Open Hannibal046 opened 2 years ago
Hi, there, a simple question about the training. When considering 2-hop retrieval, the input to the query encoder is (query,ground_truth_hop1_document) (a kind of teacher forcing?) or (query, document_retrieved_from_model) ?
(query,ground_truth_hop1_document)
(query, document_retrieved_from_model)
Thanks
I guess the former when it comes to training.
Hi, there, a simple question about the training. When considering 2-hop retrieval, the input to the query encoder is
(query,ground_truth_hop1_document)
(a kind of teacher forcing?) or(query, document_retrieved_from_model)
?Thanks