CodeSecEval is an execution-based benchmark for evaluating large language models on secure code generation and insecure-code repair.
The benchmark contains 255 Python programming tasks spanning 77 CWE vulnerability categories. Each task provides a problem specification, an insecure implementation, a secure reference implementation, executable tests, and an entry point.
SecEvalBase: 115 tasks… See the full description on the dataset page:
https://huggingface.co/datasets/JasonWang1/CodeSecEval.
