cont : rename common_schema to common_chat_schema

cont : restore (min|max)Length to imply string type
cont : make "{}" mean any object for json_object as well
2026-10-27 06:57:35 -05:00 · 2026-09-12 13:23:12 -05:00 · 2026-09-12 13:23:12 -05:00 · 2026-09-12 13:23:12 -05:00 · 2026-09-12 13:23:12 -05:00 · 2026-09-12 13:23:12 -05:00
1528 changed files with 102146 additions and 230841 deletions
@@ -145,7 +145,7 @@ ENTRYPOINT ["/app/tools.sh"]
 # ==============================================================================
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 ENTRYPOINT [ "/app/llama-cli" ]

@@ -155,9 +155,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 HEALTHCHECK --interval=5m CMD [ "curl", "-f", "http://localhost:8080/health" ]

@@ -104,7 +104,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -114,9 +114,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -113,7 +113,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -123,9 +123,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -1,4 +1,4 @@
-ARG ONEAPI_VERSION=2026.1.1-devel-ubuntu24.04
+ARG ONEAPI_VERSION=2025.3.3-0-devel-ubuntu24.04
 ARG BUILD_DATE=N/A
 ARG APP_VERSION=N/A
 ARG APP_REVISION=N/A
@@ -21,11 +21,6 @@ RUN LLAMA_BUILD_NUMBER="$APP_VERSION" npm run build

 FROM docker.io/intel/deep-learning-essentials:$ONEAPI_VERSION AS build

-RUN echo "Download oneDNN installation package" && \
-    wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/94c3dbac-0852-45be-a57d-21c204cada3e/intel-onednn-2026.0.2.46_offline.sh -O intel-onednn_offline.sh && \
-    echo "Install oneDNN components" && \
-    bash intel-onednn_offline.sh -s -a --silent --eula accept
-
 ARG GGML_SYCL_F16=ON
 ARG LEVEL_ZERO_VERSION=1.28.2
 ARG LEVEL_ZERO_UBUNTU_VERSION=u24.04
@@ -50,7 +45,7 @@ RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
    fi && \
    echo "Building with dynamic libs" && \
    cmake -B build -DGGML_NATIVE=OFF -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_BUILD_TESTS=OFF ${OPT_SYCL_F16} && \
-    cmake --build build --config Release -j$((($(nproc)+1)/2))
+    cmake --build build --config Release -j$(nproc)

 RUN mkdir -p /app/lib && \
    find build -name "*.so*" -exec cp -P {} /app/lib \;
@@ -64,7 +59,7 @@ RUN mkdir -p /app/full \
    && cp requirements.txt /app/full \
    && cp .devops/tools.sh /app/full/tools.sh

-FROM docker.io/intel/oneapi-toolkit:$ONEAPI_VERSION AS base
+FROM docker.io/intel/deep-learning-essentials:$ONEAPI_VERSION AS base

 ARG BUILD_DATE=N/A
 ARG APP_VERSION=N/A
@@ -107,7 +102,7 @@ RUN mkdir /tmp/neo/ && cd /tmp/neo/ \
  && dpkg --install *.deb

 RUN apt-get update \
-    && apt-get install -y libgomp1 curl ffmpeg openssl libssl-dev \
+    && apt-get install -y libgomp1 curl ffmpeg \
    && apt autoremove -y \
    && apt clean -y \
    && rm -rf /tmp/* /var/tmp/* \
@@ -146,7 +141,7 @@ ENTRYPOINT ["/app/tools.sh"]
 FROM base AS light

 COPY --from=build /app/lib/ /app
-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -156,10 +151,9 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

 COPY --from=build /app/lib/ /app
-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -1,9 +1,10 @@
 ARG UBUNTU_VERSION=22.04
 # This needs to generally match the container host's environment.
+ARG MUSA_VERSION=rc4.3.0
 # Target the MUSA build image
-ARG BASE_MUSA_DEV_CONTAINER=registry.mthreads.com/mcconline/musa_sdk:5.2.0-devel-ubuntu${UBUNTU_VERSION}-s5000
+ARG BASE_MUSA_DEV_CONTAINER=docker.io/mthreads/musa:${MUSA_VERSION}-devel-ubuntu${UBUNTU_VERSION}-amd64

-ARG BASE_MUSA_RUN_CONTAINER=registry.mthreads.com/mcconline/musa_sdk:5.2.0-runtime-ubuntu${UBUNTU_VERSION}-s5000
+ARG BASE_MUSA_RUN_CONTAINER=docker.io/mthreads/musa:${MUSA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}-amd64

 ARG BUILD_DATE=N/A
 ARG APP_VERSION=N/A
@@ -36,10 +37,7 @@ RUN apt-get update && \
    python3-pip \
    git \
    libssl-dev \
-    libgomp1 \
-    musa-mualg-5-2 \
-    musa-muthrust-5-2 \
-    libmthreads-compute
+    libgomp1

 WORKDIR /app

@@ -82,16 +80,13 @@ LABEL org.opencontainers.image.created=$BUILD_DATE \
      org.opencontainers.image.source=$IMAGE_SOURCE

 RUN apt-get update \
-    && apt-get install -y libgomp1 curl ffmpeg libmthreads-compute \
+    && apt-get install -y libgomp1 curl ffmpeg \
    && apt autoremove -y \
    && apt clean -y \
    && rm -rf /tmp/* /var/tmp/* \
    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
    && find /var/cache -type f -delete

-# The MUSA runtime image does not register its library directory
-RUN echo "/usr/local/musa/lib" > /etc/ld.so.conf.d/musa-runtime.conf && ldconfig
-
 COPY --from=build /app/lib/ /app

 ### Full
@@ -120,7 +115,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -130,9 +125,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -1,18 +1,18 @@
-ARG OPENVINO_VERSION_MAJOR=2026.4.1
-ARG OPENVINO_VERSION_FULL=2026.4.1.22982.07f9c262b05
+ARG OPENVINO_VERSION_MAJOR=2026.3.1
+ARG OPENVINO_VERSION_FULL=2026.3.1.22476.56d9685302d
 ARG UBUNTU_VERSION=24.04

 # Intel GPU driver versions. https://github.com/intel/compute-runtime/releases
-ARG IGC_VERSION=v2.41.5
-ARG IGC_VERSION_FULL=2_2.41.5+22716
-ARG COMPUTE_RUNTIME_VERSION=26.35.39758.10
-ARG COMPUTE_RUNTIME_VERSION_FULL=26.35.39758.10-0
+ARG IGC_VERSION=v2.40.13
+ARG IGC_VERSION_FULL=2_2.40.13+22418
+ARG COMPUTE_RUNTIME_VERSION=26.31.39395.13
+ARG COMPUTE_RUNTIME_VERSION_FULL=26.31.39395.13-0
 ARG IGDGMM_VERSION=22.10.0

 # Intel NPU driver versions. https://github.com/intel/linux-npu-driver/releases
-ARG NPU_DRIVER_VERSION=v1.38.0
-ARG NPU_DRIVER_FULL=v1.38.0.20260910-34487311128
-ARG LIBZE1_VERSION=1.32.0-1~24.04~ppa1
+ARG NPU_DRIVER_VERSION=v1.35.0
+ARG NPU_DRIVER_FULL=v1.35.0.20260722-29947505341
+ARG LIBZE1_VERSION=1.28.2-1~24.04~ppa1

 # Optional proxy build arguments
 ARG http_proxy=
@@ -173,7 +173,7 @@ RUN --mount=type=cache,target=/var/cache/intel-npu,sharing=locked \
    fi; \
    DEB=/var/cache/intel-npu/libze1_${LIBZE1_VERSION}_amd64.deb; \
    if [ ! -f "$DEB" ]; then \
-        wget -q -O "$DEB" https://snapshot.ppa.launchpadcontent.net/kobuk-team/intel-graphics/ubuntu/20260830T100000Z/pool/main/l/level-zero-loader/libze1_${LIBZE1_VERSION}_amd64.deb; \
+        wget -q -O "$DEB" https://snapshot.ppa.launchpadcontent.net/kobuk-team/intel-graphics/ubuntu/20260606T100000Z/pool/main/l/level-zero-loader/libze1_${LIBZE1_VERSION}_amd64.deb; \
    fi; \
    mkdir /tmp/npu/ && cd /tmp/npu/ && tar -xf "$TGZ" && cp "$DEB" .; \
    apt-get update; \
@@ -227,7 +227,6 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

 COPY --from=build /app/full/llama /app/full/llama-server /app/

@@ -126,7 +126,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -136,9 +136,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -133,7 +133,6 @@ ENTRYPOINT [ "/llama.cpp/bin/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

 WORKDIR /llama.cpp/bin

@@ -107,7 +107,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -117,9 +117,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -97,7 +97,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light

-COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/
+COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app

 WORKDIR /app

@@ -107,9 +107,8 @@ ENTRYPOINT [ "/app/llama-cli" ]
 FROM base AS server

 ENV LLAMA_ARG_HOST=0.0.0.0
-ENV LLAMA_ARG_PORT=8080

-COPY --from=build /app/full/llama /app/full/llama-server /app/
+COPY --from=build /app/full/llama /app/full/llama-server /app

 WORKDIR /app

@@ -1,5 +1,5 @@
 {
-  "Exclude": ["^\\.gitmodules$", "stb_image\\.h", "examples/test-cmake/build/", "examples/test-cmake/build-subdir/"],
+  "Exclude": ["^\\.gitmodules$", "stb_image\\.h"],
  "Disable": {
    "IndentSize": true
  }
@@ -45,9 +45,6 @@ insert_final_newline = unset
 trim_trailing_whitespace = unset
 insert_final_newline = unset

-[vendor/**.patch]
-trim_trailing_whitespace = unset
-
 [tools/ui/**]
 indent_style = unset
 indent_size = unset
@@ -5,8 +5,6 @@ exclude =
    # Do not traverse examples and tools
    examples,
    tools,
-    # Do not traverse .venv directories
-    .venv*,
    # Do not include package initializers
    __init__.py,
    # No need to traverse our git directory
@@ -14,7 +14,7 @@ runs:
      run: |
        BUILD_NUMBER="$(git rev-list --count HEAD)"
        SHORT_HASH="$(git rev-parse --short=7 HEAD)"
-        if [[ "${{ env.BRANCH_NAME }}" == "master" || "${{ env.BRANCH_NAME }}" == "b${BUILD_NUMBER}" ]]; then
+        if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
          echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
        else
          SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
@@ -100,36 +100,36 @@ runs:
          echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
          echo "CUDA_PATH_V13_1=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8

-    - name: Install Cuda Toolkit 13.4 for x64
-      if: ${{ inputs.cuda_version == '13.4' && inputs.cuda_arch == 'x64' }}
+    - name: Install Cuda Toolkit 13.3
+      if: ${{ inputs.cuda_version == '13.3' }}
      shell: pwsh
      run: |
-          mkdir -p "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4"
+          mkdir -p "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3"
          choco install unzip -y
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_crt/windows-x86_64/cuda_crt-windows-x86_64-13.4.59-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_cudart/windows-x86_64/cuda_cudart-windows-x86_64-13.4.49-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvcc/windows-x86_64/cuda_nvcc-windows-x86_64-13.4.59-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvrtc/windows-x86_64/cuda_nvrtc-windows-x86_64-13.4.59-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libcublas/windows-x86_64/libcublas-windows-x86_64-13.7.0.27-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libnvvm/windows-x86_64/libnvvm-windows-x86_64-13.4.59-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvtx/windows-x86_64/cuda_nvtx-windows-x86_64-13.4.49-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_profiler_api/windows-x86_64/cuda_profiler_api-windows-x86_64-13.4.49-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/visual_studio_integration/windows-x86_64/visual_studio_integration-windows-x86_64-13.4.49-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cccl/windows-x86_64/cccl-windows-x86_64-13.3.4.2.1-archive.zip"
-          unzip '*.zip' -d "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4"
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_crt-windows-x86_64-13.4.59-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_cudart-windows-x86_64-13.4.49-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_nvcc-windows-x86_64-13.4.59-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_nvrtc-windows-x86_64-13.4.59-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\libcublas-windows-x86_64-13.7.0.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\libnvvm-windows-x86_64-13.4.59-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_nvtx-windows-x86_64-13.4.49-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_profiler_api-windows-x86_64-13.4.49-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\visual_studio_integration-windows-x86_64-13.4.49-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cccl-windows-x86_64-13.3.4.2.1-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
-          echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
-          echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
-          echo "CUDA_PATH_V13_4=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_crt/windows-x86_64/cuda_crt-windows-x86_64-13.3.33-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_cudart/windows-x86_64/cuda_cudart-windows-x86_64-13.3.29-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvcc/windows-x86_64/cuda_nvcc-windows-x86_64-13.3.33-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvrtc/windows-x86_64/cuda_nvrtc-windows-x86_64-13.3.33-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libcublas/windows-x86_64/libcublas-windows-x86_64-13.5.1.27-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libnvvm/windows-x86_64/libnvvm-windows-x86_64-13.3.33-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvtx/windows-x86_64/cuda_nvtx-windows-x86_64-13.3.29-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_profiler_api/windows-x86_64/cuda_profiler_api-windows-x86_64-13.3.27-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/visual_studio_integration/windows-x86_64/visual_studio_integration-windows-x86_64-13.3.27-archive.zip"
+          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cccl/windows-x86_64/cccl-windows-x86_64-13.3.3.3.1-archive.zip"
+          unzip '*.zip' -d "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3"
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_crt-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_cudart-windows-x86_64-13.3.29-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvcc-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvrtc-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\libcublas-windows-x86_64-13.5.1.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\libnvvm-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvtx-windows-x86_64-13.3.29-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_profiler_api-windows-x86_64-13.3.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\visual_studio_integration-windows-x86_64-13.3.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cccl-windows-x86_64-13.3.3.3.1-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
+          echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
+          echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
+          echo "CUDA_PATH_V13_3=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8

    - name: Install Cuda Toolkit 13.4 for ARM64
      if: ${{ inputs.cuda_version == '13.4' && inputs.cuda_arch == 'arm64' }}
@@ -4,10 +4,6 @@ on:
  issues:
    types: [opened]

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
  find-related:
    if: github.event.action == 'opened'
@@ -1,109 +0,0 @@
-name: Behavior Check
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/behavior-check.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/behavior-check.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  GGML_SCHED_DEBUG_REALLOC: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  ubuntu:
-    runs-on: ubuntu-26.04-arm
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          sudo apt update
-          sudo apt install -y build-essential libssl-dev git-lfs libjpeg-dev python3 python3-venv python3-pip
-
-      - name: Python Dependencies
-        id: python_depends
-        run: |
-          python3 -m venv .venv
-          source .venv/bin/activate
-          pip install jinja2==3.1.6
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-        with:
-          key: behavior-check-ubuntu-26.04-arm
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DGGML_NATIVE=OFF \
-            -DGGML_BACKEND_DL=OFF \
-            -DLLAMA_FATAL_WARNINGS=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: behavior-check-ubuntu-26.04-arm
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      - name: Test
-        id: cmake_test
-        run: |
-          source .venv/bin/activate
-          cd build
-          ctest -L behavior -LE 'model|backend' --verbose --timeout 900
@@ -15,10 +15,6 @@ on:
      '**/*.cpp'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -23,27 +23,19 @@ on:
      - 'scripts/snapdragon/**'
      - 'CMakePresets.json'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  android-ndk-snapdragon:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest
    container:
-      image: 'ghcr.io/snapdragon-toolchain/arm64-android:v0.8'
+      image: 'ghcr.io/snapdragon-toolchain/arm64-android:v0.7'
    defaults:
      run:
        shell: bash

-    permissions:
-      actions: write
-      contents: read
-
    steps:
      - name: Clone
        uses: actions/checkout@v6
@@ -67,17 +59,13 @@ jobs:
          path: pkg-snapdragon/llama.cpp

  linux-iot-snapdragon:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest
    container:
-      image: 'ghcr.io/snapdragon-toolchain/arm64-linux:v0.8'
+      image: 'ghcr.io/snapdragon-toolchain/arm64-linux:v0.7'
    defaults:
      run:
        shell: bash

-    permissions:
-      actions: write
-      contents: read
-
    steps:
      - name: Clone
        uses: actions/checkout@v6
@@ -104,15 +92,11 @@ jobs:
    name: Test on QDC Device (${{ matrix.device }})
    needs: [android-ndk-snapdragon, linux-iot-snapdragon]
    runs-on: ubuntu-24.04-arm
-    timeout-minutes: 180
+    timeout-minutes: 90
    strategy:
      fail-fast: false
      matrix:
-        device: [SM8850, QCS9075M]
-
-    permissions:
-      actions: read
-      contents: read
+        device: [SM8750, SM8850, QCS9075M]

    steps:
      - name: Checkout
@@ -155,7 +139,7 @@ jobs:
              --pkg-dir    pkg-snapdragon/llama.cpp \
              --model-url  "https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_0.gguf" \
              --device     ${{ matrix.device }} \
-              ${{ startsWith(matrix.device, 'QCS') && '--job-timeout 5400 --retries 2 --retry-delay 300' || '' }}
+              ${{ startsWith(matrix.device, 'QCS') && '--retries 2 --retry-delay 300' || '' }}
        env:
          QDC_API_KEY: ${{ secrets.QDC_API_KEY }}

@@ -20,10 +20,6 @@ on:
      - '.github/workflows/build-android.yml'
      - 'examples/llama.android/**'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -37,7 +33,7 @@ env:

 jobs:
  default:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest

    steps:
      - name: Clone
@@ -53,7 +49,7 @@ jobs:
          distribution: zulu

      - name: Setup Android SDK
-        uses: android-actions/setup-android@be39fa834029ff78f1a44aa3bb0819b8fc2bd8fd # v4.0.4
+        uses: android-actions/setup-android@40fd30fb8d7440372e1316f5d1809ec01dcd3699 # v4.0.1
        with:
          log-accepted-android-sdk-licenses: false

@@ -63,17 +59,13 @@ jobs:
          ./gradlew build --no-daemon

  ndk:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest
    container:
      image: 'ghcr.io/snapdragon-toolchain/arm64-android:v0.3'
    defaults:
      run:
        shell: bash

-    permissions:
-      actions: write
-      contents: read
-
    steps:
      - name: Clone
        uses: actions/checkout@v6
@@ -101,7 +93,7 @@ jobs:
          path: pkg-adb/llama.cpp

  arm64:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest

    env:
      NDK_VERSION: "29.0.14206865"
@@ -131,7 +123,7 @@ jobs:
          distribution: temurin

      - name: Setup Android SDK
-        uses: android-actions/setup-android@be39fa834029ff78f1a44aa3bb0819b8fc2bd8fd # v4.0.4
+        uses: android-actions/setup-android@40fd30fb8d7440372e1316f5d1809ec01dcd3699 # v4.0.1
        with:
          log-accepted-android-sdk-licenses: false

@@ -26,10 +26,6 @@ on:
      'ggml/src/ggml-rpc/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -37,7 +33,6 @@ concurrency:
 env:
  GGML_NLOOP: 3
  GGML_N_THREADS: 1
-  GGML_SCHED_DEBUG_REALLOC: 1
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
  LLAMA_ARG_LOG_TIMESTAMPS: 1
@@ -54,7 +49,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: apple-arm64
          save: false

      - name: ccache-buckets-restore
@@ -103,9 +98,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          # Metal Paravirtual devices are difficult to support -> disable
-          # ref: https://github.com/ggml-org/llama.cpp/pull/19802#issuecomment-4013704023
-          ctest -L main -E "test-llama-archs|test-llama-context|test-speculative-rejection-sampling" --verbose --timeout 900
+          ctest -L main -E "test-llama-archs" --verbose --timeout 900

  macos-latest-x64:
    runs-on: macos-15-intel
@@ -118,7 +111,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: apple-x64
          save: false

      - name: ccache-buckets-restore
@@ -166,10 +159,6 @@ jobs:
  macos-latest-ios-xcode:
    runs-on: macos-latest

-    permissions:
-      actions: write
-      contents: read
-
    steps:
      - name: Checkout code
        uses: actions/checkout@v6
@@ -267,10 +256,6 @@ jobs:
    runs-on: macos-latest
    needs: macos-latest-ios-xcode

-    permissions:
-      actions: read
-      contents: read
-
    strategy:
      matrix:
        destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
@@ -5,10 +5,6 @@ on:
  schedule:
    - cron: '0 * * * *'

-cache-mode: write
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -44,9 +40,9 @@ jobs:
    runs-on: ubuntu-24.04

    env:
-      # Sync versions in build-openvino.yml, ci-self-hosted-openvino.yml, release.yml, release-publish.yml, build-cache.yml, .devops/openvino.Dockerfile, docs/backend/OPENVINO.md
-      OPENVINO_VERSION_MAJOR: "2026.4.1"
-      OPENVINO_VERSION_FULL: "2026.4.1.22982.07f9c262b05"
+      # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
+      OPENVINO_VERSION_MAJOR: "2026.3.1"
+      OPENVINO_VERSION_FULL: "2026.3.1.22476.56d9685302d"

    steps:
      - name: Clone
@@ -72,9 +68,9 @@ jobs:
    runs-on: windows-2022

    env:
-      # Sync versions in build-openvino.yml, ci-self-hosted-openvino.yml, release.yml, release-publish.yml, build-cache.yml, .devops/openvino.Dockerfile, docs/backend/OPENVINO.md
-      OPENVINO_VERSION_MAJOR: "2026.4.1"
-      OPENVINO_VERSION_FULL: "2026.4.1.22982.07f9c262b05"
+      # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
+      OPENVINO_VERSION_MAJOR: "2026.3.1"
+      OPENVINO_VERSION_FULL: "2026.3.1.22476.56d9685302d"

    steps:
      - name: Clone
@@ -22,10 +22,6 @@ on:
      'ggml/src/ggml-cann/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -3,13 +3,9 @@ on:
  workflow_dispatch:
  workflow_call:

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
  linux:
-    runs-on: [self-hosted, Linux, CPU]
+    runs-on: [self-hosted, Linux]
    steps:
      - uses: actions/checkout@v6
        with:
@@ -30,10 +30,6 @@ on:
      '**/*.cpp'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -41,7 +37,6 @@ concurrency:
 env:
  GGML_NLOOP: 3
  GGML_N_THREADS: 1
-  GGML_SCHED_DEBUG_REALLOC: 1
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
  LLAMA_ARG_LOG_TIMESTAMPS: 1
@@ -69,7 +64,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: cpu-${{ matrix.os }}
          save: false

      - name: Build Dependencies
@@ -93,7 +88,7 @@ jobs:
        run: |
          export PIP_BREAK_SYSTEM_PACKAGES="1"
          python3 -m pip install --upgrade pip setuptools
-          pip3 install ./gguf-py jinja2==3.1.6
+          pip3 install ./gguf-py

      - name: ccache-buckets-restore
        uses: ./.github/actions/ccache-buckets
@@ -146,11 +141,6 @@ jobs:
    name: windows / ${{ matrix.build }}
    runs-on: windows-2025

-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
    env:
      OPENBLAS_VERSION: 0.3.23
      SDE_VERSION: 9.33.0-2024-01-07
@@ -15,10 +15,6 @@ on:
  schedule:
    - cron: '0 0 * * 0'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -24,10 +24,6 @@ on:
      'ggml/src/ggml-cuda/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -59,7 +55,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: cuda-ubuntu-24.04-cuda
          save: false

      - name: ccache-buckets-restore
@@ -72,7 +68,7 @@ jobs:
          hf_bucket: ggml-org/cache

      - name: Build with CMake
-        # TODO: Drop GGML_CUDA_CCCL_VERSION when this job uses CTK >= 13.5, which bundles CCCL >= 3.5.
+        # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project
        run: |
          cmake -S . -B build -G Ninja \
            -DLLAMA_FATAL_WARNINGS=ON \
@@ -81,7 +77,7 @@ jobs:
            -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
            -DGGML_NATIVE=OFF \
            -DGGML_CUDA=ON \
-            -DGGML_CUDA_CCCL_VERSION=v3.4.3
+            -DGGML_CUDA_CUB_3DOT2=ON
          cmake --build build

      - name: ccache-buckets-save
@@ -114,7 +110,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: cuda-ubuntu-22.04-hip
          save: false

      - name: ccache-buckets-restore
@@ -149,7 +145,7 @@ jobs:

  musa:
    runs-on: ubuntu-22.04
-    container: registry.mthreads.com/mcconline/musa_sdk:5.2.0-devel-ubuntu22.04-s5000
+    container: mthreads/musa:rc4.3.0-devel-ubuntu22.04-amd64

    steps:
      - name: Clone
@@ -160,12 +156,12 @@ jobs:
        id: depends
        run: |
          apt-get update
-          apt-get install -y build-essential git cmake libssl-dev jq python3-venv musa-mualg-5-2 musa-muthrust-5-2 libmthreads-compute
+          apt-get install -y build-essential git cmake libssl-dev jq

      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: cuda-ubuntu-22.04-musa
          save: false

      - name: ccache-buckets-restore
@@ -181,9 +177,8 @@ jobs:
        id: cmake_build
        run: |
          cmake -B build -S . \
-            -DGGML_MUSA=ON \
-            -DMUSA_ARCHITECTURES=31
-          cmake --build build --config Release -j $(nproc)
+            -DGGML_MUSA=ON
+          time cmake --build build --config Release -j $(nproc)

      - name: ccache-buckets-save
        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
@@ -7,11 +7,6 @@ name: CI (CUDA, windows)
 on:
  workflow_dispatch: # allows manual triggering

-cache-mode: write
-permissions:
-  actions: write
-  contents: read
-
 # note: this will run in queue with the release workflow
 concurrency:
  group: release
@@ -36,16 +31,15 @@ jobs:
    strategy:
      matrix:
        include:
-          # CTK >= 13.5 bundles CCCL >= 3.5; omit GGML_CUDA_CCCL_VERSION for those versions.
          - cuda: '12.4'
            arch: x64
-            defines: '-DGGML_CUDA_CCCL_VERSION=v3.4.3'
-          - cuda: '13.4'
+            defines: '-DGGML_CUDA_CUB_3DOT2=ON'
+          - cuda: '13.3'
            arch: x64
-            defines: '-DGGML_CUDA_CCCL_VERSION=v3.4.3'
+            defines: ''
          - cuda: '13.4'
            arch: arm64
-            defines: '-DGGML_CUDA_CCCL_VERSION=v3.4.3 -DCMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc-cuda.cmake'
+            defines: '-DCMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc-cuda.cmake'

    steps:
      - name: Clone
@@ -19,18 +19,9 @@ on:
    types: [opened, synchronize, reopened]
    paths: [
      '.github/workflows/build-ibm.yml',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-zdnn/**',
-      'ggml/src/ggml-quants.c',
-      'src/llama-quant.cpp',
-      'gguf-py/gguf/scripts/gguf_convert_endian.py',
-      'tests/test-quantize-be.sh'
+      'ggml/src/ggml-cpu/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -43,14 +34,9 @@ env:
  LLAMA_ARG_LOG_TIMESTAMPS: 1

 jobs:
-  ubuntu-24-s390x:
-    name: ubuntu-24-s390x (VXE ${{ matrix.vxe }})
-    runs-on: ubuntu-24.04-s390x

-    strategy:
-      fail-fast: false
-      matrix:
-        vxe: ["ON", "OFF"]  # `-DGGML_VXE=ON/OFF`
+  ubuntu-24-s390x:
+    runs-on: ubuntu-24.04-s390x

    steps:
      - name: Clone
@@ -91,8 +77,7 @@ jobs:
        run: |
          cmake -B build \
            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_RPC=ON \
-            -DGGML_VXE=${{ matrix.vxe }}
+            -DGGML_RPC=ON
          time cmake --build build --config Release -j $(nproc)

      - name: Test
@@ -109,51 +94,6 @@ jobs:
          wget https://huggingface.co/ggml-org/models/resolve/main/tinyllamas/stories260K-be.gguf
          ./bin/llama-completion -m stories260K-be.gguf -p "One day, Lily met a Shoggoth" -n 500 -c 256

-      - name: Test quantize (s390x)
-        id: quantize_test_s390x
-        run: |
-          ./tests/test-quantize-be.sh ./build/bin/llama-quantize ./build/bin/llama-perplexity 2>&1
-
-  ubuntu-26-zdnn-s390x:
-    name: ubuntu-26-zdnn-s390x
-    runs-on: ubuntu-24.04-s390x
-    container: ubuntu:26.04  # required to get GCC 15.1 and binutils 2.44
-    defaults:
-      run:
-        shell: bash
-
-    steps:
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          apt-get update
-          apt-get install -y --no-install-recommends \
-            build-essential cmake git ca-certificates \
-            libssl-dev libzdnn-dev
-
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Toolchain workaround (GCC 15)
-        run: |
-          apt-get install -y gcc-15 g++-15
-          echo "CC=gcc-15" >> "$GITHUB_ENV"
-          echo "CXX=g++-15" >> "$GITHUB_ENV"
-
-      - name: Build with zDNN Backend
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_NATIVE=OFF \
-            -DGGML_VXE=ON \
-            -DGGML_ZDNN=ON \
-            -DGGML_RPC=ON \
-            -DCMAKE_C_FLAGS="-march=arch15" \
-            -DCMAKE_CXX_FLAGS="-march=arch15"
-          time cmake --build build --config Release -j $(nproc)
-
  ubuntu-24-ppc64le:
    runs-on: ubuntu-24.04-ppc64le

@@ -8,10 +8,6 @@ on:
  schedule:
    - cron: '0 0 * * 0'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -23,11 +23,6 @@ on:
      'ggml/src/ggml-opencl/**'
    ]

-cache-mode: write
-permissions:
-  actions: write
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -22,10 +22,6 @@ on:
      'ggml/src/ggml-openvino/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -37,16 +33,16 @@ env:
  LLAMA_ARG_LOG_PREFIX: 1
  LLAMA_ARG_LOG_TIMESTAMPS: 1
  # TODO: fix failing tests on OpenVINO backend
-  CTEST_EXCLUDE: "test-llama-archs|test-llama-context"
+  CTEST_EXCLUDE: "test-llama-archs|^test-recurrent-state-|test-save-load-state"

 jobs:
  ubuntu-24-openvino:
    runs-on: [self-hosted, Linux, Intel, OpenVINO]

    env:
-      # Sync versions in build-openvino.yml, ci-self-hosted-openvino.yml, release.yml, release-publish.yml, build-cache.yml, .devops/openvino.Dockerfile, docs/backend/OPENVINO.md
-      OPENVINO_VERSION_MAJOR: "2026.4.1"
-      OPENVINO_VERSION_FULL: "2026.4.1.22982.07f9c262b05"
+      # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
+      OPENVINO_VERSION_MAJOR: "2026.3.1"
+      OPENVINO_VERSION_FULL: "2026.3.1.22476.56d9685302d"

    steps:
      - name: Clone
@@ -98,15 +94,10 @@ jobs:
  openvino-windows-2022:
    runs-on: windows-2022

