This model contains the converted PyTorch checkpoint of the original Tensorflow model available in the TaPas repository . It is described in Herzig et al.'s (2021) paper Open Domain Question Answering over Tables via Dense Retrieval .

This model has 2 versions that can be used differing only in the table scoring head. The default one has an adapted table scoring head in order to be able to generate probabilities out of the logits. The other (non-default) version corresponds to the original checkpoint from the TaPas repository and can be accessed by setting revision="original" .

Usage

In Haystack

If you want to use this model for question-answering over tables, you can load it in Haystack :

from haystack.nodes import TableReader

table_reader = TableReader(model_name_or_path="deepset/tapas-large-nq-hn-reader")