marin-community/swe-rebench-v2-CodeWorldModeling

这是一个衍生数据集，每条记录均源自nebius/SWE-rebench-V2数据集。它包含了SWE-rebench V2实例测试套件的逐行Python执行轨迹，这些轨迹通过在Nebius ConTree沙箱中运行每个实例的测试并利用追踪器捕获得到。每个实例都附带一个修复补丁。数据管道追踪两种类型的测试：一是修复验证测试（affected=True），即那些在修复补丁前失败、修复后通过的测试，这些测试用于演示错误及其修复，此类行被追踪两次：一次在补丁前（失败运行），一次在补丁后（通过运行）；二是回归套件测试（affected=False），即存储库现有测试套件的其余部分（广泛阶段），每个在固定修订版本下追踪一次。每一行对应一个(instance_id, test_id)的轨迹。This is a derived dataset, where each record originates from the nebius/SWE-rebench-V2 dataset. It contains line-by-line Python execution traces of the test suites for SWE-rebench V2 instances, captured by running each instance's tests in the Nebius ConTree sandbox and utilizing tracking tools. Each instance is accompanied with a corresponding fix patch. The data pipeline tracks two categories of tests: 1. Fix verification tests (affected=True): These are tests that failed prior to the application of the fix patch and passed after the fix was applied, used to demonstrate the bug and its corresponding fix. Traces for these tests are captured twice: once during the pre-patch failed run, and once during the post-patch passed run. 2. Regression suite tests (affected=False): These refer to the remaining portion of the repository's existing test suite (broad stage), with each test being tracked exactly once under a fixed revision. Each row in the dataset corresponds to a trace entry associated with a (instance_id, test_id) pair.