Saved in:
Bibliographic Details
Main Authors: Li, Jialong, Tripathi, Shreyansh, Rastogi, Lakshay, Lei, Yiming, Pan, Rui, Xia, Yiting
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.17043
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929554329698304
author Li, Jialong
Tripathi, Shreyansh
Rastogi, Lakshay
Lei, Yiming
Pan, Rui
Xia, Yiting
author_facet Li, Jialong
Tripathi, Shreyansh
Rastogi, Lakshay
Lei, Yiming
Pan, Rui
Xia, Yiting
contents As machine learning models scale in size and complexity, their computational requirements become a significant barrier. Mixture-of-Experts (MoE) models alleviate this issue by selectively activating relevant experts. Despite this, MoE models are hindered by high communication overhead from all-to-all operations, low GPU utilization due to the synchronous communication constraint, and complications from heterogeneous GPU environments. This paper presents Aurora, which optimizes both model deployment and all-to-all communication scheduling to address these challenges in MoE inference. Aurora achieves minimal communication times by strategically ordering token transmissions in all-to-all communications. It improves GPU utilization by colocating experts from different models on the same device, avoiding the limitations of synchronous all-to-all communication. We analyze Aurora's optimization strategies theoretically across four common GPU cluster settings: exclusive vs. colocated models on GPUs, and homogeneous vs. heterogeneous GPUs. Aurora provides optimal solutions for three cases, and for the remaining NP-hard scenario, it offers a polynomial-time sub-optimal solution with only a 1.07x degradation from the optimal. Aurora is the first approach to minimize MoE inference time via optimal model deployment and communication scheduling across various scenarios. Evaluations demonstrate that Aurora significantly accelerates inference, achieving speedups of up to 2.38x in homogeneous clusters and 3.54x in heterogeneous environments. Moreover, Aurora enhances GPU utilization by up to 1.5x compared to existing methods.
format Preprint
id arxiv_https___arxiv_org_abs_2410_17043
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Optimizing Mixture-of-Experts Inference Time Combining Model Deployment and Communication Scheduling
Li, Jialong
Tripathi, Shreyansh
Rastogi, Lakshay
Lei, Yiming
Pan, Rui
Xia, Yiting
Machine Learning
Networking and Internet Architecture
As machine learning models scale in size and complexity, their computational requirements become a significant barrier. Mixture-of-Experts (MoE) models alleviate this issue by selectively activating relevant experts. Despite this, MoE models are hindered by high communication overhead from all-to-all operations, low GPU utilization due to the synchronous communication constraint, and complications from heterogeneous GPU environments. This paper presents Aurora, which optimizes both model deployment and all-to-all communication scheduling to address these challenges in MoE inference. Aurora achieves minimal communication times by strategically ordering token transmissions in all-to-all communications. It improves GPU utilization by colocating experts from different models on the same device, avoiding the limitations of synchronous all-to-all communication. We analyze Aurora's optimization strategies theoretically across four common GPU cluster settings: exclusive vs. colocated models on GPUs, and homogeneous vs. heterogeneous GPUs. Aurora provides optimal solutions for three cases, and for the remaining NP-hard scenario, it offers a polynomial-time sub-optimal solution with only a 1.07x degradation from the optimal. Aurora is the first approach to minimize MoE inference time via optimal model deployment and communication scheduling across various scenarios. Evaluations demonstrate that Aurora significantly accelerates inference, achieving speedups of up to 2.38x in homogeneous clusters and 3.54x in heterogeneous environments. Moreover, Aurora enhances GPU utilization by up to 1.5x compared to existing methods.
title Optimizing Mixture-of-Experts Inference Time Combining Model Deployment and Communication Scheduling
topic Machine Learning
Networking and Internet Architecture
url https://arxiv.org/abs/2410.17043