MuonIO: Principled Norm-Aware Descent for Embedding Tables and Language Model Heads

Published in arXiv, 2026

Recommended citation: Linkai Ma, Xinyu Luo, Mengbo Wang, Ananth Grama, Petros Drineas, and Brian Bullins. (2026). "MuonIO: Principled Norm-Aware Descent for Embedding Tables and Language Model Heads." arXiv preprint arXiv:2610.02705.
Download Paper