github

<details open> server: harden the file_glob_search directory walk (#26626) * server: don't walk Windows junctions in file_glob_search std::filesystem reports a junction as a plain directory, so the symlink guard misses it and a junction pointing back at an ancestor is walked u

<details open> tests: re-enable MiniMax M3 in `test-llama-archs` (#26633) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10288/llama-b10288-bin-macos-arm64.tar.gz) - macOS A

<details open> mtmd: Unlimited-OCR fix max_tiles, setting in converter (#25614) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10287/llama-b10287-bin-macos-arm64.tar.gz) - m

<details open> grammar : degrade max repetition >= 2000 to unbounded (#26613) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10286/llama-b10286-bin-macos-arm64.tar.gz) - mac

<details open> mtmd: support multi-row batching for deepseek-ocr (#26154) * mtmd: support multi-row batching for deepseek-ocr * mtmd: weave deepseek-ocr rows in one shot instead of per row (#26615) --------- Co-authored-by: Saba Fallah <sabafallah@gmail.com> </details> **W

<details open> fit: Fix memory allocation for MTP layers (#26605) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10284/llama-b10284-bin-macos-arm64.tar.gz) - macOS Apple Sil

<details open> server: Adding spec-decode counters to /metrics endpoint (#26389) * * server: add spec-decode counters to /metrics endpoint * server: fixed review comments and now aligned param names exactly with vLLM. </details> **Website:** - <https://llama.app> **macOS/iO

<details open> vendor : apply patches for subprocess.h (#26606) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10280/llama-b10280-bin-macos-arm64.tar.gz) - macOS Apple Silic

<details open> Prefer npm ci over install for security (#26601) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10276/llama-b10276-bin-macos-arm64.tar.gz) - macOS Apple Silic

<details open> server: decode Windows OEM output to UTF-8 in built-in tools (#26597) a child process writes in the OEM code page, which is not UTF-8 on a western Windows install, so accented output reaches the JSON layer as invalid bytes and gets replaced there, silently losing

<details open> mtmd: correcting duplicate empty audio chunks for short inputs (#26536) * correcting duplicate empty audio chunks for short inputs * tests.sh code restored </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://gi

<details open> sampler : remove "full-context windows" from history-based samplers (#26524) * Resolve -1 to 1024 instead of ctx-len for samplers Because of backend-sampling we initialize samplers before the complete llama_context is there. Therefore, we cannot infer the resolv

<details open> ui: CWD for agent (#26518) * server : extend file_glob_search for UI pickers * ui : add per-conversation working directory with picker * ui : add path navigation and search scope to cwd picker Treat path-like queries (starting with / or ~) as directory navigat

<details open> mtmd: support Qwen3-TTS (note: breaking change to llama-tts binary) (#26254) * convert text model * main model load ok * convert encoder ok * speaker encoder loading ok * speaker enc graph * adapt vocab for backbone (with some tricks) * add suppress_tokens

<details open> models : fix dflash wo_a reshape on load (#26577) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10269/llama-b10269-bin-macos-arm64.tar.gz) - macOS Apple Sili

<details open> ci: fix pre-built binaries no longer working on macOS 15 and below (#26375) * ci: fix pre-built binaries no longer working on macOS 15 and below * ci: add macOS deployment target to disabled KleidiAI build </details> **Website:** - <https://llama.app> **macOS

<details open> speculative : refactor enabled configs common_speculative_init (#26510) This commit contains a suggestion to reduce some code duplication in common_speculative_init when adding the enabled speculative decoding configurations. No tests were added but the existing

<details open> sync : ggml </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10265/llama-b10265-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLE

<details open> sycl: parallelize the non-contiguous concat kernel (#25852) * sycl: parallelize the non-contiguous concat kernel Launch geometry only: the non-contiguous concat kernel launched a single-lane work-group (1, 1, 1), now it will launch a (1, 1, SYCL_CONCAT_BLOCK_SIZ

<details open> Extended SYCL oneDNN SDPA to non-FP16 KV caches (Q4_0–Q8_0 and FP32) (#25874) * sycl: extend oneDNN SDPA to Q4_0-Q8_0 and F32 KV caches Extends the oneDNN SDPA path (PR #25222) to handle non-F16 KV caches by dequantizing or converting K/V to dense FP16 on-device

<details open> chat : add new template for DeepSeek V4 Flash 0731 (#26398) * common/chat: update DeepSeek V4 templates Align the DeepSeek V4 templates with the official encoders while keeping parser behavior out of this change. - Default drop_thinking for DeepSeek V4 history

<details open> vendor : update cpp-httplib to 0.52.0 (#26485) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10253/llama-b10253-bin-macos-arm64.tar.gz) - macOS Apple Silicon

<details open> vendor : update BoringSSL to 0.20260803.0 (#26523) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10252/llama-b10252-bin-macos-arm64.tar.gz) - macOS Apple Sil

<details open> model : support MTP in GLM-4.7-Flash (#24868) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10251/llama-b10251-bin-macos-arm64.tar.gz) - macOS Apple Silicon

<details open> tests: add model resolution test on synthetic repo listings (#26172) * tests: add model resolution test on synthetic repo listings Include download.cpp and arg.cpp inside a namespace with hf_cache monkey patched to serve hardcoded listings, so the resolution and

