Stack-exchange-preferences

该数据集包含来自Stack Overflow数据转储的问题和答案，用于偏好模型训练。重要的是，这些问题已被过滤以符合以下偏好模型的标准 (紧随Askell等人2021): 有> = 2个答案。此数据也可用于指令微调和语言模型训练。

This dataset comprises questions and answers extracted from Stack Overflow data dumps, intended for preference model training. Importantly, these questions have been filtered to meet the preference model criteria proposed by Askell et al. 2021: each question has at least 2 answers. This dataset can also be utilized for instruction fine-tuning and language model training.

搜集汇总

背景与挑战

背景概述

该数据集包含来自Stack Overflow的问题和答案，经过过滤以符合偏好模型训练的标准（每个问题至少有2个答案），适用于偏好模型训练、指令微调和语言模型训练。数据集由Anthropic AI于2021年12月发布，遵循CC BY-SA 4.0许可。