@article{amixtureofh1headsisbetterthanhheads,
  title = {A Mixture of $h-1$ Heads is Better than $h$ Heads},
  author = {Hao Peng and Roy Schwartz and Dianqi Li and Noah A. Smith},
  year = {2020},
  eprint = {2005.06537},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2005.06537v1},
}