SMoE: An Algorithm-System Co-Design for Pushing MoE to the Edge via Expert Substitution

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhu, Guoying, Li, Meng, Dai, Haipeng, Liu, Xuechen, Wang, Weijun, Li, Keran, xiao, Jun, Chen, Ligeng, Wang, Wei
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909010825838592
author Zhu, Guoying
Li, Meng
Dai, Haipeng
Liu, Xuechen
Wang, Weijun
Li, Keran
xiao, Jun
Chen, Ligeng
Wang, Wei
author_facet Zhu, Guoying
Li, Meng
Dai, Haipeng
Liu, Xuechen
Wang, Weijun
Li, Keran
xiao, Jun
Chen, Ligeng
Wang, Wei
contents The Mixture of Experts (MoE) architecture has emerged as a key technique for scaling Large Language Models by activating only a subset of experts per query. Deploying MoE on consumer-grade edge hardware, however, is constrained by limited device memory, making dynamic expert offloading essential. Unlike prior work that treats offloading purely as a scheduling problem, we leverage expert importance to guide decisions, substituting low-importance activated experts with functionally similar ones already cached in GPU memory, thereby preserving accuracy. As a result, this design reduces memory usage and data transfer, while largely eliminating PCIe overhead. In addition, we introduce a scheduling policy that maximizes the reuse ratio of GPU-cached experts, further boosting efficiency. Extensive evaluations show that our approach delivers 48% lower decoding latency with over 60% expert cache hit rate, while maintaining nearly lossless accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2508_18983
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SMoE: An Algorithm-System Co-Design for Pushing MoE to the Edge via Expert Substitution
Zhu, Guoying
Li, Meng
Dai, Haipeng
Liu, Xuechen
Wang, Weijun
Li, Keran
xiao, Jun
Chen, Ligeng
Wang, Wei
Artificial Intelligence
The Mixture of Experts (MoE) architecture has emerged as a key technique for scaling Large Language Models by activating only a subset of experts per query. Deploying MoE on consumer-grade edge hardware, however, is constrained by limited device memory, making dynamic expert offloading essential. Unlike prior work that treats offloading purely as a scheduling problem, we leverage expert importance to guide decisions, substituting low-importance activated experts with functionally similar ones already cached in GPU memory, thereby preserving accuracy. As a result, this design reduces memory usage and data transfer, while largely eliminating PCIe overhead. In addition, we introduce a scheduling policy that maximizes the reuse ratio of GPU-cached experts, further boosting efficiency. Extensive evaluations show that our approach delivers 48% lower decoding latency with over 60% expert cache hit rate, while maintaining nearly lossless accuracy.
title SMoE: An Algorithm-System Co-Design for Pushing MoE to the Edge via Expert Substitution
topic Artificial Intelligence
url https://arxiv.org/abs/2508.18983