This is just and example of "lying with statistics". Going by compute efficiency the gap has already closed (both in training and inference coincidentally).
This is just and example of "lying with statistics". Going by compute efficiency the gap has already closed (both in training and inference coincidentally).