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

Generalize problem_type to all sequence classification models (#14180)

Browse Source 
* Generalize problem_type to all classification models

* Missing import

* Deberta BC and fix tests

* Fix template

* Missing imports

* Revert change to reformer test

* Fix style
This commit is contained in:
Sylvain Gugger
2021-10-29 10:32:56 -04:00
committed by GitHub
parent 4ab6a4a086
commit c28bc80bbb
38 changed files with 474 additions and 191 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/test_modeling_albert.py
View File

@@ -234,8 +234,6 @@ class AlbertModelTest(ModelTesterMixin, unittest.TestCase):
fx_ready_model_classes = all_model_classes
fx_dynamic_ready_model_classes = all_model_classes
test_sequence_classification_problem_types = True
# special case for ForPreTraining model
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
inputs_dict = super()._prepare_for_class(inputs_dict, model_class, return_labels=return_labels)

1
tests/test_modeling_bert.py
View File

@@ -446,7 +446,6 @@ class BertModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
all_generative_model_classes = (BertLMHeadModel,) if is_torch_available() else ()
fx_ready_model_classes = all_model_classes
fx_dynamic_ready_model_classes = all_model_classes
test_sequence_classification_problem_types = True
# special case for ForPreTraining model
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):

1
tests/test_modeling_big_bird.py
View File

@@ -435,7 +435,6 @@ class BigBirdModelTest(ModelTesterMixin, unittest.TestCase):
# head masking & pruning is currently not supported for big bird
test_head_masking = False
test_pruning = False
test_sequence_classification_problem_types = True
# torchscript should be possible, but takes prohibitively long to test.
# Also torchscript is not an important feature to have in the beginning.

26
tests/test_modeling_common.py
View File

@@ -113,7 +113,6 @@ class ModelTesterMixin:
test_missing_keys = True
test_model_parallel = False
is_encoder_decoder = False
test_sequence_classification_problem_types = False
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
inputs_dict = copy.deepcopy(inputs_dict)
@@ -387,12 +386,13 @@ class ModelTesterMixin:
if not self.model_tester.is_training:
return
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.return_dict = True
for model_class in self.all_model_classes:
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.return_dict = True
if model_class in get_values(MODEL_MAPPING):
continue
model = model_class(config)
model.to(torch_device)
model.train()
@@ -401,14 +401,14 @@ class ModelTesterMixin:
loss.backward()
def test_training_gradient_checkpointing(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
if not self.model_tester.is_training:
return
config.use_cache = False
config.return_dict = True
for model_class in self.all_model_classes:
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.use_cache = False
config.return_dict = True
if model_class in get_values(MODEL_MAPPING) or not model_class.supports_gradient_checkpointing:
continue
model = model_class(config)
@@ -1842,9 +1842,6 @@ class ModelTesterMixin:
model.generate(**cast_to_device(inputs_dict, "cuda:0"), num_beams=2)
def test_problem_types(self):
if not self.test_sequence_classification_problem_types:
return
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
problem_types = [
@@ -1880,7 +1877,11 @@ class ModelTesterMixin:
# See https://github.com/huggingface/transformers/issues/11780
with warnings.catch_warnings(record=True) as warning_list:
loss = model(**inputs).loss
self.assertListEqual(warning_list, [])
for w in warning_list:
if "Using a target size that is different to the input size" in str(w.message):
raise ValueError(
f"Something is going wrong in the regression problem: intercepted {w.message}"
)
loss.backward()
@@ -2184,7 +2185,6 @@ class ModelPushToHubTester(unittest.TestCase):
f.write(FAKE_MODEL_CODE)
repo.push_to_hub()
print(os.listdir(tmp_dir))
new_model = AutoModel.from_pretrained(f"{USER}/test-dynamic-model", trust_remote_code=True)
for p1, p2 in zip(model.parameters(), new_model.parameters()):

