SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

TF: Finalize unpack_inputs-related changes (#16499)

Browse Source 
* Add unpack_inputs to remaining models

* removed kwargs to `call()` in TF models

* fix TF T5 tests
This commit is contained in:
Joao Gante
2022-04-04 16:37:33 +01:00
committed by GitHub
parent be9474bd35
commit dad5ca83b2
46 changed files with 78 additions and 287 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
tests/convbert/test_modeling_tf_convbert.py
View File

@@ -355,7 +355,6 @@ class TFConvBertModelTest(TFModelTesterMixin, unittest.TestCase):
for model_class in self.all_model_classes:
inputs_dict["output_attentions"] = True
inputs_dict["use_cache"] = False
config.output_hidden_states = False
model = model_class(config)
outputs = model(self._prepare_for_class(inputs_dict, model_class))

5
tests/t5/test_modeling_tf_t5.py
View File

@@ -346,6 +346,11 @@ class TFT5ModelTest(TFModelTesterMixin, unittest.TestCase):
self.assertEqual(model.get_input_embeddings().weight.shape[0], len(tokenizer))
self.assertNotEqual(model.get_input_embeddings().weight.shape[0], original_vocab_size)
# This test is run in `TFT5EncoderOnlyModelTest`, where the main layer has the same inputs as the model
@unittest.skip(reason="The inputs of the Main Layer are different.")
def test_keras_save_load(self):
pass
class TFT5EncoderOnlyModelTester:
def __init__(

40
tests/test_modeling_tf_common.py
View File

@@ -573,7 +573,12 @@ class TFModelTesterMixin:
pt_model = pt_model_class(config)
tf_inputs_dict = self._prepare_for_class(inputs_dict, model_class)
tf_inputs_dict_maybe_with_labels = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
tf_inputs_dict_maybe_with_labels = self._prepare_for_class(
inputs_dict,
model_class,
# Not all models accept "labels" in the forward pass (yet :) )
return_labels=True if "labels" in inspect.signature(model_class.call).parameters.keys() else False,
)
# Check we can load pt model in tf and vice-versa with model => model functions
tf_model = transformers.load_pytorch_model_in_tf2_model(tf_model, pt_model, tf_inputs=tf_inputs_dict)
@@ -722,7 +727,6 @@ class TFModelTesterMixin:
for model_class in self.all_model_classes:
inputs_dict["output_attentions"] = True
inputs_dict["use_cache"] = False
config.output_hidden_states = False
model = model_class(config)
outputs = model(self._prepare_for_class(inputs_dict, model_class))
@@ -944,10 +948,6 @@ class TFModelTesterMixin:
dict_inputs = self._prepare_for_class(inputs_dict, model_class)
check_equivalence(model, tuple_inputs, dict_inputs)
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs)
tuple_inputs = self._prepare_for_class(inputs_dict, model_class)
dict_inputs = self._prepare_for_class(inputs_dict, model_class)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_hidden_states": True})
@@ -956,19 +956,25 @@ class TFModelTesterMixin:
dict_inputs = self._prepare_for_class(inputs_dict, model_class)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_attentions": True})
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_hidden_states": True})
# Not all models accept "labels" in the forward pass (yet :) )
if "labels" in inspect.signature(model.call).parameters.keys():
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs)
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_attentions": True})
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_hidden_states": True})
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(
model, tuple_inputs, dict_inputs, {"output_hidden_states": True, "output_attentions": True}
)
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(model, tuple_inputs, dict_inputs, {"output_attentions": True})
tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
check_equivalence(
model, tuple_inputs, dict_inputs, {"output_hidden_states": True, "output_attentions": True}
)
def test_inputs_embeds(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()