OCP22 model crashes about ocp HOT 5 CLOSED

Hey, yes we have used exactly that linked config file, without any changes. We were also confused by that error as the lr_initial is part of the optim subtree. Actually loads of parts of the config file was not read properly according to our logs. It is a really strange that everything is working for you, while not for us. And it is also very strange, that everything is working for the GemNet-OC OCP20+22 model, but not for the OCP20 model.
We were not using OCP22 through main.py but rather through OCPCalculator:

from ocpmodels.common.relaxation.ase_utils import OCPCalculator

checkpoint_path = "/path/to/configs/oc22/s2ef/gemnet-oc/gemnet_oc.yml"
config_yml_path = "/path/to/gnoc_oc22_all_s2ef.ptl"

calc = OCPCalculator(config_yml=config_yml_path,
                       checkpoint=checkpoint_path,
                       cpu= False)

Please could you try that as well? Thank you!

from ocp.

Thanks! I was able to repro this from the OCPCalculator interface. Working on a fix

from ocp.

Hey, does the optim subtree in your config yaml have lr_initial? The config that you linked to does have it, but I'm not seeing it in the logs you pasted:

...
optim:
  energy_coefficient: 1
  force_coefficient: 1
  loss_energy: mae
  loss_force: atomwisel2
...

I also re-downloaded the GemNet-OC OC22 checkpoint at my end and ran the following to try and repro the error, but it seems to work fine:

python main.py --mode validate --config configs/oc22/s2ef/gemnet-oc/gemnet_oc.yml --checkpoint /path/to/gnoc_oc22_all_s2ef.pt --debug

from ocp.

@ondrejkrejci we just merged a fix for this in #585, could you pull the latest changes from the main branch and give it another shot please?

from ocp.

Thank you @abhshkdz ! We pull and reinstall and it is working for us!

from ocp.