PulseAugur
实时 08:56:46
English(EN) WiSP: A Working-Set View of Mixture-of-Experts Serving on Extremely Low-Resource Hardware

WiSP 使大型 MoE 模型能够在低资源 GPU 上运行

研究人员开发了 WiSP(Working-Set Paging),一个新颖的系统,旨在使大型专家混合(MoE)模型能够在低资源硬件上运行,例如 24GB RTX 3090 GPU。WiSP 将 MoE 服务视为一个运行集问题,管理路由专家权重和 KV 缓存之间在有限 VRAM 上的竞争。在相同的内存限制下,该系统实现了高达静态卸载方法两倍的解码吞吐量。此外,还引入了一种称为 MV-WSA(Marginal-Value Working-Set Allocation)的内存分配策略,用于动态地在驻留专家和 KV 缓存之间分配 VRAM,从而提高预填充和解码操作的性能。 AI

影响 使得大型 MoE 模型能够部署在消费级硬件上,可能降低本地 AI 应用的门槛。

排序理由 该集群包含一篇学术论文,详细介绍了服务大型 AI 模型的新技术方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

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

WiSP 使大型 MoE 模型能够在低资源 GPU 上运行

本文如何被排名

Signal score
15 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇学术论文,详细介绍了服务大型 AI 模型的新技术方法。[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
infra, paper
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]

  1. arXiv cs.LG TIER_1 English(EN) · Jiamu Zhang, Liang Wu, Mayank Darbari, Liangjie Hong ·

    WiSP:一种在极低资源硬件上实现混合专家模型服务的运行集视图

    arXiv:2606.21868v2 Announce Type: replace Abstract: Modern local and agentic workloads often need large-model capacity at low concurrency, but run on GPUs that cannot keep a frontier-scale model resident. Mixture-of-Experts (MoE) models are a natural fit because they activate onl…