CTRL was trained with a causal language modeling (CLM) objective and is therefore powerful at predicting the next
  token in a sequence.