Dockerfile

# infra/hf_spaces/api/Dockerfile
# Stateless API Space. No model weights — they live in the embedder and reranker Spaces.
# All secrets are injected at runtime via HF Space environment variables.

FROM python:3.11-slim

WORKDIR /app

ENV PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1

# git is required for pip to install the toon_format VCS dependency.
# Installed before the pip layer so it is cached alongside system deps.
RUN apt-get update && apt-get install -y --no-install-recommends git \
    && rm -rf /var/lib/apt/lists/*

# Install dependencies first for better Docker layer caching.
# --prefer-binary: always take a pre-built wheel over compiling from source.
# Without this, onnxruntime (pulled by fastembed) compiles from source and
# adds 20-40 min to the build, causing HF Spaces to appear stuck.
COPY requirements.txt .
RUN pip install --no-cache-dir --prefer-binary -r requirements.txt

# Copy app source after dependencies so code changes don't bust the pip cache layer.
COPY . .

EXPOSE 7860

# Single worker — HF Spaces are not auto-scaled, multiple workers waste RAM.
CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]