test_modeling_bert.py 19.3 KB
Newer Older
thomwolf's avatar
thomwolf committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
Aymeric Augustin's avatar
Aymeric Augustin committed
15
from __future__ import absolute_import, division, print_function
thomwolf's avatar
thomwolf committed
16

17
from transformers import is_torch_available
thomwolf's avatar
thomwolf committed
18

19
20
from .test_configuration_common import ConfigTester
from .test_modeling_common import CommonTestCases, floats_tensor, ids_tensor
21
from .utils import CACHE_DIR, require_torch, slow, torch_device
thomwolf's avatar
thomwolf committed
22

Aymeric Augustin's avatar
Aymeric Augustin committed
23

24
if is_torch_available():
25
26
27
28
29
30
31
32
33
34
35
    from transformers import (
        BertConfig,
        BertModel,
        BertForMaskedLM,
        BertForNextSentencePrediction,
        BertForPreTraining,
        BertForQuestionAnswering,
        BertForSequenceClassification,
        BertForTokenClassification,
        BertForMultipleChoice,
    )
36
    from transformers.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_MAP
thomwolf's avatar
thomwolf committed
37

thomwolf's avatar
thomwolf committed
38

39
@require_torch
thomwolf's avatar
thomwolf committed
40
41
class BertModelTest(CommonTestCases.CommonModelTester):

42
43
44
45
46
47
48
49
50
51
52
53
54
    all_model_classes = (
        (
            BertModel,
            BertForMaskedLM,
            BertForNextSentencePrediction,
            BertForPreTraining,
            BertForQuestionAnswering,
            BertForSequenceClassification,
            BertForTokenClassification,
        )
        if is_torch_available()
        else ()
    )
thomwolf's avatar
thomwolf committed
55

thomwolf's avatar
thomwolf committed
56
    class BertModelTester(object):
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
        def __init__(
            self,
            parent,
            batch_size=13,
            seq_length=7,
            is_training=True,
            use_input_mask=True,
            use_token_type_ids=True,
            use_labels=True,
            vocab_size=99,
            hidden_size=32,
            num_hidden_layers=5,
            num_attention_heads=4,
            intermediate_size=37,
            hidden_act="gelu",
            hidden_dropout_prob=0.1,
            attention_probs_dropout_prob=0.1,
            max_position_embeddings=512,
            type_vocab_size=16,
            type_sequence_label_size=2,
            initializer_range=0.02,
            num_labels=3,
            num_choices=4,
            scope=None,
        ):
thomwolf's avatar
thomwolf committed
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
            self.parent = parent
            self.batch_size = batch_size
            self.seq_length = seq_length
            self.is_training = is_training
            self.use_input_mask = use_input_mask
            self.use_token_type_ids = use_token_type_ids
            self.use_labels = use_labels
            self.vocab_size = vocab_size
            self.hidden_size = hidden_size
            self.num_hidden_layers = num_hidden_layers
            self.num_attention_heads = num_attention_heads
            self.intermediate_size = intermediate_size
            self.hidden_act = hidden_act
            self.hidden_dropout_prob = hidden_dropout_prob
            self.attention_probs_dropout_prob = attention_probs_dropout_prob
            self.max_position_embeddings = max_position_embeddings
            self.type_vocab_size = type_vocab_size
            self.type_sequence_label_size = type_sequence_label_size
            self.initializer_range = initializer_range
            self.num_labels = num_labels
            self.num_choices = num_choices
            self.scope = scope

        def prepare_config_and_inputs(self):
106
            input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
thomwolf's avatar
thomwolf committed
107
108
109

            input_mask = None
            if self.use_input_mask:
110
                input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
thomwolf's avatar
thomwolf committed
111
112
113

            token_type_ids = None
            if self.use_token_type_ids:
114
                token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
thomwolf's avatar
thomwolf committed
115
116
117
118
119

            sequence_labels = None
            token_labels = None
            choice_labels = None
            if self.use_labels:
120
121
122
                sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
                token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
                choice_labels = ids_tensor([self.batch_size], self.num_choices)
thomwolf's avatar
thomwolf committed
123
124

