Differences between Squad models

Hello there!

I find a lot of models trained on Squad dataset but didnt find any description for that models. https://github.com/deepmipt/DeepPavlov/tree/master/deeppavlov/configs/squad Whats differences between for example squad_ru, squad_ru_bert, squad_ru_bert_infer, squad_ru_rubert.json and squad_ru_rubert_infer.json ?


Some of the models are described in doc page (in tables with results): http://docs.deeppavlov.ai/en/master/features/models/squad.html

  • *squad* - models trained on SQuAD dataset (En)
  • squad_ru* - models trained on SDSJ Task B dataset (Ru)
  • squad_zh* - models trained on DRCD dataset (Zh)
  • *bert* - model is based on BERT (En)
  • *rubert* - model is based on RuBERT (Ru)
  • *bert_multilingual* - model is based on Multilingual BERT
  • *infer - models based on BERT that can be used for texts with more than 512 subtokens.
  • Models with multi* prefix were trained to be used in multi-paragraph mode, when you feed several paragraphs to model and then select only one answer.
1 Like