Comment on New attention mechanisms that outperform standard multi-head attentionparentComments−toxik2yThen they should have. It isn’t clear to be that this change is measurable when the code is optimized. Even in the presented numbers the difference seems meh at times.
Comments
Then they should have. It isn’t clear to be that this change is measurable when the code is optimized. Even in the presented numbers the difference seems meh at times.