Evaluating Large Language Models for Repository-Level Program Repair

The authors investigate the performance of popular LLMs in handling repository-level repair tasks and propose a new benchmark, RepoBugs, along with a context extraction method (RLCE) to enhance repair accuracy significantly.