OS-Kairos Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Wuzheng02/OS-Kairos
下载链接
链接失效反馈官方服务：
资源简介：
该数据集精心挑选了复杂场景，用以评估图形用户界面代理的表现。评估的性能指标包括动作类型成功率、逐步成功率和任务成功率。该数据集的规模为从三个复杂场景中选取的350条指令，其任务针对的是多语言语言模型的序列生成问题。
This dataset curates complex scenarios to evaluate the performance of graphical user interface (GUI) agents. The performance metrics for the evaluation include action type success rate, step-by-step success rate, and task success rate. The dataset consists of 350 instructions selected from three complex scenarios, and the tasks therein target the sequence generation problem for multilingual large language models (LLMs).
提供机构：
Authors of the paper
