Synergistic Tensor and Pipeline Parallelism

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Qi, Mengshi, Peng, Jiaxuan, Zhang, Jie, Zhu, Juan, Li, Yong, Ma, Huadong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908622567505920
author Qi, Mengshi
Peng, Jiaxuan
Zhang, Jie
Zhu, Juan
Li, Yong
Ma, Huadong
author_facet Qi, Mengshi
Peng, Jiaxuan
Zhang, Jie
Zhu, Juan
Li, Yong
Ma, Huadong
contents In the machine learning system, the hybrid model parallelism combining tensor parallelism (TP) and pipeline parallelism (PP) has become the dominant solution for distributed training of Large Language Models~(LLMs) and Multimodal LLMs (MLLMs). However, TP introduces significant collective communication overheads, while PP suffers from synchronization inefficiencies such as pipeline bubbles. Existing works primarily address these challenges from isolated perspectives, focusing either on overlapping TP communication or on flexible PP scheduling to mitigate pipeline bubbles. In this paper, we propose a new synergistic tensor and pipeline parallelism schedule that simultaneously reduces both types of bubbles. Our proposed schedule decouples the forward and backward passes in PP into fine-grained computation units, which are then braided to form a composite computation sequence. This compositional structure enables near-complete elimination of TP-related bubbles. Building upon this structure, we further design the PP schedule to minimize PP bubbles. Experimental results demonstrate that our approach improves training throughput by up to 12% for LLMs and 16% for MLLMs compared to existing scheduling methods. Our source code is avaiable at https://github.com/MICLAB-BUPT/STP.
format Preprint
id arxiv_https___arxiv_org_abs_2510_27257
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Synergistic Tensor and Pipeline Parallelism
Qi, Mengshi
Peng, Jiaxuan
Zhang, Jie
Zhu, Juan
Li, Yong
Ma, Huadong
Distributed, Parallel, and Cluster Computing
In the machine learning system, the hybrid model parallelism combining tensor parallelism (TP) and pipeline parallelism (PP) has become the dominant solution for distributed training of Large Language Models~(LLMs) and Multimodal LLMs (MLLMs). However, TP introduces significant collective communication overheads, while PP suffers from synchronization inefficiencies such as pipeline bubbles. Existing works primarily address these challenges from isolated perspectives, focusing either on overlapping TP communication or on flexible PP scheduling to mitigate pipeline bubbles. In this paper, we propose a new synergistic tensor and pipeline parallelism schedule that simultaneously reduces both types of bubbles. Our proposed schedule decouples the forward and backward passes in PP into fine-grained computation units, which are then braided to form a composite computation sequence. This compositional structure enables near-complete elimination of TP-related bubbles. Building upon this structure, we further design the PP schedule to minimize PP bubbles. Experimental results demonstrate that our approach improves training throughput by up to 12% for LLMs and 16% for MLLMs compared to existing scheduling methods. Our source code is avaiable at https://github.com/MICLAB-BUPT/STP.
title Synergistic Tensor and Pipeline Parallelism
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2510.27257