FROM python:3.10-slim
WORKDIR /app

# The user will only upload app.zip and this Dockerfile
COPY app.zip .

# Install unzip, extract the application, and clean up the zip file
# (also install build-essential for tokenizer dependencies)
RUN apt-get update && apt-get install -y --no-install-recommends unzip build-essential && \
    unzip -o app.zip && rm app.zip && \
    rm -rf /var/lib/apt/lists/*

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

ENV HF_HOME=/app/cache
RUN python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('unsloth/Llama-3.2-3B-Instruct')"

EXPOSE 7860
CMD ["gunicorn", "--preload", "--timeout", "120", "-w", "4", "-b", "0.0.0.0:7860", "app:app"]