llama.cpp releases add Vulkan, optimize matrix math, and improve server logging

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-25 09:23

b9789

<details open=""> quant : fix quantizing moe with mtp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24986">#24986</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/dow…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-25 08:47

b9788

<details open=""> sycl : support --split-mode tensor (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24152">#24152</a>) <ul> <li> Sycl tp stage1 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/issues…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-25 06:09

b9787

<details open=""> sycl : fix the failed UT cases of conv_3d (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24900">#24900</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-25 03:02

b9786

<details open=""> opencl: support non-contig rows in norm (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24965">#24965</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-25 01:23

b9785

<details open=""> chat: harden caps check (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24973">#24973</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9785/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 19:57

b9784

<details open=""> hexagon: MUL_MAT and MUL_MAT_ID rework : 32x32 tiled weight repack, kernel-params, cached graphs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24954">#24954</a>) <ul> <li> hex-mm: new weight layout and fusion upd…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 16:46

b9782

<details open=""> common: remove unused json-partial (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24968">#24968</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/down…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 15:03

b9781

<details open=""> vulkan: allow reducing the graph submission batches to avoid timeouts (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24872">#24872</a>) </details> macOS/iOS: <ul> <li><a href="https://github.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 10:21

b9780

<details open=""> vulkan: fail the build when a shader fails to compile (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24450">#24450</a>) <ul> <li>vulkan-shaders-gen: fail the build when a shader fails to compile</li> </ul> vulkan-…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 07:44

b9777

<details open=""> model : Add LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24913">#24913</a>) <ul> <li> model : Add LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M </li> <li> Res…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-24 04:01

b9776

<details open=""> vulkan: Apply bias before softmax in FA, to avoid overflow (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24909">#24909</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 16:49

b9775

<details open=""> server : check draft context creation error (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24922">#24922</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/rele…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 16:18

b9774

<details open=""> vulkan: support all backend tests for SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU/NORM (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24582">#24582</a>) <ul> <li> vulkan: make SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU use unary.com…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 15:45

b9773

<details open=""> vulkan: Support GET_ROWS_BACK (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24883">#24883</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 13:29

b9771

<details open=""> vulkan: make mul_mm ALIGNED a spec constant (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24689">#24689</a>) This trims down some of the shader variant explosion and reduces binary size. </details> <strong…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 12:27

b9770

<details open=""> server: fix remote preset handling, add test (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24938">#24938</a>) <ul> <li> server: add test for remote preset </li> <li> fix remote preset handling </li> <l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 11:45

b9769

<details open=""> vulkan: link ggml-cpu when GGML_VULKAN_CHECK_RESULTS / RUN_TESTS are enabled (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24444">#24444</a>) The result-checking and test debug paths in ggml-vulkan.cpp call ggml_…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 10:50

b9768

<details open=""> model: Granite Speech Plus (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24818">#24818</a>) <ul> <li>feat: Add conversion support for Granite Speech Plus</li> </ul> Branch: GraniteSpeechPlus AI-usage: full …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-23 09:00

b9767

<details open=""> ggml-webgpu: improve MTP inference by using mat-vec path for small batches (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24811">#24811</a>) <ul> <li> ggml-webgpu: improve small batches decoding </li> <li> …

llama.cpp — Releases TIER_1 (SO) · aldehir · 2026-06-23 05:27

b9765

server: improve user message detection and create checkpoints at ever…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 21:33

b9763

<details open=""> server : Add id to tool call responses api (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24882">#24882</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 17:04

b9761

<details open=""> server: (router) move model downloading to dedicated process (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24834">#24834</a>) <ul> <li> server: real-time model load progress tracking via /models/sse </li> <li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 15:24

b9760

<details open=""> server: refactor/generalize input file schema (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24299">#24299</a>) <ul> <li> server: refactor/generalize input file schema </li> <li> wire up input_video, accept…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 12:22

b9758

<details open=""> [SYCL] support bf16 on bin_bcast OP and unary OPs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24838">#24838</a>) <ul> <li> support bf16 on bin_bcast OP and unary OPs </li> <li> support the older Intel co…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 11:47

b9757

<details open=""> sampling : remove unconditional softmax+sort in top-n-sigma sampler (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22645">#22645</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-22 09:34

b9756

<details open=""> server: fix edit_file crash on append at end of file (line_start -1) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24893">#24893</a>) line_start -1 normalized to n+1, so append inserted at lines.begin() + n + 1,<…

llama.cpp — Releases TIER_1 (SO) · aafsmarak · 2026-06-22 03:48

b9755

docs/android.md: Add dependency <code>libandroid-spawn</code> for building in te…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 21:57

b9754

<details open=""> common/peg : implement ac parser for stricter grammar generation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24869">#24869</a>) <ul> <li> common/peg : implement ac parser </li> <li> cont : extract functi…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 19:25

b9753

<details open=""> server: fix report progress for loading spec models, add "stages" list (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24870">#24870</a>) <ul> <li> server: fix report progress for loading spec models, add "stages" …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 19:01

b9752

<details open=""> server: refactor batch construction (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24843">#24843</a>) <ul> <li> server: refactor batch construction </li> <li> wip </li> <li> wip 2 </li> <li> w…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 18:36

b9751

<details open=""> mtmd: fix mtmd_get_memory_usage (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24867">#24867</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/downloa…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 18:04

b9750

<details open=""> jinja : implement call statement (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24847">#24847</a>) <ul> <li> implement call statement </li> <li> undo unintended change </li> <li> de-lambda </li> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 17:11

b9748

<details open=""> server: add "verbose" field to schema (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24864">#24864</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/d…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 13:52

b9747

<details open=""> server: real-time model load progress tracking via /models/sse (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24828">#24828</a>) <ul> <li> server: real-time model load progress tracking via /models/sse </li> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 11:40

b9745

<details open=""> spec : Support Step3.5/3.7 flash mtp3 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24340">#24340</a>) <ul> <li> add mtp_layer_offset + include nextn flags in graph reuse </li> <li> add llama_set_mtp_layer…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-21 02:47

b9744

<details open=""> common/peg : refactor until gbnf grammar generation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24839">#24839</a>) <ul> <li> common/peg : refactor until gbnf grammar into an ac automaton </li> <li> cont …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 23:10

b9743

<details open=""> common/json-schema-to-grammar : align spacing rules with parsers (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24835">#24835</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/gg…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 22:33

b9742

<details open=""> fix(hexagon): use padded stride for ssm-conv weights (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24470">#24470</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 18:42

b9741

<details open=""> llama : use LLM_KV for quantization_version & file_type (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24802">#24802</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 18:15

b9740

<details open=""> arg: try fixing test-args-parser randomly fails (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24826">#24826</a>) <ul> <li> arg: try fixing test-args-parser randomly fails </li> <li> return ref </li> <l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 15:30

b9739

<details open=""> release: add missing link for win opencl adreno arm64 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24809">#24809</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llam…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 14:08

b9738

<details open=""> server: avoid forwarding auth headers in CORS proxy (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24373">#24373</a>) <ul> <li> server: avoid forwarding auth headers in CORS proxy </li> <li> format </li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 12:01

b9736

<details open=""> model : glm-dsa load DSA indexer tensors as optional (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24770">#24770</a>) GLM-5.2 ships the DSA "lightning indexer" on only a subset of layers (the "full" layers;…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 11:35

b9735

<details open=""> ggml : optimize AMX (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24806">#24806</a>) Flatten the partition over n_batch * M so every thread participates in the quantization <div class="snippet-clipboard…

llama.cpp — Releases TIER_1 (SO) · aldehir · 2026-06-20 10:54

b9737

docker : prebuild web UI for s390x build [no release] (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24829">#24829</a>)

