chore: bump llama.cpp to b9699

[github-actions]

llama.cpp update

• Previous pin: b9165
• New pin: b9699
• Upstream release: https://github.com/ggml-org/llama.cpp/releases/tag/b9699
• Compare: ggml-org/llama.cpp@b9165...b9699

Upstream changelog

Release notes for b9699

Details

sycl : support MUL_MAT and OUT_PROD with Q1_0 (#24721)

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 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 b9165 to b9699 (first 80)

• common : support manually triggering the reasoning budget end sequence (#23949) (5254a79)
• vulkan: Removed unused functions (#23175) (f8c0a19)
• vulkan: Block-load Q3_K/Q6_K block data and subtract on 32b ints (#23056) (1962000)
• model: Add EXAONE 4.5 implementations (#21733) (48b88c3)
• security : disable private disclosures (#23963) (02a5701)
• TP: quantized KV cache support (#23792) (8e6fff8)
• vocab: add normalizer.lowercase support to WPM (#23899) (5aba536)
• vulkan: reduce host memory lock contention (#23376) (bef69f1)
• vulkan: don't hold the device mutex while compiling pipelines (#23641) (55ac090)
• metal: template GLU kernels to support f16/f32 (#23882) (95b8b8e)
• llama: limit max outputs of llama_context (#23861) (de6f727)
• vendor : update cpp-httplib to 0.46.1 (#23980) (335abed)
• opencl: add basic support for q5_0 and q5_1 (#23548) (27d9ed8)
• nix : add nix-nodejs facilities to build Web UI (#23846) (5aa3a64)
• speculative : fix n_outputs_max and remove draft-simple auto-enable (#23988) (5dcb711)
• revert to using global_invocation_id for cpy shader (#23955) (b8275a8)
• opencl: fix compiler warnings for non-adreno path (#23922) (210a657)
• clean up unused variables warnings (#23975) (1fd5f48)
• server: real-time reasoning interruption via control endpoint (#23971) (354ebac)
• hexagon: add gelu_quick (#24007) (d178a11)
• hexagon: MUL_MAT, MUL_MAT_ID, FLASH_ATTN and GDN cleanup and optimizations for latest models (#23989) (8f7f3bf)
• llama : deprecate llama_set_warmup (#24009) (4f3a4be)
• convert : support Step3.7-Flash (#23845) (f7a0777)
• kv-cache : SWA checkpoints store only non-masked cells (#23981) (2365315)
• ui: Add Thinking mode toggle with reasoning effort levels + improvements for Chat Form Add Action UI (#23434) (f8e67fc)
• ui: simplify network error handling (#23431) (69cea5b)
• docs : update HOWTO-add-model.md (#23883) (d5ab083)
• ci : reduce self-hosted server workflow jobs (#24012) (a468b89)
• server: add SSE ping interval (#24013) (60130d1)
• common : fix state save in common_prompt_batch_decode (#23468) (0b71540)
• StepFun 3.5 MTP (#23274) (2187e00)
• model : support granite multilingual embeddings R2 (ibm-granite/granite-embedding-{97,311}m-multilingual-r2) (#22716) (bfb4308)
• model: add Mellum architecture (#23966) (4fb16ec)
• hexagon: profiler output fix and script updates (#24042) (5c394fd)
• opencl: use flat variants of q4_K and q6_K gemv for very large M (#24006) (63e66fd)
• arg : removed unecesary mmproj download when users pass --no-mmproj (#23425) (e366626)
• ci : disable ccache for msvc windows release jobs (#23911) (4da6370)
• update BoringSSL to 0.20260526.0 (#23794) (d545a2a)
• tests : add support for qwen3 SSM archs (#24031) (06938ac)
• cuda: reserve space for quantize kv-cache at startup (#23907) (f8f0a47)
• ggml-cpu: use runtime SVE width in FWHT (#24059) (3571fa5)
• Avoid PDL race conditions by disabling restrict when PDL is used (#24030) (9e58d4d)
• ui: Mermaid Diagrams in chat + interactive preview (#24032) (ee4cf70)
• mtmd, model: allow skip build_vit() (#24077) (a731805)
• mtmd: enable non-causal vision for gemma 4 unified (#24082) (c8d6a00)
• qwen35: use post-norm hidden state for MTP (#24025) (166fe29)
• mtmd: fix Gemma 4 unified FPE (#24088) (94a220c)
• sycl : Improve SYCL doc (#23025) (f478f1b)
• ggml-cpu: extend RVV quantization vec dot to higher VLENs (#22754) (3c7450c)
• ggml-webgpu: FlashAttention refactor + standardize quantization support (#23834) (e8c5489)
• metal : reduce rset heartbeat from 500ms -> 5ms (#24074) (3d19986)
• tests : refactor test-save-load-state to accept token input (#24073) (65ef50a)
• readme : add status badges (#24104) (6ddc943)
• fix(mtmd): handle Gemma 4 audio projector embedding size (#24091) (e3ba22d)
• cmake: skip cvector-generator and export-lora when CPU backend is disabled (#24053) (7ac5a42)
• server : add header to tools/server/server-http.h (#24089) (0066404)
• build : use umbrella Headers directory for XCFramework module map (#23974) (4d74287)
• webui: fix tool selector toggle/counter, key tools by stable identity (#24065) (4586479)
• agents: refactor, include more guidelines (#24111) (a121232)
• server: avoid unnecessary checkpoint restore when new tokens are present (#24110) (6f3a9f3)
• ggml: vectorize ggml_vec_dot_q4_1_q8_1 with WASM SIMD128 (#22209) (4c51309)
• convert: Fix Gemma 4 Unified conversion (#24118) (e802356)
• return filter to save memory (#24125) (0dbfa66)
• ui: added single line reasoning preview (#23601) (5269770)
• ui: Fixed packages (#24119) (21444c8)
• Move duplicated imatrix code into single common imatrix-loader.cpp (#22445) (e7bcf1c)
• webui: [a11y] fix keyboard navigation issues in chat interface and sidebar (#23132) (42b2d60)
• arg: fix double mtp downloads (#24128) (260862b)
• server : disable on-device spec checkpoints (#24108) (7c158fb)
• sycl : port multi-column MMVQ from CUDA backend (#21845) (7fe2ae4)
• ci : build-msys job slimming [no ci] (#24157) (46fa662)
• CUDA: enroll mul_mat_vec_q_moe into pdl (#24087) (2154a0f)
• kleidiai : dynamic chunck-based scheduling for hybrid execution (#23819) (3ecfb15)
• hparams : refactor hparams.n_layer (#24060) (7acb4e8)
• minor : fix lint issues (#24165) (59917d3)
• docs: Update quantization readme (#24133) (ad1b88c)
• ui: add ignore-scripts=true to npmrc (#24149) (cc7bef3)
• Fix link to available UI settings (#24169) (9c955c4)
• ui: run npm install when package-lock.json is newer than node_modules (#24171) (2016bf2)
• model : fix llama_model::n_gpu_layers() (#24188) (96fbe00)

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.