Finetuning a model that you've trained

#1

Hi, I’d like to train a DMPNN model on a large dataset and then fine tune this model on a smaller dataset but I haven’t been able to find any examples in the docs or wider web of how you can fine tune deepchem models.

Is there anyway I can do this or is it better to look for an alternate solution?

#2

It is possible, but only with a newer subset of models (those inheriting from ModularTorchModel). Check out InfoGraph, ChemBERTa for examples. We are in the middle of writing up a paper/tutorial about this functionality so this isn’t well documented, but the code/docs are all in DeepChem HEAD