Saved in:
Bibliographic Details
Main Authors: Li, Zihao, Lu, Hongyi, Guo, Yanan, Zhang, Zhenkai, Wang, Shuai, Zhang, Fengwei
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.10478
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910035622232064
author Li, Zihao
Lu, Hongyi
Guo, Yanan
Zhang, Zhenkai
Wang, Shuai
Zhang, Fengwei
author_facet Li, Zihao
Lu, Hongyi
Guo, Yanan
Zhang, Zhenkai
Wang, Shuai
Zhang, Fengwei
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2602_10478
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks
Li, Zihao
Lu, Hongyi
Guo, Yanan
Zhang, Zhenkai
Wang, Shuai
Zhang, Fengwei
Cryptography and Security
Machine Learning
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.
title GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks
topic Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2602.10478