-
Notifications
You must be signed in to change notification settings - Fork 34.4k
Expand file tree
/
Copy pathtokenization_utils_base.py
More file actions
3687 lines (3166 loc) 路 178 KB
/
Copy pathtokenization_utils_base.py
File metadata and controls
3687 lines (3166 loc) 路 178 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
# base
# Copyright 2020 The 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.
"""
Base classes common to both the slow and the fast tokenization classes: PreTrainedTokenizerBase (host all the user
fronting encoding methods) Special token mixing (host the special tokens logic) and BatchEncoding (wrap the dictionary
of output with special method for the Fast tokenizers)
"""
from __future__ import annotations
import copy
import json
import os
import re
import warnings
from collections import OrderedDict, UserDict
from collections.abc import Callable, Collection, Mapping, Sequence, Sized
from dataclasses import dataclass
from pathlib import Path
from typing import TYPE_CHECKING, Any, Generic, NamedTuple, Union, overload
import numpy as np
from huggingface_hub import is_offline_mode
from packaging import version
from typing_extensions import TypeVar
from . import __version__
from .dynamic_module_utils import custom_object_save
from .utils import (
CHAT_TEMPLATE_DIR,
CHAT_TEMPLATE_FILE,
ExplicitEnum,
PaddingStrategy,
PushToHubMixin,
TensorType,
add_end_docstrings,
cached_file,
copy_func,
extract_commit_hash,
hf_api,
is_mlx_available,
is_numpy_array,
is_protobuf_available,
is_tokenizers_available,
is_torch_available,
is_torch_device,
is_torch_tensor,
list_repo_templates,
logging,
requires_backends,
to_py_obj,
)
from .utils.chat_parsing import ResponseParser
from .utils.chat_parsing import parse_response as _template_parse_response
from .utils.chat_template_utils import render_jinja_template
if TYPE_CHECKING:
if is_torch_available():
import torch
def import_protobuf_decode_error(error_message=""):
if is_protobuf_available():
from google.protobuf.message import DecodeError
return DecodeError
return ()
def flatten(arr: list):
res = []
if len(arr) > 0:
for sub_arr in arr:
if isinstance(arr[0], (list, tuple)):
res.extend(flatten(sub_arr))
else:
res.append(sub_arr)
return res
if is_tokenizers_available() or TYPE_CHECKING:
from tokenizers import Encoding as EncodingFast
if is_tokenizers_available():
from tokenizers import AddedToken
else:
@dataclass(frozen=False, eq=True)
class AddedToken:
"""
AddedToken represents a token to be added to a Tokenizer An AddedToken can have special options defining the
way it should behave.
The `normalized` will default to `not special` if it is not specified, similarly to the definition in
`tokenizers`.
"""
def __init__(
self, content: str, single_word=False, lstrip=False, rstrip=False, special=False, normalized=None
):
self.content = content
self.single_word = single_word
self.lstrip = lstrip
self.rstrip = rstrip
self.special = special
self.normalized = normalized if normalized is not None else not special
def __getstate__(self):
return self.__dict__
def __str__(self):
return self.content
logger = logging.get_logger(__name__)
VERY_LARGE_INTEGER = int(1e30) # This is used to set the max input length for a model with infinite size input
LARGE_INTEGER = int(1e20) # This is used when we need something big but slightly smaller than VERY_LARGE_INTEGER
# Define type aliases and NamedTuples
TextInput = str
PreTokenizedInput = list[str]
EncodedInput = list[int]
TextInputPair = tuple[str, str]
PreTokenizedInputPair = tuple[list[str], list[str]]
EncodedInputPair = tuple[list[int], list[int]]
# Define type aliases for text-related non-text modalities
AudioInput = Union[np.ndarray, "torch.Tensor", list[np.ndarray], list["torch.Tensor"]]
# Slow tokenizers used to be saved in three separated files
SPECIAL_TOKENS_MAP_FILE = "special_tokens_map.json"
ADDED_TOKENS_FILE = "added_tokens.json"
TOKENIZER_CONFIG_FILE = "tokenizer_config.json"
# Fast tokenizers (provided by HuggingFace tokenizer's library) can be saved in a single file
FULL_TOKENIZER_FILE = "tokenizer.json"
_re_tokenizer_file = re.compile(r"tokenizer\.(.*)\.json")
class TruncationStrategy(ExplicitEnum):
"""
Possible values for the `truncation` argument in [`PreTrainedTokenizerBase.__call__`]. Useful for tab-completion in
an IDE.
"""
ONLY_FIRST = "only_first"
ONLY_SECOND = "only_second"
LONGEST_FIRST = "longest_first"
DO_NOT_TRUNCATE = "do_not_truncate"
class CharSpan(NamedTuple):
"""
Character span in the original string.
Args:
start (`int`): Index of the first character in the original string.
end (`int`): Index of the character following the last character in the original string.
"""
start: int
end: int
class TokenSpan(NamedTuple):
"""
Token span in an encoded string (list of tokens).
Args:
start (`int`): Index of the first token in the span.
end (`int`): Index of the token following the last token in the span.
"""
start: int
end: int
_V = TypeVar("_V", default=Any)
class BatchEncoding(UserDict, Generic[_V]):
"""
Holds the output of the [`~tokenization_utils_base.PreTrainedTokenizerBase.__call__`],
[`~tokenization_utils_base.PreTrainedTokenizerBase.encode_plus`] and
[`~tokenization_utils_base.PreTrainedTokenizerBase.batch_encode_plus`] methods (tokens, attention_masks, etc).
This class is derived from a python dictionary and can be used as a dictionary. In addition, this class exposes
utility methods to map from word/character space to token space.
