This is the same dataset as the question_generator dataset but with the context removed and the question and answer in separate fields. This is intended to be used with the question_generator repo to train the qa_evaluator model which predicts whether a question and answer pair makes sense.