Reasoning on ARC Easy (test)
53Normalized AccuracyAdamW+CWD
Evaluation results
| Method | Links | |
|---|---|---|
| AdamW+CWDOptimizer=AdamW+CWD, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 53 | |
| MuonOptimizer=Muon, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 51 | |
| Muon+CWDOptimizer=Muon+CWD, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 51 | |
| AdamWOptimizer=AdamW, Model Parameters=1B, Training Tokens=100B, Codebase=OLMo2025.10 | 50 |