Back to Search Start Over

LoongTrain: Efficient Training of Long-Sequence LLMs with Head-Context Parallelism

Authors :
Gu, Diandian
Sun, Peng
Hu, Qinghao
Huang, Ting
Chen, Xun
Xiong, Yingtong
Wang, Guoteng
Chen, Qiaoling
Zhao, Shangchun
Fang, Jiarui
Wen, Yonggang
Zhang, Tianwei
Jin, Xin
Liu, Xuanzhe
Publication Year :
2024

Abstract

Efficiently training LLMs with long sequences is important yet challenged by the massive computation and memory requirements. Sequence parallelism has been proposed to tackle these problems, but existing methods suffer from scalability or efficiency issues. We propose LoongTrain, a novel system to efficiently train LLMs with long sequences at scale. The core of LoongTrain is the 2D-Attention mechanism, which combines both head-parallel and context-parallel techniques to break the scalability constraints while maintaining efficiency. We introduce Double-Ring-Attention and analyze the performance of device placement strategies to further speed up training. We implement LoongTrain with the hybrid ZeRO and Selective Checkpoint++ techniques. Experiment results show that LoongTrain outperforms state-of-the-art baselines, i.e., DeepSpeed-Ulysses and Megatron Context Parallelism, in both end-to-end training speed and scalability, and improves Model FLOPs Utilization (MFU) by up to 2.88x.

Details

Database :
arXiv
Publication Type :
Report
Accession number :
edsarx.2406.18485
Document Type :
Working Paper