Multihead finetune pt and default models

[marcob3]

Greetings!

I am Marco Bertani, from University of Modena (Italy). I am trying to finetune the mace_agnesi_medium.model using the multihead approach.
It generates a pt_head and default models. I do not understand the difference between the two models and which one I should use for running MD after the training.

I appreciate your help.
Marco

[ilyes319]

Hey, the pt_head corresponds to the head for the pretraining replay set. This head is there to ensure that the model does not forget too much during finetuning. The default head is the head that is used for your own training set.
When you do MD you do not have to pick a head, it will automatically use the one from your data.
If you want to select another head, you can do atoms.info["head"] = "pt_head", but I advise you not to do that.

[marcob3]

Thank you very much for your reply.
So also when I convert the model for lammps it automatically uses the default one?

[ilyes319]

Yes, that is handled automatically.