Huawei CSL Releases Qwen-based MACaQ GGUF Model on Hugging Face
A GGUF quantization of Huawei CSL's MACaQ model, built on the Qwen architecture, has been published on Hugging Face. This release allows enthusiasts and researchers to run the model locally using popular inference engines. This release provides the local LLM community with access to a notable research model from Huawei CSL, optimized for efficient execution on consumer hardware. It expands the range of open-weights options available for local deployment and experimentation. The model, designated as Qwen3.6-35B-A3B-MACaQ-R-GGUF, is based on a 35-billion parameter architecture and packaged in the GGUF format. Accompanying resources include research papers on arXiv and the official GitHub repository for MACaQ.
## BACKGROUND
GGUF is a popular file format designed by the llama.cpp team to store model weights and metadata for efficient local execution of large language models. MACaQ stands for Model Agnostic Calibrated Quantization, a quantization framework developed by Huawei Computing and Software Laboratories.