Asking word
Where each head looks most
Each row names the word that head gives the largest share to, and the size of that share. Every head's six shares add up to 100%.
Part 1 of 10
AnatomyLesson 6 of 7
A word needs several things from its sentence at once. A layer runs several attention heads side by side, each with a question of its own.
Each row names the word that head gives the largest share to, and the size of that share. Every head's six shares add up to 100%.
Part 1 of 10