Skip to content

Commit 695a880

Browse files
committed
Merge branch 'main' of https://www.github.com/google/flax into pipy
2 parents 5de87fa + a6e7744 commit 695a880

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

54 files changed

+309
-7438
lines changed

CHANGELOG.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,8 @@ vNext
3333
Breaking changes:
3434
- flax.deprecated.nn is removed. Please pin to flax==0.3.6 if you are still using it.
3535
- PixelCNN++ example is removed. It was not working well on TPU.
36+
- linen Normalization layers no longer downcast double and complex floats tofloat32
37+
when computing the mean and variance.
3638

3739
New features:
3840
- Added `flax.linen.custom_vjp` for custom derivatives inside a `Module`.

docs/design_notes/arguments.md

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -85,9 +85,9 @@ It also avoids a default value which would probably cause either the train step
8585

8686

8787

88-
## Functional Core and flax.nn
88+
## Functional Core
8989

90-
The old NN api and functional core define functions rather than classes.
91-
Therefore, there is no clear distinction between hyper parameters and call time arguments.
92-
The only way to pre-determine the hyper parameters is by using `partial`.
90+
Functional core defines functions rather than classes.
91+
Therefore, there is no clear distinction between hyperparameters and call-time arguments.
92+
The only way to pre-determine the hyperparameters is by using `partial`.
9393
On the upside, there are no ambiguous cases where method arguments could also be attributes.

docs/flax.nn.rst

Lines changed: 0 additions & 115 deletions
This file was deleted.

0 commit comments

Comments
 (0)