llama.cpp releases bring OpenVINO updates, Vulkan, GGUF, and SYCL improvements

By PulseAugur Editorial · [57 sources] · 2026-09-16 13:10

The llama.cpp project has released several updates, including version b11024 which features an update to OpenVINO 2026.4 and fixes for various compiler warnings. Other recent releases, such as b11022 and b11020, introduced improvements to Vulkan support and message delimiter parsing for DeepSeek models, respectively. Version b11019 addressed issues with GGUF file alignment and LoRA loading, while b11013 resolved buffer alignment problems in Vulkan shaders. Earlier updates like b11017 and b11016 focused on Vulkan and SYCL optimizations, and b11015 included fixes for OpenCL warnings. AI

IMPACT These updates to llama.cpp enhance its performance and compatibility across various hardware and software configurations, potentially improving the efficiency of running large language models locally.

RANK_REASON The cluster consists of multiple release notes for the llama.cpp project, detailing incremental software updates and bug fixes rather than a novel product launch or significant research breakthrough.

Read on llama.cpp — Releases →

AI-generated summary · Google Gemini · from 57 sources. How we write summaries →

llama.cpp releases bring OpenVINO updates, Vulkan, GGUF, and SYCL improvements

How we ranked this

Signal score: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; The cluster consists of multiple release notes for the llama.cpp project, detailing incremental software updates and bug fixes rather than a novel product launch or significant research breakthrough.
Source corroboration: 57 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: 55 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring: +37 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.

Full methodology in our editorial standards.

COVERAGE [57]

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 19:54

b11081

<details open=""> test-llama-archs : make tensor data stdev configurable and improve help (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29133">#29133</a>) <ul> <li>test-llama-archs : make tensor data stdev configurable</li> </ul> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 19:21

b11080

<details open=""> tests/test-backend-ops : allow regex entries in the -o filter (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29204">#29204</a>) <ul> <li>tests/test-backend-ops : allow regex entries in the -o filter</li> </ul> so …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 18:52

b11078

<details open=""> args: add env vars for temperature, top-p, min-p and penalties (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/27380">#27380</a>) Allow configuring --temp, --top-p, --min-p, --repeat-penalty, --presence-penal…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 18:14

b11077

<details open=""> server : do not forward --api-key-file to router-spawned child instances (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28938">#28938</a>) In router mode, authentication belongs to the router. unset_reserved_args(…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 17:16

b11076

<details open=""> tests : remove stale comment (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29140">#29140</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] · 2026-09-21 16:46

b11075

<details open=""> ggml-metal : simplify fusion pattern op list declaration (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29206">#29206</a>) <ul> <li>ggml-metal : derive non-empty fusion ops from ops_all</li> </ul> Assisted-by: pi:…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 16:04

b11074

<details open=""> json: Fixed json enum handling (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28518">#28518</a>) <ul> <li>Fixed json enum handling</li> </ul> Added common_json_value handling for enum values. Added tests/tes…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 15:33

b11073

<details open=""> sycl : coalesce MKL-FA softmax loads instead of one work-item per row (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28918">#28918</a>) <ul> <li> sycl : coalesce MKL-FA softmax loads instead of one work-item per r…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 14:15

b11071

<details open=""> ci : Upgrade CUDA to 13.4 for Ubuntu CUDA Release Builds (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29202">#29202</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollo…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 10:53

b11070

<details open=""> hexagon: overhaul of buffer and DMA handling to support 64bit mappings + improvements (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29197">#29197</a>) <ul> <li>hex-dma64: enable support extended buffer mappings and …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 10:17

b11069

<details open=""> cuda : tune MMVQ to MMQ crossover for SM70 (Volta) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28912">#28912</a>) <ul> <li>tune MMVQ to MMQ crossover for SM70 (Volta)</li> </ul> Signed-off-by: Yangyu Chen <a hr…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 09:44

b11068

<details open=""> metal : fix deprecation warnings from macOS 27 SDK (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29136">#29136</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">htt…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-21 08:19

b11067

<details open=""> webgpu : add fused gdn + cpy (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28976">#28976</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] · 2026-09-20 21:56

b11065

<details open=""> CUDA: tune FA for Gemma 4 on Ampere or newer (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29152">#29152</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://l…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-20 15:23

