GPU-Fuzz: Finding Memory Errors in Deep Learning Frameworks
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.
Fields of Study & Subjects
Software Testing and Debugging Techniques
Parallel Computing and Optimization Techniques
Formal Methods in Verification
Publication details
- Journal / source
- arXiv (Cornell University)
- Publication date
- 11 February 2026
- Language
- EN
- Document type
- Preprint