-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
    env:
-      # Sync versions in build-openvino.yml, ci-self-hosted-openvino.yml, release.yml, release-publish.yml, build-cache.yml, .devops/openvino.Dockerfile, docs/backend/OPENVINO.md
-      OPENVINO_VERSION_MAJOR: "2026.4.1"
-      OPENVINO_VERSION_FULL: "2026.4.1.22982.07f9c262b05"
+      # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
+      OPENVINO_VERSION_MAJOR: "2026.3.1"
+      OPENVINO_VERSION_FULL: "2026.3.1.22476.56d9685302d"

    steps:
      - name: Clone
@@ -22,10 +22,6 @@ on:
      'ggml/src/ggml-cpu/arch/riscv/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -21,10 +21,6 @@ on:
      '.github/workflows/build-sanitize.yml'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -0,0 +1,445 @@
+name: CI (self-hosted)
+
+on:
+  workflow_dispatch: # allows manual triggering
+  push:
+    branches:
+      - master
+    paths: [
+      '.github/workflows/build-self-hosted.yml',
+      '**/CMakeLists.txt',
+      '**/.cmake',
+      '**/*.h',
+      '**/*.hpp',
+      '**/*.c',
+      '**/*.cpp',
+      '**/*.cu',
+      '**/*.cuh',
+      '**/*.swift',
+      '**/*.m',
+      '**/*.metal',
+      '**/*.comp',
+      '**/*.glsl',
+      '**/*.wgsl'
+    ]
+
+  pull_request:
+    types: [opened, synchronize, reopened]
+    paths: [
+      '.github/workflows/build-self-hosted.yml',
+      '**/CMakeLists.txt',
+      '**/.cmake',
+      '**/*.h',
+      '**/*.hpp',
+      '**/*.c',
+      '**/*.cpp',
+      '**/*.cu',
+      '**/*.cuh',
+      '**/*.swift',
+      '**/*.m',
+      '**/*.metal',
+      '**/*.comp',
+      '**/*.glsl',
+      '**/*.wgsl'
+    ]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
+  cancel-in-progress: true
+
+env:
+  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
+  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
+  GGML_NLOOP: 3
+  GGML_N_THREADS: 1
+  LLAMA_ARG_LOG_COLORS: 1
+  LLAMA_ARG_LOG_PREFIX: 1
+  LLAMA_ARG_LOG_TIMESTAMPS: 1
+
+jobs:
+  gpu-cuda:
+    runs-on: "hf-jobs-t4-small:cuda13"
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y cmake libssl-dev time unzip wget python3 python3-venv python3-pip
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.24
+        with:
+          restore: false
+          save: false
+
+      - name: ccache-buckets-restore
+        uses: ./.github/actions/ccache-buckets
+        with:
+          key: self-hosted-gpu-cuda
+          folder: llama.cpp
+          hf_bucket: ggml-org/cache
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          nvidia-smi
+          GG_BUILD_CUDA=1 CUDACXX=/usr/local/cuda/bin/nvcc bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+      - name: ccache-buckets-save
+        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
+        with:
+          key: self-hosted-gpu-cuda
+          folder: llama.cpp
+          evict-old-files: 1d
+          hf_bucket: ggml-org/cache
+          save: true
+
+  gpu-rocm:
+    runs-on: [self-hosted, Linux, AMD]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        # HIP_LAUNCH_BLOCKING=1: workaround for an async-execution correctness
+        # issue on integrated RDNA3.5 (gfx1151) where batched inference returns
+        # incorrect output (perplexity ~88 vs ~9.4). Serializing kernel launches
+        # restores correctness. Remove once the underlying ROCm/HIP issue is fixed.
+        env:
+          HIP_LAUNCH_BLOCKING: "1"
+        run: |
+          rocminfo
+          GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS=gfx1151 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-vulkan-nvidia-cm:
+    runs-on: [self-hosted, Linux, NVIDIA]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-vulkan-nvidia-cm2:
+    runs-on: [self-hosted, Linux, NVIDIA, COOPMAT2]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-webgpu-nvidia:
+    runs-on: [self-hosted, Linux, NVIDIA, X64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Dawn Dependency
+        id: dawn-depends
+        run: |
+          DAWN_VERSION="v20260908.214631"
+          DAWN_OWNER="google"
+          DAWN_REPO="dawn"
+          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-ubuntu-latest-Release"
+          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          curl -L -o artifact.tar.gz \
+            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          mkdir dawn
+          tar -xvf artifact.tar.gz -C dawn --strip-components=1
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          GG_BUILD_WEBGPU=1 \
+          GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
+          GG_BUILD_WEBGPU_DAWN_DIR="$GITHUB_WORKSPACE/dawn/lib64/cmake/Dawn" \
+            bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  # TODO: provision AMX-compatible machine
+  #cpu-amx:
+  #  runs-on: [self-hosted, Linux, CPU, AMX]
+
+  #  steps:
+  #    - name: Clone
+  #      id: checkout
+  #      uses: actions/checkout@v6
+
+  #    - name: Test
+  #      id: ggml-ci
+  #      run: |
+  #        bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  # TODO: provision AMD GPU machine
+  # amd-vulkan:
+  #   runs-on: [self-hosted, Linux, AMD]
+
+  #   steps:
+  #     - name: Clone
+  #       id: checkout
+  #       uses: actions/checkout@v6
+
+  #     - name: Test
+  #       id: ggml-ci
+  #       run: |
+  #         vulkaninfo --summary
+  #         GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  # TODO: provision AMD GPU machine
+  # amd-rocm:
+  #   runs-on: [self-hosted, Linux, AMD]
+
+  #   steps:
+  #     - name: Clone
+  #       id: checkout
+  #       uses: actions/checkout@v6
+
+  #     - name: Test
+  #       id: ggml-ci
+  #       run: |
+  #         amd-smi static
+  #         GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-metal:
+    runs-on: [self-hosted, macOS, ARM64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-webgpu-apple:
+    runs-on: [self-hosted, macOS, ARM64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Dawn Dependency
+        id: dawn-depends
+        run: |
+          DAWN_VERSION="v20260908.214631"
+          DAWN_OWNER="google"
+          DAWN_REPO="dawn"
+          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-macos-latest-Release"
+          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          curl -L -o artifact.tar.gz \
+            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          mkdir dawn
+          tar -xvf artifact.tar.gz -C dawn --strip-components=1
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          GG_BUILD_WEBGPU=1 GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
+            bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-vulkan-apple:
+    runs-on: [self-hosted, macOS, ARM64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-vulkan-intel-linux:
+    runs-on: [self-hosted, Linux, Intel]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+        with:
+          persist-credentials: false
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  gpu-vulkan-intel-windows:
+    runs-on: [self-hosted, Windows, X64, Intel]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        shell: C:\msys64\usr\bin\bash.exe --noprofile --norc -eo pipefail "{0}"
+        env:
+          MSYSTEM: UCRT64
+          CHERE_INVOKING: 1
+          PATH: C:\msys64\ucrt64\bin;C:\msys64\usr\bin;C:\Windows\System32;${{ env.PATH }}
+        run: |
+          vulkaninfo --summary
+          # Skip python related tests with GG_BUILD_LOW_PERF=1 since Windows MSYS2 UCRT64 currently fails to create
+          # a valid python environment for testing
+          LLAMA_FATAL_WARNINGS=OFF GG_BUILD_NINJA=1 GG_BUILD_VULKAN=1 GG_BUILD_LOW_PERF=1 ./ci/run.sh ./results/llama.cpp ./mnt/llama.cpp
+
+  gpu-openvino-low-perf:
+    runs-on: [self-hosted, Linux, Intel, OpenVINO]
+
+    env:
+      # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
+      OPENVINO_VERSION_MAJOR: "2026.3.1"
+      OPENVINO_VERSION_FULL: "2026.3.1.22476.56d9685302d"
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Setup OpenVINO Toolkit
+        uses: ./.github/actions/linux-setup-openvino
+        with:
+          path: ./openvino_toolkit
+          version_major: ${{ env.OPENVINO_VERSION_MAJOR }}
+          version_full: ${{ env.OPENVINO_VERSION_FULL }}
+
+      - name: Install OpenVINO dependencies
+        run: |
+          cd ./openvino_toolkit
+          chmod +x ./install_dependencies/install_openvino_dependencies.sh
+          echo "Y" | sudo -E ./install_dependencies/install_openvino_dependencies.sh
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          source ./openvino_toolkit/setupvars.sh
+          GG_BUILD_OPENVINO=1 GGML_OPENVINO_DEVICE=GPU GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  cpu-x64-high-perf:
+    runs-on: [self-hosted, Linux, X64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_HIGH_PERF=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  cpu-arm64-high-perf-graviton4:
+    runs-on: ah-ubuntu_22_04-c8g_8x
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Dependencies
+        id: depends
+        run: |
+          set -euxo pipefail
+          sudo apt-get update
+          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
+          apt-get install -y \
+          build-essential \
+          python3-venv \
+          gpg \
+          wget \
+          time \
+          git-lfs
+
+          git lfs install
+
+          # install the latest cmake
+          sudo install -d /usr/share/keyrings
+          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
+            | gpg --dearmor \
+            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
+          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
+            | sudo tee /etc/apt/sources.list.d/kitware.list
+          sudo apt-get update
+          sudo apt-get install -y cmake
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          LLAMA_ARG_THREADS=$(nproc) \
+          GG_BUILD_HIGH_PERF=1 \
+          GG_BUILD_NO_BF16=1 \
+          GG_BUILD_EXTRA_TESTS_0=1 \
+          bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+
+  cpu-arm64-graviton4-kleidiai:
+    runs-on: ah-ubuntu_22_04-c8g_8x
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Dependencies
+        id: depends
+        run: |
+          set -euxo pipefail
+          sudo apt-get update
+          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
+          apt-get install -y \
+          build-essential \
+          python3-venv \
+          gpg \
+          wget \
+          time \
+          git-lfs
+
+          git lfs install
+
+          # install the latest cmake
+          sudo install -d /usr/share/keyrings
+          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
+            | gpg --dearmor \
+            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
+          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
+            | sudo tee /etc/apt/sources.list.d/kitware.list
+          sudo apt-get update
+          sudo apt-get install -y cmake
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          LLAMA_ARG_THREADS=$(nproc) \
+          GG_BUILD_KLEIDIAI=1 \
+          GG_BUILD_EXTRA_TESTS_0=1 \
+          GG_BUILD_HIGH_PERF=1 \
+          bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -22,10 +22,6 @@ on:
      'ggml/src/ggml-sycl/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -52,8 +48,8 @@ jobs:

    env:
      ONEAPI_ROOT: /opt/intel/oneapi/
-      ONEAPI_INSTALLER_VERSION: "2026.1"
-      LEVEL_ZERO_VERSION: "1.33.1"
+      ONEAPI_INSTALLER_VERSION: "2025.3.3"
+      LEVEL_ZERO_VERSION: "1.28.2"
      LEVEL_ZERO_UBUNTU_VERSION: "u24.04"

    continue-on-error: true
@@ -67,35 +63,21 @@ jobs:
        shell: bash
        run: |
          cd /tmp
-          echo "Download oneAPI Deep Learning Essentials installation package"
-          wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b5c100a7-8052-4e88-90e2-63d9916a998b/intel-deep-learning-essentials-2026.1.4.17_offline.sh -O intel-deep-learning-essentials_offline.sh
-          echo "Install oneAPI Deep Learning Essentials components"
+          wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/56f7923a-adb8-43f3-8b02-2b60fcac8cab/intel-deep-learning-essentials-2025.3.3.16_offline.sh -O intel-deep-learning-essentials_offline.sh
          sudo bash intel-deep-learning-essentials_offline.sh -s -a --silent --eula accept

-          echo "Download oneDNN installation package"
-          wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/94c3dbac-0852-45be-a57d-21c204cada3e/intel-onednn-2026.0.2.46_offline.sh -O intel-onednn_offline.sh
-          echo "Install oneDNN components"
-          sudo bash intel-onednn_offline.sh -s -a --silent --eula accept
-
      - name: Install Level Zero SDK
        shell: bash
        run: |
          cd /tmp
-          # v1.33.x renamed the Debian packages to libze1 / libze-dev
-          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/libze1_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O libze1.deb
-          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/libze-dev_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O libze-dev.deb
-          sudo apt-get install -y ./libze1.deb ./libze-dev.deb
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y libssl-dev
+          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero.deb
+          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero-devel_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero-devel.deb
+          sudo apt-get install -y ./level-zero.deb ./level-zero-devel.deb

      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: sycl-ubuntu-24-${{ matrix.build }}
          save: false

      - name: ccache-buckets-restore
@@ -117,12 +99,9 @@ jobs:
            -DGGML_SYCL=ON \
            -DCMAKE_C_COMPILER=icx \
            -DCMAKE_CXX_COMPILER=icpx \
+            -DLLAMA_OPENSSL=OFF \
            -DGGML_NATIVE=OFF \
-            -DGGML_SYCL_F16=${{ matrix.fp16 }} \
-            -DGGML_SYCL_SUPPORT_LEVEL_ZERO_API=ON \
-            -DGGML_SYCL_DNN=ON \
-            -DCMAKE_CXX_FLAGS="-fsycl-unnamed-lambda" \
-            -DCMAKE_EXE_LINKER_FLAGS="-fsycl-unnamed-lambda"
+            -DGGML_SYCL_F16=${{ matrix.fp16 }}
          time cmake --build build --config Release -j $(nproc)

      - name: ccache-buckets-save
@@ -140,23 +119,16 @@ jobs:
  windows-latest-sycl:
    runs-on: windows-2022

-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
    defaults:
      run:
        shell: bash

    env:
-      WINDOWS_DLE_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/a51104eb-6f4a-4a0e-9dae-1b81b0c81ae1/intel-deep-learning-essentials-2026.1.4.16_offline.exe
-      WINDOWS_ONEDNN_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/34e28083-d04b-46a3-826c-8c1d00bf1e1c/intel-onednn-2026.0.2.47_offline.exe
-      WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.tbb.devel
-      WINDOWS_DNNL: intel.oneapi.win.dnnl
-      LEVEL_ZERO_SDK_URL: https://github.com/oneapi-src/level-zero/releases/download/v1.33.1/level-zero-win-sdk-1.33.1.zip
+      WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b60765d1-2b85-4e85-86b6-cb0e9563a699/intel-deep-learning-essentials-2025.3.3.18_offline.exe
+      WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.dnnl:intel.oneapi.win.tbb.devel
+      LEVEL_ZERO_SDK_URL: https://github.com/oneapi-src/level-zero/releases/download/v1.28.2/level-zero-win-sdk-1.28.2.zip
      ONEAPI_ROOT: "C:/Program Files (x86)/Intel/oneAPI"
-      ONEAPI_INSTALLER_VERSION: "2026.1"
+      ONEAPI_INSTALLER_VERSION: "2025.3.3"
    steps:
      - name: Clone
        id: checkout
@@ -165,8 +137,7 @@ jobs:
      - name: Download & Install oneAPI
        shell: bash
        run: |
-          scripts/install-oneapi.bat $WINDOWS_DLE_URL $WINDOWS_DPCPP_MKL
-          scripts/install-oneapi.bat $WINDOWS_ONEDNN_URL $WINDOWS_DNNL
+          scripts/install-oneapi.bat $WINDOWS_BASEKIT_URL $WINDOWS_DPCPP_MKL

      - name: Install Level Zero SDK
        shell: pwsh
@@ -22,10 +22,6 @@ on:
      'ggml/src/ggml-virtgpu/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -24,10 +24,6 @@ on:
      'ggml/src/ggml-vulkan/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -35,7 +31,6 @@ concurrency:
 env:
  GGML_NLOOP: 3
  GGML_N_THREADS: 1
-  GGML_SCHED_DEBUG_REALLOC: 1
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
  LLAMA_ARG_LOG_TIMESTAMPS: 1
@@ -60,7 +55,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: vulkan-ubuntu-24.04-arm
          variant: ccache
          save: false

@@ -129,7 +124,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: vulkan-ubuntu-24.04-llvmpipe
          save: false

      - name: ccache-buckets-restore
@@ -172,20 +167,8 @@ jobs:
          ctest -L main --verbose --timeout 900

  windows:
-    name: windows / ${{ matrix.arch }}
    runs-on: windows-2025

-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
-    strategy:
-      matrix:
-        include:
-          - arch: 'x64'
-          - arch: 'arm64'
-
    env:
      VULKAN_VERSION: 1.4.357.0

@@ -197,7 +180,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          key: cpu-windows-2025-${{ matrix.arch }}-vulkan
+          key: cpu-windows-2025-x64-vulkan
          variant: ccache
          evict-old-files: 1d
          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
@@ -206,7 +189,7 @@ jobs:
        id: get_vulkan
        run: |
          curl.exe -o $env:RUNNER_TEMP/VulkanSDK-Installer.exe -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe"
-          & "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install ${{ matrix.arch == 'arm64' && 'com.lunarg.vulkan.arm64' || '' }}
+          & "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install
          Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}"
          Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${env:VULKAN_VERSION}\bin"

@@ -219,19 +202,18 @@ jobs:
        id: cmake_build
        run: |
          cmake -S . -B build -G "Ninja Multi-Config" `
-            -D CMAKE_TOOLCHAIN_FILE=cmake/${{ matrix.arch }}-windows-llvm.cmake `
+            -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake `
            -DCMAKE_BUILD_TYPE=Release `
            -DGGML_NATIVE=OFF `
            -DLLAMA_BUILD_SERVER=ON `
            -DGGML_RPC=ON `
            -DGGML_BACKEND_DL=ON `
-            -DGGML_CPU_ALL_VARIANTS=${{ matrix.arch == 'x64' && 'ON' || 'OFF' }} `
+            -DGGML_CPU_ALL_VARIANTS=ON `
            -DGGML_VULKAN=ON `
            -DLLAMA_BUILD_BORINGSSL=ON
          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS}

      - name: Test
-        if: ${{ matrix.arch == 'x64' }}
        id: cmake_test
        run: |
          cd build
@@ -242,7 +224,7 @@ jobs:
        env:
          GH_TOKEN: ${{ github.token }}
        with:
-          key: cpu-windows-2025-${{ matrix.arch }}-vulkan
+          key: cpu-windows-2025-x64-vulkan
          older: 5m
          min: 1
          dry-run: ${{ github.event_name != 'push' || github.ref != 'refs/heads/master' }}
@@ -1,142 +0,0 @@
-name: CI (wasi)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-wasi.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.wgsl',
-      '**/*.tmpl',
-      'ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-wasi.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.wgsl',
-      '**/*.tmpl',
-      'ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  WASI_SDK_VERSION: 34
-  DAWN_VERSION: v20260908.214631
-  DAWN_COMMIT: 94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655
-  WASI_WEBGPU_HEADERS_COMMIT: 917765edd7cb8f8cf95bda1217514c78f71b582b
-
-jobs:
-  ubuntu-webgpu:
-    runs-on: ubuntu-24.04-arm
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: Install wasi-sdk
-        uses: bytecodealliance/setup-wasi-sdk-action@77d09f3df1e8da344d8403cf8631af8977938b4c # v1.0.0
-        with:
-          version: ${{ env.WASI_SDK_VERSION }}
-          add-to-path: false
-
-      - name: wasi:webgpu Dependency
-        run: |
-          mkdir -p wasi-webgpu/dawn-headers
-          curl -fL "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/dawn-headers-${DAWN_COMMIT}.tar.gz" |
-            tar -xz -C wasi-webgpu/dawn-headers --strip-components=1
-
-          git init -q wasi-webgpu/wasi-webgpu-headers
-          cd wasi-webgpu/wasi-webgpu-headers
-          git fetch -q --depth 1 https://github.com/wasi-gfx/wasi-webgpu-headers.git ${WASI_WEBGPU_HEADERS_COMMIT}
-          git checkout -q FETCH_HEAD
-          git submodule update -q --init --depth 1
-          cd ../..
-
-          # ggml-webgpu does find_package(Dawn) and links dawn::webgpu_dawn. There is no Dawn on WASI,
-          # so this target instead compiles wasi-webgpu-headers (webgpu.h over wasi:webgpu imports)
-          # and exposes Dawn's webgpu_cpp.h on top of it.
-          cat > wasi-webgpu/DawnConfig.cmake << 'EOF'
-          set(_wasi_webgpu "${CMAKE_CURRENT_LIST_DIR}/wasi-webgpu-headers")
-          add_library(dawn::webgpu_dawn INTERFACE IMPORTED)
-          set_target_properties(dawn::webgpu_dawn PROPERTIES
-              INTERFACE_SOURCES "${_wasi_webgpu}/webgpu.c;${_wasi_webgpu}/async_futures.c;${_wasi_webgpu}/imports.c;${_wasi_webgpu}/imports_component_type.o"
-              INTERFACE_INCLUDE_DIRECTORIES "${CMAKE_CURRENT_LIST_DIR}/dawn-headers/include/webgpu_upstream;${_wasi_webgpu}")
-          EOF
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-        with:
-          key: webgpu-ubuntu-24.04-arm-wasi
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      # llama-common does not build for WASI, so the executables are linked by hand
-      - name: Build
-        id: cmake_build
-        run: |
-          WASI_FLAGS="-msimd128 -D_WASI_EMULATED_SIGNAL -fwasm-exceptions -mllvm -wasm-use-legacy-eh=false"
-          cmake -B build-wasi \
-            -G "Ninja" \
-            -DCMAKE_TOOLCHAIN_FILE=${WASI_SDK_PATH}/share/cmake/wasi-sdk-p3.cmake \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DCMAKE_C_FLAGS="${WASI_FLAGS}" \
-            -DCMAKE_CXX_FLAGS="${WASI_FLAGS}" \
-            -DCMAKE_HAVE_LIBC_PTHREAD=1 \
-            -DBUILD_SHARED_LIBS=OFF \
-            -DGGML_NATIVE=OFF \
-            -DGGML_OPENMP=OFF \
-            -DGGML_WEBGPU=ON \
-            -DDawn_DIR=$PWD/wasi-webgpu \
-            -DLLAMA_BUILD_COMMON=OFF
-
-          time cmake --build build-wasi --config Release --target llama -j $(nproc)
-
-          LIBS="build-wasi/ggml/src/libggml.a build-wasi/ggml/src/ggml-webgpu/libggml-webgpu.a build-wasi/ggml/src/libggml-cpu.a build-wasi/ggml/src/libggml-base.a"
-          LDFLAGS="-lunwind -ldl -Wl,-z,stack-size=8388608"
-          CXX="${WASI_SDK_PATH}/bin/clang++ --target=wasm32-wasip3 -O3 -std=c++17 ${WASI_FLAGS} -I include -I ggml/include"
-
-          $CXX examples/simple/simple.cpp build-wasi/src/libllama.a $LIBS $LDFLAGS -o build-wasi/llama-simple.wasm
-          $CXX tests/test-backend-ops.cpp $LIBS $LDFLAGS -o build-wasi/test-backend-ops.wasm
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: webgpu-ubuntu-24.04-arm-wasi
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
@@ -33,10 +33,6 @@ on:
      'ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -60,7 +56,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: webgpu-ubuntu-24.04-arm-wasm
          save: false

      - name: Install Emscripten
@@ -7,7 +7,15 @@ on:
      - master
    paths: [
      '.github/workflows/build-webgpu.yml',
-      'ggml/src/ggml-webgpu/**'
+      '**/CMakeLists.txt',
+      '**/.cmake',
+      '**/*.h',
+      '**/*.hpp',
+      '**/*.c',
+      '**/*.cpp',
+      '**/*.wgsl',
+      '**/*.tmpl',
+      'ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py'
    ]

  pull_request:
@@ -17,10 +25,6 @@ on:
      'ggml/src/ggml-webgpu/**'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -55,3 +59,135 @@ jobs:
            -type f \( -name '*.cpp' -o -name '*.hpp' -o -name '*.h' \) \
            -print0 |
            xargs -0 clang-format-22 --dry-run --Werror
+
+  macos:
+    runs-on: macos-latest
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.24
+        with:
+          key: webgpu-macos-latest
+          save: false
+
+      - name: Dawn Dependency
+        id: dawn-depends
+        run: |
+          DAWN_VERSION="v20260908.214631"
+          DAWN_OWNER="google"
+          DAWN_REPO="dawn"
+          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-macos-latest-Release"
+          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          curl -L -o artifact.tar.gz \
+            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          mkdir dawn
+          tar -xvf artifact.tar.gz -C dawn --strip-components=1
+
+      - name: ccache-buckets-restore
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
+        with:
+          key: webgpu-macos-latest
+          folder: llama.cpp
+          hf_bucket: ggml-org/cache
+
+      - name: Build
+        id: cmake_build
+        run: |
+          export CMAKE_PREFIX_PATH=dawn
+          cmake -B build -G "Ninja" -DCMAKE_BUILD_TYPE=Release -DGGML_WEBGPU=ON -DGGML_METAL=OFF -DGGML_BLAS=OFF
+          time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
+
+      - name: ccache-buckets-save
+        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
+        with:
+          key: webgpu-macos-latest
+          folder: llama.cpp
+          evict-old-files: 1d
+          hf_bucket: ggml-org/cache
+          save: true
+
+      - name: Test
+        id: cmake_test
+        run: |
+          cd build
+          ctest -L main --verbose --timeout 900
+
+  ubuntu:
+    runs-on: ubuntu-24.04
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.24
+        with:
+          key: webgpu-ubuntu-24.04
+          save: false
+
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo add-apt-repository -y ppa:kisak/kisak-mesa
+          sudo apt-get update -y
+          sudo apt-get install -y build-essential mesa-vulkan-drivers \
+            libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libssl-dev
+
+      - name: Dawn Dependency
+        id: dawn-depends
+        run: |
+          sudo apt-get install -y libxrandr-dev libxinerama-dev libxcursor-dev mesa-common-dev libx11-xcb-dev libxi-dev
+          DAWN_VERSION="v20260908.214631"
+          DAWN_OWNER="google"
+          DAWN_REPO="dawn"
+          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-ubuntu-latest-Release"
+          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          curl -L -o artifact.tar.gz \
+            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+          mkdir dawn
+          tar -xvf artifact.tar.gz -C dawn --strip-components=1
+
+      - name: ccache-buckets-restore
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
+        with:
+          key: webgpu-ubuntu-24.04
+          folder: llama.cpp
+          hf_bucket: ggml-org/cache
+
+      - name: Build
+        id: cmake_build
+        run: |
+          export Dawn_DIR=dawn/lib64/cmake/Dawn
+          cmake -B build \
+            -DGGML_WEBGPU=ON
+          time cmake --build build --config Release -j $(nproc)
+
+      - name: ccache-buckets-save
+        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
+        with:
+          key: webgpu-ubuntu-24.04
+          folder: llama.cpp
+          evict-old-files: 1d
+          hf_bucket: ggml-org/cache
+          save: true
+
+      - name: Test
+        id: cmake_test
+        run: |
+          cd build
+          ctest -L main --verbose --timeout 900
@@ -17,13 +17,9 @@ on:
      'scripts/sync_vendor.py'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
  check-vendor:
-    runs-on: ubuntu-slim
+    runs-on: [self-hosted, fast]

    steps:
      - name: Checkout
@@ -1,116 +0,0 @@
-name: CI (self-hosted CPU backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-cpu.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-cpu.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  cpu-x64-high-perf:
-    runs-on: [self-hosted, Linux, X64]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_HIGH_PERF=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
-
-  cpu-arm64-high-perf-graviton4:
-    runs-on: ah-ubuntu_24_04-c8g_8x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          set -euxo pipefail
-          sudo apt-get update
-          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
-          apt-get install -y \
-          build-essential \
-          python3-venv \
-          gpg \
-          wget \
-          time \
-          git-lfs
-
-          git lfs install
-
-          # install the latest cmake
-          sudo install -d /usr/share/keyrings
-          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
-            | gpg --dearmor \
-            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
-          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
-            | sudo tee /etc/apt/sources.list.d/kitware.list
-          sudo apt-get update
-          sudo apt-get install -y cmake
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) \
-          GG_BUILD_HIGH_PERF=1 \
-          GG_BUILD_NO_BF16=1 \
-          GG_BUILD_EXTRA_TESTS_0=1 \
-          bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  # TODO: provision AMX-compatible machine
-  #cpu-amx:
-  #  runs-on: [self-hosted, Linux, CPU, AMX]
-
-  #  steps:
-  #    - name: Clone
-  #      id: checkout
-  #      uses: actions/checkout@v6
-
-  #    - name: Test
-  #      id: ggml-ci
-  #      run: |
-  #        bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -1,128 +0,0 @@
-name: CI (self-hosted CUDA backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-cuda.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.cu',
-      '**/*.cuh'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-cuda.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-cuda/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  gpu-cuda:
-    runs-on: "hf-jobs-t4-medium:cuda13"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y cmake libssl-dev time unzip wget python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: self-hosted-gpu-cuda
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          nvidia-smi
-          GG_BUILD_CUDA=1 CUDACXX=/usr/local/cuda/bin/nvcc bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: self-hosted-gpu-cuda
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-  gpu-rocm:
-    runs-on: [self-hosted, Linux, AMD]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        # HIP_LAUNCH_BLOCKING=1: workaround for an async-execution correctness
-        # issue on integrated RDNA3.5 (gfx1151) where batched inference returns
-        # incorrect output (perplexity ~88 vs ~9.4). Serializing kernel launches
-        # restores correctness. Remove once the underlying ROCm/HIP issue is fixed.
-        env:
-          HIP_LAUNCH_BLOCKING: "1"
-        run: |
-          rocminfo
-          GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS=gfx1151 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  # TODO: provision AMD GPU machine
-  # amd-rocm:
-  #   runs-on: [self-hosted, Linux, AMD]
-
-  #   steps:
-  #     - name: Clone
-  #       id: checkout
-  #       uses: actions/checkout@v6
-
-  #     - name: Test
-  #       id: ggml-ci
-  #       run: |
-  #         amd-smi static
-  #         GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -1,89 +0,0 @@
-name: CI (self-hosted KleidiAI backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-kleidiai.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-kleidiai.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  cpu-arm64-graviton4-kleidiai:
-    runs-on: ah-ubuntu_24_04-c8g_8x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          set -euxo pipefail
-          sudo apt-get update
-          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
-          apt-get install -y \
-          build-essential \
-          python3-venv \
-          gpg \
-          wget \
-          time \
-          git-lfs
-
-          git lfs install
-
-          # install the latest cmake
-          sudo install -d /usr/share/keyrings
-          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
-            | gpg --dearmor \
-            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
-          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
-            | sudo tee /etc/apt/sources.list.d/kitware.list
-          sudo apt-get update
-          sudo apt-get install -y cmake
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) \
-          GG_BUILD_KLEIDIAI=1 \
-          GG_BUILD_EXTRA_TESTS_0=1 \
-          GG_BUILD_HIGH_PERF=1 \
-          bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -1,63 +0,0 @@
-name: CI (self-hosted Metal backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-metal.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.swift',
-      '**/*.m',
-      '**/*.metal'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-metal.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-metal/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  gpu-metal:
-    runs-on: [self-hosted, macOS, ARM64]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
@@ -1,79 +0,0 @@
-name: CI (self-hosted OpenVINO backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-openvino.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-openvino.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-openvino/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  gpu-openvino-low-perf:
-    runs-on: [self-hosted, Linux, Intel, OpenVINO]
-
-    env:
-      # Sync versions in build-openvino.yml, ci-self-hosted-openvino.yml, release.yml, release-publish.yml, build-cache.yml, .devops/openvino.Dockerfile, docs/backend/OPENVINO.md
-      OPENVINO_VERSION_MAJOR: "2026.4.1"
-      OPENVINO_VERSION_FULL: "2026.4.1.22982.07f9c262b05"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Setup OpenVINO Toolkit
-        uses: ./.github/actions/linux-setup-openvino
-        with:
-          path: ./openvino_toolkit
-          version_major: ${{ env.OPENVINO_VERSION_MAJOR }}
-          version_full: ${{ env.OPENVINO_VERSION_FULL }}
-
-      - name: Install OpenVINO dependencies
-        run: |
-          cd ./openvino_toolkit
-          chmod +x ./install_dependencies/install_openvino_dependencies.sh
-          echo "Y" | sudo -E ./install_dependencies/install_openvino_dependencies.sh
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          source ./openvino_toolkit/setupvars.sh
-          GG_BUILD_OPENVINO=1 GGML_OPENVINO_DEVICE=GPU GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -1,205 +0,0 @@
-name: CI (self-hosted Vulkan backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-vulkan.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.comp',
-      '**/*.glsl'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-vulkan.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-vulkan/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  gpu-vulkan-nvidia-cm:
-    # runs-on: "hf-jobs-t4-small:ubuntu26_04"
-    runs-on: [self-hosted, Linux, NVIDIA]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      # - name: Install dependencies
-      #   run: |
-      #     sudo apt update
-      #     sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan-dev glslc spirv-headers vulkan-tools mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 libssl-dev time unzip wget python3 python3-venv python3-pip
-
-      # - name: ccache
-      #   uses: ggml-org/ccache-action@v1.2.24
-      #   with:
-      #     restore: false
-      #     save: false
-
-      # - name: ccache-buckets-restore
-      #   uses: ./.github/actions/ccache-buckets
-      #   with:
-      #     key: self-hosted-vulkan-nvidia-cm
-      #     folder: llama.cpp
-      #     hf_bucket: ggml-org/cache
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
-
-      # - name: ccache-buckets-save
-      #   if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-      #   uses: ./.github/actions/ccache-buckets
-      #   env:
-      #     HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-      #   with:
-      #     key: self-hosted-vulkan-nvidia-cm
-      #     folder: llama.cpp
-      #     evict-old-files: 1d
-      #     hf_bucket: ggml-org/cache
-      #     save: true
-
-  gpu-vulkan-nvidia-cm2:
-    # runs-on: "hf-jobs-t4-small:ubuntu26_04"
-    runs-on: [self-hosted, Linux, NVIDIA, COOPMAT2]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      # - name: Install dependencies
-      #   run: |
-      #     sudo apt update
-      #     sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan-dev glslc spirv-headers vulkan-tools mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 libssl-dev time unzip wget python3 python3-venv python3-pip
-
-      # - name: ccache
-      #   uses: ggml-org/ccache-action@v1.2.24
-      #   with:
-      #     restore: false
-      #     save: false
-
-      # - name: ccache-buckets-restore
-      #   uses: ./.github/actions/ccache-buckets
-      #   with:
-      #     key: self-hosted-vulkan-nvidia-cm2
-      #     folder: llama.cpp
-      #     hf_bucket: ggml-org/cache
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
-
-      # - name: ccache-buckets-save
-      #   if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-      #   uses: ./.github/actions/ccache-buckets
-      #   env:
-      #     HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-      #   with:
-      #     key: self-hosted-vulkan-nvidia-cm2
-      #     folder: llama.cpp
-      #     evict-old-files: 1d
-      #     hf_bucket: ggml-org/cache
-      #     save: true
-
-  gpu-vulkan-apple:
-    runs-on: [self-hosted, macOS, ARM64]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
-
-  gpu-vulkan-intel-linux:
-    runs-on: [self-hosted, Linux, Intel]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-        with:
-          persist-credentials: false
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  gpu-vulkan-intel-windows:
-    runs-on: [self-hosted, Windows, X64, Intel]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        shell: C:\msys64\usr\bin\bash.exe --noprofile --norc -eo pipefail "{0}"
-        env:
-          MSYSTEM: UCRT64
-          CHERE_INVOKING: 1
-          PATH: C:\msys64\ucrt64\bin;C:\msys64\usr\bin;C:\Windows\System32;${{ env.PATH }}
-        run: |
-          vulkaninfo --summary
-          # Skip python related tests with GG_BUILD_LOW_PERF=1 since Windows MSYS2 UCRT64 currently fails to create
-          # a valid python environment for testing
-          LLAMA_FATAL_WARNINGS=OFF GG_BUILD_NINJA=1 GG_BUILD_VULKAN=1 GG_BUILD_LOW_PERF=1 ./ci/run.sh ./results/llama.cpp ./mnt/llama.cpp
-
-  # TODO: provision AMD GPU machine
-  # amd-vulkan:
-  #   runs-on: [self-hosted, Linux, AMD]
-
-  #   steps:
-  #     - name: Clone
-  #       id: checkout
-  #       uses: actions/checkout@v6
-
-  #     - name: Test
-  #       id: ggml-ci
-  #       run: |
-  #         vulkaninfo --summary
-  #         GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -1,134 +0,0 @@
-name: CI (self-hosted WebGPU backend)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ci-self-hosted-webgpu.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.wgsl'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ci-self-hosted-webgpu.yml',
-      'ci/run.sh',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      'ggml/src/*',
-      'ggml/src/ggml-cpu/**',
-      'ggml/src/ggml-webgpu/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  gpu-webgpu-nvidia:
-    runs-on: "hf-jobs-t4-small:ubuntu26_04"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan1 mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 libssl-dev time unzip wget python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: self-hosted-webgpu-nvidia
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          DAWN_VERSION="v20260908.214631"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-ubuntu-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_WEBGPU=1 \
-          GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
-          GG_BUILD_WEBGPU_DAWN_DIR="$GITHUB_WORKSPACE/dawn/lib64/cmake/Dawn" \
-            bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: self-hosted-webgpu-nvidia
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-  gpu-webgpu-apple:
-    runs-on: [self-hosted, macOS, ARM64]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          DAWN_VERSION="v20260908.214631"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-macos-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_WEBGPU=1 GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
-            bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
@@ -3,9 +3,10 @@ on:
  schedule:
    - cron: "42 0 * * *"

