HY-MT1.5

This repository contains quantized versions of Tencent’s Hunyuan Translation Model (HY-MT1.5).

HY-MT1.5 is a state-of-the-art translation model supporting mutual translation across 33 languages and 5 ethnic/dialect variations. It is optimized for explanatory translation, mixed-language scenarios, terminology intervention, and formatted translation.

Performance

Quick Start

Pull and run the model using Ollama:

# For the 1.8B model (Fast, efficient, edge-ready)
ollama run ali6parmak/hy-mt1.5:1.8b

# For the 7B model (Higher accuracy, complex scenarios)
ollama run ali6parmak/hy-mt1.5:7b

Model Details

• Original Developer: Tencent Hunyuan
• Architectures: 1.8B (Edge/Real-time) and 7B (High Performance)
• Capabilities:
  • Standard Translation: High-quality translation across 33 languages.
  • Terminology Intervention: Force specific translations for terms.
  • Contextual Translation: Translate based on provided background info.
  • Formatted Translation: Preserve XML/HTML-like tags during translation.

Prompt Templates

This model works best with specific prompt structures. Do not use a system prompt; instead, use the formats below in the user prompt.

1. Standard Translation (Chinese ⇔ Other)

Template:

将以下文本翻译为{target_language}，注意只需要输出翻译后的结果，不要额外解释：

{source_text}

2. Standard Translation (Non-Chinese ⇔ Non-Chinese)

Template:

Translate the following segment into {target_language}, without additional explanation.

{source_text}

3. Terminology Intervention

Use this to force specific terms to be translated a certain way.

Template:

参考下面的翻译：
{source_term} 翻译成 {target_term}

将以下文本翻译为{target_language}，注意只需要输出翻译后的结果，不要额外解释：
{source_text}

4. Contextual Translation

Use this when the translation depends on previous context.

Template:

{context}
参考上面的信息，把下面的文本翻译成{target_language}，注意不需要翻译上文，也不要额外解释：
{source_text}

Supported Languages

License & Citation

This model is based on the work by Tencent Hunyuan. Please refer to the Official Repository for license details.

@misc{hy-mt1.5,
      title={HY-MT1.5 Technical Report}, 
      author={Mao Zheng and Zheng Li and Tao Chen and Mingyang Song and Di Wang},
      year={2025},
      eprint={2512.24092},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}