一句话：提出 Helix，将异构 GPU 集群 LLM 服务建模为最大流问题，用 MILP 联合优化模型放置与调度，显著提升吞吐、降低延迟。
本文针对异构 GPU 集群中大规模语言模型（LLM）服务的高吞吐、低延迟需求。现有方法难以在 GPU 与网络异构条件下联合优化模型放置和请求调度，二者高度耦合。Helix 将 LLM 在异构 GPU 与网络连接上的推理计算建模为有向加权图上的最大流问题：节点表示 GPU 实例，边容量刻画 GPU 与网络异构性。随后使用混合整数线性规划（MILP）算法搜索高度优化的服务策略，从而联合优化模型放置与请求调度。在 24 至 42 个 GPU 节点的多个异构集群上评估，Helix 相比现有方法将服务吞吐量最高提升 3.3 倍，并将提示延迟和解码延迟分别最高降低 66% 和 24%。系统已开源。
This paper introduces Helix, a distributed system for high-throughput, low-latency large language model (LLM) serving in heterogeneous GPU clusters. The key idea behind Helix is to formulate inference computation of LLMs over heterogeneous GPUs and network connections as a max-flow problem on directed, weighted graphs, whose nodes represent GPU instances and edges capture both GPU and network heterogeneity through their capacities. Helix then uses a mixed integer linear programming (MILP) algorithm to discover highly optimized strategies to serve LLMs on heterogeneous GPUs. This approach allows Helix to jointly optimize model placement and request scheduling, two highly entangled tasks in heterogeneous LLM serving. Our evaluation on several heterogeneous clusters ranging from 24 to 42 GPU nodes shows that Helix improves serving throughput by up to 3.3x and reduces prompting and decoding latency by up to 66% and 24%, respectively, compared to existing approaches. Helix is available at https://github.com/Thesys-lab/Helix-ASPLOS25.
DOI 原文 · arXiv · PDF（开放获取） · DBLP
@inproceedings{Mei0MYJV25,
title = {Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flow},
author = {Yixuan Mei and Yonghao Zhuang and Xupeng Miao and Juncheng Yang and Zhihao Jia and Rashmi Vinayak},
booktitle = {ASPLOS},
year = {2025},
doi = {10.1145/3669940.3707215}
}