ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Yang, Yuchen, Zhao, Yaru, Yang, Pu, Wang, Shaowei, Zhou, Zhi-Hua
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913153622736896
author Yang, Yuchen
Zhao, Yaru
Yang, Pu
Wang, Shaowei
Zhou, Zhi-Hua
author_facet Yang, Yuchen
Zhao, Yaru
Yang, Pu
Wang, Shaowei
Zhou, Zhi-Hua
contents While Mixture-of-Experts (MoE) architectures substantially bolster the expressive power of large-language models, their prohibitive memory footprint severely impedes the practical deployment on resource-constrained edge devices, especially when model behavior must be preserved without relying on lossy quantization. In this paper, we present ZipMoE, an efficient and semantically lossless on-device MoE serving system. ZipMoE exploits the synergy between the hardware properties of edge devices and the statistical redundancy inherent to MoE parameters via a caching-scheduling co-design with provable performance guarantee. Fundamentally, our design shifts the paradigm of on-device MoE inference from an I/O-bound bottleneck to a compute-centric workflow that enables efficient parallelization. We implement a prototype of ZipMoE and conduct extensive experiments on representative edge computing platforms using popular open-source MoE models and real-world workloads. Our evaluation reveals that ZipMoE achieves up to $72.77\%$ inference latency reduction and up to $6.76\times$ higher throughput than the state-of-the-art systems.Our code is available at: https://github.com/npnothard/ZipMoE-ICML26.
format Preprint
id arxiv_https___arxiv_org_abs_2601_21198
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling
Yang, Yuchen
Zhao, Yaru
Yang, Pu
Wang, Shaowei
Zhou, Zhi-Hua
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
While Mixture-of-Experts (MoE) architectures substantially bolster the expressive power of large-language models, their prohibitive memory footprint severely impedes the practical deployment on resource-constrained edge devices, especially when model behavior must be preserved without relying on lossy quantization. In this paper, we present ZipMoE, an efficient and semantically lossless on-device MoE serving system. ZipMoE exploits the synergy between the hardware properties of edge devices and the statistical redundancy inherent to MoE parameters via a caching-scheduling co-design with provable performance guarantee. Fundamentally, our design shifts the paradigm of on-device MoE inference from an I/O-bound bottleneck to a compute-centric workflow that enables efficient parallelization. We implement a prototype of ZipMoE and conduct extensive experiments on representative edge computing platforms using popular open-source MoE models and real-world workloads. Our evaluation reveals that ZipMoE achieves up to $72.77\%$ inference latency reduction and up to $6.76\times$ higher throughput than the state-of-the-art systems.Our code is available at: https://github.com/npnothard/ZipMoE-ICML26.
title ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.21198