nebius/SWE-rebench-leaderboard

SWE-rebench-leaderboard是一个针对软件工程代理在现实世界任务上进行基准测试的持续更新、精选的数据集子集。它包含了用于SWE-rebench排行榜的任务，并为每个任务提供了预构建的Docker镜像。数据集结构包括实例ID、补丁、仓库、基础提交、测试补丁、问题陈述、创建时间等多个字段，以及用于设置和安装环境的安装配置字段。

SWE-rebench-leaderboard is a continuously updated, curated subset of the SWE-rebench corpus, designed for benchmarking software engineering agents on real-world tasks. It includes tasks used in the SWE-rebench leaderboard and provides pre-built Docker images for each task. The dataset structure features fields such as instance ID, patch, repository, base commit, test patch, problem statement, creation date, and installation configuration for setting up and installing the environment.