CodeMorph: Mitigating Data Leakage in Large Language Model Assessment

TL;DR

CodeMorph is a novel approach that uses semantic-preserving transformations and a genetic algorithm to generate diverse, complex code variations across multiple languages, effectively reducing data leakage and providing more reliable LLM evaluation metrics.

Contribution

It introduces a multi-language, dependency-preserving perturbation framework with an adaptive selection algorithm to improve code variation diversity for LLM assessment.

Findings

LLM accuracy on code tasks decreased by 24.67% after applying CodeMorph.

PESO-optimized code has 7.01% lower similarity scores than random perturbations.

Significant reduction in similarity scores, up to 42.86%, demonstrating effective diversification.

Abstract

Concerns about benchmark leakage in large language models for code (Code LLMs) have raised issues of data contamination and inflated evaluation metrics. The diversity and inaccessibility of many training datasets make it difficult to prevent data leakage entirely, even with time lag strategies. Consequently, generating new datasets through code perturbation has become essential. However, existing methods often fail to produce complex and diverse variations, struggle with complex cross-file dependencies, and lack support for multiple programming languages, which limits their effectiveness in enhancing LLM evaluations for coding tasks. To fill this gap, we propose CodeMorph, an approach designed to support multiple programming languages while preserving cross-file dependencies to mitigate data leakage. CodeMorph consists of two main components that work together to enhance the perturbation process. The first component employs 26 semantic-preserving transformation methods to iteratively perturb code, generating diverse variations while ensuring that the modified code remains compilable. The second component introduces a genetic algorithm-based selection algorithm, PESO, to identify the more effective perturbation method for each iteration by targeting lower similarity scores between the perturbed and original code, thereby enhancing overall perturbation effectiveness. Experimental results demonstrate that after applying CodeMorph, the accuracy of the LLM on code completion tasks across five programming languages decreased by an average of 24.67%, with Python showing the most significant reduction at 45%. The similarity score of code optimized by PESO is, on average, 7.01% lower than that of randomly perturbed code, peaking at a reduction of 42.86%.