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-20 00:17

b9733

<details open=""> ggml-webgpu: add adapter toggles for F16 on Vulkan + NVIDIA </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9733/llama-b9733-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 23:45

b9732

<details open=""> server: refactor child --> router communication (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24821">#24821</a>) <ul> <li> server: refactor child --> router communication </li> <li> fix wakeup case</…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 21:56

b9731

<details open=""> server : optimize get_token_probabilities (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24796">#24796</a>) Use std::partial_sort to order only the requested top-n tokens instead of the full vocabulary <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 21:20

b9730

<details open=""> mtmd, arg: fix utf8 handling on windows (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24779">#24779</a>) <ul> <li> mtmd, arg: fix utf8 handling on windows </li> <li> also fix ggml_fopen </li> <li> f…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 20:49

b9729

<details open=""> server: remove all internal mentions about "webui" (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24817">#24817</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 16:20

b9728

<details open=""> arg: Add comment line support to --api-key-file (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23168">#23168</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 15:47

b9727

<details open=""> vendor : update cpp-httplib to 0.48.0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24787">#24787</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/d…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 15:17

b9726

<details open=""> server: add --agent arg, remove redundant webui naming compat (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24801">#24801</a>) <ul> <li> server: add --agent arg, remove redundant webui naming compat </li> <li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 11:42

b9724

<details open=""> mtmd: several bug fixes (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24784">#24784</a>) <ul> <li> mtmd: several bug fixes </li> <li> fix build </li> <li> fix gemma4ua </li> <li> add sanity c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 11:02

b9723

<details open=""> spec: support eagle3 for qwen3.5 & 3.6 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24593">#24593</a>) <ul> <li> spec: support qwen3.5 & 3.6 eagle3 draft </li> <li> eagle3: Add deferred boundary c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 09:27

b9722

<details open=""> server: fix non-bound n_discard value (ctx shifting) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24786">#24786</a>) <ul> <li> server: fix non-bound n_discard value </li> <li> Update tools/server/server-c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 08:11

b9721

<details open=""> sync : ggml </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9721/llama-b9721-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI ena…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 07:34

b9718

<details open=""> server : consolidate slot selection into get_available_slot (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24755">#24755</a>) Absorb get_slot_by_id logic into get_available_slot so slot selection is handled …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 06:34

b9717

<details open=""> ggml-cpu: support K tails in power10 Q8/Q4 MMA matmul (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24753">#24753</a>) <ul> <li>ggml-cpu: support K tails in Power10 MMA Q8/Q4 matmul</li> </ul> This patch removes …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 03:56

b9716

<details open=""> mtmd: add batching support for internvl (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24775">#24775</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 03:27

b9715

<details open=""> Ggml/cuda col2im 1d (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24417">#24417</a>) <ul> <li> cuda: add GGML_OP_COL2IM_1D, follow-up to the CPU op </li> <li> cuda: col2im_1d use fast_div_modulo for the in…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 02:58

b9714

<details open=""> server: add "X-Accel-Buffering": "no" header to streaming endpoints (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24774">#24774</a>) <ul> <li>server: add "X-Accel-Buffering": "no" header to streaming endpoints</li> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 02:27

b9713

<details open=""> mtmd: add batching for mtmd-cli, add video tests (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24778">#24778</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 01:59

b9712

<details open=""> cmake : fix ui build with read-only source (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24752">#24752</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 01:35

b9711

<details open=""> mtmd: refactor llava-uhd overview image handling (always use ov_img_first) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24769">#24769</a>) <ul> <li> add dedicated "overview" for mtmd_image_preproc_out </li> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-19 00:28

b9707

<details open=""> server: add "schema" and validation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24150">#24150</a>) <ul> <li> wip </li> <li> working </li> <li> correct some limits </li> <li> add field name …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 15:48

b9704

<details open=""> server : return HTTP 400 on invalid grammar (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/issues/24144">#24144</a>) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24154">#24154</a>)…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 15:15

b9703

<details open=""> server: (router) rework -hf preset repo (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24739">#24739</a>) <ul> <li> server: temporary remove HF remote preset </li> <li> rework remove preset.ini support …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 14:36

b9702

<details open=""> server: fix router args not being forwarded to child instances (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24760">#24760</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 13:42

b9701

<details open=""> mtmd: refactor preprocessor, add mtmd_image_preproc_out (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24736">#24736</a>) <ul> <li> add mtmd_image_preproc_out </li> <li> add dev docs </li> <li> remov…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 11:14

b9700

<details open=""> [SYCL] rename GGML_SYCL_SUPPORT_LEVEL_ZERO (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24719">#24719</a>) <ul> <li> rename GGML_SYCL_SUPPORT_LEVEL_ZERO to GGML_SYCL_SUPPORT_LEVEL_ZERO_API, and GGML_SYCL_ENABLE_…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 09:55

b9699

<details open=""> sycl : support MUL_MAT and OUT_PROD with Q1_0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24721">#24721</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/re…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 09:27

b9698

<details open=""> app : enable self-update only when built with llama-install.sh (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24754">#24754</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">angt@huggingface…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 08:46

b9697

<details open=""> ci : fix check-release message parsing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24751">#24751</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 07:30

b9694

<details open=""> ci : fix Windows x64 (OpenVINO) release link (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24731">#24731</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/rel…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-18 06:44

b9693

<details open=""> metal : check for BF16 support in concat kernel (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24747">#24747</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 21:29

b9692

<details open=""> mtmd: llava_uhd should no longer use batch dim (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24732">#24732</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/r…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 20:56

b9691

<details open=""> ggml-cpu: Conditionally enable power11 backend based on compiler support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24687">#24687</a>) <ul> <li>ggml: Conditionally enable power11 backend based on compiler support…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 20:01

b9690

<details open=""> metal : implement rope_back operator (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24725">#24725</a>) Reuse existing rope kernels with a function constant to toggle forward/backward rotation, avoiding dupli…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 19:33

b9689

<details open=""> metal : add f16 and bf16 support for concat operator (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24724">#24724</a>) <ul> <li>metal : add f16 and bf16 support for concat operator</li> </ul> Extend the Metal back…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 18:56

b9688

<details open=""> server: (router) add model management API (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23976">#23976</a>) <ul> <li> wip </li> <li> server: (router) add SSE realtime updates API </li> <li> nits …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 18:26

