llama.cpp 项目发布多个预发布更新，包含各种修复和优化

作者 PulseAugur 编辑部 · [8 个来源] · 2026-10-13 21:33

llama.cpp 项目已发布多个预发布更新，包括 b11609，该更新修复了 WebGPU 的持续集成问题并解决了重复的专家 ID。版本 b11608 解决了转义特殊标记时递归正则表达式替换引起的服务器崩溃问题，防止了 SIGSEGV 错误。其他更新包括 Vulkan 池的优化、子图重新分配的改进以及 Hexagon 后端的配置电源模式。 AI

影响对一个流行的开源 LLM 推理引擎的次要改进。

排序理由这是一系列特定项目的软件发布说明，而非重大的行业事件。

在 llama.cpp — Releases 阅读 →

AI 生成摘要 · Google Gemini · 来自 8 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 4 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 这是一系列特定项目的软件发布说明，而非重大的行业事件。
Source corroboration: 8 independent sources; Strong cross-source corroboration — multiple independent publishers covered this within the clustering window.
Topics: product, 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: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [8]

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-14 00:55

b11609

<details open=""> webgpu: (Fix CI) fix mul_mat_id with duplicate expert ids (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30451">#30451</a>) <ul> <li> Fix mm-id of webgpu in the case of ids duplication </li> <li> formatting…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-14 00:38

b11608

<details open=""> server : escape the special tokens without std::regex (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30243">#30243</a>) The d1-omni and kev prompts escape "<|name|>" in the request text with std::regex…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-14 00:21

b11607

<details open=""> chat: replace output string with parts (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30389">#30389</a>) <ul> <li> chat: replace output string with parts </li> <li> reserve first </li> </ul> </details> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-13 22:39

b11606

<details open=""> ggml : simplify backend search path error handling (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28367">#28367</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-13 22:22

b11604

<details open=""> kleidiai: add SME2 Q4_K and Q6_K matmul support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29219">#29219</a>) Change-Id: Ib4aa4c9814c010f9cf084783cc80612e364f9a27 Change-Id: I406a6606c3fea51f8cd32331c1ed…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-13 22:04

b11603

<details open=""> vulkan: Optimize pool_2d when the pooling window is large (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30119">#30119</a>) When the window is larger than a subgroup, use a whole subgroup to compute each out…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-13 21:50

b11602

<details open=""> meta: recreate all subgraphs when the arena is reallocated (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30453">#30453</a>) <ul> <li> meta: recreate all subgraphs when the arena is reallocated </li> <li> c…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-13 21:33

b11601

<details open=""> hexagon: add configurable HTP DCVS/HMX power mode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/30423">#30423</a>) <ul> <li> hexagon: add configurable HTP DCVS/HMX power mode </li> <li> hexagon: address PR…

本文如何被排名

报道来源 [8]

相关实体

相关话题