Dockerfile-CUDA11.3

FROM pytorch/pytorch:1.11.0-cuda11.3-cudnn8-runtime
LABEL github="https://github.com/CBICA/GaNDLF"
LABEL docs="https://cbica.github.io/GaNDLF/"
LABEL version=1.0

# Install instructions for NVIDIA Container Toolkit allowing you to use the host's GPU: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
# Note that to do this on a Windows host you need experimental feature "CUDA on WSL" -- not yet stable.

# Explicitly install python3.8 (this uses 11.1 for now, as PyTorch LTS 1.8.2 is built against it)
RUN apt-get update && apt-get install -y python3.8 python3-pip libvips libjpeg8-dev zlib1g-dev python3-dev libpython3.8-dev libffi-dev
RUN python3.8 -m pip install --upgrade pip
RUN python3.8 -m pip install torch==1.8.2 torchvision==0.9.2 torchaudio===0.8.2 --extra-index-url https://download.pytorch.org/whl/lts/1.8/cu111
COPY . /GaNDLF
WORKDIR /GaNDLF
RUN python3.8 -m pip install openvino-dev==2022.1.0
RUN python3.8 -m pip install -e .

# Entrypoint forces all commands given via "docker run" to go through python, CMD forces the default entrypoint script argument to be gandlf_run
# If a user calls "docker run gandlf:[tag] gandlf_anonymize", it will resolve to running "python gandlf_anonymize" instead.
# CMD is inherently overridden by args to "docker run", entrypoint is constant.
ENTRYPOINT python3.8
CMD gandlf_run

# The below force the container commands to run as a nonroot user with UID > 10000.
# This greatly reduces UID collision probability between container and host, helping prevent privilege escalation attacks.
# As a side benefit this also decreases the likelihood that users on a cluster won't be able to access their files.
# See https://github.com/hexops/dockerfile as a best practices guide.
RUN addgroup --gid 10001 --system nonroot \
 && adduser  --uid 10000 --system --ingroup nonroot --home /home/nonroot nonroot
 
USER nonroot