AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Block Rotation is All You Need for MXFP4 Quantization

Yuantian Shao, Peisong Wang, Yuanteng Chen, Chang Xu, Zhihui Wei, Jian Cheng

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 0 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
apply random Hadamard rotation (QuaRot) before RTN under MXFP4RTN MXFP4 · LLaMA-2 7B WikiText2 / avg accuracyperplexity / accuracy: 7.08/57.26 -> 13.09/50.32Hurt
replace global rotation with block-wise rotation (BRQ) (ablation)QuaRot+ MXFP4 · LLaMA-3.2 1B WikiText2 / avg accuracyperplexity / accuracy: 12.78/48.83 -> 11.95/49.87Improved
replace global rotation with block-wise rotation (BRQ) (ablation)QuaRot+ MXFP4 · Qwen2.5-1.5B WikiText2 / avg accuracyperplexity / accuracy: 12.80/53.50 -> 12.15/54.83Improved
replace global rotation with block-wise rotation (BRQ) (ablation)QuaRot+ MXFP4 · LLaMA-3 8B WikiText2perplexity: 7.68 -> 7.14Improved
replace global rotation with block-wise rotation (BRQ) (ablation)QuaRot+ MXFP4 · LLaMA-2 70B WikiText2perplexity: 3.76 -> 3.62Improved
use random block rotation (BRQ) instead of optimized global rotation (SpinQuant) (ablation)SpinQuant MXFP4 · LLaMA-3 8B WikiText2 / avg accuracyperplexity / accuracy: 7.62/61.93 -> 7.14/63.54Improved
optimize block rotation matrices within SpinQuant (BRQ_Spin)BRQ · LLaMA-3.2 3B WikiText2perplexity: 9.41 -> 9.08Improved
optimize block rotation matrices within SpinQuant (BRQ_Spin)BRQ · LLaMA-3 8B WikiText2perplexity: 7.14 -> 7.13No clear effect
use rotation dimension equal to MXFP4 block size (32) instead of larger or smaller (ablation)BRQ · LLaMA-3 / 3.2 WikiText2perplexityImproved
add GPTQ compensation to RTN under MXFP4RTN MXFP4 · LLaMA-3.2 1B WikiText2 / avg accuracyperplexity / accuracy: 15.91/46.89 -> 13.35/48.52Improved
use block-wise rotation instead of global rotation for online rotation (ablation)QuaRot · LLaMA-2 7B prefill latency (batch 1, SeqLen 512)overhead: 7.03% -> 4.22%Improved
apply SmoothQuant before RTN under MXFP4RTN MXFP4 · LLaMA-3.2 1B WikiText2perplexity: 15.91 -> 16.86Hurt

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