Repository navigation
Expand file tree
/
Copy pathtest_tokenizer_common.py
More file actions
2236 lines (1866 loc) Β· 109 KB
/
Copy pathtest_tokenizer_common.py
File metadata and controls
2236 lines (1866 loc) Β· 109 KB
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
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
# Copyright 2019 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.
import inspect
import itertools
import json
import os
import pickle
import re
import shutil
import sys
import tempfile
import unittest
from itertools import takewhile
from pathlib import Path
from typing import Any, Dict, List, Tuple
from paddlenlp.transformers import PretrainedFastTokenizer, PretrainedTokenizer
from paddlenlp.transformers.tokenizer_utils import AddedToken, Trie
from paddlenlp.transformers.tokenizer_utils_base import PretrainedTokenizerBase
from ..testing_utils import get_tests_dir
sys.path.append(str(Path(__file__).parent.parent / "utils"))
NON_ENGLISH_TAGS = ["chinese", "dutch", "french", "finnish", "german", "multilingual"]
SMALL_TRAINING_CORPUS = [
["This is the first sentence.", "This is the second one."],
["This sentence (contains #) over symbols and numbers 12 3.", "But not this one."],
]
def filter_non_english(_, pretrained_name: str):
"""Filter all the model for non-english language"""
return not any([lang in pretrained_name for lang in NON_ENGLISH_TAGS])
def filter_roberta_detectors(_, pretrained_name: str):
return "detector" not in pretrained_name
class TokenizerTesterMixin:
tokenizer_class = None
fast_tokenizer_class = None
test_fast_tokenizer = False
space_between_special_tokens = False
from_pretrained_kwargs = None
from_pretrained_filter = None
from_pretrained_vocab_key = "vocab_file"
# set to True to test a sentencepiece tokenizer
test_sentencepiece = False
# set to True to ignore casing when testing a sentencepiece tokenizer
# test_sentencepiece must also be set to True
test_sentencepiece_ignore_case = False
test_offsets = True
only_english_character: bool = True
def setUp(self) -> None:
tokenizers_list = [
(
self.tokenizer_class,
pretrained_name,
self.from_pretrained_kwargs if self.from_pretrained_kwargs is not None else {},
)
for pretrained_name in self.tokenizer_class.pretrained_resource_files_map[
self.from_pretrained_vocab_key
].keys()
if self.from_pretrained_filter is None
or (self.from_pretrained_filter is not None and self.from_pretrained_filter(pretrained_name))
]
self.tokenizers_list = tokenizers_list[:1]
with open(f"{get_tests_dir()}/sample_text.txt", encoding="utf-8") as f_data:
self._data = f_data.read().replace("\n\n", "\n").strip()
self.tmpdirname = tempfile.mkdtemp()
def tearDown(self):
shutil.rmtree(self.tmpdirname)
def get_input_output_texts(self, tokenizer):
input_txt = self.get_clean_sequence(tokenizer)[0]
return input_txt, input_txt
def get_clean_sequence(self, tokenizer, with_prefix_space=False, max_length=20, min_length=5) -> Tuple[str, list]:
toks = [(i, tokenizer.decode([i], clean_up_tokenization_spaces=False)) for i in range(len(tokenizer))]
# filter the english only character
if self.only_english_character:
toks = list(filter(lambda t: re.match(r"^[ a-zA-Z]+$", t[1]), toks))
toks = list(
filter(
lambda t: [t[0]]
== tokenizer.encode(t[1], return_token_type_ids=None, add_special_tokens=False)["input_ids"],
toks,
)
)
if max_length is not None and len(toks) > max_length:
toks = toks[:max_length]
if min_length is not None and len(toks) < min_length and len(toks) > 0:
while len(toks) < min_length:
toks = toks + toks
# toks_str = [t[1] for t in toks]
toks_ids = [t[0] for t in toks]
# Ensure consistency
output_txt = tokenizer.decode(toks_ids, clean_up_tokenization_spaces=False)
if " " not in output_txt and len(toks_ids) > 1:
output_txt = (
tokenizer.decode([toks_ids[0]], clean_up_tokenization_spaces=False)
+ " "
+ tokenizer.decode(toks_ids[1:], clean_up_tokenization_spaces=False)
)
if with_prefix_space:
output_txt = " " + output_txt
output_ids = tokenizer.encode(output_txt, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
return output_txt, output_ids
def get_tokenizers(self, **kwargs) -> List[PretrainedTokenizerBase]:
return [self.get_tokenizer(**kwargs)]
def get_tokenizer(self, **kwargs) -> PretrainedTokenizer:
return self.tokenizer_class.from_pretrained(self.tmpdirname, **kwargs)
def get_fast_tokenizer(self, **kwargs) -> PretrainedFastTokenizer:
return self.fast_tokenizer_class.from_pretrained(self.tmpdirname, **kwargs)
def tokenizer_integration_test_util(
self,
expected_encoding: Dict,
model_name: str,
sequences: List[str] = None,
decode_kwargs: Dict[str, Any] = None,
padding: bool = True,
):
"""
Util for integration test.
Text is tokenized and then reverted back to text. Both results are then checked.
Args:
expected_encoding:
The expected result of the tokenizer output.
model_name:
The model name of the tokenizer to load and use.
sequences:
Can overwrite the texts that are used to check the tokenizer.
This is useful if the tokenizer supports non english languages
like france.
decode_kwargs:
Additional args for the ``decode`` function which reverts the
tokenized text back to a string.
padding:
Activates and controls padding of the tokenizer.
"""
decode_kwargs = {} if decode_kwargs is None else decode_kwargs
if sequences is None:
sequences = [
"Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides "
"general-purpose architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet...) for Natural "
"Language Understanding (NLU) and Natural Language Generation (NLG) with over 32+ pretrained "
"models in 100+ languages and deep interoperability between Jax, PyTorch and TensorFlow.",
"BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly "
"conditioning on both left and right context in all layers.",
"The quick brown fox jumps over the lazy dog.",
]
if self.test_sentencepiece_ignore_case:
sequences = [sequence.lower() for sequence in sequences]
tokenizer_classes = [self.tokenizer_class]
for tokenizer_class in tokenizer_classes:
tokenizer = tokenizer_class.from_pretrained(model_name)
encoding = tokenizer(sequences, padding=padding, return_attention_mask=True)
decoded_sequences = [
tokenizer.decode(seq, skip_special_tokens=True, **decode_kwargs) for seq in encoding["input_ids"]
]
encoding_data = encoding.data
self.assertDictEqual(encoding_data, expected_encoding)
for expected, decoded in zip(sequences, decoded_sequences):
if self.test_sentencepiece_ignore_case:
expected = expected.lower()
self.assertEqual(expected, decoded)
def assert_padded_input_match(self, input_r: list, input_p: list, max_length: int, pad_token_id: int):
# Ensure we match max_length
self.assertEqual(len(input_r), max_length)
self.assertEqual(len(input_p), max_length)
# Ensure the number of padded tokens is the same
padded_tokens_r = list(takewhile(lambda i: i == pad_token_id, reversed(input_r)))
padded_tokens_p = list(takewhile(lambda i: i == pad_token_id, reversed(input_p)))
self.assertSequenceEqual(padded_tokens_r, padded_tokens_p)
def assert_batch_padded_input_match(
self,
input_r: dict,
input_p: dict,
max_length: int,
pad_token_id: int,
model_main_input_name: str = "input_ids",
):
for i_r in input_r.values():
self.assertEqual(len(i_r), 2), self.assertEqual(len(i_r[0]), max_length), self.assertEqual(
len(i_r[1]), max_length
)
self.assertEqual(len(i_r), 2), self.assertEqual(len(i_r[0]), max_length), self.assertEqual(
len(i_r[1]), max_length
)
for i_r, i_p in zip(input_r[model_main_input_name], input_p[model_main_input_name]):
self.assert_padded_input_match(i_r, i_p, max_length, pad_token_id)
for i_r, i_p in zip(input_r["attention_mask"], input_p["attention_mask"]):
self.assertSequenceEqual(i_r, i_p)
@staticmethod
def convert_batch_encode_plus_format_to_encode_plus(batch_encode_plus_sequences):
# Switch from batch_encode_plus format: {'input_ids': [[...], [...]], ...}
# to the list of examples/ encode_plus format: [{'input_ids': [...], ...}, {'input_ids': [...], ...}]
return [
{value: batch_encode_plus_sequences[value][i] for value in batch_encode_plus_sequences.keys()}
for i in range(len(batch_encode_plus_sequences["input_ids"]))
]
# TODO: this test can be combined with `test_sentencepiece_tokenize_and_convert_tokens_to_string` after the latter is extended to all tokenizers.
def test_tokenize_special_tokens(self):
"""Test `tokenize` with special tokens."""
tokenizers = self.get_tokenizers(do_lower_case=True)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
SPECIAL_TOKEN_1 = "[SPECIAL_TOKEN_1]"
SPECIAL_TOKEN_2 = "[SPECIAL_TOKEN_2]"
# TODO:
# Can we combine `unique_no_split_tokens` and `all_special_tokens`(and properties related to it)
# with one variable(property) for a better maintainability?
# `add_tokens` method stores special tokens only in `tokenizer.unique_no_split_tokens`. (in tokenization_utils.py)
tokenizer.add_tokens([SPECIAL_TOKEN_1], special_tokens=True)
# `add_special_tokens` method stores special tokens in `tokenizer.additional_special_tokens`,
# which also occur in `tokenizer.all_special_tokens`. (in tokenization_utils_base.py)
tokenizer.add_special_tokens({"additional_special_tokens": [SPECIAL_TOKEN_2]})
token_1 = tokenizer.tokenize(SPECIAL_TOKEN_1)
token_2 = tokenizer.tokenize(SPECIAL_TOKEN_2)
self.assertEqual(len(token_1), 1)
self.assertEqual(len(token_2), 1)
self.assertEqual(token_1[0], SPECIAL_TOKEN_1)
self.assertEqual(token_2[0], SPECIAL_TOKEN_2)
# TODO: this test could be extended to all tokenizers - not just the sentencepiece
def test_sentencepiece_tokenize_and_convert_tokens_to_string(self):
"""Test ``_tokenize`` and ``convert_tokens_to_string``."""
