21st century AD
gradient
is now guaranteed to return a tuple with one gradient per argument, even if the gradients are nothing
.nothing
returned from a now will now be consistently collapsed into a single nothing
.Merged pull requests:
Expr(:boundscheck)
(#1462) (@ToucheSir)Zygote.accum
(#1491) (@lxvm)@adjoint
s for sort
and filter
(#1493) (@lkdvos)gradient
(#1495) (@ToucheSir)Closed issues:
sort(x; rev=true)
is not supported (#1492)plan_rfft(x) * x
(#1496)Merged pull requests:
Closed issues:
Closed issues:
gradient
very slow when function call involves indexing (#1436)Merged pull requests:
f
to return a Tuple in withgradient(f, args...)
(#1419) (@mcabbott)parent
for LowerTriangular
and UpperTriangular
(#1444) (@torfjelde)Closed issues:
(nothing,)
on a Lux Network. Gradient works. (#1422)Merged pull requests: