GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks

Zihao Li; Hongyi Lu; Yanan Guo; Zhenkai Zhang; Shuai Wang; Fengwei Zhang

GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks

Cryptography and Security 2026-03-03 v3 Machine Learning

Authors: Zihao Li , Hongyi Lu , Yanan Guo , Zhenkai Zhang , Shuai Wang , Fengwei Zhang

Abstract

GPU memory errors are a critical threat to deep learning (DL) frameworks, leading to crashes or even security issues. We introduce GPU-Fuzz, a fuzzer locating these issues efficiently by modeling operator parameters as formal constraints. GPU-Fuzz utilizes a constraint solver to generate test cases that systematically probe error-prone boundary conditions in GPU kernels. Applied to PyTorch, TensorFlow, and PaddlePaddle, we uncovered 13 unknown bugs, demonstrating the effectiveness of GPU-Fuzz in finding memory errors.

Keywords

fuzzing gpu computing

Cite

@article{arxiv.2602.10478,
  title  = {GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks},
  author = {Zihao Li and Hongyi Lu and Yanan Guo and Zhenkai Zhang and Shuai Wang and Fengwei Zhang},
  journal= {arXiv preprint arXiv:2602.10478},
  year   = {2026}
}

Related papers

View all related →