AI Paper Atlas

Research map / Architectures and efficiency / Quantization

MicroMix: Efficient Mixed-Precision Quantization with Microscaling Formats for Large Language Models

Wenyuan Liu, Haoqian Meng, Yilun Luo, Peng Zhang, Xindian Ma

ICLR 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 11 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use MXFP8 E4M3 and MXFP6 E3M2 instead of E5M2 and E3M2 (ablation)MicroMix · WikiText2, Llama3.1-8Bperplexity: 6.84 -> 6.72Improved
use MXFP6 E3M2 instead of E2M3 with MXFP8 E4M3 (ablation)MicroMix · Llama3.1-8B zero-shot and WikiText2Winogrande: 71.51 -> 72.14Mixed
use MXFP6 E2M3 instead of E3M2 with MXFP8 E5M2 (ablation)MicroMix · WikiText2, Llama3.1-8Bperplexity: 6.84 -> 6.81No clear effect
calibrate on Pile instead of WikiText2 (ablation)MicroMix · WikiText2, Qwen2.5-14Bperplexity: 5.87 -> 5.92No clear effect
calibrate on C4 instead of WikiText2 (ablation)MicroMix · WikiText2, Qwen2.5-14Bperplexity: 5.87 -> 5.92No clear effect
apply Atom or QUIK fixed-channel mixed precision to MXFP formats instead of MicroMix thresholds (ablation)Atom on MXFP · WikiText2, Llama3.1-8Bperplexity: 6.72 -> 8.02Hurt
vary average bit-width from 4.48 to 6.01 (ablation)MicroMix · Qwen2.5-32B zero-shot avgaccuracy: 74.89 -> 75.20No clear effect
replace FP16 with MicroMixMixtral-8x7B-Instruct · zero-shot avgaccuracy: 78.58 -> 78.20No clear effect
replace FP16 with MicroMixMixtral-8x7B-Instruct · execution timetime: 5min 18s -> 2min 03sImproved
quantize KV cache to INT4 on top of MicroMixMicroMix · Llama3.1-8B WikiText2perplexity: 6.72 -> 6.97Hurt
use MicroMix instead of 6-bit INT quantizationINT6 · Llama3.1-8B MMLUaccuracy: 58.67 -> 62.65Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy