primeqa.qg.models.qg_model.QGModel

class primeqa.qg.models.qg_model.QGModel(model_path, modality='table', lang='en')

Bases: object

Methods

generate_questions
prune_hallucinations	Our approach to pruning hallucinated questions uses an entity lookup to check that a generated entity or a part of it is present in the hybrid context.

Attributes

model	Propery of TableQG model.
tokenizer	Property of TableQG model.

property model

Propery of TableQG model.

Returns

Sequence to sequence model object (based on model name)

prune_hallucinations(qdicts, num_instances=5, hallucination_prop=0.25)

Our approach to pruning hallucinated questions uses an entity lookup to check that a generated entity or a part of it is present in the hybrid context. If the entity is not present, the question and the entity are considered to be hallucinated.

Parameters

qdicts (list) – A list of dicts, each dict contains, an answer, context and (num_instances) generated questions.

Returns

Updated qdicts without any hallucinated questions. Each dict will have only one question now.

property tokenizer

Property of TableQG model.

Returns

Tokenizer class object based on the model name/ path