ECP Annual Meeting 21 - EXARL Tutorial
Getting Training Account on NERSC
Enter the randomly-generated 4-letter code for this tutorial - dBXa
Fill out the rest of the form appropriately
This will create a temporary account for the user, with the username trainXXX where XXX is a 3-digit number. They will be members of the project ntrain. NERSC has already given read access to members of ntrain to the relevant data sets for this tutorial in the directory /global/cfs/cdirs/m3363/exarl_data. To access the cori machine, you need to download google authenticator app/authy. Refer https://docs.nersc.gov/connect/mfa/.
$ ssh -XY username@cori.nersc.gov.
when prompted, please type your password+ssh token from google authenticator. No spaces.
Installation on NERSC Cori
Go to the scratch directory
$ cd $SCRATCH
Create a directory for tutorial and enter the directory
$ mkdir EXARL_tutorial
$ cd EXARL_tutorial
Clone the EXARL repository
$ git clone --recursive https://github.com/exalearn/ExaRL.git
$ cd ExaRL
$ git lfs fetch
$ git lfs pull
Start Shifter image
$ shifter --image=registry.nersc.gov/apg/exarl-ngc:0.1 /bin/bash
Install dependencies
$ pip install -e .
Exit out of the Shifter image
$ exit
Go to the top level and create a directory for your requirements
$ cd ..
$ mkdir Runs
$ cd Runs
Submit job
$ sbatch scripts/cori_tutorial_cpu.sh
For training on GPUs
$ module load cgpu
$ sbatch scripts/cori_tutorial_cpu.sh