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

Support various BERT relative position embeddings (2nd) (#8276)

Browse Source 
* Support BERT relative position embeddings

* Fix typo in README.md

* Address review comment

* Fix failing tests

* [tiny] Fix style_doc.py check by adding an empty line to configuration_bert.py

* make fix copies

* fix configs of electra and albert and fix longformer

* remove copy statement from longformer

* fix albert

* fix electra

* Add bert variants forward tests for various position embeddings

* [tiny] Fix style for test_modeling_bert.py

* improve docstring

* [tiny] improve docstring and remove unnecessary dependency

* [tiny] Remove unused import

* re-add to ALBERT

* make embeddings work for ALBERT

* add test for albert

Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
This commit is contained in:
zhiheng-huang
2020-11-24 05:40:53 -08:00
committed by GitHub
parent 9e71aa2f8f
commit 2c83b3c38d
16 changed files with 327 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
tests/test_modeling_albert.py
View File

@@ -106,7 +106,7 @@ class AlbertModelTester:
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
def create_and_check_albert_model(
def create_and_check_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = AlbertModel(config=config)
@@ -118,7 +118,7 @@ class AlbertModelTester:
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
self.parent.assertEqual(result.pooler_output.shape, (self.batch_size, self.hidden_size))
def create_and_check_albert_for_pretraining(
def create_and_check_for_pretraining(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = AlbertForPreTraining(config=config)
@@ -134,7 +134,7 @@ class AlbertModelTester:
self.parent.assertEqual(result.prediction_logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
self.parent.assertEqual(result.sop_logits.shape, (self.batch_size, config.num_labels))
def create_and_check_albert_for_masked_lm(
def create_and_check_for_masked_lm(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = AlbertForMaskedLM(config=config)
@@ -143,7 +143,7 @@ class AlbertModelTester:
result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
def create_and_check_albert_for_question_answering(
def create_and_check_for_question_answering(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = AlbertForQuestionAnswering(config=config)
@@ -159,7 +159,7 @@ class AlbertModelTester:
self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
def create_and_check_albert_for_sequence_classification(
def create_and_check_for_sequence_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_labels = self.num_labels
@@ -169,7 +169,7 @@ class AlbertModelTester:
result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
def create_and_check_albert_for_token_classification(
def create_and_check_for_token_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_labels = self.num_labels
@@ -179,7 +179,7 @@ class AlbertModelTester:
result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
def create_and_check_albert_for_multiple_choice(
def create_and_check_for_multiple_choice(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_choices = self.num_choices
@@ -250,29 +250,35 @@ class AlbertModelTest(ModelTesterMixin, unittest.TestCase):
def test_config(self):
self.config_tester.run_common_tests()
def test_albert_model(self):
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_model(*config_and_inputs)
self.model_tester.create_and_check_model(*config_and_inputs)
def test_for_pretraining(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_for_pretraining(*config_and_inputs)
self.model_tester.create_and_check_for_pretraining(*config_and_inputs)
def test_for_masked_lm(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_for_masked_lm(*config_and_inputs)
self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)
def test_for_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_for_multiple_choice(*config_and_inputs)
self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
def test_for_question_answering(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_for_question_answering(*config_and_inputs)
self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
def test_for_sequence_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_albert_for_sequence_classification(*config_and_inputs)
self.model_tester.create_and_check_for_sequence_classification(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
@slow
def test_model_from_pretrained(self):

46
tests/test_modeling_bert.py
View File

@@ -404,6 +404,12 @@ class BertModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_as_decoder(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_model_as_decoder(*config_and_inputs)
@@ -477,3 +483,43 @@ class BertModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = BertModel.from_pretrained(model_name)
self.assertIsNotNone(model)
@require_torch
class BertModelIntegrationTest(unittest.TestCase):
@slow
def test_inference_no_head_absolute_embedding(self):
model = BertModel.from_pretrained("bert-base-uncased")
input_ids = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
output = model(input_ids)[0]
expected_shape = torch.Size((1, 11, 768))
self.assertEqual(output.shape, expected_shape)
expected_slice = torch.tensor(
[[[-0.0483, 0.1188, -0.0313], [-0.0606, 0.1435, 0.0199], [-0.0235, 0.1519, 0.0175]]]
)
self.assertTrue(torch.allclose(output[:, :3, :3], expected_slice, atol=1e-4))
@slow
def test_inference_no_head_relative_embedding_key(self):
model = BertModel.from_pretrained("zhiheng-huang/bert-base-uncased-embedding-relative-key")
input_ids = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
output = model(input_ids)[0]
expected_shape = torch.Size((1, 11, 768))
self.assertEqual(output.shape, expected_shape)
expected_slice = torch.tensor(
[[[0.3492, 0.4126, -0.1484], [0.2274, -0.0549, 0.1623], [0.5889, 0.6797, -0.0189]]]
)
self.assertTrue(torch.allclose(output[:, :3, :3], expected_slice, atol=1e-4))
@slow
def test_inference_no_head_relative_embedding_key_query(self):
model = BertModel.from_pretrained("zhiheng-huang/bert-base-uncased-embedding-relative-key-query")
input_ids = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
output = model(input_ids)[0]
expected_shape = torch.Size((1, 11, 768))
self.assertEqual(output.shape, expected_shape)
expected_slice = torch.tensor([[[1.1677, 0.5129, 0.9524], [0.6659, 0.5958, 0.6688], [1.1714, 0.1764, 0.6266]]])
self.assertTrue(torch.allclose(output[:, :3, :3], expected_slice, atol=1e-4))

6
tests/test_modeling_electra.py
View File

@@ -309,6 +309,12 @@ class ElectraModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_electra_model(*config_and_inputs)
def test_electra_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_electra_model(*config_and_inputs)
def test_for_masked_lm(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_electra_for_masked_lm(*config_and_inputs)

6
tests/test_modeling_layoutlm.py
View File

@@ -199,6 +199,12 @@ class LayoutLMModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
def test_for_masked_lm(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)

10
tests/test_modeling_roberta.py
View File

@@ -17,7 +17,7 @@
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow, torch_device
from transformers.testing_utils import require_torch, slow, torch_device
from .test_configuration_common import ConfigTester
from .test_generation_utils import GenerationTesterMixin
@@ -295,6 +295,12 @@ class RobertaModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCas
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_as_decoder(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_model_as_decoder(*config_and_inputs)
@@ -395,8 +401,6 @@ class RobertaModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCas
self.assertTrue(torch.all(torch.eq(position_ids, expected_positions)))
@require_sentencepiece
@require_tokenizers
@require_torch
class RobertaModelIntegrationTest(unittest.TestCase):
@slow