Seq1F1B: Efficient Sequence-Level Pipeline Parallelism for Large Language Model Training

2024/06/05 by Ao Sun, Weilin Zhao, Sun, Ao +13 · 1 voice · 5 citations
Computer Science · #Distributed #FOS: Computer and information sciences #Natural Language Processing Techniques #Parallel #Speech Recognition and Synthesis #Topic Modeling #and Cluster Computing (cs.DC) #cs.DC

paper · pdf · doi:10.48550/arxiv.2406.03488

openalex publication_date 2024/06/05 · arxiv published 2024/06/05 · arxiv updated 2024/11/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

The emergence of large language models (LLMs) relies heavily on distributed training strategies, among which pipeline parallelism plays a crucial role. As LLMs' training sequence length extends to 32k or even 128k, the current pipeline parallel methods face severe bottlenecks, including high memory footprints and substantial pipeline bubbles, greatly hindering model scalability and training throughput. To enhance memory efficiency and training throughput, in this work, we introduce an efficient sequence-level one-forward-one-backward (1F1B) pipeline scheduling method tailored for training LLMs on long sequences named Seq1F1B. Seq1F1B decomposes batch-level schedulable units into finer sequence-level units, reducing bubble size and memory footprint. Considering that Seq1F1B may produce slight extra bubbles if sequences are split evenly, we design a computation-wise strategy to partition input sequences and mitigate this side effect. Compared to competitive pipeline baseline methods such as Megatron 1F1B pipeline parallelism, our method achieves higher training throughput with less memory footprint. Notably, Seq1F1B efficiently trains a LLM with 30B parameters on sequences up to 64k using 64 NVIDIA A100 GPUs without recomputation strategies, a feat unachievable with existing methods. Our source code is based on Megatron-LM, and now is avaiable at: https://github.com/MayDomine/Seq1F1B.git.

Cited by

• InfiniPipe: Elastic Pipeline Parallelism for Efficient Variable-Length Long-Context LLM Training
• TokenSmith: Streamlining Data Editing, Search, and Inspection for Large-Scale Language Model Training and Interpretability
• CrossPipe: Towards Optimal Pipeline Schedules for Cross-Datacenter Training
• NoLoCo: No-all-reduce Low Communication Training Method for Large Models
• SlimPipe: Memory-Thrifty and Efficient Pipeline Parallelism for Long-Context LLM Training

Discussions

• Seq1F1B: Efficient Sequence-Level Pipeline Parallelism for LLM Training [hn, 1 points, 0 comments]

Related

• SPPO:Efficient Long-sequence LLM Training via Adaptive Sequence Pipeline Parallel Offloading
• SlimPipe: Memory-Thrifty and Efficient Pipeline Parallelism for Long-Context LLM Training
• Efficient Parallelization Layouts for Large-Scale Distributed Model Training
• Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM
• Efficient Long Context Fine-tuning with Chunk Flow