MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cao, Shiyi, Liu, Shu, Griggs, Tyler, Schafhalter, Peter, Liu, Xiaoxuan, Sheng, Ying, Gonzalez, Joseph E., Zaharia, Matei, Stoica, Ion
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912126029791232
author Cao, Shiyi
Liu, Shu
Griggs, Tyler
Schafhalter, Peter
Liu, Xiaoxuan
Sheng, Ying
Gonzalez, Joseph E.
Zaharia, Matei
Stoica, Ion
author_facet Cao, Shiyi
Liu, Shu
Griggs, Tyler
Schafhalter, Peter
Liu, Xiaoxuan
Sheng, Ying
Gonzalez, Joseph E.
Zaharia, Matei
Stoica, Ion
contents Efficient deployment of large language models, particularly Mixture of Experts (MoE), on resource-constrained platforms presents significant challenges, especially in terms of computational efficiency and memory utilization. The MoE architecture, renowned for its ability to increase model capacity without a proportional increase in inference cost, greatly reduces the token generation latency compared with dense models. However, the large model size makes MoE models inaccessible to individuals without high-end GPUs. In this paper, we propose a high-throughput MoE batch inference system, that significantly outperforms past work. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization, and a performance model, HRM, based on a Hierarchical Roofline Model we introduce to help find policies with higher throughput than existing systems. MoE-Lightning can achieve up to 10.3x higher throughput than state-of-the-art offloading-enabled LLM inference systems for Mixtral 8x7B on a single T4 GPU (16GB). When the theoretical system throughput is bounded by the GPU memory, MoE-Lightning can reach the throughput upper bound with 2-3x less CPU memory, significantly increasing resource utilization. MoE-Lightning also supports efficient batch inference for much larger MoEs (e.g., Mixtral 8x22B and DBRX) on multiple low-cost GPUs (e.g., 2-4 T4).
format Preprint
id arxiv_https___arxiv_org_abs_2411_11217
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs
Cao, Shiyi
Liu, Shu
Griggs, Tyler
Schafhalter, Peter
Liu, Xiaoxuan
Sheng, Ying
Gonzalez, Joseph E.
Zaharia, Matei
Stoica, Ion
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
Efficient deployment of large language models, particularly Mixture of Experts (MoE), on resource-constrained platforms presents significant challenges, especially in terms of computational efficiency and memory utilization. The MoE architecture, renowned for its ability to increase model capacity without a proportional increase in inference cost, greatly reduces the token generation latency compared with dense models. However, the large model size makes MoE models inaccessible to individuals without high-end GPUs. In this paper, we propose a high-throughput MoE batch inference system, that significantly outperforms past work. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization, and a performance model, HRM, based on a Hierarchical Roofline Model we introduce to help find policies with higher throughput than existing systems. MoE-Lightning can achieve up to 10.3x higher throughput than state-of-the-art offloading-enabled LLM inference systems for Mixtral 8x7B on a single T4 GPU (16GB). When the theoretical system throughput is bounded by the GPU memory, MoE-Lightning can reach the throughput upper bound with 2-3x less CPU memory, significantly increasing resource utilization. MoE-Lightning also supports efficient batch inference for much larger MoEs (e.g., Mixtral 8x22B and DBRX) on multiple low-cost GPUs (e.g., 2-4 T4).
title MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2411.11217