AI-ML-HPC-Workflows/workflows/sciml-bench/def_files/sciml-bench-cu11-tf27-mlnx.def

Bootstrap: localimage
From: ompi4-cu11-mlnx.sif

%files
# -----------------------------------------------------------------------------------
    requirements.txt /sciml-benchmarks/requirements.txt
    MANIFEST.in /sciml-benchmarks/MANIFEST.in
    setup.py /sciml-benchmarks/setup.py
    doc /sciml-benchmarks/doc
    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=2.7.0
    export PYTORCH_VERSION=1.10.0+cu113

%post
# -----------------------------------------------------------------------------------
# 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=2.7.0
    export PYTORCH_VERSION=1.10.0+cu113

    echo "deb https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 /" > /etc/apt/sources.list.d/nvidia-ml.list

    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
    
    ls /usr/local/
    whereis nvcc
    export PATH="/usr/local/cuda-11.5/bin:$PATH"
    whereis nvcc

# Install Horovod, temporarily using CUDA stubs
    ldconfig /usr/local/cuda-11.4/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

# Download examples
    cd / && \
    apt-get install -y --no-install-recommends subversion && \
    svn checkout https://github.com/uber/horovod/trunk/examples && \
    rm -rf /examples/.svn

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

%runscript
    sciml-bench $@