Feature Request: New Model

[idontkonwher]

I found that MolXP（https://arxiv.org/pdf/2305.10688.pdf）has SOTA proformence in molecular property prediction, here is my idea!
I think we can modling the molecular property prediction task as a translaton task, use graphormer as SMILES's encoder and other word should be tokenized, use transformer decoder as our model's decoder. Use molecular SMILES as input and next word in PubMed papper's text as output!
I really appreciate Microsoft's progress in AI！