You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Motivating example is RNN which returns a tuple of (last_hidden, all_hiddens), but usually only one of these is actually used in further computations. In such settings, gradient w.r.t. to non-target element of the tuple can be replaced with some kind of symbolic zero instead of actually allocated object.
The text was updated successfully, but these errors were encountered:
Motivating example is RNN which returns a tuple of
(last_hidden, all_hiddens)
, but usually only one of these is actually used in further computations. In such settings, gradient w.r.t. to non-target element of the tuple can be replaced with some kind of symbolic zero instead of actually allocated object.The text was updated successfully, but these errors were encountered: