File size: 12,909 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
# coding=utf-8
# Copyright 2018 HuggingFace Inc..
#
# 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.
"""
isort:skip_file
"""
import os
import pickle
import tempfile
import unittest
from typing import Callable, Optional

import numpy as np

from transformers import (
    BatchEncoding,
    BertTokenizer,
    BertTokenizerFast,
    PreTrainedTokenizer,
    PreTrainedTokenizerFast,
    TensorType,
    TokenSpan,
    is_tokenizers_available,
)
from transformers.models.gpt2.tokenization_gpt2 import GPT2Tokenizer
from transformers.testing_utils import CaptureStderr, require_flax, require_tf, require_tokenizers, require_torch, slow


if is_tokenizers_available():
    from tokenizers import Tokenizer
    from tokenizers.models import WordPiece


class TokenizerUtilsTest(unittest.TestCase):
    def check_tokenizer_from_pretrained(self, tokenizer_class):
        s3_models = list(tokenizer_class.max_model_input_sizes.keys())
        for model_name in s3_models[:1]:
            tokenizer = tokenizer_class.from_pretrained(model_name)
            self.assertIsNotNone(tokenizer)
            self.assertIsInstance(tokenizer, tokenizer_class)
            self.assertIsInstance(tokenizer, PreTrainedTokenizer)

            for special_tok in tokenizer.all_special_tokens:
                self.assertIsInstance(special_tok, str)
                special_tok_id = tokenizer.convert_tokens_to_ids(special_tok)
                self.assertIsInstance(special_tok_id, int)

    def assert_dump_and_restore(self, be_original: BatchEncoding, equal_op: Optional[Callable] = None):
        batch_encoding_str = pickle.dumps(be_original)
        self.assertIsNotNone(batch_encoding_str)

        be_restored = pickle.loads(batch_encoding_str)

        # Ensure is_fast is correctly restored
        self.assertEqual(be_restored.is_fast, be_original.is_fast)

        # Ensure encodings are potentially correctly restored
        if be_original.is_fast:
            self.assertIsNotNone(be_restored.encodings)
        else:
            self.assertIsNone(be_restored.encodings)

        # Ensure the keys are the same
        for original_v, restored_v in zip(be_original.values(), be_restored.values()):
            if equal_op:
                self.assertTrue(equal_op(restored_v, original_v))
            else:
                self.assertEqual(restored_v, original_v)

    @slow
    def test_pretrained_tokenizers(self):
        self.check_tokenizer_from_pretrained(GPT2Tokenizer)

    def test_tensor_type_from_str(self):
        self.assertEqual(TensorType("tf"), TensorType.TENSORFLOW)
        self.assertEqual(TensorType("pt"), TensorType.PYTORCH)
        self.assertEqual(TensorType("np"), TensorType.NUMPY)

    @require_tokenizers
    def test_batch_encoding_pickle(self):
        import numpy as np

        tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
        tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")

        # Python no tensor
        with self.subTest("BatchEncoding (Python, return_tensors=None)"):
            self.assert_dump_and_restore(tokenizer_p("Small example to encode"))

        with self.subTest("BatchEncoding (Python, return_tensors=NUMPY)"):
            self.assert_dump_and_restore(
                tokenizer_p("Small example to encode", return_tensors=TensorType.NUMPY), np.array_equal
            )

        with self.subTest("BatchEncoding (Rust, return_tensors=None)"):
            self.assert_dump_and_restore(tokenizer_r("Small example to encode"))

        with self.subTest("BatchEncoding (Rust, return_tensors=NUMPY)"):
            self.assert_dump_and_restore(
                tokenizer_r("Small example to encode", return_tensors=TensorType.NUMPY), np.array_equal
            )

    @require_tf
    @require_tokenizers
    def test_batch_encoding_pickle_tf(self):
        import tensorflow as tf

        def tf_array_equals(t1, t2):
            return tf.reduce_all(tf.equal(t1, t2))

        tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
        tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")

        with self.subTest("BatchEncoding (Python, return_tensors=TENSORFLOW)"):
            self.assert_dump_and_restore(
                tokenizer_p("Small example to encode", return_tensors=TensorType.TENSORFLOW), tf_array_equals
            )

        with self.subTest("BatchEncoding (Rust, return_tensors=TENSORFLOW)"):
            self.assert_dump_and_restore(
                tokenizer_r("Small example to encode", return_tensors=TensorType.TENSORFLOW), tf_array_equals
            )

    @require_torch
    @require_tokenizers
    def test_batch_encoding_pickle_pt(self):
        import torch

        tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
        tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")

        with self.subTest("BatchEncoding (Python, return_tensors=PYTORCH)"):
            self.assert_dump_and_restore(
                tokenizer_p("Small example to encode", return_tensors=TensorType.PYTORCH), torch.equal
            )

        with self.subTest("BatchEncoding (Rust, return_tensors=PYTORCH)"):
            self.assert_dump_and_restore(
                tokenizer_r("Small example to encode", return_tensors=TensorType.PYTORCH), torch.equal
            )

    @require_tokenizers
    def test_batch_encoding_is_fast(self):
        tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
        tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")

        with self.subTest("Python Tokenizer"):
            self.assertFalse(tokenizer_p("Small example to_encode").is_fast)

        with self.subTest("Rust Tokenizer"):
            self.assertTrue(tokenizer_r("Small example to_encode").is_fast)

