Installation and setup#

  1. Clone the catlas repo

    git clone https://github.com/ulissigroup/catlas.git
OR
git clone git@github.com:ulissigroup/catlas.git

  2. Install catlas locally

    cd ~/catlas && python setup.py develop

  3. Add model checkpoints to a folder in the base directory called ocp_checkpoints

Adding parity data#

The parity data files have been omitted from the repo because they are large. We will have them as downloadable links. If you need them now. Please email bwander@andrew.cmu.edu. They should be put in catlas/catlas/parity/df_pkls/ and should be named to match the model checkpoint name.

Advanced options#

If you plan to use custom model architectures, they must be compatible with the OCP infra. This is currently supported. To do this you have to

  1. clone the ocp repo

  2. install it locally cd ~/ocp && python setup.py develop

  3. make your changes & train your model. Then just use the checkpoint as you would for any other catlas run