Dockerfile

FROM python:3.12-slim

WORKDIR /app

# git is needed for huggingface_hub model downloads.
# No cmake/build-essential needed -- no C++ compilation.
# LLM runs locally via transformers.
RUN apt-get update && apt-get install -y git \
    && rm -rf /var/lib/apt/lists/*

COPY requirements.txt .

RUN pip install --upgrade pip
# -- Step 1: CPU-only PyTorch ------------------------------------------------
# sentence-transformers + transformers both need torch.
# Pre-installing the CPU wheel (~190 MB) prevents pip from resolving the
# default CUDA bundle (~3.5 GB) which would blow the build disk quota.
RUN pip install torch --index-url https://download.pytorch.org/whl/cpu \
    --no-cache-dir

# -- Step 2: Everything else -------------------------------------------------
RUN pip install -r requirements.txt --no-cache-dir

# -- Step 3: Pin transformers to a stable release ---------------------------
# Llama 3.2 works with any recent PyPI release; no git-main needed.
RUN pip install --force-reinstall --no-cache-dir "transformers>=4.43.0,<5.0.0"

COPY . .

EXPOSE 7860

CMD ["uvicorn", "app:app", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "--workers", "1", \
     "--timeout-keep-alive", "120"]