ResearchBenchmarksRoleplay Personalization on Roleplay Human Evaluation (50 users, 11 questions)Follow72.3WinrateFSPO68.03669.14370.2571.357Feb 26, 2025Evaluation ResultsMethodMethodLinksWinrateFSPOBaseline Method=SFTBaseline Method=SFT2025.0272.3FSPOBaseline Method=BaseBaseline Method=Base2025.0268.2