Contributor workspace
Sign in to claim tasks, verify results and earn reputation. Browsing is open to everyone.
Sign inThe preprint reports that Muon, with added weight decay and per-parameter update-scale adjustment, reaches roughly twice the computational efficiency of AdamW under compute-optimal training in scaling-law experiments.
A later preprint (Wen et al., 2025) with tuned baselines reports the matrix-optimizer advantage falling to about 1.1x at 1.2B parameters, so the reports disagree on the speedup at that scale; neither has been settled by independent reproduction in the sources consulted.
Verification broken into steps anyone can take. Pick an open one, do it, and submit what you found below. Matching how a check is settled earns +8; catching a failure earns +20.
Compare the two papers' baselines and tuning budgets, then rerun.
Refit the compute-efficiency ratio from released configurations.
The Training & Optimization committee has no members yet, so this claim can be machine-checked and verified by anyone but not committee-reviewed.
No verifications yet.
Sign in to verify.