ConDefects
收藏资源简介:
ConDefects是由北京化工大学和伦敦国王学院联合创建的数据集,专注于解决大型语言模型(LLMs)在故障定位和程序修复中的数据泄露问题。该数据集包含1254个Java错误程序和1625个Python错误程序,均来源于AtCoder在线竞赛平台,涵盖2021年10月至2023年9月间的数据。每个错误程序都配有错误位置和修复版本,适用于故障定位和自动程序修复研究。ConDefects支持根据时间窗口和编程任务难度选择子集,提供用户友好的接口进行测试案例执行和覆盖收集,广泛应用于评估不同故障定位和程序修复方法。
ConDefects is a dataset jointly created by Beijing University of Chemical Technology and King's College London, focusing on addressing data leakage issues in large language models (LLMs) during fault localization and program repair. It contains 1,254 faulty Java programs and 1,625 faulty Python programs, all sourced from the AtCoder online contest platform, covering data from October 2021 to September 2023. Each faulty program is equipped with its bug location and fixed version, making it applicable to research on fault localization and automated program repair. ConDefects supports subset selection based on time windows and programming task difficulty, and provides a user-friendly interface for test case execution and coverage collection, which is widely used to evaluate different fault localization and program repair methods.

- 1ConDefects: A New Dataset to Address the Data Leakage Concern for LLM-based Fault Localization and Program Repair北京化工大学 · 2023年



