test_modeling_albert.py 12.5 KB
Newer Older
Lysandre's avatar
Lysandre committed
1
# coding=utf-8
Sylvain Gugger's avatar
Sylvain Gugger committed
2
# Copyright 2020 The HuggingFace Team. All rights reserved.
Lysandre's avatar
Lysandre committed
3
4
5
6
7
8
9
10
11
12
13
14
#
# 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

Lysandre's avatar
Lysandre committed
16

17
18
import unittest

Lysandre's avatar
Lysandre committed
19
from transformers import is_torch_available
20
from transformers.models.auto import get_values
21
from transformers.testing_utils import require_torch, slow, torch_device
Lysandre's avatar
Lysandre committed
22

23
from .test_configuration_common import ConfigTester
24
from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
Lysandre's avatar
Lysandre committed
25

Aymeric Augustin's avatar
Aymeric Augustin committed
26

Lysandre's avatar
Lysandre committed
27
if is_torch_available():
28
29
    import torch

30
    from transformers import (
31
        MODEL_FOR_PRETRAINING_MAPPING,
32
33
        AlbertConfig,
        AlbertForMaskedLM,
34
        AlbertForMultipleChoice,
35
36
        AlbertForPreTraining,
        AlbertForQuestionAnswering,
37
        AlbertForSequenceClassification,
38
        AlbertForTokenClassification,
39
        AlbertModel,
40
    )
Sylvain Gugger's avatar
Sylvain Gugger committed
41
    from transformers.models.albert.modeling_albert import ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST
Lysandre's avatar
Lysandre committed
42
43


44
45
class AlbertModelTester:
    def __init__(
Lysandre's avatar
Lysandre committed
46
47
        self,
        parent,
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
    ):
        self.parent = parent
        self.batch_size = 13
        self.seq_length = 7
        self.is_training = True
        self.use_input_mask = True
        self.use_token_type_ids = True
        self.use_labels = True
        self.vocab_size = 99
        self.embedding_size = 16
        self.hidden_size = 36
        self.num_hidden_layers = 6
        self.num_hidden_groups = 6
        self.num_attention_heads = 6
        self.intermediate_size = 37
        self.hidden_act = "gelu"
        self.hidden_dropout_prob = 0.1
        self.attention_probs_dropout_prob = 0.1
        self.max_position_embeddings = 512
        self.type_vocab_size = 16
        self.type_sequence_label_size = 2
        self.initializer_range = 0.02
        self.num_labels = 3
        self.num_choices = 4
        self.scope = None

    def prepare_config_and_inputs(self):
        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)

        input_mask = None
        if self.use_input_mask:
79
            input_mask = random_attention_mask([self.batch_size, self.seq_length])
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109

        token_type_ids = None
        if self.use_token_type_ids:
            token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)

        sequence_labels = None
        token_labels = None
        choice_labels = None
        if self.use_labels:
            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)

        config = AlbertConfig(
            vocab_size=self.vocab_size,
            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,
            initializer_range=self.initializer_range,
            num_hidden_groups=self.num_hidden_groups,
        )

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

110
    def create_and_check_model(
111
112
113
114
115
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = AlbertModel(config=config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
116
117
118
        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
        result = model(input_ids, token_type_ids=token_type_ids)
        result = model(input_ids)
Stas Bekman's avatar
Stas Bekman committed
119
120
        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))
121

122
    def create_and_check_for_pretraining(
123
124
125
126
127
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = AlbertForPreTraining(config=config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
128
        result = model(
129
130
131
132
133
134
            input_ids,
            attention_mask=input_mask,
            token_type_ids=token_type_ids,
            labels=token_labels,
            sentence_order_label=sequence_labels,
        )
Stas Bekman's avatar
Stas Bekman committed
135
136
        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))
137

138
    def create_and_check_for_masked_lm(
139
140
141
142
143
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = AlbertForMaskedLM(config=config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
144
        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
Stas Bekman's avatar
Stas Bekman committed
145
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
146

147
    def create_and_check_for_question_answering(
148
149
150
151
152
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = AlbertForQuestionAnswering(config=config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
153
        result = model(
154
155
156
157
158
159
            input_ids,
            attention_mask=input_mask,
            token_type_ids=token_type_ids,
            start_positions=sequence_labels,
            end_positions=sequence_labels,
        )
Stas Bekman's avatar
Stas Bekman committed
160
161
        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))
162

163
    def create_and_check_for_sequence_classification(
164
165
166
167
168
169
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        config.num_labels = self.num_labels
        model = AlbertForSequenceClassification(config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
170
        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
Stas Bekman's avatar
Stas Bekman committed
171
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
172

173
    def create_and_check_for_token_classification(
174
175
176
177
178
179
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        config.num_labels = self.num_labels
        model = AlbertForTokenClassification(config=config)
        model.to(torch_device)
        model.eval()
Sylvain Gugger's avatar
Sylvain Gugger committed
180
        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
Stas Bekman's avatar
Stas Bekman committed
181
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
182

183
    def create_and_check_for_multiple_choice(
184
185
186
187
188
189
190
191
192
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        config.num_choices = self.num_choices
        model = AlbertForMultipleChoice(config=config)
        model.to(torch_device)
        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()
Sylvain Gugger's avatar
Sylvain Gugger committed
193
        result = model(
194
195
196
197
198
            multiple_choice_inputs_ids,
            attention_mask=multiple_choice_input_mask,
            token_type_ids=multiple_choice_token_type_ids,
            labels=choice_labels,
        )
Stas Bekman's avatar
Stas Bekman committed
199
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215

    def prepare_config_and_inputs_for_common(self):
        config_and_inputs = self.prepare_config_and_inputs()
        (
            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}
        return config, inputs_dict


216
@require_torch
217
class AlbertModelTest(ModelTesterMixin, unittest.TestCase):
Lysandre's avatar
Lysandre committed
218

219
220
221
222
223
224
225
226
227
228
229
230
231
    all_model_classes = (
        (
            AlbertModel,
            AlbertForPreTraining,
            AlbertForMaskedLM,
            AlbertForMultipleChoice,
            AlbertForSequenceClassification,
            AlbertForTokenClassification,
            AlbertForQuestionAnswering,
        )
        if is_torch_available()
        else ()
    )
Lysandre's avatar
Lysandre committed
232

233
234
    test_sequence_classification_problem_types = True

235
236
237
238
239
    # 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)

        if return_labels:
240
            if model_class in get_values(MODEL_FOR_PRETRAINING_MAPPING):
241
242
243
244
245
246
247
248
                inputs_dict["labels"] = torch.zeros(
                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
                )
                inputs_dict["sentence_order_label"] = torch.zeros(
                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
                )
        return inputs_dict

Lysandre's avatar
Lysandre committed
249
    def setUp(self):
250
        self.model_tester = AlbertModelTester(self)
Lysandre's avatar
Lysandre committed
251
252
253
254
255
        self.config_tester = ConfigTester(self, config_class=AlbertConfig, hidden_size=37)

    def test_config(self):
        self.config_tester.run_common_tests()

256
    def test_model(self):
Lysandre's avatar
Lysandre committed
257
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
258
        self.model_tester.create_and_check_model(*config_and_inputs)
Lysandre's avatar
Lysandre committed
259

260
261
    def test_for_pretraining(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
262
        self.model_tester.create_and_check_for_pretraining(*config_and_inputs)
263

Lysandre's avatar
Lysandre committed
264
265
    def test_for_masked_lm(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
266
        self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)
Lysandre's avatar
Lysandre committed
267

268
269
    def test_for_multiple_choice(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
270
        self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
271

272
273
    def test_for_question_answering(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
274
        self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
275
276
277

    def test_for_sequence_classification(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
278
279
280
281
282
283
284
        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)
285

286
    @slow
Lysandre's avatar
Lysandre committed
287
    def test_model_from_pretrained(self):
288
        for model_name in ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
289
            model = AlbertModel.from_pretrained(model_name)
Lysandre's avatar
Lysandre committed
290
            self.assertIsNotNone(model)
291
292
293
294
295
296


@require_torch
class AlbertModelIntegrationTest(unittest.TestCase):
    @slow
    def test_inference_no_head_absolute_embedding(self):
297
        model = AlbertModel.from_pretrained("albert-base-v2")
298
        input_ids = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
299
300
301
        attention_mask = torch.tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])
        output = model(input_ids, attention_mask=attention_mask)[0]
        expected_shape = torch.Size((1, 11, 768))
302
303
        self.assertEqual(output.shape, expected_shape)
        expected_slice = torch.tensor(
304
            [[[-0.6513, 1.5035, -0.2766], [-0.6515, 1.5046, -0.2780], [-0.6512, 1.5049, -0.2784]]]
305
306
        )

307
        self.assertTrue(torch.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4))