DOCE: Finding the Sweet Spot for Execution-Based Code Generation

TL;DR

This paper introduces DOCE, a comprehensive framework for execution-based code generation that compares various decoding and reranking methods, emphasizing execution-based evaluation and self-debugging to improve performance.

Contribution

The paper proposes a unified framework for execution-based code generation, systematically evaluating different components and introducing self-debugging for state-of-the-art reranking results.

Findings

Execution-based methods outperform execution-free approaches.

Filtering with trial unit tests significantly improves code quality.

Self-debugging on multiple candidates achieves state-of-the-art reranking performance.

Abstract

Recently, a diverse set of decoding and reranking procedures have been shown effective for LLM-based code generation. However, a comprehensive framework that links and experimentally compares these methods is missing. We address this by proposing Decoding Objectives for Code Execution, a comprehensive framework that includes candidate generation, $n$-best reranking, minimum Bayes risk (MBR) decoding, and self-debugging as the core components. We then study the contributions of these components through execution-based evaluation metrics. Our findings highlight the importance of execution-based methods and the difference gap between execution-based and execution-free methods. Furthermore, we assess the impact of filtering based on trial unit tests, a simple and effective strategy that has been often overlooked in prior works. We also propose self-debugging on multiple candidates, obtaining state-of-the-art performance on reranking for code generation. We expect our framework to provide a solid guideline for future research on code generation.