When BERT Plays the Lottery, All Tickets Are Winning Large Transformer-based models were shown to be reducible to a smaller number of self-attention h