if not self.test_sentencepiece:
return
tokenizer = self.get_tokenizer()
text = "This is text to test the tokenizer."
if self.test_sentencepiece_ignore_case:
text = text.lower()
tokens = tokenizer.tokenize(text)
self.assertTrue(len(tokens) > 0)
# check if converting back to original text works
reverse_text = tokenizer.convert_tokens_to_string(tokens)
if self.test_sentencepiece_ignore_case:
reverse_text = reverse_text.lower()
self.assertEqual(reverse_text, text)
def test_subword_regularization_tokenizer(self) -> None:
if not self.test_sentencepiece:
return
# Subword regularization is only available for the slow tokenizer.
sp_model_kwargs = {"enable_sampling": True, "alpha": 0.1, "nbest_size": -1}
tokenizer = self.get_tokenizer(sp_model_kwargs=sp_model_kwargs)
self.assertTrue(hasattr(tokenizer, "sp_model_kwargs"))
self.assertIsNotNone(tokenizer.sp_model_kwargs)
self.assertTrue(isinstance(tokenizer.sp_model_kwargs, dict))
self.assertEqual(tokenizer.sp_model_kwargs, sp_model_kwargs)
self.check_subword_sampling(tokenizer)
def test_pickle_subword_regularization_tokenizer(self) -> None:
if not self.test_sentencepiece:
return
"""Google pickle __getstate__ __setstate__ if you are struggling with this."""
# Subword regularization is only available for the slow tokenizer.
sp_model_kwargs = {"enable_sampling": True, "alpha": 0.1, "nbest_size": -1}
tokenizer = self.get_tokenizer(sp_model_kwargs=sp_model_kwargs)
tokenizer_bin = pickle.dumps(tokenizer)
del tokenizer
tokenizer_new = pickle.loads(tokenizer_bin)
self.assertTrue(hasattr(tokenizer_new, "sp_model_kwargs"))
self.assertIsNotNone(tokenizer_new.sp_model_kwargs)
self.assertTrue(isinstance(tokenizer_new.sp_model_kwargs, dict))
self.assertEqual(tokenizer_new.sp_model_kwargs, sp_model_kwargs)
self.check_subword_sampling(tokenizer_new)
def test_save_sentencepiece_tokenizer(self) -> None:
if not self.test_sentencepiece:
return
# We want to verify that we will be able to save the tokenizer even if the original files that were used to
# build the tokenizer have been deleted in the meantime.
text = "This is text to test the tokenizer."
tokenizer_slow_1 = self.get_tokenizer()
encoding_tokenizer_slow_1 = tokenizer_slow_1(text)
tmpdirname_1 = tempfile.mkdtemp()
tmpdirname_2 = tempfile.mkdtemp()
tokenizer_slow_1.save_pretrained(tmpdirname_1)
tokenizer_slow_2 = self.tokenizer_class.from_pretrained(tmpdirname_1)
encoding_tokenizer_slow_2 = tokenizer_slow_2(text)
shutil.rmtree(tmpdirname_1)
tokenizer_slow_2.save_pretrained(tmpdirname_2)
tokenizer_slow_3 = self.tokenizer_class.from_pretrained(tmpdirname_2)
encoding_tokenizer_slow_3 = tokenizer_slow_3(text)
shutil.rmtree(tmpdirname_2)
self.assertEqual(encoding_tokenizer_slow_1, encoding_tokenizer_slow_2)
self.assertEqual(encoding_tokenizer_slow_1, encoding_tokenizer_slow_3)
def test_model_input_names_signature(self):
accepted_model_main_input_names = [
"input_ids", # nlp models
"input_values", # speech models
]
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
# first name of model_input_names has to correspond to main model input name
# to make sure `tokenizer.pad(...)` works correctly
self.assertTrue(tokenizer.model_input_names[0] in accepted_model_main_input_names)
def test_tokenizer_slow_store_full_signature(self):
signature = inspect.signature(self.tokenizer_class.__init__)
tokenizer = self.get_tokenizer()
for parameter_name, parameter in signature.parameters.items():
if parameter.default != inspect.Parameter.empty:
self.assertIn(parameter_name, tokenizer.init_kwargs)
def test_tokenizers_common_properties(self):
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
attributes_list = [
"bos_token",
"eos_token",
"unk_token",
"sep_token",
"pad_token",
"cls_token",
"mask_token",
]
for attr in attributes_list:
self.assertTrue(hasattr(tokenizer, attr))
self.assertTrue(hasattr(tokenizer, attr + "_id"))
self.assertTrue(hasattr(tokenizer, "additional_special_tokens"))
self.assertTrue(hasattr(tokenizer, "additional_special_tokens_ids"))
attributes_list = [
"model_max_length",
"init_inputs",
"init_kwargs",
]
attributes_list += [
"added_tokens_encoder",
"added_tokens_decoder",
]
for attr in attributes_list:
self.assertTrue(hasattr(tokenizer, attr))
def test_tokenizers_common_ids_setters(self):
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
attributes_list = [
"bos_token",
"eos_token",
"unk_token",
"sep_token",
"pad_token",
"cls_token",
"mask_token",
]
vocab = tokenizer.get_vocab()
token_id_to_test_setters = next(iter(vocab.values()))
token_to_test_setters = tokenizer.convert_ids_to_tokens(
token_id_to_test_setters, skip_special_tokens=False
