HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference

The Mixture of Experts (MoE) architecture has demonstrated significant advantages as it enables to increase the model capacity without a proportional increase in computation. However, the large MoE model size still introduces substantial memory demands, which usually requires expert offloading on re...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2025 62nd ACM/IEEE Design Automation Conference (DAC) s. 1 - 7
Hlavní autoři: Zhong, Shuzhang, Sun, Yanfan, Liang, Ling, Wang, Runsheng, Huang, Ru, Li, Meng
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 22.06.2025
Témata:
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Abstract The Mixture of Experts (MoE) architecture has demonstrated significant advantages as it enables to increase the model capacity without a proportional increase in computation. However, the large MoE model size still introduces substantial memory demands, which usually requires expert offloading on resource-constrained platforms and incurs significant overhead. Hybrid CPU-GPU inference has been proposed to leverage CPU computation to reduce expert loading overhead but faces major challenges: on one hand, the expert activation patterns of MoE models are highly unstable, rendering the fixed mapping strategies in existing works inefficient; on the other hand, the hybrid CPU-GPU schedule for MoE is inherently complex due to the diverse expert sizes, structures, uneven workload distribution, etc. To address these challenges, in this paper, we propose HybriMoE, a hybrid CPU-GPU inference framework that improves resource utilization through a novel CPU-GPU scheduling and cache management system. HybriMoE introduces (i) a dynamic intra-layer scheduling strategy to balance workloads across CPU and GPU, (ii) an impact-driven inter-layer prefetching algorithm, and (iii) a score-based caching algorithm to mitigate expert activation instability. We implement HybriMoE on top of the kTransformers framework and evaluate it on three widely used MoE-based LLMs. Experimental results demonstrate that HybriMoE achieves an average speedup of \mathbf{1. 3 3} \times in the prefill stage and 1.70 \times in the decode stage compared to state-of-the-art hybrid MoE inference framework. Our code is available at: https://github.com/PKU-SEC-Lab/HybriMoE.
AbstractList The Mixture of Experts (MoE) architecture has demonstrated significant advantages as it enables to increase the model capacity without a proportional increase in computation. However, the large MoE model size still introduces substantial memory demands, which usually requires expert offloading on resource-constrained platforms and incurs significant overhead. Hybrid CPU-GPU inference has been proposed to leverage CPU computation to reduce expert loading overhead but faces major challenges: on one hand, the expert activation patterns of MoE models are highly unstable, rendering the fixed mapping strategies in existing works inefficient; on the other hand, the hybrid CPU-GPU schedule for MoE is inherently complex due to the diverse expert sizes, structures, uneven workload distribution, etc. To address these challenges, in this paper, we propose HybriMoE, a hybrid CPU-GPU inference framework that improves resource utilization through a novel CPU-GPU scheduling and cache management system. HybriMoE introduces (i) a dynamic intra-layer scheduling strategy to balance workloads across CPU and GPU, (ii) an impact-driven inter-layer prefetching algorithm, and (iii) a score-based caching algorithm to mitigate expert activation instability. We implement HybriMoE on top of the kTransformers framework and evaluate it on three widely used MoE-based LLMs. Experimental results demonstrate that HybriMoE achieves an average speedup of \mathbf{1. 3 3} \times in the prefill stage and 1.70 \times in the decode stage compared to state-of-the-art hybrid MoE inference framework. Our code is available at: https://github.com/PKU-SEC-Lab/HybriMoE.
Author Zhong, Shuzhang
Wang, Runsheng
Liang, Ling
Li, Meng
Sun, Yanfan
Huang, Ru
Author_xml – sequence: 1
  givenname: Shuzhang
  surname: Zhong
  fullname: Zhong, Shuzhang
  organization: Institute for Artificial Intelligence, Peking University,Beijing,China
– sequence: 2
  givenname: Yanfan
  surname: Sun
  fullname: Sun, Yanfan
  organization: Beihang University,School of Computer Science and Engineering,Beijing,China
