The reaction can be either represented through a single column reaction_SMILES or using four columns (ligand, ...).

@kjappelbaum how do you want me to include this information?

Hey! Thanks for looking into this!

I think it would be good to have the rxn SMILES as an identifier such that we can aggregate on this across all rxn datasets. However, for rxns, I would go via prompt templates that you can specify in the yml. That is, we would, by default, sample a prompt that uses rxn-SMILES and one of the targets (here yield) to construct a prompt. However, you might suggest more that leverage the structure of this dataset.

I can later (this weekend) commit an example of this to your PR if this helps!

Thanks, again 💯