Academic Graham Neubig endorsed the findings on Muon's performance
@gneubig @parthsh_ We call the lower bounds in opt theory 😎 still didn’t go very close to Muon.
@parthsh_ @DimitrisPapail I think that's fine-tuning the parameters of SGD on the test set without tuning the baseline parameters? 😅
Users thank researchers for interesting experiments in the debate over SGD lower bounds versus Muon optimizer performance.
@DimitrisPapail @parthsh_ Love the posts though @DimitrisPapail, thanks for the interesting experiments!
@gneubig @parthsh_ We call the lower bounds in opt theory 😎 still didn’t go very close to Muon.