Dockerfile

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

ENV DEBIAN_FRONTEND=noninteractive \
    PIP_NO_CACHE_DIR=1 \
    PYTHONUNBUFFERED=1 \
    CUDA_HOME=/usr/local/cuda \
    FORCE_CUDA=1 \
    TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;8.9" \
    CMAKE_BUILD_PARALLEL_LEVEL=8 \
    # ✅ make all caches writable (fixes PermissionError: '/.cache')
    HF_HOME=/tmp/hf \
    HUGGINGFACE_HUB_CACHE=/tmp/hf/hub \
    TRANSFORMERS_CACHE=/tmp/hf/transformers \
    XDG_CACHE_HOME=/tmp/.cache \
    MPLCONFIGDIR=/tmp/mpl \
    # ✅ silence tokenizer/OMP warnings
    TOKENIZERS_PARALLELISM=false \
    OMP_NUM_THREADS=1

# Python 3.11 + headers + toolchain + Cairo dev
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3.11 python3.11-venv python3.11-dev python3-pip \
    git git-lfs build-essential ninja-build cmake pkg-config \
    libgl1 libsm6 libxext6 ffmpeg \
    libcairo2-dev libpango1.0-dev libffi-dev \
 && rm -rf /var/lib/apt/lists/* && git lfs install \
 && ln -sf /usr/bin/python3.11 /usr/bin/python \
 && python -m pip install -U pip wheel setuptools packaging pybind11 psutil \
 && mkdir -p /tmp/hf/hub /tmp/hf/transformers /tmp/.cache /tmp/mpl \
 && chmod -R 777 /tmp

WORKDIR /app
COPY requirements.txt /app/requirements.txt

# 1) Torch (CUDA 12.1) first
RUN pip install --index-url https://download.pytorch.org/whl/cu121 \
    torch==2.5.1 torchvision==0.20.1

# 2) flash-attn (needs Torch, CUDA toolchain, psutil)
RUN pip install --no-build-isolation --no-cache-dir flash-attn==2.7.3

# 3) Cairo wheels to avoid pycairo source builds
RUN pip install --no-cache-dir pycairo==1.26.0 rlpycairo==0.4.0 svglib==1.6.0

# 4) Remaining deps (no torch here)
RUN pip install --no-cache-dir -r requirements.txt

# 5) StarVector package
RUN pip install --no-cache-dir git+https://github.com/joanrod/star-vector.git@main

# 6) App code
COPY . /app

ENV PORT=7860
EXPOSE 7860
CMD ["python", "app.py"]