)
for attr in attributes_list:
setattr(tokenizer, attr + "_id", None)
self.assertEqual(getattr(tokenizer, attr), None)
self.assertEqual(getattr(tokenizer, attr + "_id"), None)
setattr(tokenizer, attr + "_id", token_id_to_test_setters)
self.assertEqual(getattr(tokenizer, attr), token_to_test_setters)
self.assertEqual(getattr(tokenizer, attr + "_id"), token_id_to_test_setters)
setattr(tokenizer, "additional_special_tokens_ids", [])
self.assertListEqual(getattr(tokenizer, "additional_special_tokens"), [])
self.assertListEqual(getattr(tokenizer, "additional_special_tokens_ids"), [])
setattr(tokenizer, "additional_special_tokens_ids", [token_id_to_test_setters])
self.assertListEqual(getattr(tokenizer, "additional_special_tokens"), [token_to_test_setters])
self.assertListEqual(getattr(tokenizer, "additional_special_tokens_ids"), [token_id_to_test_setters])
def test_save_and_load_tokenizer(self):
# safety check on max_len default value so we are sure the test works
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
self.assertNotEqual(tokenizer.model_max_length, 42)
# Now let's start the test
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
# Isolate this from the other tests because we save additional tokens/etc
tmpdirname = tempfile.mkdtemp()
sample_text = " He is very happy, UNwant\u00E9d,running"
before_tokens = tokenizer.encode(sample_text, add_special_tokens=False)
before_vocab = tokenizer.get_vocab()
tokenizer.save_pretrained(tmpdirname)
after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
after_tokens = after_tokenizer.encode(sample_text, add_special_tokens=False)
after_vocab = after_tokenizer.get_vocab()
self.assertListEqual(before_tokens["input_ids"], after_tokens["input_ids"])
self.assertDictEqual(before_vocab, after_vocab)
shutil.rmtree(tmpdirname)
tokenizers = self.get_tokenizers(model_max_length=42)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
# Isolate this from the other tests because we save additional tokens/etc
tmpdirname = tempfile.mkdtemp()
sample_text = " He is very happy, UNwant\u00E9d,running"
tokenizer.add_tokens(["bim", "bambam"])
additional_special_tokens = tokenizer.additional_special_tokens
additional_special_tokens.append("new_additional_special_token")
tokenizer.add_special_tokens({"additional_special_tokens": additional_special_tokens})
before_tokens = tokenizer.encode(sample_text, add_special_tokens=False)
before_vocab = tokenizer.get_vocab()
tokenizer.save_pretrained(tmpdirname)
after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
after_tokens = after_tokenizer.encode(sample_text, add_special_tokens=False)
after_vocab = after_tokenizer.get_vocab()
self.assertListEqual(before_tokens["input_ids"], after_tokens["input_ids"])
self.assertDictEqual(before_vocab, after_vocab)
self.assertIn("bim", after_vocab)
self.assertIn("bambam", after_vocab)
self.assertIn("new_additional_special_token", after_tokenizer.additional_special_tokens)
self.assertEqual(after_tokenizer.model_max_length, 42)
tokenizer = tokenizer.__class__.from_pretrained(tmpdirname, model_max_length=43)
self.assertEqual(tokenizer.model_max_length, 43)
shutil.rmtree(tmpdirname)
# Test that we can also use the non-legacy saving format for fast tokenizers
tokenizers = self.get_tokenizers(model_max_length=42)
for tokenizer in tokenizers:
if not tokenizer.is_fast:
continue
with self.subTest(f"{tokenizer.__class__.__name__}"):
# Isolate this from the other tests because we save additional tokens/etc
tmpdirname = tempfile.mkdtemp()
sample_text = " He is very happy, UNwant\u00E9d,running"
tokenizer.add_tokens(["bim", "bambam"])
additional_special_tokens = tokenizer.additional_special_tokens
additional_special_tokens.append("new_additional_special_token")
tokenizer.add_special_tokens({"additional_special_tokens": additional_special_tokens})
before_tokens = tokenizer.encode(sample_text, add_special_tokens=False)
before_vocab = tokenizer.get_vocab()
tokenizer.save_pretrained(tmpdirname)
after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
after_tokens = after_tokenizer.encode(sample_text, add_special_tokens=False)
after_vocab = after_tokenizer.get_vocab()
self.assertListEqual(before_tokens, after_tokens)
self.assertDictEqual(before_vocab, after_vocab)
self.assertIn("bim", after_vocab)
self.assertIn("bambam", after_vocab)
self.assertIn("new_additional_special_token", after_tokenizer.additional_special_tokens)
self.assertEqual(after_tokenizer.model_max_length, 42)
tokenizer = tokenizer.__class__.from_pretrained(tmpdirname, model_max_length=43)
self.assertEqual(tokenizer.model_max_length, 43)
shutil.rmtree(tmpdirname)
