🔧🔗https://github.com/cognitivecomputations/Muon
Muon optimizer: +>30% sample efficiency with <3% wallclock overhead
