Dockerfile.arm64

# Stage 1: Build environment
# using standard ubuntu which supports arm64 natively
FROM ubuntu:24.04 AS build

# Prevent interactive prompts during package installation
ENV DEBIAN_FRONTEND=noninteractive

# Install build dependencies
# Note: explicit openblas isn't strictly necessary as llama.cpp has great internal 
# ARM kernels, but build-essential and cmake are critical.
RUN apt-get update && \
    apt-get install -y \
    build-essential \
    cmake \
    python3 \
    python3-pip \
    python3-venv \
    git \
    pkg-config \
    libcurl4-openssl-dev \
    && apt-get clean && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# Ensure latest code is fetched and submodules are updated
RUN echo "Cloning latest llama.cpp..." && \
    git clone --depth 1 https://github.com/ggml-org/llama.cpp.git . && \
    echo "Updating submodules..." && \
    git submodule update --init --recursive

# Configure and build llama.cpp for CPU (ARM64)
RUN echo "Configuring and building llama.cpp for ARM64..." && \
    mkdir -p build && \
    cd build && \
    cmake .. \
    # Disable CUDA specifically
    -DGGML_CUDA=OFF \
    # Enable Shared Libs so we generate the .so files your runtime stage expects
    -DBUILD_SHARED_LIBS=ON \
    # Ensure CURL support is compiled in
    -DGGML_CURL=ON \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_EXE_LINKER_FLAGS="-Wl,--allow-shlib-undefined" && \
    cmake --build . --config Release -j $(nproc)

# --- Final Artifact Collection ---
RUN echo "Collecting build artifacts from /app/build/bin/..." && \
    rm -rf /app/artifacts && \
    mkdir -p /app/artifacts && \
    # Copy binaries
    cp -v /app/build/bin/* /app/artifacts/ && \
    # Copy shared libraries (often located in build/src or build/ depending on cmake version)
    # We use find here to be robust against cmake folder structure changes
    find /app/build -name "*.so" -exec cp -v {} /app/artifacts/ \;

# Stage 2: Runtime environment
FROM ubuntu:24.04

# Prevent interactive prompts
ENV DEBIAN_FRONTEND=noninteractive

# Install runtime dependencies (No CUDA libs needed here)
RUN apt-get update && \
    apt-get install -y \
    libcurl4 \
    libgomp1 \
    python3 \
    python3-pip \
    python3-venv \
    python3-requests \
    && apt-get clean && rm -rf /var/lib/apt/lists/*

# Set up a virtual environment
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

# Add virtualenv activation to .bashrc
RUN echo 'source /opt/venv/bin/activate' >> /root/.bashrc

# Copy requirements file into the image
COPY requirements.txt /tmp/requirements.txt

# Install Python dependencies
RUN pip install --no-cache-dir --upgrade pip && \
    pip install --no-cache-dir -r /tmp/requirements.txt && \
    rm /tmp/requirements.txt

# --- Final Artifact Copying ---
# Copy shared libraries (.so) to /usr/local/lib
COPY --from=build /app/artifacts/*.so /usr/local/lib/
# Copy executables to /usr/local/bin
# We exclude .so files from this specific copy to keep bin clean
COPY --from=build /app/artifacts/llama-* /usr/local/bin/

# Set up library path and ensure ldconfig recognizes new libs
ENV LD_LIBRARY_PATH=/usr/local/lib
RUN ldconfig

# Create workspace directory structure
RUN mkdir -p /workspace/config /workspace/models /workspace/examples/test_images

# Copy Python connector files, config folder, and examples
COPY llama_cli_connector.py llama_server_connector.py /workspace/
COPY config/ /workspace/config/
COPY models/ /workspace/models/
COPY examples/ /workspace/examples/

# Set workspace as working directory
WORKDIR /workspace

# Entrypoint script
RUN echo '#!/bin/bash\necho "Activating Python virtual environment..."\nsource /opt/venv/bin/activate\nPS1="(venv) \u@\h:\w\\$ "\nexec "$@"' > /entrypoint.sh && \
    chmod +x /entrypoint.sh

ENTRYPOINT ["/entrypoint.sh", "/bin/bash", "-l"]