OpenAssistant Conversation Dataset （OASST1）

为了使大规模对齐的研究民主化，我们发布了OpenAssistant Conversations (OASST1)，这是一种人类生成的，人类注释的助手风格的对话语料库，由35种不同语言的161,443条消息组成，并以461,292的质量评级进行注释，导致超过10,000个完全注释的对话树。该语料库是涉及13,500多名志愿者的全球众包工作的产物。

To democratize research on large-scale alignment, we release OpenAssistant Conversations (OASST1), a human-generated, human-annotated assistant-style conversational corpus consisting of 161,443 messages across 35 distinct languages, annotated with 461,292 quality ratings, resulting in over 10,000 fully annotated conversation trees. This corpus is the product of a global crowdsourcing effort involving more than 13,500 volunteers.

搜集汇总

背景与挑战

背景概述

OpenAssistant Conversations (OASST1) 是一个人类生成和注释的助手风格对话语料库，包含35种语言的161,443条消息和461,292个质量评级，覆盖超过10,000个对话树。该数据集由13,500多名志愿者通过全球众包工作创建，旨在推动大规模对齐研究的民主化。