test2

数据集介绍
简介
大模型多轮对话SFT数据集_中文由来自中国7个省份 (江苏、四川、山东、山西、北京、广东、海南)的663名说话人独家贡献，北京晴数智慧科技有限公司进行采集。每组对话由两名说话人围绕一个主题展开，历史的对话与当前的内容密切相关。适用于训练大模型多轮对话 (back and forth conversation)、上下文逻辑推理能力。

数据组成
包含16932轮中文自然对话句子，涉及【个人特质、家庭生活、健康、教育、经济、军事战争、科技、科学技术、气候环境、人际关系、数码产品、体育竞技、文旅、休闲娱乐、衣食住行、艺术美术、政治法律、职业发展】18个主题。

Statistics MagicData-CLAM-Conversation_CN
#Dialogue 343
#Turns 16932
#Topics 18
Avg. #Turns per Dialogue 49
Avg. #Tokens per Turn 80

Dataset Introduction
Overview
Chinese Multi-turn Dialogue SFT Dataset for Large Models is exclusively contributed by 663 speakers from 7 provinces in China (Jiangsu, Sichuan, Shandong, Shanxi, Beijing, Guangdong, Hainan), and collected by Beijing Qingshu Wisdom Technology Co., Ltd. Each dialogue is conducted by two speakers around a specific topic, with historical utterances closely related to the current content. This dataset is suitable for training the multi-turn dialogue (back-and-forth conversation) capability and contextual logical reasoning ability of large language models.

Data Composition
This dataset contains 16,932 turns of natural Chinese conversational sentences, covering 18 topics including: personal traits, family life, health, education, economy, military warfare, technology, climate and environment, interpersonal relationships, digital products, sports competitions, cultural tourism, leisure and entertainment, basic necessities of life, art and fine arts, politics and law, career development.

Statistics
MagicData-CLAM-Conversation_CN
Number of Dialogues: 343
Number of Turns: 16,932
Number of Topics: 18
Average Number of Turns per Dialogue: 49
Average Number of Tokens per Turn: 80

搜集汇总

背景与挑战

背景概述

该数据集名为test2，由Huanhuan/DataLab创建，属于文本自然语言处理与对话生成类别，目前为公开状态但暂无具体数据文件。