- Last updated 2022-04-18 *
First, login to the traverse cluster headnode via ssh:
ssh -XC <yourusername>@traverse.princeton.edu
The instructions below work with these modules (as of April 2022)
(frnn) [rkube@traverse site-packages]$ module list
Currently Loaded Modulefiles:
1) anaconda3/2021.11 2) openmpi/cuda-11.0/gcc/4.0.4/64 3) cudatoolkit/11.0 4) cudnn/cuda-11.x/8.2.0
Next, check out the source code from github and enter the directory of the local copy:
git clone git://github.com/PPPLDeepLearning/plasma-python
cd plasma-python
After that, create a new conda environment and install the relevant packages, starting with pytorch.
conda create --name conda create --name frnn --channel https://opence.mit.edu/#/ pytorch
Next, activate the envrionment can add other required packages
conda activate frnn
conda install pandas scipy flake8 h5py pyparsing pyyaml cython matplotlib scikit-learn joblib xgboost
Now we can install plasma-python:
python setup.py install
Once this command is done, the module is availabe in python from within the environment
(frnn) [rkube@traverse ~]$ ipython
importPython 3.8.13 (default, Mar 28 2022, 11:00:56)
Type 'copyright', 'credits' or 'license' for more information
IPython 8.2.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: import plasma
In [2]: print(plasma.__file__)
/home/user/.conda/envs/frnn/lib/python3.8/site-packages/plasma-1.0.0-py3.8.egg/plasma/__init__.py