Hy3-GGUF is a quantized language model hosted on Hugging Face in GGUF format. It employs a custom tokenization scheme that defines numerous specialized tokens for structured interactions.
The model includes distinct tokens for sentence beginnings, padding, user and assistant roles, as well as tags for reasoning segments, tool call blocks, individual tool calls, argument keys and values, and tool responses. These tokens support formatted output such as reasoning tags and tool-related structures. The implementation appears in a template that generates these tokens dynamically using a placeholder variable.
It is provided as part of the Hugging Face ecosystem of models. The page belongs to the organization or user AngelSlim and sits within the platform's collection of foundation models intended for open-source use and local inference runners compatible with the GGUF format. Hugging Face positions its model repository as a means to advance artificial intelligence through open source and open science.
No pricing, licensing details, or specific end-user roles beyond the general model-hosting context are stated.
In the Tool calling space, Hy3 takes a focused approach. It focuses on running large language models efficiently on consumer hardware using GGUF quantization. Hy3 is an open-source project aimed at developers. Hy3 is open source under the MIT license. It ships for the web, the command line, and API.
AngelSlim builds and maintains Hy3, and it first shipped in 2023. Development happens publicly on GitHub with 121.1k stars and 1.2k commits in the last 90 days. Among its 3 catalogued features are Quantized Model, Custom Tokenization, and Tool Calling.
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