analiza aiqbench
This research addresses the computational bottleneck of homomorphic encryption in secure LLM inference by introducing MOAI, a framework that drastically reduces expensive HE rotations through optimized matrix packing and rotation-free algorithms for Softmax and LayerNorm. By achieving a 52.8% reduction in evaluation time compared to state-of-the-art methods, the work makes privacy-preserving cloud inference significantly more practical for transformer models like BERT and LLaMA-3. The approach represents a tangible efficiency breakthrough in applying cryptographic security directly to AI inference pipelines without interactive protocols.
MOAI: Module-Optimizing Architecture for non-interactive secure transformer Inference
De la arXiv
Text integralarXiv:2611.13641v1 Announce Type: cross Abstract: Privacy concerns have been raised in Large Language Models (LLM) inference when models are deployed in Cloud Service Providers (CSP). Homomorphic encryption (HE) offers a promising solution by enabling secure inference directly over encrypted inputs. However, the high computational overhead of HE remains a major bottleneck. To address this challenge, we propose MOAI, an efficient HE-based, non-interactive framework for secure transformer inference. MOAI gains significant efficiency improvement from: (1) a novel evaluation flow that combines column and diagonal packing with consistent strategies across all layers, eliminating expensive format conversions.
(2) rotation-free algorithms for Softmax and LayerNorm that significantly reduce the number of costly HE rotations, removing 2,448 HE rotations in BERT-base inference. (3) Column packing removes rotations in plaintext-ciphertext matrix multiplications and interleaved batching further reduces the rotations in ciphertext-ciphertext matrix multiplications. MOAI uses at least 1.7x fewer HE rotations compared to the state-of-the-art works across all matrix multiplications of BERT-base. As a result, we achieve a $52.8 \%$ reduction in evaluation time compared to the state-of-the-art HE-based non-interactive secure transformer inference, THOR (Moon et al., CCS 25).
We then apply MOAI on the Powerformer's framework and achieve a $55.7 \%$ reduction in evaluation time compared to Powerformer (Park et al., ACL 25), which approximates Softmax and LayerNorm with simpler functions in transformer and proposes HE-based non-interactive transformer inference. We report an amortized time of 2.36 minutes per input on a single GPU environment. We show the extendibility by applying MOAI in LLaMA-3-8B. Our implementation is publicly available as open source.
Abstract:Privacy concerns have been raised in Large Language Models (LLM) inference when models are deployed in Cloud Service Providers (CSP). Homomorphic encryption (HE) offers a promising solution by enabling secure inference directly over encrypted inputs. However, the high computational overhead of HE remains a major bottleneck. To address this challenge, we propose MOAI, an efficient HE-based, non-interactive framework for secure transformer inference. MOAI gains significant efficiency improvement from: (1) a novel evaluation flow that combines column and diagonal packing with consistent strategies across all layers, eliminating expensive format conversions.
(2) rotation-free algorithms for Softmax and LayerNorm that significantly reduce the number of costly HE rotations, removing 2,448 HE rotations in BERT-base inference. (3) Column packing removes rotations in plaintext-ciphertext matrix multiplications and interleaved batching further reduces the rotations in ciphertext-ciphertext matrix multiplications. MOAI uses at least 1.7x fewer HE rotations compared to the state-of-the-art works across all matrix multiplications of BERT-base. As a result, we achieve a $52.8 \%$ reduction in evaluation time compared to the state-of-the-art HE-based non-interactive secure transformer inference, THOR (Moon et al., CCS 25).
We then apply MOAI on the Powerformer's framework and achieve a $55.7 \%$ reduction in evaluation time compared to Powerformer (Park et al., ACL 25), which approximates Softmax and LayerNorm with simpler functions in transformer and proposes HE-based non-interactive transformer inference. We report an amortized time of 2.36 minutes per input on a single GPU environment. We show the extendibility by applying MOAI in LLaMA-3-8B. Our implementation is publicly available as open source.
Published elsewhere. Minor revision. ICLR 2026
Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
arXiv:2611.13641 [cs.CR]
(or arXiv:2611.13641v1 [cs.CR] for this version)
https://doi.org/10.48550/arXiv.2611.13641 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 355 cuvinte