forked from huggingface/transformers
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_data_collator.py
292 lines (237 loc) · 13.6 KB
/
test_data_collator.py
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
# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# 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 os
import shutil
import tempfile
import unittest
from transformers import BertTokenizer, is_torch_available, set_seed
from transformers.testing_utils import require_torch
if is_torch_available():
import torch
from transformers import (
DataCollatorForLanguageModeling,
DataCollatorForPermutationLanguageModeling,
DataCollatorForTokenClassification,
DataCollatorWithPadding,
default_data_collator,
)
@require_torch
class DataCollatorIntegrationTest(unittest.TestCase):
def setUp(self):
self.tmpdirname = tempfile.mkdtemp()
vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"]
self.vocab_file = os.path.join(self.tmpdirname, "vocab.txt")
with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
def tearDown(self):
shutil.rmtree(self.tmpdirname)
def test_default_with_dict(self):
features = [{"label": i, "inputs": [0, 1, 2, 3, 4, 5]} for i in range(8)]
batch = default_data_collator(features)
self.assertTrue(batch["labels"].equal(torch.tensor(list(range(8)))))
self.assertEqual(batch["labels"].dtype, torch.long)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 6]))
# With label_ids
features = [{"label_ids": [0, 1, 2], "inputs": [0, 1, 2, 3, 4, 5]} for i in range(8)]
batch = default_data_collator(features)
self.assertTrue(batch["labels"].equal(torch.tensor([[0, 1, 2]] * 8)))
self.assertEqual(batch["labels"].dtype, torch.long)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 6]))
# Features can already be tensors
features = [{"label": i, "inputs": torch.randint(10, [10])} for i in range(8)]
batch = default_data_collator(features)
self.assertTrue(batch["labels"].equal(torch.tensor(list(range(8)))))
self.assertEqual(batch["labels"].dtype, torch.long)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 10]))
# Labels can already be tensors
features = [{"label": torch.tensor(i), "inputs": torch.randint(10, [10])} for i in range(8)]
batch = default_data_collator(features)
self.assertEqual(batch["labels"].dtype, torch.long)
self.assertTrue(batch["labels"].equal(torch.tensor(list(range(8)))))
self.assertEqual(batch["labels"].dtype, torch.long)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 10]))
def test_default_classification_and_regression(self):
data_collator = default_data_collator
features = [{"input_ids": [0, 1, 2, 3, 4], "label": i} for i in range(4)]
batch = data_collator(features)
self.assertEqual(batch["labels"].dtype, torch.long)
features = [{"input_ids": [0, 1, 2, 3, 4], "label": float(i)} for i in range(4)]
batch = data_collator(features)
self.assertEqual(batch["labels"].dtype, torch.float)
def test_default_with_no_labels(self):
features = [{"label": None, "inputs": [0, 1, 2, 3, 4, 5]} for i in range(8)]
batch = default_data_collator(features)
self.assertTrue("labels" not in batch)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 6]))
# With label_ids
features = [{"label_ids": None, "inputs": [0, 1, 2, 3, 4, 5]} for i in range(8)]
batch = default_data_collator(features)
self.assertTrue("labels" not in batch)
self.assertEqual(batch["inputs"].shape, torch.Size([8, 6]))
def test_data_collator_with_padding(self):
tokenizer = BertTokenizer(self.vocab_file)
features = [{"input_ids": [0, 1, 2]}, {"input_ids": [0, 1, 2, 3, 4, 5]}]
data_collator = DataCollatorWithPadding(tokenizer)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 6]))
self.assertEqual(batch["input_ids"][0].tolist(), [0, 1, 2] + [tokenizer.pad_token_id] * 3)
data_collator = DataCollatorWithPadding(tokenizer, padding="max_length", max_length=10)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 10]))
data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 8]))
def test_data_collator_for_token_classification(self):
tokenizer = BertTokenizer(self.vocab_file)
features = [
{"input_ids": [0, 1, 2], "labels": [0, 1, 2]},
{"input_ids": [0, 1, 2, 3, 4, 5], "labels": [0, 1, 2, 3, 4, 5]},
]
data_collator = DataCollatorForTokenClassification(tokenizer)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 6]))
self.assertEqual(batch["input_ids"][0].tolist(), [0, 1, 2] + [tokenizer.pad_token_id] * 3)
self.assertEqual(batch["labels"].shape, torch.Size([2, 6]))
self.assertEqual(batch["labels"][0].tolist(), [0, 1, 2] + [-100] * 3)
data_collator = DataCollatorForTokenClassification(tokenizer, padding="max_length", max_length=10)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 10]))
self.assertEqual(batch["labels"].shape, torch.Size([2, 10]))
data_collator = DataCollatorForTokenClassification(tokenizer, pad_to_multiple_of=8)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 8]))
self.assertEqual(batch["labels"].shape, torch.Size([2, 8]))
data_collator = DataCollatorForTokenClassification(tokenizer, label_pad_token_id=-1)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size([2, 6]))
self.assertEqual(batch["input_ids"][0].tolist(), [0, 1, 2] + [tokenizer.pad_token_id] * 3)
self.assertEqual(batch["labels"].shape, torch.Size([2, 6]))
self.assertEqual(batch["labels"][0].tolist(), [0, 1, 2] + [-1] * 3)
def _test_no_pad_and_pad(self, no_pad_features, pad_features):
