Metraipolis › archive › dc_fea0438bd8a5
KV-Pipe: On the Relation Between KV Sharing and Pipeline Parallel Efficiency in LLMs — arXiv:2608.15943
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.15943
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T19:02:06.950Z
- SHA-256 of the archived text
1698d8d89f07af29e584d536c90ace1e3c1520bb6b294f8ff43f6fba9855a917- Stable URL of this page
https://metraipolis.com/documento/dc_fea0438bd8a5
How to cite
KV-Pipe: On the Relation Between KV Sharing and Pipeline Parallel Efficiency in LLMs — arXiv:2608.15943 — arXiv (en), https://arxiv.org/abs/2608.15943 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fea0438bd8a5 (Metraipolis archive dc_fea0438bd8a5, sha256 1698d8d89f07af29e584d536c90ace1e3c1520bb6b294f8ff43f6fba9855a917)
BibTeX
@misc{dcfea0438bd8a5,
title = {KV-Pipe: On the Relation Between KV Sharing and Pipeline Parallel Efficiency in LLMs — arXiv:2608.15943},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.15943},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fea0438bd8a5; sha256 1698d8d89f07af29e584d536c90ace1e3c1520bb6b294f8ff43f6fba9855a917},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15943. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Pipeline parallelism (PP) is widely used to scale large language model (LLM) training, but its efficiency is often limited by stage imbalance and pipeline bubbles. Meanwhile, cross-layer KV sharing has primarily been studied as a mechanism for reducing KV-cache costs during inference, without examining how KV reuse reshapes pipeline workloads. We present \textbf{KV-Pipe}, a stage-aware KV-sharing mechanism that turns KV reuse into a pipeline-balancing control knob. KV-Pipe starts from the tail stage, converts selected attention layers to cross-layer KV sharing in a tail-first order, and iteratively retargets the current bottleneck to drive the FLOPs Imbalance Ratio (FIR) toward $1$. The procedure is performed offline and requires only a pipeline partition and per-layer FLOPs estimates, introducing negligible runtime overhead and requiring no online tuning. Across multiple pipeline-parallel configurations, KV-Pipe consistently improves utilization and throughput, achieving up to \textbf{9.2\%} higher training MFU and up to a \textbf{9.8\%} reduction in iteration time, with larger gains at higher pipeline-parallel degrees where stage imbalance is amplified. Furthermore, the same KV-sharing mechanism provides an inference-side benefit by reducing KV-cache growth and redundant KV projection work, resulting in higher decoding throughput for long-context workloads. These results identify KV layout as a system--architecture degree of freedom for jointly improving pipeline-parallel training efficiency and long-context inference.
Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-08-18 in cs.DC).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2608.15943 — 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 2026-10-18T18:07:16.074Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).