A question about few-shot learning experiments

[JhuoW]

Thank you for your excellent work. In Tables 4 and 5, I noticed that Prodigy was not evaluated on the node classification task for the Cora dataset, nor on the link classification task for WN18RR. Could you please clarify why the paper does not report Prodigy’s results on these tasks?

[LechengKong]

Hi @JhuoW ,
Sorry for the delayed reply. We didn't report simply because Prodigy didn't include code for those two datasets, prodigy has a pretrained model based on a sentence transformer, and for the pretrained model to work for downstream tasks, we also need the same sentence transformer, and process the downstream task first. We didn't have a chance to incorporate that yet.