    @require_tokenizers
    def test_batch_encoding_word_to_tokens(self):
        tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
        encoded = tokenizer_r(["Test", "\xad", "test"], is_split_into_words=True)

        self.assertEqual(encoded.word_to_tokens(0), TokenSpan(start=1, end=2))
        self.assertEqual(encoded.word_to_tokens(1), None)
        self.assertEqual(encoded.word_to_tokens(2), TokenSpan(start=2, end=3))

    def test_batch_encoding_with_labels(self):
        batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
        tensor_batch = batch.convert_to_tensors(tensor_type="np")
        self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
        self.assertEqual(tensor_batch["labels"].shape, (2,))
        # test converting the converted
        with CaptureStderr() as cs:
            tensor_batch = batch.convert_to_tensors(tensor_type="np")
        self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")

        batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
        tensor_batch = batch.convert_to_tensors(tensor_type="np", prepend_batch_axis=True)
        self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
        self.assertEqual(tensor_batch["labels"].shape, (1,))

    @require_torch
    def test_batch_encoding_with_labels_pt(self):
        batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
        tensor_batch = batch.convert_to_tensors(tensor_type="pt")
        self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
        self.assertEqual(tensor_batch["labels"].shape, (2,))
        # test converting the converted
        with CaptureStderr() as cs:
            tensor_batch = batch.convert_to_tensors(tensor_type="pt")
        self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")

        batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
        tensor_batch = batch.convert_to_tensors(tensor_type="pt", prepend_batch_axis=True)
        self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
        self.assertEqual(tensor_batch["labels"].shape, (1,))

    @require_tf
    def test_batch_encoding_with_labels_tf(self):
        batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
        tensor_batch = batch.convert_to_tensors(tensor_type="tf")
        self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
        self.assertEqual(tensor_batch["labels"].shape, (2,))
        # test converting the converted
        with CaptureStderr() as cs:
            tensor_batch = batch.convert_to_tensors(tensor_type="tf")
        self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")

        batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
        tensor_batch = batch.convert_to_tensors(tensor_type="tf", prepend_batch_axis=True)
        self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
        self.assertEqual(tensor_batch["labels"].shape, (1,))

    @require_flax
    def test_batch_encoding_with_labels_jax(self):
        batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
        tensor_batch = batch.convert_to_tensors(tensor_type="jax")
        self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
        self.assertEqual(tensor_batch["labels"].shape, (2,))
        # test converting the converted
        with CaptureStderr() as cs:
            tensor_batch = batch.convert_to_tensors(tensor_type="jax")
        self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")

        batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
        tensor_batch = batch.convert_to_tensors(tensor_type="jax", prepend_batch_axis=True)
        self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
        self.assertEqual(tensor_batch["labels"].shape, (1,))

    def test_padding_accepts_tensors(self):
        features = [{"input_ids": np.array([0, 1, 2])}, {"input_ids": np.array([0, 1, 2, 3])}]
        tokenizer = BertTokenizer.from_pretrained("bert-base-cased")

        batch = tokenizer.pad(features, padding=True)
        self.assertTrue(isinstance(batch["input_ids"], np.ndarray))
        self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
        batch = tokenizer.pad(features, padding=True, return_tensors="np")
        self.assertTrue(isinstance(batch["input_ids"], np.ndarray))
        self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])

    @require_torch
    def test_padding_accepts_tensors_pt(self):
        import torch

        features = [{"input_ids": torch.tensor([0, 1, 2])}, {"input_ids": torch.tensor([0, 1, 2, 3])}]
        tokenizer = BertTokenizer.from_pretrained("bert-base-cased")

        batch = tokenizer.pad(features, padding=True)
        self.assertTrue(isinstance(batch["input_ids"], torch.Tensor))
        self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
        batch = tokenizer.pad(features, padding=True, return_tensors="pt")
        self.assertTrue(isinstance(batch["input_ids"], torch.Tensor))
        self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])

    @require_tf
    def test_padding_accepts_tensors_tf(self):
        import tensorflow as tf

        features = [{"input_ids": tf.constant([0, 1, 2])}, {"input_ids": tf.constant([0, 1, 2, 3])}]
        tokenizer = BertTokenizer.from_pretrained("bert-base-cased")

        batch = tokenizer.pad(features, padding=True)
        self.assertTrue(isinstance(batch["input_ids"], tf.Tensor))
        self.assertEqual(batch["input_ids"].numpy().tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
        batch = tokenizer.pad(features, padding=True, return_tensors="tf")
        self.assertTrue(isinstance(batch["input_ids"], tf.Tensor))
        self.assertEqual(batch["input_ids"].numpy().tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])

    @require_tokenizers
    def test_instantiation_from_tokenizers(self):
        bert_tokenizer = Tokenizer(WordPiece(unk_token="[UNK]"))
        PreTrainedTokenizerFast(tokenizer_object=bert_tokenizer)

    @require_tokenizers
    def test_instantiation_from_tokenizers_json_file(self):
        bert_tokenizer = Tokenizer(WordPiece(unk_token="[UNK]"))
        with tempfile.TemporaryDirectory() as tmpdirname:
            bert_tokenizer.save(os.path.join(tmpdirname, "tokenizer.json"))
            PreTrainedTokenizerFast(tokenizer_file=os.path.join(tmpdirname, "tokenizer.json"))