List your company

Science Question Answering on GPQA (Mean Accuracy)

24.6Mean AccuracyOPD
0.0566.42812.819.172Feb 16, 2026

Evaluation results

MethodLinks
OPDTraining Budget=sufficient, Rollout=full, Zero-shot=true2026.0224.6
OPD (prefix scheduling)Training Budget=sufficient, Zero-shot=true2026.0223.6
SeqKDSteps=300, Training Budget=limited, Zero-shot=true2026.0220.2
SeqKDTraining Budget=sufficient, Zero-shot=true2026.0219.8
OPDSteps=10, Training Budget=limited, Zero-shot=true2026.0218.3
Qwen3-1.7B-BaseZero-shot=true2026.0211.2
OPDPrefix Length=2048, Training Budget=limited, Zero-shot=true2026.024.5
OPDPrefix Length=512, Training Budget=limited, Zero-shot=true2026.021.1
OPDPrefix Length=1024, Training Budget=limited, Zero-shot=true2026.021.1
OPDPrefix Length=256, Training Budget=limited, Zero-shot=true2026.021