ResearchBenchmarksDecision Inference on MMLUFollow0.772AccuracyOurs0.685680.708090.73050.75291Feb 18, 2025Evaluation ResultsMethodMethodLinksAccuracyOurs2025.020.772KTOFramework=RLHFFramework=RLHF2025.020.753DPOFramework=RLHFFramework=RLHF2025.020.746ReFTFramework=ReasoningFramework=Reasoning2025.020.743SkyworkFramework=RLAIFFramework=RLAIF2025.020.737PPOFramework=RLHFFramework=RLHF2025.020.723DeepSeek2025.020.721o3-mini2025.020.707CoT+SFTFramework=ReasoningFramework=Reasoning2025.020.705Proxy LLM2025.020.703SFT2025.020.689