Hy-MT2-30B-A3B
Hy-MT2-30B-A3B is part of Tencent’s fast-response translation family. It supports translation among 33 languages and instructions written in multiple languages.
Model design and access
The sparse release has 30B total parameters and about 3B active parameters. It needs the full expert weights despite its lower active computation. The release includes a training pipeline and LLaMA-Factory integration. Official low-precision formats are alternate distributions of the same model.Hy-MT2-30B-A3B official model card
Practical use
Practical guidance. Measure both memory and throughput. A low active parameter count does not mean it fits in the memory required by a dense 3B model.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Apache 2.0Hy-MT2-30B-A3B official model card |
| Adaptation | Full-parameter and LoRA fine-tuning toolsHy-MT2-30B-A3B official model card |
Related models
- Hy-MT2-Lite
Tencent · Hy-MT2 API
Hy-MT2-Lite is a hosted model for short-latency translation.
Launch date not confirmedAPI - Hy-MT2-Plus
Tencent · Hy-MT2 API
Hy-MT2-Plus is a hosted model for instruction-following translation.
Launch date not confirmedAPI - Hy-MT2-Pro
Tencent · Hy-MT2 API
Hy-MT2-Pro is a hosted model for specialist and quality-focused translation.
Launch date not confirmedAPI