b9687

<details open=""> llama : skip main_gpu validation when no devices are available (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23405">#23405</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 17:33

b9686

<details open=""> spec: fix segfault error on long prompts for eagle3 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24707">#24707</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 17:02

b9685

<details open=""> [SYCL] add dev2dev memcpy by SYCL API (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24476">#24476</a>) <ul> <li> add dev2dev memcpy by SYCL API </li> <li> mv GGML_SYCL_DEV2DEV_MEMCPY to runntime table …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 16:28

b9684

<details open=""> [SYCL] Add conv_3d (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24691">#24691</a>) <ul> <li> add conv_3d </li> <li> optimize </li> <li> update ops.md </li> <li> restore test script </li>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 13:34

b9682

<details open=""> vulkan: record actual memory properties during buffer creation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24326">#24326</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 12:19

b9678

<details open=""> opencl: optimize mul_mat_f16_f32_l4 for decode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24504">#24504</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/r…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 11:49

b9677

<details open=""> common: update logging to enforce max_capacity and optimize queue resizing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24490">#24490</a>) <ul> <li> common: update logging to enforce max_capacity and optimize qu…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 11:09

b9675

<details open=""> sycl : Enable to support fp16 by OPs: SQR, SQRT, LOG, SIN, COS, CLAMP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24692">#24692</a>) </details> macOS/iOS: <ul> <li><a href="https://github.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 10:36

b9674

<details open=""> SYCL: fix use-after-free bug with async memcpy in MoE prefill (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24676">#24676</a>) <ul> <li> SYCL: fix a bug with async memcpy </li> <li> make mmid_row_mapping_h…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-17 07:34

b9673

<details open=""> sycl: Add optional USM system allocations (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22526">#22526</a>) This introduces an optional feature to allocate large GPU buffers (≥ 1GB) using USM system allocati…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 18:56

b9672

<details open=""> vendor : update BoringSSL to 0.20260616.0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24693">#24693</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 13:13

b9670

<details open=""> Fix and restrict NVFP4 edge-cases in llama-graph (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24331">#24331</a>) <ul> <li>Move post-GEMM MUL required for dequant b4 lora and bias add</li> </ul> see <a class="iss…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 12:27

b9669

<details open=""> spec: add backend sampling support for eagle3 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24655">#24655</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/re…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 11:50

b9668

<details open=""> vulkan: prefer host-visible memory buffers on UMA devices (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22930">#22930</a>) <ul> <li> implement UMA host-visible memory </li> <li> update based on 0cc4m's sug…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 11:12

b9667

<details open=""> vulkan: Support gated_delta_net with S_v=16 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24581">#24581</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/rele…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 10:13

b9665

<details open=""> bench : add --offline (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24511">#24511</a>) <ul> <li>bench : add --offline</li> </ul> Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">angt@huggingfac…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 08:11

b9664

<details open=""> sycl: support reordered Q4_K/Q5_K/Q6_K MoE MUL_MAT_ID (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24452">#24452</a>) <ul> <li>sycl: support reordered Q4_K and Q5_K MoE MUL_MAT_ID</li> </ul> Extend reordered-wei…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 06:14

b9663

<details open=""> [SYCL] Support OP EXPM1, support all UT cases of FLOOR, TRUNC, ROUND (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24363">#24363</a>) <ul> <li> support OP EXPM1, support all UT cases of FLOOR, TRUNC, ROUND </…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-16 05:08

b9661

<details open=""> vulkan: add col2im_1d op (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24425">#24425</a>) <ul> <li> vulkan: add GGML_OP_COL2IM_1D, follow-up to the CPU op </li> <li> vulkan: col2im_1d bounded gather loop i…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 22:06

b9660

<details open=""> chat : fix LFM2 tool-call parsing double-escaping (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24667">#24667</a>) <ul> <li> Add escape test cases </li> <li> chat : fix LFM2 tool-call parsing double-escapi…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 21:34

b9659

<details open=""> mtmd: fix miscounting n_tokens (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24656">#24656</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 21:01

b9658

<details open=""> chat: include full unparsed prompt in debug (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24650">#24650</a>) message on parse error </details> macOS/iOS: <ul> <li><a href="https://gith…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 20:29

b9656

<details open=""> chat: harden peg-native tool call parsing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24329">#24329</a>) <ul> <li>chat: harden peg-native tool call parsing</li> </ul> accept an optional leading type: function f…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 19:57

b9655

<details open=""> chat: fix an "oldie but goodie" grammar generator bug that surfaced during last changes (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24653">#24653</a>) <ul> <li> chat: fix an "oldie but goodie" grammar generator…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 19:26

b9654

<details open=""> mtmd : add post-decode callback (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24645">#24645</a>) Assisted-by: pi:llama.cpp/Qwen3.6-27B </details> macOS/iOS: <ul> <li><a href="https://g…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 18:51

b9653

<details open=""> vulkan: support more CONCAT types (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24579">#24579</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/downl…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 18:21

b9652

<details open=""> wasm : fix fallback symbol collision (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24639">#24639</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 17:47

b9651

<details open=""> SYCL: use native subgroup size for K-quant DMMV (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/21700">#21700</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 17:19

b9650

<details open=""> sycl: fix soft_max_f32 max reduction (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24451">#24451</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 16:45

b9649

<details open=""> sycl : fix reorder function; add fp32/fp16 in build script (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24578">#24578</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 15:52

b9647

<details open=""> [SYCL] add to support pool_1d, move pool_1d/2d code to pool.cpp/hpp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24584">#24584</a>) <ul> <li> add to support pool_1d, move pool_1d/2d code to pool.cpp/hpp </li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 15:20

b9646

<details open=""> [SYCL]: Remove per-allocation Level Zero runtime checks (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23399">#23399</a>) <ul> <li> [SYCL] Centralize Level Zero detection in ggml_sycl_init </li> <li> use th…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 14:43

b9645

<details open=""> metal : add repeat bf16 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24638">#24638</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9645/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 12:58

b9644

<details open=""> chat: fix whitespace problems once and for all (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24624">#24624</a>) <ul> <li> chat: fix whitespace problems once and for all </li> <li> Purge trailing spaces fro…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-15 09:47

b9642

<details open=""> CUDA: only support F32/F16 for GGML_OP_REPEAT (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24533">#24533</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/re…

llama.cpp — Releases TIER_1 (SO) · yomaytk · 2026-06-15 01:15

b9641

ggml-webgpu: improve i-quants mul_mat performance and speed up prefil…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-14 18:51

b9637

<details open=""> chat: add dedicated Cohere2MoE (North Code) parser (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24615">#24615</a>) <ul> <li> chat: add dedicated Cohere2MoE (North Code) parser </li> <li> Some renames to m…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-14 15:17

b9632

<details open=""> jinja : add count/d/e filter aliases (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24606">#24606</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-14 10:20

b9631

<details open=""> cli : fix not copying preserved tokens (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24258">#24258</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-14 07:32

