LCA Code Editing

该数据集包含经过精心挑选的提交信息，这些信息作为自然语言指令，并对应着作为目标的代码变更。该数据集特别关注大规模的代码改动，相较于SWE-bench Lite，其黄金补丁中的平均行数显著增多，这使得上下文检索更具挑战性。该数据集的规模为119个数据点，其任务聚焦于仓库级别的代码编辑。

This dataset contains carefully curated commit messages that serve as natural language instructions paired with corresponding target code changes. This dataset specifically focuses on large-scale code modifications. Compared with SWE-bench Lite, the average number of lines in its golden patches is significantly higher, which makes context retrieval more challenging. The dataset consists of 119 data points, with its tasks focused on repository-level code editing.