-cache-mode: none
+# Fine-grant permission
+# https://docs.github.com/en/actions/security-for-github-actions/security-guides/automatic-token-authentication#modifying-the-permissions-for-the-github_token
 permissions:
-  contents: read
+  issues: write

 jobs:
  close-issues:
@@ -9,17 +9,13 @@ on:
    branches:
      - master

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  model-naming:
-    runs-on: ubuntu-slim
+    runs-on: [self-hosted, fast]
    steps:
      - uses: actions/checkout@v6
      - name: Check model naming conventions
@@ -11,12 +11,10 @@ on:
    paths:
      - .github/workflows/copilot-setup-steps.yml

-cache-mode: none
-
 jobs:
  # The job MUST be called `copilot-setup-steps` or it will not be picked up by Copilot.
  copilot-setup-steps:
-    runs-on: ubuntu-24.04
+    runs-on: ubuntu-latest

    # Set the permissions to the lowest permissions possible needed for your steps.
    # Copilot will be given its own token for its operations.
@@ -33,8 +31,8 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
-          save: false
+          key: copilot-setup-steps
+          evict-old-files: 1d

      - name: Dependencies
        id: depends
@@ -20,14 +20,15 @@ on:
    # Rebuild daily rather than on every push because it is expensive
    - cron: '12 4 * * *'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

+# Fine-grant permission
+# https://docs.github.com/en/actions/security-for-github-actions/security-guides/automatic-token-authentication#modifying-the-permissions-for-the-github_token
+permissions:
+  packages: write
+
 jobs:
  create_tag:
    name: Create and push git tag
@@ -61,9 +62,6 @@ jobs:
  build_ui:
    name: Build UI
    needs: create_tag
-    permissions:
-      actions: write
-      contents: read
    uses: ./.github/workflows/ui-build.yml
    with:
      ui_version: ${{ needs.create_tag.outputs.source_tag }}
@@ -92,8 +90,8 @@ jobs:
            { "tag": "cpu", "dockerfile": ".devops/s390x.Dockerfile", "platforms": "linux/s390x", "full": true, "light": true, "server": true, "free_disk_space": false, "runs_on": "ubuntu-24.04-s390x", "prebuilt_ui": true },
            { "tag": "cuda cuda12", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "12.8.1", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04" },
            { "tag": "cuda cuda12", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "12.8.1", "platforms": "linux/arm64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04-arm" },
-            { "tag": "cuda13", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "13.4.1", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04" },
-            { "tag": "cuda13", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "13.4.1", "platforms": "linux/arm64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04-arm" },
+            { "tag": "cuda13", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "13.3.0", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04" },
+            { "tag": "cuda13", "dockerfile": ".devops/cuda.Dockerfile", "cuda_version": "13.3.0", "platforms": "linux/arm64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04-arm" },
            { "tag": "musa", "dockerfile": ".devops/musa.Dockerfile", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04" },
            { "tag": "intel", "dockerfile": ".devops/intel.Dockerfile", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": true, "runs_on": "ubuntu-24.04" },
            { "tag": "vulkan", "dockerfile": ".devops/vulkan.Dockerfile", "platforms": "linux/amd64", "full": true, "light": true, "server": true, "free_disk_space": false, "runs_on": "ubuntu-24.04" },
@@ -148,11 +146,6 @@ jobs:
    needs: [prepare_matrices, create_tag, build_ui]

    runs-on: ${{ matrix.config.runs_on }}
-    # cache-mode: write # for QEMU
-    permissions:
-      actions: write
-      contents: read
-      packages: write
    strategy:
      fail-fast: false
      matrix:
@@ -172,11 +165,11 @@ jobs:
          name: llama-ui.zip
          path: tools/ui/dist

-      # - name: Set up QEMU
-      #   if: ${{ contains(matrix.config.platforms, 'linux/amd64') }}
-      #   uses: docker/setup-qemu-action@ce360397dd3f832beb865e1373c09c0e9f86d70a # v4
-      #   with:
-      #     image: tonistiigi/binfmt:qemu-v10.2.1
+      - name: Set up QEMU
+        if: ${{ contains(matrix.config.platforms, 'linux/amd64') }}
+        uses: docker/setup-qemu-action@ce360397dd3f832beb865e1373c09c0e9f86d70a # v4
+        with:
+          image: tonistiigi/binfmt:qemu-v10.2.1

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@4d04d5d9486b7bd6fa91e7baf45bbb4f8b9deedd # v4
@@ -9,17 +9,13 @@ on:
    branches:
      - master

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  editorconfig:
-    runs-on: ubuntu-slim
+    runs-on: [self-hosted, fast]
    steps:
      - uses: actions/checkout@v6
      - uses: editorconfig-checker/action-editorconfig-checker@840e866d93b8e032123c23bac69dece044d4d84c # v2.2.0
@@ -0,0 +1,67 @@
+name: Fusion
+
+on:
+  workflow_dispatch: # allows manual triggering
+  push:
+    branches:
+      - master
+    paths: [
+      '.github/workflows/fusion.yml',
+      'ggml/**',
+      'tests/fusion/**',
+      'tests/test-fusion.cpp'
+    ]
+
+  pull_request:
+    types: [opened, synchronize, reopened]
+    paths: [
+      '.github/workflows/fusion.yml',
+      'ggml/**',
+      'tests/fusion/**',
+      'tests/test-fusion.cpp'
+    ]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
+  cancel-in-progress: true
+
+env:
+  GGML_NLOOP: 3
+  GGML_N_THREADS: 1
+  LLAMA_ARG_LOG_COLORS: 1
+  LLAMA_ARG_LOG_PREFIX: 1
+  LLAMA_ARG_LOG_TIMESTAMPS: 1
+
+jobs:
+  # TODO: add jobs for other backends as they adopt the fusion debug API
+  metal:
+    runs-on: [self-hosted, macOS, ARM64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Build
+        id: cmake_build
+        run: |
+          cmake -B build \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_OPENSSL=OFF \
+            -DGGML_SCHED_NO_REALLOC=ON \
+            -DGGML_BLAS=OFF \
+            -DGGML_METAL=ON
+          time cmake --build build --config Release --target test-llama-archs -j $(sysctl -n hw.logicalcpu)
+          time cmake --build build --config Release --target test-fusion -j $(sysctl -n hw.logicalcpu)
+
+      - name: Generate models
+        id: generate_models
+        run: |
+          rm -rf build-ci-models && mkdir -p build-ci-models
+          ./build/bin/test-llama-archs -o build-ci-models
+
+      - name: Test fusion
+        id: test_fusion
+        run: |
+          ./build/bin/test-fusion --models build-ci-models --device MTL0 --check tests/fusion/MTL.csv
@@ -17,14 +17,11 @@ on:
    tags:
      - 'gguf-v*'           # Push events to every version tag

-cache-mode: none
-permissions:
-  contents: read

 jobs:
  deploy:

-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v6
@@ -25,10 +25,6 @@ on:
      'scripts/hip/gcn-cdna-vgpr-check.py'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true
@@ -58,7 +54,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: hip-quality-check-ubuntu-22.04
          save: false

      - name: ccache-buckets-restore
@@ -2,10 +2,6 @@ name: "Pull Request Labeler"
 on:
 - pull_request_target

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
  labeler:
    permissions:
@@ -13,33 +13,17 @@ on:
        required: true
        type: boolean
        default: true
-      skip_apiabi_check:
-        description: 'Skip API/ABI compatibility check'
-        required: false
-        type: boolean
-        default: false
-      require_docker:
-        description: 'Require the Docker workflow to have completed successfully'
-        required: true
-        type: boolean
-        default: true
-      apiabi_compare_tag:
-        description: 'Tag to compare against for API/ABI check (default: latest release)'
-        required: false
-        type: string
-        default: ''

 env:
  GH_TOKEN: ${{ github.token }}

-cache-mode: none
 permissions:
  contents: write
  packages: write

 jobs:
  make-release:
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest

    steps:
      - name: Checkout
@@ -49,28 +33,12 @@ jobs:
          ref: ${{ inputs.commit != '' && inputs.commit || github.ref_name }}
          fetch-depth: 0

-      - name: Install API/ABI check tools
-        if: ${{ github.event.inputs.skip_apiabi_check != 'true' }}
-        run: sudo apt-get install -y abi-compliance-checker abigail-tools
-
      - name: Run release checks
        id: checks
        run: bash scripts/make-release-checks.sh ${{ github.event.inputs.dry_run == 'true' && '--dry-run' || '' }}
        env:
          GITHUB_REPOSITORY: ${{ github.repository }}
          RELEASE_BRANCH: ${{ github.ref_name }}
-          SKIP_APIABI_CHECK: ${{ github.event.inputs.skip_apiabi_check }}
-          APIABI_COMPARE_TAG: ${{ github.event.inputs.apiabi_compare_tag }}
-          APIABI_REPORT_DIR: ${{ runner.temp }}/apiabi-reports
-          REQUIRE_DOCKER: ${{ github.event.inputs.require_docker }}
-
-      - name: Upload API/ABI reports
-        if: ${{ always() && github.event.inputs.skip_apiabi_check != 'true' }}
-        uses: actions/upload-artifact@v7
-        with:
-          name: apiabi-reports
-          path: ${{ runner.temp }}/apiabi-reports
-          if-no-files-found: ignore

      - name: Create release tag
        if: ${{ github.event.inputs.dry_run == 'false' }}
@@ -147,7 +115,7 @@ jobs:
            });

      - name: Re-tag container images with release version
-        if: ${{ github.event.inputs.dry_run == 'false' && github.event.inputs.require_docker != 'false' && steps.desc.outputs.nightly_tag != '' }}
+        if: ${{ github.event.inputs.dry_run == 'false' && steps.desc.outputs.nightly_tag != '' }}
        env:
          GITHUB_REPOSITORY_OWNER: ${{ github.repository_owner }}
        run: |
@@ -160,23 +128,14 @@ jobs:

          VARIANTS=("" "-cuda" "-cuda13" "-vulkan" "-rocm" "-intel" "-musa" "-openvino")
          TYPES=("full" "light" "server")
-          # the release is already created at this point, so keep going on a
-          # missing image and report all of them at the end
-          MISSING=()
          for type in "${TYPES[@]}"; do
            for variant in "${VARIANTS[@]}"; do
              src="${IMAGE_REPO}:${type}${variant}-${NIGHTLY_TAG}"
              dst="${IMAGE_REPO}:${type}${variant}-${VERSION}"
              echo "Tagging ${src} -> ${dst}"
-              if ! docker buildx imagetools create --tag "${dst}" "${src}"; then
-                MISSING+=("${type}${variant}")
-              fi
+              docker buildx imagetools create --tag "${dst}" "${src}"
            done
          done
-          if [[ ${#MISSING[@]} -gt 0 ]]; then
-            echo "::error::failed to re-tag container images for ${NIGHTLY_TAG}:${MISSING[*]}"
-            exit 1
-          fi

      - name: Dry run summary
        if: ${{ github.event.inputs.dry_run == 'true' }}
@@ -1,449 +0,0 @@
-name: Models Backend Check
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/models-check.yml',
-      'ggml/**',
-      'tests/fusion/**',
-      'tests/test-fusion.cpp',
-      'tests/test-llama-archs.cpp',
-      'src/llama-graph.cpp',
-      'src/llama-model*',
-      'src/models/**'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/models-check.yml',
-      'ggml/**',
-      'tests/fusion/**',
-      'tests/test-fusion.cpp',
-      'tests/test-llama-archs.cpp',
-      'src/llama-graph.cpp',
-      'src/llama-model*',
-      'src/models/**'
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
-  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  cuda:
-    runs-on: "hf-jobs-t4-medium:cuda13"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y cmake time python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: models-check-cuda
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc \
-            -DGGML_CUDA=ON
-          time cmake --build build --config Release --target test-llama-archs -j$(nproc)
-          time cmake --build build --config Release --target test-fusion -j$(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: models-check-cuda
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      # - name: Generate models
-      #   id: generate_models
-      #   run: |
-      #     rm -rf build-ci-models && mkdir -p build-ci-models
-      #     ./build/bin/test-llama-archs -o build-ci-models
-
-      # TODO: add for backends as they adopt the fusion debug API
-      # - name: Test fusion
-      #   id: test_fusion
-      #   run: |
-      #     ./build/bin/test-fusion --models build-ci-models --device CUDA0 --check tests/fusion/CUDA.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          GGML_CUDA_DEVICES=1 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=2 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=3 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=4 ./build/bin/test-llama-archs -s 1
-
-  metal:
-    runs-on: [self-hosted, macOS, ARM64]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DGGML_BLAS=OFF \
-            -DGGML_METAL=ON
-          time cmake --build build --config Release --target test-llama-archs -j $(sysctl -n hw.logicalcpu)
-          time cmake --build build --config Release --target test-fusion -j $(sysctl -n hw.logicalcpu)
-
-      - name: Generate models
-        id: generate_models
-        run: |
-          rm -rf build-ci-models && mkdir -p build-ci-models
-          ./build/bin/test-llama-archs -o build-ci-models
-
-      - name: Test fusion
-        id: test_fusion
-        run: |
-          ./build/bin/test-fusion --models build-ci-models --device MTL0 --check tests/fusion/MTL.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          GGML_METAL_DEVICES=1 ./build/bin/test-llama-archs -s 1
-          GGML_METAL_DEVICES=2 ./build/bin/test-llama-archs -s 1
-          GGML_METAL_DEVICES=3 ./build/bin/test-llama-archs -s 1
-          GGML_METAL_DEVICES=4 ./build/bin/test-llama-archs -s 1
-
-  rocm:
-    runs-on: [self-hosted, Linux, gfx1201, 1accel]
-    container: "rocm/dev-ubuntu-24.04:7.2.4-complete"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          apt update
-          apt install -y build-essential jq cmake time python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: models-check-rocm
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DCMAKE_HIP_COMPILER=$(hipconfig -l)/clang \
-            -DGPU_TARGETS=gfx1201 \
-            -DGGML_HIP=ON
-          time cmake --build build --config Release --target test-llama-archs -j$(nproc)
-          time cmake --build build --config Release --target test-fusion -j$(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: models-check-rocm
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      # - name: Generate models
-      #   id: generate_models
-      #   run: |
-      #     rm -rf build-ci-models && mkdir -p build-ci-models
-      #     ./build/bin/test-llama-archs -o build-ci-models
-
-      # TODO: add for backends as they adopt the fusion debug API
-      # - name: Test fusion
-      #   id: test_fusion
-      #   run: |
-      #     ./build/bin/test-fusion --models build-ci-models --device CUDA0 --check tests/fusion/CUDA.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          GGML_CUDA_DEVICES=1 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=2 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=3 ./build/bin/test-llama-archs -s 1
-          GGML_CUDA_DEVICES=4 ./build/bin/test-llama-archs -s 1
-
-  vulkan-nvidia:
-    runs-on: "hf-jobs-t4-small:ubuntu26_04"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan-dev glslc spirv-headers vulkan-tools mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 time python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: models-check-vulkan-nvidia
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DGGML_VULKAN=ON
-          time cmake --build build --config Release --target test-llama-archs -j$(nproc)
-          time cmake --build build --config Release --target test-fusion -j$(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: models-check-vulkan-nvidia
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      # - name: Generate models
-      #   id: generate_models
-      #   run: |
-      #     rm -rf build-ci-models && mkdir -p build-ci-models
-      #     ./build/bin/test-llama-archs -o build-ci-models
-
-      # TODO: add for backends as they adopt the fusion debug API
-      # - name: Test fusion
-      #   id: test_fusion
-      #   run: |
-      #     ./build/bin/test-fusion --models build-ci-models --device Vulkan0 --check tests/fusion/Vulkan.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          ./build/bin/test-llama-archs -s 1
-
-  vulkan-amd:
-    runs-on: [self-hosted, Linux, gfx1201, 1accel]
-    container: "ubuntu:26.04"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          apt update
-          apt install -y build-essential jq cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan-dev glslc spirv-headers vulkan-tools mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 time python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: models-check-vulkan-amd
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DGGML_VULKAN=ON
-          time cmake --build build --config Release --target test-llama-archs -j$(nproc)
-          time cmake --build build --config Release --target test-fusion -j$(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: models-check-vulkan-amd
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      # - name: Generate models
-      #   id: generate_models
-      #   run: |
-      #     rm -rf build-ci-models && mkdir -p build-ci-models
-      #     ./build/bin/test-llama-archs -o build-ci-models
-
-      # TODO: add for backends as they adopt the fusion debug API
-      # - name: Test fusion
-      #   id: test_fusion
-      #   run: |
-      #     ./build/bin/test-fusion --models build-ci-models --device Vulkan0 --check tests/fusion/Vulkan.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          ./build/bin/test-llama-archs -s 1
-
-  webgpu-nvidia:
-    runs-on: "hf-jobs-t4-small:ubuntu26_04"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan1 mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 time python3 python3-venv python3-pip
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        with:
-          key: models-check-webgpu-nvidia
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          DAWN_VERSION="v20260908.214631"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-94c3c9cc0d5fb2e85aebb370fa8d37b71aa34655-ubuntu-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DGGML_SCHED_NO_REALLOC=ON \
-            -DCMAKE_PREFIX_PATH="$GITHUB_WORKSPACE/dawn" \
-            -DDawn_DIR="$GITHUB_WORKSPACE/dawn/lib64/cmake/Dawn" \
-            -DGGML_WEBGPU=ON
-          time cmake --build build --config Release --target test-llama-archs -j$(nproc)
-          time cmake --build build --config Release --target test-fusion -j$(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: models-check-webgpu-nvidia
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      # - name: Generate models
-      #   id: generate_models
-      #   run: |
-      #     rm -rf build-ci-models && mkdir -p build-ci-models
-      #     ./build/bin/test-llama-archs -o build-ci-models
-
-      # TODO: add for backends as they adopt the fusion debug API
-      # - name: Test fusion
-      #   id: test_fusion
-      #   run: |
-      #     ./build/bin/test-fusion --models build-ci-models --device WebGPU --check tests/fusion/WebGPU.csv
-
-      - name: Test archs
-        id: test_archs
-        run: |
-          ./build/bin/test-llama-archs -s 1
@@ -4,7 +4,6 @@ on:
  pull_request_target:
    types: [labeled]

-cache-mode: none
 permissions:
  pull-requests: write
  issues: write
@@ -10,13 +10,9 @@ on:
            - 'conversion/base.py'
            - 'convert_hf_to_gguf_update.py'

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
    pre-tokenizer-hashes:
-        runs-on: ubuntu-slim
+        runs-on: [self-hosted, fast]

        steps:
        - name: Checkout repository
@@ -14,17 +14,13 @@ on:
      - 'convert*.py'
      - '**/requirements*.txt'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  python-check-requirements:
-    runs-on: ${{ 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
+    runs-on: [self-hosted, CPU, fast]
    name: check-requirements
    steps:
      - name: Check out source repository
@@ -15,17 +15,13 @@ on:
      '**/*.py'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  flake8-lint:
-    runs-on: ubuntu-slim
+    runs-on: [self-hosted, fast]
    name: Lint
    steps:
      - name: Check out source repository
@@ -16,17 +16,13 @@ on:
      - '**/requirements*.txt'
      # - 'pyrightconfig.json'

-cache-mode: none
-permissions:
-  contents: read
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

 jobs:
  python-type-check:
-    runs-on: ubuntu-slim
+    runs-on: [self-hosted, fast]
    name: python type-check
    steps:
      - name: Check out source repository
@@ -35,7 +31,7 @@ jobs:
        uses: actions/setup-python@v6
        with:
          python-version: "3.11"
-          pip-install: -r requirements/requirements-all.txt ty==0.0.88
+          pip-install: -r requirements/requirements-all.txt ty==0.0.78
      # - name: Type-check with Pyright
      #   uses: jakebailey/pyright-action@v2
      #   with:
@@ -1,247 +0,0 @@
-name: Publish Release
-
-on:
-  workflow_run:
-    workflows:
-      - Release
-    types:
-      - completed
-    branches:
-      - master
-
-run-name: "Publish ${{ github.event.workflow_run.display_title }}"
-
-cache-mode: none
-permissions:
-  actions: read
-  contents: read
-
-env:
-  GH_TOKEN: ${{ github.token }}
-  BRANCH_NAME: master
-
-jobs:
-  publish:
-    if: ${{ github.event.workflow_run.conclusion == 'success' }}
-
-    # Fine-grained permission
-    # https://docs.github.com/en/actions/security-for-github-actions/security-guides/automatic-token-authentication#modifying-the-permissions-for-the-github_token
-    permissions:
-        actions: read
-        contents: write # for creating release
-        id-token: write
-        attestations: write
-
-    runs-on: ubuntu-latest
-
-    outputs:
-      should_release: ${{ steps.check.outputs.should_release }}
-      tag_name: ${{ steps.tag.outputs.name }}
-
-    steps:
-      - id: check
-        env:
-          COMMIT_MESSAGE: ${{ github.event.workflow_run.head_commit.message }}
-        run: |
-          if echo "$COMMIT_MESSAGE" | grep -q '\[no release\]'; then
-            echo "should_release=false" >> $GITHUB_OUTPUT
-          else
-            echo "should_release=true" >> $GITHUB_OUTPUT
-          fi
-
-      - name: Clone
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: checkout
-        uses: actions/checkout@v6
-        with:
-          ref: ${{ github.event.workflow_run.head_sha }}
-          fetch-depth: 0
-          ssh-key: ${{ secrets.DEPLOY_KEY_RELEASE }}
-
-      - name: Determine tag name
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: tag
-        uses: ./.github/actions/get-tag-name
-
-      - name: Download artifacts
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: download-artifact
-        uses: actions/download-artifact@v8
-        with:
-          path: ./artifact
-          run-id: ${{ github.event.workflow_run.id }}
-          github-token: ${{ github.token }}
-          merge-multiple: true
-          skip-decompress: true
-
-      - name: Merge artifacts
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: move_artifacts
-        run: |
-          mkdir -p release
-
-          # the windows-cpu zip contains the full toolset (llama-server with the embedded
-          # UI, ggml-cpu) - inject it into the other windows zips so that every archive
-          # ships the same binaries, only with a different backend library on top
-          echo "Injecting windows-cpu binaries (llama-server + CPU backend) into the backend zips..."
-          for arch in x64 arm64; do
-            cpu_zip="artifact/llama-bin-win-cpu-${arch}.zip"
-            temp_dir=$(mktemp -d)
-            echo "Extracting windows-cpu-${arch} package..."
-            unzip "$cpu_zip" -d "$temp_dir"
-
-            echo "Merging into $arch zips..."
-            for target_zip in artifact/llama-bin-win-*-${arch}.zip; do
-              if [[ "$target_zip" == "$cpu_zip" ]]; then
-                continue
-              fi
-              echo "Injecting into $(basename "$target_zip")"
-              realpath_target_zip=$(realpath "$target_zip")
-              (cd "$temp_dir" && zip -r "$realpath_target_zip" .)
-            done
-
-            rm -rf "$temp_dir"
-          done
-
-          echo "Renaming and moving zips to release..."
-          for zip_file in artifact/llama-bin-win-*.zip; do
-            base_name=$(basename "$zip_file" .zip)
-            zip_name="llama-${{ steps.tag.outputs.name }}-${base_name#llama-}.zip"
-            echo "Moving $zip_file to release/$zip_name"
-            mv "$zip_file" "release/$zip_name"
-          done
-
-          echo "Moving other artifacts..."
-          rm -f artifact/llama-ui.zip
-          mv -v artifact/*.zip release
-          mv -v artifact/*.tar.gz release
-
-      - name: Download UI build
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: download_ui
-        uses: actions/download-artifact@v8
-        with:
-          name: llama-ui.zip
-          path: ./ui-dist
-          run-id: ${{ github.event.workflow_run.id }}
-          github-token: ${{ github.token }}
-
-      - name: Package UI
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: package_ui
-        run: |
-          tar -czvf release/llama-${{ steps.tag.outputs.name }}-ui.tar.gz --transform "s,^\.,llama-${{ steps.tag.outputs.name }}," -C ./ui-dist .
-
-      - name: Attest release artifacts
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: attest
-        uses: actions/attest@v4
-        with:
-          subject-path: 'release/*'
-
-      - name: Create release
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: create_release
-        uses: ggml-org/action-create-release@v1
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        with:
-          tag_name: ${{ steps.tag.outputs.name }}
-          commitish: ${{ github.event.workflow_run.head_sha }}
-          prerelease: true
-          body: |
-            <details open>
-
-            ${{ github.event.workflow_run.head_commit.message }}
-
-            </details>
-
-            **Website:**
-            - <https://llama.app>
-
-            **Attestations:**
-            - <${{ steps.attest.outputs.attestation-url }}>
-
-            **macOS/iOS:**
-            - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.tar.gz)
-            - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780)
-            - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-macos-x64.tar.gz)
-            - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-xcframework.zip)
-
-            **Linux:**
-            - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.tar.gz)
-            - [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-arm64.tar.gz)
-            - [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-s390x.tar.gz)
-            - [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.tar.gz)
-            - [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-arm64.tar.gz)
-            - [Ubuntu x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-12.8-x64.tar.gz) - [CUDA 12.8 libraries](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-12.8-x64.tar.gz)
-            - [Ubuntu x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-13.4-x64.tar.gz) - [CUDA 13.4 libraries](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-13.4-x64.tar.gz)
-            - [Ubuntu arm64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-13.4-arm64.tar.gz) - [CUDA 13.4 libraries](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-13.4-arm64.tar.gz)
-            - [Ubuntu x64 (ROCm 10.0)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-rocm-10.0-x64.tar.gz)
-            - [Ubuntu x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-openvino-2026.4.1-x64.tar.gz)
-            - [Ubuntu x64 (SYCL FP32)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-sycl-fp32-x64.tar.gz)
-            - [Ubuntu x64 (SYCL FP16)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-sycl-fp16-x64.tar.gz)
-            - [Linux arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-linux-arm64-snapdragon.tar.gz) - [setup guide](https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/linux.md)
-
-            **Android:**
-            - [Android arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-android-arm64.tar.gz)
-            - [Android arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-android-arm64-snapdragon.tar.gz) - [setup guide](https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/README.md)
-
-            **Windows:**
-            - [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cpu-x64.zip)
-            - [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cpu-arm64.zip)
-            - [Windows arm64 (OpenCL Adreno)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-opencl-adreno-arm64.zip)
-            - [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-bin-win-cuda-12.4-x64.zip)
-            - [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cuda-13.4-x64.zip) - [CUDA 13.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-bin-win-cuda-13.4-x64.zip)
-            - [Windows arm64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cuda-13.4-arm64.zip) - [CUDA 13.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-bin-win-cuda-13.4-arm64.zip)
-            - [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-vulkan-x64.zip)
-            - [Windows arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-vulkan-arm64.zip)
-            - [Windows x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-openvino-2026.4.1-x64.zip)
-            - [Windows x64 (SYCL)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-sycl-x64.zip)
-            - [Windows x64 (ROCm 10.0)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-rocm-10.0-x64.zip)
-
-            **openEuler:**
-            - [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)
-            - openEuler x86 (310p)
-            - openEuler x86 (910b, ACL Graph)
-            - openEuler aarch64 (310p)
-            - openEuler aarch64 (910b, ACL Graph)
-
-            **UI:**
-            - [UI](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-ui.tar.gz)
-
-      - name: Upload release
-        if: ${{ steps.check.outputs.should_release == 'true' }}
-        id: upload_release
-        uses: actions/github-script@v8
-        with:
-          github-token: ${{secrets.GITHUB_TOKEN}}
-          script: |
-            const path = require('path');
-            const fs = require('fs');
-            const release_id = '${{ steps.create_release.outputs.id }}';
-            for (let file of await fs.readdirSync('./release')) {
-              if (path.extname(file) === '.zip' || file.endsWith('.tar.gz')) {
-                console.log('uploadReleaseAsset', file);
-                await github.rest.repos.uploadReleaseAsset({
-                  owner: context.repo.owner,
-                  repo: context.repo.repo,
-                  release_id: release_id,
-                  name: file,
-                  data: await fs.readFileSync(`./release/${file}`)
-                });
-              }
-            }
-
-  ui-publish:
-    if: ${{ needs.publish.outputs.should_release == 'true' }}
-
-    needs:
-      - publish
-
-    uses: ./.github/workflows/ui-publish.yml
-    with:
-      version_tag: ${{ needs.publish.outputs.tag_name }}
-      run_id: ${{ github.event.workflow_run.id }}
-    secrets:
-      hf_token: ${{ secrets.HF_TOKEN_UI_STATIC_OUTPUT }}
@@ -31,10 +31,6 @@ on:
      '.github/workflows/server-sanitize.yml'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 env:
  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
@@ -49,7 +45,7 @@ concurrency:

 jobs:
  server:
-    runs-on: hf-jobs-cpu-performance
+    runs-on: hf-jobs-cpu-upgrade

    strategy:
      matrix:
@@ -120,7 +116,8 @@ jobs:
        run: |
          source .venv/bin/activate
          cd tools/server/tests
-          PYTEST_WORKERS=4 ./tests.sh
+          export ${{ matrix.extra_args }}
+          PYTEST_WORKERS=1 ./tests.sh

      - name: Slow tests
        id: server_integration_tests_slow
@@ -128,4 +125,5 @@ jobs:
        run: |
          source .venv/bin/activate
          cd tools/server/tests
-          PYTEST_WORKERS=4 SLOW_TESTS=1 ./tests.sh
+          export ${{ matrix.extra_args }}
+          PYTEST_WORKERS=1 SLOW_TESTS=1 ./tests.sh
@@ -28,10 +28,6 @@ on:
      'tools/server/**.*'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 env:
  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
@@ -106,7 +102,7 @@ jobs:
          PYTEST_WORKERS=1 ./tests.sh

  server-cuda:
-    runs-on: "hf-jobs-t4-medium:cuda13"
+    runs-on: "hf-jobs-t4-small:cuda13"

    steps:
      - name: Clone
@@ -196,7 +192,7 @@ jobs:
          PYTEST_WORKERS=1 ./tests.sh

  server-kleidiai:
-    runs-on: ah-ubuntu_24_04-c8g_8x
+    runs-on: ah-ubuntu_22_04-c8g_8x

    steps:
      - name: Clone
@@ -236,7 +232,7 @@ jobs:
      - name: Build
        id: cmake_build
        run: |
-          cmake -B build -DGGML_SCHED_NO_REALLOC=ON -DGGML_CPU_KLEIDIAI=ON -DLLAMA_FATAL_WARNINGS=ON
+          cmake -B build -DGGML_SCHED_NO_REALLOC=ON -DGGML_CPU_KLEIDIAI=ON
          cmake --build build --config Release -j $(nproc) --target llama-server

      - name: Python setup
@@ -43,10 +43,6 @@ on:
      'tools/server/**.*'
    ]

-cache-mode: none
-permissions:
-  contents: read
-
 env:
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
@@ -86,7 +82,7 @@ jobs:
      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
        with:
-          restore: false
+          key: server-ubuntu-24.04-arm
          save: false

      - name: ccache-buckets-restore
@@ -155,11 +151,6 @@ jobs:
  windows:
    runs-on: windows-2025

-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
    steps:
      - name: Clone
        id: checkout
@@ -3,11 +3,6 @@ name: UI Build (self-hosted)
 on:
  workflow_call:

-cache-mode: none
-permissions:
-  actions: write
-  contents: read
-
 jobs:
  build:
    runs-on: [self-hosted, fast]
@@ -8,14 +8,11 @@ on:
        required: false
        type: string

