File size: 10,546 Bytes
a1d409e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
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
79
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
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
# coding=utf-8
# Copyright 2018 Salesforce and HuggingFace Inc. team.
# 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.


import gc
import unittest

from transformers import CTRLConfig, is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device

from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin


if is_torch_available():
    import torch

    from transformers import (
        CTRL_PRETRAINED_MODEL_ARCHIVE_LIST,
        CTRLForSequenceClassification,
        CTRLLMHeadModel,
        CTRLModel,
    )


class CTRLModelTester:
    def __init__(
        self,
        parent,
        batch_size=14,
        seq_length=7,
        is_training=True,
        use_token_type_ids=True,
        use_input_mask=True,
        use_labels=True,
        use_mc_token_ids=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,
    ):
        self.parent = parent
        self.batch_size = batch_size
        self.seq_length = seq_length
        self.is_training = is_training
        self.use_token_type_ids = use_token_type_ids
        self.use_input_mask = use_input_mask
        self.use_labels = use_labels
        self.use_mc_token_ids = use_mc_token_ids
        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
        self.pad_token_id = self.vocab_size - 1

    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:
            input_mask = random_attention_mask([self.batch_size, self.seq_length])

        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)

        mc_token_ids = None
        if self.use_mc_token_ids:
            mc_token_ids = ids_tensor([self.batch_size, self.num_choices], self.seq_length)

        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 = self.get_config()

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

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

    def get_config(self):
        return CTRLConfig(
            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,
            # type_vocab_size=self.type_vocab_size,
            # initializer_range=self.initializer_range,
            pad_token_id=self.pad_token_id,
        )

    def create_and_check_ctrl_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
        model = CTRLModel(config=config)
        model.to(torch_device)
        model.eval()

        model(input_ids, token_type_ids=token_type_ids, head_mask=head_mask)
        model(input_ids, token_type_ids=token_type_ids)
        result = model(input_ids)
        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
        self.parent.assertEqual(len(result.past_key_values), config.n_layer)

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

        result = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
        self.parent.assertEqual(result.loss.shape, ())
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))

    def prepare_config_and_inputs_for_common(self):
        config_and_inputs = self.prepare_config_and_inputs()

        (
            config,
            input_ids,
            input_mask,
            head_mask,
            token_type_ids,
            mc_token_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

    def create_and_check_ctrl_for_sequence_classification(self, config, input_ids, head_mask, token_type_ids, *args):
        config.num_labels = self.num_labels
        model = CTRLForSequenceClassification(config)
        model.to(torch_device)
        model.eval()
        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))


@require_torch
class CTRLModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
    all_model_classes = (CTRLModel, CTRLLMHeadModel, CTRLForSequenceClassification) if is_torch_available() else ()
    all_generative_model_classes = (CTRLLMHeadModel,) if is_torch_available() else ()
    pipeline_model_mapping = (
        {
            "feature-extraction": CTRLModel,
            "text-classification": CTRLForSequenceClassification,
            "text-generation": CTRLLMHeadModel,
            "zero-shot": CTRLForSequenceClassification,
        }
        if is_torch_available()
        else {}
    )
    test_pruning = True
    test_resize_embeddings = False
    test_head_masking = False

    # TODO: Fix the failed tests
    def is_pipeline_test_to_skip(
        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
    ):
        if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
            # Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
            # `CTRLConfig` was never used in pipeline tests, either because of a missing checkpoint or because a tiny
            # config could not be created.
            return True

        return False

    def setUp(self):
        self.model_tester = CTRLModelTester(self)
        self.config_tester = ConfigTester(self, config_class=CTRLConfig, n_embd=37)

    def tearDown(self):
        super().tearDown()
        # clean-up as much as possible GPU memory occupied by PyTorch
        gc.collect()
        torch.cuda.empty_cache()

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

    def test_ctrl_model(self):
        config_and_inputs = self.model_tester.prepare_config_and_inputs()
        self.model_tester.create_and_check_ctrl_model(*config_and_inputs)

    def test_ctrl_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)

    @slow
    def test_model_from_pretrained(self):
        for model_name in CTRL_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
            model = CTRLModel.from_pretrained(model_name)
            self.assertIsNotNone(model)


@require_torch
class CTRLModelLanguageGenerationTest(unittest.TestCase):
    def tearDown(self):
        super().tearDown()
        # clean-up as much as possible GPU memory occupied by PyTorch
        gc.collect()
        torch.cuda.empty_cache()

    @slow
    def test_lm_generate_ctrl(self):
        model = CTRLLMHeadModel.from_pretrained("ctrl")
        model.to(torch_device)
        input_ids = torch.tensor(
            [[11859, 0, 1611, 8]], dtype=torch.long, device=torch_device
        )  # Legal the president is
        expected_output_ids = [
            11859,
            0,
            1611,
            8,
            5,
            150,
            26449,
            2,
            19,
            348,
            469,
            3,
            2595,
            48,
            20740,
            246533,
            246533,
            19,
            30,
            5,
        ]  # Legal the president is a good guy and I don't want to lose my job. \n \n I have a

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