            config = BertConfig(
thomwolf's avatar
thomwolf committed
125
                vocab_size=self.vocab_size,
thomwolf's avatar
thomwolf committed
126
127
128
129
130
131
132
133
134
                hidden_size=self.hidden_size,
                num_hidden_layers=self.num_hidden_layers,
                num_attention_heads=self.num_attention_heads,
                intermediate_size=self.intermediate_size,
                hidden_act=self.hidden_act,
                hidden_dropout_prob=self.hidden_dropout_prob,
                attention_probs_dropout_prob=self.attention_probs_dropout_prob,
                max_position_embeddings=self.max_position_embeddings,
                type_vocab_size=self.type_vocab_size,
135
                is_decoder=False,
136
137
                initializer_range=self.initializer_range,
            )
thomwolf's avatar
thomwolf committed
138
139
140

            return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels

141
        def prepare_config_and_inputs_for_decoder(self):
142
143
144
145
146
147
148
149
150
            (
                config,
                input_ids,
                token_type_ids,
                input_mask,
                sequence_labels,
                token_labels,
                choice_labels,
            ) = self.prepare_config_and_inputs()
151
152
153
154
155

            config.is_decoder = True
            encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
            encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)

156
157
158
159
160
161
162
163
164
165
166
            return (
                config,
                input_ids,
                token_type_ids,
                input_mask,
                sequence_labels,
                token_labels,
                choice_labels,
                encoder_hidden_states,
                encoder_attention_mask,
            )
167

thomwolf's avatar
thomwolf committed
168
        def check_loss_output(self, result):
169
            self.parent.assertListEqual(list(result["loss"].size()), [])
thomwolf's avatar
thomwolf committed
170

171
172
173
        def create_and_check_bert_model(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
174
            model = BertModel(config=config)
175
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
176
            model.eval()
177
178
            sequence_output, pooled_output = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
            sequence_output, pooled_output = model(input_ids, token_type_ids=token_type_ids)
thomwolf's avatar
thomwolf committed
179
            sequence_output, pooled_output = model(input_ids)
thomwolf's avatar
thomwolf committed
180
181
182
183
184
185

            result = {
                "sequence_output": sequence_output,
                "pooled_output": pooled_output,
            }
            self.parent.assertListEqual(
186
187
                list(result["sequence_output"].size()), [self.batch_size, self.seq_length, self.hidden_size]
            )
thomwolf's avatar
thomwolf committed
188
189
            self.parent.assertListEqual(list(result["pooled_output"].size()), [self.batch_size, self.hidden_size])

190
191
192
193
194
195
196
197
198
199
200
201
        def create_and_check_bert_model_as_decoder(
            self,
            config,
            input_ids,
            token_type_ids,
            input_mask,
            sequence_labels,
            token_labels,
            choice_labels,
            encoder_hidden_states,
            encoder_attention_mask,
        ):
202
            model = BertModel(config)
203
            model.to(torch_device)
204
            model.eval()
205
206
207
208
209
210
211
212
213
214
215
216
217
            sequence_output, pooled_output = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                encoder_hidden_states=encoder_hidden_states,
                encoder_attention_mask=encoder_attention_mask,
            )
            sequence_output, pooled_output = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                encoder_hidden_states=encoder_hidden_states,
            )
218
219
220
221
222
223
224
            sequence_output, pooled_output = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)

            result = {
                "sequence_output": sequence_output,
                "pooled_output": pooled_output,
            }
            self.parent.assertListEqual(
225
226
                list(result["sequence_output"].size()), [self.batch_size, self.seq_length, self.hidden_size]
            )
227
228
            self.parent.assertListEqual(list(result["pooled_output"].size()), [self.batch_size, self.hidden_size])

229
230
231
        def create_and_check_bert_for_masked_lm(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
232
            model = BertForMaskedLM(config=config)
233
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
234
            model.eval()
235
236
237
            loss, prediction_scores = model(
                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, masked_lm_labels=token_labels
            )
thomwolf's avatar
thomwolf committed
238
239
240
241
242
            result = {
                "loss": loss,
                "prediction_scores": prediction_scores,
            }
            self.parent.assertListEqual(
243
244
                list(result["prediction_scores"].size()), [self.batch_size, self.seq_length, self.vocab_size]
            )
thomwolf's avatar
thomwolf committed
245
246
            self.check_loss_output(result)

247
248
249
250
251
252
253
254
255
256
257
258
        def create_and_check_bert_model_for_masked_lm_as_decoder(
            self,
            config,
            input_ids,
            token_type_ids,
            input_mask,
            sequence_labels,
            token_labels,
            choice_labels,
            encoder_hidden_states,
            encoder_attention_mask,
        ):
259
            model = BertForMaskedLM(config=config)
260
            model.to(torch_device)
261
            model.eval()
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
            loss, prediction_scores = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                masked_lm_labels=token_labels,
                encoder_hidden_states=encoder_hidden_states,
                encoder_attention_mask=encoder_attention_mask,
            )
            loss, prediction_scores = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                masked_lm_labels=token_labels,
                encoder_hidden_states=encoder_hidden_states,
            )
277
278
279
280
281
            result = {
                "loss": loss,
                "prediction_scores": prediction_scores,
            }
            self.parent.assertListEqual(
282
283
                list(result["prediction_scores"].size()), [self.batch_size, self.seq_length, self.vocab_size]
            )
284
285
            self.check_loss_output(result)

286
287
288
        def create_and_check_bert_for_next_sequence_prediction(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
289
            model = BertForNextSentencePrediction(config=config)
290
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
291
            model.eval()
292
293
294
295
296
297
            loss, seq_relationship_score = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                next_sentence_label=sequence_labels,
            )
thomwolf's avatar
thomwolf committed
298
299
300
301
            result = {
                "loss": loss,
                "seq_relationship_score": seq_relationship_score,
            }
302
            self.parent.assertListEqual(list(result["seq_relationship_score"].size()), [self.batch_size, 2])
thomwolf's avatar
thomwolf committed
303
304
            self.check_loss_output(result)

305
306
307
        def create_and_check_bert_for_pretraining(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
308
            model = BertForPreTraining(config=config)
309
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
310
            model.eval()
311
312
313
314
315
316
317
            loss, prediction_scores, seq_relationship_score = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                masked_lm_labels=token_labels,
                next_sentence_label=sequence_labels,
            )
thomwolf's avatar
thomwolf committed
318
319
320
321
322
323
            result = {
                "loss": loss,
                "prediction_scores": prediction_scores,
                "seq_relationship_score": seq_relationship_score,
            }
            self.parent.assertListEqual(
324
325
326
                list(result["prediction_scores"].size()), [self.batch_size, self.seq_length, self.vocab_size]
            )
            self.parent.assertListEqual(list(result["seq_relationship_score"].size()), [self.batch_size, 2])
thomwolf's avatar
thomwolf committed
327
328
            self.check_loss_output(result)

329
330
331
        def create_and_check_bert_for_question_answering(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
332
            model = BertForQuestionAnswering(config=config)
333
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
334
            model.eval()
335
336
337
338
339
340
341
            loss, start_logits, end_logits = model(
                input_ids,
                attention_mask=input_mask,
                token_type_ids=token_type_ids,
                start_positions=sequence_labels,
                end_positions=sequence_labels,
            )
thomwolf's avatar
thomwolf committed
342
343
344
345
346
            result = {
                "loss": loss,
                "start_logits": start_logits,
                "end_logits": end_logits,
            }
347
348
            self.parent.assertListEqual(list(result["start_logits"].size()), [self.batch_size, self.seq_length])
            self.parent.assertListEqual(list(result["end_logits"].size()), [self.batch_size, self.seq_length])
thomwolf's avatar
thomwolf committed
349
350
            self.check_loss_output(result)

351
352
353
        def create_and_check_bert_for_sequence_classification(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
354
355
            config.num_labels = self.num_labels
            model = BertForSequenceClassification(config)
356
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
357
            model.eval()
358
359
360
            loss, logits = model(
                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels
            )
thomwolf's avatar
thomwolf committed
361
362
363
364
            result = {
                "loss": loss,
                "logits": logits,
            }
365
            self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.num_labels])
thomwolf's avatar
thomwolf committed
366
367
            self.check_loss_output(result)

368
369
370
        def create_and_check_bert_for_token_classification(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
371
372
            config.num_labels = self.num_labels
            model = BertForTokenClassification(config=config)
373
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
374
            model.eval()
375
376
377
            loss, logits = model(
                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels
            )
thomwolf's avatar
thomwolf committed
378
379
380
381
382
            result = {
                "loss": loss,
                "logits": logits,
            }
            self.parent.assertListEqual(
383
384
                list(result["logits"].size()), [self.batch_size, self.seq_length, self.num_labels]
            )
thomwolf's avatar
thomwolf committed
385
386
            self.check_loss_output(result)

