DeepLearnPhysics/larcv3-singularity:centos7-cuda-torch-mpich

$ singularity pull shub://DeepLearnPhysics/larcv3-singularity:centos7-cuda-torch-mpich

Singularity Recipe

Bootstrap: shub
From: DeepLearnPhysics/larcv3-singularity:centos7-cuda-core-mpich

%help
Centos7 with cuda9.0 cudnn7
ML/DL packages  : tensorflow keras sc-learn nccl
Sci.  packages  : numpy pandas sc-image matplotlib opencv-python ROOT
Basic python    : ipython jupyter yaml pygments six zmq wheel h5py tqdm mpi4py horovod
Development kit : g++/gcc cython nvcc libqt4-dev python-dev
Utility kit     : git wget emacs vim openssh-client mpich

To start your container simply try
singularity exec THIS_CONTAINER.simg bash

To use GPUs, try
singularity exec --nv THIS_CONTAINER.simg bash

%labels
Maintainer coreyjadams
Version centos7-cuda-torch-mpich-py36

#------------
# Global installation
#------------
%environment
 

    # for MPICH:
    export PATH=/usr/local/mpich/install/bin/:${PATH}
    export LD_LIBRARY_PATH=/usr/local/mpich/install/lib/:${LD_LIBRARY_PATH}

    # This line is necessary to run on Cooley:
    export NCCL_P2P_DISABLE=1

%post

    scl enable devtoolset-4 bash
    scl enable rh-python36 bash
    
    # This is a dependency for building horovod from source
    yum install -y eigen3-devel


    # Using the app area to store software:
    mkdir /app
    cd /app

    scl enable devtoolset-4 bash
    # Install the necessary hash functions:
    git clone https://github.com/sparsehash/sparsehash.git
    cd sparsehash
    ./configure
    make -j install
    cd -

    pip3 --no-cache-dir --disable-pip-version-check install torch
    pip3 --no-cache-dir --disable-pip-version-check install torchvision

    # This is for tensorboardX
    pip3 --disable-pip-version-check --no-cache-dir install tensorboardX


    # This is a mild hack, but it allows me to install the cuda-built versions
    # of SparseConvNet and Horovod without having cuda available at build time.

    # I made the container without these, built the wheel files, uploaded to github,
    # and then added the prebuilt binaries to this recipe for installation.

    # Definitely a bootstrap method, and not easily maintained since the build is by hand.

    # Download the prebuilt binary:
    wget https://github.com/DeepLearnPhysics/larcv3-singularity/raw/master/wheels/cuda/pytorch/scn_cuda10-0.2.12182018-cp36-cp36m-linux_x86_64.whl
    pip3 --no-cache-dir --disable-pip-version-check install scn_cuda10-0.2.12182018-cp36-cp36m-linux_x86_64.whl 
    rm scn_cuda10-0.2.12182018-cp36-cp36m-linux_x86_64.whl


    # Install all of the horovod dependencies:
    pip3 --disable-pip-version-check --no-cache-dir install cffi pycparser cloudpickle


    # Download prebuilt libraries for horovod:
    wget https://github.com/DeepLearnPhysics/larcv3-singularity/raw/master/wheels/cuda/pytorch/horovod-0.16.2-cp36-cp36m-linux_x86_64.whl


    pip3 install horovod-0.16.2-cp36-cp36m-linux_x86_64.whl
    rm horovod-0.16.2-cp36-cp36m-linux_x86_64.whl


    # Done!

Collection

• Name: DeepLearnPhysics/larcv3-singularity
• License: GNU General Public License v3.0

Metrics

Feedback

Was this page helpful?

Glad to hear it! Please tell us how we can improve.

Sorry to hear that. Please tell us how we can improve.