tokenizer = BertTokenizer(self.vocab_file)
data_collator = DataCollatorForLanguageModeling(tokenizer, mlm=False)
batch = data_collator(no_pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
batch = data_collator(pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
data_collator = DataCollatorForLanguageModeling(tokenizer, mlm=False, pad_to_multiple_of=8)
batch = data_collator(no_pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 16)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 16)))
batch = data_collator(pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 16)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 16)))
tokenizer._pad_token = None
data_collator = DataCollatorForLanguageModeling(tokenizer, mlm=False)
with self.assertRaises(ValueError):
# Expect error due to padding token missing
data_collator(pad_features)
set_seed(42) # For reproducibility
tokenizer = BertTokenizer(self.vocab_file)
data_collator = DataCollatorForLanguageModeling(tokenizer)
batch = data_collator(no_pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
masked_tokens = batch["input_ids"] == tokenizer.mask_token_id
self.assertTrue(torch.any(masked_tokens))
self.assertTrue(all(x == -100 for x in batch["labels"][~masked_tokens].tolist()))
batch = data_collator(pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
masked_tokens = batch["input_ids"] == tokenizer.mask_token_id
self.assertTrue(torch.any(masked_tokens))
self.assertTrue(all(x == -100 for x in batch["labels"][~masked_tokens].tolist()))
data_collator = DataCollatorForLanguageModeling(tokenizer, pad_to_multiple_of=8)
batch = data_collator(no_pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 16)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 16)))
masked_tokens = batch["input_ids"] == tokenizer.mask_token_id
self.assertTrue(torch.any(masked_tokens))
self.assertTrue(all(x == -100 for x in batch["labels"][~masked_tokens].tolist()))
batch = data_collator(pad_features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 16)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 16)))
masked_tokens = batch["input_ids"] == tokenizer.mask_token_id
self.assertTrue(torch.any(masked_tokens))
self.assertTrue(all(x == -100 for x in batch["labels"][~masked_tokens].tolist()))
def test_data_collator_for_language_modeling(self):
no_pad_features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]
pad_features = [{"input_ids": list(range(5))}, {"input_ids": list(range(10))}]
self._test_no_pad_and_pad(no_pad_features, pad_features)
no_pad_features = [list(range(10)), list(range(10))]
pad_features = [list(range(5)), list(range(10))]
self._test_no_pad_and_pad(no_pad_features, pad_features)
def test_plm(self):
tokenizer = BertTokenizer(self.vocab_file)
no_pad_features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]
pad_features = [{"input_ids": list(range(5))}, {"input_ids": list(range(10))}]
data_collator = DataCollatorForPermutationLanguageModeling(tokenizer)
batch = data_collator(pad_features)
self.assertIsInstance(batch, dict)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["perm_mask"].shape, torch.Size((2, 10, 10)))
self.assertEqual(batch["target_mapping"].shape, torch.Size((2, 10, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
batch = data_collator(no_pad_features)
self.assertIsInstance(batch, dict)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["perm_mask"].shape, torch.Size((2, 10, 10)))
self.assertEqual(batch["target_mapping"].shape, torch.Size((2, 10, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))
example = [torch.randint(5, [5])]
with self.assertRaises(ValueError):
# Expect error due to odd sequence length
data_collator(example)
def test_nsp(self):
tokenizer = BertTokenizer(self.vocab_file)
features = [
{"input_ids": [0, 1, 2, 3, 4], "token_type_ids": [0, 1, 2, 3, 4], "next_sentence_label": i}
for i in range(2)
]
data_collator = DataCollatorForLanguageModeling(tokenizer)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 5)))
self.assertEqual(batch["token_type_ids"].shape, torch.Size((2, 5)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 5)))
self.assertEqual(batch["next_sentence_label"].shape, torch.Size((2,)))
data_collator = DataCollatorForLanguageModeling(tokenizer, pad_to_multiple_of=8)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 8)))
self.assertEqual(batch["token_type_ids"].shape, torch.Size((2, 8)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 8)))
self.assertEqual(batch["next_sentence_label"].shape, torch.Size((2,)))
def test_sop(self):
tokenizer = BertTokenizer(self.vocab_file)
features = [
{
"input_ids": torch.tensor([0, 1, 2, 3, 4]),
"token_type_ids": torch.tensor([0, 1, 2, 3, 4]),
"sentence_order_label": i,
}
for i in range(2)
]
data_collator = DataCollatorForLanguageModeling(tokenizer)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 5)))
self.assertEqual(batch["token_type_ids"].shape, torch.Size((2, 5)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 5)))
self.assertEqual(batch["sentence_order_label"].shape, torch.Size((2,)))
data_collator = DataCollatorForLanguageModeling(tokenizer, pad_to_multiple_of=8)
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 8)))
self.assertEqual(batch["token_type_ids"].shape, torch.Size((2, 8)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 8)))
self.assertEqual(batch["sentence_order_label"].shape, torch.Size((2,)))