– sequence: 3
  givenname: Ling
  surname: Liang
  fullname: Liang, Ling
  organization: Peking University,School of Integrated Circuits,Beijing,China
– sequence: 4
  givenname: Runsheng
  surname: Wang
  fullname: Wang, Runsheng
  organization: Peking University,School of Integrated Circuits,Beijing,China
– sequence: 5
  givenname: Ru
  surname: Huang
  fullname: Huang, Ru
  organization: Peking University,School of Integrated Circuits,Beijing,China
– sequence: 6
  givenname: Meng
  surname: Li
  fullname: Li, Meng
  email: meng.li@pku.edu.cn
  organization: Institute for Artificial Intelligence, Peking University,Beijing,China
BookMark eNo1j81OAjEUhWuiC0XewJi-wGDb25m27sg4AglEEpk1udPeYhMoZsQFb-_4tzrnfIsvOTfsMh8zMXYvxURK4R6epnUFVruJEqockARQRl-wsTPOAshSgND2mm3m565Pq2PzyH9a4PW6LWbrlr_6Nwqf-5R3HPOAcdh8hRl3dKB84vHY8ybG5NP3Ggx8kSP1lD3dsquI-w8a_-WItc_Npp4Xy5fZop4uC5TGnYqq0ip6bRHRCQckCIWzwTpNwhtTKh2cAinQha5CMKILptS-gi5KGSnCiN39ehMRbd_7dMD-vP0_C19zL0zt
ContentType Conference Proceeding
DBID 6IE
6IH
CBEJK
RIE
RIO
DOI 10.1109/DAC63849.2025.11133274
DatabaseName IEEE Electronic Library (IEL) Conference Proceedings
IEEE Proceedings Order Plan (POP) 1998-present by volume
IEEE Xplore All Conference Proceedings
IEEE/IET Electronic Library
IEEE Proceedings Order Plans (POP) 1998-present
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Electronic Library (IEL)
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
EISBN 9798331503048
EndPage 7
ExternalDocumentID 11133274
Genre orig-research
GrantInformation_xml – fundername: Research and Development
  funderid: 10.13039/100006190
GroupedDBID 6IE
6IH
CBEJK
RIE
RIO
ID FETCH-LOGICAL-a179t-6642fc48aaa9093e0ea098d894e0c77524d92310a9db6a370bd754c63bf11fef3
IEDL.DBID RIE
IngestDate Wed Oct 01 07:05:15 EDT 2025
IsPeerReviewed false
IsScholarly true
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-a179t-6642fc48aaa9093e0ea098d894e0c77524d92310a9db6a370bd754c63bf11fef3
PageCount 7
ParticipantIDs ieee_primary_11133274
PublicationCentury 2000
PublicationDate 2025-June-22
PublicationDateYYYYMMDD 2025-06-22
PublicationDate_xml – month: 06
  year: 2025
  text: 2025-June-22
  day: 22
PublicationDecade 2020
PublicationTitle 2025 62nd ACM/IEEE Design Automation Conference (DAC)
PublicationTitleAbbrev DAC
PublicationYear 2025
Publisher IEEE
Publisher_xml – name: IEEE
Score 2.2953632
Snippet The Mixture of Experts (MoE) architecture has demonstrated significant advantages as it enables to increase the model capacity without a proportional increase...
SourceID ieee
SourceType Publisher
StartPage 1
SubjectTerms Computational modeling
Dynamic scheduling
Hands
Heuristic algorithms
Inference algorithms
Load modeling
Prefetching
Rendering (computer graphics)
Resource management
Schedules
Title HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference
URI https://ieeexplore.ieee.org/document/11133274
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1LSwMxEA62ePCkYsU3OXhNm93NbhJvUlvrwbJgC72VyUt62Uofgv_eJN21ePDgbRLygEnCTJLvm0HoPgk2KgAcMsNywrRJiPRuB-EmcYZ7m0DBxWQTfDwWs5ksa7J65MJYayP4zHaDGP_yzVJvw1NZL6RFz_w1qoVanPMdWatm_SZU9p4e-343sUA_SfNu0_hX2pRoNYbH_5zvBHX2_Dtc_liWU3RgqzM0GX2p1eJ1OXjAUTK4X07JcznFb17xJiDK3zFUvjrEaMZ7XAv2fikexFARoeRHwC_NNB00HQ4m_RGpcyIQ8EdnQwp_X3CaCQCQVGaWWqBSGCGZpZrzPGUmumwgjSog41QZnjNdZMolibMuO0ftalnZC4RTUAogT31vx5RWogDBnDRcKw4i15eoE1Qy_9iFvZg32rj6o_4aHQXFBxxVmt6g9ma1tbfoUH9uFuvVXVysb1iTlY4
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1LTwIxEG4UTfSkRoxve_C6sI9223ozCEIEsomQcCPTl-GyGFxM_Pe2BSQePHhrmz6SmTYzbb9vBqH7xNsoD3DINKERUTqJhHM7IqYTq5mzCTHYkGyCDYd8MhHFmqweuDDGmAA-Mw1fDH_5eq6W_qms6dOiZ-4atYv2KCFpsqJrrXm_SSyaT48tt5-IJ6CktLHp_itxSrAbnaN_rniM6lsGHi5-bMsJ2jHlKRp1v-RiNpi3H3AoadwqxtFzMcavTvTaY8rfMJSu2UdpxltkC3aeKW6HYBG-5mbAvc0ydTTutEetbrTOihCBOzxVlLsbg1WEA4CIRWZiA7HgmgtiYsUYTYkOThsILXPIWCw1o0TlmbRJYo3NzlCtnJfmHOEUpASgqRttiVSS58CJFZopyYBTdYHqXiTT91Xgi-lGGpd_tN-hg-5o0J_2e8OXK3ToleBRVWl6jWrVYmlu0L76rGYfi9uguG9QMpjV
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=proceeding&rft.title=2025+62nd+ACM%2FIEEE+Design+Automation+Conference+%28DAC%29&rft.atitle=HybriMoE%3A+Hybrid+CPU-GPU+Scheduling+and+Cache+Management+for+Efficient+MoE+Inference&rft.au=Zhong%2C+Shuzhang&rft.au=Sun%2C+Yanfan&rft.au=Liang%2C+Ling&rft.au=Wang%2C+Runsheng&rft.date=2025-06-22&rft.pub=IEEE&rft.spage=1&rft.epage=7&rft_id=info:doi/10.1109%2FDAC63849.2025.11133274&rft.externalDocID=11133274