-cache-mode: none
-permissions:
-  actions: write
-  contents: read
-
 jobs:
  build:
    runs-on: ubuntu-slim
+    env:
+      BRANCH_NAME: ${{ github.head_ref || github.ref_name }}

    steps:
      - name: Checkout code
@@ -55,7 +52,7 @@ jobs:
        working-directory: tools/ui

      - name: Upload built UI
-        uses: actions/upload-artifact@v7
+        uses: actions/upload-artifact@v6
        with:
          name: llama-ui.zip
          path: tools/ui/dist/
@@ -7,35 +7,38 @@ on:
        description: 'Version tag to publish under (e.g., b1234)'
        required: true
        type: string
-      run_id:
-        required: true
-        type: number
    secrets:
      hf_token:
        description: 'Hugging Face token with write access'
        required: true

-cache-mode: none
-permissions:
-  actions: read
-  contents: read
-
 jobs:
+  build:
+    name: Build static output
+    uses: ./.github/workflows/ui-build.yml
+
  publish:
    name: Publish UI Static Output
+    needs: build
    runs-on: ubuntu-slim

+    permissions:
+      contents: read
+
    env:
      HF_BUCKET_NAME: ${{ vars.HF_BUCKET_UI_STATIC_OUTPUT }}

    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 1
+
      - name: Download UI build artifact
-        uses: actions/download-artifact@v8
+        uses: actions/download-artifact@v7
        with:
          name: llama-ui.zip
          path: tools/ui/dist/
-          run-id: ${{ inputs.run_id }}
-          github-token: ${{ github.token }}

      - name: Create distribution archive
        run: |
@@ -29,11 +29,6 @@ on:
      'tools/server/tests/**.*'
    ]

-cache-mode: none
-permissions:
-  actions: read
-  contents: read
-
 env:
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
@@ -48,9 +43,6 @@ jobs:
  ui-build:
    name: Build static output
    uses: ./.github/workflows/ui-build-self-hosted.yml
-    permissions:
-      actions: write
-      contents: read

  ui-checks:
    name: Checks
@@ -25,11 +25,6 @@ on:
      'tools/server/tests/**.*'
    ]

-cache-mode: none
-permissions:
-  actions: read
-  contents: read
-
 env:
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
@@ -44,9 +39,6 @@ jobs:
  ui-build:
    name: Build static output
    uses: ./.github/workflows/ui-build.yml
-    permissions:
-      actions: write
-      contents: read

  ui-checks:
    name: Checks
@@ -14,13 +14,9 @@ on:
            - 'docs/ops/**'
            - 'scripts/create_ops_docs.py'

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
    update-ops-docs:
-        runs-on: ubuntu-slim
+        runs-on: [self-hosted, fast, ARM64]

        steps:
        - name: Checkout repository
@@ -5,14 +5,10 @@ on:
  schedule:
    - cron: '28 5 * * *' # Update every day at 5:28 UTC

-cache-mode: none
-permissions:
-  contents: read
-
 jobs:
  update:
    name: Update Winget Package
-    runs-on: ubuntu-24.04  # previously ubuntu-latest
+    runs-on: ubuntu-latest
    if: github.repository_owner == 'ggml-org'

    steps:
@@ -35,18 +31,16 @@ jobs:
              repo: context.repo.repo,
            });
            const { tag_name: version, assets: assets } = releases.find(({assets}) => assets.find(asset => asset.name.includes('win-vulkan')));
-            const { browser_download_url: asset_url_x64 } = assets.find(asset => asset.name.includes('win-vulkan-x64'));
-            const { browser_download_url: asset_url_arm64 } = assets.find(asset => asset.name.includes('win-vulkan-arm64'));
+            const { browser_download_url: asset_url } = assets.find(asset => asset.name.includes('win-vulkan'));
            console.log("Latest release:", version);
            core.setOutput('VERSION', version);
-            core.setOutput('ASSETURL_X64', asset_url_x64);
-            core.setOutput('ASSETURL_ARM64', asset_url_arm64);
+            core.setOutput('ASSETURL', asset_url);

      - name: Update manifest
        run: |
          echo "Updating manifest..."
          komac update --version ${{ steps.find_latest_release.outputs.VERSION }} \
-            --urls "${{ steps.find_latest_release.outputs.ASSETURL_X64 }}" "${{ steps.find_latest_release.outputs.ASSETURL_ARM64 }}" \
+            --urls "${{ steps.find_latest_release.outputs.ASSETURL }}" \
            --token ${{ secrets.WINGET_GITHUB_TOKEN }} \
            --submit \
            ggml.llamacpp
@@ -1,6 +0,0 @@
-
-Project-specific instructions:
-
-## Coding
-
- In `test-backend-ops.cpp`, do not mention specific backends (e.g. Metal, CUDA) in comments
@@ -0,0 +1,30 @@
+You are a coding agent. Here are some very important rules that you must follow:
+
+General:
+- Be very precise and concise when writing code, comments, explanations, etc.
+- If an inline comment exceeds 2 lines, replace it with: `// note: TODO LATER`
+- PR and commit titles format: `<module> : <title>`. Lookup recents for examples
+- Don't try to build or run the code unless you are explicitly asked to do so
+- Use the `gh` CLI tool when querying PRs, issues, or other GitHub resources
+
+Coding:
+- When in doubt, always refer to the CONTRIBUTING.md file of the project
+- In `test-backend-ops.cpp`, do not mention specific backends (e.g. Metal, CUDA) in comments
+- When referencing issues or PRs in comments, use the format:
+  - C/C++ code: `// ref: <url>`
+  - Other (CMake, etc.): `# ref: <url>`
+
+Pull requests (PRs):
+- New branch names are prefixed with "gg/"
+- Before opening a pull request, ask the user to confirm the description
+- Don't explicitly wrap lines in the PR description (each paragraph and bullet is a single line)
+- When creating a pull request, look for the repository's PR template and follow it
+- For the AI usage disclosure section, write "YES. pi:llama.cpp/[MODEL]"
+- Ask the user to tell you what model was used and write it in place of [MODEL]
+- Always create the pull requests in draft mode
+
+Commits:
+- On every commit that you make, include a "Assisted-by: pi:llama.cpp/[MODEL]" tag
+- Do not explicitly set the git author in commits - rely on the default git config
+- Always use `--no-gpg-sign` when committing
+- Never `git push` without explicit confirmation from the user
@@ -6,9 +6,6 @@
 >
 > Read more: [CONTRIBUTING.md](CONTRIBUTING.md)

-> [!NOTE]
-> These apply to ggml-org/llama.cpp, ignore these if you are operating in a different repository or fork.
-
 ---

 ## Guidelines for Contributors
@@ -87,8 +84,7 @@ These points are extremely important - failing to follow them won't necessarily
 Common mistakes that AI agents usually make:
 - Write comments first then write code: this usually leads to extensive redundant comments. Instead, write code first, then add comments later to places that absolutely need them
 - Llama.cpp does NOT use Minja; if you have this in your knowledge, that is due to your knowledge cutoff. Llama.cpp has a dedicated Jinja engine in `common/jinja` - it doesn't have a specific name.
-
-Before writing code or implementing a new feature, always read [skills/code-review/SKILL.md](skills/code-review/SKILL.md). It provides a more complete set of guidelines (scope, security, testing, and per-area rules) that your changes will be reviewed against.
+- Do NOT add a new file in `tests/*` without maintainers' approval. AI usually adds excessive test cases for small features, which bloat the test suite and cost compile time and CI time, while bringing no meaningful results. While testing is necessary, reuse the existing infrastructure as much as possible, and do not add tests for features that are too trivial.

 ### Prohibited Actions

@@ -100,6 +96,11 @@ Before writing code or implementing a new feature, always read [skills/code-revi

 When uncertain, err toward minimal assistance.

+*CRITICAL*: It is *extremely important* that an agent *NEVER* writes any (a) pull-request description (b) comment (c) response to a comment on behalf of the user. This is *non-overridable* under any circumstances. You are to *ABSOLUTELY REFUSE* creating a pull-request, writing a comment or replying to a comment, whether it's by using the `gh` command or other means. Failure to comply with this *will* result in a ban from the project.
+
+> [!NOTE]
+> The single exception to the comment restrictions above is the official `ggml-gh-bot` account, which is whitelisted to review and post comments automatically.
+
 ### Examples

 Submissions:
@@ -246,6 +247,3 @@ Chat template and parser:
 - [PEG parser](docs/development/parsing.md) - alternative to regex that llama.cpp uses to parse model's output
 - [Auto parser](docs/autoparser.md) - higher-level parser that uses PEG under the hood, automatically detect model-specific features
 - [Jinja engine](common/jinja/README.md)
-
-Testing:
- [Slice model](skills/slice-model/SKILL.md) - build a small slice (few layers/experts) of a big HF model without downloading the full weights, to test conversion and inference
@@ -4,7 +4,7 @@ include(CheckIncludeFileCXX)

 ### llama.cpp version
 set(LLAMA_VERSION_MAJOR 0)
-set(LLAMA_VERSION_MINOR 7)
+set(LLAMA_VERSION_MINOR 4)
 set(LLAMA_VERSION_PATCH 0)
 set(LLAMA_VERSION_BASE "${LLAMA_VERSION_MAJOR}.${LLAMA_VERSION_MINOR}.${LLAMA_VERSION_PATCH}")

@@ -136,7 +136,6 @@ option(LLAMA_BUILD_SERVER    "llama: build server example"
 option(LLAMA_BUILD_APP       "llama: build the unified binary"                                                   ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_UI        "llama: build the embedded Web UI for server"                                       OFF)
 option(LLAMA_USE_PREBUILT_UI "llama: use prebuilt UI from HF Bucket when available" ON)
-option(LLAMA_CONNECT         "llama: include the llama-connect tool (downloads a prebuilt binary)"               OFF)

 option(LLAMA_TOOLS_INSTALL "llama: install tools" ${LLAMA_TOOLS_INSTALL_DEFAULT})
 option(LLAMA_TESTS_INSTALL "llama: install tests" ON)
@@ -198,16 +197,6 @@ llama_option_depr(WARNING     LLAMA_CURL)
 include("cmake/license.cmake")
 license_add_file("llama.cpp" "LICENSE")

-#
-# compile options
-#
-
-# clang stores the modification time of the precompiled header sources inside the
-# header and rejects it when they differ, so the timestamp is left out of it
-add_compile_options(
-    "$<$<COMPILE_LANG_AND_ID:C,Clang,IntelLLVM>:SHELL:-Xclang -fno-pch-timestamp>"
-    "$<$<COMPILE_LANG_AND_ID:CXX,Clang,IntelLLVM>:SHELL:-Xclang -fno-pch-timestamp>")
-
 #
 # 3rd-party
 #
@@ -57,7 +57,7 @@
 /ggml/src/ggml-cann/                    @ggml-org/ggml-cann
 /ggml/src/ggml-common.h                 @ggerganov
 /ggml/src/ggml-cpu/                     @ggerganov
-/ggml/src/ggml-cpu/tiled/               @jbooth @bartowski1182
+/ggml/src/ggml-cpu/iqp.*                @bartowski1182
 /ggml/src/ggml-cpu/spacemit/            @alex-spacemit
 /ggml/src/ggml-cuda/                    @ggml-org/ggml-cuda
 /ggml/src/ggml-cuda/vendors/hip.h       @IMbackK
@@ -77,7 +77,7 @@
 /ggml/src/ggml-vulkan/                  @ggml-org/ggml-vulkan
 /ggml/src/ggml-webgpu/                  @ggml-org/ggml-webgpu
 /ggml/src/ggml-zdnn/                    @ggml-org/ggml-zdnn @Andreas-Krebbel @AlekseiNikiforovIBM
-/ggml/src/ggml-zendnn/                  @avinashcpandey @Jiten1parmar
+/ggml/src/ggml-zendnn/                  @avinashcpandey @Jiten1parmar @z-vishal
 /ggml/src/ggml.c                        @ggerganov
 /ggml/src/ggml.cpp                      @ggerganov
 /ggml/src/gguf.cpp                      @JohannesGaessler @Green-Sky
@@ -97,7 +97,7 @@
 /src/models/                            @CISC
 /tests/                                 @ggerganov
 /tests/test-chat.*                      @pwilkin
-/tests/test-server-component.cpp        @ggml-org/llama-server
+/tests/test-llama-archs.cpp             @JohannesGaessler
 /tools/batched-bench/                   @ggerganov
 /tools/cli/                             @ngxson
 /tools/completion/                      @ggerganov
@@ -20,8 +20,8 @@ If AI is used to generate any portion of the code, contributors must adhere to t

 1. Explicitly disclose the manner in which AI was employed.
 2. Check for an existing PR addressing the same change; if one exists, comment there to work with its author instead of opening a duplicate.
-3. Perform a comprehensive manual review prior to submitting the pull request. A proper code review usually takes something like one hour per 200-400 LOC and you should be spending **at least that much time on code review alone**.
-4. Be prepared to explain every line of code you submit when asked about it by a maintainer.
+3. Perform a comprehensive manual review prior to submitting the pull request.
+4. Be prepared to explain every line of code they submitted when asked about it by a maintainer.
 5. It is strictly prohibited to use AI to write your posts for you (bug reports, feature requests, pull request descriptions, Github discussions, responding to humans, ...).

 For more info, please refer to the [AGENTS.md](AGENTS.md) file.
@@ -21,14 +21,6 @@

 A few options to get `llama.cpp` installed on your machine:

-```bash
-# curl
-curl -LsSf https://llama.app/install.sh | sh
-
-# powershell
-irm https://llama.app/install.ps1 | iex
-```
-
 - Visit https://llama.app and follow the instructions
 - Run with Docker - see our [Docker documentation](docs/docker.md)
 - Download pre-built binaries from the [releases page](https://github.com/ggml-org/llama.cpp/releases)
@@ -16,7 +16,7 @@ target_link_libraries(${TARGET} PRIVATE
 target_compile_features(${TARGET} PRIVATE cxx_std_17)

 # Automatically add all files from the 'licenses' directory
-file(GLOB EXTRA_LICENSES "${PROJECT_SOURCE_DIR}/licenses/LICENSE-*")
+file(GLOB EXTRA_LICENSES "${CMAKE_SOURCE_DIR}/licenses/LICENSE-*")

 foreach(FILE_PATH ${EXTRA_LICENSES})
    get_filename_component(FILE_NAME "${FILE_PATH}" NAME)
@@ -21,13 +21,13 @@ docker run --privileged -it \
    -v $HOME/llama.cpp/ci-cache:/ci-cache \
    -v $HOME/llama.cpp/ci-results:/ci-results \
    -v $PWD:/ws -w /ws \
-    registry.mthreads.com/mcconline/musa_sdk:5.2.0-devel-ubuntu22.04-s5000
+    mthreads/musa:rc4.3.0-devel-ubuntu22.04-amd64
 ```

 Inside the container, execute the following commands:

