AmuroEita/SWEBench-FileTree

SWEBench-FileTree数据集是SWEbenchCodeRetrieval数据集的一个仅包含路径的版本，专注于路由基准测试而非代码检索。它通过GitHub问题文本和仓库快照的文件路径列表（不包含内容），评估路径令牌、文件名、扩展名和目录结构在识别需要修改的文件方面的效果。数据集包含多种配置的查询、语料库和相关文件，详细说明了模式、评估指标和难度等级。此外，还提供了数据集的大小、结构、加载方法以及预期用途和限制。

SWEBench-FileTree is a path-only version of the SWEbenchCodeRetrieval dataset, focusing on routing benchmarks rather than code retrieval. It measures the effectiveness of path tokens, filenames, extensions, and directory structure in identifying files that need modification based on GitHub issue texts and repository snapshots represented by file paths alone. The dataset includes various configurations for queries, corpus, and qrels, with detailed information on the schema, evaluation metrics, and difficulty tiers. It also provides specifics on the datasets size, structure, loading instructions, intended use, and limitations.