{"name":"MACaQ","repo":"huawei-csl/MACaQ","url":"https://midorreal.com/project/huawei-csl-macaq","repository":"https://github.com/huawei-csl/MACaQ","description":"SOTA model-agnostic quantization for GGUF, local inference and beyond. Quantize and run text, image, video and audio models with llama.cpp, vLLM, Diffusers and PyTorch.","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/huawei-csl/MACaQ","checked":"22 Nov 2026"},"language":{"value":"C++","source":"https://github.com/huawei-csl/MACaQ","checked":"22 Nov 2026"},"homepage":{"value":"https://huawei-csl.com/blog/macaq/","source":"https://huawei-csl.com/blog/macaq/","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/huawei-csl/MACaQ/releases","checked":"21 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/huawei-csl/MACaQ/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/huawei-csl/MACaQ","checked":"22 Nov 2026"}},"stars":387.0,"writeup":{"what_it_is":"MACaQ is a model-agnostic quantization project that compresses text, image, video and audio models. It produces GGUF files for llama.cpp and checkpoints for vLLM, Diffusers and PyTorch.","audience":"Developers who want to quantize and run local or served AI models","claims":[{"kind":"specific","claim":"Can quantize a 122B-parameter model in 64 minutes using 32.5 GB of device memory","excerpt":"MACaQ can quantize a **122B-parameter model in 64 minutes using just 32.5 GB of device memory**","status":"not_checked"},{"kind":"specific","claim":"Achieves lower perplexity than MR-GPTQ and DuQuant++, or matches their best Llama-3-8B perplexity with 11× less quantization time","excerpt":"You can get lower perplexity than MR-GPTQ\nand DuQuant++, or match their best Llama-3-8B perplexity with **11× less\nquantization time**.","status":"not_checked"},{"kind":"specific","claim":"Produces MACaQ-R GGUFs that run with a standard llama.cpp installation","excerpt":"Its GGUFs run with a\n[standard llama.cpp installation]","status":"not_checked"},{"kind":"specific","claim":"Qwen3.6-35B-A3B MACaQ-R Q4_K_XL has lower KL divergence to FP16 than Unsloth UD-Q4_K_XL at the same file size","excerpt":"| **MACaQ-R Q4_K_XL** | **22.36 GB** | **0.013367** |","status":"not_checked"},{"kind":"specific","claim":"Quantizes speech generators to W4A4 and saves compact checkpoints that load without repeating calibration","excerpt":"Saved checkpoints load without repeating calibration.","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-12"},"why_now":null,"owner_supplied":null,"score":{"verdict":"too_early","hype":94,"reality":31,"gap":63,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":{"estimate":true,"lean":"too_close","gap_low":29,"gap_high":70,"rule":"early-2.0"},"changes":[{"date":"2026-11-10T08:52:22+00:00","text":"Repository created","source":"https://github.com/huawei-csl/MACaQ"}],"cite":"MACaQ is a GitHub project at https://github.com/huawei-csl/MACaQ written mainly in C++, described there as \"SOTA model-agnostic quantization for GGUF, local inference and beyond. Quantize and run text, image, video and audio models with llama.cpp, vLLM, Diffusers and PyTorch\".","attribution":"Data from Mid or Real, https://midorreal.com/project/huawei-csl-macaq, checked 22 November 2026. Attribution with a link is required.","still_real":null}