paper-with-me

홈 › Papers

HETHUB: A Distributed Training System with Heterogeneous Cluster for Large-Scale Models

2024-05-25 · Si Xu, Zixiao Huang, Yan Zeng, Shengen Yan, Xuefei Ning, Quanlu Zhang, Haolin Ye, Sipei Gu, Chunsheng Shui, Zhezheng Lin, Hao Zhang, Sheng Wang, Guohao Dai, Yu Wang

Training large-scale models relies on a vast number of computing resources. For example, training the GPT-4 model (1.8 trillion parameters) requires 25000 A100 GPUs . It is a challenge to build a large-scale cluster with one type of GPU-accelerator. Using multiple types of GPU-accelerators to construct a large-scale cluster is an effective way to solve the problem of insufficient homogeneous GPU-accelerators. However, the existing distributed training systems for large-scale models only support homogeneous GPU-accelerators, not support heterogeneous GPU-accelerators. To address the problem, this paper proposes a distributed training system with hybrid parallelism, HETHUB, for large-scale models, which supports heterogeneous cluster, including AMD, Nvidia GPU and other types of GPU-accelerators . It introduces a distributed unified communicator to realize the communication between heterogeneous GPU-accelerators, a distributed performance predictor, and an automatic parallel planner to develop and train models efficiently with heterogeneous GPU-accelerators. Compared to the distributed training system with homogeneous GPU-accelerators, our system can support six combinations of heterogeneous GPU-accelerators. We train the Llama-140B model on a heterogeneous cluster with 768 GPU-accelerators(128 AMD and 640 GPU-accelerator A). The experiment results show that the optimal performance of our system in the heterogeneous cluster has achieved up to 97.49% of the theoretical upper bound performance.

📄 PDF Abstract BibTeX arXiv:2405.16256

Code (0)

등록된 구현이 없습니다.

Tasks

GPU

Methods 이 논문이 사용한 방법론

Linear Layer A Linear Layer is a projection $\mathbf{XW + b}$.
BPE Byte Pair Encoding, or BPE, is a subword segmentation algorithm that encodes rare and unknown words as sequences of subword units. The intuition is that various word…
Label Smoothing Label Smoothing is a regularization technique that introduces noise for the labels. This accounts for the fact that datasets may have mistakes in them, so maximizing the…
Adam 설명 없음
Attention 설명 없음
Residual Connection 설명 없음
Position-Wise Feed-Forward Layer 설명 없음
Multi-Head Attention 설명 없음

Similar Papers 제목 키워드 기반

OmniLearn: A Framework for Distributed Deep Learning over Heterogeneous Clusters

2025-03-21 · Sahil Tyagi, Prateek Sharma

Deep learning systems are optimized for clusters with homogeneous resources. However, heterogeneity is prevalent in computing infrastructure across edge, cloud and HPC. When training neural networks using stochastic grad…

Deep Learning

Contrastive encoder pre-training-based clustered federated learning for heterogeneous data

2023-11-28 · Ye Lin Tun, Minh N. H. Nguyen, Chu Myaet Thwal, Jinwoo Choi 외

Federated learning (FL) is a promising approach that enables distributed clients to collaboratively train a global model while preserving their data privacy. However, FL often suffers from data heterogeneity problems, wh…

ClusteringContrastive LearningFederated Learning

Taming Resource Heterogeneity In Distributed ML Training With Dynamic Batching

2023-05-20 · Sahil Tyagi, Prateek Sharma

Current techniques and systems for distributed model training mostly assume that clusters are comprised of homogeneous servers with a constant resource availability. However, cluster heterogeneity is pervasive in computi…

CPUGPU

Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms

2025-08-07 · Jie Xiao, Changyuan Fan, Qingnan Ren, Alfred Long 외 arxiv

Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This se…

Expediting Distributed DNN Training with Device Topology-Aware Graph Deployment

2023-02-13 · Shiwei Zhang, Xiaodong Yi, Lansong Diao, Chuan Wu 외

This paper presents TAG, an automatic system to derive optimized DNN training graph and its deployment onto any device topology, for expedited training in device- and topology- heterogeneous ML clusters. We novelly combi…

Combinatorial OptimizationGraph Neural NetworkTAG