llama.cpp 发布包括服务器修复、OpenVINO 优化和平台支持扩展

作者 PulseAugur 编辑部 · [73 个来源] · 2026-09-30 21:25

llama.cpp 项目发布了多个更新，包括 b11379、b11378、b11377、b11376、b11375、b11374、b11372、b11371 和 b11370 版本。这些版本在不同平台和功能上引入了各种改进和修复。值得注意的变化包括服务器端中止修复、JSON schema 聊天解析增强、OpenVINO 优化以及 CUDA 和 Metal 后端更新。该项目还继续扩展对 Snapdragon 和 openEuler 等不同硬件架构和操作系统的支持。 AI

影响持续开发和优化本地 LLM 推理工具。

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 73 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 0 / 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: 73 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: 7 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring: +17 source(s) since last score; New sources have picked up this story since our last re-score. Score will update on the next scoring pass.

完整方法见我们的编辑标准。

报道来源 [73]

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 22:56

b11399

<details open=""> CUDA: refactor swizzling code (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29612">#29612</a>) <ul> <li> CUDA: refactor swizzling code </li> <li> fix templates/loop bounds </li> </ul> </details> <st…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 20:07

b11398

<details open=""> ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29806">#29806</a>) <ul> <li> ggml-cpu: vectorize BF16 K tails in tinyBLAS </li> <li> tests: Skip ti…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 19:50

b11397

<details open=""> cuda : move neu_padded to where it is used (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29940">#29940</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </detail…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 17:49

b11396

<details open=""> ci : windows llvm build requires ninja multi-config (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29959">#29959</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">ht…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 16:25

b11393

<details open=""> chat-peg-parser : clear current_tool when pending_tool_call is reset (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29942">#29942</a>) A TOOL_ID node that arrives after TOOL_CLOSE wrote through <code>current_tool<…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 14:43

b11392

<details open=""> ci : set default permissions (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29945">#29945</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] · 2026-10-04 14:12

b11391

<details open=""> cuda : move blocks_per_col to where it is used (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29939">#29939</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </de…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 12:56

b11390

<details open=""> CUDA: fix MMQ memory fault if n_expert >> n_ubatch (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29941">#29941</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollo…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 12:34

b11389

<details open=""> vulkan: fix rdna4 mat_vec tuning (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29934">#29934</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] · 2026-10-04 10:42

b11388

<details open=""> imatrix: calculate activation-based statistics for new format (GGUF) imatrices (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/14891">#14891</a>) <ul> <li>Use activations to calculate the stats</li> <li>Determine calc…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 10:22

b11387

<details open=""> spec : fix n-gram drafts rejected at temp > 0 after truncation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29924">#29924</a>) Co-authored-by: Pranesh Gonegandla <a href="mailto:[email protected]">pgoneg…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 10:03

b11386

<details open=""> common : prepare load_from_models_dir() for path conversion (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29674">#29674</a>) This is part of the fs::path modernization series. That was also the opportunity …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 09:37

b11385

<details open=""> server : fix dead LLAMA_ARG_HF_REPO_FILE key in preset allow-list (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29938">#29938</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">angt@huggingf…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-04 08:06

b11384

<details open=""> ci : pushing tag needs deploy key (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29937">#29937</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] · 2026-10-04 00:37

b11382

<details open=""> webgpu: add f16 support to fill/set_rows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29897">#29897</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 21:35

b11381

<details open=""> mtmd : fix deprecated strdup warning on Windows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29863">#29863</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </d…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 21:09

b11380

<details open=""> vendor : update cpp-httplib to 0.59.0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29886">#29886</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.ap…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 15:54

b11379

<details open=""> server : fix laya abort by limiting n_batch to n_ubatch (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29903">#29903</a>) <ul> <li>server : fix laya abort by limiting n_batch to n_ubatch</li> </ul> Fixes <a class=…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 15:02

b11378

<details open=""> common : add common_is_tty() helper and fix deprecated warnings on Windows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29860">#29860</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">angt…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 14:27

b11377

<details open=""> chat : honor json_schema in Ling 3.0 parser (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29813">#29813</a>) <ul> <li>chat : honor json_schema in Ling 3.0 parser</li> </ul> Ling 3.0 only built a grammar for tool …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 13:27

b11376

<details open=""> ci: fix flaky ADD_ADD f16 by using the fused ADD tolerance (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29904">#29904</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofol…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 12:40

b11375

<details open=""> graph: gather the recurrent states once so the reserve covers every split (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29856">#29856</a>) build_rs gathered the extra states (n_rs - n_seqs rows) with their own<br…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 11:54

b11374

<details open=""> ggml-openvino: update to 2026.4.1, optimize performance, expand ops, improve device listing. (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29852">#29852</a>) <ul> <li>ggml-openvino : Qwen3.5 MoE perf (<a class="issu…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 11:28

