Nitral-Archive/Project_Persona-ShareGPT

---
license: other
language:
- en
---
Data sourced from: https://huggingface.co/datasets/proj-persona/PersonaHub (the persona, tools json were not used. Since its redundant or not applicable in this setup )

Do not train on this as a direct dataset.
This is meant to be used in a synthetic data creation pipeline.
The 'human' turn corresponds to an input describing a persona, while the 'gpt' turn adds onto the relevant input providing additional (instruction, reasoning, mathmatics, knowledge, npc generation). These are broken up over several conversation keys in the dataset.
Allowing you to use this in a two shot setup for generating additional synthetic data bound to specific persona's.

Slop/Refusals/Duplicates and keys ending on human filtered.
Grammar Corrected and Converted.
The above was done with: https://github.com/The-Chaotic-Neutrals/ShareGPT-Formaxxing

许可证：其他
语言：
- 英语

数据集来源：https://huggingface.co/datasets/proj-persona/PersonaHub（本数据集未使用其中的persona、tools json文件，因其在当前场景中存在冗余或不适用的情况）

请勿将本数据集直接用于模型训练。本数据集仅用于合成数据生成流水线。

对话中的“人类轮次”对应描述人设（persona）的输入内容，而“GPT轮次”则基于该相关输入补充提供额外信息（涵盖指令、推理、数学、知识、非玩家角色（NPC）生成等内容）。上述内容被拆分至数据集的多个对话键中，可将本数据集用于少样本（Few-shot）场景，以生成绑定至特定人设的额外合成数据。

已过滤低质量内容、模型拒答样本、重复样本及以human结尾的对话键。已完成语法修正与格式转换。上述处理均通过工具 https://github.com/The-Chaotic-Neutrals/ShareGPT-Formaxxing 完成。