Mathematical Reasoning on AIME 24 (Mean Accuracy)
10.6Mean AccuracyOPD (prefix scheduling)
Evaluation Results
| Method | Links | |
|---|---|---|
| OPD (prefix scheduling)Training Budget=sufficient, Zero-shot=true2026.02 | 10.6 | |
| OPDTraining Budget=sufficient, Rollout=full, Zero-shot=true2026.02 | 10 | |
| SeqKDTraining Budget=sufficient, Zero-shot=true2026.02 | 8.8 | |
| OPDPrefix Length=1024, Training Budget=limited, Zero-shot=true2026.02 | 8.1 | |
| OPDPrefix Length=2048, Training Budget=limited, Zero-shot=true2026.02 | 7.3 | |
| OPDSteps=10, Training Budget=limited, Zero-shot=true2026.02 | 5.4 | |
| OPDPrefix Length=256, Training Budget=limited, Zero-shot=true2026.02 | 4.8 | |
| SeqKDSteps=300, Training Budget=limited, Zero-shot=true2026.02 | 4.6 | |
| OPDPrefix Length=512, Training Budget=limited, Zero-shot=true2026.02 | 4.2 | |
| Qwen3-1.7B-BaseZero-shot=true2026.02 | 0.2 |