# def test_pickle_tokenizer(self):
# """Google pickle __getstate__ __setstate__ if you are struggling with this."""
# tokenizers = self.get_tokenizers()
# for tokenizer in tokenizers:
# with self.subTest(f"{tokenizer.__class__.__name__}"):
# self.assertIsNotNone(tokenizer)
#
# text = "Munich and Berlin are nice cities"
# subwords = tokenizer.tokenize(text)
#
# filename = os.path.join(self.tmpdirname, "tokenizer.bin")
# with open(filename, "wb") as handle:
# pickle.dump(tokenizer, handle)
#
# with open(filename, "rb") as handle:
# tokenizer_new = pickle.load(handle)
#
# subwords_loaded = tokenizer_new.tokenize(text)
#
# self.assertListEqual(subwords, subwords_loaded)
def test_pickle_added_tokens(self):
tok1 = AddedToken("<s>", rstrip=True, lstrip=True, normalized=False, single_word=True)
tok2 = pickle.loads(pickle.dumps(tok1))
self.assertEqual(tok1.__getstate__(), tok2.__getstate__())
def test_added_tokens_do_lower_case(self):
tokenizers = self.get_tokenizers(do_lower_case=True)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
if not hasattr(tokenizer, "do_lower_case") or not tokenizer.do_lower_case:
continue
special_token = tokenizer.all_special_tokens[0]
text = special_token + " aaaaa bbbbbb low cccccccccdddddddd l " + special_token
text2 = special_token + " AAAAA BBBBBB low CCCCCCCCCDDDDDDDD l " + special_token
toks_before_adding = tokenizer.tokenize(text) # toks before adding new_toks
new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd", "AAAAA BBBBBB", "CCCCCCCCCDDDDDDDD"]
added = tokenizer.add_tokens([AddedToken(tok, lstrip=True, rstrip=True) for tok in new_toks])
toks_after_adding = tokenizer.tokenize(text)
toks_after_adding2 = tokenizer.tokenize(text2)
# Rust tokenizers dont't lowercase added tokens at the time calling `tokenizer.add_tokens`,
# while python tokenizers do, so new_toks 0 and 2 would be treated as the same, so do new_toks 1 and 3.
self.assertIn(added, [2, 4])
self.assertListEqual(toks_after_adding, toks_after_adding2)
self.assertTrue(
len(toks_before_adding) > len(toks_after_adding), # toks_before_adding should be longer
)
# Check that none of the special tokens are lowercased
sequence_with_special_tokens = "A " + " yEs ".join(tokenizer.all_special_tokens) + " B"
