what's the train data in metaQA?

[ToneLi]

Dear author, I have a question, in metaQA, I have read your code, you used hop1 data to train a hop1 model, hop2 data to train a hop2 model, hop3 data to train a hop3 model, is it right? Should all these 3 datasets be used to train one model? Let me see, give a question, we do not know the hop number of this question in advance. Can you explain it?