test_modeling_openai.py 8.8 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

thomwolf's avatar
thomwolf committed
16

17
18
import unittest

19
from transformers import is_torch_available
20
from transformers.testing_utils import require_torch, slow, torch_device
thomwolf's avatar
thomwolf committed
21

22
from .test_configuration_common import ConfigTester
23
from .test_generation_utils import GenerationTesterMixin
24
from .test_modeling_common import ModelTesterMixin, ids_tensor
Aymeric Augustin's avatar
Aymeric Augustin committed
25
26


27
if is_torch_available():
28
    import torch
29

30
    from transformers import (
31
        OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
32
        OpenAIGPTConfig,
33
        OpenAIGPTDoubleHeadsModel,
34
        OpenAIGPTForSequenceClassification,
35
36
        OpenAIGPTLMHeadModel,
        OpenAIGPTModel,
37
38
    )

39

40
41
class OpenAIGPTModelTester:
    def __init__(
Lysandre's avatar
Lysandre committed
42
43
        self,
        parent,
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
    ):
        self.parent = parent
        self.batch_size = 13
        self.seq_length = 7
        self.is_training = True
        self.use_token_type_ids = True
        self.use_labels = True
        self.vocab_size = 99
        self.hidden_size = 32
        self.num_hidden_layers = 5
        self.num_attention_heads = 4
        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
66
        self.pad_token_id = self.vocab_size - 1
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92

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

        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 = OpenAIGPTConfig(
            vocab_size=self.vocab_size,
            n_embd=self.hidden_size,
            n_layer=self.num_hidden_layers,
            n_head=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,
            n_positions=self.max_position_embeddings,
Sylvain Gugger's avatar
Sylvain Gugger committed
93
            n_ctx=self.max_position_embeddings,
94
95
            # type_vocab_size=self.type_vocab_size,
            # initializer_range=self.initializer_range
96
            pad_token_id=self.pad_token_id,
Sylvain Gugger's avatar
Sylvain Gugger committed
97
            return_dict=True,
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
        )

        head_mask = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2)

        return (
            config,
            input_ids,
            head_mask,
            token_type_ids,
            sequence_labels,
            token_labels,
            choice_labels,
        )

    def create_and_check_openai_gpt_model(self, config, input_ids, head_mask, token_type_ids, *args):
        model = OpenAIGPTModel(config=config)
        model.to(torch_device)
        model.eval()

Sylvain Gugger's avatar
Sylvain Gugger committed
117
118
119
        result = model(input_ids, token_type_ids=token_type_ids, head_mask=head_mask)
        result = model(input_ids, token_type_ids=token_type_ids)
        result = model(input_ids)
120

Stas Bekman's avatar
Stas Bekman committed
121
        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
122
123
124
125
126
127

    def create_and_check_lm_head_model(self, config, input_ids, head_mask, token_type_ids, *args):
        model = OpenAIGPTLMHeadModel(config)
        model.to(torch_device)
        model.eval()

Sylvain Gugger's avatar
Sylvain Gugger committed
128
        result = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
Stas Bekman's avatar
Stas Bekman committed
129
130
        self.parent.assertEqual(result.loss.shape, ())
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
131
132
133
134
135
136

    def create_and_check_double_lm_head_model(self, config, input_ids, head_mask, token_type_ids, *args):
        model = OpenAIGPTDoubleHeadsModel(config)
        model.to(torch_device)
        model.eval()

Sylvain Gugger's avatar
Sylvain Gugger committed
137
        result = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
138
139
        self.parent.assertEqual(result.loss.shape, ())
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
140

141
142
143
144
145
146
147
148
149
150
151
152
    def create_and_check_openai_gpt_for_sequence_classification(
        self, config, input_ids, head_mask, token_type_ids, *args
    ):
        config.num_labels = self.num_labels
        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))

153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
    def prepare_config_and_inputs_for_common(self):
        config_and_inputs = self.prepare_config_and_inputs()
        (
            config,
            input_ids,
            head_mask,
            token_type_ids,
            sequence_labels,
            token_labels,
            choice_labels,
        ) = config_and_inputs
        inputs_dict = {
            "input_ids": input_ids,
            "token_type_ids": token_type_ids,
            "head_mask": head_mask,
        }

        return config, inputs_dict


173
@require_torch
174
class OpenAIGPTModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
175

176
    all_model_classes = (
177
178
179
        (OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
        if is_torch_available()
        else ()
180
    )
181
182
183
    all_generative_model_classes = (
        (OpenAIGPTLMHeadModel,) if is_torch_available() else ()
    )  # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
184
185

    def setUp(self):
186
        self.model_tester = OpenAIGPTModelTester(self)
187
        self.config_tester = ConfigTester(self, config_class=OpenAIGPTConfig, n_embd=37)
thomwolf's avatar
thomwolf committed
188
189

    def test_config(self):
190
        self.config_tester.run_common_tests()
thomwolf's avatar
thomwolf committed
191

192
193
194
195
196
197
198
199
200
201
202
    def test_openai_gpt_model(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_openai_gpt_model(*config_and_inputs)

    def test_openai_gpt_lm_head_model(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_lm_head_model(*config_and_inputs)

    def test_openai_gpt_double_lm_head_model(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_double_lm_head_model(*config_and_inputs)
thomwolf's avatar
thomwolf committed
203

204
205
206
207
    def test_openai_gpt_classification_model(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*config_and_inputs)

208
    @slow
209
    def test_model_from_pretrained(self):
210
        for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
211
            model = OpenAIGPTModel.from_pretrained(model_name)
212
            self.assertIsNotNone(model)
213
214


215
@require_torch
216
217
218
219
class OPENAIGPTModelLanguageGenerationTest(unittest.TestCase):
    @slow
    def test_lm_generate_openai_gpt(self):
        model = OpenAIGPTLMHeadModel.from_pretrained("openai-gpt")
220
        model.to(torch_device)
patrickvonplaten's avatar
patrickvonplaten committed
221
        input_ids = torch.tensor([[481, 4735, 544]], dtype=torch.long, device=torch_device)  # the president is
222
223
        expected_output_ids = [
            481,
patrickvonplaten's avatar
patrickvonplaten committed
224
            4735,
225
            544,
patrickvonplaten's avatar
patrickvonplaten committed
226
227
228
229
230
231
232
233
234
235
236
            246,
            963,
            870,
            762,
            239,
            244,
            40477,
            244,
            249,
            719,
            881,
237
            487,
patrickvonplaten's avatar
patrickvonplaten committed
238
            544,
239
            240,
patrickvonplaten's avatar
patrickvonplaten committed
240
241
242
243
244
245
            244,
            603,
            481,
        ]  # the president is a very good man. " \n " i\'m sure he is, " said the

        output_ids = model.generate(input_ids, do_sample=False)
246
        self.assertListEqual(output_ids[0].tolist(), expected_output_ids)