/catlas

files for dask parallelization of OCP energy predictions.

Primary LanguagePythonMIT LicenseMIT

catlas

catlas overview

Repo graphical overview

catlas overview

Detailed documentation

https://ulissigroup.cheme.cmu.edu/catlas/intro.html

Installation

  • Install ocp:
  • $ git clone git@github.com: Open-Catalyst-Project/ocp
  • $ cd ~/ocp && python setup.py develop
  • Install catlas:
  • $ git clone git@github.com: ulissigroup/catlas
  • $ cd ~/catlas && python setup.py develop

Large file handling

Add large files to their appropriate place in the repo

  • Model checkpoints -> catlas/ocp_checkpoints
  • Inference on validation data for parity (link to download coming soon!)

Usage

Local / kubernetes cluster

$ python bin/predictions.py configs/path/to/config.yml configs/path/to/cluster.py

more info about using dask operator to manage resources on a kubectl cluster: kubernetes.dask.org

HPC (Perlmutter)

$ sbatch configs/dask_cluster/perlmutter/catlas_run.sh

For additional options see NERSC documentation

Monitoring

The default dashboard port is 8787

  • localhost:8787/status