Tencent-HY-MT1.5 - Tencent's open-source translation model

What is Tencent-HY-MT1.5?

Tencent-HY-MT1.5 is an open-source translation model from Tencent Hunyuan, comprising two versions: Tencent-HY-MT1.5-1.8B and Tencent-HY-MT1.5-7B. The model supports translation between 33 international languages and 5 minority-Chinese/dialect languages, covering a variety of less commonly spoken languages. Version 1.8B, after quantization processing, requires only 1GB of memory to achieve real-time offline translation on consumer devices such as mobile phones, offering fast processing speed. Version 7B is designed for high-performance scenarios, boasting high translation accuracy and effectively reducing the inclusion of annotations and language mixing in the translation.

Main functions of Tencent-HY-MT1.5

• Multilingual supportIt covers translation between 33 languages and 5 ethnic minority languages/dialects, including common languages such as Chinese, English, and Japanese, as well as less common languages such as Czech, Marathi, Estonian, and Icelandic.
• End-side deployment capabilityThe HY-MT1.5-1.8B, after quantization, only requires 1GB of memory to achieve offline real-time translation on consumer devices such as mobile phones. The average processing time for 50 tokens is only 0.18 seconds, which is much faster than the 0.4 seconds of mainstream commercial translation APIs.
• High translation qualityThe HY-MT1.5-1.8B achieves the 90th percentile performance of ultra-large closed-source models such as Gemini-3.0-Pro in authoritative test sets such as FLORES-200, and comprehensively surpasses medium-sized open-source models.
• Enhanced practicalityIt supports custom terminology databases, contextual long text understanding, and formatted text translation. Users can import professional terminology lists, and the model can optimize subsequent translations based on the preceding context while preserving the original formatting.

Technical Principles of Tencent-HY-MT1.5

• On-Policy Distillation StrategyThe 1.8B model employs an On-Policy Distillation strategy, where the 7B large model acts as the teacher to guide training in real time. This avoids the small model simply memorizing the standard answer. By correcting the bias in predicting the distribution of sequences, the small model learns from its mistakes, significantly improving its generalization ability and translation quality.
• Custom terminology databaseUsers can build their own terminology lookup tables for different industries and professional scenarios. After importing the terminology database, the model will prioritize the adoption of user-defined standard terms during the translation process to ensure the accuracy and consistency of key terms.
• Context translationThe model possesses the ability to understand long texts and dialogue contexts, and can continuously optimize subsequent translation results based on the preceding context. It is suitable for translating long texts such as meeting minutes, interview content, novel chapters, and technical documents.
• Formatted translationThe model can maintain the format information before and after translation, and is suitable for translating structured content such as web pages and documents.

How to use Tencent-HY-MT1.5

• Get the modelVisit the Tencent Hunyuan official website for online experience; or visit GitHub or HuggingFace to download the required Tencent-HY-MT1.5-1.8B or Tencent-HY-MT1.5-7B model for deployment.
• End-side deploymentTencent-HY-MT1.5-1.8B, after quantization, requires only 1GB of memory to achieve direct deployment on consumer devices such as mobile phones and real-time offline translation.
• Cloud-based deploymentTencent-HY-MT1.5-7B is suitable for high-performance scenarios, can be deployed in the cloud, and can be used in conjunction with the 1.8B model to improve the consistency and stability of the results.

The project address for Tencent-HY-MT1.5

• GitHub repository(https://github.com/Tencent-Hunyuan/HY-MT)
• HuggingFace model libraryhttps://huggingface.co/collections/tencent/hy-mt15

Application Scenarios of Tencent-HY-MT1.5

• Mobile device translationThe Tencent-HY-MT1.5-1.8B, after quantization processing, only requires 1GB of memory to achieve real-time offline translation on consumer devices such as mobile phones. It is suitable for instant translation needs on mobile devices, such as language communication during travel and cross-language dialogue in instant messaging.
• High-performance translation serviceThe Tencent-HY-MT1.5-7B is designed for high-performance scenarios and is suitable for deployment in the cloud. It provides services for enterprise-level applications that require high translation quality, such as document translation for multinational corporations and technical document translation in professional fields.
• Multilingual content creationIt supports mutual translation between 33 international languages and 5 ethnic minority/dialect translations, which can meet the creative needs of content creators in a multilingual environment, such as multilingual video subtitle production and cross-language literary creation.
• Translation of professional terminologyIt supports custom terminology databases, allowing users to import professional terminology lists. The model will prioritize the adoption of user-defined standard terms during the translation process, making it suitable for translation needs in professional fields such as medicine, law, and science and technology.
• Long text translationIt possesses the ability to understand long texts and dialogue contexts, and can continuously optimize subsequent translation results based on the preceding context. It is suitable for translating long content such as meeting minutes, interview content, novel chapters, and technical documents.
• Formatted content translationIt can maintain the original formatting information before and after translation, and is suitable for translating structured content such as web pages and documents, ensuring that the formatting of the translated text is consistent with the original text.