Skip to content

QingpingZheng/pytorch-on-phx

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Running DL project with pyenv+pytorch+tensorboardX on Phoenix HPC

Table of Contents

Phoenix Login

Register for an account by submitting a form on this page.

After your account is activated, login clone this project and copy phx/.bashrc into your home folder.

$ git clone https://github.com/stanstarks/pytorch-on-phx
$ cd pytorch-on-phx; cp phx/.bashrc ~
$ source ~/.bashrc

Install pyenv

Pyenv is an easy to use Python version management tool integrating virtualenv and Anaconda. To install it, simply use pyenv installer.

Since there is limited space under home folder on phx. It is recommended to install pyenv under /fast partition. To so do, set PYENV_ROOT environment variable to be your fast folder:

$ echo 'export PYENV_ROOT=/fast/users/a17XXXXX/.pyenv' >> ~/.bashrc
$ source ~/.bashrc

Now you can install pyenv from Git:

$ curl -L https://github.com/pyenv/pyenv-installer/raw/master/bin/pyenv-installer | bash

Remember to make pyenv take effect by modifying .bashrc again:

$ echo 'export PATH="$PYENV_ROOT/bin:$PATH"\neval "$(pyenv init -)"\n eval "$(pyenv virtualenv-init -)"
source ~/.bashrc

How we can install desired Python version. The best behaviour is to use conda mostly because of off the shelf mkl support. However, latest conda-3 uses Python 3.7 which is not officially supported by TensorFlow. So just install 3.6.6 for now.

pyenv install 3.6.6

Use this python globally:

pyenv global 3.6.6

Install PyTorch

We will use PyTorch in this demo. Install it with conda:

pip install torch torchvision

We will use tensorboardX for visualization:

pip install git+https://github.com/lanpa/tensorboardX

Run pytorch on Phoenix

Please refer to word_language_model folder for further instructions.

About

Demo for pytorch on Phoenix

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 56.4%
  • Python 38.8%
  • Shell 4.8%