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.
@article{arxiv.2406.18485,
title = {LoongTrain: Efficient Training of Long-Sequence LLMs with Head-Context Parallelism},
author = {Diandian Gu and Peng Sun and Qinghao Hu and Ting Huang and Xun Chen and Yingtong Xiong and Guoteng Wang and Qiaoling Chen and Shangchun Zhao and Jiarui Fang and Yonggang Wen and Tianwei Zhang and Xin Jin and Xuanzhe Liu},
journal= {arXiv preprint arXiv:2406.18485},
year = {2024}
}