FROM python:3.11-slim

WORKDIR /app

# 1. CPU-only torch on its own line: huge, but cached as a single reusable layer.
#    The CPU index drops the NVIDIA CUDA stack you don't need on a CPU server.
RUN pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu

# 2. The rest of the dependencies (torch is already satisfied above).
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# 3. Application code last — the layer that changes most often.
COPY . .

EXPOSE 7860

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