Arena-Hard
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/lm-sys/arena-hard
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个包含805个问题的基准测试集，专为通用指令遵循评估而设计。在评估时，采用了长度控制的胜率（LC）作为评价指标，任务类型为指令遵循。
This dataset is a benchmark test set containing 805 questions, specifically designed for general instruction following evaluation. During the evaluation process, the Length-Controlled Win Rate (LC) was adopted as the evaluation metric, and the task type of this benchmark is instruction following.
