llama.cpp 发布带来 Vulkan SDK 更新和性能优化 · 跟踪 10 个来源

作者 PulseAugur 编辑部 · [15 个来源] · 2026-07-31 09:47

llama.cpp 项目发布了多个更新，包括 b10211 版本，该版本将 Vulkan SDK 更新至 1.4.357.0。之前的版本，如 b10210，包括了对服务器端令牌重放的修复。其他更新则侧重于优化 SYCL 和 CUDA 性能，改进各种硬件和数据类型的 flash attention，以及增强矩阵乘法运算。 AI

影响 llama.cpp 的持续优化和错误修复提高了本地 LLM 推理的性能和稳定性。

排序理由该集群包含 llama.cpp 项目的多个发布说明，详细介绍了软件更新和错误修复，而不是新产品发布或研究突破。

在 llama.cpp — Releases 阅读 →

AI 生成摘要 · Google Gemini · 来自 15 个来源。我们如何撰写摘要 →

llama.cpp 发布带来 Vulkan SDK 更新和性能优化 · 跟踪 10 个来源

报道来源 [15]

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 22:06

b10216

<details open=""> vulkan: add POOL_1D op (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25431">#25431</a>) <ul> <li>vulkan : add pool1d push constants and pipeline field</li> </ul> Declared data structures needed for POOL1D OP, whi…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 21:33

b10215

<details open=""> vulkan: Introduce driver version check for Windows Intel GPU to mitigate crashing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25192">#25192</a>) <ul> <li>Removed crash guard for Intel</li> </ul> Crash fixed fro…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 20:47

b10214

<details open=""> mtmd: add n_embd_head (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26342">#26342</a>) Co-authored-by: Daniel Han <a href="mailto:[email protected]">[email protected]</a> </details> Website:</s…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 20:12

b10213

<details open=""> Support rotated kv cache quant (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26180">#26180</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 19:27

b10212

<details open=""> llama : load MTP tensors only if they are really used (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26296">#26296</a>) <ul> <li> llama : load MTP tensors only if they are really used </li> <li> llama : ski…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 18:53

b10211

<details open=""> vulkan: update vulkan sdk to 1.4.357.0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26303">#26303</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.a…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 18:11

b10210

<details open=""> server: correct accepted tokens when need draft token replay (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26320">#26320</a>) <ul> <li> spec: correct accepted tokens when need draft token replay </li> <li> <p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 17:30

b10209

<details open=""> cuda: extract Q2_0 elements via __byte_perm (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25603">#25603</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://ll…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 16:05

b10208

<details open=""> SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt proc… (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25025">#25025</a>) <ul> <li> SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 15:14

b10207

<details open=""> [SYCL] support the missed types in cpy (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26005">#26005</a>) <ul> <li> support the missed types in cpy </li> <li> use correct funct </li> <li> rm unused co…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 13:28

b10205

<details open=""> ggml-zendnn : group matmul direct API for mul_mat_id (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25918">#25918</a>) <ul> <li> ggml-zendnn : group matmul API for mul_mat_id </li> <li> ggml-zendnn : scale …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 12:46

b10204

<details open=""> sycl : support dev2dev memcpy by DEV2DEV_MEMCPY_FORWARD (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26234">#26234</a>) Co-authored-by: Neo Zhang Jianyu <a href="mailto:[email protected]">jianyu.zhang@intel…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 12:16

b10203

<details open=""> [SYCL] Support q2 mul_mat (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26231">#26231</a>) <ul> <li> support q2_0 in mul_mat </li> <li> support more q2_0 case </li> </ul> </details> Website:…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 11:43

b10202

<details open=""> sycl: fuse RMS_NORM + MUL (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26015">#26015</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></li> <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-07-31 09:47

b10201

<details open=""> ggml-webgpu: improve flash_attn_vec for quantized KV at long contexts (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25956">#25956</a>) <ul> <li> improve fa of quantized kv cache </li> <li> Fix some bugs an…

报道来源 [15]

相关实体

相关话题