Sampling realistic text from large pretrained models is non-trivial. I came across this paper in one of ACL 2019 workshops: https://arxiv.org/pdf/1904.09751.pd

Sampling realistic text from large pretrained models is non-trivial. I came across this paper in one of ACL 2019 workshops: https://arxiv.org/pdf/1904.09751.pdf https://arxiv.org/pdf/1904.09751.pdf