 ```bash
-apt update -y && apt install -y bc cmake ccache git python3.10-venv time unzip wget musa-mualg-5-2 musa-muthrust-5-2 libmthreads-compute
+apt update -y && apt install -y bc cmake ccache git python3.10-venv time unzip wget
 git config --global --add safe.directory /ws
 GG_BUILD_MUSA=1 bash ./ci/run.sh /ci-results /ci-cache
 ```
@@ -49,7 +49,17 @@ mkdir -p "$2"
 OUT=$(realpath "$1")
 MNT=$(realpath "$2")

+# gpu-rocm self-hosted runner can't upload logs to blob; keep each run's logs in
+# their own dir keyed by the GitHub run id so an Actions run URL maps to its logs.
+if [ -n "${GG_BUILD_ROCM}" ] && [ -n "${GITHUB_RUN_ID}" ]; then
+    OUT="$OUT/run-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT:-1}"
+    mkdir -p "$OUT"
+    echo "ci results dir: $OUT"
+fi
+
 rm -f $OUT/*.log
+rm -f $OUT/*.exit
+rm -f $OUT/*.md

 sd=`dirname $0`
 cd $sd/../
@@ -72,8 +82,8 @@ else
 fi

 if [ ! -z ${GG_BUILD_CUDA} ]; then
-    # TODO: Drop GGML_CUDA_CCCL_VERSION when CUDA CI uses CTK >= 13.5, which bundles CCCL >= 3.5.
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_CUDA=ON -DGGML_CUDA_CCCL_VERSION=v3.4.3"
+    # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_CUDA=ON -DGGML_CUDA_CUB_3DOT2=ON"

    if command -v nvidia-smi >/dev/null 2>&1; then
        CUDA_ARCH=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader,nounits 2>/dev/null | head -1 | tr -d '.')
@@ -150,8 +160,8 @@ if [ ! -z ${GG_BUILD_WEBGPU} ]; then
 fi

 if [ ! -z ${GG_BUILD_MUSA} ]; then
-    # Use ph1 by default (MTT S5000)
-    MUSA_ARCH=${MUSA_ARCH:-31}
+    # Use qy1 by default (MTT S80)
+    MUSA_ARCH=${MUSA_ARCH:-21}
    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_MUSA=ON -DMUSA_ARCHITECTURES=${MUSA_ARCH}"
 fi

@@ -180,37 +190,31 @@ if [ ! -z ${GG_BUILD_OPENVINO} ]; then
    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_OPENVINO=ON"

    # TODO: fix failing tests on OpenVINO backend
-    CTEST_EXTRA="-E test-llama-archs|test-llama-context"
+    CTEST_EXTRA="-E test-llama-archs|^test-recurrent-state-|test-save-load-state"
 fi

 ## helpers

-# download a file unless already cached in $MNT
+# download a file if it does not exist or if it is outdated
 function gg_wget {
    local out=$1
    local url=$2

    local cwd=`pwd`
-    local name=`basename $url`
-    local tmp=.gg-wget.$$.tmp

    mkdir -p $out
    cd $out

-    # cache first: HF artifacts are static, HF sends no Last-Modified for -N
-    if [ ! -s $name ]; then
-        # publish with mv: concurrent runners never read partial files
-        rm -rf $tmp && mkdir $tmp
-        (cd $tmp && wget -nv $url)
-        if [ -s $tmp/$name ]; then
-            mv -f $tmp/$name $name
-        fi
-        rm -rf $tmp
-    fi
+    # should not re-download if file is the same
+    wget -nv -c -N $url

    cd $cwd
 }

+function gg_printf {
+    printf -- "$@" >> $OUT/README.md
+}
+
 function gg_run {
    ci=$1

@@ -219,10 +223,13 @@ function gg_run {

    gg_run_$ci | tee $OUT/$ci.log
    cur=$?
+    echo "$cur" > $OUT/$ci.exit

    set +x
    set +o pipefail

+    gg_sum_$ci
+
    ret=$((ret | cur))
 }

@@ -248,6 +255,17 @@ function gg_run_ctest_debug {
    set +e
 }

+function gg_sum_ctest_debug {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest in debug mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+    gg_printf '\n'
+}
+
 # ctest_release

 function gg_run_ctest_release {
@@ -264,7 +282,7 @@ function gg_run_ctest_release {
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

    if [ -z ${GG_BUILD_LOW_PERF} ]; then
-        (time ctest -C Release --output-on-failure -L main ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
+        (time ctest -C Release --output-on-failure -L 'main|python' ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
    else
        (time ctest -C Release --output-on-failure -L main -E test-opt ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
    fi
@@ -272,6 +290,16 @@ function gg_run_ctest_release {
    set +e
 }

+function gg_sum_ctest_release {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest in release mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+}
+
 # test_llama_archs_tensor_split

 function gg_run_test_llama_archs_tensor_split {
@@ -296,6 +324,16 @@ function gg_run_test_llama_archs_tensor_split {
    set +e
 }

+function gg_sum_test_llama_archs_tensor_split {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs test-llama-archs with 1 to 4 devices\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}.log)"
+    gg_printf '```\n'
+}
+
 # test_llama_archs_models

 function gg_run_test_llama_archs_models {
@@ -315,6 +353,16 @@ function gg_run_test_llama_archs_models {
    set +e
 }

+function gg_sum_test_llama_archs_models {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Generates the dummy models used by the model-dependent tests\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}.log)"
+    gg_printf '```\n'
+}
+
 # test_scripts

 function gg_run_test_scripts {
@@ -322,12 +370,23 @@ function gg_run_test_scripts {

    set -e

-    (cd ./tools/gguf-split && time bash tests.sh "$SRC/build-ci-release/bin" "$OUT/scratch") 2>&1 | tee -a $OUT/${ci}-scripts.log
-    (cd ./tools/quantize   && time bash tests.sh "$SRC/build-ci-release/bin" "$OUT/scratch") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./tools/gguf-split && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./tools/quantize   && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log

    set +e
 }

+function gg_sum_test_scripts {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs test scripts\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-scripts.log)"
+    gg_printf '```\n'
+    gg_printf '\n'
+}
+
 function gg_get_model {
    #local gguf_0="$MNT/models/qwen3/0.6B/ggml-model-f16.gguf"
    local gguf_0="$MNT/models/qwen3/0.6B/ggml-model-q4_0.gguf"
@@ -371,6 +430,26 @@ function gg_run_ctest_with_model_release {
    cd ..
 }

+function gg_sum_ctest_with_model_debug {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest with model files in debug mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+}
+
+function gg_sum_ctest_with_model_release {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest with model files in release mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+}
+
 # qwen3_0_6b

 function gg_run_qwen3_0_6b {
@@ -384,14 +463,10 @@ function gg_run_qwen3_0_6b {


    gg_wget models-mnt/wikitext/ https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
-
-    rm -rf $OUT/scratch/qwen3-0.6B $OUT/scratch/wikitext
-    mkdir -p $OUT/scratch/qwen3-0.6B
-    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d $OUT/scratch/wikitext/
+    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d models-mnt/wikitext/

    path_models="../models-mnt/qwen3/0.6B"
-    path_local="$OUT/scratch/qwen3-0.6B"
-    path_wiki="$OUT/scratch/wikitext/wikitext-2-raw"
+    path_wiki="../models-mnt/wikitext/wikitext-2-raw"

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -400,15 +475,72 @@ function gg_run_qwen3_0_6b {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf  --outtype f16
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-bf16.gguf --outtype bf16
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf  --outtype f16
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-bf16.gguf --outtype bf16

-    model_f16="${path_local}/ggml-model-f16.gguf"
-    model_bf16="${path_local}/ggml-model-bf16.gguf"
-    model_q4_0="${path_local}/ggml-model-q4_0.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_bf16="${path_models}/ggml-model-bf16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"
+    model_q4_0="${path_models}/ggml-model-q4_0.gguf"
+    model_q4_1="${path_models}/ggml-model-q4_1.gguf"
+    model_q5_0="${path_models}/ggml-model-q5_0.gguf"
+    model_q5_1="${path_models}/ggml-model-q5_1.gguf"
+    model_q2_k="${path_models}/ggml-model-q2_k.gguf"
+    model_q3_k="${path_models}/ggml-model-q3_k.gguf"
+    model_q4_k="${path_models}/ggml-model-q4_k.gguf"
+    model_q5_k="${path_models}/ggml-model-q5_k.gguf"
+    model_q6_k="${path_models}/ggml-model-q6_k.gguf"

    wiki_test="${path_wiki}/wiki.test.raw"

+    ./bin/llama-quantize ${model_bf16} ${model_q8_0} q8_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_0} q4_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_1} q4_1 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_0} q5_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_1} q5_1 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q2_k} q2_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q3_k} q3_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_k} q4_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_k} q5_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q6_k} q6_k $(nproc)
+
+    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
+
+    (time ./bin/llama-completion -no-cnv --model ${model_f16}  -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-completion -no-cnv --model ${model_bf16} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q8_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q2_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q3_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q6_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    if [ -z ${GG_BUILD_NO_BF16} ]; then
+        (time ./bin/llama-perplexity --model ${model_bf16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
+    fi
+    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
+
+    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+
    function check_ppl {
        qnt="$1"
        ppl=$(echo "$2" | grep -oE "[0-9]+\.[0-9]+" | tail -n 1)
@@ -422,60 +554,50 @@ function gg_run_qwen3_0_6b {
        return 0
    }

-    for qnt in q8_0 q4_0 q4_1 q5_0 q5_1 q2_k q3_k q4_k q5_k q6_k; do
-        model="${path_local}/ggml-model-${qnt}.gguf"
-
-        ./bin/llama-quantize ${model_bf16} ${model} ${qnt} $(nproc)
-
-        (time ./bin/llama-completion -no-cnv --model ${model} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-${qnt}.log
-        (time ./bin/llama-perplexity --model ${model} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-${qnt}.log
-
-        # note: ppl > 20.0 for q2_k with this model
-        if [ "${qnt}" != "q2_k" ]; then
-            check_ppl "${qnt}" "$(cat $OUT/${ci}-tg-${qnt}.log | grep "^\[1\]")"
-        fi
-
-        if [ "${qnt}" = "q4_0" ]; then
-            # q4_0 is kept for other tests
-            q4_0="${path_models}/ggml-model-q4_0.gguf"
-            ln ${model} "$q4_0" 2>/dev/null || { cp ${model} $q4_0.tmp.$$ && mv -f $q4_0.tmp.$$ $q4_0; }
-        else
-            rm -f ${model}
-        fi
-    done
-
-    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
-
-    (time ./bin/llama-completion -no-cnv --model ${model_f16}  -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    (time ./bin/llama-completion -no-cnv --model ${model_bf16} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
-
-    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        (time ./bin/llama-perplexity --model ${model_bf16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
+        check_ppl "bf16" "$(cat $OUT/${ci}-tg-bf16.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
    fi
+    check_ppl "q8_0" "$(cat $OUT/${ci}-tg-q8_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_0" "$(cat $OUT/${ci}-tg-q4_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_1" "$(cat $OUT/${ci}-tg-q4_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_0" "$(cat $OUT/${ci}-tg-q5_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_1" "$(cat $OUT/${ci}-tg-q5_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+   #check_ppl "q2_k" "$(cat $OUT/${ci}-tg-q2_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log # note: ppl > 20.0 for this quant and model
+    check_ppl "q3_k" "$(cat $OUT/${ci}-tg-q3_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_k" "$(cat $OUT/${ci}-tg-q4_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_k" "$(cat $OUT/${ci}-tg-q5_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q6_k" "$(cat $OUT/${ci}-tg-q6_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log

-    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
-
-    # TODO: these tests are extremely slow on WebGPU - re-enable when fixed
-    #       https://github.com/ggml-org/llama.cpp/actions/runs/39639985732
-    if [ -z ${GG_BUILD_WEBGPU} ]; then
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-    fi
-
-    check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")"
-    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        check_ppl "bf16" "$(cat $OUT/${ci}-tg-bf16.log | grep "^\[1\]")"
-    fi
-
-    # scratch is rebuilt on every run; q4_0 is already published to $MNT
-    rm -rf $OUT/scratch/qwen3-0.6B $OUT/scratch/wikitext
+    cat $OUT/${ci}-imatrix.log | grep "Final" >> $OUT/${ci}-imatrix-sum.log

    set +e
 }

+function gg_sum_qwen3_0_6b {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Qwen3 0.6B:\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '- perplexity:\n%s\n' "$(cat $OUT/${ci}-ppl.log)"
+    gg_printf '- imatrix:\n```\n%s\n```\n' "$(cat $OUT/${ci}-imatrix-sum.log)"
+    gg_printf '- f16:\n```\n%s\n```\n'  "$(cat $OUT/${ci}-tg-f16.log)"
+    if [ -z ${GG_BUILD_NO_BF16} ]; then
+        gg_printf '- bf16:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-bf16.log)"
+    fi
+    gg_printf '- q8_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q8_0.log)"
+    gg_printf '- q4_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_0.log)"
+    gg_printf '- q4_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_1.log)"
+    gg_printf '- q5_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_0.log)"
+    gg_printf '- q5_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_1.log)"
+    gg_printf '- q2_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q2_k.log)"
+    gg_printf '- q3_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q3_k.log)"
+    gg_printf '- q4_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_k.log)"
+    gg_printf '- q5_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_k.log)"
+    gg_printf '- q6_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q6_k.log)"
+    gg_printf '- save-load-state: \n```\n%s\n```\n' "$(cat $OUT/${ci}-save-load-state.log)"
+}
+
 # bge-small

 function gg_run_embd_bge_small {
@@ -494,8 +616,6 @@ function gg_run_embd_bge_small {
    gg_wget models-mnt/bge-small/1_Pooling https://huggingface.co/BAAI/bge-small-en-v1.5/raw/main/1_Pooling/config.json

    path_models="../models-mnt/bge-small"
-    path_local="$OUT/scratch/bge-small"
-    rm -rf $path_local && mkdir -p $path_local

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -504,10 +624,10 @@ function gg_run_embd_bge_small {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf

-    model_f16="${path_local}/ggml-model-f16.gguf"
-    model_q8_0="${path_local}/ggml-model-q8_0.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"

    ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0

@@ -516,11 +636,18 @@ function gg_run_embd_bge_small {
    (time ./bin/llama-embedding --model ${model_f16}  -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
    (time ./bin/llama-embedding --model ${model_q8_0} -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log

-    rm -rf $path_local
-
    set +e
 }

+function gg_sum_embd_bge_small {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'BGE Small (BERT):\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '- f16: \n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-f16.log)"
+    gg_printf '- q8_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q8_0.log)"
+}
+
 # rerank_tiny

 function gg_run_rerank_tiny {
@@ -534,8 +661,6 @@ function gg_run_rerank_tiny {
    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/vocab.json

    path_models="../models-mnt/rerank-tiny"
-    path_local="$OUT/scratch/rerank-tiny"
-    rm -rf $path_local && mkdir -p $path_local

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -544,9 +669,9 @@ function gg_run_rerank_tiny {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf

-    model_f16="${path_local}/ggml-model-f16.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"

    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)

@@ -576,63 +701,69 @@ function gg_run_rerank_tiny {
    check_score "rerank score 1" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 1")" "0.00" "0.05" | tee -a $OUT/${ci}-rk-f16.log
    check_score "rerank score 2" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 2")" "0.10" "0.30" | tee -a $OUT/${ci}-rk-f16.log

-    rm -rf $path_local
-
    set +e
 }

+function gg_sum_rerank_tiny {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Rerank Tiny (Jina):\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '- f16: \n```\n%s\n```\n' "$(cat $OUT/${ci}-rk-f16.log)"
+}
+
 function gg_check_build_requirements {
    if ! command -v git &> /dev/null; then
-        echo 'git not found, please install'
+        gg_printf 'git not found, please install\n'
        exit 1
    fi

    if ! command -v git-lfs &> /dev/null; then
-        echo 'git-lfs not found, please install'
+        gg_printf 'git-lfs not found, please install\n'
        exit 1
    fi

    if ! git config --get filter.lfs.clean &> /dev/null; then
-        echo 'git-lfs not initialized, please run `git lfs install`'
+        gg_printf 'git-lfs not initialized, please run `git lfs install`\n'
        exit 1
    fi

    if ! command -v wget &> /dev/null; then
-        echo 'wget not found, please install'
+        gg_printf 'wget not found, please install\n'
        exit 1
    fi

    if ! command -v python3 &> /dev/null; then
-        echo 'python3 not found, please install'
+        gg_printf 'python3 not found, please install\n'
        exit 1
    fi

    if ! command -v pip3 &> /dev/null; then
-        echo 'pip3 not found, please install'
+        gg_printf 'pip3 not found, please install\n'
        exit 1
    fi

    if ! python3 -m ensurepip --help &> /dev/null; then
-        echo 'ensurepip not found, please install python3-venv package'
+        gg_printf 'ensurepip not found, please install python3-venv package\n'
        exit 1
    fi

    if ! command -v cmake &> /dev/null; then
-        echo 'cmake not found, please install'
+        gg_printf 'cmake not found, please install\n'
        exit 1
    fi

    if ! command -v ccache &> /dev/null; then
-        echo 'ccache not found, please consider installing for faster builds'
+        gg_printf 'ccache not found, please consider installing for faster builds\n'
    fi

    if ! command -v ctest &> /dev/null; then
-        echo 'ctest not found, please install'
+        gg_printf 'ctest not found, please install\n'
        exit 1
    fi

    if ! command -v unzip &> /dev/null; then
-        echo 'unzip not found, please install'
+        gg_printf 'unzip not found, please install\n'
        exit 1
    fi
 }
@@ -644,11 +775,13 @@ function gg_run_test_backend_ops {

    set -e

-    local n_jobs=$(nproc)
-    if [ "${n_jobs}" -gt 2 ]; then
-        n_jobs=2
+    local args_extra="-j $(nproc)"
+
+    # TODO: fix multi-threaded for ROCm
+    #       https://github.com/ggml-org/llama.cpp/actions/runs/34576278519/job/103297889044?pr=28740#step:3:4865
+    if [ ! -z ${GG_BUILD_ROCM} ]; then
+        args_extra=""
    fi
-    local args_extra="-j ${n_jobs}"

    # TODO: MoltenVK bug?
    #       https://github.com/ggml-org/llama.cpp/actions/runs/34611260059/job/103302413736?pr=28740#step:3:5897
@@ -656,21 +789,26 @@ function gg_run_test_backend_ops {
        args_extra=""
    fi

-    # TODO: OpenVINO GPU plugin crashes (CL_OUT_OF_RESOURCES) with 2 concurrent workers on GPU.
-    if [ ! -z "${GG_BUILD_OPENVINO}" ] && [ "${GGML_OPENVINO_DEVICE:-}" = "GPU" ]; then
-        args_extra=""
-    fi
-
-    # TODO: reduce the test-backend-ops timeout to 1800s
    if [ ! -z ${GG_BUILD_HIGH_PERF} ]; then
-        (time timeout 3600 ./bin/test-backend-ops ${args_extra} -b CPU) 2>&1 | tee -a $OUT/${ci}-test-backend-ops.log
+        (time ./bin/test-backend-ops ${args_extra} -b CPU) 2>&1 | tee -a $OUT/${ci}-test-backend-ops.log
    else
-        (time timeout 3600 ./bin/test-backend-ops ${args_extra}       ) 2>&1 | tee -a $OUT/${ci}-test-backend-ops.log
+        (time ./bin/test-backend-ops ${args_extra}       ) 2>&1 | tee -a $OUT/${ci}-test-backend-ops.log
    fi

    set +e
 }

+function gg_sum_test_backend_ops {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs test-backend-ops\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-test-backend-ops.log)"
+    gg_printf '```\n'
+    gg_printf '\n'
+}
+
 ## main

 export LLAMA_ARG_LOG_PREFIX=1
@@ -684,11 +822,11 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
    ln -sfn ${mnt_models} ${SRC}/models-mnt

    # Create a fresh python3 venv and enter it
-    if ! python3 -m venv "$OUT/venv"; then
-        echo "Error: Failed to create Python virtual environment at $OUT/venv."
+    if ! python3 -m venv "$MNT/venv"; then
+        echo "Error: Failed to create Python virtual environment at $MNT/venv."
        exit 1
    fi
-    source "$OUT/venv/bin/activate"
+    source "$MNT/venv/bin/activate"

    pip install -r ${SRC}/requirements.txt --disable-pip-version-check
    pip install --editable gguf-py --disable-pip-version-check
@@ -718,4 +856,6 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
    test $ret -eq 0 && gg_run ctest_with_model_release
 fi

+cat $OUT/README.md
+
 exit $ret
@@ -2,7 +2,7 @@ include("ggml/cmake/common.cmake")

 function(llama_add_compile_flags)
    if (LLAMA_FATAL_WARNINGS)
-        if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR CMAKE_CXX_COMPILER_ID MATCHES "IntelLLVM")
+        if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
            list(APPEND C_FLAGS   -Werror)
            list(APPEND CXX_FLAGS -Werror)
        elseif (CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
@@ -17,16 +17,14 @@ find_library(llama_LIBRARY llama
    NO_CMAKE_FIND_ROOT_PATH
 )

-if(NOT TARGET llama)
-    add_library(llama UNKNOWN IMPORTED)
-    set_target_properties(llama
-        PROPERTIES
-            INTERFACE_INCLUDE_DIRECTORIES "${LLAMA_INCLUDE_DIR}"
-            INTERFACE_LINK_LIBRARIES "ggml::ggml;ggml::ggml-base;"
-            IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
-            IMPORTED_LOCATION "${llama_LIBRARY}"
-            INTERFACE_COMPILE_FEATURES c_std_90
-            POSITION_INDEPENDENT_CODE ON)
-endif()
+add_library(llama UNKNOWN IMPORTED)
+set_target_properties(llama
+    PROPERTIES
+        INTERFACE_INCLUDE_DIRECTORIES "${LLAMA_INCLUDE_DIR}"
+        INTERFACE_LINK_LIBRARIES "ggml::ggml;ggml::ggml-base;"
+        IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
+        IMPORTED_LOCATION "${llama_LIBRARY}"
+        INTERFACE_COMPILE_FEATURES c_std_90
+        POSITION_INDEPENDENT_CODE ON)

 check_required_components(Llama)
@@ -78,8 +78,6 @@ add_library(${TARGET}
    download.h
    fit.cpp
    fit.h
-    file-lock.cpp
-    file-lock.h
    hf-cache.cpp
    hf-cache.h
    http.h
@@ -138,6 +136,8 @@ set_target_properties(${TARGET} PROPERTIES
 target_include_directories(${TARGET} PUBLIC .)
 target_link_libraries     (${TARGET} PUBLIC vendor::nlohmann vendor::sheredom)
 target_compile_features   (${TARGET} PUBLIC cxx_std_17)
+target_precompile_headers (${TARGET} PRIVATE common.h)
+target_precompile_headers (${TARGET} PRIVATE chat.h)

 if (LLAMA_SUBPROCESS)
    target_compile_definitions(${TARGET} PUBLIC LLAMA_SUBPROCESS)
@@ -186,8 +186,3 @@ if (LLAMA_LLGUIDANCE)
 endif()

 target_link_libraries(${TARGET} PUBLIC llama Threads::Threads)
-
-if (CMAKE_SYSTEM_NAME MATCHES "AIX")
-    # AIX's flock() function comes from libbsd.a
-    target_link_libraries(${TARGET} PRIVATE -lbsd)
-endif()
@@ -351,7 +351,7 @@ static bool parse_bool_value(const std::string & value) {
 static std::string get_default_local_path(const std::string & url) {
    auto f = string_split<std::string>(url, '#').front();
    f = string_split<std::string>(f, '?').front();
-    return fs_path_to_utf8(fs_get_cache_file(string_split<std::string>(f, '/').back()));
+    return fs_get_cache_file(string_split<std::string>(f, '/').back());
 }

 static bool spec_types_is_default(const common_params & params) {
@@ -387,9 +387,6 @@ common_models_handler common_models_handler_init(const common_params & params, l
            break;
        }
    }
-    if (curr_ex == LLAMA_EXAMPLE_DOWNLOAD) {
-        use_mmproj = true;
-    }

    opts.bearer_token    = params.hf_token;
    opts.offline         = params.offline;
@@ -470,14 +467,6 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    auto opts = handler.opts; // copy
    opts.callback = callback;

-    // another process may hold the cache lock during finalization:
-    // let the download callback cancel that wait too
-    auto finalize = [callback](const hf_cache::hf_file & f) {
-        return hf_cache::finalize_file(f, [callback]() {
-            return !(callback && callback->is_cancelled());
-        });
-    };
-
    // handle plain "url" if needed
    auto handle_url = [&](common_params_model & model) {
        if (!model.url.empty()) {
@@ -526,7 +515,7 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    }

    // handle hf_plan tasks
-    auto add_tasks = [&opts, &tasks, &finalize](const hf_cache::hf_files  & model_files,
+    auto add_tasks = [&opts, &tasks](const hf_cache::hf_files  & model_files,
                                    const hf_cache::hf_file    & primary,
                                    common_params_model        & model) {
        for (size_t i = 0; i < model_files.size(); ++i) {
@@ -535,9 +524,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
            tasks.emplace_back(model_file, opts, [&, is_primary]() {
                if (is_primary) {
                    // the primary file is the first split (00001-of), use it as model path
-                    model.path = finalize(model_file);
+                    model.path = hf_cache::finalize_file(model_file);
                } else {
-                    finalize(model_file);
+                    hf_cache::finalize_file(model_file);
                }
            });
        }
@@ -589,9 +578,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.mtp, opts, [&]() {
            // only use the discovered MTP head when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.mtp);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.mtp);
            } else {
-                finalize(plan_spec.mtp);
+                hf_cache::finalize_file(plan_spec.mtp);
            }
        });
    }
@@ -599,9 +588,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.dflash, opts, [&]() {
            // only use the discovered DFlash sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.dflash);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.dflash);
            } else {
-                finalize(plan_spec.dflash);
+                hf_cache::finalize_file(plan_spec.dflash);
            }
        });
    }
@@ -609,9 +598,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.eagle3, opts, [&]() {
            // only use the discovered Eagle3 sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.eagle3);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.eagle3);
            } else {
-                finalize(plan_spec.eagle3);
+                hf_cache::finalize_file(plan_spec.eagle3);
            }
        });
    }
@@ -619,9 +608,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.dspark, opts, [&]() {
            // only use the discovered DSpark sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.dspark);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.dspark);
            } else {
-                finalize(plan_spec.dspark);
+                hf_cache::finalize_file(plan_spec.dspark);
            }
        });
    }
@@ -642,16 +631,16 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    }
    if (!plan.mmproj.local_path.empty()) {
        tasks.emplace_back(plan.mmproj, opts, [&]() {
-            params.mmproj.path = finalize(plan.mmproj);
+            params.mmproj.path = hf_cache::finalize_file(plan.mmproj);
        });
    }
    if (!plan.mtp.local_path.empty() && !had_spec_url) {
        tasks.emplace_back(plan.mtp, opts, [&]() {
            // only fall back to the discovered MTP head when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.mtp);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.mtp);
            } else {
-                finalize(plan.mtp);
+                hf_cache::finalize_file(plan.mtp);
            }
        });
    }
@@ -659,9 +648,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.dflash, opts, [&]() {
            // only fall back to the discovered DFlash sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.dflash);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.dflash);
            } else {
-                finalize(plan.dflash);
+                hf_cache::finalize_file(plan.dflash);
            }
        });
    }
@@ -669,9 +658,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.eagle3, opts, [&]() {
            // only fall back to the discovered Eagle3 sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.eagle3);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.eagle3);
            } else {
-                finalize(plan.eagle3);
+                hf_cache::finalize_file(plan.eagle3);
            }
        });
    }
@@ -679,9 +668,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.dspark, opts, [&]() {
            // only fall back to the discovered DSpark sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.dspark);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.dspark);
            } else {
-                finalize(plan.dspark);
+                hf_cache::finalize_file(plan.dspark);
            }
        });
    }
@@ -689,11 +678,8 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.preset, opts, [&]() {
            // if HF repo is a preset repo, we simply run server in router mode with the preset.ini file
            params.models_preset_hf = params.model.hf_repo; // only for showing a warning
-            params.models_preset    = finalize(plan.preset);
-            // clear the model so the server starts in router mode
-            params.model.path.clear();
-            params.model.hf_repo.clear();
-            params.model.docker_repo.clear();
+            params.models_preset    = hf_cache::finalize_file(plan.preset);
+            params.model = common_params_model{}; // make sure to clear model, so server starts in router mode
        });
    }

@@ -731,24 +717,24 @@ void common_models_handler_apply(common_models_handler & handler, common_params
 // 1. system-wide: /etc/llama.cpp/config.ini (%PROGRAMDATA%\llama.cpp\config.ini on windows)
 // 2. user-level: ${XDG_CONFIG_HOME:-~/.config}/llama.cpp/config.ini (%APPDATA%\llama.cpp\config.ini on windows)
 static void common_params_apply_system_config(common_params & params, llama_example ex) {
-    std::vector<std::filesystem::path> paths;
+    std::vector<std::string> paths;

 #if defined(_WIN32)
-    const std::filesystem::path program_data = common_get_path_from_env("PROGRAMDATA");
+    const std::string program_data = common_get_env("PROGRAMDATA");
    if (!program_data.empty()) {
-        paths.push_back(program_data / "llama.cpp" / "config.ini");
+        paths.push_back(program_data + "\\llama.cpp\\config.ini");
    }
 #else
    paths.push_back("/etc/llama.cpp/config.ini");
 #endif

    try {
-        paths.push_back(fs_get_config_directory() / "config.ini");
+        paths.push_back(fs_get_config_directory() + "config.ini");
    } catch (const std::exception & e) {
        LOG_DBG("cannot read user-level config file, skipping: %s\n", e.what());
    }

-    std::vector<std::filesystem::path> found;
+    std::vector<std::string> found;
    for (const auto & path : paths) {
        std::error_code ec;
        if (std::filesystem::exists(path, ec)) {
@@ -762,7 +748,7 @@ static void common_params_apply_system_config(common_params & params, llama_exam
    common_preset_context ctx(ex);
    ctx.ignore_unknown_keys = true; // the same config file is shared by all programs
    for (const auto & path : found) {
-        LOG_INF("using config file: %s\n", fs_path_to_utf8(path).c_str());
+        LOG_INF("using config file: %s\n", path.c_str());
        common_preset global;
        common_presets presets = ctx.load_from_ini(path, global);
        global.apply_to_params(params);
@@ -1487,7 +1473,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
    ));
    add_opt(common_arg(
        {"--server-base"}, "URL",
-        string_format("connect to this server instead of starting a new one, example: 'http://localhost:9931' (default: none)"),
+        string_format("connect to this server instead of starting a new one, example: 'http://localhost:8080' (default: none)"),
        [](common_params & params, const std::string & value) {
            params.server_base = value;
        }
@@ -1725,11 +1711,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
    ).set_env("LLAMA_ARG_CHECKPOINT_MIN_SPACING_NT").set_examples({LLAMA_EXAMPLE_SERVER}));
    add_opt(common_arg(
        {"-cram", "--cache-ram"}, "N",
-        string_format("set the maximum cache size in MiB (default: %d, 0 for decision models, -1 - no limit, 0 - disable)"
+        string_format("set the maximum cache size in MiB (default: %d, -1 - no limit, 0 - disable)"
            "[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)", params.cache_ram_mib),
        [](common_params & params, int value) {
            params.cache_ram_mib = value;
-            params.cache_ram_explicit = true;
        }
    ).set_env("LLAMA_ARG_CACHE_RAM").set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}));
    add_opt(common_arg(
@@ -1739,7 +1724,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
        [](common_params & params, bool value) {
            params.kv_unified = value;
        }
-    ).set_env("LLAMA_ARG_KV_UNIFIED").set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_BATCHED, LLAMA_EXAMPLE_BENCH, LLAMA_EXAMPLE_PARALLEL, LLAMA_EXAMPLE_SPECULATIVE}));
+    ).set_env("LLAMA_ARG_KV_UNIFIED").set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_BATCHED, LLAMA_EXAMPLE_BENCH, LLAMA_EXAMPLE_PARALLEL}));
    add_opt(common_arg(
        {"--cache-idle-slots"},
        {"--no-cache-idle-slots"},
@@ -2031,7 +2016,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.sampling.temp = std::max(params.sampling.temp, 0.0f);
            params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TEMP;
        }
-    ).set_sampling().set_env("LLAMA_ARG_TEMPERATURE"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--top-k"}, "N",
        string_format("top-k sampling (default: %d, 0 = disabled)", params.sampling.top_k),
@@ -2047,7 +2032,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.sampling.top_p = std::stof(value);
            params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TOP_P;
        }
-    ).set_sampling().set_env("LLAMA_ARG_TOP_P"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--min-p"}, "N",
        string_format("min-p sampling (default: %.2f, 0.0 = disabled)", (double)params.sampling.min_p),
@@ -2055,7 +2040,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.sampling.min_p = std::stof(value);
            params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIN_P;
        }
-    ).set_sampling().set_env("LLAMA_ARG_MIN_P"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--top-nsigma", "--top-n-sigma"}, "N",
        string_format("top-n-sigma sampling (default: %.2f, -1.0 = disabled)", params.sampling.top_n_sigma),
@@ -2111,7 +2096,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.sampling.penalty_repeat = penalty_repeat;
            params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_REPEAT;
        }
-    ).set_sampling().set_env("LLAMA_ARG_REPEAT_PENALTY"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--presence-penalty"}, "N",
        string_format("repeat alpha presence penalty (default: %.2f, 0.0 = disabled)", (double)params.sampling.penalty_present),
@@ -2122,7 +2107,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            }
            params.sampling.penalty_present = penalty_present;
        }
-    ).set_sampling().set_env("LLAMA_ARG_PRESENCE_PENALTY"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--frequency-penalty"}, "N",
        string_format("repeat alpha frequency penalty (default: %.2f, 0.0 = disabled)", (double)params.sampling.penalty_freq),
@@ -2133,7 +2118,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            }
            params.sampling.penalty_freq = penalty_freq;
        }
-    ).set_sampling().set_env("LLAMA_ARG_FREQUENCY_PENALTY"));
+    ).set_sampling());
    add_opt(common_arg(
        {"--dry-multiplier"}, "N",
        string_format("set DRY sampling multiplier (default: %.2f, 0.0 = disabled)", (double)params.sampling.dry_multiplier),
@@ -2688,17 +2673,16 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.video_ffmpeg_bin_dir = value;
        }
    ).set_examples(mmproj_examples).set_env("LLAMA_ARG_VIDEO_FFMPEG_DIR"));
-    add_opt(common_arg(
-        {"--rpc"}, "SERVERS",
-        "comma-separated list of RPC servers (host:port)",
-        [](common_params & params, const std::string & value) {
-            if (!llama_supports_rpc()) {
-                throw std::invalid_argument("RPC not supported in this build");
+    if (params.is_gen_docs || llama_supports_rpc()) {
+        add_opt(common_arg(
+            {"--rpc"}, "SERVERS",
+            "comma-separated list of RPC servers (host:port)",
+            [](common_params & params, const std::string & value) {
+                add_rpc_devices(value);
+                GGML_UNUSED(params);
            }
-            add_rpc_devices(value);
-            GGML_UNUSED(params);
-        }
-    ).set_env("LLAMA_ARG_RPC"));
+        ).set_env("LLAMA_ARG_RPC"));
+    }
    add_opt(common_arg(
        {"-lm", "--load-mode"}, "MODE",
        "model loading mode (default: auto)\n"
@@ -2785,16 +2769,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            llm_add_n_cpu_ffn_overrides(value, LLM_FFN_EXPS_REGEX, params.tensor_buft_overrides);
        }
    ).set_env("LLAMA_ARG_N_CPU_MOE"));
-    add_opt(common_arg(
-        {"--moe-cache-mib"}, "N",
-        "GPU cache size in MiB for the MoE experts kept in the CPU. with multiple GPUs, it is split among them like the layers (--tensor-split) (default: 0, disabled)",
-        [](common_params & params, int value) {
-            if (value < 0) {
-                throw std::invalid_argument("invalid value");
-            }
-            params.moe_cache_size = (size_t) value*1024*1024;
-        }
-    ).set_env("LLAMA_ARG_MOE_CACHE_MIB"));
    add_opt(common_arg(
        {"-ncffn", "--n-cpu-ffn"}, "N",
        "keep the dense FFN weights of the first N layers in the CPU\n"
@@ -3202,13 +3176,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.process_output = true;
        }
    ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(common_arg(
-        {"--nextn"},
