ResearchBenchmarksHuman Preference Alignment on PKU-SafetyFollow67.1Win RateDPO-HPS55.6658.6361.664.57Feb 20, 2025Evaluation ResultsMethodMethodLinksWin RateLose RateTie RateDPO-HPSComparison Baseline=SF...Comparison Baseline=SFT, setting=fine-tuning2025.0267.15.1527.75DPO-HPSComparison Baseline=DP...Comparison Baseline=DPO-PL, setting=fine-tuning2025.0257.6510.3532DPO-HPSComparison Baseline=DP...Comparison Baseline=DPO-BT, setting=fine-tuning2025.0256.111.1532.75