SQUAD model with defolt confugurations gets all the answers only from the first paragraph. Should be the context for each question divided on paragraphs? I supposed multisquad model is more appropriate in this case but it requires contexts with questions in train dataset. I coudn’t find any information in documentation, neither on github about the format of of train-v1.json and dev-v1.json files for multisquad_none. Could you, please, describe it or provide an example?
Check GitHub Issue #805: https://github.com/deepmipt/DeepPavlov/issues/805