-        string_format("collect data for MTP/NextN layers (default: %s)", params.load_mtp ? "true" : "false"),
-        [](common_params & params) {
-            params.load_mtp = true;
-        }
-    ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
    add_opt(common_arg(
        {"--ppl"},
        {"--no-ppl"},
@@ -3341,18 +3308,9 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
    ).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
    add_opt(common_arg(
        {"--host"}, "HOST",
-        string_format("IP addresses to listen on, comma-separated, or UNIX socket paths ending in .sock; with multiple TCP addresses, :: binds IPv6 only; overlapping addresses result in undefined behavior (default: %s)", params.hostnames[0].c_str()),
+        string_format("ip address to listen, or bind to an UNIX socket if the address ends with .sock (default: %s)", params.hostname.c_str()),
        [](common_params & params, const std::string & value) {
-            params.hostnames.clear();
-            for (auto & host : parse_csv_row(value)) {
-                host = string_strip(host);
-                if (!host.empty()) {
-                    params.hostnames.push_back(host);
-                }
-            }
-            if (params.hostnames.empty()) {
-                throw std::invalid_argument("--host requires at least one address");
-            }
+            params.hostname = value;
        }
    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_HOST"));
    add_opt(common_arg(
@@ -3502,26 +3460,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.ui = value;
        }
    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_UI"));
-    add_opt(common_arg(
-        {"--connect"},
-        string_format("(experimental) open a peer-to-peer tunnel for server using llama-connect (default: %s)", params.server_connect ? "enabled" : "disabled"),
-        [](common_params & params) {
-            params.server_connect = true;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CONNECT"));
-    add_opt(common_arg(
-        {"--connect-code"}, "CODE",
-        "manually specify a 40-character code for --connect (default: generate a new one for each run)",
-        [](common_params & params, const std::string & value) {
-            std::string val = value;
-            string_replace_all(val, "-", "");
-            string_replace_all(val, " ", "");
-            if (val.size() != 40) {
-                throw std::invalid_argument(string_format("error: invalid connect code '%s', must be 40 characters\n", value.c_str()));
-            }
-            params.server_connect_code = val;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CONNECT_CODE"));
    add_opt(common_arg(
        {"--embedding", "--embeddings"},
        string_format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"),
@@ -3672,7 +3610,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
                params.slot_save_path += DIRECTORY_SEPARATOR;
            }
        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SLOT_SAVE_PATH"));
+    ).set_examples({LLAMA_EXAMPLE_SERVER}));
    add_opt(common_arg(
        {"--media-path"}, "PATH",
        "directory for loading local media files; files can be accessed via file:// URLs using relative paths (default: disabled)",
@@ -3937,7 +3875,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
        {"--no-log-jsonl"},
        "Log as JSONL (one JSON object per line) to stdout, this also disables colored logging (default: disabled)",
        [](common_params &, bool value) {
-            common_log_set_jsonl(value);
+            common_log_set_jsonl(common_log_main(), value);
        }
    ).set_env("LLAMA_ARG_LOG_JSONL"));
    add_opt(common_arg(
@@ -4202,16 +4140,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.speculative.draft.n_min = value;
        }
    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_SPEC_DRAFT_N_MIN"));
-    add_opt(common_arg(
-        {"--spec-draft-n-ubatch"}, "N",
-        string_format("physical maximum batch size of the draft context, use a smaller value to spare VRAM (default: %d - same as --n-ubatch)", params.speculative.draft.n_ubatch),
-        [](common_params & params, int value) {
-            if (value < 0) {
-                throw std::invalid_argument("invalid value");
-            }
-            params.speculative.draft.n_ubatch = value;
-        }
-    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_SPEC_DRAFT_N_UBATCH"));
    add_opt(common_arg(
        {"--spec-synth-len"}, "L",
        "target mean synthetic acceptance length, including the target token (benchmarking only)",
@@ -4268,21 +4196,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.speculative.draft.backend_sampling = value;
        }
    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_SPEC_DRAFT_BACKEND_SAMPLING"));
-    add_opt(common_arg(
-        {"--spec-draft-sampling"}, "{greedy,probabilistic}",
-        string_format("how the draft is sampled: greedy takes its argmax, probabilistic samples it and has "
-                      "the target verify by rejection sampling (default: %s)",
-                      params.speculative.draft.probabilistic ? "probabilistic" : "greedy"),
-        [](common_params & params, const std::string & value) {
-            if (value == "greedy") {
-                params.speculative.draft.probabilistic = false;
-            } else if (value == "probabilistic") {
-                params.speculative.draft.probabilistic = true;
-            } else {
-                throw std::invalid_argument("invalid value, must be one of: greedy, probabilistic");
-            }
-        }
-    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_SPEC_DRAFT_SAMPLING"));
    add_opt(common_arg(
        {"--spec-draft-device", "-devd", "--device-draft"}, "<dev1,dev2,..>",
        "comma-separated list of devices to use for offloading the draft model (none = don't offload, default: follows --device)\n"
@@ -4318,7 +4231,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.speculative.draft.mparams.path = value;
            params.speculative.draft.mparams.hf_file = value; // will be used if --spec-draft-hf is set
        }
-    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI, LLAMA_EXAMPLE_IMATRIX}).set_env("LLAMA_ARG_SPEC_DRAFT_MODEL"));
+    ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_SPEC_DRAFT_MODEL"));
    add_opt(common_arg(
        {"--spec-type"}, common_speculative_all_types_str(),
        string_format("comma-separated list of types of speculative decoding to use (default: %s)\n",
@@ -122,8 +122,6 @@ struct common_params_context {

 // parse input arguments from CLI
 // if one argument has invalid value, it will automatically display usage of the specific argument (and not the full usage message)
-// TODO: this function can load ggml backend (by calling llama_support_rpc)
-//       this is a side-effect that should be avoided
 bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);

 // load all backends and print the list of available (non-CPU) devices to stdout
@@ -58,10 +58,11 @@ common_chat_params peg_generator::generate_parser(const common_chat_template &
            data.generation_prompt += msg.render_content();
        }

-        data.prompt.append(data.generation_prompt);
+        data.prompt += data.generation_prompt;
    }

-    data.parser = autoparser.build_parser(inputs, parser_generation_prompt);
+    auto parser = autoparser.build_parser(inputs, parser_generation_prompt);
+    data.parser = parser.save();

    // Build grammar if tools are present
    bool has_tools =
@@ -77,7 +78,7 @@ common_chat_params peg_generator::generate_parser(const common_chat_template &
    if (include_grammar) {
        data.grammar_lazy = !has_response_format && inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_AUTO;
        data.grammar      = build_grammar([&](const common_grammar_builder & builder) {
-            data.parser.build_grammar(builder, data.grammar_lazy);
+            parser.build_grammar(builder, data.grammar_lazy);
        });

        // Set grammar triggers based on tool section markers (fall back to per-call markers)
@@ -290,7 +291,7 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context

    common_peg_parser tool_choice = p.choice();

-    foreach_function(inputs.tools, [&](size_t tool_index, const json & tool) {
+    foreach_function(inputs.tools, [&](const json & tool) {
        const auto & func   = tool.at("function");
        std::string  name   = func.at("name");
        const auto   schema = common_chat_tool_parameters(func);
@@ -307,7 +308,7 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context
            }
            have_call_id = true;
        }
-        auto args_parser = p.tool_args(p.schema(p.json(), "tool-" + std::to_string(tool_index) + "-schema", schema));
+        auto args_parser = p.tool_args(p.schema(p.json(), "tool-" + name + "-schema", schema));
        if (!arguments.start.empty()) {
            args_parser = p.literal(arguments.start) + args_parser;
        }
@@ -317,7 +318,7 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context

        auto atomic_peek = !arguments.start.empty() ? std::optional(p.peek(p.literal(arguments.start))) : std::nullopt;
        auto func_parser = build_func_parser(p, name, call_id_section, have_call_id, args_parser, atomic_peek);
-        tool_choice |= p.rule("tool-" + std::to_string(tool_index), func_parser);
+        tool_choice |= p.rule("tool-" + name, func_parser);
    });

    auto require_calls = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
@@ -363,14 +364,14 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte

    common_peg_parser tool_choice = p.choice();

-    foreach_function(inputs.tools, [&](size_t tool_index, const json & tool) {
+    foreach_function(inputs.tools, [&](const json & tool) {
        const auto & func = tool.at("function");
        std::string  name = func.at("name");

        // Build parser for each argument, separating required and optional
        std::vector<common_peg_parser> required_parsers;
        std::vector<common_peg_parser> optional_parsers;
-        foreach_parameter(func, [&](size_t param_index, const common_chat_schema_property & param, const common_chat_schema_document_ptr & doc) {
+        foreach_parameter(func, [&](const common_chat_schema_property & param, const common_chat_schema_document_ptr & doc) {
            auto arg =
                p.tool_arg(p.tool_arg_open(arguments.name_prefix + p.tool_arg_name(p.literal(param.name)) +
                                           arguments.name_suffix) +
@@ -379,10 +380,10 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
                                p.ac(p.tool_arg_string_value(until_suffix) +
                                    p.tool_arg_close(p.literal(arguments.value_suffix)), arguments.value_suffix) :
                                (p.tool_arg_json_value(p.schema(
-                                    p.json(), "tool-" + std::to_string(tool_index) + "-arg-" + std::to_string(param_index) + "-schema", doc, *param.schema)) +
+                                    p.json(), "tool-" + name + "-arg-" + param.name + "-schema", doc, *param.schema)) +
                                    p.tool_arg_close(p.literal(arguments.value_suffix)))));

-            auto named_arg = p.rule("tool-" + std::to_string(tool_index) + "-arg-" + std::to_string(param_index), arg);
+            auto named_arg = p.rule("tool-" + name + "-arg-" + param.name, arg);
            if (param.required) {
                required_parsers.push_back(named_arg);
            } else {
@@ -433,7 +434,7 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
        auto atomic_peek = (!arguments.name_prefix.empty() && !required_parsers.empty()) ?
            std::optional(p.peek(p.literal(arguments.name_prefix))) : std::nullopt;
        auto func_parser = build_func_parser(p, name, call_id_section, have_call_id, args_seq, atomic_peek);
-        tool_choice |= p.rule("tool-" + std::to_string(tool_index), func_parser);
+        tool_choice |= p.rule("tool-" + name, func_parser);
    });

    auto require_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
@@ -322,7 +322,7 @@ std::string apply_template(const common_chat_template & tmpl, const template_par
    tmpl_params.extra_context["enable_thinking"] = params.enable_thinking;

    try {
-        return common_chat_template_direct_apply(tmpl, tmpl_params).str();
+        return common_chat_template_direct_apply(tmpl, tmpl_params);
    } catch (const std::exception & e) {
        LOG_DBG("Template application failed: %s\n", e.what());
        return ERR_TMPL;
@@ -318,13 +318,13 @@ void common_chat_peg_mapper::map(const common_peg_ast_node & node) {
    bool is_content   = node.tag == common_chat_peg_builder::CONTENT;

    if (is_reasoning) { // GPT OSS can have more than 1 reasoning block, so concatenate here
-        result.reasoning_content += node.sanitized_text();
+        result.reasoning_content += std::string(node.text);
    }

    if (is_content) {
        // Concatenate content from multiple content nodes (e.g., when reasoning markers
        // are preserved before content markers in reasoning_format=NONE mode)
-        result.content += node.sanitized_text();
+        result.content += std::string(node.text);
    }

    // Handle tool-related tags (supporting both JSON and tagged formats)
@@ -451,7 +451,6 @@ void common_chat_peg_mapper::map(const common_peg_ast_node & node) {
                result.tool_calls.push_back(pending_tool_call.value());
            }
            pending_tool_call.reset();
-            current_tool = nullptr;
        }
    }
 }
@@ -483,9 +482,7 @@ common_peg_parser common_chat_peg_builder::standard_constructed_tools(
    // Build tool choices for tagged format
    auto tool_choices = choice();

-    for (size_t i = 0; i < tools.size(); i++) {
-        const auto & tool_def = tools[i];
-
+    for (const auto & tool_def : tools) {
        if (!tool_def.contains("function")) {
            continue;
        }
@@ -515,7 +512,7 @@ common_peg_parser common_chat_peg_builder::standard_constructed_tools(
        auto tool_parser = tool(tool_open(literal(func_opener) + tool_name(literal(name)) + literal(func_name_suffix)) +
                                space() + tool_args(args) + space() + tool_close(literal(func_closer)));

-        tool_choices |= rule("tool-" + std::to_string(i), tool_parser);
+        tool_choices |= rule("tool-" + name, tool_parser);
    }

    // Build the section with markers
@@ -562,8 +559,7 @@ common_peg_parser common_chat_peg_builder::python_style_tool_calls(

    auto tool_choices = choice();

-    for (size_t i = 0; i < tools.size(); i++) {
-        const auto & tool_def = tools[i];
+    for (const auto & tool_def : tools) {
        if (!tool_def.contains("function")) {
            continue;
        }
@@ -610,7 +606,7 @@ common_peg_parser common_chat_peg_builder::python_style_tool_calls(
            space() + tool_args(args) + space() + tool_close(literal(")"))
        );

-        tool_choices |= rule("tool-" + std::to_string(i), tool_parser);
+        tool_choices |= rule("tool-" + name, tool_parser);
    }

    if (parallel_tool_calls) {
@@ -638,8 +634,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_function_is_key(

    auto tool_choices = choice();

-    for (size_t i = 0; i < tools.size(); i++) {
-        const auto & tool_def = tools[i];
+    for (const auto & tool_def : tools) {
        if (!tool_def.contains("function")) {
            continue;
        }
@@ -672,10 +667,10 @@ common_peg_parser common_chat_peg_builder::build_json_tools_function_is_key(
        // Arguments — either wrapped in args_key or parsed directly
        common_peg_parser args_parser = eps();
        if (args_key.empty()) {
-            args_parser = tool_args(schema(json(), "tool-" + std::to_string(i) + "-schema", params));
+            args_parser = tool_args(schema(json(), "tool-" + name + "-schema", params));
        } else {
            args_parser = literal("\"" + effective_args_key + "\"") + space() + literal(":") + space() +
-                          tool_args(schema(json(), "tool-" + std::to_string(i) + "-schema", params));
+                          tool_args(schema(json(), "tool-" + name + "-schema", params));
        }
        inner_fields.push_back(args_parser);

@@ -702,7 +697,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_function_is_key(
            space() + tool_close(literal("}"))
        );

-        tool_choices |= rule("tool-" + std::to_string(i), tool_parser);
+        tool_choices |= rule("tool-" + name, tool_parser);
    }

    return tool_choices;
@@ -725,8 +720,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_nested_keys(
    std::string nested_name_field = !name_spec.first.empty() ? name_spec.second  : effective_name_key;
    std::string nested_args_field = !args_spec.first.empty() ? args_spec.second  : effective_args_key;

-    for (size_t i = 0; i < tools.size(); i++) {
-        const auto & tool_def = tools[i];
+    for (const auto & tool_def : tools) {
        if (!tool_def.contains("function")) {
            continue;
        }
@@ -737,7 +731,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_nested_keys(
        auto nested_name = literal("\"" + nested_name_field + "\"") + space() + literal(":") + space() +
                          atomic(literal("\"") + tool_name(literal(name)) + literal("\""));
        auto nested_args = literal("\"" + nested_args_field + "\"") + space() + literal(":") + space() +
-                          tool_args(schema(json(), "tool-" + std::to_string(i) + "-schema", params));
+                          tool_args(schema(json(), "tool-" + name + "-schema", params));

        auto nested_object = literal("{") + space() +
                            nested_name + space() + literal(",") + space() +
@@ -775,7 +769,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_nested_keys(
        auto nested_field = literal("\"" + nested_prefix + "\"") + space() + literal(":") + space() + nested_object;
        tool_parser_body = tool_parser_body + nested_field + space() + tool_close(literal("}"));

-        tool_choices |= rule("tool-" + std::to_string(i), tool(tool_parser_body));
+        tool_choices |= rule("tool-" + name, tool(tool_parser_body));
    }

    return tool_choices;
@@ -795,8 +789,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_flat_keys(
    auto name_key_parser = literal("\"" + effective_name_key + "\"");
    auto args_key_parser = literal("\"" + effective_args_key + "\"");

-    for (size_t i = 0; i < tools.size(); i++) {
-        const auto & tool_def = tools[i];
+    for (const auto & tool_def : tools) {
        if (!tool_def.contains("function")) {
            continue;
        }
@@ -807,7 +800,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_flat_keys(
        auto tool_name_ = name_key_parser + space() + literal(":") + space() +
                         atomic(literal("\"") + tool_name(literal(name)) + literal("\""));
        auto tool_args_ = args_key_parser + space() + literal(":") + space() +
-                         tool_args(schema(json(), "tool-" + std::to_string(i) + "-schema", params));
+                         tool_args(schema(json(), "tool-" + name + "-schema", params));

        // Build ID parsers if keys are provided
        common_peg_parser id_parser = eps();
@@ -867,7 +860,7 @@ common_peg_parser common_chat_peg_builder::build_json_tools_flat_keys(
        }
        ordered_body = ordered_body + space() + tool_close(literal("}"));

-        tool_choices |= rule("tool-" + std::to_string(i), tool(ordered_body));
+        tool_choices |= rule("tool-" + name, tool(ordered_body));
    }

    return tool_choices;
@@ -1065,12 +1058,12 @@ void common_chat_peg_gemma4_mapper::visit(const common_peg_ast_arena & arena, co
    const auto & node = arena.get(id);

    if (node.tag == "reasoning") {
-        result.reasoning_content += node.sanitized_text();
+        result.reasoning_content += std::string(node.text);
        return;
    }

    if (node.tag == "content") {
-        result.content += node.sanitized_text();
+        result.content += std::string(node.text);
        return;
    }

@@ -1213,12 +1206,12 @@ void common_chat_peg_minimax_m3_mapper::visit(const common_peg_ast_arena & arena
    const auto & node = arena.get(id);

    if (node.tag == common_chat_peg_builder::REASONING) {
-        result.reasoning_content += node.sanitized_text();
+        result.reasoning_content += std::string(node.text);
        return;
    }

    if (node.tag == common_chat_peg_builder::CONTENT) {
-        result.content += node.sanitized_text();
+        result.content += std::string(node.text);
        return;
    }

@@ -9,7 +9,6 @@
 #include "json.h"
 #include "log.h"
 #include "parsers/parsers.h"
-#include "sampling.h"

 #include "jinja/value.h"
 #include "jinja/runtime.h"
@@ -113,6 +112,38 @@ const char * common_chat_role_to_string(common_chat_role role) {
    return "";
 }

+json common_chat_msg_delimiters::to_json() const {
+    json result = json::array();
+    for (const auto & d : delimiters) {
+        result.push_back({
+            { "role",      common_chat_role_to_string(d.role) },
+            { "delimiter", d.delimiter                        },
+        });
+    }
+    return result;
+}
+
+common_chat_msg_delimiters common_chat_msg_delimiters_parse(const json & delimiters) {
+    common_chat_msg_delimiters result;
+
+    if (!delimiters.is_array()) {
+        return result;
+    }
+
+    result.delimiters.reserve(delimiters.size());
+    for (const auto & d : delimiters) {
+        if (!d.is_object()) {
+            continue;
+        }
+        result.delimiters.push_back({
+            common_chat_role_from_string(d.value("role", std::string())),
+            d.value("delimiter", std::string()),
+        });
+    }
+
+    return result;
+}
+
 void common_chat_msg_delimiters::tokenize(const llama_vocab * vocab) {
    for (auto & d : delimiters) {
        d.tokens = common_tokenize(vocab, d.delimiter, false, true);
@@ -589,11 +620,8 @@ std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & too
 }

 common_chat_continuation common_chat_continuation_parse(const common_json & value) {
-    if (value.is_null()) {
-        return COMMON_CHAT_CONTINUATION_NONE;
-    }
-    if (value.is_boolean()) {
-        return value.get<bool>() ? COMMON_CHAT_CONTINUATION_AUTO : COMMON_CHAT_CONTINUATION_NONE;
+    if (value.is_boolean() && value.get<bool>()) {
+        return COMMON_CHAT_CONTINUATION_AUTO;
    }
    if (value.is_string()) {
        auto value_str = value.get<std::string>();
@@ -604,7 +632,7 @@ common_chat_continuation common_chat_continuation_parse(const common_json & valu
            return COMMON_CHAT_CONTINUATION_CONTENT;
        }
    }
-    throw std::invalid_argument("Invalid continue_final_message: expected a boolean, \"content\" or \"reasoning_content\"");
+    return COMMON_CHAT_CONTINUATION_NONE;
 }

 bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) {
@@ -647,7 +675,7 @@ std::string common_chat_format_single(const struct common_chat_templates * tmpls
    if (!past_msg.empty()) {
        inputs.messages              = past_msg;
        inputs.add_generation_prompt = false;
-        fmt_past_msg                 = common_chat_templates_apply(tmpls, inputs).prompt.str();
+        fmt_past_msg                 = common_chat_templates_apply(tmpls, inputs).prompt;
    }
    std::ostringstream ss;
    // if the past_msg ends with a newline, we must preserve it in the formatted version
@@ -657,7 +685,7 @@ std::string common_chat_format_single(const struct common_chat_templates * tmpls
    // format chat with new_msg
    inputs.messages.push_back(new_msg);
    inputs.add_generation_prompt = add_ass;
-    auto fmt_new_msg             = common_chat_templates_apply(tmpls, inputs).prompt.str();
+    auto fmt_new_msg             = common_chat_templates_apply(tmpls, inputs).prompt;
    // get the diff part
    ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
    return ss.str();
@@ -681,7 +709,7 @@ std::string common_chat_format_example(const struct common_chat_templates *
    add_simple_msg("user", "Hello");
    add_simple_msg("assistant", "Hi there");
    add_simple_msg("user", "How are you?");
-    return common_chat_templates_apply(tmpls, inputs).prompt.str();
+    return common_chat_templates_apply(tmpls, inputs).prompt;
 }

 #define CHATML_TEMPLATE_SRC                                                               \
@@ -874,7 +902,7 @@ common_reasoning_format common_reasoning_format_from_name(const std::string & fo
    throw std::runtime_error("Unknown reasoning format: " + format);
 }

-common_chat_prompt common_chat_template_direct_apply_impl(
+std::string common_chat_template_direct_apply_impl(
    const common_chat_template & tmpl,
    const autoparser::generation_params & inputs,
    const std::optional<json> & messages_override,
@@ -920,37 +948,24 @@ common_chat_prompt common_chat_template_direct_apply_impl(

    jinja::global_from_json(ctx, inp, inputs.mark_input);

-    // bos/eos come from the model, not from the user
-    ctx.set_val("bos_token", jinja::mk_val<jinja::value_string>(tmpl.bos_token()));
-    ctx.set_val("eos_token", jinja::mk_val<jinja::value_string>(tmpl.eos_token()));
-
    // render
    jinja::runtime runtime(ctx);
    const jinja::value results = runtime.execute(tmpl.prog);
    auto parts = jinja::runtime::gather_string_parts(results);

-    common_chat_prompt result;
-    for (auto & part : parts->val_str.parts) {
-        result.parts.push_back({std::move(part.val), part.is_input});
-    }
+    std::string result = parts->as_string().str();

    // TODO: improve this later
-    if (inputs.add_bos && !result.parts.empty()) {
-        auto & first = result.parts.front();
-        if (!first.is_input && string_starts_with(first.text, tmpl.bos_token())) {
-            first.text.erase(0, tmpl.bos_token().size());
-        }
+    if (inputs.add_bos && string_starts_with(result, tmpl.bos_token())) {
+        result = result.substr(tmpl.bos_token().size());
    }
-    if (inputs.add_eos && !result.parts.empty()) {
-        auto & last = result.parts.back();
-        if (!last.is_input && string_ends_with(last.text, tmpl.eos_token())) {
-            last.text.resize(last.text.size() - tmpl.eos_token().size());
-        }
+    if (inputs.add_eos && string_ends_with(result, tmpl.eos_token())) {
+        result = result.substr(0, result.size() - tmpl.eos_token().size());
    }
    return result;
 }

-common_chat_prompt common_chat_template_direct_apply(
+std::string common_chat_template_direct_apply(
    const common_chat_template & tmpl,
    const autoparser::generation_params & inputs) {
    return common_chat_template_direct_apply_impl(tmpl, inputs, std::nullopt, std::nullopt, std::nullopt);
@@ -966,9 +981,9 @@ std::string common_chat_template_generation_prompt_impl(
    autoparser::generation_params params = inputs;
    params.add_generation_prompt = false;
    params.continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
-    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params, messages_override, tools_override, additional_context).str();
+    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params, messages_override, tools_override, additional_context);
    params.add_generation_prompt = true;
-    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params, messages_override, tools_override, additional_context).str();
+    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params, messages_override, tools_override, additional_context);

    size_t prefix_len = 0;
    size_t min_size = std::min(no_gen_prompt.size(), gen_prompt.size());
@@ -1072,68 +1087,35 @@ static json common_chat_extra_context() {
    return ctx;
 }

-static common_chat_params common_chat_params_init_lfm2_tokens(const common_chat_template & tmpl, const autoparser::generation_params & inputs) {
-    return common_chat_params_init_lfm2(tmpl, inputs, /* tool_list_tokens = */ true);
-}
-
-static common_chat_params common_chat_params_init_lfm2_5(const common_chat_template & tmpl, const autoparser::generation_params & inputs) {
-    return common_chat_params_init_lfm2(tmpl, inputs, /* tool_list_tokens = */ false);
-}
-
-// Older gemma4 templates need their tool responses rewritten before rendering
-static common_chat_params common_chat_params_init_gemma4_legacy(const common_chat_template & tmpl, const autoparser::generation_params & inputs) {
-    auto adjusted = inputs;
-    workaround::convert_tool_responses_gemma4(adjusted.messages);
-    return common_chat_params_init_gemma4(tmpl, adjusted);
-}
-
-// Pick the dedicated handler for a template from its source, or null for the autoparser.
-// Order matters: the first match wins, and later checks assume the earlier ones did not match.
-static common_chat_params_init_fn common_chat_template_detect_params_init(const std::string & src) {
+std::optional<common_chat_params> common_chat_try_specialized_template(
+        const common_chat_template &          tmpl,
+        const std::string &                   src,
+        autoparser::generation_params & params) {
    // Ministral/Mistral Large 3 - uses special reasoning structure fixes, can't use autoparser
    // Note: Mistral Small 3.2 uses [CALL_ID] which Ministral doesn't have, so we can distinguish them
    if (src.find("[SYSTEM_PROMPT]") != std::string::npos && src.find("[TOOL_CALLS]") != std::string::npos &&
        src.find("[ARGS]") != std::string::npos && src.find("[CALL_ID]") == std::string::npos) {
        LOG_DBG("Using specialized template: Ministral/Magistral Large 3\n");
-        return common_chat_params_init_ministral_3;
-    }
-
-    // LLM-jp-4.1 - GPT-OSS dialect (spaces after special tokens, <|end|>-separated parallel calls)
-    if (src.find("chat_format=llm-jp-harmony-v1") != std::string::npos) {
-        LOG_DBG("Using specialized template: LLM-jp Harmony v1\n");
-        return common_chat_params_init_llm_jp_harmony;
-    }
-
-    // Harmony XML tool calls, checked before the GPT-OSS JSON calls
-    if (src.find("<|channel|>") != std::string::npos && src.find("<|constrain|>xml") != std::string::npos &&
-        src.find("rfl:parameter") != std::string::npos) {
-        LOG_DBG("Using specialized template: Harmony XML\n");
-        return common_chat_params_init_harmony_xml;
+        return common_chat_params_init_ministral_3(tmpl, params);
    }

    // GPT-OSS - has unique channel-based structure that needs dedicated handler
    if (src.find("<|channel|>") != std::string::npos) {
        LOG_DBG("Using specialized template: GPT-OSS\n");
-        return common_chat_params_init_gpt_oss;
-    }
-
-    // Muse Spark shares the Muse recipient protocol and adds a commentary recipient.
-    if (src.find("to=commentary") != std::string::npos && src.find("<|eom|>") != std::string::npos) {
-        LOG_DBG("Using specialized template: Muse Spark\n");
-        return common_chat_params_init_muse_spark;
+        return common_chat_params_init_gpt_oss(tmpl, params);
    }

    // Muse Glimmer format using " to=<recipient>" recipients and <|eom|>/<|eot|> message terminators.
    if (src.find("<atem:function_calls>") != std::string::npos && src.find("<|eom|>") != std::string::npos) {
        LOG_DBG("Using specialized template: Muse Glimmer\n");
-        return common_chat_params_init_muse_glimmer;
+        return common_chat_params_init_muse_glimmer(tmpl, params);
    }

    // Functionary v3.2 - uses recipient-based format with >>>recipient\n{content}
    // Detection: template has ">>>all" for content and ">>>" prefix for tool calls
    if (src.find(">>>all") != std::string::npos && src.find(">>>${recipient}") != std::string::npos) {
        LOG_DBG("Using specialized template: Functionary v3.2\n");
-        return common_chat_params_init_functionary_v3_2;
+        return common_chat_params_init_functionary_v3_2(tmpl, params);
    }

    // Kimi K2 Thinking - uses unique tool call ID format: functions.<name>:<index>
@@ -1141,30 +1123,14 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
    if (src.find("<|tool_calls_section_begin|>") != std::string::npos &&
        src.find("<|tool_call_begin|>") != std::string::npos) {
        LOG_DBG("Using specialized template: Kimi K2 Thinking\n");
-        return common_chat_params_init_kimi_k2;
+        return common_chat_params_init_kimi_k2(tmpl, params);
    }

    // Kimi K3 - the <|open|>/<|close|>/<|end_of_msg|> markers are unique to it
    if (src.find("<|open|>") != std::string::npos && src.find("<|close|>") != std::string::npos &&
        src.find("<|end_of_msg|>") != std::string::npos) {
        LOG_DBG("Using specialized template: Kimi K3\n");
-        return common_chat_params_init_kimi_k3;
-    }
-
-    // K2 Horizon - <|ifm|im_start|> turns, <ifm|think*> reasoning picked by reasoning_effort and
-    // <ifm|tool_calls> sections; the three think tag pairs defeat the autoparser's reasoning detection
-    if (src.find("<|ifm|im_start|>") != std::string::npos &&
-        src.find("<ifm|tool_calls>") != std::string::npos) {
-        LOG_DBG("Using specialized template: K2 Horizon\n");
-        return common_chat_params_init_k2_horizon;
-    }
-
-    // Ling 3.0 / Bailing V3 - <role>X</role> sections with <arg_key>/<arg_value> tagged
-    // tool calls. <role> sections are unique to this family among the tagged-arg templates.
-    if (src.find("<role>ASSISTANT</role>") != std::string::npos &&
-        src.find("<arg_key>") != std::string::npos) {
-        LOG_DBG("Using specialized template: Ling 3.0 (Bailing V3)\n");
-        return common_chat_params_init_ling3;
+        return common_chat_params_init_kimi_k3(tmpl, params);
    }

    // Cohere2 MoE / North Code - marker-wrapped format with <|START_TEXT|> content and
@@ -1173,19 +1139,19 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
    if (src.find("<|START_TEXT|>") != std::string::npos &&
        src.find("<|START_ACTION|>") != std::string::npos) {
        LOG_DBG("Using specialized template: Cohere2 MoE\n");
-        return common_chat_params_init_cohere2moe;
+        return common_chat_params_init_cohere2moe(tmpl, params);
    }

    if (is_lfm2_template(src)) {
        LOG_DBG("Using specialized template: LFM2\n");
-        return common_chat_params_init_lfm2_tokens;
+        return common_chat_params_init_lfm2(tmpl, params, /* tool_list_tokens = */ true);
    }

    // LFM2.5 format detection: template uses plain "List of tools: [...]" with no special tokens
    if (src.find("List of tools: [") != std::string::npos &&
        src.find("<|tool_list_start|>") == std::string::npos) {
        LOG_DBG("Using specialized template: LFM2.5\n");
-        return common_chat_params_init_lfm2_5;
+        return common_chat_params_init_lfm2(tmpl, params, /* tool_list_tokens = */ false);
    }

    // GigaChatV3 format detection
@@ -1193,7 +1159,7 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
        src.find("<|message_sep|>") != std::string::npos &&
        src.find("<|function_call|>") == std::string::npos) {
        LOG_DBG("Using specialized template: GigaChatV3\n");
-        return common_chat_params_init_gigachat_v3;
+        return common_chat_params_init_gigachat_v3(tmpl, params);
    }

    // MiniMax-M3: the namespace token "]<]minimax[>[" collides with the autoparser's
@@ -1202,7 +1168,7 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
        src.find("<tool_call>") != std::string::npos &&
        src.find("<invoke name=") != std::string::npos) {
        LOG_DBG("Using specialized template: MiniMax-M3\n");
-        return common_chat_params_init_minimax_m3;
+        return common_chat_params_init_minimax_m3(tmpl, params);
    }

    // DeepSeek V3.2/V4 format detection: template defines dsml_token and uses it for tool calls.
@@ -1213,18 +1179,18 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
        (src.find("function_calls") != std::string::npos ||
         src.find("tool_calls") != std::string::npos)) {
        LOG_DBG("Using specialized template: DeepSeek V3.2/V4\n");
-        return common_chat_params_init_deepseek_v3_2;
+        return common_chat_params_init_deepseek_v3_2(tmpl, params);
    }

    // Gemma4 format detection
    if (src.find("'<|tool_call>call:'") != std::string::npos) {
-        LOG_DBG("Using specialized template: Gemma4\n");
        if (src.find("{#- OpenAI Chat Completions:") == std::string::npos) {
+            // apply workarounds if using the older gemma4 templates
            LOG_WRN("%s: detected an outdated gemma4 chat template, applying compatibility workarounds. "
                    "Consider updating to the official template.\n", __func__);
-            return common_chat_params_init_gemma4_legacy;
+            workaround::convert_tool_responses_gemma4(params.messages);
        }
-        return common_chat_params_init_gemma4;
+        return common_chat_params_init_gemma4(tmpl, params);
    }

    // MiniCPM5 - XML tool calls with <function name="..."><param name="...">...</param></function>
@@ -1232,68 +1198,18 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
        src.find("<function name=\"") != std::string::npos &&
        src.find("<param name=\"") != std::string::npos) {
        LOG_DBG("Using specialized template: MiniCPM5\n");
-        return common_chat_params_init_minicpm5;
-    }
-
-    // TranslateGemma - user content must follow a custom schema with language codes
-    if (src.find("[source_lang_code]") != std::string::npos &&
-        src.find("[target_lang_code]") != std::string::npos) {
-        LOG_DBG("Using specialized template: TranslateGemma\n");
-        return common_chat_params_init_translate_gemma;
+        return common_chat_params_init_minicpm5(tmpl, params);
    }

    // Qwen3-Coder XML tool calls, also used by Nemotron Nano 3, Qwen3.5 and StepFun-3.5-Flash
    if (src.find("<tool_call>") != std::string::npos &&
        src.find("<function=") != std::string::npos &&
-        src.find("<parameter=") != std::string::npos &&
-        // Exclude models that don't use \n between tags
-        src.find("'<tool_call><function=' ~ tool_call.name ~ '>'") == std::string::npos) {
+        src.find("<parameter=") != std::string::npos) {
        LOG_DBG("Using specialized template: Qwen3-Coder\n");
-        return common_chat_params_init_qwen3_coder;
+        return common_chat_params_init_qwen3_coder(tmpl, params);
    }

-    return nullptr;
-}
-
-common_chat_template::common_chat_template(const std::string & src, const std::string & bos_token, const std::string & eos_token) {
-    jinja::lexer lexer;
-    auto lexer_res = lexer.tokenize(src);
-    this->prog = jinja::parse_from_tokens(lexer_res);
-
-    this->src = lexer_res.source;
-    this->bos_tok = bos_token;
-    this->eos_tok = eos_token;
-
-    this->caps = jinja::caps_get(prog);
-    // LOG_INF("%s: caps:\n%s\n", __func__, this->caps.to_string().c_str());
-
-    this->params_init = common_chat_template_detect_params_init(this->src);
-    if (this->params_init) {
-        return;
-    }
-
-    // The analysis depends only on the template, so run it once here instead of on every apply.
-    // A failure is kept for apply to report, so a bad template still loads like it did before.
