Summarize from Feedback
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/openai/summarize-from-feedback
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了针对摘要任务的由人类编写的指令跟随示例和偏好配对。具体来说，它包括了一个训练分割，其中包含人类编写的完成情况，以及一个用于偏好学习的偏好分割。数据集的规模为116千个示例（SFT分割）和93千个偏好配对，专门针对摘要任务而设计。
This dataset comprises human-written instruction-following examples and preference pairs tailored for summarization tasks. Specifically, it includes two splits: a Supervised Fine-Tuning (SFT) training split containing human-written completions, and a preference split dedicated to preference learning. The dataset consists of 116 thousand examples in the SFT split and 93 thousand preference pairs, and is specifically designed for summarization tasks.
提供机构：
OpenAI
