ResearchBenchmarksHelpfulness on PKU-SafeRLHF 30KFollow84.5Win RateSafeDPO65.135270.162675.1980.2174May 26, 2025Evaluation ResultsMethodMethodLinksWin RateTie RateLose RateSafeDPOBaseline=DPO-HELPFULBaseline=DPO-HELPFUL2025.0584.50.1315.38SafeDPOBaseline=DPO-HARMLESSBaseline=DPO-HARMLESS2025.0583.389.2516.38SafeDPOBaseline=DPO-SAFEBETTERBaseline=DPO-SAFEBETTER2025.0582.070.2517.68SafeDPOBaseline=SACPOBaseline=SACPO2025.05760.3823.63SafeDPOBaseline=P-SACPOBaseline=P-SACPO2025.0570.75029.25SafeDPOBaseline=SafeRLHFBaseline=SafeRLHF2025.0565.88034.13