Dockerfile

FROM python:3.11-slim

WORKDIR /app

RUN apt-get update && apt-get install -y \
    build-essential \
    cmake \
    libopenblas-dev \
    pkg-config \
    git \
    curl \
    wget \
    && rm -rf /var/lib/apt/lists/*

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Use latest llama-cpp-python — 0.3.8 does NOT support Gemma 4 MoE (A4B) architecture
RUN CMAKE_BUILD_PARALLEL_LEVEL=4 \
    CMAKE_ARGS="-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS" \
    pip install --no-cache-dir "llama-cpp-python"

RUN mkdir -p /app/models

COPY app.py .

EXPOSE 7860

ENV SPACE_URL=""
ENV N_CTX="4096"
ENV N_THREADS="2"

CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]