b11372

<details open=""> qwen4exp : halve the indexer score memory (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29825">#29825</a>) <ul> <li>qwen4exp : halve the indexer score memory</li> </ul> The indexer scored all heads in one product…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 08:36

b11371

<details open=""> model: add support for clef decision model (text-only) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29831">#29831</a>) <ul> <li> init support for clef (text only) </li> <li> more static graph </li> <l…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 06:27

b11370

<details open=""> CUDA: fuse shared experts into MMVQ (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29184">#29184</a>) <ul> <li> CUDA: fuse shared experts into MMVQ </li> <li> check if buffer is null </li> <li> move …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 02:38

b11368

<details open=""> spec : add probabilistic sampling for simple draft and MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/27694">#27694</a>) <ul> <li> Make the drafter probabilistic and the target verify by rejection sampling…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 01:58

b11366

<details open=""> ggml-quants : avoid invalid rounding in qkx3 scale search (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29817">#29817</a>) <ul> <li>ggml-quants : avoid invalid rounding in qkx3 scale search</li> </ul> The imatrix…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 01:32

b11365

<details open=""> ggml-cpu : fix soft_max_back wrong output when dst aliases src1 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/27096">#27096</a>) <ul> <li>ggml-cpu : fix soft_max_back wrong output when dst aliases src1</li> </ul> <p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 01:06

b11364

<details open=""> model: support nimble decision model (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29844">#29844</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-03 00:28

b11362

<details open=""> metal : add tensor API flash attention kernel for F16 KV (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29570">#29570</a>) <ul> <li> metal : add tensor API flash attention kernel for F16 KV </li> <li> cont …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 23:57

b11361

<details open=""> llama, server: add /v1/systemone API (models: laya, julia-1, lev, openjev, kev) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29818">#29818</a>) <ul> <li> init conversion </li> <li> convert: ok </li> <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 23:18

b11355

<details open=""> vulkan: disable large matmul tile on Samsung GPUs with 32KB shared memory (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28531">#28531</a>) Assisted-by: Claude Opus </details> Website: …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 22:23

b11352

<details open=""> qwen4exp : optimize mask constructions (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29824">#29824</a>) <ul> <li> qwen4exp : optimize mask constructions </li> <li> cont : apply the same change for GLM5-nex…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 18:55

b11351

<details open=""> ggml : add <code>alloc_buffer_n</code> to buffer type interface (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23671">#23671</a>) <ul> <li>ggml : add <code>alloc_buffer_n</code> to buffer type interface</li> </ul> <p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 15:43

b11349

<details open=""> vulkan: add logging to pipeline compile issues (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29794">#29794</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https:/…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 13:52

b11346

<details open=""> qwen4exp: fix tests (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29819">#29819</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></li> </ul> <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 13:11

b11345

<details open=""> hexagon: add q2_k and q3_k quant type support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29717">#29717</a>) <ul> <li> hexagon: add q2_k and q3_k quant type support </li> <li> hex-qk: consistent allocati…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 08:56

b11344

<details open=""> CUDA: fix 2 broken Volta FA cases (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29803">#29803</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] · 2026-10-02 05:09

b11342

<details open=""> common,rpc : fix cache dir creation through symlinks on buggy libstdc++ (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29816">#29816</a>) See <a href="https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101510" rel="nofo…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 04:32

b11339

<details open=""> llama : clamp kpool re-pool bound to existing pools (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29805">#29805</a>) <ul> <li> tests : simplify function signature </li> <li> llama : clamp kpool re-pool bou…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 04:07

b11338

<details open=""> hexagon: shared strided DMA copy for CPY and CONCAT, any-dim CONCAT via DMA (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29685">#29685</a>) <ul> <li> hexagon: shared strided DMA copy for CPY and CONCAT, any-dim …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 03:37

b11337

<details open=""> server: return HTTP 400 for invalid embedding requests (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29060">#29060</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow"…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 03:10

b11335

<details open=""> cuda : route sm70 to the Turing MMVQ nwarps table (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29753">#29753</a>) <ul> <li>cuda : route sm70 to the Turing MMVQ nwarps table</li> </ul> Volta (sm_70) has no MMVQ p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 02:37

b11334

<details open=""> metal : release temporary private transfer buffers (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29777">#29777</a>) <ul> <li>metal : release temporary private transfer buffers</li> </ul> Assisted-by: OpenAI Codex…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 01:57

b11333

<details open=""> webgpu: add bfloat16 support for MUL_MAT/MUL_MAT_ID/GET_ROWS- <a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29358">#29358</a> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29358">#…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 01:24

b11332

<details open=""> llama : fix invalid assert in recurrent memory (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29799">#29799</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https:/…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 00:51

b11331

<details open=""> CUDA: Handle compute type for NVFP4 on cublass path (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29173">#29173</a>) <ul> <li>CUDA: Handle compute type for NVFP4 on cublass path</li> </ul> Signed-off-by: ynankani…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-02 00:21

