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

Output hidden states (#4978)

Browse Source 
* Configure all models to use output_hidden_states as argument passed to foward()

* Pass all tests

* Remove cast_bool_to_primitive in TF Flaubert model

* correct tf xlnet

* add pytorch test

* add tf test

* Fix broken tests

* Configure all models to use output_hidden_states as argument passed to foward()

* Pass all tests

* Remove cast_bool_to_primitive in TF Flaubert model

* correct tf xlnet

* add pytorch test

* add tf test

* Fix broken tests

* Refactor output_hidden_states for mobilebert

* Reset and remerge to master

Co-authored-by: Joseph Liu <joseph.liu@coinflex.com>
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>
This commit is contained in:
Joseph Liu
2020-06-22 22:10:45 +08:00
committed by GitHub
parent 866a8ccabb
commit f4e1f02210
34 changed files with 814 additions and 349 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
tests/test_modeling_common.py
View File

@@ -143,14 +143,13 @@ class ModelTesterMixin:
for model_class in self.all_model_classes:
inputs_dict["output_attentions"] = True
config.output_hidden_states = False
inputs_dict["output_hidden_states"] = False
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
attentions = outputs[-1]
self.assertEqual(model.config.output_hidden_states, False)
self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
# check that output_attentions also work using config
@@ -162,7 +161,6 @@ class ModelTesterMixin:
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
attentions = outputs[-1]
self.assertEqual(model.config.output_hidden_states, False)
self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
if chunk_length is not None:
@@ -201,14 +199,13 @@ class ModelTesterMixin:
# Check attention is always last and order is fine
inputs_dict["output_attentions"] = True
config.output_hidden_states = True
inputs_dict["output_hidden_states"] = True
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
self.assertEqual(out_len + (2 if self.is_encoder_decoder else 1), len(outputs))
self.assertEqual(model.config.output_hidden_states, True)
self_attentions = outputs[-1]
self.assertEqual(len(self_attentions), self.model_tester.num_hidden_layers)
@@ -493,19 +490,16 @@ class ModelTesterMixin:
self.assertDictEqual(model.config.pruned_heads, {0: [0], 1: [1, 2], 2: [1, 2]})
def test_hidden_states_output(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
config.output_hidden_states = True
def check_hidden_states_output(inputs_dict, config, model_class):
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
hidden_states = outputs[-1]
self.assertEqual(model.config.output_hidden_states, True)
self.assertEqual(len(hidden_states), self.model_tester.num_hidden_layers + 1)
self.assertEqual(len(hidden_states), self.model_tester.num_hidden_layers + 1)
if hasattr(self.model_tester, "encoder_seq_length"):
seq_length = self.model_tester.encoder_seq_length
if hasattr(self.model_tester, "chunk_length") and self.model_tester.chunk_length > 1:
@@ -517,6 +511,18 @@ class ModelTesterMixin:
list(hidden_states[0].shape[-2:]), [seq_length, self.model_tester.hidden_size],
)
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
inputs_dict["output_hidden_states"] = True
check_hidden_states_output(inputs_dict, config, model_class)
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
config.output_hidden_states = True
check_hidden_states_output(inputs_dict, config, model_class)
def test_resize_tokens_embeddings(self):
(original_config, inputs_dict,) = self.model_tester.prepare_config_and_inputs_for_common()
if not self.test_resize_embeddings: