utkarshd/roleplay_small

这是一个小型数据集，旨在增强大型语言模型（LLMs）的角色扮演能力。其主要目标是确保LLMs在角色扮演中不脱离角色，避免过度帮助。这是通过确保模型在问题或对话超出其角色定义时（例如，向二战士兵询问Python语言）礼貌地回应来实现的。数据集是通过从Hugging Face Hub上的两个特定数据集（IlyaGusev/gpt_roleplay_realm和dim/roleplay_instruct_v2_final）中选择特定实例并手动注释一些对话来创建的。

This is a small dataset meant for enhancing roleplay abilities of LLMs. The primary objective is to ensure LLMs stick to the roleplay and do not break character in order to be overly helpful. This is achieved by making sure the LLMs respond (politely, in some cases) that the question/conversation is beyond their character definition - such as asking a World War II soldier about the Python language. The dataset is created by selecting specific instances from the following datasets from the Hub and manually annotating some conversations: IlyaGusev/gpt_roleplay_realm and dim/roleplay_instruct_v2_final.

原始信息汇总

数据集概述

数据集用途

• 用于增强大型语言模型（LLMs）的角色扮演能力。
• 确保LLMs在角色扮演中不脱离角色，避免过度帮助。

数据集特点

• LLMs在超出角色定义的问题或对话中，会礼貌地回应，表明问题超出其角色范围。

数据集来源

• 从以下数据集中选择特定实例并手动注释：
  1. IlyaGusev/gpt_roleplay_realm
  2. dim/roleplay_instruct_v2_final

许可证

• CC BY 4.0