reassessment

This paper evaluates the Muon optimizer on low-rank matrix factorization, finding it does not consistently outperform AdamW, challenging earlier claims about its advantages in large-scale deep learning.