Rethinking Code Refinement: Learning to Judge Code Efficiency

TL;DR

This paper introduces a new approach using code language models to efficiently evaluate and compare the efficiency of different code versions, reducing the need for time-consuming execution comparisons.

Contribution

It proposes a novel method for judging code efficiency with language models, eliminating the need for running multiple code versions for comparison.

Findings

Effective distinction between more and less efficient code versions

Validated across multiple programming languages and refinement steps

Reduces time-consuming code execution comparisons

Abstract

Large Language Models (LLMs) have demonstrated impressive capabilities in understanding and generating codes. Due to these capabilities, many recent methods are proposed to automatically refine the codes with LLMs. However, we should rethink that the refined codes (from LLMs and even humans) are not always more efficient than their original versions. On the other hand, running two different versions of codes and comparing them every time is not ideal and time-consuming. Therefore, in this work, we propose a novel method based on the code language model that is trained to judge the efficiency between two different codes (generated across humans and machines) by either classifying the superior one or predicting the relative improvement. We validate our method on multiple programming languages with multiple refinement steps, demonstrating that the proposed method can effectively distinguish between more and less efficient versions of code.