Hy-MT2
Tencent Hunyuan MT model training code + IFMTBench.
Last verified:
What is Hy-MT2?
Hy-MT2 is Tencent Hunyuan's family of open-source 'fast-thinking' multilingual translation models built for complex real-world scenarios. It comes in three sizes (1.8B, 7B, and 30B-A3B MoE) supporting translation across 33 languages, with weights available on HuggingFace and ModelScope.
Hy-MT2 pricing
Pricing model: Freemium
Free and open-source (GitHub repository with downloadable model weights)
Hy-MT2 pros
- Three model sizes including an efficient MoE variant (30B-A3B)
- Supports translation among 33 languages
- Open-source with weights on HuggingFace and ModelScope
- Includes IFMTBench evaluation toolkit and training code
Hy-MT2 cons
- Content truncated; full benchmark and deployment details not visible
- Self-hosting required; no hosted API in this repo
- Limited community traction (675 stars) relative to larger translation models
Frequently asked questions about Hy-MT2
What model sizes are available for Hy-MT2?
Three sizes: 1.8B, 7B, and 30B-A3B (a Mixture-of-Experts variant).
Where can I download Hy-MT2 model weights?
Weights are available on HuggingFace and ModelScope under Tencent / Tencent-Hunyuan collections.
How many languages does Hy-MT2 support?
It supports translation among 33 languages.