HeterMoE: Efficient Training of Mixture-of-Experts Models on Heterogeneous GPUs

• URL: http://arxiv.org/abs/2504.03871v1
• Date: Fri, 04 Apr 2025 18:55:52 GMT
• Title: HeterMoE: Efficient Training of Mixture-of-Experts Models on Heterogeneous GPUs
• Authors: Yongji Wu, Xueshen Liu, Shuowei Jin, Ceyu Xu, Feng Qian, Z. Morley Mao, Matthew Lentz, Danyang Zhuo, Ion Stoica,
• Abstract summary: HeterMoE is a system to efficiently train MoE models on heterogeneous GPUs.<n>We show that HeterMoE achieves up to 2.3x speed-up compared to existing MoE training systems.
• Score: 26.00138507762884
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The Mixture-of-Experts (MoE) architecture has become increasingly popular as a method to scale up large language models (LLMs). To save costs, heterogeneity-aware training solutions have been proposed to utilize GPU clusters made up of both newer and older-generation GPUs. However, existing solutions are agnostic to the performance characteristics of different MoE model components (i.e., attention and expert) and do not fully utilize each GPU's compute capability. In this paper, we introduce HeterMoE, a system to efficiently train MoE models on heterogeneous GPUs. Our key insight is that newer GPUs significantly outperform older generations on attention due to architectural advancements, while older GPUs are still relatively efficient for experts. HeterMoE disaggregates attention and expert computation, where older GPUs are only assigned with expert modules. Through the proposed zebra parallelism, HeterMoE overlaps the computation on different GPUs, in addition to employing an asymmetric expert assignment strategy for fine-grained load balancing to minimize GPU idle time. Our evaluation shows that HeterMoE achieves up to 2.3x speed-up compared to existing MoE training systems, and 1.4x compared to an optimally balanced heterogeneity-aware solution. HeterMoE efficiently utilizes older GPUs by maintaining 95% training throughput on average, even with half of the GPUs in a homogeneous A40 cluster replaced with V100.

Related papers

• MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs [55.95879347182669]
  MoE architecture is renowned for its ability to increase model capacity without a proportional increase in inference cost. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization. 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)
  arXiv  Detail & Related papers  (2024-11-18T01:06:12Z)
• Forecasting GPU Performance for Deep Learning Training and Inference [10.741682409837612]
  NeuSight is a framework to predict the performance of various deep learning models, for both training and inference, on unseen GPUs without requiring actual execution.<n>NeuSight decomposes a single deep learning kernel prediction into smaller working sets called tiles, which are executed independently on the GPU.<n>It reduces the percentage error from 121.4% and 30.8% to 2.3% in predicting the latency of GPT3 model for training and inference on H100, compared to state-of-the-art prior work.
  arXiv  Detail & Related papers  (2024-07-18T18:47:52Z)
• Optimized thread-block arrangement in a GPU implementation of a linear solver for atmospheric chemistry mechanisms [0.0]
  Earth system models (ESM) demand significant hardware resources and energy consumption to solve atmospheric chemistry processes. Recent studies have shown improved performance from running these models on GPU accelerators. This study proposes an optimized distribution of the chemical solver's computational load on the GPU, named Block-cells.
  arXiv  Detail & Related papers  (2024-05-27T17:12:59Z)
• HETHUB: A Distributed Training System with Heterogeneous Cluster for Large-Scale Models [17.08669201975141]
  Training large-scale models relies on a vast number of computing resources. It is a challenge to build a large-scale cluster with one type of GPU-accelerator. This paper proposes a distributed training system with hybrid parallelism, HETHUB, for large-scale models.
  arXiv  Detail & Related papers  (2024-05-25T14:36:35Z)
• NeRF-XL: Scaling NeRFs with Multiple GPUs [72.75214892939411]
  We present NeRF-XL, a principled method for distributing Neural Radiance Fields (NeRFs) across multiple GPU. We show improvements in reconstruction quality with larger parameter counts and speed improvements with more GPU. We demonstrate the effectiveness of NeRF-XL on a wide variety of datasets, including the largest open-source dataset to date, MatrixCity, containing 258K images covering a 25km2 city area.
  arXiv  Detail & Related papers  (2024-04-24T21:43:15Z)
• Benchmarking GPUs on SVBRDF Extractor Model [0.0]
  In this work, we try to differentiate the performance of different GPUs on neural network models that operate on bigger input images (256x256) In this work, we tried to differentiate the performance of different GPUs on neural network models that operate on bigger input images (256x256)
  arXiv  Detail & Related papers  (2023-10-19T17:09:06Z)
• FusionAI: Decentralized Training and Deploying LLMs with Massive Consumer-Level GPUs [57.12856172329322]
  We envision a decentralized system unlocking the potential vast untapped consumer-level GPU. This system faces critical challenges, including limited CPU and GPU memory, low network bandwidth, the variability of peer and device heterogeneity.
  arXiv  Detail & Related papers  (2023-09-03T13:27:56Z)
• An Analysis of Collocation on GPUs for Deep Learning Training [0.0]
  Multi-Instance GPU (MIG) is a new technology introduced by NVIDIA that can partition a GPU to better-fit workloads. In this paper, we examine the performance of a MIG-enabled A100 GPU under deep learning workloads containing various sizes and combinations of models.
  arXiv  Detail & Related papers  (2022-09-13T14:13:06Z)
• Adaptive Elastic Training for Sparse Deep Learning on Heterogeneous Multi-GPU Servers [65.60007071024629]
  We show that Adaptive SGD outperforms four state-of-the-art solutions in time-to-accuracy. We show experimentally that Adaptive SGD outperforms four state-of-the-art solutions in time-to-accuracy.
  arXiv  Detail & Related papers  (2021-10-13T20:58:15Z)
• Kernel methods through the roof: handling billions of points efficiently [94.31450736250918]
  Kernel methods provide an elegant and principled approach to nonparametric learning, but so far could hardly be used in large scale problems. Recent advances have shown the benefits of a number of algorithmic ideas, for example combining optimization, numerical linear algebra and random projections. Here, we push these efforts further to develop and test a solver that takes full advantage of GPU hardware.
  arXiv  Detail & Related papers  (2020-06-18T08:16:25Z)
• MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models [96.1052289276254]
  This work introduces a new MAP-solver, based on the popular Dual Block-Coordinate Ascent principle. Surprisingly, by making a small change to the low-performing solver, we derive the new solver MPLP++ that significantly outperforms all existing solvers by a large margin.
  arXiv  Detail & Related papers  (2020-04-16T16:20:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.