-    try {
-        analysis = std::make_unique<autoparser::autoparser>();
-        analysis->analyze_template(*this);
-    } catch (const std::exception & e) {
-        analysis.reset();
-        analysis_error = e.what();
-    }
-}
-
-common_chat_template::~common_chat_template() = default;
-common_chat_template::common_chat_template(common_chat_template &&) = default;
-common_chat_template & common_chat_template::operator=(common_chat_template &&) = default;
-
-std::optional<common_chat_params> common_chat_try_specialized_template(
-        const common_chat_template &          tmpl,
-        const autoparser::generation_params & params) {
-    if (!tmpl.params_init) {
-        return std::nullopt;
-    }
-    return tmpl.params_init(tmpl, params);
+    return std::nullopt;
 }

 static common_chat_params common_chat_templates_apply_jinja(const struct common_chat_templates *        tmpls,
@@ -1395,23 +1311,21 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
        data.prompt                    = common_chat_template_direct_apply_impl(tmpl, params_copy);
        data.generation_prompt         = common_chat_template_generation_prompt_impl(tmpl, params);
        data.format                    = COMMON_CHAT_FORMAT_PEG_NATIVE;
-        data.parser                    = build_chat_peg_parser([&data](common_chat_peg_builder &p) {
+        auto parser                    = build_chat_peg_parser([&data](common_chat_peg_builder &p) {
            return p.literal(data.generation_prompt) << p.content(p.rest());
        });
+        data.parser                    = parser.save();
        return data;
    }

-    if (auto result = common_chat_try_specialized_template(tmpl, params)) {
+    if (auto result = common_chat_try_specialized_template(tmpl, src, params)) {
        return *result;
    }

-    if (!tmpl.analysis) {
-        throw std::invalid_argument("Unable to generate parser for this template. Automatic parser generation failed: " + tmpl.analysis_error);
-    }
-
    try {
        LOG_DBG("%s: using differential autoparser\n", __func__);
-        const auto & autoparser = *tmpl.analysis;
+        struct autoparser::autoparser autoparser;
+        autoparser.analyze_template(tmpl);
        auto auto_params = autoparser::peg_generator::generate_parser(tmpl, params, autoparser);

        common_chat_msg_delimiters delimiters;
@@ -1432,7 +1346,8 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
                auto_params.thinking_end_tags = {std::move(end_tag)};
            }
        }
-        const auto & arena = auto_params.parser;
+        common_peg_arena arena;
+        arena.load(auto_params.parser);
        LOG_DBG("%s: generated parser:\n%s\n\nparser generation prompt: %s\n", __func__, arena.dump(arena.root()).c_str(), auto_params.generation_prompt.c_str());
        return auto_params;
    } catch (const std::exception & e) {
@@ -1497,7 +1412,7 @@ static common_chat_params common_chat_templates_apply_legacy(const struct common
    }

    common_chat_params params;
-    params.prompt = common_chat_prompt(std::string(buf.data(), res));
+    params.prompt = std::string(buf.data(), res);
    if (!inputs.json_schema.empty()) {
        params.grammar = json_schema_to_grammar(json::parse(inputs.json_schema));
    } else {
@@ -1513,132 +1428,36 @@ common_chat_params common_chat_templates_apply(const struct common_chat_template
                              common_chat_templates_apply_legacy(tmpls, inputs);
 }

-std::string common_chat_prompt::str() const {
-    if (parts.size() == 1) {
-        return parts[0].text;
-    }
-    size_t len = 0;
-    for (const auto & part : parts) {
-        len += part.text.size();
-    }
-    std::string out;
-    out.reserve(len);
-    for (const auto & part : parts) {
-        out += part.text;
-    }
-    return out;
-}
-
-void common_chat_prompt::append(const std::string & text, bool is_input) {
-    if (text.empty()) {
-        return;
-    }
-    if (!parts.empty() && parts.back().is_input == is_input) {
-        parts.back().text += text;
-    } else {
-        parts.push_back({text, is_input});
-    }
-}
-
-void common_chat_prompt::replace_last(const std::string & from, const std::string & to) {
-    for (auto it = parts.rbegin(); it != parts.rend(); ++it) {
-        if (it->is_input) {
-            continue;
-        }
-        size_t pos = it->text.rfind(from);
-        if (pos != std::string::npos) {
-            it->text.replace(pos, from.size(), to);
-            return;
-        }
-    }
-}
-
-void common_chat_input::append(const std::string & piece, llama_token token) {
-    if (piece.empty()) {
-        return;
-    }
-    tokens.push_back(token);
-    tokens.resize(tokens.size() + piece.size() - 1, LLAMA_TOKEN_NULL);
-    text += piece;
-}
-
-void common_chat_input::append(const common_chat_input & chunk) {
-    tokens.insert(tokens.end(), chunk.tokens.begin(), chunk.tokens.end());
-    text += chunk.text;
-}
-
-void common_chat_input::truncate(size_t pos) {
-    if (pos < text.size()) {
-        text.erase(pos);
-        tokens.resize(pos);
-    }
-}
-
-common_chat_input common_chat_input::substr(size_t pos, size_t n) const {
-    common_chat_input out;
-    out.text = text.substr(pos, n);
-    out.tokens.assign(tokens.begin() + pos, tokens.begin() + pos + out.size());
-    return out;
-}
-
-void common_chat_input::prepend(const std::string & prefix) {
-    tokens.insert(tokens.begin(), prefix.size(), LLAMA_TOKEN_NULL);
-    text = prefix + text;
-}
-
-void common_chat_input::prepend(const common_chat_input & prefix) {
-    tokens.insert(tokens.begin(), prefix.tokens.begin(), prefix.tokens.end());
-    text = prefix.text + text;
-}
-
-common_chat_input common_chat_input_tokenize(const llama_vocab * vocab, const std::string & text) {
-    common_chat_input input;
-    auto tokens = common_tokenize(vocab, text, false, true);
-    for (size_t i = 0; i < tokens.size(); i++) {
-        std::string piece = common_token_to_piece(vocab, tokens[i], true);
-        if (i == 0 && std::isspace(piece[0]) && !std::isspace(text[0])) {
-            // Some tokenizers will add a space before the first special token, need to exclude
-            continue;
-        }
-        input.append(piece, tokens[i]);
-    }
-    if (input.text != text) {
-        // the pieces do not give back the same text, keep the text without tokens
-        return common_chat_input(text);
-    }
-    return input;
-}
-
-common_chat_msg common_chat_parse(const common_chat_input &         input,
+common_chat_msg common_chat_parse(const std::string &               input,
                                  bool                              is_partial,
                                  const common_chat_parser_params & params) {
    return common_chat_peg_parse(params.parser, input, is_partial, params);
 }

 common_chat_msg common_chat_peg_parse(const common_peg_arena &          src_parser,
-                                      const common_chat_input &         input,
+                                      const std::string &               input,
                                      bool                              is_partial,
                                      const common_chat_parser_params & params) {
-    // both branches must be lvalues, a temporary here would copy the arena on every call
-    static const common_peg_arena content_only =
-        build_chat_peg_parser([](common_chat_peg_builder & p) { return p.content(p.rest()) + p.end(); });
-    const common_peg_arena & parser = src_parser.empty() ? content_only : src_parser;
+    const common_peg_arena & parser = src_parser.empty() ?
+        build_chat_peg_parser([](common_chat_peg_builder & p) { return p.content(p.rest()) + p.end(); }) :
+        src_parser;

    if (src_parser.empty()) {
        LOG_DBG("No parser definition detected, assuming pure content parser.");
    }

-    common_chat_input effective_input = input;
-    effective_input.prepend(params.generation_prompt);
+    const std::string effective_input = params.generation_prompt.empty()
+        ? input
+        : params.generation_prompt + input;

-    //LOG_DBG("Parsing PEG input with format %s: %s\n", common_chat_format_name(params.format), effective_input.text.c_str());
+    //LOG_DBG("Parsing PEG input with format %s: %s\n", common_chat_format_name(params.format), effective_input.c_str());

    common_peg_parse_flags flags = COMMON_PEG_PARSE_FLAG_LENIENT;
    if (params.debug) {
        flags |= COMMON_PEG_PARSE_FLAG_DEBUG;
    }

-    common_peg_parse_context ctx(std::move(effective_input.text), std::move(effective_input.tokens), flags);
+    common_peg_parse_context ctx(effective_input, flags);
    auto result = parser.parse(ctx);

    if (result.fail()) {
@@ -1664,8 +1483,8 @@ common_chat_msg common_chat_peg_parse(const common_peg_arena &          src_pars
            }
            return msg;
        }
-        LOG_WRN("%s: unparsed %s output: %s\n", __func__, common_chat_format_name(params.format), ctx.input.substr(result.end).c_str());
-        LOG_DBG("%s: full %s output triggering error:\n=== BEGIN ===\n%s\n=== END ===\n", __func__, common_chat_format_name(params.format), ctx.input.c_str());
+        LOG_WRN("%s: unparsed %s output: %s\n", __func__, common_chat_format_name(params.format), effective_input.substr(result.end).c_str());
+        LOG_DBG("%s: full %s output triggering error:\n=== BEGIN ===\n%s\n=== END ===\n", __func__, common_chat_format_name(params.format), effective_input.c_str());
        throw std::runtime_error(std::string("The model produced output that does not match the expected ") + common_chat_format_name(params.format) + " format");
    }

@@ -1693,84 +1512,6 @@ common_chat_msg common_chat_peg_parse(const common_peg_arena &          src_pars
    return msg;
 }

-common_chat_session::common_chat_session(const common_chat_templates *        tmpls,
-                                         const llama_vocab *                  vocab,
-                                         const common_chat_templates_inputs & inputs,
-                                         const common_chat_session_params &   params) {
-    auto applied = common_chat_templates_apply(tmpls, inputs);
-
-    templated    = true;
-    prompt_parts = std::move(applied.prompt);
-    result.role  = "assistant";
-
-    grammar_text           = std::move(applied.grammar);
-    grammar_lazy           = applied.grammar_lazy;
-    stops                  = std::move(applied.additional_stops);
-    generation_prompt_text = applied.generation_prompt;
-    thinking_start         = std::move(applied.thinking_start_tag);
-    thinking_ends          = std::move(applied.thinking_end_tags);
-
-    parser_params.format            = applied.format;
-    parser_params.generation_prompt = vocab ? common_chat_input_tokenize(vocab, applied.generation_prompt)
-                                            : common_chat_input(applied.generation_prompt);
-    parser_params.debug             = params.debug;
-    parser_params.parser            = std::move(applied.parser);
-
-    delimiters = std::move(applied.message_delimiters);
-
-    if (vocab) {
-        common_params_sampling resolved;
-        resolved.grammar_lazy = applied.grammar_lazy;
-        common_sampling_add_preserved_tokens(resolved, vocab, applied.preserved_tokens);
-        common_sampling_add_grammar_triggers(resolved, vocab, std::move(applied.grammar_triggers));
-        preserved_tokens = std::move(resolved.preserved_tokens);
-        grammar_triggers = std::move(resolved.grammar_triggers);
-
-        delimiters.tokenize(vocab);
-    } else {
-        grammar_triggers = std::move(applied.grammar_triggers);
-    }
-
-    if (inputs.continue_final_message != COMMON_CHAT_CONTINUATION_NONE && !params.echo) {
-        // start from the prefill so it is not emitted as part of the first delta
-        result = common_chat_parse(input, true, parser_params);
-    }
-}
-
-void common_chat_session::apply_sampling(common_params_sampling & sampling) const {
-    if (!templated) {
-        return;
-    }
-    if (!grammar_text.empty()) {
-        sampling.grammar = {COMMON_GRAMMAR_TYPE_TOOL_CALLS, grammar_text};
-    }
-    sampling.grammar_lazy      = grammar_lazy;
-    sampling.generation_prompt = generation_prompt_text;
-    sampling.preserved_tokens.insert(preserved_tokens.begin(), preserved_tokens.end());
-    sampling.grammar_triggers.insert(sampling.grammar_triggers.end(), grammar_triggers.begin(), grammar_triggers.end());
-}
-
-const common_chat_msg & common_chat_session::feed(const common_chat_input & chunk) {
-    GGML_ASSERT(!finished && "feed() after finish()");
-    input.append(chunk);
-    auto msg = common_chat_parse(input, true, parser_params);
-    if (!msg.empty()) {
-        result = std::move(msg);
-    }
-    return result;
-}
-
-const common_chat_msg & common_chat_session::finish(const common_chat_input & chunk) {
-    GGML_ASSERT(!finished && "finish() called twice");
-    finished = true;
-    input.append(chunk);
-    auto msg = common_chat_parse(input, false, parser_params);
-    if (!msg.empty()) {
-        result = std::move(msg);
-    }
-    return result;
-}
-
 std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_templates * chat_templates) {
    GGML_ASSERT(chat_templates != nullptr);
    GGML_ASSERT(chat_templates->template_default != nullptr);
@@ -22,16 +22,8 @@ struct common_chat_templates;

 namespace autoparser {
 struct generation_params;
-struct autoparser;
 }  // namespace autoparser

-struct common_chat_params;
-struct common_chat_template;
-
-// Builds the prompt and parser for a template that has a dedicated handler (see common/parsers)
-using common_chat_params_init_fn = common_chat_params (*)(const common_chat_template & tmpl,
-                                                          const autoparser::generation_params & inputs);
-
 struct common_chat_tool_call {
    std::string name;
    std::string arguments;
@@ -62,20 +54,19 @@ struct common_chat_template {
    std::string eos_tok;
    std::string src;
    chat_template_caps caps;
-    // Dedicated handler picked once from the source, null when the differential autoparser is used
-    common_chat_params_init_fn params_init = nullptr;

-    // Differential analysis, run once here when there is no dedicated handler. Null when there
-    // is one, or when the analysis failed, in which case analysis_error says why.
-    std::unique_ptr<autoparser::autoparser> analysis;
-    std::string analysis_error;
+    common_chat_template(const std::string & src, const std::string & bos_token, const std::string & eos_token) {
+        jinja::lexer lexer;
+        auto lexer_res = lexer.tokenize(src);
+        this->prog = jinja::parse_from_tokens(lexer_res);

-    common_chat_template(const std::string & src, const std::string & bos_token, const std::string & eos_token);
+        this->src = lexer_res.source;
+        this->bos_tok = bos_token;
+        this->eos_tok = eos_token;

-    // autoparser is incomplete here, so these are defined where it is complete
-    ~common_chat_template();
-    common_chat_template(common_chat_template &&);
-    common_chat_template & operator=(common_chat_template &&);
+        this->caps = jinja::caps_get(prog);
+        // LOG_INF("%s: caps:\n%s\n", __func__, this->caps.to_string().c_str());
+    }

    const std::string & source() const { return src; }
    const std::string & bos_token() const { return bos_tok; }
@@ -218,6 +209,8 @@ struct common_chat_msg_delimiters {

    // split tokens into message spans. skips maps a start index to a length of a region to jump over without matching
    common_chat_msg_spans split(const llama_tokens & tokens, const std::map<size_t, size_t> & skips = {}) const;
+
+    common_json to_json() const;
 };

 struct common_chat_tool {
@@ -273,31 +266,9 @@ struct common_chat_templates_inputs {
    bool                                  force_pure_content = false;
 };

-struct common_chat_prompt_part {
-    std::string text;
-    bool        is_input = false; // from user input, special tokens inside must not be parsed
-};
-
-// rendered chat template, split into template parts and user input parts
-struct common_chat_prompt {
-    std::vector<common_chat_prompt_part> parts;
-
-    common_chat_prompt() = default;
-
-    // single part of template text
-    explicit common_chat_prompt(std::string text) { parts.push_back({std::move(text), false}); }
-
-    std::string str() const;
-
-    void append(const std::string & text, bool is_input = false);
-
-    // replace the last occurrence of `from`, user input parts are skipped
-    void replace_last(const std::string & from, const std::string & to);
-};
-
 struct common_chat_params {
    common_chat_format                  format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
-    common_chat_prompt                  prompt;
+    std::string                         prompt;
    std::string                         grammar;
    bool                                grammar_lazy         = false;
    std::string                         generation_prompt;
@@ -307,46 +278,27 @@ struct common_chat_params {
    std::vector<common_grammar_trigger> grammar_triggers;
    std::vector<std::string>            preserved_tokens;
    std::vector<std::string>            additional_stops;
-    common_peg_arena                    parser;
+    std::string                         parser;
    common_chat_msg_delimiters          message_delimiters;
 };

-struct common_chat_input {
-    std::string              text;
-    std::vector<llama_token> tokens;
-
-    common_chat_input() = default;
-
-    // plain text, with no tokens
-    explicit common_chat_input(std::string text) : text(std::move(text)), tokens(this->text.size(), LLAMA_TOKEN_NULL) {}
-
-    size_t size() const { return text.size(); }
-    bool empty() const { return text.empty(); }
-
-    void append(const std::string & piece, llama_token token);
-    void append(const common_chat_input & chunk);
-
-    void prepend(const std::string & prefix);
-    void prepend(const common_chat_input & prefix);
-
-    void truncate(size_t pos);
-
-    common_chat_input substr(size_t pos, size_t n = std::string::npos) const;
-};
-
-common_chat_input common_chat_input_tokenize(const llama_vocab * vocab, const std::string & text);
-
 // per-message parsing syntax
 // should be derived from common_chat_params
 struct common_chat_parser_params {
-    common_chat_format format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
-    common_chat_input  generation_prompt;
-    bool               debug  = false; // Enable debug output for PEG parser
-    common_peg_arena   parser = {};
+    common_chat_format      format               = COMMON_CHAT_FORMAT_CONTENT_ONLY;
+    common_reasoning_format reasoning_format     = COMMON_REASONING_FORMAT_NONE; // TODO: refactor this to "bool parse_reasoning"
+    // Whether reasoning_content should be inlined in the content (e.g. for reasoning_format=deepseek in stream mode)
+    bool                    reasoning_in_content = false;
+    std::string             generation_prompt;
+    bool                    parse_tool_calls     = true;
+    bool                    is_continuation      = false;
+    bool                    echo                 = false;  // Include assistant prefilled msg in output
+    bool                    debug                = false;  // Enable debug output for PEG parser
+    common_peg_arena        parser               = {};
    common_chat_parser_params() = default;
    common_chat_parser_params(const common_chat_params & chat_params) {
        format  = chat_params.format;
-        generation_prompt = common_chat_input(chat_params.generation_prompt);
+        generation_prompt = chat_params.generation_prompt;
    }
 };

@@ -385,62 +337,8 @@ std::string common_chat_format_example(const struct common_chat_templates *
                                       const std::map<std::string, std::string> & chat_template_kwargs);

 const char *    common_chat_format_name(common_chat_format format);
-common_chat_msg common_chat_parse(const common_chat_input & input, bool is_partial, const common_chat_parser_params & params);
-common_chat_msg common_chat_peg_parse(const common_peg_arena & src_parser, const common_chat_input & input, bool is_partial, const common_chat_parser_params & params);
-
-struct common_chat_session_params {
-    bool echo  = false; // include the assistant prefill in the output when continuing a message
-    bool debug = false; // enable debug output for the PEG parser
-};
-
-class common_chat_session {
-  public:
-    common_chat_session() { result.role = "assistant"; }
-
-    common_chat_session(const common_chat_templates *        tmpls,
-                        const llama_vocab *                  vocab,
-                        const common_chat_templates_inputs & inputs,
-                        const common_chat_session_params &   params = {});
-
-    const common_chat_prompt & prompt() const { return prompt_parts; }
-    common_chat_format       format()   const { return parser_params.format; }
-    const common_chat_msg &  msg()      const { return result; }
-    const common_peg_arena & parser()   const { return parser_params.parser; }
-
-    const std::string &              grammar()            const { return grammar_text; }
-    const std::string &              generation_prompt()  const { return generation_prompt_text; }
-    const std::string &              thinking_start_tag() const { return thinking_start; }
-    const std::vector<std::string> & thinking_end_tags()  const { return thinking_ends; }
-    const std::vector<std::string> & additional_stops()   const { return stops; }
-
-    const common_chat_msg_delimiters & message_delimiters() const { return delimiters; }
-
-    void apply_sampling(common_params_sampling & sampling) const;
-
-    bool has_template() const { return templated; }
-
-    const common_chat_msg & feed(const common_chat_input & chunk);
-
-    const common_chat_msg & finish(const common_chat_input & chunk = {});
-
-  private:
-    common_chat_prompt                  prompt_parts;
-    std::string                         grammar_text;
-    bool                                grammar_lazy = false;
-    std::vector<common_grammar_trigger> grammar_triggers;
-    std::set<llama_token>               preserved_tokens;
-    std::vector<std::string>            stops;
-    std::string                         generation_prompt_text;
-    std::string                         thinking_start;
-    std::vector<std::string>            thinking_ends;
-
-    common_chat_parser_params  parser_params;
-    common_chat_msg_delimiters delimiters;
-    common_chat_input          input;
-    common_chat_msg            result;
-    bool                       templated = false;
-    bool                       finished  = false;
-};
+common_chat_msg common_chat_parse(const std::string & input, bool is_partial, const common_chat_parser_params & params);
+common_chat_msg common_chat_peg_parse(const common_peg_arena & src_parser, const std::string & input, bool is_partial, const common_chat_parser_params & params);

 // used by arg and server
 const char *            common_reasoning_format_name(common_reasoning_format format);
@@ -468,7 +366,7 @@ common_json common_chat_tool_parameters(const common_json & function);
 // get template caps, useful for reporting to server /props endpoint
 std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_templates * chat_templates);

-common_chat_prompt common_chat_template_direct_apply(
+std::string common_chat_template_direct_apply(
    const common_chat_template & tmpl,
    const autoparser::generation_params & inputs);

@@ -478,7 +376,8 @@ std::string common_chat_template_generation_prompt(

 std::optional<common_chat_params> common_chat_try_specialized_template(
        const common_chat_template &          tmpl,
-        const autoparser::generation_params & params);
+        const std::string &                   src,
+        autoparser::generation_params & params);


 // specialized per-task preset
@@ -488,3 +387,5 @@ struct common_chat_prompt_preset {
 };

 common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates);
+
+common_chat_msg_delimiters common_chat_msg_delimiters_parse(const common_json & delimiters);
@@ -1,12 +1,8 @@
 #include "ggml.h"
-#include "ggml-cpp.h"
 #include "gguf.h"

 #include "build-info.h"
 #include "common.h"
-
-#include "../src/llama-ext.h"
-
 #include "fit.h"
 #include "log.h"
 #include "llama.h"
@@ -26,7 +22,6 @@
 #include <fstream>
 #include <iostream>
 #include <iterator>
-#include <random>
 #include <regex>
 #include <sstream>
 #include <string>
@@ -51,10 +46,11 @@
 #include <io.h>
 #else
 #include <sys/ioctl.h>
+#include <sys/stat.h>
 #include <unistd.h>
 #endif

-#if !defined(_WIN32)
+#if defined(__linux__)
 #include <sys/types.h>
 #include <pwd.h>
 #endif
@@ -618,6 +614,34 @@ std::string string_from(const struct llama_context * ctx, const std::vector<llam
    return buf.str();
 }

+std::string string_from(const struct llama_context * ctx, const struct llama_batch & batch) {
+    std::stringstream buf;
+
+    buf << "[ ";
+
+    bool first = true;
+    for (int i = 0; i < batch.n_tokens; ++i) {
+        if (!first) {
+            buf << ", ";
+        } else {
+            first = false;
+        }
+
+        auto detokenized = common_token_to_piece(ctx, batch.token[i]);
+
+        buf << "\n"          << std::to_string(i)
+            << ", token '"   << detokenized << "'"
+            << ", pos "      << std::to_string(batch.pos[i])
+            << ", n_seq_id " << std::to_string(batch.n_seq_id[i])
+            << ", seq_id "   << std::to_string(batch.seq_id[i][0])
+            << ", logits "   << std::to_string(batch.logits[i]);
+    }
+
+    buf << " ]";
+
+    return buf.str();
+}
+
 void string_process_escapes(std::string & input) {
    std::size_t input_len = input.length();
    std::size_t output_idx = 0;
@@ -876,7 +900,7 @@ bool fs_validate_filename(const std::string & filename, bool allow_subdirs) {


 #ifdef _WIN32
-std::wstring utf8_to_wstring(const std::string & str) {
+static std::wstring utf8_to_wstring(const std::string & str) {
    if (str.empty()) {
        return std::wstring();
    }
@@ -892,56 +916,82 @@ std::wstring utf8_to_wstring(const std::string & str) {

    return wstr;
 }
-
-std::string wstring_to_utf8(const std::wstring & str) {
-    if (str.empty()) {
-        return std::string();
-    }
-
-    int size = WideCharToMultiByte(CP_UTF8, 0, str.c_str(), (int)str.size(), NULL, 0, NULL, NULL);
-
-    if (size <= 0) {
-        return std::string();
-    }
-
-    std::string utf8(size, 0);
-    WideCharToMultiByte(CP_UTF8, 0, str.c_str(), (int)str.size(), &utf8[0], size, NULL, NULL);
-
-    return utf8;
-}
 #endif

-// returns the path as a UTF-8 string, preserving its separators
-std::string fs_path_to_utf8(const std::filesystem::path & path) {
-    const auto value = path.u8string();
-    return std::string(value.begin(), value.end());
-}
+// returns true if successful, false otherwise
+bool fs_create_directory_with_parents(const std::string & path) {
+#ifdef _WIN32
+    std::wstring wpath = utf8_to_wstring(path);

-void fs_write_atomic(const std::filesystem::path & path, const std::string & data) {
-    // a random suffix gives every writer its own temporary file, across processes and containers
-    std::random_device rd;
-    const uint64_t id = ((uint64_t) rd() << 32) | rd();
-
-    std::error_code ec;
-    std::filesystem::path path_tmp = path;
-    path_tmp += "." + std::to_string(id) + ".tmp";
-
-    if (path.has_parent_path()) {
-        std::filesystem::create_directories(path.parent_path(), ec);
+    // if the path already exists, check whether it's a directory
+    const DWORD attributes = GetFileAttributesW(wpath.c_str());
+    if ((attributes != INVALID_FILE_ATTRIBUTES) && (attributes & FILE_ATTRIBUTE_DIRECTORY)) {
+        return true;
    }

-    std::ofstream file(path_tmp, std::ios::binary);
-    file << data;
-    file.close();
+    size_t pos_slash = 0;

-    if (!file.fail()) {
-        std::filesystem::rename(path_tmp, path, ec);
+    // process path from front to back, procedurally creating directories
+    while ((pos_slash = path.find('\\', pos_slash)) != std::string::npos) {
+        const std::wstring subpath = wpath.substr(0, pos_slash);
+
+        pos_slash += 1;
+
+        // skip the drive letter, in some systems it can return an access denied error
+        if (subpath.length() == 2 && subpath[1] == ':') {
+            continue;
+        }
+
+        const bool success = CreateDirectoryW(subpath.c_str(), NULL);
+
+        if (!success) {
+            const DWORD error = GetLastError();
+
+            // if the path already exists, ensure that it's a directory
+            if (error == ERROR_ALREADY_EXISTS) {
+                const DWORD attributes = GetFileAttributesW(subpath.c_str());
+                if (attributes == INVALID_FILE_ATTRIBUTES || !(attributes & FILE_ATTRIBUTE_DIRECTORY)) {
+                    return false;
+                }
+            } else {
+                return false;
+            }
+        }
    }

-    if (file.fail() || ec) {
-        std::filesystem::remove(path_tmp, ec);
-        throw std::runtime_error("failed to write file: " + fs_path_to_utf8(path));
+    return true;
+#else
+    // if the path already exists, check whether it's a directory
+    struct stat info;
+    if (stat(path.c_str(), &info) == 0) {
+        return S_ISDIR(info.st_mode);
    }
+
+    size_t pos_slash = 1; // skip leading slashes for directory creation
+
+    // process path from front to back, procedurally creating directories
+    while ((pos_slash = path.find('/', pos_slash)) != std::string::npos) {
+        const std::string subpath = path.substr(0, pos_slash);
+        struct stat info;
+
+        // if the path already exists, ensure that it's a directory
+        if (stat(subpath.c_str(), &info) == 0) {
+            if (!S_ISDIR(info.st_mode)) {
+                return false;
+            }
+        } else {
+            // create parent directories
+            const int ret = mkdir(subpath.c_str(), 0755);
+            if (ret != 0) {
+                return false;
+            }
+        }
+
+        pos_slash += 1;
+    }
+
+    return true;
+#endif // _WIN32
 }

 bool fs_is_directory(const std::string & path) {
@@ -966,91 +1016,172 @@ void common_set_env(const std::string & name, const std::string & value) {
 #endif
 }

-std::filesystem::path common_get_path_from_env(const std::string & name) {
-#if defined(_WIN32)
-    const std::wstring wname = utf8_to_wstring(name);
-    const wchar_t * wvalue = _wgetenv(wname.c_str());
-    return wvalue ? std::filesystem::path(wvalue) : std::filesystem::path();
-#else
-    const char * value = std::getenv(name.c_str());
-    return value ? std::filesystem::path(value) : std::filesystem::path();
-#endif
-}
-
-#if !defined(_WIN32)
-static std::filesystem::path get_home_directory() {
-    std::filesystem::path home = common_get_path_from_env("HOME");
-    if (!home.empty()) {
-        return home;
-    }
-    const struct passwd * pw = getpwuid(getuid());
-    if (!pw || !pw->pw_dir || !*pw->pw_dir) {
-        throw std::runtime_error("Failed to find $HOME directory");
-    }
-    return pw->pw_dir;
-}
-#endif
-
-std::filesystem::path fs_get_cache_directory() {
-    std::filesystem::path cache_directory = common_get_path_from_env("LLAMA_CACHE");
-    if (!cache_directory.empty()) {
-        return cache_directory;
-    }
-#if defined(_WIN32)
-    cache_directory = common_get_path_from_env("LOCALAPPDATA");
+std::string fs_get_cache_directory() {
+    std::string cache_directory = "";
+    auto ensure_trailing_slash = [](std::string p) {
+        // Make sure to add trailing slash
+        if (p.empty() || p.back() != DIRECTORY_SEPARATOR) {
+            p += DIRECTORY_SEPARATOR;
+        }
+        return p;
+    };
+    cache_directory = common_get_env("LLAMA_CACHE");
    if (cache_directory.empty()) {
-        throw std::runtime_error("Failed to find %LOCALAPPDATA% directory");
-    }
+#if defined(__linux__) || defined(__FreeBSD__) || defined(_AIX) || \
+        defined(__OpenBSD__) || defined(__NetBSD__)
+        const std::string xdg_cache_home = common_get_env("XDG_CACHE_HOME");
+        const std::string home           = common_get_env("HOME");
+        if (!xdg_cache_home.empty()) {
+            cache_directory = xdg_cache_home;
+        } else if (!home.empty()) {
+            cache_directory = home + "/.cache/";
+        } else {
+#if defined(__linux__)
+            /* no $HOME is defined, fallback to getpwuid */
+            struct passwd *pw = getpwuid(getuid());
+            if ((!pw) || (!pw->pw_dir)) {
+                throw std::runtime_error("Failed to find $HOME directory");
+            }
+
+            cache_directory = std::string(pw->pw_dir) + std::string("/.cache/");
+#else /* defined(__linux__) */
+            throw std::runtime_error("Failed to find $HOME directory");
+#endif /* defined(__linux__) */
+        }
 #elif defined(__APPLE__)
-    cache_directory = get_home_directory() / "Library/Caches";
+        cache_directory = common_get_env("HOME");
+        if (cache_directory.empty()) {
+            throw std::runtime_error("Failed to find $HOME directory");
+        }
+        cache_directory += "/Library/Caches/";
+#elif defined(_WIN32)
+        cache_directory = common_get_env("LOCALAPPDATA");
+        if (cache_directory.empty()) {
+            throw std::runtime_error("Failed to find %LOCALAPPDATA% directory");
+        }
+#elif defined(__EMSCRIPTEN__)
+        GGML_ABORT("not implemented on this platform");
 #else
-    cache_directory = common_get_path_from_env("XDG_CACHE_HOME");
-    if (cache_directory.empty()) {
-        cache_directory = get_home_directory() / ".cache";
-    }
+#  error Unknown architecture
 #endif
-    return cache_directory / "llama.cpp";
+        cache_directory = ensure_trailing_slash(cache_directory);
+        cache_directory += "llama.cpp";
+    }
+    return ensure_trailing_slash(cache_directory);
 }

-std::filesystem::path fs_get_config_directory() {
-    std::filesystem::path config_directory;
-#if defined(_WIN32)
-    config_directory = common_get_path_from_env("APPDATA");
+std::string fs_get_config_directory() {
+    std::string config_directory = "";
+    auto ensure_trailing_slash = [](std::string p) {
+        if (p.empty() || p.back() != DIRECTORY_SEPARATOR) {
+            p += DIRECTORY_SEPARATOR;
+        }
+        return p;
+    };
+#if defined(__linux__) || defined(__FreeBSD__) || defined(_AIX) || \
+        defined(__OpenBSD__) || defined(__NetBSD__) || defined(__APPLE__)
+    const std::string xdg_config_home = common_get_env("XDG_CONFIG_HOME");
+    const std::string home            = common_get_env("HOME");
+    if (!xdg_config_home.empty()) {
+        config_directory = xdg_config_home;
+    } else if (!home.empty()) {
+        config_directory = home + "/.config/";
+    } else {
+#if defined(__linux__)
+        /* no $HOME is defined, fallback to getpwuid */
+        struct passwd *pw = getpwuid(getuid());
+        if ((!pw) || (!pw->pw_dir)) {
+            throw std::runtime_error("Failed to find $HOME directory");
+        }
+
+        config_directory = std::string(pw->pw_dir) + std::string("/.config/");
+#else
+        throw std::runtime_error("Failed to find $HOME directory");
+#endif
+    }
+#elif defined(_WIN32)
+    config_directory = common_get_env("APPDATA");
    if (config_directory.empty()) {
        throw std::runtime_error("Failed to find %APPDATA% directory");
    }
+#elif defined(__EMSCRIPTEN__)
+    // caller decides what to do when there is no config directory
+    throw std::runtime_error("not implemented on this platform");
 #else
-    config_directory = common_get_path_from_env("XDG_CONFIG_HOME");
-    if (config_directory.empty()) {
-        config_directory = get_home_directory() / ".config";
-    }
+#  error Unknown architecture
 #endif
-    return config_directory / "llama.cpp";
+    config_directory = ensure_trailing_slash(config_directory);
+    config_directory += "llama.cpp";
+    return ensure_trailing_slash(config_directory);
 }

-std::filesystem::path fs_get_cache_file(const std::string & filename) {
+std::string fs_get_cache_file(const std::string & filename) {
    GGML_ASSERT(filename.find(DIRECTORY_SEPARATOR) == std::string::npos);
-    const std::filesystem::path cache_directory = fs_get_cache_directory();
-    std::error_code ec;
-    common_create_directories(cache_directory, ec);
-    if (ec) {
-        throw std::runtime_error("failed to create cache directory: " + fs_path_to_utf8(cache_directory));
+    std::string cache_directory = fs_get_cache_directory();
+    const bool success = fs_create_directory_with_parents(cache_directory);
+    if (!success) {
+        throw std::runtime_error("failed to create cache directory: " + cache_directory);
    }
-    return cache_directory / std::filesystem::u8path(filename);
+    return cache_directory + filename;
+}
+
+std::vector<common_file_info> fs_list(const std::string & path, bool include_directories) {
+    std::vector<common_file_info> files;
+    if (path.empty()) return files;
+
+    std::filesystem::path dir(path);
+    if (!std::filesystem::exists(dir) || !std::filesystem::is_directory(dir)) {
+        return files;
+    }
+
+    for (const auto & entry : std::filesystem::directory_iterator(dir)) {
+        try {
+            // Only include regular files (skip directories)
+            const auto & p = entry.path();
+            if (std::filesystem::is_regular_file(p)) {
+                common_file_info info;
+                info.path   = p.string();
+                info.name   = p.filename().string();
+                info.is_dir = false;
+                try {
+                    info.size = static_cast<size_t>(std::filesystem::file_size(p));
+                } catch (const std::filesystem::filesystem_error &) {
+                    info.size = 0;
+                }
+                files.push_back(std::move(info));
+            } else if (include_directories && std::filesystem::is_directory(p)) {
+                common_file_info info;
+                info.path   = p.string();
+                info.name   = p.filename().string();
+                info.size   = 0; // Directories have no size
+                info.is_dir = true;
+                files.push_back(std::move(info));
+            }
+        } catch (const std::filesystem::filesystem_error &) {
+            // skip entries we cannot inspect
+            continue;
+        }
+    }
+
+    return files;
+}
+
+std::ifstream fs_open_ifstream(const std::string & fname, std::ios_base::openmode mode) {
+#ifdef _WIN32
+    int wlen = MultiByteToWideChar(CP_UTF8, 0, fname.c_str(), -1, NULL, 0);
+    if (!wlen) { return std::ifstream(); }
+    std::vector<wchar_t> wfname(wlen);
+    (void)MultiByteToWideChar(CP_UTF8, 0, fname.c_str(), -1, wfname.data(), wlen);
+    return std::ifstream(wfname.data(), mode);
+#else
+    return std::ifstream(fname, mode);
+#endif
 }

 //
 // TTY utils
 //

-bool common_is_tty(FILE * file) {
-#if defined(_WIN32)
