Mathematical Reasoning on MATH (Accuracy, Loss)
40AccuracyAdam+Nexus
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Adam+NexusOptimizer=Adam+Nexus, Model Size=3B2026.04 | 40 | 1.026 | |
| MuonOptimizer=Muon, Model Size=3B2026.04 | 38 | 1.035 | |
| AdamWOptimizer=AdamW, Model Size=3B2026.04 | 32 | 1.054 |