Ensuring Functional Correctness of Large Code Models with Selective Generation

TL;DR

This paper introduces a method to improve the correctness of large code generation models by automatically generating unit tests for evaluating and controlling code quality, reducing hallucinations and increasing safety.

Contribution

It proposes a selective code generation approach that abstains from uncertain outputs based on functional correctness, and introduces FuzzEval, a paradigm using generated unit tests for evaluation and learning.

Findings

Effective control of code hallucination demonstrated

Generated unit tests improve evaluation precision

Selective generation increases safety and efficiency

Abstract

The hallucination of code generation models hinders their applicability to systems requiring higher safety standards. One critical bottleneck in addressing code hallucination is the difficulty of identifying the functional correctness of generated code, due to its unnatural form. We address this core bottleneck by automatically generating unit tests using dynamic code analysis tools, leveraging the \emph{executable nature} of code. Accordingly, we propose \emph{selective code generator} that abstains from uncertain generations -- based on the functional correctness evaluated by generated unit tests -- to theoretically control the correctness among non-abstained answers, \ie the false discovery rate. Finally, we propose to use generated unit tests in evaluation as well as in learning for precise code evaluation, calling this paradigm \emph{FuzzEval}. We demonstrate the efficacy of our method along with the controllability of code hallucination and reasonable selection efficiency.