Metraipolis › archive › dc_001038370488

LiquidCP: Decoupling Attention Tasks from Communication and Kernel Boundaries in Context Parallelism — arXiv:2611.04401

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.04401
License (as archived): CC0-1.0
Archived: 2026-11-04T21:45:42.609Z
SHA-256 of the archived text: 35bc84c3477a293f4fe9dc4591d00f524c5ddeff006b323d41f1f138a152987a
Stable URL of this page: https://metraipolis.com/documento/dc_001038370488

How to cite

LiquidCP: Decoupling Attention Tasks from Communication and Kernel Boundaries in Context Parallelism — arXiv:2611.04401 — arXiv (en), https://arxiv.org/abs/2611.04401 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_001038370488 (Metraipolis archive dc_001038370488, sha256 35bc84c3477a293f4fe9dc4591d00f524c5ddeff006b323d41f1f138a152987a)

BibTeX

@misc{dc001038370488,
  title = {LiquidCP: Decoupling Attention Tasks from Communication and Kernel Boundaries in Context Parallelism — arXiv:2611.04401},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04401},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_001038370488; sha256 35bc84c3477a293f4fe9dc4591d00f524c5ddeff006b323d41f1f138a152987a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04401. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Context parallelism (CP) partitions long sequences across GPUs for LLM training and inference. For variable-length training and full prefill, attention-balanced placement can leave token loads uneven, while ring execution repeats overhead at each computation and communication step. In stage-serialized decode CP (DCP), long-running tasks delay dependent work and leave substantial GPU hardware idle. We propose LiquidCP, which unifies training and inference through a common model of Transfer, Compute, and Reduction tasks with tile-level dependencies, decoupling attention tasks from communication and kernel boundaries. For training and full prefill, MegaRing incorporates BuddyRing placement to balance attention and token loads, and coalesces ready KV regions during forward attention to amortize overhead. For decoding and mixed batches, MegaDCP selects history-KV splits by jointly modeling the completion of attention and its dependent reductions. A shared persistent GPU-resident engine executes both designs. Compared with the fastest baseline per configuration on eight H100 GPUs, MegaRing achieves geometric-mean core-attention speedups of $1.23\times$ forward and $1.18\times$ backward, and accelerates a Transformer layer by $1.08$--$1.16\times$. MegaDCP achieves geometric-mean core-attention speedups of $1.20\times$ for decode-only and $1.35\times$ for mixed batches. In vLLM serving with KV prefix cache, it achieves an average time-to-first-token speedup of $1.22\times$.

Tema degli esploratori: calcolo e infrastrutture (elenco arXiv cs.DC, nuovi lavori del giorno).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2611.04401 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-04T21:36:19.374Z dalla pagina https://arxiv.org/list/cs.DC/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).