Reproduce “MOAI: Module-Optimizing Architecture for non-interactive secure transformer Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Linru Zhang, Xiangning Wang, Jun Jie Sim, Zhicong Huang, Jiahao Zhong, Huaxiong Wang, Pu Duan, Kwok-Yan Lam
MOAI reduces the overhead of encrypted transformer inference while preserving the original BERT components. It keeps data in compatible column and diagonal layouts across stages, and computes Softmax and LayerNorm without homomorphic-encryption rotations. On one NVIDIA A100, its BERT-base evaluation took 283.95 seconds versus THOR’s 602.26 seconds, a 52.8% reduction under the reported setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.