Hy MT2 30B A3B is a foundation model hosted on Hugging Face under the tencent organization. The model uses a specialized tokenization scheme and includes a defined chat template with distinct tokens for user input, assistant responses, beginning and end of sentence markers, padding, and end-of-sequence handling.
Its chat template supports conditional generation of structured markers for reasoning segments and tool-use interactions. These include dedicated tokens for the start and end of thinking blocks, tool call blocks, individual tool calls, argument keys, and separators when a specific flag is enabled. The configuration explicitly names an eos token, pad token, and other custom identifiers that encode an md5-derived component for certain dynamic tokens.
The model is delivered as a repository on the Hugging Face platform, where it can be accessed for download or inference. No details on parameter counts, training data, performance benchmarks, licensing terms, or target user groups appear in the source page.
Hy MT2 30B A3B is a Foundation models & chat project. It focuses on delivering a high-capacity, efficient language model with specialized expert routing for complex tasks. It is built as an open-source project for AI researchers and developers. Hy MT2 30B A3B is open source under the Open Source license. It ships for the web, the command line, and API.
Behind Hy MT2 30B A3B is Tencent, based in China, and it first shipped in 2026. The project is developed in the open on GitHub with 499 stars and 13 commits in the last 90 days. Among its 3 catalogued features are mixture of Experts, Tool Calling, and Long Context.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do