Muown adds row-norm control to Muon via explicit l_infty updates on row-magnitude vectors, proving optimal non-convex rates and delivering better perplexity plus stability on GPT pre-training from 124M to 2.7B parameters.
year = 2004, type =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Muown: Row-Norm Control for Muon Optimization
Muown adds row-norm control to Muon via explicit l_infty updates on row-magnitude vectors, proving optimal non-convex rates and delivering better perplexity plus stability on GPT pre-training from 124M to 2.7B parameters.