Efficient CPU-GPU Collaborative Inference for MoE-based LLMs on Memory-Limited Systems

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Huang, En-Ming, Lin, Li-Shang, Lee, Chun-Yi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914207300059136
author Huang, En-Ming
Lin, Li-Shang
Lee, Chun-Yi
author_facet Huang, En-Ming
Lin, Li-Shang
Lee, Chun-Yi
contents Large Language Models (LLMs) have achieved impressive results across various tasks, yet their high computational demands pose deployment challenges, especially on consumer-grade hardware. Mixture of Experts (MoE) models provide an efficient solution through selective activation of parameter subsets, which reduces computation requirements. Despite this efficiency, state-of-the-art MoE models still require substantial memory beyond typical consumer GPU capacities. Traditional offloading methods that transfer model weights between CPU and GPU introduce latency, limiting inference performance. This paper presents a novel CPU-GPU collaborative inference framework that incorporates an expert caching mechanism on the GPU to reduce data transfer requirements and enable faster inference through cache hits. Computations are offloaded to CPU for efficient cache miss handling, which benefits from CPU multithreading optimizations. The evaluations of our framework demonstrate performance improvements and highlight the potential of CPU-GPU collaboration to maximize hardware utilization for single-request inference scenarios on consumer-grade systems. The implementation of our framework is available at https://github.com/elsa-lab/MoE-CPU-GPU-Collaborative-Inference.
format Preprint
id arxiv_https___arxiv_org_abs_2512_16473
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Efficient CPU-GPU Collaborative Inference for MoE-based LLMs on Memory-Limited Systems
Huang, En-Ming
Lin, Li-Shang
Lee, Chun-Yi
Distributed, Parallel, and Cluster Computing
Large Language Models (LLMs) have achieved impressive results across various tasks, yet their high computational demands pose deployment challenges, especially on consumer-grade hardware. Mixture of Experts (MoE) models provide an efficient solution through selective activation of parameter subsets, which reduces computation requirements. Despite this efficiency, state-of-the-art MoE models still require substantial memory beyond typical consumer GPU capacities. Traditional offloading methods that transfer model weights between CPU and GPU introduce latency, limiting inference performance. This paper presents a novel CPU-GPU collaborative inference framework that incorporates an expert caching mechanism on the GPU to reduce data transfer requirements and enable faster inference through cache hits. Computations are offloaded to CPU for efficient cache miss handling, which benefits from CPU multithreading optimizations. The evaluations of our framework demonstrate performance improvements and highlight the potential of CPU-GPU collaboration to maximize hardware utilization for single-request inference scenarios on consumer-grade systems. The implementation of our framework is available at https://github.com/elsa-lab/MoE-CPU-GPU-Collaborative-Inference.
title Efficient CPU-GPU Collaborative Inference for MoE-based LLMs on Memory-Limited Systems
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2512.16473