Mathematical Reasoning on MATH 500, GSM8K, AIME 24, AMC 23, OlympiadBench, and Minerva Math (test)
46.92Final AccuracyAdaRFT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AdaRFTModel=Qwen2.5 7B, Setup=uniform2025.04 | 46.92 | 234.16 | 0 | 0 | 0 | |
| AdaRFTModel=Qwen2.5 7B, Setup=skew-difficult2025.04 | 46.83 | 239.92 | 0 | 0 | 0 | |
| AdaRFTModel=Qwen2.5 7B, Setup=skew-easy2025.04 | 45.94 | 247.44 | 0 | 0 | 0 | |
| PPO (w/ FC)Model=Qwen2.5 7B, Setup=skew-difficult2025.04 | 45.54 | 243.12 | 22.5 | 9 | 2,188.08 | |
| PPO (w/ Filter)Model=Qwen2.5 7B, Setup=skew-difficult2025.04 | 45.35 | 254.22 | 62.5 | 25 | 6,355.5 | |
| PPOModel=Qwen2.5 7B, Setup=skew-easy2025.04 | 45.07 | 235.27 | 50 | 20 | 4,705.4 | |
| PPO (w/ Filter)Model=Qwen2.5 7B, Setup=skew-easy2025.04 | 44.98 | 233.13 | 42.5 | 17 | 3,963.21 | |
| PPOModel=Qwen2.5 7B, Setup=uniform2025.04 | 44.7 | 243.82 | 32.5 | 13 | 3,169.66 | |
| PPO (w/ Filter)Model=Qwen2.5 7B, Setup=uniform2025.04 | 44.57 | 263.11 | 57.5 | 23 | 6,051.53 | |
| PPO (w/ FC)Model=Qwen2.5 7B, Setup=uniform2025.04 | 44.32 | 240.62 | 17.5 | 7 | 1,684.34 | |
| PPOModel=Qwen2.5 7B, Setup=skew-difficult2025.04 | 44.17 | 246.21 | 60 | 24 | 5,909.04 | |
| PPO (w/ FC)Model=Qwen2.5 7B, Setup=skew-easy2025.04 | 43.46 | 240.66 | 57.5 | 23 | 5,535.18 | |
| AdaRFTModel=Qwen2.5 Math 1.5B, Setup=uniform2025.04 | 41.11 | 121.31 | 0 | 0 | 0 | |
| AdaRFTModel=Qwen2.5 Math 1.5B, Setup=skew-difficult2025.04 | 40.48 | 122.24 | 0 | 0 | 0 | |
| AdaRFTModel=Qwen2.5 Math 1.5B, Setup=skew-easy2025.04 | 39.18 | 120.52 | 0 | 0 | 0 | |
| PPOModel=Qwen2.5 Math 1.5B, Setup=skew-easy2025.04 | 38.46 | 121.15 | 26.7 | 16 | 1,938.4 | |
| PPO (w/ Filter)Model=Qwen2.5 Math 1.5B, Setup=skew-easy2025.04 | 38.15 | 115.12 | 35 | 21 | 2,417.52 | |
| PPO (w/ FC)Model=Qwen2.5 Math 1.5B, Setup=skew-easy2025.04 | 38.09 | 121.99 | 58.3 | 35 | 4,269.65 | |
| PPO (w/ Filter)Model=Qwen2.5 Math 1.5B, Setup=uniform2025.04 | 37.87 | 126.35 | 86.7 | 52 | 6,570.2 | |
| PPOModel=Qwen2.5 Math 1.5B, Setup=skew-difficult2025.04 | 37.41 | 132.95 | 71.7 | 43 | 5,716.85 | |
| PPO (w/ FC)Model=Qwen2.5 Math 1.5B, Setup=uniform2025.04 | 37.27 | 126.4 | 80 | 48 | 6,067.2 | |
| PPO (w/ Filter)Model=Qwen2.5 Math 1.5B, Setup=skew-difficult2025.04 | 37.22 | 128.2 | 81.7 | 49 | 6,281.8 | |
| PPO (w/ FC)Model=Qwen2.5 Math 1.5B, Setup=skew-difficult2025.04 | 37.22 | 130.91 | 26.7 | 16 | 2,094.56 | |
| PPOModel=Qwen2.5 Math 1.5B, Setup=uniform2025.04 | 37.2 | 126.82 | 56.7 | 34 | 4,311.88 |