chore: bump llama.cpp to b9864

[github-actions]

llama.cpp update

• Previous pin: b9699
• New pin: b9864
• Upstream release: https://github.com/ggml-org/llama.cpp/releases/tag/b9864
• Compare: ggml-org/llama.cpp@b9699...b9864

Upstream changelog

Release notes for b9864

Details

server + ui: ping silent SSE streams every 1s and kick only after 3s so slow prefill never drops healthy connections (#25241)

• server + ui: ping silent SSE streams every 1s and kick only after 3s so slow prefill never drops healthy connections

• server + ui: sse_ping_interval becomes a per-request body field

Address review from ngxson: the global default returns to 30 so API
clients see no behavior change, and the WebUI sends sse_ping_interval: 1
in the request body since it owns the 3s visibility-kick contract and
declares the cadence it needs. Positive values keep the existing > 0
gate, -1 keeps its disabled semantics.

• server: move sse_ping_interval into the request schema

Address review from ngxson: the field is now a typed field_num with
hard limits (-1, INT32_MAX) bound to task_params, seeded from the CLI
default alongside the other inherited parameters. The raw json_value
read and its redundant comment are gone, and schema evaluation brings
type and range validation for free.

macOS/iOS:

• macOS Apple Silicon (arm64)
• macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED
• macOS Intel (x64)
• iOS XCFramework

Linux:

• Ubuntu x64 (CPU)
• Ubuntu arm64 (CPU)
• Ubuntu s390x (CPU)
• Ubuntu x64 (Vulkan)
• Ubuntu arm64 (Vulkan)
• Ubuntu x64 (ROCm 7.2)
• Ubuntu x64 (OpenVINO)
• Ubuntu x64 (SYCL FP32)
• Ubuntu x64 (SYCL FP16)

Android:

• Android arm64 (CPU)

Windows:

• Windows x64 (CPU)
• Windows arm64 (CPU)
• Windows arm64 (OpenCL Adreno)
• Windows x64 (CUDA 12) - CUDA 12.4 DLLs
• Windows x64 (CUDA 13) - CUDA 13.3 DLLs
• Windows x64 (Vulkan)
• Windows x64 (OpenVINO)
• Windows x64 (SYCL)
• Windows x64 (HIP)

openEuler:

• DISABLED
• openEuler x86 (310p)
• openEuler x86 (910b, ACL Graph)
• openEuler aarch64 (310p)
• openEuler aarch64 (910b, ACL Graph)

UI:

• UI

Commit range

Commits from b9699 to b9864 (first 80)

