That requires to modify the kbp37, tacred, and multitacred pie dataset builders because they follow this format. The preprocessing pipelines of these datasets should use src.utils.documnt.token_based_document_with_entities_and_relations_to_text_based instead of the variants from src.document.conversion (see usage _token_based_document_with_entities_and_relations_to_text_based and remove that). Finally, src.utils.documnt.token_based_document_with_entities_and_relations_to_text_based should be moved back to src.document.conversion.
ArneBinder
commented
1 year ago
That requires to modify the
kbp37,tacred, andmultitacredpie dataset builders because they follow this format. The preprocessing pipelines of these datasets should usesrc.utils.documnt.token_based_document_with_entities_and_relations_to_text_basedinstead of the variants fromsrc.document.conversion(see usage_token_based_document_with_entities_and_relations_to_text_basedand remove that). Finally,src.utils.documnt.token_based_document_with_entities_and_relations_to_text_basedshould be moved back tosrc.document.conversion.