MACaQ quantitative method released
The story so far
Qwen released MACaQ, a fast, model-agnostic calibration quantization method that reduces cost while maintaining accuracy in language and diffusion models by refining the independent quantization of each linear layer through online activation transformation and Hessian perception. MACaQ achieves only a 2.3% end-to-end overhead on Qwen3-32B with vLLM, has a lower perplexity than DuQuant++ and MR-GPTQ, and reaches the best perplexity of Llama-3-8B in 11 times shorter quantization time.
AI-written from the reports · Updated 7 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face Daily PapersMACaQ: Fast, Model Agnostic Calibrated Quantization for Low-Precision Weights and Activations
MACaQ is a post-training quantization method for MXFP4 W4A4, which reduces costs while maintaining accuracy in language and diffusion models by using online activation transformations and Hessian-aware refinement to independently quantize each linear layer. MACaQ achieves only a 2.3% end-to-end overhead increase on Qwen3-32B with vLLM, has a lower perplexity than DuQuant++ and MR-GPTQ, and reaches the best perplexity of Llama-3-8B in 11 times shorter quantization time.
Heat over time
- Now
- 3
- Peak
- 512 Nov, 08:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.