MuonEq: Balancing Before Orthogonalization with Lightweight Equilibration

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chang, Da, Shi, Qiankun, Zhang, Lvgang, Li, Yu, Zhang, Ruijie, Lu, Yao, Liu, Yongxiang, Yuan, Ganzhao
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915997147987968
author Chang, Da
Shi, Qiankun
Zhang, Lvgang
Li, Yu
Zhang, Ruijie
Lu, Yao
Liu, Yongxiang
Yuan, Ganzhao
author_facet Chang, Da
Shi, Qiankun
Zhang, Lvgang
Li, Yu
Zhang, Ruijie
Lu, Yao
Liu, Yongxiang
Yuan, Ganzhao
contents Orthogonalized-update optimizers such as Muon improve training of matrix-valued parameters, but existing extensions typically either rescale updates after orthogonalization or use heavier whitening-based preconditioners before it. We introduce {\method}, a lightweight family of pre-orthogonalization equilibration schemes for Muon with three forms: two-sided row/column normalization (RC), row normalization (R), and column normalization (C). By rebalancing the momentum matrix before finite-step Newton--Schulz orthogonalization, {\method} improves the geometry seen by orthogonalization. We show that finite-step orthogonalization is governed by the input spectrum, especially stable rank and condition number, and that row/column normalization acts as a zeroth-order surrogate for whitening. For hidden matrix weights, R is the default variant. Theoretically, {\method} (R) retains the standard $\widetilde{\mathcal O}(T^{-1/4})$ Muon-type nonconvex stationarity guarantee with decoupled weight decay and a horizon-free diminishing learning-rate schedule, and extends it to finite-step NS5 up to an explicit inexactness constant. In LLaMA2 pretraining on C4, {\method} (R) consistently outperforms Muon on 130M, 350M, and 1B models, with faster convergence and lower validation perplexity. The code is available at the \href{https://github.com/MaeChd/muon-eq}{MuonEq codebase}.
format Preprint
id arxiv_https___arxiv_org_abs_2603_28254
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MuonEq: Balancing Before Orthogonalization with Lightweight Equilibration
Chang, Da
Shi, Qiankun
Zhang, Lvgang
Li, Yu
Zhang, Ruijie
Lu, Yao
Liu, Yongxiang
Yuan, Ganzhao
Machine Learning
Orthogonalized-update optimizers such as Muon improve training of matrix-valued parameters, but existing extensions typically either rescale updates after orthogonalization or use heavier whitening-based preconditioners before it. We introduce {\method}, a lightweight family of pre-orthogonalization equilibration schemes for Muon with three forms: two-sided row/column normalization (RC), row normalization (R), and column normalization (C). By rebalancing the momentum matrix before finite-step Newton--Schulz orthogonalization, {\method} improves the geometry seen by orthogonalization. We show that finite-step orthogonalization is governed by the input spectrum, especially stable rank and condition number, and that row/column normalization acts as a zeroth-order surrogate for whitening. For hidden matrix weights, R is the default variant. Theoretically, {\method} (R) retains the standard $\widetilde{\mathcal O}(T^{-1/4})$ Muon-type nonconvex stationarity guarantee with decoupled weight decay and a horizon-free diminishing learning-rate schedule, and extends it to finite-step NS5 up to an explicit inexactness constant. In LLaMA2 pretraining on C4, {\method} (R) consistently outperforms Muon on 130M, 350M, and 1B models, with faster convergence and lower validation perplexity. The code is available at the \href{https://github.com/MaeChd/muon-eq}{MuonEq codebase}.
title MuonEq: Balancing Before Orthogonalization with Lightweight Equilibration
topic Machine Learning
url https://arxiv.org/abs/2603.28254