CodeGrad: Integrating Multi-Step Verification with Gradient-Based LLM Refinement

TL;DR

CodeGrad presents a novel framework that integrates verification with gradient-based refinement in LLMs to produce more correct, robust, and mathematically justified code solutions, significantly improving performance on benchmark datasets.

Contribution

It introduces a new method that treats code as a differentiable variable, enabling iterative refinement guided by structured feedback and constraints.

Findings

Achieves up to 27% improvement on HumanEval

Attains 41% relative improvement on LiveCodeBench V6

Generates mathematically justified and robust code

Abstract

While Large Language Models (LLMs) have demonstrated remarkable capabilities in code generation, they often produce solutions that lack guarantees of correctness, robustness, and efficiency. This limitation is particularly acute in domains requiring strict constraints. CodeGrad introduces a principled framework that integrates rigorous verification techniques directly into an iterative LLM-based generation loop. It uniquely treats code as a differentiable variable, converting structured feedback and mathematical constraints into a textual pseudo-gradient. This gradient guides the model to iteratively refine solutions, ensuring they are not only functional but also robust and mathematically justified. We evaluate CodeGrad on the HumanEval, HumanEval+, and LiveCodeBench benchmarks. Our implementation outperforms strong baselines, achieving an absolute improvement of up to 27% on HumanEval and a 41% relative improvement on the challenging LiveCodeBench V6. StructuredGrad generates mathematically justified code that is robust and efficient, paving the way for reliable AI-assisted software development in high-stakes applications.