# Use official PyTorch image with CUDA if available, else CPU-only python base
# For portability default to python:3.10-slim; users with GPU can adapt.
FROM python:3.10-slim

ENV PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    PIP_NO_CACHE_DIR=1 \
    HF_HOME=/app/.cache/huggingface \
    HF_TOKEN=""

# System deps
RUN apt-get update && apt-get install -y --no-install-recommends \
        build-essential \
        git \
        libgl1 \
        && rm -rf /var/lib/apt/lists/*

# Ensure HF cache dir exists and is writable
RUN mkdir -p /app/.cache/huggingface && chmod -R 777 /app/.cache

WORKDIR /app

# Copy only requirements first for better caching
COPY requirements.txt /app/requirements.txt

# Install Python dependencies
RUN pip install --no-cache-dir --upgrade pip && \
    pip install --no-cache-dir -r requirements.txt

# Copy application code
COPY . /app
# Optional: allow passing HF_TOKEN at runtime for private models
# ENV HF_TOKEN=""


# Expose port
EXPOSE 7860

# Model weights will be downloaded on first run
CMD ["uvicorn", "server:app", "--host", "0.0.0.0", "--port", "7860"]