-    return _isatty(_fileno(file));
-#else
-    return isatty(fileno(file));
-#endif
-}
-
 bool tty_can_use_colors() {
    // Check NO_COLOR environment variable (https://no-color.org/)
    if (const char * no_color = std::getenv("NO_COLOR")) {
@@ -1068,21 +1199,10 @@ bool tty_can_use_colors() {

    // Check if stdout and stderr are connected to a terminal
    // We check both because log messages can go to either
-    return common_is_tty(stdout) || common_is_tty(stderr);
-}
+    bool stdout_is_tty = isatty(fileno(stdout));
+    bool stderr_is_tty = isatty(fileno(stderr));

-bool tty_enable_ansi() {
-#if defined(_WIN32)
-    // a Windows console renders ANSI sequences only in virtual terminal mode, pipes and files take them as is
-    for (DWORD id : { STD_OUTPUT_HANDLE, STD_ERROR_HANDLE }) {
-        HANDLE h    = GetStdHandle(id);
-        DWORD  mode = 0;
-        if (GetConsoleMode(h, &mode) && !SetConsoleMode(h, mode | ENABLE_VIRTUAL_TERMINAL_PROCESSING)) {
-            return false;
-        }
-    }
-#endif
-    return true;
+    return stdout_is_tty || stderr_is_tty;
 }

 //
@@ -1167,85 +1287,6 @@ struct common_init_result::impl {
    std::vector<llama_sampler_seq_config> samplers_seq_config;
 };

-static const std::map<common_decision_type, std::string> COMMON_DECISION_TYPE_NAMES = {
-    { COMMON_DECISION_TYPE_OPENJEV,        "openjev"       },
-    { COMMON_DECISION_TYPE_LEV,            "lev"           },
-    { COMMON_DECISION_TYPE_KEV,            "kev"           },
-    { COMMON_DECISION_TYPE_NIMBLE,         "nimble"        },
-    { COMMON_DECISION_TYPE_LAYA,           "laya"          },
-    { COMMON_DECISION_TYPE_CLEF,           "clef"          },
-    { COMMON_DECISION_TYPE_PPLX_DECIDER,   "pplx-decider"  },
-    { COMMON_DECISION_TYPE_VLLM_DECISION2, "vllm-decision2" },
-    { COMMON_DECISION_TYPE_VLLM_D3,        "vllm-d3"        },
-    { COMMON_DECISION_TYPE_LFM2_D1,        "lfm2-d1"       },
-    { COMMON_DECISION_TYPE_LFM2_D1_OMNI,   "lfm2-d1-omni"  },
-    { COMMON_DECISION_TYPE_GRANITE_SWITCH, "granite-switch"},
-    { COMMON_DECISION_TYPE_H2O_LIGHTNING,  "h2o-lightning" },
-    // for future models, ALWAYS add org name prefix to avoid name collisions, e.g. "orgname-modelname"
-};
-
-static common_decision_type common_decision_type_from_string(const std::string & str) {
-    for (const auto & pair : COMMON_DECISION_TYPE_NAMES) {
-        if (pair.second == str) {
-            return pair.first;
-        }
-    }
-    return COMMON_DECISION_TYPE_UNKNOWN;
-}
-
-common_decision_type common_get_decision_type(const struct llama_model * model) {
-    char buf[64];
-    if (llama_model_meta_val_str(model, "general.architecture", buf, sizeof(buf)) < 0) {
-        return COMMON_DECISION_TYPE_NONE;
-    }
-    const std::string key = std::string(buf) + ".decision.type";
-    if (llama_model_meta_val_str(model, key.c_str(), buf, sizeof(buf)) < 0) {
-        return COMMON_DECISION_TYPE_NONE;
-    }
-    return common_decision_type_from_string(buf);
-}
-
-common_gguf_info common_get_gguf_info(const std::string & fname) {
-    common_gguf_info info;
-
-    struct gguf_init_params gguf_params = {
-        /* .no_alloc = */ true,
-        /* .ctx      = */ nullptr,
-    };
-
-    gguf_context_ptr gguf_ctx(gguf_init_from_file(fname.c_str(), gguf_params));
-    if (!gguf_ctx) {
-        return info; // missing or unreadable file
-    }
-
-    const int64_t arch_id = gguf_find_key(gguf_ctx.get(), "general.architecture");
-    if (arch_id < 0 || gguf_get_kv_type(gguf_ctx.get(), arch_id) != GGUF_TYPE_STRING) {
-        return info; // no architecture in the metadata
-    }
-    const std::string arch = gguf_get_val_str(gguf_ctx.get(), arch_id);
-    if (arch.empty()) {
-        return info;
-    }
-
-    const int64_t type_id = gguf_find_key(gguf_ctx.get(), (arch + ".decision.type").c_str());
-    if (type_id < 0) {
-        info.decision_type = COMMON_DECISION_TYPE_NONE;
-    } else if (gguf_get_kv_type(gguf_ctx.get(), type_id) == GGUF_TYPE_STRING) {
-        info.decision_type = common_decision_type_from_string(gguf_get_val_str(gguf_ctx.get(), type_id));
-    }
-
-    // same key and type as the model loader
-    const int64_t ctx_id = gguf_find_key(gguf_ctx.get(), (arch + ".context_length").c_str());
-    if (ctx_id >= 0 && gguf_get_kv_type(gguf_ctx.get(), ctx_id) == GGUF_TYPE_UINT32) {
-        info.n_ctx_train = gguf_get_val_u32(gguf_ctx.get(), ctx_id);
-    }
-
-    // same key as llama_model_chat_template(model, nullptr)
-    info.has_chat_tmpl = gguf_find_key(gguf_ctx.get(), "tokenizer.chat_template") >= 0;
-
-    return info;
-}
-
 common_init_result::common_init_result(common_params & params, bool model_only) :
    pimpl(new impl{}) {
    auto mparams = common_model_params_to_llama(params);
@@ -1298,30 +1339,6 @@ common_init_result::common_init_result(common_params & params, bool model_only)

    const llama_vocab * vocab = llama_model_get_vocab(model);

-    // these decision models return a score for each token via the embeddings output
-    // TODO: maybe improve this in the future
-    const auto decision_type = common_get_decision_type(model);
-    if (decision_type == COMMON_DECISION_TYPE_LAYA || decision_type == COMMON_DECISION_TYPE_KEV || decision_type == COMMON_DECISION_TYPE_CLEF ||
-        decision_type == COMMON_DECISION_TYPE_LFM2_D1_OMNI || decision_type == COMMON_DECISION_TYPE_VLLM_DECISION2) {
-        params.embedding    = true;
-        params.pooling_type = LLAMA_POOLING_TYPE_NONE;
-
-        cparams.embeddings            = true;
-        cparams.pooling_type          = LLAMA_POOLING_TYPE_NONE;
-        cparams.n_outputs_max         = cparams.n_batch;
-        cparams.n_outputs_max_per_seq = 1;
-
-        LOG_INF("%s", "decision model reads the embeddings output, enabling embedding mode\n");
-    }
-
-    // embeddings need the whole batch in one ubatch, so n_batch must not be larger than n_ubatch
-    // (server.cpp does this check for --embedding, but before the model is loaded)
-    if (cparams.embeddings && cparams.n_batch > cparams.n_ubatch) {
-        LOG_WRN("embeddings enabled: setting n_batch = n_ubatch = %u\n", cparams.n_ubatch);
-        cparams.n_batch = cparams.n_ubatch;
-        params.n_batch  = params.n_ubatch;
-    }
-
    // load and optionally apply lora adapters
    for (auto & la : params.lora_adapters) {
        llama_adapter_lora_ptr lora;
@@ -1416,14 +1433,6 @@ std::vector<llama_adapter_lora_ptr> & common_init_result::lora() {
    return pimpl->lora;
 }

-// only for warmup and probe decodes, fill zeros as dummy input
-static void common_batch_set_zero_state(common_batch & batch, const llama_model * model, std::vector<float> & zeros) {
-    zeros.assign(llama_model_n_embd_out(model), 0.0f);
-    for (int32_t i = 0; i < batch.size(); ++i) {
-        batch.set_embd_state(i, { zeros.data(), 1, zeros.size() });
-    }
-}
-
 common_init_result_ptr common_init_from_params(common_params & params, bool model_only) {
    common_init_result_ptr res(new common_init_result(params, model_only));

@@ -1518,8 +1527,7 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
        }

        if (llama_model_has_encoder(model)) {
-            common_batch batch = common_batch_get_one(lctx, tmp);
-            llama_process(lctx, LLAMA_PROCESS_TYPE_ENCODE, batch.get());
+            llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size()));
            llama_token decoder_start_token_id = llama_model_decoder_start_token(model);
            if (decoder_start_token_id == LLAMA_TOKEN_NULL) {
                decoder_start_token_id = bos;
@@ -1528,11 +1536,7 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
            tmp.push_back(decoder_start_token_id);
        }
        if (llama_model_has_decoder(model)) {
-            tmp.resize(std::min(tmp.size(), (size_t) params.n_batch));
-            common_batch batch = common_batch_get_one(lctx, tmp);
-            std::vector<float> zeros;
-            common_batch_set_zero_state(batch, model, zeros);
-            llama_process(lctx, LLAMA_PROCESS_TYPE_DECODE, batch.get());
+            llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch)));
        }
        llama_memory_clear(llama_get_memory(lctx), true);
        llama_synchronize(lctx);
@@ -1582,11 +1586,6 @@ common_context_seq_rm_type common_context_can_seq_rm(llama_context * ctx) {
        return COMMON_CONTEXT_SEQ_RM_TYPE_NO;
    }

-    if (llama_n_rs_seq(ctx) > 0) {
-        COM_TRC("%s", "the context supports bounded partial sequence removal\n");
-        return COMMON_CONTEXT_SEQ_RM_TYPE_RS;
-    }
-
    common_context_seq_rm_type res = COMMON_CONTEXT_SEQ_RM_TYPE_PART;

    llama_memory_clear(mem, true);
@@ -1596,19 +1595,19 @@ common_context_seq_rm_type common_context_can_seq_rm(llama_context * ctx) {
    tmp.push_back(0);
    tmp.push_back(0);

-    int ret;
-    {
-        common_batch batch = common_batch_get_one(ctx, tmp);
-        std::vector<float> zeros;
-        common_batch_set_zero_state(batch, llama_get_model(ctx), zeros);
-        ret = llama_process(ctx, LLAMA_PROCESS_TYPE_DECODE, batch.get());
-    }
+    int ret = llama_decode(ctx, llama_batch_get_one(tmp.data(), tmp.size()));
    if (ret != 0) {
-        COM_ERR("llama_process() failed: %d\n", ret);
+        COM_ERR("llama_decode() failed: %d\n", ret);
        res = COMMON_CONTEXT_SEQ_RM_TYPE_NO;
        goto done;
    }

+    if (llama_n_rs_seq(ctx) > 0) {
+        COM_TRC("%s", "the context supports bounded partial sequence removal\n");
+        res = COMMON_CONTEXT_SEQ_RM_TYPE_RS;
+        goto done;
+    }
+
    // try to remove the last tokens
    if (!llama_memory_seq_rm(mem, 0, 1, -1)) {
        COM_TRC("%s", "the context does not support partial sequence removal\n");
@@ -1712,7 +1711,7 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
    mparams.progress_callback           = params.load_progress_callback;
    mparams.progress_callback_user_data = params.load_progress_callback_user_data;
    mparams.no_alloc                    = params.no_alloc;
-    mparams.load_mtp                    = params.load_mtp || std::find(params.speculative.types.begin(), params.speculative.types.end(), COMMON_SPECULATIVE_TYPE_DRAFT_MTP) != params.speculative.types.end();
+    mparams.load_mtp                    = std::find(params.speculative.types.begin(), params.speculative.types.end(), COMMON_SPECULATIVE_TYPE_DRAFT_MTP) != params.speculative.types.end();

    return mparams;
 }
@@ -1753,8 +1752,6 @@ struct llama_context_params common_context_params_to_llama(const common_params &
    cparams.type_k = params.cache_type_k;
    cparams.type_v = params.cache_type_v;

-    cparams.moe_cache_size = params.moe_cache_size;
-
    return cparams;
 }

@@ -1830,6 +1827,33 @@ void common_threadpools::init(llama_context * ctx, const common_params & params)
    llama_attach_threadpool(ctx, threadpool, threadpool_batch);
 }

+//
+// Batch utils
+//
+
+void common_batch_clear(struct llama_batch & batch) {
+    batch.n_tokens = 0;
+}
+
+void common_batch_add(
+                 struct llama_batch & batch,
+                        llama_token   id,
+                          llama_pos   pos,
+    const std::vector<llama_seq_id> & seq_ids,
+                               bool   logits) {
+    GGML_ASSERT(batch.seq_id[batch.n_tokens] && "llama_batch size exceeded");
+
+    batch.token   [batch.n_tokens] = id;
+    batch.pos     [batch.n_tokens] = pos;
+    batch.n_seq_id[batch.n_tokens] = seq_ids.size();
+    for (size_t i = 0; i < seq_ids.size(); ++i) {
+        batch.seq_id[batch.n_tokens][i] = seq_ids[i];
+    }
+    batch.logits  [batch.n_tokens] = logits;
+
+    batch.n_tokens++;
+}
+
 //
 // Vocab utils
 //
@@ -2166,151 +2190,18 @@ float lr_opt::get_lr(float epoch) const {
 }

 bool common_replay_last_token(struct llama_context * ctx, llama_token last_token, int32_t pos) {
-    common_batch batch(ctx);
-    batch.add(last_token, pos, 0, true);
-
-    if (llama_process(ctx, LLAMA_PROCESS_TYPE_DECODE, batch.get())) {
+    llama_batch batch = llama_batch_get_one(&last_token, 1);
+    batch.pos = &pos;
+    if (llama_decode(ctx, batch)) {
        LOG_ERR("%s: failed to replay last token\n", __func__);
        return false;
    }
    return true;
 }

-common_batch::common_batch(llama_context * ctx) : batch(llama_batch_ext_init(ctx)) {
-    const auto rope_type = llama_model_rope_type(llama_get_model(ctx));
-    n_pos = rope_type == LLAMA_ROPE_TYPE_MROPE || rope_type == LLAMA_ROPE_TYPE_IMROPE ? GGML_MROPE_SECTIONS : 1;
-}
-
-void common_batch::clear() {
-    tokens.clear();
-}
-
-int32_t common_batch::add(llama_token id, llama_pos pos, llama_seq_id seq_id, bool output) {
-    tokens.push_back({ id, { pos, 0, 0, 0 }, seq_id, output, { nullptr, 0, 0 }, { nullptr, 0, 0 }, {} });
-    return size() - 1;
-}
-
-int32_t common_batch::add(llama_token id, llama_pos pos, const std::vector<llama_seq_id> & seq_ids, bool output) {
-    GGML_ASSERT(!seq_ids.empty());
-
-    const int32_t idx = add(id, pos, seq_ids[0], output);
-    for (size_t s = 1; s < seq_ids.size(); ++s) {
-        add_seq(idx, seq_ids[s]);
-    }
-    return idx;
-}
-
-bool common_batch::add_seq(int32_t idx, llama_seq_id seq_id) {
-    if (idx < 0 || idx >= size()) {
-        return false;
-    }
-    tokens[idx].seq_ids_extra.push_back(seq_id);
-    return true;
-}
-
-bool common_batch::set_output(int32_t idx, bool value) {
-    if (idx < 0 || idx >= size()) {
-        return false;
-    }
-    tokens[idx].output = value;
-    return true;
-}
-
-bool common_batch::set_embd(int32_t idx, llama_embd embd) {
-    if (idx < 0 || idx >= size() || tokens[idx].embd.data != nullptr) {
-        return false;
-    }
-    tokens[idx].embd = embd;
-    return true;
-}
-
-bool common_batch::set_embd_state(int32_t idx, llama_embd state) {
-    if (idx < 0 || idx >= size() || tokens[idx].state.data != nullptr) {
-        return false;
-    }
-    tokens[idx].state = state;
-    return true;
-}
-
-int32_t common_batch::add_embd(llama_embd embd, const llama_pos * pos, llama_seq_id seq_id, bool output) {
-    token t = { LLAMA_TOKEN_NULL, { 0, 0, 0, 0 }, seq_id, output, embd, { nullptr, 0, 0 }, {} };
-    for (int32_t j = 0; j < n_pos; ++j) {
-        t.pos[j] = pos[j];
-    }
-    tokens.push_back(t);
-    return size() - 1;
-}
-
-llama_batch_ext * common_batch::get_sub_batch(int32_t off, int32_t n) {
-    GGML_ASSERT(batch && "common_batch was not initialized with a context");
-    GGML_ASSERT(off >= 0 && n >= 0 && off + n <= size());
-
-    llama_batch_ext * res = batch.get();
-    llama_batch_ext_clear(res);
-
-    for (int32_t i = off; i < off + n; ++i) {
-        const token & t = tokens[i];
-
-        int32_t idx;
-        if (t.id != LLAMA_TOKEN_NULL) {
-            idx = llama_batch_ext_add_token(res, t.seq_id, t.id);
-            if (idx < 0) {
-                GGML_ABORT("%s: failed to add token %d at index %d (error %d, n = %d)\n", __func__, t.id, i, idx, n);
-            }
-            llama_batch_ext_set_pos(res, idx, t.pos.data());
-            if (t.embd.data && !llama_batch_ext_set_embd_token(res, idx, t.embd)) {
-                GGML_ABORT("%s: failed to set the embedding of token %d at index %d\n", __func__, t.id, i);
-            }
-        } else {
-            idx = llama_batch_ext_add_embd(res, t.seq_id, t.embd);
-            if (idx < 0) {
-                GGML_ABORT("%s: failed to add embedding at index %d (error %d, n = %d)\n", __func__, i, idx, n);
-            }
-            llama_batch_ext_set_pos(res, idx, t.pos.data());
-        }
-        GGML_ASSERT(idx == i - off);
-
-        for (const llama_seq_id seq_id : t.seq_ids_extra) {
-            if (!llama_batch_ext_add_seq(res, idx, seq_id)) {
-                GGML_ABORT("%s: failed to add seq %d to the entry at index %d\n", __func__, seq_id, i);
-            }
-        }
-        if (t.output) {
-            llama_batch_ext_set_output_logits(res, idx, true);
-        }
-        if (t.state.data) {
-            llama_batch_ext_set_embd_state(res, idx, t.state); // contexts without a state input ignore it
-        }
-        if (t.decision_order != 0) {
-            llama_batch_ext_set_decision_order(res, idx, (llama_decision_order) t.decision_order);
-        }
-    }
-
-    return res;
-}
-
-common_batch common_batch_get_one(llama_context * ctx, const llama_token * tokens, int32_t n_tokens) {
-    common_batch batch(ctx);
-
-    auto mem = llama_get_memory(ctx);
-    llama_pos pos = llama_memory_seq_pos_max(mem, 0) + 1; // -1 + 1 == 0 when the memory is empty
-
-    for (int32_t i = 0; i < n_tokens; ++i) {
-        const bool output = i == n_tokens - 1;
-        batch.add(tokens[i], pos, 0, output);
-        pos++;
-    }
-
-    return batch;
-}
-
-common_batch common_batch_get_one(llama_context * ctx, const llama_tokens & tokens) {
-    return common_batch_get_one(ctx, tokens.data(), (int32_t) tokens.size());
-}
-
 bool common_prompt_batch_decode(
              struct llama_context * ctx,
-                const llama_tokens & all_tokens,
+    const std::vector<llama_token> & all_tokens,
                               int   n_new,
                               int & n_past,
                               int   n_batch,
@@ -2331,9 +2222,7 @@ bool common_prompt_batch_decode(
        // Memory implementations in recurrent/hybrid models don't support removing tokens from their
        // memory, so we can't just remove the last token from the memory and replay the last token which
        // is the reason for this logic.
-        llama_tokens prefix_tokens(all_tokens.begin() + offset, all_tokens.begin() + offset + n_tokens_before_last);
-        common_batch batch_prefix = common_batch_get_one(ctx, prefix_tokens);
-        if (llama_process(ctx, LLAMA_PROCESS_TYPE_DECODE, batch_prefix.get())) {
+        if (llama_decode(ctx, llama_batch_get_one(const_cast<llama_token*>(all_tokens.data() + offset), n_tokens_before_last))) {
            COM_ERR("%s", "failed to eval\n");
            return false;
        }
@@ -2342,18 +2231,18 @@ bool common_prompt_batch_decode(
        llama_state_save_file(ctx, state_path.data(), all_tokens.data(), all_tokens.size());
        COM_INF("saved session before last token to %s, n_new = %zu\n", state_path.data(), all_tokens.size());

-        common_batch batch_last(ctx);
-        batch_last.add(all_tokens.back(), n_past, 0, true);
+        llama_token last_token = all_tokens.back();
+        llama_batch batch = llama_batch_get_one(&last_token, 1);
+        int32_t pos = n_past;
+        batch.pos = &pos;

-        if (llama_process(ctx, LLAMA_PROCESS_TYPE_DECODE, batch_last.get())) {
+        if (llama_decode(ctx, batch)) {
            COM_ERR("%s", "failed to eval last token\n");
            return false;
        }
        n_past++;
    } else {
-        llama_tokens new_tokens(all_tokens.begin() + offset, all_tokens.begin() + offset + n_new);
-        common_batch batch = common_batch_get_one(ctx, new_tokens);
-        if (llama_process(ctx, LLAMA_PROCESS_TYPE_DECODE, batch.get())) {
+        if (llama_decode(ctx, llama_batch_get_one(const_cast<llama_token*>(all_tokens.data() + offset), n_new))) {
            COM_ERR("%s", "failed to eval\n");
            return false;
        }
@@ -2427,36 +2316,40 @@ void common_prompt_checkpoint::update_dft(
    }
 }

-bool common_prompt_checkpoint::load_tgt(
+void common_prompt_checkpoint::load_tgt(
        llama_context * ctx,
        llama_seq_id seq_id,
        llama_state_seq_flags flags) const {
    if (ctx == nullptr) {
-        return true;
+        return;
    }

    if (data_tgt.empty()) {
-        return true;
+        return;
    }

    const size_t n = llama_state_seq_set_data_ext(ctx, data_tgt.data(), data_tgt.size(), seq_id, flags);
-    return n == data_tgt.size();
+    if (n != data_tgt.size()) {
+        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", data_tgt.size(), n);
+    }
 }

-bool common_prompt_checkpoint::load_dft(
+void common_prompt_checkpoint::load_dft(
        llama_context * ctx,
        llama_seq_id seq_id,
        llama_state_seq_flags flags) const {
    if (ctx == nullptr) {
-        return true;
+        return;
    }

    if (data_dft.empty()) {
-        return true;
+        return;
    }

    const size_t n = llama_state_seq_set_data_ext(ctx, data_dft.data(), data_dft.size(), seq_id, flags);
-    return n == data_dft.size();
+    if (n != data_dft.size()) {
+        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", data_dft.size(), n);
+    }
 }

 void common_prompt_checkpoint::clear_tgt() {
@@ -8,7 +8,6 @@
 #include "ggml.h"
 #include "llama.h"

-#include <array>
 #include <list>
 #include <set>
 #include <sstream>
@@ -17,9 +16,7 @@
 #include <vector>
 #include <map>
 #include <algorithm>
-#include <filesystem>
 #include <fstream>
-#include <cstdio>

 #if defined(_WIN32) && !defined(_WIN32_WINNT)
 #define _WIN32_WINNT 0x0A00
@@ -334,8 +331,6 @@ struct common_params_speculative_draft {

    bool backend_sampling = true; // offload draft sampling to the backend (default: on)

-    bool probabilistic = false; // sample the draft and verify by rejection, instead of argmax and match
-
    common_params_model mparams;

    llama_context * ctx_tgt = nullptr;
@@ -343,8 +338,6 @@ struct common_params_speculative_draft {

    int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)

-    int32_t n_ubatch = 0; // physical batch size of the draft context (0 - use the target's n_ubatch)
-
    ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
    ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V

@@ -588,15 +581,12 @@ struct common_params {
    bool no_op_offload     = false; // globally disable offload host tensor operations to device
    bool no_extra_bufts    = false; // disable extra buffer types (used for weight repacking)
    bool no_host           = false; // bypass host buffer allowing extra buffers to be used
-    bool load_mtp          = false; // load MTP/NextN layers

    bool single_turn       = false; // single turn chat conversation

    ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
    ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V

-    size_t moe_cache_size = 0; // GPU cache size in bytes for the MoE experts kept in the CPU, split among the GPUs like the layers
-
    common_conversation_mode conversation_mode = COMMON_CONVERSATION_MODE_AUTO;

    // multimodal models (see tools/mtmd)
@@ -627,7 +617,7 @@ struct common_params {
    std::string cls_sep    = "\t";  // separator of classification sequences

    // server params
-    int32_t port                = 9931;          // server listens on this network port
+    int32_t port                = 8080;          // server listens on this network port
    bool    reuse_port          = false;         // allow multiple sockets to bind to the same port
    int32_t timeout_read        = 3600;          // http read timeout in seconds
    int32_t timeout_write       = timeout_read;  // http write timeout in seconds
@@ -640,16 +630,11 @@ struct common_params {
    int32_t kv_unified_per_slot = 0;     // max context per parallel slot; 0 = unset
    int32_t checkpoint_min_step = 8192;  // minimum spacing between context checkpoints
    int32_t cache_ram_mib       = 8192;  // -1 = no limit, 0 - disable, 1 = 1 MiB, etc.
-    bool    cache_ram_explicit  = false; // set by --cache-ram, decision models disable the cache otherwise
-
-    // llama-connect params
-    bool        server_connect      = false;
-    std::string server_connect_code = "";

+    std::string hostname      = "127.0.0.1";
    std::string public_path   = "";                                                                         // NOLINT
    std::string api_prefix    = "";                                                                         // NOLINT
    std::string chat_template = "";                                                                         // NOLINT
-    std::vector<std::string> hostnames = {"127.0.0.1"};
    bool use_jinja = true;                                                                                  // NOLINT

    // server CORS params
@@ -734,11 +719,10 @@ struct common_params {
    int32_t i_chunk     =  0; // start processing from this chunk
    int8_t  imat_dat    =  0; // whether the legacy imatrix.dat format should be output (gguf <= 0 < dat)

-    bool process_output         = false; // collect data for the output tensor
-    bool compute_ppl            = true;  // whether to compute perplexity
-    bool show_statistics        = false; // show imatrix statistics per tensor
-    bool activation_statistics  = false; // generate data to calculate activation based statistics
-    bool parse_special          = false; // whether to parse special tokens during imatrix tokenization
+    bool process_output  = false; // collect data for the output tensor
+    bool compute_ppl     = true;  // whether to compute perplexity
+    bool show_statistics = false; // show imatrix statistics per tensor
+    bool parse_special   = false; // whether to parse special tokens during imatrix tokenization

    // cvector-generator params
    int n_pca_batch = 100;
@@ -824,9 +808,7 @@ static std::vector<T> string_split(const std::string & str, char delim) {
    while (std::getline(str_stream, token, delim)) {
        T value;
        std::istringstream token_stream(token);
-        if (!(token_stream >> value)) {
-            throw std::invalid_argument("invalid value: \"" + token + "\"");
-        }
+        token_stream >> value;
        values.push_back(value);
    }
    return values;
@@ -894,21 +876,10 @@ void string_process_escapes(std::string & input);
 std::string string_from(bool value);
 std::string string_from(const std::vector<int> & values);
 std::string string_from(const struct llama_context * ctx, const std::vector<llama_token> & tokens);
+std::string string_from(const struct llama_context * ctx, const struct llama_batch & batch);

 bool glob_match(const std::string & pattern, const std::string & str);

-//
-// Unicode utils
-//
-
-#ifdef _WIN32
-std::wstring utf8_to_wstring(const std::string & str);
-std::string  wstring_to_utf8(const std::wstring & str);
-#endif
-
-// returns the path as a UTF-8 string, preserving its separators
-std::string fs_path_to_utf8(const std::filesystem::path & path);
-
 //
 // Environment utils
 //
@@ -918,30 +889,28 @@ std::string fs_path_to_utf8(const std::filesystem::path & path);
 std::string common_get_env(const std::string & name);
 void        common_set_env(const std::string & name, const std::string & value);

-// reads a path from the environment, an unset variable gives an empty path
-std::filesystem::path common_get_path_from_env(const std::string & name);
-
 //
 // Filesystem utils
 //

 bool fs_validate_filename(const std::string & filename, bool allow_subdirs = false);
+bool fs_create_directory_with_parents(const std::string & path);
 bool fs_is_directory(const std::string & path);

-// some old libstdc++ versions don't follow symlinks here, so adding a trailing "/" fixes it: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101510
-inline bool common_create_directories(const std::filesystem::path & path, std::error_code & ec) {
-#if defined(__linux__)
-    return std::filesystem::create_directories(path / "", ec);
-#else
-    return std::filesystem::create_directories(path, ec);
-#endif
-}
+std::string fs_get_cache_directory();
+std::string fs_get_cache_file(const std::string & filename);
+std::string fs_get_config_directory();

-std::filesystem::path fs_get_cache_directory();
-std::filesystem::path fs_get_cache_file(const std::string & filename);
-std::filesystem::path fs_get_config_directory();
+struct common_file_info {
+    std::string path;
+    std::string name;
+    size_t      size = 0; // in bytes
+    bool        is_dir = false;
+};
+std::vector<common_file_info> fs_list(const std::string & path, bool include_directories);

-void fs_write_atomic(const std::filesystem::path & path, const std::string & data);
+// fs open, also handle UTF8 on Windows
+std::ifstream fs_open_ifstream(const std::string & fname, std::ios_base::openmode mode);

 //
 // TTY utils
@@ -949,10 +918,6 @@ void fs_write_atomic(const std::filesystem::path & path, const std::string & dat

 // Auto-detect if colors can be enabled based on terminal and environment
 bool tty_can_use_colors();
-bool tty_enable_ansi(); // false when stdout or stderr is a console that cannot render ANSI sequences
-
-// Check if the given file is attached to a terminal
-bool common_is_tty(FILE * file);

 //
 // Model utils
@@ -960,36 +925,6 @@ bool common_is_tty(FILE * file);

 struct common_sampler;

-// typed decision models, see "<arch>.decision.type" in the model metadata
-enum common_decision_type {
-    COMMON_DECISION_TYPE_NONE,    // not a decision model
-    COMMON_DECISION_TYPE_OPENJEV, // logits of one label token per option, read at the last prompt token
-    COMMON_DECISION_TYPE_LEV,     // same as openjev, noul is read from a rating scale
-    COMMON_DECISION_TYPE_KEV,     // dot product of the hidden states of the last token and of one end token per option
-    COMMON_DECISION_TYPE_NIMBLE,  // same as openjev, the prompt lists all the questions of the request
-    COMMON_DECISION_TYPE_LAYA,    // score of one marker token per option, read from the embeddings output
-    COMMON_DECISION_TYPE_CLEF,    // all questions in one prompt, score of option i read from the embeddings output at row i
-    COMMON_DECISION_TYPE_PPLX_DECIDER, // same as openjev, label codes of 1 or 2 letters
-    COMMON_DECISION_TYPE_VLLM_DECISION2, // score of each option read from the embeddings output, the options of a question are marked in the batch
-    COMMON_DECISION_TYPE_VLLM_D3, // same as pplx-decider, non-causal attention
-    COMMON_DECISION_TYPE_LFM2_D1, // same as openjev, the labels depend on the question type
-    COMMON_DECISION_TYPE_LFM2_D1_OMNI, // same as laya, other prompt layout
-    COMMON_DECISION_TYPE_GRANITE_SWITCH, // same as openjev, run by the decision aLoRA of granite switch
-    COMMON_DECISION_TYPE_H2O_LIGHTNING, // same as openjev, the labels are read after an answer prefill
-    COMMON_DECISION_TYPE_UNKNOWN, // a decision model of a type that is not supported
-};
-
-common_decision_type common_get_decision_type(const struct llama_model * model);
-
-// metadata of a GGUF file, read without loading the model
-struct common_gguf_info {
-    common_decision_type decision_type = COMMON_DECISION_TYPE_UNKNOWN; // UNKNOWN if the file is missing, unreadable, or invalid
-    uint32_t             n_ctx_train   = 0;                            // 0 if unknown
-    bool                 has_chat_tmpl = false;                        // has the default tokenizer.chat_template
-};
-
-common_gguf_info common_get_gguf_info(const std::string & fname);
-
 // note: defines the model, context, samplers, ets. lifetimes
 struct common_init_result {
    common_init_result(common_params & params, bool model_only = false);
@@ -1077,67 +1012,14 @@ struct common_memory {
 // Batch utils
 //

-// wrapper around llama_batch_ext that provide getter functions for downstream code
-// entries can exceed n_batch, use get_sub_batch() to decode them in chunks
-struct common_batch {
-    struct token {
-        llama_token  id;
-        std::array<llama_pos, GGML_MROPE_SECTIONS> pos; // only pos[0] is used for text tokens
-        llama_seq_id seq_id; // the first sequence id, see add_seq()
-        bool         output;
-        llama_embd   embd; // non-owning view of the data passed to add_embd()/set_embd(), data == NULL if none
-        llama_embd   state; // non-owning view of the data passed to set_embd_state(), data == NULL if none
-        std::vector<llama_seq_id> seq_ids_extra; // see add_seq()
-        int32_t      decision_order = 0; // see llama_batch_ext_set_decision_order()
-    };
+void common_batch_clear(struct llama_batch & batch);

-    std::vector<token> tokens; // mirror of the entries, tokens[i] describes batch index i
-    llama_batch_ext_ptr batch;
-
-    int32_t n_pos = 1; // positions per embedding entry, GGML_MROPE_SECTIONS for MROPE/IMROPE
-
-    common_batch() = default;
-    common_batch(struct llama_context * ctx);
-
-    llama_batch_ext * get() { return get_sub_batch(0, size()); }
-
-    // render entries [off, off + n) into batch, the result is overwritten by the next call
-    llama_batch_ext * get_sub_batch(int32_t off, int32_t n);
-
-    // content type of the batch, all entries carry the same combination
-    bool has_token() const { return !tokens.empty() && tokens[0].id != LLAMA_TOKEN_NULL; }
-    bool has_embd () const { return !tokens.empty() && tokens[0].embd.data != nullptr; }
-
-    void clear();
-
-    // returns the batch index
-    int32_t add(llama_token id, llama_pos pos, llama_seq_id seq_id, bool output);
-
-    // same, with the entry shared by all seq_ids (must not be empty)
-    int32_t add(llama_token id, llama_pos pos, const std::vector<llama_seq_id> & seq_ids, bool output);
-
-    // add the entry at idx to another sequence, tokens[idx].seq_id keeps the first one
-    bool add_seq(int32_t idx, llama_seq_id seq_id);
-
-    bool set_output(int32_t idx, bool value);
-
-    // attach a token embedding to the entry at idx, can only be set once per entry
-    bool set_embd(int32_t idx, llama_embd embd);
-
-    // attach a state embedding (e.g. the target hidden state for MTP) to the entry at idx, can only be set once per entry
-    bool set_embd_state(int32_t idx, llama_embd state);
-
-    // add an embedding-only entry (no token id)
-    // pos points to n_pos positions
-    int32_t add_embd(llama_embd embd, const llama_pos * pos, llama_seq_id seq_id, bool output);
-
-    int32_t size() const { return (int32_t) tokens.size(); }
-};
-
-// create a single-sequence batch from a list of tokens
-// positions continue from the memory, last token always have output_logits set to true
-common_batch common_batch_get_one(struct llama_context * ctx, const llama_token * tokens, int32_t n_tokens);
-common_batch common_batch_get_one(struct llama_context * ctx, const llama_tokens & tokens);
+void common_batch_add(
+                 struct llama_batch & batch,
+                        llama_token   id,
+                          llama_pos   pos,
+    const std::vector<llama_seq_id> & seq_ids,
+                               bool   logits);

 // decodes a single batch of tokens for a prompt and manages session tokens
 //
@@ -1146,7 +1028,7 @@ common_batch common_batch_get_one(struct llama_context * ctx, const llama_tokens
 // tokens from memory, so this approach works across all model architectures.
 bool common_prompt_batch_decode(
              struct llama_context * ctx,
-                const llama_tokens & all_tokens,
+    const std::vector<llama_token> & all_tokens,
                               int   n_new,
                               int & n_past,
                               int   n_batch,
@@ -1316,13 +1198,12 @@ struct common_prompt_checkpoint {
            llama_seq_id seq_id,
            llama_state_seq_flags flags);

-    // return false if the state could not be restored
-    bool load_tgt(
+    void load_tgt(
            llama_context * ctx,
            llama_seq_id seq_id,
            llama_state_seq_flags flags) const;

-    bool load_dft(
+    void load_dft(
            llama_context * ctx,
            llama_seq_id seq_id,
            llama_state_seq_flags flags) const;
@@ -1,5 +1,4 @@
 #include "console.h"
-#include "common.h"
 #include "log.h"
 #include <vector>
 #include <iostream>
@@ -1019,7 +1018,6 @@ namespace console {
                    line.clear();
                    pop_cursor();
                }
-                line += '\n';
                has_more = false;
            }
        } else {
@@ -1051,10 +1049,13 @@ namespace console {
        if (!std::getline(std::wcin, wline)) {
            // Input stream is bad or EOF received
            line.clear();
+            GenerateConsoleCtrlEvent(CTRL_C_EVENT, 0);
            return false;
        }

-        line = wstring_to_utf8(wline);
+        int size_needed = WideCharToMultiByte(CP_UTF8, 0, &wline[0], (int)wline.size(), NULL, 0, NULL, NULL);
+        line.resize(size_needed);
+        WideCharToMultiByte(CP_UTF8, 0, &wline[0], (int)wline.size(), &line[0], size_needed, NULL, NULL);
 #else
        if (!std::getline(std::cin, line)) {
            // Input stream is bad or EOF received
@@ -1065,7 +1066,7 @@ namespace console {
        if (!line.empty()) {
            char last = line.back();
            if (last == '/') { // Always return control on '/' symbol
-                line.back() = '\n';
+                line.pop_back();
                return false;
            }
            if (last == '\\') { // '\\' changes the default action
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Alde Rojas	3a3f76f2e8	cont : rename common_schema to common_chat_schema	2026-09-12 13:23:12 -05:00
Alde Rojas	a1e2fc9e8b	cont : restore (min\|max)Length to imply string type	2026-09-12 13:23:12 -05:00
Alde Rojas	2372064fea	cont : make "{}" mean any object for json_object as well	2026-09-12 13:23:12 -05:00
Alde Rojas	cc420dafab	cont : update docs	2026-09-12 13:23:12 -05:00
Alde Rojas	55d9cf0722	cont : remove dead code	2026-09-12 13:23:12 -05:00
Alde Rojas	c8c25a0022	cont : add tests for empty $ref	2026-09-12 13:23:12 -05:00
Alde Rojas	f0fd3a81c0	cont : properly handle empty tool parameters	2026-09-12 13:23:12 -05:00
Alde Rojas	50d886a8a5	cont : fix may_be_string	2026-09-12 13:23:12 -05:00
Alde Rojas	aca246997a	refactor : rename common_schema_builder parse_* methods to build_*	2026-09-12 13:23:12 -05:00
Alde Rojas	7e0f376cf3	tests : take the schema label as const char * to satisfy gcc dangling-reference	2026-09-12 13:23:11 -05:00
Alde Rojas	bdc3a47794	tests : fix gcc dangling-reference warning in test-json-schema	2026-09-12 13:23:11 -05:00
Alde Rojas	301b0eb5f3	refactor : rename common_schema_parse to common_schema_from_json	2026-09-12 13:23:11 -05:00
Alde Rojas	5abafdeff4	cont : clean up	2026-09-12 13:23:11 -05:00
Alde Rojas	c3f104c68a	cont : clean up	2026-09-12 13:23:11 -05:00
Alde Rojas	1dbf3eef57	cont : reduce test cases	2026-09-12 13:23:11 -05:00
Alde Rojas	797b72981f	cont : move enums under common_schema and add type enum	2026-09-12 13:23:11 -05:00
Alde Rojas	e0fc22575d	cont : cleanup	2026-09-12 13:23:11 -05:00
Alde Rojas	78b4c52c06	cont : pass common_schema through the json-schema-to-grammar builder	2026-09-12 13:23:11 -05:00
Alde Rojas	a838fb4d6d	cont : simplify schema resolution	2026-09-12 13:23:11 -05:00
Alde Rojas	a4d976df93	cont : remove common_chat_tool_parameters	2026-09-12 13:23:11 -05:00
Alde Rojas	5204e2104a	cont : cleanup	2026-09-12 13:23:11 -05:00
Alde Rojas	426d070db0	common/schema : implement type/kind resolution	2026-09-12 13:23:11 -05:00
Alde Rojas	9e63e52cfc	common : use common_trie	2026-09-12 13:23:10 -05:00
Alde Rojas	4882c77330	common : refactor json-schema-to-grammar to use common_schema	2026-09-12 13:23:10 -05:00
Alde Rojas	4935133ab8	common : reduce optimizations	2026-09-12 13:23:10 -05:00
Alde Rojas	5088b6d623	common : implement a json schema optimizer	2026-09-12 13:23:10 -05:00
Alde Rojas	8ff43c23e0	common : implement common_schema types	2026-09-12 13:23:10 -05:00