# Convert the tokenized list to str as some special tokens are tokenized like normal tokens
# which have a prefix spacee e.g. the mask token of Albert, and cannot match the original
# special tokens exactly.
tokenized_sequence = "".join(tokenizer.tokenize(sequence_with_special_tokens))
for special_token in tokenizer.all_special_tokens:
self.assertTrue(special_token in tokenized_sequence)
tokenizers = self.get_tokenizers(do_lower_case=True)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
if hasattr(tokenizer, "do_lower_case") and tokenizer.do_lower_case:
continue
special_token = tokenizer.all_special_tokens[0]
text = special_token + " aaaaa bbbbbb low cccccccccdddddddd l " + special_token
text2 = special_token + " AAAAA BBBBBB low CCCCCCCCCDDDDDDDD l " + special_token
toks_before_adding = tokenizer.tokenize(text) # toks before adding new_toks
new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd", "AAAAA BBBBBB", "CCCCCCCCCDDDDDDDD"]
added = tokenizer.add_tokens([AddedToken(tok, lstrip=True, rstrip=True) for tok in new_toks])
self.assertIn(added, [2, 4])
toks_after_adding = tokenizer.tokenize(text)
toks_after_adding2 = tokenizer.tokenize(text2)
self.assertEqual(len(toks_after_adding), len(toks_after_adding2)) # Length should still be the same
self.assertNotEqual(
toks_after_adding[1], toks_after_adding2[1]
) # But at least the first non-special tokens should differ
self.assertTrue(
len(toks_before_adding) > len(toks_after_adding), # toks_before_adding should be longer
)
def test_add_tokens_tokenizer(self):
tokenizers = self.get_tokenizers(do_lower_case=False)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
vocab_size = tokenizer.vocab_size
all_size = len(tokenizer)
self.assertNotEqual(vocab_size, 0)
# We usually have added tokens from the start in tests because our vocab fixtures are
# smaller than the original vocabs - let's not assert this
# self.assertEqual(vocab_size, all_size)
new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd"]
added_toks = tokenizer.add_tokens(new_toks)
vocab_size_2 = tokenizer.vocab_size
all_size_2 = len(tokenizer)
self.assertNotEqual(vocab_size_2, 0)
self.assertEqual(vocab_size, vocab_size_2)
self.assertEqual(added_toks, len(new_toks))
self.assertEqual(all_size_2, all_size + len(new_toks))
tokens = tokenizer.encode(
"aaaaa bbbbbb low cccccccccdddddddd l", return_token_type_ids=None, add_special_tokens=False
)["input_ids"]
self.assertGreaterEqual(len(tokens), 4)
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
new_toks_2 = {"eos_token": ">>>>|||<||<<|<<", "pad_token": "<<<<<|||>|>>>>|>"}
added_toks_2 = tokenizer.add_special_tokens(new_toks_2)
vocab_size_3 = tokenizer.vocab_size
all_size_3 = len(tokenizer)
self.assertNotEqual(vocab_size_3, 0)
self.assertEqual(vocab_size, vocab_size_3)
self.assertEqual(added_toks_2, len(new_toks_2))
self.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
tokens = tokenizer.encode(
">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l",
return_token_type_ids=None,
add_special_tokens=False,
)["input_ids"]
self.assertGreaterEqual(len(tokens), 6)
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
self.assertGreater(tokens[0], tokens[1])
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
self.assertGreater(tokens[-2], tokens[-3])
self.assertEqual(tokens[0], tokenizer.eos_token_id)
self.assertEqual(tokens[-2], tokenizer.pad_token_id)
def test_add_special_tokens(self):
tokenizers = self.get_tokenizers(do_lower_case=False)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
input_text, ids = self.get_clean_sequence(tokenizer)
special_token = "[SPECIAL_TOKEN]"
tokenizer.add_special_tokens({"cls_token": special_token})
encoded_special_token = tokenizer.encode(
special_token, return_token_type_ids=None, add_special_tokens=False
)["input_ids"]
self.assertEqual(len(encoded_special_token), 1)
text = tokenizer.decode(ids + encoded_special_token, clean_up_tokenization_spaces=False)
encoded = tokenizer.encode(text, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
input_encoded = tokenizer.encode(input_text, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
]
special_token_id = tokenizer.encode(
special_token, return_token_type_ids=None, add_special_tokens=False
)["input_ids"]
self.assertEqual(encoded, input_encoded + special_token_id)
decoded = tokenizer.decode(encoded, skip_special_tokens=True)
self.assertTrue(special_token not in decoded)
def test_internal_consistency(self):
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
input_text, output_text = self.get_input_output_texts(tokenizer)
tokens = tokenizer.tokenize(input_text)
ids = tokenizer.convert_tokens_to_ids(tokens)
ids_2 = tokenizer.encode(input_text, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
self.assertListEqual(ids, ids_2)
tokens_2 = tokenizer.convert_ids_to_tokens(ids)
self.assertNotEqual(len(tokens_2), 0)
text_2 = tokenizer.decode(ids)
self.assertIsInstance(text_2, str)
self.assertEqual(text_2, output_text)
def test_encode_decode_with_spaces(self):
tokenizers = self.get_tokenizers(do_lower_case=False)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
new_toks = [
AddedToken("[ABC]", normalized=False),
AddedToken("[DEF]", normalized=False),
AddedToken("GHI IHG", normalized=False),
]
tokenizer.add_tokens(new_toks)
input = "[ABC][DEF][ABC]GHI IHG[DEF]"
if self.space_between_special_tokens:
output = "[ABC] [DEF] [ABC] GHI IHG [DEF]"
else:
output = input
encoded = tokenizer.encode(input, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
decoded = tokenizer.decode(encoded, spaces_between_special_tokens=self.space_between_special_tokens)
self.assertIn(decoded, [output, output.lower()])
def test_pretrained_model_lists(self):
# We should have at least one default checkpoint for each tokenizer
# We should specify the max input length as well (used in some part to list the pretrained checkpoints)
self.assertGreaterEqual(len(self.tokenizer_class.pretrained_resource_files_map), 1)
self.assertGreaterEqual(len(list(self.tokenizer_class.pretrained_resource_files_map.values())[0]), 1)
self.assertEqual(
len(list(self.tokenizer_class.pretrained_resource_files_map.values())[0]),
len(self.tokenizer_class.max_model_input_sizes),
)
weights_list = list(self.tokenizer_class.max_model_input_sizes.keys())
weights_lists_2 = []
for file_id, map_list in self.tokenizer_class.pretrained_resource_files_map.items():
weights_lists_2.append(list(map_list.keys()))
weights_list.sort()
for weights_list_2 in weights_lists_2:
weights_list_2.sort()
self.assertListEqual(weights_list, weights_list_2)
def test_mask_output(self):
tokenizers = self.get_tokenizers(do_lower_case=False)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
if (
tokenizer.build_inputs_with_special_tokens.__qualname__.split(".")[0] != "PretrainedTokenizer"
and "token_type_ids" in tokenizer.model_input_names
):
seq_0 = "Test this method."
