Hacky fix for dictionary output with tf 2.14 #933
Conversation
Codecov ReportPatch coverage:
Additional details and impacted files@@ Coverage Diff @@
## main keras-team/keras-core#933 +/- ##
==========================================
+ Coverage 76.82% 83.83% +7.00%
==========================================
Files 329 318 -11
Lines 31427 28743 -2684
Branches 6112 5491 -621
==========================================
- Hits 24144 24096 -48
+ Misses 5719 3141 -2578
+ Partials 1564 1506 -58
Flags with carried forward coverage won't be shown. Click here to find out more.
☔ View full report in Codecov by Sentry. |
| @@ -46,10 +47,20 @@ class Function(Operation): | |||
| def __init__(self, inputs, outputs, name=None): | |||
| super().__init__(name=name) | |||
|
|
|||
| if backend() == "tensorflow": | |||
There was a problem hiding this comment.
Can we do this in keras_core/backend/tensorflow/layer.py::TFLayer.__init__ instead?
There was a problem hiding this comment.
I don't think so, because we can't turn off tf auto tracking overall for functional models. We only want to turn it off for the input/output structs here to avoid the tf bug. Essentially we need a way to run a hook inside Function.__init__.
The nice way to avoid all this would be to use Tracker for our tf tracking. This could save a lot of bugs I think, but is more of an effort. https://github.com/keras-team/keras-core/issues/934#issuecomment-1728464445
There was a problem hiding this comment.
Maybe we could just not depend on AutoTrackable (just Trackable) and then manually track everything according to our own Tracker.
There was a problem hiding this comment.
Yeah, I think this would take some work, but really improve the overall state if we did.
mattdangerw
force-pushed
the
fix-tf-2.14-breakage
branch
from
4a040bb
to
55e9c46
Compare
| @@ -46,10 +47,20 @@ class Function(Operation): | |||
| def __init__(self, inputs, outputs, name=None): | |||
| super().__init__(name=name) | |||
|
|
|||
| if backend() == "tensorflow": | |||
There was a problem hiding this comment.
Maybe we could just not depend on AutoTrackable (just Trackable) and then manually track everything according to our own Tracker.
See keras-team/keras#18399 for more info. And other workaround we could consider.