Answer Level Calibration

AbstractPre-trained language models have recently shown that training on large corpora using the language modeling objective enables few-shot and zero-shot capabilities on a variety of NLP tasks, including commonsense reasoning …...