CodeSearchNet-Python
收藏arXiv2025-08-16 更新2025-11-27 收录
下载链接：
https://zenodo.org/records/13755652
下载链接
链接失效反馈官方服务：
资源简介：
CodeSearchNet-Python数据集是从真实的GitHub项目中收集的，包含2百万个代码与注释对，涵盖了Python等六种编程语言。该数据集已被广泛用作大型语言模型训练的基准数据集。论文中发现该数据集中存在大量的代码异味，并开发了名为SmellCC的代码异味清理工具，用于自动重构和清理代码中的异味，从而提高数据集质量，进而提升大型语言模型在代码生成、代码补全等下游任务中的性能。
The CodeSearchNet-Python dataset is collected from real-world GitHub projects, containing 2 million code-comment pairs covering six programming languages including Python. This dataset has been widely utilized as a benchmark dataset for training large language models. It was found in the literature that there are a large number of code smells in this dataset, and a code smell cleaning tool named SmellCC was developed to automatically refactor and remove code smells, thereby improving the quality of the dataset and further enhancing the performance of large language models in downstream tasks such as code generation and code completion.
提供机构：
浙江大学，中国
创建时间：
2025-08-16
