FROM python:3.11-slim

# Install system dependencies
RUN apt-get update && apt-get install -y --no-install-recommends \
        gcc libc6-dev \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# Install Python dependencies
RUN pip install --no-cache-dir \
        flask flask-cors Pillow litert-lm "huggingface_hub[cli]"

# Create dummy Vulkan library to satisfy litert-lm on CPU
RUN printf '%s\n' \
    'void vkGetInstanceProcAddr(){}' \
    'void vkCreateInstance(){}' \
    'void vkDestroyInstance(){}' \
    'void vkEnumeratePhysicalDevices(){}' \
    'void vkGetDeviceProcAddr(){}' > /tmp/vk.c \
    && gcc -shared -fPIC -o /app/libvulkan.so.1 /tmp/vk.c \
    && rm /tmp/vk.c

# Create model directory
RUN mkdir -p /app/models/qwen

# Download Qwen 3.5 2B LiteRT-LM (community conversion)
RUN huggingface-cli download paulsp94/Qwen3.5-2B-LiteRT-LM model.litertlm \
        --local-dir /app/models/qwen

# Copy the server code
COPY server.py .

# Setup permissions for Hugging Face Spaces (requires user 1000)
RUN useradd -m -u 1000 user && chown -R user:user /app
USER user

# Environment
ENV PORT=7860 \
    LD_LIBRARY_PATH=/app \
    GEMMA_MODEL_PATH=/app/models/qwen/model.litertlm \
    HF_HOME=/app/.cache/huggingface

EXPOSE 7860

CMD ["python3", "server.py"]