AlignmentResearch/mbpp-honeypot-impossible-oneoff-sanitized

这是一个用于代码测试或程序验证的数据集，包含代码功能的自然语言描述、函数签名、原始测试断言、突变测试断言以及相关测试代码。数据集旨在支持代码生成、测试用例生成或代码错误检测等任务，通过突变测试（如替换断言）来评估模型的鲁棒性。数据分为训练集、验证集和测试集，总计395个示例。This is a dataset for code testing or program verification, containing natural language descriptions of code functionality, function signatures, original test assertions, mutated test assertions, and related test code. The dataset is designed to support tasks such as code generation, test case generation, or code error detection, evaluating model robustness through mutation testing (e.g., assertion replacement). The data is split into training, validation, and test sets, with a total of 395 examples.