b9630

<details open=""> Add cohere2moe to llama-vocab for TINY_AYA (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24601">#24601</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-14 02:02

b9628

<details open=""> add sycl to check-release (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24583">#24583</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b962…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 23:27

b9627

<details open=""> ui : fix llama-ui-embed crash when no asset dir is given (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24597">#24597</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 18:20

b9626

<details open=""> Add arch support for cohere2-MoE (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24260">#24260</a>) <ul> <li> Add arch support for cohere2-MoE </li> <li> Removed redundant gating_func checks </li> <li> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 17:08

b9625

<details open=""> jinja : fix negative step slice with start/stop values (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24580">#24580</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/lla…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 16:07

b9624

<details open=""> ui: build-time gzip compression (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24571">#24571</a>) <ul> <li> ui: keep original file name and path </li> <li> fix nocache </li> <li> ui: build-time gzip …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 15:34

b9623

<details open=""> jinja : fix split and replace with empty first arg (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24574">#24574</a>) <ul> <li> fix split and replace with empty first arg </li> <li> fix reserve size </li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 14:18

b9622

<details open=""> vulkan: support non-contig unary/glu ops (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24215">#24215</a>) <ul> <li>vulkan: support non-contig unary/glu ops</li> </ul> Change unary/glu ops to pass in all strides a…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 13:11

b9621

<details open=""> ui: keep original file name and path (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24568">#24568</a>) <ul> <li> ui: keep original file name and path </li> <li> fix nocache </li> </ul> </details> <st…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-13 10:14

b9620

<details open=""> server: clean up static assets handling (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24550">#24550</a>) <ul> <li> server: clean up static assets handling </li> <li> nits </li> <li> simplify file na…

llama.cpp — Releases TIER_1 (SO) · ggerganov · 2026-06-13 05:09

b9619

fit : wrap llama_device_memory_data (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24522">#24522</a>)

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-12 22:21

b9616

<details open=""> ci : unbreak release harder (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24545">#24545</a>) <ul> <li> unbreak release harder </li> <li> missed one </li> <li> remove missing test for now </li> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-12 14:05

b9611

<details open=""> fit : avoid including llama-ext.h in fit.h (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24506">#24506</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-12 13:35

b9610

<details open=""> sync : ggml </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9610/llama-b9610-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI ena…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-11 18:58

b9601

<details open=""> vulkan: ifdef eMesaHoneykrisp (build fix) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24479">#24479</a>) Fixes build/CI after <a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-11 10:10

b9596

<details open=""> server: skip unused log lines on router mode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24463">#24463</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/rel…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-11 08:28

b9594

<details open=""> vocab : refactor normalizer flags into options struct, add strip_accents (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24371">#24371</a>) <ul> <li> vocab : refactor normalizer flags into options struct, add strip…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-10 20:56

b9592

<details open=""> vendor : update LibreSSL to 4.3.2 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24397">#24397</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </details> <st…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-10 20:21

b9591

<details open=""> Remove padding and multiple D2D copies for MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24086">#24086</a>) <ul> <li>Make ggml_gated_delta_net take only the initial recurrent state (D, 1, n_seqs) and passes the …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-10 14:50

b9590

<details open=""> chat: fix LFM2/LFM2.5 ignoring json_schema (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24377">#24377</a>) The LFM2 specialized template handler only built a grammar for tool-calling, silently ignoring jso…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-10 14:14

b9589

<details open=""> CUDA: Fix ssm_scan_f32 data-races (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24360">#24360</a>) <ul> <li>Add missing syncthreads before resuing cub_temp_storage</li> </ul> __syncthreads() is required before be…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-10 08:14

b9587

<details open=""> speculative : fix "ngram-map-k4v" name in logging (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24253">#24253</a>) This is a non-functional change. When using <code>--spec-type ngram-map-k4v</code>, the lo…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 18:16

b9585

<details open=""> graph: Fix granite speech model inference by applying embedding scale when deepstack is not used (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24357">#24357</a>) <ul> <li> llama-graph : apply embedding scale when…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 17:08

b9584

<details open=""> ci : fix windows release (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24369">#24369</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9584…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 13:45

b9581

<details open=""> vulkan: reduce iq1 shared memory usage for mul_mm (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24287">#24287</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 13:09

b9580

<details open=""> vulkan: add <code>v_dot2_f32_f16</code> support in matrix-matrix multiplication and Flash Attention (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24123">#24123</a>) <ul> <li> vulkan: add support for valve fp16 do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 12:42

b9578

<details open=""> mtmd: refactor video subproc handling (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24316">#24316</a>) <ul> <li> mtmd: refactor video subproc handling </li> <li> Update tools/mtmd/mtmd-helper.cpp </li>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 12:11

b9577

<details open=""> server: log prompts to directory (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22031">#22031</a>) <ul> <li>server: log prompts to directory</li> </ul> Add <code>--log-prompts-dir</code> to write each prompt to a …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 11:43

b9575

<details open=""> ggml : add GGML_OP_COL2IM_1D (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24206">#24206</a>) <ul> <li>cpu: add GGML_OP_COL2IM_1D</li> </ul> Add the overlap-add (scatter-add) step of a 1D transposed convolution.<…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 09:11

b9574

<details open=""> server : do not clear slots without unified KV cache (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24190">#24190</a>) <ul> <li>Always export idle slots to RAM</li> </ul> Without this, a slot's VRAM cache may not …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 08:41

b9573

<details open=""> models : fix plamo2 attention_key/value_length regression (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24317">#24317</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 08:08

b9572

<details open=""> ggml-cpu : fix rms_norm_back wrong output under in-place aliasing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24305">#24305</a>) <ul> <li> ggml-cpu : fix rms_norm_back wrong output under in-place aliasing <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 06:07

b9571

<details open=""> Remove case for GGML_TYPE_Q4_K in mvvq.cu (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23528">#23528</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-09 04:18

b9570

<details open=""> ggml-webgpu: Add clang-format job (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24308">#24308</a>) <ul> <li> Add clang-format job </li> <li> try local formatting </li> </ul> </details> macOS…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 21:12

b9568

<details open=""> mtp: support for gemma-4 E2B and E4B assistants (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24282">#24282</a>) <ul> <li> models: update converter to support smaller assistants </li> <li> models: add mask…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 19:04

b9567

<details open=""> server : do not parse when flushing http headers (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24281">#24281</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 18:09

b9566

<details open=""> graph: guard iswa kq_mask on its own buffer (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24294">#24294</a>) A SWA-only draft head (e.g. StepFun MTP) leaves the base sub-cache empty, so its kq_mask buffer s…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 17:37

b9565

<details open=""> [ggml-webgpu] Handle buffer overlap / buffer aliasing for concat operator (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24000">#24000</a>) <ul> <li> Only run webgpu CI on my fork </li> <li> Add webgpu only…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 17:10

b9564

<details open=""> [ggml-webgpu] Implement 2D workgroups for scale, binary, and unary ops (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24044">#24044</a>) <ul> <li> Only run webgpu CI on my fork </li> <li> Add webgpu only wo…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 16:42

b9562

<details open=""> mtmd : add video input support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24269">#24269</a>) <ul> <li> wip </li> <li> ok: lazy bitmap API </li> <li> remember to free lazy text </li> <li> w…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 16:16

b9561

<details open=""> sync : ggml </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9561/llama-b9561-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI ena…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 15:26

b9559

<details open=""> cli: fix spinner not show during prompt processing (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24283">#24283</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.c…

llama.cpp — Releases TIER_1 (SO) · ngxson · 2026-06-08 14:59

b9563

docker: install ffmpeg in the released image (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24302">#24302</a>)

