Reasoning on ARC Challenge (test)
28Normalized AccuracyMuon+CWD
Evaluation results
| Method | Links | |
|---|---|---|
| Muon+CWDOptimizer=Muon+CWD, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 28 | |
| AdamW+CWDOptimizer=AdamW+CWD, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 27 | |
| MuonOptimizer=Muon, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 26 | |
| AdamWOptimizer=AdamW, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 25 |