[R] Is there any research on allowing Transformers to spent more compute on more difficult to predict tokens? Submitted by Chemont t3_109z8om on January 12, 2023 at 1:07 PM in MachineLearning 16 comments 19