How to implement a simple inference?

[18930215187]

How to implement a simple inference process with uni-mol2? I want to input only a SMILES sequence and get the output from the trunk network (features), without involving the head network in any specific tasks.

[dalessioluca]

Is there any update on this?
uni-mol2 seems an wonderful model and we are all eager to try it on different tasks.

Thanks!

[Naplessss]

You can try this branch, https://github.com/deepmodeling/Uni-Mol/tree/yn-unimol2

[dalessioluca]

This branch must have been merged to main.
Right now you can extract the representations from unimol2 using unimol_tools with the code below:

params = {
        'data_type': 'molecule',
        'remove_hs': False,
        'model_name': 'unimolv2',
        'model_size': '84m',
}
smiles_list = ["C(=O)=O", "[C-]#[O+]"]
clf = UniMolRepr(use_gpu=use_gpu, **params)
unimol_repr = clf.get_repr(data=smiles_list, return_atomic_reprs=False)
cls_repr = numpy.array(unimol_repr["cls_repr"])
N, D = cls_repr.shape
assert len(smiles_list) == N

Thanks for the nice job releasing and maintaining this model!