@inproceedings{huang-etal-2025-apb,
    title = "{APB}: Accelerating Distributed Long-Context Inference by Passing Compressed Context Blocks across {GPU}s",
    author = "Huang, Yuxiang  and
      Li, Mingye  and
      Han, Xu  and
      Xiao, Chaojun  and
      Zhao, Weilin  and
      Ao, Sun  and
      Zhou, Hao  and
      Zhou, Jie  and
      Liu, Zhiyuan  and
      Sun, Maosong",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.525/",
    pages = "10708--10727",
    ISBN = "979-8-89176-251-0",
    abstract = "While long-context inference is crucial for advancing large language model (LLM) applications, its prefill speed remains a significant bottleneck. Current approaches, including sequence parallelism strategies and compute reduction through approximate attention mechanisms, still fall short of delivering optimal inference efficiency. This hinders scaling the inputs to longer sequences and processing long-context queries in a timely manner. To address this, we introduce APB, an efficient long-context inference framework that leverages multi-host approximate attention to enhance prefill speed by reducing compute and enhancing parallelism simultaneously. APB introduces a communication mechanism for essential key-value pairs within a sequence parallelism framework, enabling a faster inference speed while maintaining task performance. We implement APB by incorporating a tailored FlashAttn kernel alongside optimized distribution strategies, supporting diverse models and parallelism configurations. APB achieves speedups of up to 9.2$\times$, 4.2$\times$, and 1.6$\times$ compared with FlashAttn, RingAttn, and StarAttn, respectively, without any observable task performance degradation."
}