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 14:46

b9558

<details open=""> vulkan: Use cm2 decode_vector for mul_mat_id B matrix loads (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23991">#23991</a>) This allows vec4 loads of the B elements. Also increase BK to 64 when this is ena…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 14:19

b9557

<details open=""> cuda: reset cuda context after reading memory size (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23935">#23935</a>) <ul> <li> cuda: reset device in get_memory function if no backend is active </li> <li> al…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 13:50

b9556

<details open=""> HIP: add gfx1152 and gfx1153 to RDNA3.5 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24129">#24129</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-08 11:10

b9555

<details open=""> metal : fix im2col 1D case (audio models) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24220">#24220</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 21:11

b9553

<details open=""> common : relax sampler name matching (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23744">#23744</a>) <ul> <li>common : relax sampler name matching</li> </ul> Currently, in some cases, the alternative names for s…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 19:28

b9551

<details open=""> kv-cache : avoid kv cells copies (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24277">#24277</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/downlo…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 15:57

b9550

<details open=""> kv-cache: follow the source cache size when sharing cells (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24267">#24267</a>) A fitted target context can end up smaller than the draft default, the oversized as…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 13:40

b9549

<details open=""> llama : add Gemma4 MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23398">#23398</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9549/l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 12:11

b9548

<details open=""> spec : fix vocab compatibility check (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24256">#24256</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-07 09:40

b9547

<details open=""> arg: Skip mmproj download when user supplied mmproj (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24239">#24239</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 21:21

b9544

<details open=""> common/chat : fix LFM2/LFM2.5 reasoning round-trip and leak (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24234">#24234</a>) <ul> <li>common/chat : fix LFM2 reasoning round-trip and stray leak</li> <li>Gate by reaso…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 19:50

b9543

<details open=""> mtmd: support "frame merge" for qwen-vl-based models (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/21858">#21858</a>) <ul> <li> feat: add video support for Qwen3.5 </li> <li> various clean up </li> <li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 10:48

b9542

<details open=""> completion : remove useless statics (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24226">#24226</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </details> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 09:56

b9541

<details open=""> completion : fix format specifier in LOG_INF (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24213">#24213</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </deta…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 04:58

b9538

<details open=""> model : rename local n_layer_all variable (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24209">#24209</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-06 04:35

b9537

<details open=""> context : fix off-by-one comparisons to n_gpu_layers (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24208">#24208</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 21:10

b9536

<details open=""> opencl: improve get_rows, cpy, concat and q6_k flat gemv (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24160">#24160</a>) <ul> <li> opencl: allow multiple workgroups for large rows </li> <li> opencl: impro…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 20:07

b9535

<details open=""> common/chat : unify and fix LFM2/LFM2.5 tool parser (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24178">#24178</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 18:17

b9534

<details open=""> vulkan: add fwht support for Intel with shmem reduction (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23964">#23964</a>) <ul> <li> vulkan: add fwht support for Intel with shmem reduction </li> <li> don't u…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 17:29

b9533

<details open=""> model: fix build failed (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24193">#24193</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9533/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 16:47

b9531

<details open=""> TP: round up granularity to 128 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24180">#24180</a>) <ul> <li> TP: round up granularity to 128 </li> <li> remove assert </li> </ul> </details> mac…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 16:09

b9530

<details open=""> cli: fix model params not propagated (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23893">#23893</a>) Fixes <a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/issues/23847">#23847</a>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 14:43

b9529

<details open=""> model : fix llama_model::n_gpu_layers() (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24188">#24188</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 13:19

b9528

<details open=""> ui: run npm install when package-lock.json is newer than node_modules (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24171">#24171</a>) </details> macOS/iOS: <ul> <li><a href="https://github.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 09:19

b9524

<details open=""> minor : fix lint issues (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24165">#24165</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9524/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 08:45

b9523

<details open=""> hparams : refactor <code>hparams.n_layer</code> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24060">#24060</a>) <ul> <li> hparams : refactor hparams.n_layer </li> <li> cont : remove <code>n_layer_kv()</co…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 07:45

b9522

<details open=""> kleidiai : dynamic chunck-based scheduling for hybrid execution (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23819">#23819</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggm…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 06:56

b9521

<details open=""> CUDA: enroll mul_mat_vec_q_moe into pdl (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24087">#24087</a>) <ul> <li>Enroll mul_mat_vec_q_moe into PDL, boosting MTP performance on BW</li> </ul> Data collected on a B…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-05 05:46

b9519

<details open=""> sycl : port multi-column MMVQ from CUDA backend (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/21845">#21845</a>) mmvq: Port the ncols_dst optimization from ggml-cuda/mmvq.cu to SYCL. Read weights onc…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 17:15

b9518

<details open=""> server : disable on-device spec checkpoints (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24108">#24108</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/rele…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 16:20

b9515

<details open=""> Move duplicated imatrix code into single common imatrix-loader.cpp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22445">#22445</a>) <ul> <li> Deduplicate imatrix loading code </li> <li> Add back LLAMA_TRAC…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 15:15

b9512

<details open=""> return filter to save memory (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24125">#24125</a>) Co-authored-by: lvyichen <a href="mailto:[email protected]">[email protected]</a> </details> macO…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 14:46

b9510

<details open=""> ggml: vectorize ggml_vec_dot_q4_1_q8_1 with WASM SIMD128 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22209">#22209</a>) <ul> <li>ggml: vectorize ggml_vec_dot_q4_1_q8_1 with WASM SIMD128</li> </ul> Optimize the …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 13:55

b9509

<details open=""> server: avoid unnecessary checkpoint restore when new tokens are present (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24110">#24110</a>) <ul> <li>server: avoid unnecessary checkpoint restore when new tokens are pre…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 11:33

b9505

<details open=""> server : add header to tools/server/server-http.h (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24089">#24089</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 10:59

b9504

<details open=""> cmake: skip cvector-generator and export-lora when CPU backend is disabled (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24053">#24053</a>) </details> macOS/iOS: <ul> <li><a href="https://git…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 10:14

b9503

<details open=""> fix(mtmd): handle Gemma 4 audio projector embedding size (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24091">#24091</a>) <ul> <li> mtmd: handle Gemma 4 audio projector embedding size </li> <li> rm project…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 06:09