1
tests/test_modeling_convbert.py
View File

@@ -262,7 +262,6 @@ class ConvBertModelTest(ModelTesterMixin, unittest.TestCase):
)
test_pruning = False
test_head_masking = False
test_sequence_classification_problem_types = True
def setUp(self):
self.model_tester = ConvBertModelTester(self)

1
tests/test_modeling_distilbert.py
View File

@@ -214,7 +214,6 @@ class DistilBertModelTest(ModelTesterMixin, unittest.TestCase):
test_pruning = True
test_torchscript = True
test_resize_embeddings = True
test_sequence_classification_problem_types = True
test_resize_position_embeddings = True
def setUp(self):

1
tests/test_modeling_electra.py
View File

@@ -291,7 +291,6 @@ class ElectraModelTest(ModelTesterMixin, unittest.TestCase):
)
fx_ready_model_classes = all_model_classes
fx_dynamic_ready_model_classes = all_model_classes
test_sequence_classification_problem_types = True
# special case for ForPreTraining model
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):

1
tests/test_modeling_funnel.py
View File

@@ -362,7 +362,6 @@ class FunnelModelTest(ModelTesterMixin, unittest.TestCase):
if is_torch_available()
else ()
)
test_sequence_classification_problem_types = True
# special case for ForPreTraining model
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):

1
tests/test_modeling_longformer.py
View File

@@ -278,7 +278,6 @@ class LongformerModelTester:
class LongformerModelTest(ModelTesterMixin, unittest.TestCase):
test_pruning = False # pruning is not supported
test_torchscript = False
test_sequence_classification_problem_types = True
all_model_classes = (
(

1
tests/test_modeling_mobilebert.py
View File

@@ -271,7 +271,6 @@ class MobileBertModelTest(ModelTesterMixin, unittest.TestCase):
)
fx_ready_model_classes = all_model_classes
fx_dynamic_ready_model_classes = all_model_classes
test_sequence_classification_problem_types = True
# special case for ForPreTraining model
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):

2
tests/test_modeling_openai.py
View File

@@ -143,7 +143,7 @@ class OpenAIGPTModelTester:
model = OpenAIGPTForSequenceClassification(config)
model.to(torch_device)
model.eval()
# print(config.num_labels, sequence_labels.size())
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
result = model(input_ids, token_type_ids=token_type_ids, labels=sequence_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))

4
tests/test_modeling_reformer.py
View File

@@ -795,6 +795,10 @@ class ReformerLSHAttnModelTest(ReformerTesterMixin, ModelTesterMixin, Generation
[expected_shape] * len(iter_hidden_states),
)
def test_problem_types(self):
# Fails because the sequence length is not a multiple of 4
pass
@require_torch
@require_sentencepiece

1
tests/test_modeling_roberta.py
View File

@@ -356,7 +356,6 @@ class RobertaModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCas
else ()
)
all_generative_model_classes = (RobertaForCausalLM,) if is_torch_available() else ()
test_sequence_classification_problem_types = True
def setUp(self):
self.model_tester = RobertaModelTester(self)

1
tests/test_modeling_squeezebert.py
View File

@@ -232,7 +232,6 @@ class SqueezeBertModelTest(ModelTesterMixin, unittest.TestCase):
test_torchscript = True
test_resize_embeddings = True
test_head_masking = False
test_sequence_classification_problem_types = True
def setUp(self):
self.model_tester = SqueezeBertModelTester(self)

1
tests/test_modeling_xlm.py
View File

@@ -350,7 +350,6 @@ class XLMModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
all_generative_model_classes = (
(XLMWithLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Check other models whether language generation is also applicable
test_sequence_classification_problem_types = True
# XLM has 2 QA models -> need to manually set the correct labels for one of them here
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):

1
tests/test_modeling_xlnet.py
View File

@@ -527,7 +527,6 @@ class XLNetModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase)
(XLNetLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Check other models whether language generation is also applicable
test_pruning = False
test_sequence_classification_problem_types = True
# XLNet has 2 QA models -> need to manually set the correct labels for one of them here
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):