• [SYCL] rename GGML_SYCL_SUPPORT_LEVEL_ZERO (#24719) (9724f66)
• mtmd: refactor preprocessor, add mtmd_image_preproc_out (#24736) (24bba7b)
• server: fix router args not being forwarded to child instances (#24760) (968c438)
• server: (router) rework -hf preset repo (#24739) (552258c)
• server : return HTTP 400 on invalid grammar (#24144) (#24154) (1078621)
• ui: provide touch accessible model selection UI (#24604) (2083217)
• server : add last-5-seconds generation speed display (#24291) (0802307)
• server: add "schema" and validation (#24150) (e1efd09)
• server: (router) fix stopping_thread potentially hang (#24728) (fe7c8b2)
• docs: fix export-lora --lora-scaled syntax [no release] (#24703) (7b6c5a2)
• hexagon: support for op-trace (fine-grain tracing of HVX/HMX/DMA events) (#24592) (d2c6795)
• mtmd: refactor llava-uhd overview image handling (always use ov_img_first) (#24769) (060ce1b)
• cmake : fix ui build with read-only source (#24752) (32eddaf)
• mtmd: add batching for mtmd-cli, add video tests (#24778) (a6b3260)
• server: add "X-Accel-Buffering": "no" header to streaming endpoints (#24774) (40f3aaf)
• Ggml/cuda col2im 1d (#24417) (3a3edc9)
• mtmd: add batching support for internvl (#24775) (db52540)
• ggml-cpu: support K tails in power10 Q8/Q4 MMA matmul (#24753) (8141e73)
• server : consolidate slot selection into get_available_slot (#24755) (80452d6)
• pi : remove docs from system prompt (#24791) (5bd21b8)
• ggml : bump version to 0.15.2 (ggml/1548) (1868af1)
• sync : ggml (5fd2dc2)
• server: fix non-bound n_discard value (ctx shifting) (#24786) (159d093)
• spec: support eagle3 for qwen3.5 & 3.6 (#24593) (b14e3fb)
• mtmd: several bug fixes (#24784) (e2e7a9b)
• docker : build the UI (#24794) (38724ab)
• server: add --agent arg, remove redundant webui naming compat (#24801) (8c2d6f6)
• vendor : update cpp-httplib to 0.48.0 (#24787) (0d2d9cc)
• arg: Add comment line support to --api-key-file (#23168) (fabde3b)
• server: remove all internal mentions about "webui" (#24817) (175147e)
• mtmd, arg: fix utf8 handling on windows (#24779) (e475fa2)
• server : optimize get_token_probabilities (#24796) (4b48a53)
• server: refactor child --> router communication (#24821) (2b686a9)
• ggml-webgpu: add adapter toggles for F16 on Vulkan + NVIDIA (f449e05)
• convert : more consistent handling of rope_parameters (#24833) (f4043fe)
• ggml : optimize AMX (#24806) (37a77fb)
• model : glm-dsa load DSA indexer tensors as optional (#24770) (796f41b)
• docker : prebuild web UI for s390x build [no release] (#24829) (67e9fd3)
• server: avoid forwarding auth headers in CORS proxy (#24373) (e27f308)
• release: add missing link for win opencl adreno arm64 (#24809) (8452824)
• arg: try fixing test-args-parser randomly fails (#24826) (75f460a)
• llama : use LLM_KV for quantization_version & file_type (#24802) (84de01a)
• fix(hexagon): use padded stride for ssm-conv weights (#24470) (4a80943)
• common/json-schema-to-grammar : align spacing rules with parsers (#24835) (c576070)
• common/peg : refactor until gbnf grammar generation (#24839) (063d9c1)
• spec : Support Step3.5/3.7 flash mtp3 (#24340) (d789527)
• minor : clean-up whitespaces (#24862) (8a118ee)
• server: real-time model load progress tracking via /models/sse (#24828) (d6d8995)
• server: add "verbose" field to schema (#24864) (bfa3219)
• mtmd: add load progress callback (#24865) (2f89acc)
• jinja : implement call statement (#24847) (bf53382)
• mtmd: fix mtmd_get_memory_usage (#24867) (0d135df)
• server: refactor batch construction (#24843) (bddfd2b)
• server: fix report progress for loading spec models, add "stages" list (#24870) (7c082bc)
• common/peg : implement ac parser for stricter grammar generation (#24869) (52b3df0)
• docs/android.md: Add dependency libandroid-spawn for building in termux (#21812) (0ef6f06)
• server: fix edit_file crash on append at end of file (line_start -1) (#24893) (d0f9d2e)
• sampling : remove unconditional softmax+sort in top-n-sigma sampler (#22645) (37957e8)
• [SYCL] support bf16 on bin_bcast OP and unary OPs (#24838) (f8cc15f)
• ui: model status and load progress via /models/sse feed (#24878) (099b579)
• server: refactor/generalize input file schema (#24299) (6ee0f65)
• server: (router) move model downloading to dedicated process (#24834) (721354f)
• ui: Prioritize favorite models in model selection (#24766) (9c0ac88)
• server : Add id to tool call responses api (#24882) (dec5ca5)
• opencl: q8_0 gemv precision improvement (#24923) (23ee879)
• server: improve user message detection and create checkpoints at every user message (#24176) (73618f2)
• codeowners: add yomaytk to ggml-webgpu (#24930) (035cd8f)
• ggml-webgpu: improve MTP inference by using mat-vec path for small batches (#24811) (7c90850)
• model: Granite Speech Plus (#24818) (a3900a6)
• vulkan: link ggml-cpu when GGML_VULKAN_CHECK_RESULTS / RUN_TESTS are enabled (#24444) (c926ad0)
• server: fix remote preset handling, add test (#24938) (75ad0b2)
• vulkan: make mul_mm ALIGNED a spec constant (#24689) (0eb874d)
• vulkan: support CONV_3D (#24612) (c560636)
• vulkan: Support GET_ROWS_BACK (#24883) (92e854a)
• vulkan: support all backend tests for SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU/NORM (#24582) (72a9269)
• server : check draft context creation error (#24922) (be4a6a6)
• vulkan: Apply bias before softmax in FA, to avoid overflow (#24909) (ac4105d)
• model : Add LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M (#24913) (88636e1)
• ui: New Logo + Navigation cleanup & Mobile UI/UX improvements (#24897) (ef9c13d)
• ui: loading bar below the model picker (#24931) (00139b6)

Web bridge review focus

Please pay extra attention to upstream changes touching:

• WebGPU, WASM, Emscripten, pthreads, or memory64 build behavior
• ggml backend APIs used by the bridge
• model loading, tokenizer, chat template, context/state persistence, or cache semantics
• CMake/build flags that can affect the generated JS/WASM artifacts

Validation

• Emscripten build passed
• Browser WebGPU/state-persistence smoke passed
• Generated bridge artifacts include wasm32 and memory64 outputs
• No stale hard-coded llama.cpp tag remains in CI/publish defaults

Automation behavior

This PR is managed from the stable branch automation/bump-llama-cpp. If another llama.cpp release appears before merge, the scheduled workflow updates this same PR instead of opening a duplicate. The workflow skips if a non-automation PR already changes llama_cpp.version.