b11064

<details open=""> metal : support arbitrary hc in dsv4_hc_pre (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29169">#29169</a>) the dsv4_hc_pre kernels hardcoded hc = 4 via a constexpr used with simd_shuffle, so the op was re…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-20 12:31

b11063

<details open=""> common/peg : handle invalid utf-8 sequences in the AST (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29161">#29161</a>) <ul> <li> common/peg : handle invalid utf-8 sequences in the AST </li> <li> cont : re…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-20 09:50

b11062

<details open=""> CUDA: enable sparse fa for qwen4 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28770">#28770</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] · 2026-09-20 06:09

b11060

<details open=""> fix(mamba) : make time-step projection input contiguous (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28832">#28832</a>) <ul> <li>mamba : make time-step projection input contiguous</li> </ul> Assisted-by: ChatGPT…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-20 05:44

b11059

<details open=""> metal: add F16 input to the FWHT (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29094">#29094</a>) <ul> <li>metal: add F16 input to the FWHT</li> </ul> The Metal FWHT kernel accepts F32 input only. This change mak…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-20 00:01

b11057

<details open=""> chat : add dedicated Ling 3.0 (Bailing V3) parser (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28682">#28682</a>) <ul> <li>chat: add dedicated Ling 3.0 (Bailing V3) parser</li> </ul> Ling 3.0 Flash templates pre…
llama.cpp — Releases TIER_1 (SO) · aldehir · 2026-09-19 23:59

b11058

chat : fix gemma4 required tool grammar (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29115">#29115</a>)
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-19 17:55

b11056

<details open=""> hexagon: enable I32 GET_ROWS (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29116">#29116</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] · 2026-09-19 17:25

b11055

<details open=""> hexagon: add support for GEGLU_QUICK (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29114">#29114</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-19 16:50

b11054

<details open=""> hexagon: enable support for TOP_K op (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29113">#29113</a>) <ul> <li> hexagon: enable support for TOP_K op </li> <li> hex-topk: thread single-row TOP_K, raise VTCM…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 15:05

b11040

<details open=""> ggml : check for allocation failures to prevent crashes (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28149">#28149</a>) <ul> <li> ggml : check for allocation failures to prevent crashes </li> <li> wording…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 14:13

b11039

<details open=""> Model-Saver: Write the SWA pattern, 15 more architectures roundtrip (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29042">#29042</a>) <ul> <li>llama: read the SWA pattern as a period or a per-layer array</li> </ul> <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 12:16

b11037

<details open=""> ggml-webgpu: fix supports_op condition for GET_ROWS (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28978">#28978</a>) <ul> <li> fix get_rows vec4 handling </li> <li> Add src strides checking to vec4_aligned…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 11:37

b11036

<details open=""> ggml : handle graph buffer reservation failure (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/26070">#26070</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https:/…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 11:07

b11035

<details open=""> vulkan: add IQ3_S MMQ matmul kernels (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28822">#28822</a>) <ul> <li> vulkan: add IQ3_S MMQ matmul kernels </li> <li> Make block_a_to_shmem do 2-byte loads (110 by…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 10:40

b11034

<details open=""> vocab : add ufakzeka pre-tokenizer (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29033">#29033</a>) <ul> <li> vocab : add ufakzeka pre-tokenizer </li> <li> vocab : move ufakzeka to the models list and rege…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 09:58

b11033

<details open=""> cmake : fix build when GGML_CPU=OFF and GGML_CUDA=ON (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29026">#29026</a>) <ul> <li> fix: build fails when GGML_CPU=OFF and GGML_CUDA=ON </li> <li> fix: eol in ex…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 08:16

b11030

<details open=""> ci : bump android-actions/setup-android to 4.0.4 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29065">#29065</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-18 07:43

b11029

<details open=""> vulkan: raise the hoisted row-id limit for mul_mat_id from 256 to 1024 experts (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28501">#28501</a>) <ul> <li>vulkan: raise the hoisted row-id limit for mul_mat_id to 512 e…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 17:30

b11028

<details open=""> ci : add missing evict-old-files (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29041">#29041</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] · 2026-09-17 16:03

b11027

<details open=""> rpc : skip ACCEL devices (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29020">#29020</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] · 2026-09-17 13:34

b11026

<details open=""> model : skip gate_up_exps if TENSOR_SKIP is set (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29014">#29014</a>) required for qwen35moe if MTP tensors are fused but not loaded </details> Website:</…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 13:06