seq_1 = "With these inputs."
information = tokenizer.encode(seq_0, seq_1, add_special_tokens=True)
sequences, mask = information["input_ids"], information["token_type_ids"]
self.assertEqual(len(sequences), len(mask))
def test_token_type_ids(self):
tokenizers = self.get_tokenizers()
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
seq_0 = "Test this method."
# We want to have sequence 0 and sequence 1 are tagged
# respectively with 0 and 1 token_ids
# (regardless of whether the model use token type ids)
# We use this assumption in the QA pipeline among other place
output = tokenizer(seq_0, return_token_type_ids=True)
self.assertIn(0, output["token_type_ids"])
def test_number_of_added_tokens(self):
tokenizers = self.get_tokenizers(do_lower_case=False)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
seq_0 = "Test this method."
seq_1 = "With these inputs."
sequences = tokenizer.encode(seq_0, seq_1, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
]
attached_sequences = tokenizer.encode(seq_0, seq_1, add_special_tokens=True)["input_ids"]
# Method is implemented (e.g. not GPT-2)
if len(attached_sequences) != 2:
self.assertEqual(
tokenizer.num_special_tokens_to_add(pair=True), len(attached_sequences) - len(sequences)
)
def test_maximum_encoding_length_single_input(self):
tokenizers = self.get_tokenizers(do_lower_case=False, model_max_length=100)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
seq_0, ids = self.get_clean_sequence(tokenizer, max_length=20)
sequence = tokenizer.encode(seq_0, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
total_length = len(sequence)
self.assertGreater(total_length, 4, "Issue with the testing sequence, please update it it's too short")
# Test with max model input length
model_max_length = tokenizer.model_max_length
self.assertEqual(model_max_length, 100)
seq_1 = seq_0 * model_max_length
sequence1 = tokenizer(seq_1, return_token_type_ids=None, add_special_tokens=False, truncation=False)
total_length1 = len(sequence1["input_ids"])
self.assertGreater(
total_length1, model_max_length, "Issue with the testing sequence, please update it it's too short"
)
# Simple
padding_strategies = (
[False, True, "longest"] if tokenizer.pad_token and tokenizer.pad_token_id >= 0 else [False]
)
for padding_state in padding_strategies:
with self.subTest(f"Padding: {padding_state}"):
for truncation_state in [True, "longest_first", "only_first"]:
with self.subTest(f"Truncation: {truncation_state}"):
output = tokenizer(seq_1, padding=padding_state, truncation=truncation_state)
self.assertEqual(len(output["input_ids"]), model_max_length)
output = tokenizer([seq_1], padding=padding_state, truncation=truncation_state)
self.assertEqual(len(output["input_ids"][0]), model_max_length)
# Simple with no truncation
# Reset warnings
tokenizer.deprecation_warnings = {}
with self.assertLogs("PaddleNLP", level="WARNING") as cm:
output = tokenizer(seq_1, padding=padding_state, truncation=False)
self.assertNotEqual(len(output["input_ids"]), model_max_length)
self.assertEqual(len(cm.records), 1)
self.assertTrue(
cm.records[0].message.startswith(
"Token indices sequence length is longer than the specified maximum sequence length for this model"
)
)
tokenizer.deprecation_warnings = {}
with self.assertLogs("PaddleNLP", level="WARNING") as cm:
output = tokenizer([seq_1], padding=padding_state, truncation=False)
self.assertNotEqual(len(output["input_ids"][0]), model_max_length)
self.assertEqual(len(cm.records), 1)
self.assertTrue(
cm.records[0].message.startswith(
"Token indices sequence length is longer than the specified maximum sequence length for this model"
)
)
# Overflowing tokens
stride = 2
information = tokenizer(
seq_0,
max_length=total_length - 2,
return_token_type_ids=None,
add_special_tokens=False,
stride=stride,
truncation="longest_first",
return_overflowing_tokens=True,
# add_prefix_space=False,
)
# Overflowing tokens are handled quite differently in slow and fast tokenizers
truncated_sequence = information["input_ids"]
overflowing_tokens = information["overflowing_tokens"]
self.assertEqual(len(truncated_sequence), total_length - 2)
self.assertEqual(truncated_sequence, sequence[:-2])
self.assertEqual(len(overflowing_tokens), 2 + stride)
self.assertEqual(overflowing_tokens, sequence[-(2 + stride) :])
def test_maximum_encoding_length_pair_input(self):
tokenizers = self.get_tokenizers(do_lower_case=False, model_max_length=100)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
# Build a sequence from our model's vocabulary
stride = 2
seq_0, ids = self.get_clean_sequence(tokenizer, max_length=20)
if len(ids) <= 2 + stride:
seq_0 = (seq_0 + " ") * (2 + stride)
ids = None
seq0_tokens = tokenizer.encode(seq_0, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
]
self.assertGreater(len(seq0_tokens), 2 + stride)
seq_1 = "This is another sentence to be encoded."
