Dockerfile

FROM nvidia/cuda:12.1.1-cudnn8-devel-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive

# Install Python and ML dependencies
RUN apt-get update && \
    apt-get install -y \
        python3.10 \
        python3.10-venv \
        python3.10-distutils \
        python3-pip \
        python3-dev \
        wget \
        git \
        libgl1 \
        fontconfig \
        libglib2.0-0 \
        libxrender1 \
        libsm6 \
        libxext6 \
        poppler-utils \
        libjpeg-dev \
        libpng-dev && \
    rm -rf /var/lib/apt/lists/*

RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.10 1

# Create user with UID 1000 (required for Hugging Face Spaces)
RUN useradd -m -u 1000 user

# Create necessary directories and set permissions
RUN mkdir -p /app /app/docker_mineru/output/images /home/user/.cache/huggingface /home/user/.cache/torch && \
    chown -R user:user /app /home/user

WORKDIR /app

# Copy requirements first
COPY --chown=user:user requirements.txt .

# Upgrade pip and install PyTorch dependencies first
# Use versions compatible with CUDA 12.1 for NVIDIA L4 support
RUN pip install --no-cache-dir --upgrade pip && \
    pip install --no-cache-dir \
        torch==2.1.2 \
        torchvision==0.16.2 \
        torchaudio==2.1.2 \
        --extra-index-url https://download.pytorch.org/whl/cu121

# Install other requirements including gunicorn
RUN pip install --no-cache-dir -r requirements.txt && \
    pip install --no-cache-dir gunicorn

# Copy the rest of the application code
COPY --chown=user:user . .

# Ensure output directory exists and has correct permissions (redundant but safe)
RUN mkdir -p /app/docker_mineru/output/images && \
    chown -R user:user /app/docker_mineru/output

# Create marker static directory and set proper permissions (fix for font download error)
RUN mkdir -p /usr/local/lib/python3.10/dist-packages/static && \
    chmod -R 777 /usr/local/lib/python3.10/dist-packages/static

# Set the user
USER user

# Environment variables for caching (optional, might help with model downloads)
ENV HF_HOME=/home/user/.cache/huggingface
ENV TORCH_HOME=/home/user/.cache/torch
# Add environment variable for marker font path (alternative fix)
ENV MARKER_FONT_PATH=/home/user/.cache/marker_fonts
# Add PyTorch memory optimization environment variables
ENV PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:128
ENV CUDA_VISIBLE_DEVICES=0

# Expose the port
EXPOSE 7860

# Command to run the application with Gunicorn and Uvicorn workers
# Reduced workers to 4 (from 16) to avoid OOM errors
CMD ["gunicorn", "-w", "4", "-k", "uvicorn.workers.UvicornWorker", "app.main:app", "--bind", "0.0.0.0:7860"]