b11025

<details open=""> model : extend Nemotron MTP support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29018">#29018</a>) <ul> <li> first fix </li> <li> removed unnecessary declarations </li> </ul> </details> We…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 12:33

b11024

<details open=""> openvino : Update OpenVINO to 2026.4;fix clangd,MSVC warnings; (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29009">#29009</a>) <ul> <li> Update to openvino-2026.4 </li> <li> Update OV docs </li> <li> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 11:39

b11022

<details open=""> vulkan: split buffers and debug code into separate files, add shared headers (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28732">#28732</a>) </details> Website: <ul> <li><a href="https://lla…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 11:10

b11020

<details open=""> chat : add message delimiters to the DeepSeek V3.2/V4 parser (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29008">#29008</a>) <ul> <li>chat : add message delimiters to the DeepSeek V3.2/V4 parser</li> </ul> Assis…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 10:39

b11019

<details open=""> gguf : align the data section relative to the GGUF start, not the file (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28993">#28993</a>) <ul> <li>gguf : align the data section relative to the GGUF start, not the file…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 10:03

b11018

<details open=""> sycl : fix the B70 mem allocate error when >19.3GB (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28953">#28953</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 09:31

b11017

<details open=""> vulkan: skip unneeded MoE work in mul_mm coopmat1 path (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25483">#25483</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow"…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 09:01

b11016

<details open=""> sycl: ssm_conv: fuse the SiLU epilogue into the ssm_conv kernel (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28929">#28929</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 08:34

b11015

<details open=""> opencl: fix various warnings (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28984">#28984</a>) <ul> <li> opencl: fix warnings </li> <li> opencl: fix warnings for non adreno </li> </ul> </details> <st…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 07:54

b11013

<details open=""> vulkan: fix buffer_reference alignment in im2col shaders (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28996">#28996</a>) Both im2col.comp and im2col_3d.comp declare D_ptr without an explicit buffer_referen…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 06:29

b11012

<details open=""> vulkan: support qwen4exp hc ops (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28988">#28988</a>) <ul> <li> vulkan: support qwen4exp hc ops </li> <li> fix stale comment [no-ci] </li> </ul> </details> <p…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 02:50

b11011

<details open=""> [SYCL] Fix function signature for <code>ggml_backend_sycl_split_buffer_type</code> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28981">#28981</a>) </details> Website: <ul> <li><a href="https…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-17 00:05

b11010

<details open=""> vulkan: work around NV bug with argsort_large.comp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28975">#28975</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">htt…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 19:41

b11009

<details open=""> TP: fix split state and granularity for fused QKV gemma4, qwen35 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28965">#28965</a>) <ul> <li>model: calculate split states for attn_qkv from n_head * n_embd_head_k</li> …
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 17:49

b11007

<details open=""> Enable CUDA graph for MTP draft (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28549">#28549</a>) <ul> <li> Improve CUDA graph usage for MTP </li> <li> Rename field </li> <li> Address review feedback…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 17:22

b11006

<details open=""> hexagon: Support for K-Quants Q4_K and Q6_K (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28994">#28994</a>) implement q6k/q4k kernels Squashed from: feat: implement q6k kernel hex-q6k: improve…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 16:53

b11005

<details open=""> hexagon: accept the zeroed rope probe in supports_op (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28995">#28995</a>) llama probes weight placement with a rope where all params are 0, so rejecting n_dims ==…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 15:31

b11003

<details open=""> model : add support for HrmTextForCausalLM (DFM Mimir 1B) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/27625">#27625</a>) <ul> <li>model : add support for HrmTextForCausalLM (DFM Mimir 1B)</li> </ul> HRM-Text ru…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 14:41

b11002

<details open=""> CUDA/HIP: improve access patterns in im2col (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/28013">#28013</a>) </details> Website: <ul> <li><a href="https://llama.app" rel="nofollow">https://ll…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 14:05

b11001

<details open=""> spacemit : fix wrong transpose function for int16 data (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/25161">#25161</a>) The <code>sizeof(int16_t)</code> branch in <code>permute_transpose_impl</code> calls <…
llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-09-16 13:10

b11000

<details open=""> rpc : invalidate cached compute graph when a referenced buffer is freed (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24292">#24292</a>) The server caches the most recent compute graph per device so that GR…

How we ranked this

COVERAGE [57]

RELATED ENTITIES

RELATED TOPICS