b9499

<details open=""> ggml-webgpu: FlashAttention refactor + standardize quantization support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23834">#23834</a>) <ul> <li> Start work on flash_attn refactor </li> <li> Refactor …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-04 05:43

b9498

<details open=""> ggml-cpu: extend RVV quantization vec dot to higher VLENs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22754">#22754</a>) <ul> <li> ggml-cpu: add rvv 512b,1024b impls for iq4_xs </li> <li> ggml-cpu: refac…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 20:25

b9496

<details open=""> mtmd: fix Gemma 4 unified FPE (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24088">#24088</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 18:14

b9495

<details open=""> qwen35: use post-norm hidden state for MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24025">#24025</a>) <ul> <li> qwen35: use post-norm hidden state for MTP </li> <li> rename pre_norm to nextn </li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 17:37

b9494

<details open=""> mtmd: enable non-causal vision for gemma 4 unified (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24082">#24082</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 15:45

b9493

<details open=""> mtmd, model: allow skip build_vit() (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24077">#24077</a>) <ul> <li> add model </li> <li> nits </li> </ul> </details> macOS/iOS: <ul> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 14:17

b9491

<details open=""> Avoid PDL race conditions by disabling restrict when PDL is used (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24030">#24030</a>) <ul> <li> Removes restrict from PDL kernel heade…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 11:48

b9490

<details open=""> ggml-cpu: use runtime SVE width in FWHT (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24059">#24059</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 11:23

b9489

<details open=""> cuda: reserve space for quantize kv-cache at startup (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23907">#23907</a>) <ul> <li> cuda: reserve space for quantize kv-cache at startup </li> <li> address revie…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 07:49

b9488

<details open=""> tests : add support for qwen3 SSM archs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24031">#24031</a>) <ul> <li> tests : add support for qwen3 SSM archs </li> <li> arch : add LLM_KV_ATTENTION_RECURRENT_L…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 06:26

b9487

<details open=""> update BoringSSL to 0.20260526.0 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23794">#23794</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/downlo…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 06:03

b9486

<details open=""> ci : disable ccache for msvc windows release jobs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23911">#23911</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-03 05:38

b9485

<details open=""> arg : removed unecesary mmproj download when users pass --no-mmproj (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23425">#23425</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 22:02

b9484

<details open=""> opencl: use flat variants of q4_K and q6_K gemv for very large M (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24006">#24006</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/gg…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 21:36

b9483

<details open=""> hexagon: profiler output fix and script updates (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24042">#24042</a>) <ul> <li> hex-ops: fix profiler output (ie remove the redundant NONEs) </li> <li> hex-prof: …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 19:44

b9482

<details open=""> model: add Mellum architecture (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23966">#23966</a>) <ul> <li> model: support for Mellum architecture </li> <li> model: improve mellum.py formatting </li> <li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 16:58

b9481

<details open=""> model : support granite multilingual embeddings R2 (ibm-granite/granite-embedding-{97,311}m-multilingual-r2) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22716">#22716</a>) <ul> <li> Add support for the ibm-gran…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 16:25

b9480

<details open=""> StepFun 3.5 MTP (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23274">#23274</a>) <ul> <li> StepFun 3.5 MTP </li> <li> Simplify to single layer </li> <li> Rollback core changes </li> <li> fix …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 14:22

b9479

<details open=""> common : fix state save in common_prompt_batch_decode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23468">#23468</a>) <ul> <li>common : fix state save in common_prompt_batch_decode</li> </ul> This commit address…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 13:03

b9478

<details open=""> server: add SSE ping interval (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24013">#24013</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 11:23

b9474

<details open=""> ui: Add Thinking mode toggle with reasoning effort levels + improvements for Chat Form Add Action UI (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23434">#23434</a>) <ul> <li> feat: Add "Thinking" toggle and stat…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 10:52

b9473

<details open=""> kv-cache : SWA checkpoints store only non-masked cells (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23981">#23981</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/lla…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 10:22

b9471

<details open=""> llama : deprecate <code>llama_set_warmup</code> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24009">#24009</a>) <ul> <li> llama : deprecate <code>llama_set_warmup</code> </li> <li> cont : fix type </l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 09:37

b9470

<details open=""> hexagon: MUL_MAT, MUL_MAT_ID, FLASH_ATTN and GDN cleanup and optimizations for latest models (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23989">#23989</a>) <ul> <li> hex-mm: initial support for F32 * F32 -> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 07:18

b9469

<details open=""> hexagon: add gelu_quick (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/24007">#24007</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9469/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 05:54

b9468

<details open=""> server: real-time reasoning interruption via control endpoint (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23971">#23971</a>) <ul> <li>server: real-time reasoning interruption via control endpoint</li> </ul> Bui…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 03:31

b9467

<details open=""> clean up unused variables warnings (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23975">#23975</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/down…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-02 02:44

b9466

<details open=""> opencl: fix compiler warnings for non-adreno path (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23922">#23922</a>) <ul> <li> opencl: fix compiler warnings for non-adreno path </li> <li> opencl: fix const c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 19:58

b9464

<details open=""> speculative : fix n_outputs_max and remove draft-simple auto-enable (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23988">#23988</a>) <ul> <li>speculative : add common_speculative_n_max helper function</li> </ul> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 19:25

b9460

<details open=""> llama: limit max outputs of <code>llama_context</code> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23861">#23861</a>) <ul> <li> llama: save more VRAM by reserving n_outputs == n_seqs when possible </li> <li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 19:00

b9459

<details open=""> metal: template GLU kernels to support f16/f32 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23882">#23882</a>) Drops the hardcoded f32 GLU kernels in favor of a single template. We now load/store in the native t…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 18:32

b9458

<details open=""> vulkan: don't hold the device mutex while compiling pipelines (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23641">#23641</a>) <ul> <li>vulkan: don't hold the device mutex while compiling pipelines</li> </ul> We …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 18:08

b9457

<details open=""> vulkan: reduce host memory lock contention (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23376">#23376</a>) <ul> <li> vulkan: reduces lock contention </li> <li> replace unique_lock with lock_guard </li…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 17:22

b9455

<details open=""> TP: quantized KV cache support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23792">#23792</a>) <ul> <li> TP: quantized KV cache support </li> <li> fix partial view </li> <li> remove overly strict a…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 14:57

b9453

<details open=""> model: Add EXAONE 4.5 implementations (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/21733">#21733</a>) <ul> <li> Add EXAONE 4.5 and Add GQA for MMproj </li> <li> mtmd: EXAONE 4.5 vision markers and project…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 14:22

b9452

<details open=""> vulkan: Block-load Q3_K/Q6_K block data and subtract on 32b ints (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23056">#23056</a>) Q2_K/Q3_K/Q6_K do much better when using MMVQ on Intel BMG even though they'…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-06-01 13:15

b9451

<details open=""> vulkan: Removed unused functions (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23175">#23175</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/downlo…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-31 21:53

b9444

<details open=""> server : handle If-None-Match weak ETags (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23916">#23916</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/release…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-31 11:08