b11330

<details open=""> Qwen4Exp: add MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29761">#29761</a>) <ul> <li> Qwen4Exp: add MTP </li> <li> remove has_state member, check via ctx_bufs being non-empty </li> <li> consi…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 20:59

b11327

<details open=""> mtmd: cap max_image to ubatch for non_causal models (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29773">#29773</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">ht…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 20:33

b11326

<details open=""> meta: clear inactive AllReduce shards with FILL, not SCALE (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29793">#29793</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofol…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 20:02

b11325

<details open=""> jinja : skip copying loop scope unless a loop filter needs it (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29776">#29776</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="no…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 19:28

b11324

<details open=""> llama-mmap : avoid a second full-size copy of each tensor with direct-io (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29749">#29749</a>) Assisted-by: Claude Co-authored-by: Pranesh Gonegandla <a href="mai…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 18:48

b11323

<details open=""> HIP: avoid treating CDNA as dgx spark for gqa_ratio 20 in fattn_mma dqk 576 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29572">#29572</a>) </details> Website: <ul> <li><a href="https://llam…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 18:03

b11322

<details open=""> hex-workqueue: fix race condition in seqn getting out of sync with idx_read/write (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29785">#29785</a>) </details> Website: <ul> <li><a href="https:…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 14:13

b11321

<details open=""> BLAS : Document AOCL-BLAS build and label the device AOCL-BLAS (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29640">#29640</a>) <ul> <li> BLAS : Document AOCL-BLAS build and label the device AOCL-BLAS </li> <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 11:23

b11320

<details open=""> common : add LLM-jp-4.1 Harmony dialect handler (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29681">#29681</a>) LLM-jp-4.1 uses the GPT-OSS format, but its tokenizer decodes a space after every special tok…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 10:48

b11319

<details open=""> opencl: mark vec subgroup bcast as supproted for Adreno E17 compiler (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29698">#29698</a>) </details> Website: <ul> <li><a href="https://llama.app" …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 10:20

b11318

<details open=""> vocab : honor BOS/EOS settings for PLaMo-2 and PLaMo-3 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29734">#29734</a>) <ul> <li>vocab : honor BOS/EOS settings for PLaMo-2 and PLaMo-3</li> </ul> The original toke…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 09:36

b11317

<details open=""> llama-bench : fix verbosity filter to show GGML_LOG_ERROR (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28229">#28229</a>) <ul> <li> bench : fix verbosity filter to show GGML_LOG_ERROR (<a class="issue-link js-is…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 09:01

b11316

<details open=""> metal : use bf16 math for mxfp4 mul-mat (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29770">#29770</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 08:08

b11313

<details open=""> model : re-enable -sm tensor for qwen4exp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28569">#28569</a>) <a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/27941">#27941</a> di…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 03:11

b11312

<details open=""> webgpu: fix SSM_SCAN binding aliasing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29750">#29750</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.ap…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 02:49

b11311

<details open=""> ggml-opencl : replace alloca() with std::vector (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29765">#29765</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </d…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 02:09

b11310

<details open=""> cuda: guard the iq4_nl dequantize row kernel against short rows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29683">#29683</a>) dequantize_block_iq4_nl writes QK_K values per block, but a row can be shorter than…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 01:38

b11309

<details open=""> Hexagon: optimize ALLREDUCE with support for safe scatter mode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29757">#29757</a>) <ul> <li> hex-allreduce: add support for safe scatter mode </li> <li> hex-all…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 01:09

b11308

<details open=""> args: fix cli download mmproj arg (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28977">#28977</a>) <ul> <li> tests: add tests for cli download arg parsing </li> <li> args: fix cli download mmproj arg <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-10-01 00:37

b11307

<details open=""> llama : preserve original batch order for speculative decoding layer inputs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29019">#29019</a>) <ul> <li>llama: preserve original batch order for layer inputs</li> </ul> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-30 23:36

b11306

<details open=""> test-llama-archs : toggle causal_attn to catch graph shape changes (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29724">#29724</a>) After the device decode, flip causal_attn off, decode n_ubatch/2 then n_ub…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-30 23:08

b11304

<details open=""> llama: properly handle KV on training (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28520">#28520</a>) <ul> <li> llama: properly handle KV on training </li> <li> improve </li> </ul> </details> <stro…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-30 22:40

b11303

<details open=""> batch: migrate the rest of examples to llama_batch_ext (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29601">#29601</a>) <ul> <li> migrate the rest </li> <li> test-thread-safety </li> <li> rm common_…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-30 21:59

b11302

<details open=""> glm5-next: give dead indexer slots unique scatter rows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29745">#29745</a>) The sparse indexer mask is built with a set_rows scatter. Padded pools, absent sequenc…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-30 21:25

b11301

<details open=""> ggml/gguf : fix integer overflow (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29384">#29384</a>) <ul> <li> ggml: fix integer overflow guard for zero-element tensors </li> <li> ggml: validate number of ele…

本文如何被排名

报道来源 [73]

相关实体

相关话题