AmazonScience/SWE-PolyBench

SWE-PolyBench是一个面向代码仓库级别的多语言软件工程基准测试数据集，包含了Python、Java、JavaScript和TypeScript四种编程语言。数据集由两个部分组成：完整的数据集和包含500个实例的分层抽样数据集。数据集的每一行记录了包括实例ID、补丁、仓库信息、基础提交哈希、提示文本、创建日期、测试补丁、问题陈述、测试通过情况列表、编程语言、Dockerfile、测试命令、任务类别以及一些用于评估检索指标的信息。

SWE-PolyBench is a multi-language software engineering benchmark for repository-level evaluation, which includes four programming languages: Python, Java, JavaScript, and TypeScript. The dataset consists of two parts: the full dataset and a stratified sampled dataset with 500 instances. Each row in the dataset records information such as instance ID, patch, repository information, base commit hash, hint text, creation date, test patch, problem statement, test pass lists, programming language, Dockerfile, test command, task category, and some information for evaluating retrieval metrics.