English(EN) b11980: [SYCL] add DPAS flash attention kernel for quantized KV cache (#31125)

llama.cpp 为量化KV缓存添加DPAS闪存注意力内核

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-29 06:01

llama.cpp项目发布了b11980版本，为量化KV缓存引入了新的DPAS闪存注意力内核。此优化允许对小批量和量化KV块进行更快处理，从而提高在各种硬件和操作系统上的性能。更新特别针对XMX上的s8 x s4和s8 x s8 DPAS的Q.K操作，而V.P保持f16。 AI

影响提高了在消费级硬件上LLM的推理速度和效率。

排序理由这是一个开源项目的软件发布，用于优化现有模型的性能，而不是新模型发布或核心研究。

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: 18 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 这是一个开源项目的软件发布，用于优化现有模型的性能，而不是新模型发布或核心研究。
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, model release; 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。

llama.cpp — Releases TIER_1 English(EN) · Torchit1 · 2026-10-29 06:01

b11980: [SYCL] 为量化KV缓存添加DPAS闪存注意力内核 (#31125)

<ul> <li>sycl : add DPAS flash attention kernel for quantized KV cache</li> </ul> <p>Decode and small batches (1 to 31 query tokens) now run flash attention<br /> straight on q4_0 and q8_0 KV blocks, instead of scalar dequant or<br /> converting the whole cache to f16 per call. Q…