NoCode-bench
收藏资源简介:
NoCode-bench是一个用于评估语言模型在自然语言驱动特征添加任务上的能力的基准数据集。该数据集包含634个实例,涉及10个开源项目,总共约114k代码更改。每个实例都包含用户文档更改和相应的代码实现,并可以使用开发者编写的测试用例进行验证。数据集的构建过程包括五个阶段:项目选择、实例收集、环境构建、实例筛选和输入精炼。此外,还提供了一个名为NoCode-bench Verified的人工验证子集,包含114个高质量实例,其任务清晰度和评估有效性已手动验证。该数据集可用于推动基于语言模型软件工程的研究,特别是在无代码特征添加方面。
NoCode-bench is a benchmark dataset for evaluating the capability of language models on natural language-driven feature addition tasks. This dataset contains 634 instances, covering 10 open-source projects, with a total of approximately 114k code changes. Each instance includes user documentation changes and corresponding code implementations, and can be validated using test cases written by developers. The construction process of the dataset consists of five stages: project selection, instance collection, environment setup, instance filtering, and input refinement. In addition, a manually verified subset named NoCode-bench Verified is provided, which contains 114 high-quality instances whose task clarity and evaluation validity have been manually verified. This dataset can be used to advance research in language model-based software engineering, particularly in the area of no-code feature addition.




