Abstract: Large language models (LLMs) have shown impressive capabilities in coding tasks, including code understanding and generation. However, these models are also susceptible to input ...