← Quit0 / 0 correct
LLMs & AIHard1 of 409

In scaled dot-product attention, why divide the scores by √d (d = key dimension)?