ResearchBenchmarksHarmlessness Evaluation on PKU-SafeRLHF-30KFollow87.25Win RateSafeDPO56.705264.635172.56580.4949May 26, 2025Evaluation ResultsMethodMethodLinksWin RateTie RateLose RateSafeDPOBaseline=DPO-HELPFULBaseline=DPO-HELPFUL2025.0587.253.759SafeDPOBaseline=DPO-SAFEBETTERBaseline=DPO-SAFEBETTER2025.0585.983.7910.23SafeDPOBaseline=DPO-HARMLESSBaseline=DPO-HARMLESS2025.0579.754.515.75SafeDPOBaseline=SACPOBaseline=SACPO2025.0574.755.7519.5SafeDPOBaseline=P-SACPOBaseline=P-SACPO2025.05726.521.5SafeDPOBaseline=SafeRLHFBaseline=SafeRLHF2025.0557.885.7536.38