English(EN) KVBuffer: IO-aware Serving for Linear Attention

KVBuffer 增强了长上下文AI模型的线性注意力服务

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-06 05:00

研究人员推出了一种新颖的服务机制 KVBuffer，旨在提高线性注意力模型的效率，特别是在长上下文推理方面。这种感知IO的方法缓冲了最近的键和值，实现了分块计算和推测解码的并行验证。KVBuffer 在 Qwen 3.5 和 Nemotron 3 等模型中通过 SGLang 实现，在批量解码方面显示出显著的加速，并增加了推测解码的服务容量。 AI

影响提高了长上下文AI模型的服务效率和延迟，可能促进线性注意力架构的更广泛应用。

排序理由这是一篇详细介绍AI模型服务新技术的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 14 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 这是一篇详细介绍AI模型服务新技术的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, infra; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Longwei Zou, Lin Zhong · 2026-11-06 05:00

KVBuffer：面向线性注意力（Linear Attention）的 IO 感知服务

arXiv:2605.19049v2 Announce Type: replace-cross Abstract: Linear attention has recently gained significant attention for long-context inference due to its constant decoding cost with respect to context length. However, existing serving systems typically serve linear attention by …

本文如何被排名

报道来源 [1]

KVBuffer：面向线性注意力（Linear Attention）的 IO 感知服务

相关实体

相关话题