-
Notifications
You must be signed in to change notification settings - Fork 6
Open
Description
if (N / C) < 1: x = ((q @ k.transpose(-2, -1)) @ v).transpose(1, 2).reshape(B, N, C) else: x = (q @ (k.transpose(-2, -1) @ v)).transpose(1, 2).reshape(B, N, C)
should be
if (N / (C/H)) < 1: x = ((q @ k.transpose(-2, -1)) @ v).transpose(1, 2).reshape(B, N, C) else: x = (q @ (k.transpose(-2, -1) @ v)).transpose(1, 2).reshape(B, N, C)
?
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels