Spaces:
Running
Running
Ctrl+K
- claim-1-normuon-algorithm-muon-orthogonalization-neuron-wise-2nd-moment-normalization-alg-1
- claim-2-muon-leaves-high-per-neuron-update-norm-variance-normuon-normalizes-it-fig-1
- claim-3-21-74-step-efficiency-gain-over-adam-11-31-over-muon-at-1-1b-table-1
- claim-4-normuon-improves-validation-loss-trajectories-at-5-4b-fig-2
- claim-5-optimizer-memory-near-muon-2-9-step-time-overhead-vs-adamw-table-2
- claim-6-normuon-outperforms-muon-in-modded-nanogpt-at-124m-and-350m-fig-5
- 1.12 kB