Outline, Then Details: Syntactically Guided Coarse-To-Fine Code Generation

TL;DR

This paper introduces ChainCoder, a multi-pass, coarse-to-fine code generation approach that mimics human outlining, significantly improving the quality of generated Python code by leveraging hierarchical syntactic representations and progressive refinement.

Contribution

It proposes a novel multi-pass, hierarchical code generation method inspired by human outlining, enhancing large language models' ability to produce structured and accurate code.

Findings

Outperforms state-of-the-art code generation models

Progressive, multi-pass generation improves reasoning and code quality

Hierarchical syntactic representation aids in structured code synthesis

Abstract

For a complicated algorithm, its implementation by a human programmer usually starts with outlining a rough control flow followed by iterative enrichments, eventually yielding carefully generated syntactic structures and variables in a hierarchy. However, state-of-the-art large language models generate codes in a single pass, without intermediate warm-ups to reflect the structured thought process of "outline-then-detail". Inspired by the recent success of chain-of-thought prompting, we propose ChainCoder, a program synthesis language model that generates Python code progressively, i.e. from coarse to fine in multiple passes. We first decompose source code into layout frame components and accessory components via abstract syntax tree parsing to construct a hierarchical representation. We then reform our prediction target into a multi-pass objective, each pass generates a subsequence, which is concatenated in the hierarchy. Finally, a tailored transformer architecture is leveraged to jointly encode the natural language descriptions and syntactically aligned I/O data samples. Extensive evaluations show that ChainCoder outperforms state-of-the-arts, demonstrating that our progressive generation eases the reasoning procedure and guides the language model to generate higher-quality solutions. Our codes are available at: https://github.com/VITA-Group/ChainCoder.