Code generation aims to produce code that fulfills requirements written in natural language automatically. Large Language Models (LLMs) like ChatGPT have demonstrated promising effectiveness in this area. Nonetheless, the LLMs often fail to ensure the syntactic and semantic correctness of the generated code. Recently, researchers proposed multi-agent frameworks that guide LLMs with different prompts to analyze programming tasks, generate code, and perform testing in a sequential workflow. However, the performance of the workflow is not robust as code generation depends on the performance of each agent. To address this challenge, we propose CodeCoR, a multi-agent framework that prunes intermediate outputs at different stages to reduce error propagation in sequential code generation workflows. Specifically, for a given task description, four agents in CodeCoR generate prompts, code, test cases, and repair advice, respectively. Each agent generates more than one output and prunes away the low-quality ones. The generated code is tested in the local environment: the code that fails to pass the generated test cases is sent to the repair agent, and the coding agent re-generates the code based on repair advice. Finally, the code that passes the highest number of generated test cases is returned to the users. Our experiments on four widely used datasets, HumanEval, HumanEval-ET, MBPP, and MBPP-ET, demonstrate that CodeCoR outperforms existing baselines (e.g., CodeCoT and MapCoder), achieving an average Pass@1 score of 77.13%.

翻译：暂无翻译