Adding new train_step logic to make things less confusing for users #15994
sgugger
approved these changes
on 2022-03-29
gante
approved these changes
on 2022-03-29
Adding new train_step logic to make things less confusing for users
fae607ae
DO NOT ASK WHY WE NEED THAT SUBCLASS
02505441
Metrics now working, at least for single-output models with type anno…
0e294f5f
Updates and TODOs for the new train_step
125235ed
Make fixup
119cbf24
Temporary test workaround until T5 has types
892176d1
Temporary test workaround until T5 has types
ec766956
I think this actually works! Needs a lot of tests though
1053bdc7
MAke style/quality
9557efdd
Revert changes to T5 tests
6e3f09a2
Deleting the aforementioned unmentionable subclass
5851cc36
Deleting the aforementioned unmentionable subclass
b8137b14
Adding a Keras API test
98935fe3
Style fixes
82cc3c0e
Removing unneeded TODO and comments
2f5a3896
Update test_step too
a3556a8f
Stop trying to compute metrics with the dummy_loss, patch up test
f5dc36ee
Make style
680f6d22
make fixup
8d805d8b
Docstring cleanup
8ed327fa
make fixup
8226e7b1
make fixup
ca6bd5bf
Stop expanding 1D input tensors when using dummy loss
b7ae0618
Adjust T5 test given the new compile()
b015f5cb
make fixup
28d58da8
Skipping test for convnext
8666fdba
Removing old T5-specific Keras test now that we have a common one
6bb0621d
make fixup
d34d4282
make fixup
3c4465a2
Only skip convnext test on CPU
bd414d34
Update src/transformers/modeling_tf_utils.py
11a608f8
Update src/transformers/modeling_tf_utils.py
5e01947b
Avoiding TF import issues
0b5287af
make fixup
196a90b0
Update compile() to support TF 2.3
9548480a
Skipping model.fit() on template classes for now
c03492c8
Skipping model.fit() on template class tests for now
81ad430c
Replace ad-hoc solution with find_labels
e961c739
make fixup
f73c90ab
Rocketknight1
deleted the tf_train_step_copies branch 3 years ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub