Abstract: Large language models (LLMs) have made significant strides in code generation but often struggle with API hallucination issues, especially for the third-party library. Existing approaches ...