Mathematical Reasoning on Numina (test)
53.1Pass@8Muonp
Evaluation results
| Method | Links | |
|---|---|---|
| MuonpLearning rate=1 × 10−4, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 53.1 | |
| MuonLearning rate=1 × 10−4, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 50.6 | |
| MuonpLearning rate=5 × 10−5, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 49.6 | |
| MuonLearning rate=5 × 10−5, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 45.5 | |
| MuonpLearning rate=1 × 10−5, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 36.1 | |
| MuonLearning rate=1 × 10−5, Backbone=Llama3-3B, Training tokens=786 million, Temperature=0.8, Generations=322026.06 | 29.6 |