b9442

<details open=""> vocab : add tokenizer support for jina-embeddings-v2-base-zh (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/18756">#18756</a>) <ul> <li> vocab : add jina-embeddings-v2-base-zh (whitespace tokenizer) </li> <li>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-31 09:50

b9441

<details open=""> ui: fix ETag truncation with MSVC compiler (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23917">#23917</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-31 06:58

b9439

<details open=""> llama: only use one iGPU device by default (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23897">#23897</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/relea…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 20:57

b9437

<details open=""> Support <code>-fa auto</code> in llama-bench (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23714">#23714</a>) <ul> <li>Support <code>-fa auto</code> in llama-bench</li> </ul> Make the default value of <code>-ngl<…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 17:45

b9436

<details open=""> opencl: support bf16 by converting to f16 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23839">#23839</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releas…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 14:27

b9434

<details open=""> TP: fix granularity for Qwen 3.5/3.6 + 3 GPUs (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23843">#23843</a>) <ul> <li> TP: fix granularity for Qwen 3.5/3.6 + 3 GPUs </li> <li> fix afmoe TP </li> </ul…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 13:40

b9433

<details open=""> metal : restore im2col implementation for large kernels (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23901">#23901</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/ll…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 11:14

b9432

<details open=""> test: (test-llama-archs) log the config name first (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23885">#23885</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 10:47

b9431

<details open=""> ci : update ios-xcode release job to macos-26 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23906">#23906</a>) <ul> <li> ci : disable libcommon build from xcframework </li> <li> ocd : fix name </li> <l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 10:13

b9430

<details open=""> ggml : add some lsx support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23798">#23798</a>) <ul> <li>loongarch : optimize LSX fp16 load/store with native intrinsics</li> </ul> Use __lsx_vfcvtl_s_h and __lsx_vfcv…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-30 08:52

b9428

<details open=""> ci : fix s390x release job (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23898">#23898</a>) <ul> <li> ci : fix s390x release job </li> <li> ci : multi-thread build for <code>ios-xcode</code> </li> <li>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 23:38

b9415

<details open=""> download: add option to skip_download (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23059">#23059</a>) <ul> <li> download: add option to skip_download </li> <li> fix </li> <li> fix 2 </li> <li> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 21:03

b9414

<details open=""> mtmd: Add DeepSeekOCR 2 Support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/20975">#20975</a>) <ul> <li> mtmd: DeepSeek-OCR 2 support, with multi-tile dynamic resolution </li> <li> introduced clip_image_…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 18:43

b9413

<details open=""> CUDA: Check PTX version on host side to guard PDL dispatch (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23530">#23530</a>) <ul> <li>CUDA: Check PTX version on host side to guard PDL dispatch</li> </ul> Checking …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 16:22

b9412

<details open=""> server: bump timeout to 3600s (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23842">#23842</a>) <ul> <li> server: bump timeout to 3600s </li> <li> nits: change wording </li> </ul> </details> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 15:32

b9411

<details open=""> model : support for DeepseekV32ForCausalLM with generic DeepSeek Sparse Attention (DSA) implementation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23346">#23346</a>) <ul> <li> llama : support DeepSeek V3.2 mode…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 14:42

b9410

<details open=""> llama: use f16 mask for FA to save VRAM (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23764">#23764</a>) <ul> <li> llama: use f16 mask for FA </li> <li> review: add llama_cast + formatting </li> <li> <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 13:49

b9409

<details open=""> sync : ggml </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9409/llama-b9409-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI ena…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 13:00

b9406

<details open=""> llama: add llm_graph_input_mtp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23643">#23643</a>) <ul> <li> llama: add llm_graph_input_mtp </li> <li> rename input_mtp -> input_token_embd </li> <li> <p…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 12:10

b9405

<details open=""> app : move licences to llama-app (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23824">#23824</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </details> <str…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 10:27

b9403

<details open=""> meta : Add missing <code>buffer</code> set in allreduce fallback !COMPUTE clear (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23480">#23480</a>) Without this at least the vulkan backend will skip the <code>* 0</c…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 08:47

b9402

<details open=""> hexagon: basic/generic op fusion support and RMS_NORM+MUL fusion (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23835">#23835</a>) Updating infra to enable op fusion and using RMS_NORM+MUL as the use-case. </d…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 06:53

b9401

<details open=""> mtmd-debug: add color and rainbow mode (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23829">#23829</a>) <ul> <li> mtmd-debug: add color and rainbow mode </li> <li> fix M_PI </li> <li> max_dist <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 05:55

b9400

<details open=""> mtmd: fix gemma 4 projector pre_norm (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23822">#23822</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 05:18

b9399

<details open=""> opencl: move backend info printing into its own function (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23702">#23702</a>) <ul> <li> opencl: move backend info print into its own function </li> <li> opencl: …

llama.cpp — Releases TIER_1 (SO) · aendk · 2026-05-29 04:46

b9404

cuda : disables launch_fattn PDL enrollment due to compiler bug (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23825">#23825</a>)

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 04:44

b9395

<details open=""> app : improve help output (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23805">#23805</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </details> mac…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 04:10

b9394

<details open=""> mtmd: n_head_kv defaults to n_head (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23782">#23782</a>) removed AI-generated comment </details> macOS/iOS: <ul> <li><a href="https://github.…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 03:29

b9393

<details open=""> mtmd: fix gemma 4 audio rms norm eps (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23815">#23815</a>) <ul> <li> mtmd: fix gemma 4 audio rms norm eps </li> <li> Update tools/mtmd/clip.cpp </li> </ul> <p…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 02:53

b9391

<details open=""> arg: Add LLAMA_ARG_API_KEY_FILE environment variable for --api-key-file (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23167">#23167</a>) </details> macOS/iOS: <ul> <li><a href="https://github…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 02:18

b9389

<details open=""> ggml: auto apply iGPU flag CUDA/HIP if integrated device (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23007">#23007</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 01:45

b9388

<details open=""> mmvq Optim: add MMVQ_PARAMETERS_TURING(mmvq_parameter_table_id) for … (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23729">#23729</a>) <ul> <li> mmvq Optim: add MMVQ_PARAMETERS_TURING(mmvq_parameter_table_id) for…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 01:06

b9387

<details open=""> CUDA: route batch>=4 quantized matmul to MMQ on AMD MFMA hardware (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23227">#23227</a>) <ul> <li>CUDA: per-quant MMVQ/MMQ batch threshold on AMD MFMA hardware</li> </ul>…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-29 00:30

b9386

<details open=""> server: minor tweaks to use more cpp features (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23785">#23785</a>) <ul> <li> misc(server): add default port to impl RAII </li> <li> misc(server): register_gcp_co…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 20:30

b9384

<details open=""> vulkan: fast path for walsh-hadamard transform (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23687">#23687</a>) <ul> <li> vulkan: fast path for walsh-hadamard transform </li> <li> disable for intel due to …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 19:58

b9383

<details open=""> chat : add Granite 4.1 chat template (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23518">#23518</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 18:58

