CodeProjectEval
收藏资源简介:
CodeProjectEval数据集由北京航空航天大学和华为中央软件研究院共同构建,旨在支持端到端的项目级代码生成任务。该数据集由18个高质量的真实世界代码库组成,每个任务平均包含12.7个源文件和2388.6行代码,并补充了反映用户需求的相关文档。与现有数据集相比,CodeProjectEval更好地捕捉了现实软件开发中的规模和结构复杂性,并提供了可执行的测试用例,以支持自动评估。
The CodeProjectEval dataset, co-constructed by Beihang University and Huawei Central Software Institute, is designed to support end-to-end project-level code generation tasks. Comprising 18 high-quality real-world code repositories, each task in this dataset contains an average of 12.7 source files and 2388.6 lines of code, and is supplemented with relevant documentation reflecting user requirements. Compared with existing datasets, CodeProjectEval better captures the scale and structural complexity of real-world software development, and provides executable test cases to support automatic evaluation.




