3 ms·Speeding Up Transformer Training and Inference by Increasing Model Size3 points by bjourne 6y ago