Mathematical Reasoning on MATH 500 (M→M-500 and G8→M-500 Accuracy)
78.2Accuracy (M→M-500)GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPOModel Size=7B2025.12 | 78.2 | 45.6 | |
| iCLPModel Size=7B2025.12 | 68.6 | 46.2 | |
| GRPOModel Size=3B2025.12 | 60.8 | 40.4 | |
| FT-SModel Size=7B2025.12 | 58.4 | 44 | |
| iCLPModel Size=3B2025.12 | 55.4 | 44.2 | |
| FT-EModel Size=7B2025.12 | 52 | 43.8 | |
| FT-SModel Size=3B2025.12 | 48.4 | 40 | |
| FT-EModel Size=3B2025.12 | 43 | 39.8 | |
| BaseModel Size=7B2025.12 | 42.4 | 42.4 | |
| BaseModel Size=3B2025.12 | 39.6 | 39.6 | |
| GRPOModel Size=0.5B2025.12 | 34.4 | 17.9 | |
| iCLPModel Size=0.5B2025.12 | 28.1 | 20.8 | |
| FT-SModel Size=0.5B2025.12 | 27.2 | 18.4 | |
| FT-EModel Size=0.5B2025.12 | 18.8 | 15.8 | |
| BaseModel Size=0.5B2025.12 | 15.8 | 15.8 | |
| FT-LEModel Size=0.5B2025.12 | 11.2 | 9.6 |