Timezone: »

 
Poster
Outline, Then Details: Syntactically Guided Coarse-To-Fine Code Generation
Wenqing Zheng · S P Sharan · Ajay Jaiswal · Kevin Wang · Yihan Xi · Dejia Xu · Zhangyang “Atlas” Wang

Wed Jul 26 02:00 PM -- 03:30 PM (PDT) @ Exhibit Hall 1 #230
in Poster Session 3 »

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.

[ PDF]  [ Poster] 

Author Information

Wenqing Zheng (The University of Texas At Austin)
S P Sharan (University of Texas at Austin)
Ajay Jaiswal (University of Texas at Austin)
Kevin Wang (University of Texas at Austin)
Yihan Xi (University of Texas at Austin)
Dejia Xu (University of Texas at Austin)
Zhangyang “Atlas” Wang (University of Texas at Austin)

More from the Same Authors