FROM python:3.11-slim

# Install minimal build tools
RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    cmake \
    pkg-config \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# Install Python deps
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Build the LATEST llama-cpp-python from source.
# We disable BLAS (OpenBLAS/OpenCL) to significantly speed up compilation 
# and avoid the 20-minute Hugging Face build timeout. 
# Standard CPU optimization is still very fast for a 4B model.
RUN CMAKE_ARGS="-DGGML_BLAS=OFF" pip install --no-cache-dir llama-cpp-python

COPY . .

EXPOSE 7860
CMD ["python", "-c", "import uvicorn; uvicorn.run('app:app', host='0.0.0.0', port=7860)"]