387
388
389
        def create_and_check_bert_for_multiple_choice(
            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
        ):
thomwolf's avatar
thomwolf committed
390
391
            config.num_choices = self.num_choices
            model = BertForMultipleChoice(config=config)
392
            model.to(torch_device)
thomwolf's avatar
thomwolf committed
393
394
395
396
            model.eval()
            multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
            multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
            multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
397
398
399
400
401
402
            loss, logits = model(
                multiple_choice_inputs_ids,
                attention_mask=multiple_choice_input_mask,
                token_type_ids=multiple_choice_token_type_ids,
                labels=choice_labels,
            )
thomwolf's avatar
thomwolf committed
403
404
405
406
            result = {
                "loss": loss,
                "logits": logits,
            }
407
            self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.num_choices])
thomwolf's avatar
thomwolf committed
408
409
            self.check_loss_output(result)

thomwolf's avatar
thomwolf committed
410
411
        def prepare_config_and_inputs_for_common(self):
            config_and_inputs = self.prepare_config_and_inputs()
412
413
414
415
416
417
418
419
420
421
            (
                config,
                input_ids,
                token_type_ids,
                input_mask,
                sequence_labels,
                token_labels,
                choice_labels,
            ) = config_and_inputs
            inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
thomwolf's avatar
thomwolf committed
422
            return config, inputs_dict
thomwolf's avatar
thomwolf committed
423

thomwolf's avatar
thomwolf committed
424
425
426
    def setUp(self):
        self.model_tester = BertModelTest.BertModelTester(self)
        self.config_tester = ConfigTester(self, config_class=BertConfig, hidden_size=37)
thomwolf's avatar
thomwolf committed
427
428

    def test_config(self):
thomwolf's avatar
thomwolf committed
429
        self.config_tester.run_common_tests()
thomwolf's avatar
thomwolf committed
430

431
    def test_bert_model(self):
thomwolf's avatar
thomwolf committed
432
433
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_model(*config_and_inputs)
thomwolf's avatar
thomwolf committed
434

435
436
437
438
    def test_bert_model_as_decoder(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
        self.model_tester.create_and_check_bert_model_as_decoder(*config_and_inputs)

thomwolf's avatar
thomwolf committed
439
440
441
    def test_for_masked_lm(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_masked_lm(*config_and_inputs)
thomwolf's avatar
thomwolf committed
442

443
444
445
446
    def test_for_masked_lm_decoder(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
        self.model_tester.create_and_check_bert_model_for_masked_lm_as_decoder(*config_and_inputs)

thomwolf's avatar
thomwolf committed
447
448
449
    def test_for_multiple_choice(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_multiple_choice(*config_and_inputs)
thomwolf's avatar
thomwolf committed
450

thomwolf's avatar
thomwolf committed
451
452
453
    def test_for_next_sequence_prediction(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_next_sequence_prediction(*config_and_inputs)
thomwolf's avatar
thomwolf committed
454

thomwolf's avatar
thomwolf committed
455
456
457
    def test_for_pretraining(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_pretraining(*config_and_inputs)
thomwolf's avatar
thomwolf committed
458

thomwolf's avatar
thomwolf committed
459
460
461
    def test_for_question_answering(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_question_answering(*config_and_inputs)
thomwolf's avatar
thomwolf committed
462

thomwolf's avatar
thomwolf committed
463
464
465
    def test_for_sequence_classification(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_sequence_classification(*config_and_inputs)
thomwolf's avatar
thomwolf committed
466

thomwolf's avatar
thomwolf committed
467
468
469
    def test_for_token_classification(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_bert_for_token_classification(*config_and_inputs)
thomwolf's avatar
thomwolf committed
470

471
    @slow
thomwolf's avatar
thomwolf committed
472
473
    def test_model_from_pretrained(self):
        for model_name in list(BERT_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
474
            model = BertModel.from_pretrained(model_name, cache_dir=CACHE_DIR)
thomwolf's avatar
thomwolf committed
475
            self.assertIsNotNone(model)