# not needed anymore - we are using the pre-baked image from llamacpp official repo
ARG BUILD_IMAGE=ubuntu:24.04
ARG RUNTIME_IMAGE=ubuntu:24.04

FROM ${BUILD_IMAGE} AS build

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    gcc-14 \
    g++-14 \
    cmake \
    git \
    curl \
    libopenblas-dev \
    pkg-config \
    && rm -rf /var/lib/apt/lists/*

# gcc-14 is required as CUDA host compiler on Ubuntu 24.04
ENV CC=gcc-14 CXX=g++-14 CUDAHOSTCXX=g++-14

WORKDIR /build
COPY llama.cpp/ .

# Use build args to decide between CUDA and BLAS
ARG GGML_CUDA=OFF
ARG CUDA_DOCKER_ARCH=all
RUN if [ "$GGML_CUDA" = "ON" ]; then \
        cmake -B build -DGGML_CUDA=ON -DCUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH} \
            -DGGML_NATIVE=OFF \
            -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
            -DCMAKE_BUILD_TYPE=Release; \
    else \
        cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DCMAKE_BUILD_TYPE=Release; \
    fi && \
    cmake --build build --config Release -j$(nproc) --target llama-server

FROM ${RUNTIME_IMAGE}

ENV DEBIAN_FRONTEND=noninteractive
# Include CUDA compat libs injected by nvidia-container-toolkit
ENV LD_LIBRARY_PATH=/usr/local/cuda/compat:/usr/local/cuda/lib64:${LD_LIBRARY_PATH:-}

RUN apt-get update && apt-get install -y --no-install-recommends \
    libopenblas0 \
    libgomp1 \
    curl \
    && rm -rf /var/lib/apt/lists/*

# llama.cpp may place the binary in build/bin/ or build/bin/Release/
COPY --from=build /build/build/bin/llama-server /usr/local/bin/
COPY --from=build /build/build/bin/*.so* /usr/local/lib/
RUN ldconfig

EXPOSE 7474
HEALTHCHECK --interval=30s --timeout=10s --retries=5 --start-period=120s \
    CMD curl -sf http://localhost:7474/health || exit 1

ENTRYPOINT ["llama-server"]