Dockerfile

################################################################################
# llama.cpp Studio - Multi-stage Docker build
################################################################################
ARG BASE_IMAGE=ubuntu:24.04

################################################################################
# Stage 1: Frontend Builder
# Purpose: Compile Vue.js frontend with Vite
################################################################################
FROM node:20-slim AS frontend-builder

WORKDIR /build

# Copy package files and install dependencies (including devDependencies for build tools)
COPY package*.json ./
RUN if [ -f package-lock.json ] || [ -f npm-shrinkwrap.json ]; then \
        npm ci; \
    else \
        npm install; \
    fi

# Copy frontend source using the same layout as the repo root scripts expect
COPY frontend/ ./frontend/
RUN npm run build

################################################################################
# Stage 2: Python Builder
# Purpose: Create isolated venv with all Python dependencies
################################################################################
FROM ${BASE_IMAGE} AS python-builder

ENV DEBIAN_FRONTEND=noninteractive

# Install build dependencies required for python wheels and runtime compilation (llama.cpp)
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 \
    python3-venv \
    python3-pip \
    python3-dev \
    build-essential \
    cmake \
    gcc \
    g++ \
    pkg-config \
    libssl-dev \
    libffi-dev \
    libcurl4-openssl-dev \
    libopenblas-dev \
    git \
    curl \
    wget \
    ca-certificates \
    pciutils \
    lsb-release \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

# Install Rust (required for tokenizers)
RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | \
    sh -s -- -y --profile minimal --default-toolchain stable
ENV PATH="/root/.cargo/bin:${PATH}"

# Create venv and install Python packages
ENV VENV_PATH=/opt/venv
RUN python3 -m venv ${VENV_PATH}
ENV PATH="${VENV_PATH}/bin:${PATH}"

COPY requirements.txt /tmp/requirements.txt
RUN pip install --no-cache-dir --upgrade pip setuptools wheel && \
    pip install --no-cache-dir -r /tmp/requirements.txt && \
    rm /tmp/requirements.txt

################################################################################
# Stage 3: Runtime
# Purpose: Minimal production image with compiled artifacts
################################################################################
FROM ${BASE_IMAGE} AS runtime

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    CUDA_VISIBLE_DEVICES=all \
    NVIDIA_VISIBLE_DEVICES=all \
    NVIDIA_DRIVER_CAPABILITIES=compute,utility \
    HF_HOME=/app/data/hf-cache \
    HUGGINGFACE_HUB_CACHE=/app/data/hf-cache/hub \
    VENV_PATH=/opt/venv \
    PYTHONPATH=/app \
    PATH="/app/data/cuda/current/bin:${PATH}" \
    LD_LIBRARY_PATH="/app/data/cuda/current/lib64:${LD_LIBRARY_PATH}"

# Install runtime dependencies (retain build toolchain for llama.cpp builds and GPU detection)
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 \
    python3-venv \
    python3-pip \
    python3-dev \
    build-essential \
    git \
    pkg-config \
    ninja-build \
    curl \
    ca-certificates \
    wget \
    # Core libs for Python packages
    libssl3 \
    libssl-dev \
    libffi8 \
    libcurl4 \
    libcurl4-openssl-dev \
    libopenblas0 \
    # GPU acceleration support (CUDA only)
    ocl-icd-libopencl1 \
    libnuma1 \
    pciutils \
    # Optional: ROCm (fails gracefully if unavailable)
    && (apt-get install -y --no-install-recommends rocminfo rocm-smi || echo "ROCm unavailable") \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

# CUDA installation is handled at runtime via the built-in CUDA installer
# All CUDA components (toolkit, cuDNN, TensorRT) are installed to /app/data/cuda/
# which is volume-backed and persists across container restarts

# Install modern CMake (3.28+) required for CUDA 12.x support
# Ubuntu 24.04 may have a newer cmake, but we install a specific version for consistency
# Placed here to avoid re-downloading when application code changes
ARG CMAKE_VERSION=3.31.3
RUN curl -fsSL "https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}-linux-x86_64.sh" -o /tmp/cmake.sh \
    && chmod +x /tmp/cmake.sh \
    && /tmp/cmake.sh --skip-license --prefix=/usr/local \
    && rm /tmp/cmake.sh \
    && cmake --version

# Install llama-swap binary
ARG LLAMA_SWAP_VERSION=197
RUN curl -fsSL "https://github.com/mostlygeek/llama-swap/releases/download/v${LLAMA_SWAP_VERSION}/llama-swap_${LLAMA_SWAP_VERSION}_linux_amd64.tar.gz" -o /tmp/llama-swap.tar.gz && \
    tar -xzf /tmp/llama-swap.tar.gz -C /tmp && \
    mv /tmp/llama-swap /usr/local/bin/llama-swap && \
    chmod +x /usr/local/bin/llama-swap && \
    rm -rf /tmp/* && \
    llama-swap --version

# Copy Python venv from builder
COPY --from=python-builder ${VENV_PATH} ${VENV_PATH}
ENV PATH="${VENV_PATH}/bin:${PATH}"

# Set up application directory
WORKDIR /app

# Copy application code (excluding data via .dockerignore)
COPY backend/ ./backend/
COPY --from=frontend-builder /build/frontend/dist ./frontend/dist
COPY frontend/public ./frontend/public

# Copy and setup entrypoint script and CUDA environment helper
COPY docker-entrypoint.sh /usr/local/bin/docker-entrypoint.sh
COPY setup-cuda-env.sh /app/setup-cuda-env.sh
# Ensure line endings are Unix format and file is executable
RUN sed -i 's/\r$//' /usr/local/bin/docker-entrypoint.sh /app/setup-cuda-env.sh && \
    chmod 755 /usr/local/bin/docker-entrypoint.sh /app/setup-cuda-env.sh && \
    [ -f /usr/local/bin/docker-entrypoint.sh ] && \
    [ -x /usr/local/bin/docker-entrypoint.sh ] && \
    head -n 1 /usr/local/bin/docker-entrypoint.sh | grep -q "^#!/bin/bash"

# Create python symlink for compatibility
RUN ln -sf /usr/bin/python3 /usr/bin/python

# Create non-root user and data directory structure
RUN useradd -m -s /bin/bash appuser && \
    mkdir -p /app/data/models /app/data/config /app/data/logs /app/data/llama-cpp /app/data/hf-cache/hub && \
    chown -R appuser:appuser /app && \
    # Ensure entrypoint script is accessible to appuser
    chmod 755 /usr/local/bin/docker-entrypoint.sh
# Note: /app/data is a volume mount, so permissions will be set by the entrypoint or at runtime

# Expose API port
EXPOSE 8080

# Declare volume for persistent data
VOLUME ["/app/data"]

# Switch to non-root user
USER appuser

# Health check
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -f http://localhost:8080/api/status || exit 1

# Set entrypoint to configure CUDA environment before starting application
ENTRYPOINT ["/usr/local/bin/docker-entrypoint.sh"]

# Start application
CMD ["python", "backend/main.py"]