
• The models shall not be used to improve non-Hunyuan series AI models. You shall bear sole responsibility for any non-compliant use.

I. Project Overview

HY-MT1.5 is a next-generation open-source translation model project developed by Tencent-Hunyuan. It consists of two core models: HY-MT1.5-1.8B (1.8B parameters) and HY-MT1.5-7B (7B parameters). Open-sourced on GitHub, the project provides complete model files, technical documents, and deployment tools, aiming to deliver an efficient, accurate, and flexible multilingual translation solution for global users, covering diverse needs such as daily communication and professional scenarios.

II. Core Features

1. Wide-ranging Multilingual Translation：Supports bidirectional translation between 33 mainstream languages (e.g., Chinese-English, French-Spanish, Japanese-Korean, Russian-Arabic) and 5 ethnic languages/dialects (Tibetan, Kazakh, Mongolian, Uyghur, Cantonese). Whether you’re reading foreign news, chatting with international friends, or processing ethnic language content, it can handle it effortlessly.
2. Three Practical Translation Modes for Different Scenarios：
   • Terminology Intervention: Specify custom translations (e.g., "人工智能" must be translated as "AI"), which the model will strictly follow—ideal for professional document translation.
   • Contextual Translation: Refers to prior context for accurate translation, avoiding ambiguity caused by polysemy (e.g., correctly distinguishing "book" in "book a ticket" and "a good book").
   • Formatted Translation: Preserves the original text’s tags, layout, and other formatting during translation, suitable for structured content (e.g., code comments, document templates).
3. Small Model with High Performance, Flexible Deployment：Despite its small size (less than 1/3 of the 7B model), the 1.8B model delivers translation quality comparable to large models and most commercial translation tools. After quantization, it can run on edge devices (e.g., local computers, small servers) to achieve real-time translation without relying on cloud networks.
4. Upgraded Large Model for Professional Scenarios：The 7B model is an enhanced version of the WMT25 translation competition champion model. It performs exceptionally well in mixed-language and explanatory translation scenarios (e.g., technical documents, academic papers) with higher accuracy and clearer logic.

III. Technical Foundation & Development Team

• Development Team：Tencent-Hunyuan, the core team of Tencent dedicated to large model research and development, with extensive experience in natural language processing and model optimization.
• Underlying Technology：Built on the transformers deep learning framework, adopting the CausalLM architecture. Model quantization and compression are implemented via Tencent’s self-developed AngelSlim tool (supporting FP8, INT4, etc.), balancing performance and deployment efficiency. It also supports fine-tuning with the LLaMA-Factory tool, enabling users to customize the model according to their specific needs.