RoleBench
收藏资源简介:
RoleBench是由中国科学院大学团队开发的首个系统性和细粒度的角色扮演基准数据集,包含168,093个样本,涵盖多种角色和语言风格。该数据集通过四个阶段构建:角色简介构建、基于上下文的指令生成、角色提示使用GPT(RoleGPT)和角色条件指令调整(RoCIT)。RoleBench旨在通过Context-Instruct和RoleGPT生成的高质量QA对,提取角色特定知识,增强开源模型的角色扮演能力,解决现有模型在角色扮演优化上的限制。该数据集的应用领域包括自然语言处理和人工智能,特别是在提升大型语言模型在复杂任务如角色扮演中的表现。
RoleBench is the first systematic and fine-grained role-play benchmark dataset developed by the team from the University of Chinese Academy of Sciences, containing 168,093 samples covering diverse roles and linguistic styles. This dataset is constructed in four stages: role profile construction, context-based instruction generation, role prompt utilization via GPT (dubbed RoleGPT), and role-conditioned instruction tuning (RoCIT). RoleBench aims to extract role-specific knowledge from high-quality QA pairs generated by Context-Instruct and RoleGPT, so as to enhance the role-playing capabilities of open-source models and address the limitations of existing models in role-play optimization. Its application fields include natural language processing and artificial intelligence, particularly for improving the performance of large language models in complex tasks such as role-playing.

- 1RoleLLM: Benchmarking, Eliciting, and Enhancing Role-Playing Abilities of Large Language Models中国科学院大学 · 2024年



