Faith: An Efficient Framework for Transformer Verification on GPUs

TL;DR

Faith is a GPU-optimized framework that significantly accelerates transformer verification by leveraging semantic-aware graph transformations, specialized kernels, and autotuning, enabling faster robustness checks against adversarial attacks.

Contribution

The paper introduces Faith, a novel framework that improves transformer verification efficiency on GPUs through semantic-aware graph transformation, specialized kernel design, and autotuning techniques.

Findings

Achieves 2.1x to 3.4x speedup over existing frameworks

Utilizes semantic-aware computation graph transformation for efficiency

Employs verification-specialized GPU kernels and autotuning

Abstract

Transformer verification draws increasing attention in machine learning research and industry. It formally verifies the robustness of transformers against adversarial attacks such as exchanging words in a sentence with synonyms. However, the performance of transformer verification is still not satisfactory due to bound-centric computation which is significantly different from standard neural networks. In this paper, we propose Faith, an efficient framework for transformer verification on GPUs. We first propose a semantic-aware computation graph transformation to identify semantic information such as bound computation in transformer verification. We exploit such semantic information to enable efficient kernel fusion at the computation graph level. Second, we propose a verification-specialized kernel crafter to efficiently map transformer verification to modern GPUs. This crafter exploits a set of GPU hardware supports to accelerate verification specialized operations which are usually memory-intensive. Third, we propose an expert-guided autotuning to incorporate expert knowledge on GPU backends to facilitate large search space exploration. Extensive evaluations show that Faith achieves $2.1\times$ to $3.4\times$ ($2.6\times$ on average) speedup over state-of-the-art frameworks.