Given attention weights and pair of key/value, how to calculate query?

[hadaev8 (Had)]

I mean pytorch default multi head attention layer.