seq1_tokens = tokenizer.encode(seq_1, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
]
if abs(len(seq0_tokens) - len(seq1_tokens)) <= 2:
seq1_tokens = seq1_tokens + seq1_tokens
seq_1 = tokenizer.decode(seq1_tokens, clean_up_tokenization_spaces=False)
seq1_tokens = tokenizer.encode(seq_1, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
]
self.assertGreater(len(seq1_tokens), 2 + stride)
# We are not using the special tokens - a bit too hard to test all the tokenizers with this
# TODO try this again later
sequence = tokenizer.encode(seq_0, seq_1, return_token_type_ids=None, add_special_tokens=False)[
"input_ids"
] # , add_prefix_space=False)
# Test with max model input length
model_max_length = tokenizer.model_max_length
self.assertEqual(model_max_length, 100)
seq_2 = seq_0 * model_max_length
self.assertGreater(len(seq_2), model_max_length)
sequence1 = tokenizer(seq_1, return_token_type_ids=None, add_special_tokens=False, truncation=False)
total_length1 = len(sequence1["input_ids"])
sequence2 = tokenizer(
seq_2, seq_1, return_token_type_ids=None, add_special_tokens=False, truncation=False
)
total_length2 = len(sequence2["input_ids"])
self.assertLess(
total_length1, model_max_length - 10, "Issue with the testing sequence, please update it."
)
self.assertGreater(
total_length2, model_max_length, "Issue with the testing sequence, please update it."
)
# Simple
padding_strategies = (
[False, True, "longest"] if tokenizer.pad_token and tokenizer.pad_token_id >= 0 else [False]
)
for padding_state in padding_strategies:
with self.subTest(f"{tokenizer.__class__.__name__} Padding: {padding_state}"):
for truncation_state in [True, "longest_first", "only_first"]:
with self.subTest(f"{tokenizer.__class__.__name__} Truncation: {truncation_state}"):
output = tokenizer(seq_2, seq_1, padding=padding_state, truncation=truncation_state)
self.assertEqual(len(output["input_ids"]), model_max_length)
output = tokenizer(
[seq_2], [seq_1], padding=padding_state, truncation=truncation_state
)
self.assertEqual(len(output["input_ids"][0]), model_max_length)
# Simple
output = tokenizer(seq_1, seq_2, padding=padding_state, truncation="only_second")
self.assertEqual(len(output["input_ids"]), model_max_length)
output = tokenizer([seq_1], [seq_2], padding=padding_state, truncation="only_second")
self.assertEqual(len(output["input_ids"][0]), model_max_length)
# Simple with no truncation
# Reset warnings
tokenizer.deprecation_warnings = {}
with self.assertLogs("PaddleNLP", level="WARNING") as cm:
output = tokenizer(seq_1, seq_2, padding=padding_state, truncation=False)
self.assertNotEqual(len(output["input_ids"]), model_max_length)
self.assertEqual(len(cm.records), 1)
self.assertTrue(
cm.records[0].message.startswith(
"Token indices sequence length is longer than the specified maximum sequence length for this model"
)
)
tokenizer.deprecation_warnings = {}
with self.assertLogs("PaddleNLP", level="WARNING") as cm:
output = tokenizer([seq_1], [seq_2], padding=padding_state, truncation=False)
self.assertNotEqual(len(output["input_ids"][0]), model_max_length)
self.assertEqual(len(cm.records), 1)
self.assertTrue(
cm.records[0].message.startswith(
"Token indices sequence length is longer than the specified maximum sequence length for this model"