ProCoder

• 📙Paper: Iterative Refinement of Project-Level Code Context for Precise Code Generation with Compiler Feedback
• 📚Publisher: arxiv
• 🏠Author Affiliation: Huazhong University of Science and Technology
• Contribution:
  • An empirical study to analyze the error distribution in self-contained and repository-level code generation, highlighting the significance of precise and grounded program context in generating code at the project level;
  • A new iterative generation-verification-retrieval method that leverages the program compiler to eliminate context-related errors in repositorylevel code generation;
  • Extensive experiments and analysis based on two LLMs, i.e., GPT-3.5-Turbo and Code Llama (13B), validate the effectiveness of our proposed ProCoder.