AI-ML-HPC-Workflows/workflows/sciml-bench/def_files/sciml-bench-clean.def

Bootstrap: docker
From: ubuntu:18.04
#Stage: spython-base

%files
# -----------------------------------------------------------------------------------
    sciml-bench/requirements.txt /sciml-benchmarks/requirements.txt
    sciml-bench/MANIFEST.in /sciml-benchmarks/MANIFEST.in
    sciml-bench/setup.py /sciml-benchmarks/setup.py
    sciml-bench/doc /sciml-benchmarks/doc
    sciml-bench/sciml_bench /sciml-benchmarks/sciml_bench


%environment
# -----------------------------------------------------------------------------------

    export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
    export LC_ALL=C
    export HOROVOD_GPU_ALLREDUCE=NCCL
    export HOROVOD_GPU_ALLGATHER=MPI
    export HOROVOD_GPU_BROADCAST=MPI
    #export HOROVOD_NCCL_HOME=/usr/local/cuda/nccl
    #export HOROVOD_NCCL_INCLUDE=/usr/local/cuda/nccl/include
    #export HOROVOD_NCCL_LIB=/usr/local/cuda/nccl/lib
    export PYTHON_VERSION=3.8
    export TENSORFLOW_VERSION=2.3.0
    export PYTORCH_VERSION=1.10.0+cu113

%post
# -----------------------------------------------------------------------------------

export SINGULARITY_BINDPATH=$SINGULARITY_BINDPATH,$(echo /usr/bin/ | sed -e 's/ /,/g')

MLNX_ARCH=ubuntu18.04-x86_64
MLNX_VERSION=5.2-1.0.4.0

# Python 3.7 is supported by Ubuntu Bionic out of the box
python=3.7
PYTHON_VERSION=${python}

# Set default shell to /bin/bash
#SHELL ["/bin/bash", "-cu"]

TZ="Europe/London"

RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-downgrades --allow-change-held-packages --no-install-recommends \
build-essential \
cmake \
g++-7 \
git \
curl \
vim \
wget \
ca-certificates \
python${PYTHON_VERSION} \
python${PYTHON_VERSION}-dev \
python${PYTHON_VERSION}-distutils \
libpython-stdlib \
python \
udev \
automake \
gfortran \
autotools-dev \
chrpath \
pkg-config \
libnl-3-dev \
tcl \
pciutils \
tk \
libnl-route-3-dev \
libltdl-dev \
bison \
autoconf \
flex \
graphviz \
libssl1.0.0 \
kmod \
debhelper \
libgfortran4 \
ethtool \
swig \
lsof \
libnl-route-3-200 \
m4 \
libnl-3-200 \
dpatch \
libnuma-dev \
iputils-ping \
iproute2 \
apt-utils

ln -sf /usr/bin/python${PYTHON_VERSION} /usr/bin/python

curl -O https://bootstrap.pypa.io/get-pip.py && \
python get-pip.py && \
rm get-pip.py

wget http://www.mellanox.com/downloads/ofed/MLNX_OFED-${MLNX_VERSION}/MLNX_OFED_LINUX-${MLNX_VERSION}-${MLNX_ARCH}.tgz && \
tar -xzvf MLNX_OFED_LINUX-${MLNX_VERSION}-${MLNX_ARCH}.tgz --no-same-owner && \
cd  MLNX_OFED_LINUX-${MLNX_VERSION}-${MLNX_ARCH} && \
./mlnxofedinstall --user-space-only --without-fw-update --skip-repo --force && \
cd .. && \
rm -rf MLNX_OFED-LINUX-${MLNX_VERSION}-${MLNX_ARCH} && \
rm -rf *.tgz

# Install Open MPI version 4.1
wget https://www.open-mpi.org/software/ompi/v4.1/downloads/openmpi-4.1.0.tar.gz && \
tar zxf openmpi-4.1.0.tar.gz && \
cd openmpi-4.1.0 && \
./configure --enable-mpirun-prefix-by-default \
            --enable-shared \
            --without-verbs && \
make -j $(nproc) all && \
make install && \
ldconfig && \
rm -rf /tmp/openmpi

cd /
pip --no-cache-dir --disable-pip-version-check install mpi4py

LD_LIBRARY_PATH=/usr/local/lib/:/usr/local/lib/openmpi:$LD_LIBRARY_PATH

# --------------------------------------------------------------------------

# this will install all necessary packages and prepare the container

# TensorFlow version is tightly coupled to CUDA and cuDNN so it should be selected carefully
# Python 2.7 or 3.5 is supported by Ubuntu Xenial out of the box

    export PYTHON_VERSION=3.8
    export TENSORFLOW_VERSION=2.3.0
    export PYTORCH_VERSION=1.10.0+cu113

    wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
    mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
    apt-get update && apt-get install -y gnupg
    apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
    apt-get install -y software-properties-common
    apt-get update
    add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/ /"
    apt-get update
    export DEBIAN_FRONTEND=noninteractive
    apt-get -y install cuda

    apt-get update && apt-get install -y libcudnn8 libcudnn8-dev
    apt install libnccl2 libnccl-dev

    
    apt-get -y update && apt-get install -y --allow-downgrades --allow-change-held-packages --no-install-recommends \
        build-essential \
        cmake \
        git \
        curl \
        vim \
        wget \
        ca-certificates \
        libjpeg-dev \
        libpng-dev \
        python${PYTHON_VERSION} \
        python${PYTHON_VERSION}-dev

    ln -sf /usr/bin/python${PYTHON_VERSION} /usr/bin/python

    curl -O https://bootstrap.pypa.io/get-pip.py && \
    python get-pip.py && \
    rm get-pip.py

# Install TensorFlow, Keras and PyTorch
    pip install torch==${PYTORCH_VERSION} torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
    pip install mxnet-cu112 tensorflow-gpu==${TENSORFLOW_VERSION} keras h5py filelock matplotlib scikit-learn
    
    export PATH="/usr/local/cuda-11.5/bin:$PATH"

# Install Horovod, temporarily using CUDA stubs
    ldconfig /usr/local/cuda-11.5/targets/x86_64-linux/lib/stubs && \
    HOROVOD_GPU_ALLREDUCE=NCCL HOROVOD_WITH_MXNET=1 HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITH_PYTORCH=1 pip install --no-cache-dir horovod && \
    ldconfig

# Set default NCCL parameters
    echo NCCL_DEBUG=INFO >> /etc/nccl.conf && \
    echo NCCL_SOCKET_IFNAME=^docker0 >> /etc/nccl.conf

# Clean up cuda
    ls /usr/local/
    sudo apt-get --purge -y remove "cuda*"
    sudo apt-get --purge -y remove "nvidia*"

# Install sciml-bench
    cd /sciml-benchmarks && pip install .

%environment

%runscript
    sciml-bench $@