Dockerfile

FROM python:3.9-slim

WORKDIR /app

# System deps (minimal and safe)
# libgl1 and libglib2.0-0 are required for OpenCV
# libgomp1 is required for PaddlePaddle math operations
RUN apt-get update && apt-get install -y \
    libgl1 \
    libglib2.0-0 \
    libgomp1 \
    git \
    && rm -rf /var/lib/apt/lists/*

# Upgrade pip stack
RUN pip install --no-cache-dir --upgrade pip setuptools wheel

# 🔒 HARD PIN numpy BEFORE paddle
# This prevents the binary incompatibility with newer numpy 2.x
RUN pip install --no-cache-dir "numpy==1.26.4"

# PaddlePaddle CPU (stable version 2.6.2 matches well with OCR 2.7.x)
RUN pip install --no-cache-dir paddlepaddle==2.6.2

# PaddleOCR + API stack
# 2.7.3 is a "Golden" version: very stable, no weird dependency bugs
RUN pip install --no-cache-dir \
    paddleocr==2.7.3 \
    opencv-python-headless==4.11.0.86 \
    fastapi \
    uvicorn \
    python-multipart \
    pillow

# Copy app code
COPY main.py .

# 🚫 DO NOT preload models during build
# NOTE: The first time you send a request, it will take ~20-40s to 
# download the detection and classification models. 
# Subsequent requests will be fast.

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