# syntax=docker/dockerfile:1
#
# C++ worker + backend bundle image.
#
# One container runs:
#   - llama.cpp-omni llama-omni-server on 127.0.0.1:${BACKEND_PORT:-22500}
#   - worker.py on 0.0.0.0:${WORKER_PORT:-22400}, forwarding to the local backend
#
# Model weights are mounted at runtime; they are not baked into the image.

ARG CUDA_VERSION=12.8.1
ARG UBUNTU_VERSION=22.04

FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION} AS cpp-build

ARG LLAMA_OMNI_REPO=https://github.com/tc-mb/llama.cpp-omni.git
ARG LLAMA_OMNI_REFSPEC=master
ARG LLAMA_OMNI_REF=origin/master
ARG LLAMA_OMNI_SERVER_TARGET=llama-omni-server
ARG LLAMA_OMNI_SERVER_BIN=llama-omni-server
ARG CMAKE_CUDA_ARCHITECTURES=80;120

RUN apt-get update && apt-get install -y --no-install-recommends \
        build-essential \
        ca-certificates \
        cmake \
        curl \
        git \
        libgomp1 \
        ninja-build \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /src

RUN git init llama.cpp-omni \
    && cd llama.cpp-omni \
    && git remote add origin "${LLAMA_OMNI_REPO}" \
    && git fetch --depth=1 origin "${LLAMA_OMNI_REFSPEC}" \
    && git checkout --detach "${LLAMA_OMNI_REF}" \
    && git rev-parse HEAD

WORKDIR /src/llama.cpp-omni

RUN cmake -S . -B build -G Ninja \
        -DCMAKE_BUILD_TYPE=Release \
        -DGGML_CUDA=ON \
        -DGGML_NATIVE=OFF \
        -DLLAMA_CURL=OFF \
        -DLLAMA_BUILD_SERVER=ON \
        -DLLAMA_BUILD_TESTS=OFF \
        -DLLAMA_BUILD_EXAMPLES=OFF \
        -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
        -DCMAKE_CUDA_ARCHITECTURES="${CMAKE_CUDA_ARCHITECTURES}" \
    && cmake --build build --target "${LLAMA_OMNI_SERVER_TARGET}" -j "$(nproc)" \
    && mkdir -p /opt/llama.cpp-omni/bin \
    && cp -a "build/bin/${LLAMA_OMNI_SERVER_BIN}" build/bin/*.so* /opt/llama.cpp-omni/bin/

FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}

ENV DEBIAN_FRONTEND=noninteractive \
    LD_LIBRARY_PATH=/opt/llama.cpp-omni/bin:/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
    PATH=/opt/venv/bin:${PATH} \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    PYTHONPATH=/app \
    PYTHONUNBUFFERED=1

RUN apt-get update && apt-get install -y --no-install-recommends \
        ca-certificates \
        curl \
        libgomp1 \
        python3 \
        python3-pip \
        python3-venv \
    && rm -rf /var/lib/apt/lists/* \
    && python3 -m venv /opt/venv

WORKDIR /app

RUN --mount=type=cache,target=/root/.cache/pip \
    pip install \
        "fastapi>=0.128.0" \
        "httpx>=0.28.0" \
        "numpy>=2.2.0" \
        "pydantic>=2.11.0" \
        python-multipart \
        "uvicorn>=0.40.0" \
        "websockets>=16.0"

COPY --from=cpp-build /opt/llama.cpp-omni /opt/llama.cpp-omni
COPY . .

RUN chmod +x docker/entrypoint-cpp-worker-backend.sh \
    && mkdir -p /app/logs

EXPOSE 22500
EXPOSE 22400

HEALTHCHECK --interval=15s --timeout=5s --start-period=120s --retries=40 \
    CMD curl -sf "http://127.0.0.1:${WORKER_PORT:-22400}/health" >/dev/null || exit 1

ENTRYPOINT ["docker/entrypoint-cpp-worker-backend.sh"]