IconicAI/janet-24oct

该数据集包含39个由GPT-4生成的多轮对话，每个对话代表狙击手和中央操作员Janet之间的独特对话。对话的背景是狙击手在屋顶上执行任务，Janet指导狙击手识别并消除隐藏在人群中的危险目标。数据集以两种格式提供：原始的`.txt`格式和Axolotl兼容的ShareGPT版本`.sharegpt.jsonl`。

This dataset contains 39 multi-turn conversations generated by GPT-4, each representing a distinct dialogue between a sniper and a central operator named Janet. The context of each conversation is that the sniper is perched on a rooftop, on a mission to eliminate a dangerous person hiding in the crowd below. Janet, who knows the appearance of the person, guides the sniper through the process of identifying and eliminating the target. The dataset is in JSON format, including the conversation ID and content, with each dialogue line marked by the speaker type and content.

原始信息汇总

数据集描述

39个由GPT-4生成的多轮对话。每个对话代表两个人之间的独特对话——一个狙击手和一个名为Janet的中央操作员。每个对话的背景相同——狙击手位于屋顶上，执行任务以消灭隐藏在下方人群中的危险人物。Janet知道该人物的外观，并通过对话引导狙击手识别并消灭目标。

数据结构

json { "id": "字符串", "conversation": [{ "from": "字符串", # system - system-prompt, assistant - LLM, user - human "value": "字符串", # 对话内容 }] }

实现细节

原始格式为.txt，Axolotl兼容的ShareGPT版本为.sharegpt.jsonl。

引用

@misc{Janet24Oct, title = {Janet24Oct: A Dataset of Multi-Turn Conversations for Conversational AI}, author = {Kieran Donaldson and Mikel Bober-Izar and Piotr Trochim}, year = {2023}, publisher = {HuggingFace}, journal = {HuggingFace repository}, howpublished = {url{https://huggingface.co/datasets/IconicAI/janet-24oct}}, }