Human Response-Guided Evaluation of Instruction Following (HREF)
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/datasets/allenai/href_preference
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4,258个样本，覆盖了11个任务类别，旨在提高自动评估在遵循指令任务中的可靠性。数据集中不仅包含了人类编写的回应和模型生成的回应，还包含了人类评判员对模型偏好进行的标注。这一规模的数据集为指令遵循评估任务提供了丰富的数据支持。
This dataset contains 4,258 samples spanning 11 task categories, aiming to enhance the reliability of automatic evaluation for instruction-following tasks. It includes not only human-written responses and model-generated responses, but also annotations of human annotators' preferences towards model outputs. The scale of this dataset provides ample data support for instruction-following evaluation tasks.
提供机构：
Allen Institute for AI
