Felipe97/llama-cpp-compiled

1ARG OPENVINO_VERSION_MAJOR=2026.42ARG OPENVINO_VERSION_FULL=2026.4.0.22959.99c81491cc33ARG UBUNTU_VERSION=24.044 5# Intel GPU driver versions. https://github.com/intel/compute-runtime/releases6ARG IGC_VERSION=v2.40.137ARG IGC_VERSION_FULL=2_2.40.13+224188ARG COMPUTE_RUNTIME_VERSION=26.31.39395.139ARG COMPUTE_RUNTIME_VERSION_FULL=26.31.39395.13-010ARG IGDGMM_VERSION=22.10.011 12# Intel NPU driver versions. https://github.com/intel/linux-npu-driver/releases13ARG NPU_DRIVER_VERSION=v1.38.014ARG NPU_DRIVER_FULL=v1.38.0.20260910-3448731112815ARG LIBZE1_VERSION=1.32.0-1~24.04~ppa116 17# Optional proxy build arguments18ARG http_proxy=19ARG https_proxy=20 21ARG BUILD_DATE=N/A22ARG APP_VERSION=N/A23ARG APP_REVISION=N/A24 25ARG NODE_VERSION=2426 27FROM docker.io/node:$NODE_VERSION AS web28 29ARG APP_VERSION30 31WORKDIR /app/tools/ui32 33COPY tools/ui/package.json tools/ui/package-lock.json ./34RUN npm ci35 36COPY tools/ui/ ./37RUN LLAMA_BUILD_NUMBER="$APP_VERSION" npm run build38 39## Build Image40FROM docker.io/ubuntu:${UBUNTU_VERSION} AS build41 42# Pass proxy args to build stage43ARG http_proxy44ARG https_proxy45 46RUN apt-get update && \47    apt-get install -y --no-install-recommends \48        ca-certificates \49        gnupg \50        wget \51        git \52        cmake \53        ninja-build \54        build-essential \55        libtbb12 \56        libssl-dev \57        ocl-icd-opencl-dev \58        opencl-headers \59        opencl-clhpp-headers \60        intel-opencl-icd && \61    rm -rf /var/lib/apt/lists/*62 63# OpenVINO toolkit and GPU/NPU drivers are cached via BuildKit cache mounts to avoid re-downloading on rebuilds.64# Install OpenVINO for Ubuntu 24.04.65ARG OPENVINO_VERSION_MAJOR66ARG OPENVINO_VERSION_FULL67RUN --mount=type=cache,target=/var/cache/openvino,sharing=locked \68    mkdir -p /opt/intel && \69    TGZ=/var/cache/openvino/openvino_toolkit_ubuntu24_${OPENVINO_VERSION_FULL}_x86_64.tgz && \70    if [ ! -f "$TGZ" ]; then \71        wget -O "$TGZ" https://storage.openvinotoolkit.org/repositories/openvino/packages/${OPENVINO_VERSION_MAJOR}/linux/openvino_toolkit_ubuntu24_${OPENVINO_VERSION_FULL}_x86_64.tgz; \72    fi && \73    tar -xf "$TGZ" -C /opt/intel/ && \74    mv /opt/intel/openvino_toolkit_ubuntu24_${OPENVINO_VERSION_FULL}_x86_64 /opt/intel/openvino_${OPENVINO_VERSION_MAJOR} && \75    cd /opt/intel/openvino_${OPENVINO_VERSION_MAJOR} && \76    echo "Y" | ./install_dependencies/install_openvino_dependencies.sh && \77    cd - && \78    ln -s /opt/intel/openvino_${OPENVINO_VERSION_MAJOR} /opt/intel/openvino79 80ENV OpenVINO_DIR=/opt/intel/openvino81 82WORKDIR /app83 84COPY . .85 86COPY --from=web /app/tools/ui/dist tools/ui/dist87 88# Build Stage89RUN bash -c "source ${OpenVINO_DIR}/setupvars.sh && \90    cmake -B build/ReleaseOV -G Ninja \91        -DCMAKE_BUILD_TYPE=Release \92        -DLLAMA_BUILD_TESTS=OFF \93        -DGGML_NATIVE=OFF \94        -DGGML_BACKEND_DL=ON \95        -DGGML_CPU_ALL_VARIANTS=ON \96        -DGGML_OPENVINO=ON && \97    cmake --build build/ReleaseOV --parallel "98 99# Copy all necessary libraries (build outputs + OpenVINO runtime libs)100RUN mkdir -p /app/lib && \101    find build/ReleaseOV -name '*.so*' -exec cp -P {} /app/lib \; && \102    find "${OpenVINO_DIR}/runtime/lib/intel64" -name '*.so*' -exec cp -P {} /app/lib \;103 104# Create runtime directories and copy binaries105RUN mkdir -p /app/full \106    && cp build/ReleaseOV/bin/* /app/full/ \107    && cp *.py /app/full \108    && cp -r conversion /app/full \109    && cp -r gguf-py /app/full \110    && cp -r requirements /app/full \111    && cp requirements.txt /app/full \112    && cp .devops/tools.sh /app/full/tools.sh113 114## Base Runtime Image115FROM docker.io/ubuntu:${UBUNTU_VERSION} AS base116 117# Pass proxy args to runtime stage118ARG http_proxy119ARG https_proxy120ARG BUILD_DATE=N/A121ARG APP_VERSION=N/A122ARG APP_REVISION=N/A123ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp124ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp125LABEL org.opencontainers.image.created=$BUILD_DATE \126      org.opencontainers.image.version=$APP_VERSION \127      org.opencontainers.image.revision=$APP_REVISION \128      org.opencontainers.image.title="llama.cpp" \129      org.opencontainers.image.description="LLM inference in C/C++" \130      org.opencontainers.image.url=$IMAGE_URL \131      org.opencontainers.image.source=$IMAGE_SOURCE132 133RUN apt-get update \134    && apt-get install -y libgomp1 libtbb12 curl wget ffmpeg ocl-icd-libopencl1 \135    && apt autoremove -y \136    && apt clean -y \137    && rm -rf /tmp/* /var/tmp/* \138    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \139    && find /var/cache -type f -delete140 141# Install GPU drivers142ARG IGC_VERSION143ARG IGC_VERSION_FULL144ARG COMPUTE_RUNTIME_VERSION145ARG COMPUTE_RUNTIME_VERSION_FULL146ARG IGDGMM_VERSION147RUN --mount=type=cache,target=/var/cache/intel-gpu,sharing=locked \148    set -eux; \149    cd /var/cache/intel-gpu; \150    for url in \151        https://github.com/intel/intel-graphics-compiler/releases/download/${IGC_VERSION}/intel-igc-core-${IGC_VERSION_FULL}_amd64.deb \152        https://github.com/intel/intel-graphics-compiler/releases/download/${IGC_VERSION}/intel-igc-opencl-${IGC_VERSION_FULL}_amd64.deb \153        https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-ocloc_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb \154        https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-opencl-icd_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb \155        https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/libigdgmm12_${IGDGMM_VERSION}_amd64.deb \156        https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/libze-intel-gpu1_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb ; do \157        f=$(basename "$url"); \158        [ -f "$f" ] || wget -q -O "$f" "$url"; \159    done; \160    apt-get update; \161    apt-get install -y --no-install-recommends ./*.deb; \162    rm -rf /var/lib/apt/lists/*163 164# Install NPU drivers165ARG NPU_DRIVER_VERSION166ARG NPU_DRIVER_FULL167ARG LIBZE1_VERSION168RUN --mount=type=cache,target=/var/cache/intel-npu,sharing=locked \169    set -eux; \170    TGZ=/var/cache/intel-npu/linux-npu-driver-${NPU_DRIVER_FULL}-ubuntu2404.tar.gz; \171    if [ ! -f "$TGZ" ]; then \172        wget -q -O "$TGZ" https://github.com/intel/linux-npu-driver/releases/download/${NPU_DRIVER_VERSION}/linux-npu-driver-${NPU_DRIVER_FULL}-ubuntu2404.tar.gz; \173    fi; \174    DEB=/var/cache/intel-npu/libze1_${LIBZE1_VERSION}_amd64.deb; \175    if [ ! -f "$DEB" ]; then \176        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; \177    fi; \178    mkdir /tmp/npu/ && cd /tmp/npu/ && tar -xf "$TGZ" && cp "$DEB" .; \179    apt-get update; \180    apt-get install -y --no-install-recommends ./*.deb; \181    rm -rf /tmp/npu/ /var/lib/apt/lists/*182 183COPY --from=build /app/lib/ /app/184 185### Full (all binaries)186FROM base AS full187 188ARG http_proxy189ARG https_proxy190 191COPY --from=build /app/full /app/192 193WORKDIR /app194 195RUN apt-get update && \196    apt-get install -y --no-install-recommends \197    git \198    python3 \199    python3-venv \200    python3-pip && \201    python3 -m venv /openvino-venv && \202    /openvino-venv/bin/pip install --no-cache-dir --upgrade pip setuptools wheel && \203    /openvino-venv/bin/pip install --no-cache-dir -r requirements.txt && \204    apt-get autoremove -y && \205    apt-get clean && \206    rm -rf /tmp/* /var/tmp/* && \207    find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete && \208    find /var/cache -type f -delete209 210# Activate the venv211ENV VIRTUAL_ENV=/openvino-venv \212    PATH=/openvino-venv/bin:$PATH213 214ENTRYPOINT ["/app/tools.sh"]215 216 217### Light, CLI only218FROM base AS light219 220COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app/221 222WORKDIR /app223 224ENTRYPOINT [ "/app/llama-cli" ]225 226### Server, Server only227FROM base AS server228 229ENV LLAMA_ARG_HOST=0.0.0.0230 231COPY --from=build /app/full/llama /app/full/llama-server /app/232 233WORKDIR /app234 235HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ]236 237ENTRYPOINT [ "/app/llama-server" ]238