dropout in attention function

[kwonmha]

I saw scores = dropout(scores) in line 331 of akt.py
This is the first time I see dropout applied to attention weight.
Any reference or reason for this?