ResearchBenchmarksRole-playing Instruction Following on RoleBench English Role GeneralizationFollow64.5Win Rate (GPT-4)RoleLLaMA6.57221.61136.6551.689Oct 1, 2023Evaluation ResultsMethodMethodLinksWin Rate (GPT-4)Win Rate (Human)RoleLLaMAParameters=7BParameters=7B2023.1064.556.1Vicuna2023.103132.4Alpaca2023.101228.2ChatPLUG2023.108.812