Extraction

We also expose a standard question answering model that can retrieve a short span for a given query from a larger retrieved object. We use a BERT-base architecture that is by default trained on Squad2.0. Performance of the model can vary.