<details open> server: add get_info tool (#26522) * server: add get_info tool * fix --rpc in docs * server: harden get_info probe result handling Report the OS as unknown when the probe process fails to spawn or times out, so the diagnostic text from run() is never returned

<details open> vocab : validate default special token ids (#26506) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10248/llama-b10248-bin-macos-arm64.tar.gz) - macOS Apple Si

<details open> ggml: use dynamic allocation for split graph inputs (#22789) * ggml: use dynamic allocation for split graph inputs Replace fixed-size GGML_SCHED_MAX_SPLIT_INPUTS arrays with dynamically allocated buffers in the backend scheduler. This fixes crashes when loading

<details open> model: MTP support for Qwen3-Next (#25589) * mtp for qwen3nex * fix for python type-check * Fix to compute num_mtp from directly mtp layer * define opt_num_mtp_layers in _QwenMtpMixin and fix some comments * Fix for python type check * Update gguf-py/gguf/co

<details open> llama : MTP support for DeepSeek V3.2 (#26457) * llama : MTP support for DeepSeek V3.2 * model : no need to include MTP layers during DeepSeek V3.2 model type discovery --------- Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com> </details> **Website:**

<details open> metal: implement DSv4 Lightning Indexer (#25893) * metal: implement F16 Lightning Indexer - Implement GGML_OP_LIGHTNING_INDEXER for 128-dimensional, 64-head inputs with F32 queries and weights plus F16 keys and masks. - Add tiled and tail kernels and test KV l

<details open> metal : add SILU_BACK (#25982) * feat(silu_back): implemented silu_back op for f32 * fix(silu_back): removed redundant asserts in ggml-metal-ops.cpp function ggml_metal_op_silu_back. </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple

<details open> metal : add F16 support for bin ops (#26465) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10234/llama-b10234-bin-macos-arm64.tar.gz) - macOS Apple Silicon (

<details open> opencl: limit local workgroup size for GLU operation (#26383) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10233/llama-b10233-bin-macos-arm64.tar.gz) - macO

<details open> metal: implement DeepSeek V4 hyper-connections (#26459) - Implement GGML_OP_DSV4_HC_COMB, GGML_OP_DSV4_HC_PRE, and GGML_OP_DSV4_HC_POST with SIMDgroup register and shuffle optimized kernels. - Add Metal dispatch and support plumbing and test the production Sink

<details open> common: support the DSpark sidecar resolution (#26458) The dspark- files resolve like the other speculative sidecars: the -hfd tag applies to them, a requested sidecar resolves without a full model at the tag, and an explicit -md selection disables the discovery.

<details open> opencl: bugfix increment ref_count in ggml_backend_opencl_init() (#26162) Incrementing `ref_count` at the beginning is important later in the `free()` method of the `ggml_backend_opencl_context` at program end. If we do not increment the `ref_count`, the result

<details open> DeepseekV4 MTP + DSpark (#25784) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10228/llama-b10228-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, Kleid

<details open> ggml-webgpu: add support for f16 repeat (#26307) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10224/llama-b10224-bin-macos-arm64.tar.gz) - macOS Apple Silic

<details open> test: fix some CI errors (#26415) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10223/llama-b10223-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, Klei

<details open> vendor : update BoringSSL to 0.20260730.0 (#26353) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10221/llama-b10221-bin-macos-arm64.tar.gz) - macOS Apple Sil

<details open> cli : persist reasoning_content in chat history (#26362) * cli : persist reasoning_content in chat history llama-cli collected reasoning from the stream for display but only stored assistant content in messages, so --reasoning-preserve could not re-inject prior

<details open> mtmd: add minicpmv46 downsample (#25993) * add minicpmv46 downsample Signed-off-by: tc-mb <tianchi_cai@icloud.com> * put downsample mode inside gguf. Signed-off-by: tc-mb <tianchi_cai@icloud.com> * build mtmd_image_preprocessor_llava_uhd Signed-off-by: tc-mb

<details open> chat : enable tool call in thinking for DS4 (#26269) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10217/llama-b10217-bin-macos-arm64.tar.gz) - macOS Apple S

<details open> vulkan: add POOL_1D op (#25431) * vulkan : add pool1d push constants and pipeline field Declared data structures needed for POOL1D OP, which are the vk_op_pool1d_push_constants struct and pipeline_pool1d_f32 field. * vulkan : add pool1d compute shader Added po

<details open> vulkan: Introduce driver version check for Windows Intel GPU to mitigate crashing (#25192) * Removed crash guard for Intel Crash fixed from driver 32.0.101.8860 * Added driver version check for windows * Change to convert from driverVersion rather than string

<details open> mtmd: add n_embd_head (#26342) Co-authored-by: Daniel Han <unslothai@gmail.com> </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10214/llama-b10214-bin-macos-a

<details open> Support rotated kv cache quant (#26180) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10213/llama-b10213-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64

<details open> llama : load MTP tensors only if they are really used (#26296) * llama : load MTP tensors only if they are really used * llama : skip loading MTP (if not used) in remaining models that support MTP --------- Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.co

<details open> vulkan: update vulkan sdk to 1.4.357.0 (#26303) </details> **Website:** - <https://llama.app> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10211/llama-b10211-bin-macos-arm64.tar.gz) - macOS Apple Silico