b9382

<details open=""> vulkan: fix wrong index variable in inner loop (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23665">#23665</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/r…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 18:19

b9381

<details open=""> vulkan: Fix memory logger unsafe iterator access (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23667">#23667</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 17:44

b9380

<details open=""> server, ui : Add support for HTTP ETags in llama-server (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23701">#23701</a>) <ul> <li> allow caching of ui elements in llama-server </li> <li> use fnv_hash <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 17:05

b9378

<details open=""> cuda : fix KQ mask offset integer overflow in fattn MMA kernel (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23610">#23610</a>) Co-authored-by: Stanisław Szymczyk <a href="mailto:[email protected]">sszymczy@gmai…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 16:24

b9377

<details open=""> perplexity : fix format specifier in LOG_ERR (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23788">#23788</a>) Signed-off-by: Adrien Gallouët <a href="mailto:[email protected]">[email protected]</a> </deta…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 15:53

b9375

<details open=""> ggml: fixed Arm SVE usage bug in vec.h, vec.cpp (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22841">#22841</a>) <ul> <li>Updated vec.h/vec.cpp code to accumulate to F32 rather than F16</li> </ul> Change-Id: I0cb…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-28 12:52

b9374

<details open=""> ci : refactor (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23789">#23789</a>) <ul> <li> ci : separate CUDA windows workflow + fix names </li> <li> ci : rename workflow </li> <li> ci : prefix cache …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 23:47

b9371

<details open=""> ggml-webgpu: remove legacy constants (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23672">#23672</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/do…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 18:25

b9370

<details open=""> hexagon: add support for Q4_1 in MUL_MAT and MUL_MAT_ID (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23647">#23647</a>) <ul> <li> hex-mm: add support for Q4_1 matmul/matvec, hvx-only for now </li> <li> hm…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 18:13

b9368

<details open=""> vulkan: Switch MUL_MAT_VEC to 4 K per iteration for F16/32 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22887">#22887</a>) <ul> <li>vulkan: Switch MUL_MAT_VEC to 4 K per iteration for F16/32</li> </ul> Against m…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 17:42

b9369

<details open=""> ggml-webgpu: Fix how to dispatch WG to some ops (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23750">#23750</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 17:37

b9367

<details open=""> vulkan: use GL_NV_cooperative_matrix_decode_vector for faster matmul (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23541">#23541</a>) </details> macOS/iOS: <ul> <li><a href="https://github.co…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 17:34

b9366

<details open=""> vulkan: add REPEAT op support for f16 to f16. (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23298">#23298</a>) <ul> <li> feat: extend repeat op for vulkan </li> <li> feat: add repeat_f16 vulkan pipeline</p…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 17:06

b9365

<details open=""> ci : move ARM jobs to self-hosted + disable kleidiai mac release (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23780">#23780</a>) <ul> <li> ci : move ARM jobs to 3rd-party runners + disable kleidiai release <…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 13:15

b9360

<details open=""> common : fix env names to all have LLAMA_ARG_ prefix (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23778">#23778</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 10:19

b9357

<details open=""> vulkan: avoid preferring transfer queue on AMD UMA devices (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22455">#22455</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 06:02

b9354

<details open=""> convert: add MiniCPM5 tokenizer support (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23384">#23384</a>) Add minicpm5 pre-tokenizer hash via convert_hf_to_gguf_update.py and implement hardcoded regex handli…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-27 05:50

b9353

<details open=""> server : fix the log message when using SSL (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23393">#23393</a>) When llama-server is started with SSL key and cert, the log says that it listens on http instead …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 23:49

b9352

<details open=""> ggml-zendnn : fixed naming of matmul function (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/20964">#20964</a>) <ul> <li> ggml-zendnn: fixed naming of matmul function </li> <li> ggml-zendnn: fixed naming of…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 19:56

b9351

<details open=""> </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9351/llama-b9351-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/dow…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 09:02

b9334

<details open=""> CUDA: missing PDL sync for FWHT, better fallback (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23690">#23690</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 06:14

b9333

<details open=""> metal : add apple device id (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23566">#23566</a>) Co-authored-by: lvyichen <a href="mailto:[email protected]">[email protected]</a> </details> macOS…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 05:11

b9331

<details open=""> ci : reduce PR jobs by matching backend paths (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23675">#23675</a>) <ul> <li> ci : disable SYCL f16 builds </li> <li> ci : extract android and hip into separate w…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 02:49

b9330

<details open=""> model: tag ffn_latent as MUL_MAT to fix buft probe (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23664">#23664</a>) ffn_latent_down/up are declared GGML_OP_MUL in LLM_TENSOR_INFOS but nemotron-h feeds them …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 02:02

b9329

<details open=""> CUDA: add fast walsh-hadamard transform (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23615">#23615</a>) <ul> <li> CUDA: add fast walsh-hadamard transform </li> <li> review: add unrolls + change size_t -&g…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 00:33

b9326

<details open=""> sync : ggml </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b9326/llama-b9326-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li><a href="https://github.com/ggml-org/lla…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-26 00:00

b9320

<details open=""> TP: fix ggml context size calculation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22616">#22616</a>) <ul> <li> TP: fix ggml context size calculation, memory leak </li> <li> move split state cache back in…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 23:09

b9319

<details open=""> ggml: <code>gguf_init_from_callback</code> and <code>gguf_init_from_buffer</code> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22341">#22341</a>) <ul> <li> ggml: implement <code>gguf_init_from_buffer</code> …

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 20:27

b9318

<details open=""> server: MTP layer kv-cache should respect draft type ctk (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23646">#23646</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/l…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 18:40

b9315

<details open=""> llama : document that only one on-device state can be saved per sequence (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23520">#23520</a>) </details> macOS/iOS: <ul> <li><a href="https://githu…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 17:14

b9313

<details open=""> ggml : Parallelize quant LUT init (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23595">#23595</a>) <ul> <li>Use OpenMP to parallelize iq2xs_init_impl and iq3xs_init_impl.</li> <li>Move the OpenMP detection from ggml…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 15:57

b9311

<details open=""> vendor : update cpp-httplib to 0.45.1 (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23639">#23639</a>) </details> macOS/iOS: <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/d…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-25 11:52

b9310

<details open=""> server: fix checkpoints creation (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22929">#22929</a>) <ul> <li> common : add common_chat_split_by_role </li> <li> cont : fix spans to reach end of message </…

llama.cpp — Releases TIER_1 (SO) · github-actions[bot] · 2026-05-24 11:33

b9305

<details open=""> cmake : fix ui build (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/23592">#23592</a>) <ul> <li> cmake/ui : add -fPIC to llama-ui static lib </li> <li> cmake : rename host compiled embed helper </li> </…

llama.cpp — Releases TIER_1 (SO) · njsyw1997 · 2026-05-24 02:56

b9301

hexagon: apply repl optimization in flash attn softmax as <a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/22993">#22993</a> (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/issues/23">#23</a>…

COVERAGE [309]