Mathematical Reasoning on AIME 24, AIME 25, AMC, Minerva, MATH-500, Olympiad, Gaokao23 (test)
43.3AIME 2024 ScoreScaf-GRPO
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Scaf-GRPOBackbone=Qwen2.5-Math-7B2025.10 | 43.3 | 20 | 70 | 36.4 | 80 | 43.3 | 63.4 | 50.9 | |
| DAPOBackbone=Qwen2.5-Math-7B2025.10 | 36.7 | 20 | 62.5 | 35.6 | 79.6 | 42.3 | 63 | 48.5 | |
| Vanilla GRPOBackbone=Qwen2.5-Math-7B2025.10 | 30 | 13.3 | 60 | 33.4 | 75.8 | 41.3 | 62.6 | 45.2 |