Args:
data (`dict`, *optional*):
Dictionary of lists/arrays/tensors returned by the `__call__`/`encode_plus`/`batch_encode_plus` methods
('input_ids', 'attention_mask', etc.).
encoding (`tokenizers.Encoding` or `Sequence[tokenizers.Encoding]`, *optional*):
If the tokenizer is a fast tokenizer which outputs additional information like mapping from word/character
space to token space the `tokenizers.Encoding` instance or list of instance (for batches) hold this
information.
tensor_type (`Union[None, str, TensorType]`, *optional*):
You can give a tensor_type here to convert the lists of integers in PyTorch/Numpy Tensors at
initialization.
prepend_batch_axis (`bool`, *optional*, defaults to `False`):
Whether or not to add a batch axis when converting to tensors (see `tensor_type` above). Note that this
parameter has an effect if the parameter `tensor_type` is set, *otherwise has no effect*.
n_sequences (`Optional[int]`, *optional*):
The number of input sequences represented by each encoding (`None` for unknown, `1` for a single sequence and `2` for a pair of sequences).
"""
def __init__(
self,
data: dict[str, Any] | None = None,
encoding: EncodingFast | Sequence[EncodingFast] | None = None,
tensor_type: None | str | TensorType = None,
prepend_batch_axis: bool = False,
n_sequences: int | None = None,
):
super().__init__(data)
# If encoding is not None, the fast tokenization is used
if encoding is not None and isinstance(encoding, EncodingFast):
encoding = [encoding]
self._encodings = encoding
if n_sequences is None and encoding is not None and encoding:
n_sequences = encoding[0].n_sequences
self._n_sequences = n_sequences
self.convert_to_tensors(tensor_type=tensor_type, prepend_batch_axis=prepend_batch_axis)
@property
def n_sequences(self) -> int | None:
"""
`Optional[int]`: The number of sequences used to generate each sample from the batch encoded in this
[`BatchEncoding`]. Currently can be one of `None` (unknown), `1` (a single sentence) or `2` (a pair of
sentences)
"""
return self._n_sequences
@overload
def __getitem__(self, item: str) -> _V: ...
@overload
def __getitem__(self, item: int) -> EncodingFast: ...
@overload
def __getitem__(self, item: slice) -> dict[str, _V]: ...
def __getitem__(self, item: int | str | slice) -> _V | EncodingFast | dict[str, _V]:
"""
If the key is a string, returns the value of the dict associated to `key` ('input_ids', 'attention_mask',
etc.).
If the key is an integer, get the `tokenizers.Encoding` for batch item with index `key`.
If the key is a slice, returns the value of the dict associated to `key` ('input_ids', 'attention_mask', etc.)
with the constraint of slice.
"""
if isinstance(item, str):
return self.data[item]
elif self._encodings is not None:
return self._encodings[item]
elif isinstance(item, slice):
return {key: self.data[key][item] for key in self.data}
else:
raise KeyError(
"Invalid key. Only three types of key are available: "
"(1) string, (2) integers for backend Encoding, and (3) slices for data subsetting."
)
def __getattr__(self, item: str):
try:
return self.data[item]
except KeyError:
raise AttributeError
def __getstate__(self):
return {"data": self.data, "encodings": self._encodings}
def __setstate__(self, state):
if "data" in state:
self.data = state["data"]
if "encodings" in state:
self._encodings = state["encodings"]
# After this point:
# Extended properties and methods only available for fast (Rust-based) tokenizers
# provided by HuggingFace tokenizers library.
@property
def is_fast(self) -> bool:
"""
TODO: ita i will rm this `bool`: Whether or not this BatchEncoding was created by a fast tokenizer.
"""
return self._encodings is not None
@property
def encodings(self) -> list[EncodingFast] | None:
"""
`Optional[list[tokenizers.Encoding]]`: The list all encodings from the tokenization process. Returns `None` if
the input was tokenized through Python (i.e., not a fast) tokenizer.
"""
return self._encodings
def tokens(self, batch_index: int = 0) -> list[str]:
"""
Return the list of tokens (sub-parts of the input strings after word/subword splitting and before conversion to
integer indices) at a given batch index (only works for the output of a fast tokenizer).
Args:
batch_index (`int`, *optional*, defaults to 0): The index to access in the batch.
Returns:
`list[str]`: The list of tokens at that index.
"""
if not self._encodings:
raise ValueError("tokens() is not available when using Python based tokenizers")
return self._encodings[batch_index].tokens
def sequence_ids(self, batch_index: int = 0) -> list[int | None]:
"""
Return a list mapping the tokens to the id of their original sentences:
- `None` for special tokens added around or between sequences,
- `0` for tokens corresponding to words in the first sequence,
- `1` for tokens corresponding to words in the second sequence when a pair of sequences was jointly
encoded.
Args:
batch_index (`int`, *optional*, defaults to 0): The index to access in the batch.
Returns:
`list[Optional[int]]`: A list indicating the sequence id corresponding to each token. Special tokens added
by the tokenizer are mapped to `None` and other tokens are mapped to the index of their corresponding
sequence.
"""
if not self._encodings:
raise ValueError("sequence_ids() is not available when using Python based tokenizers")
return self._encodings[batch_index].sequence_ids
def word_ids(self, batch_index: int = 0) -> list[int | None]:
"""
Return a list mapping the tokens to their actual word in the initial sentence for a fast tokenizer.
Args:
batch_index (`int`, *optional*, defaults to 0): The index to access in the batch.
Returns:
`list[Optional[int]]`: A list indicating the word corresponding to each token. Special tokens added by the
tokenizer are mapped to `None` and other tokens are mapped to the index of their corresponding word
(several tokens will be mapped to the same word index if they are parts of that word).
"""
if not self._encodings:
raise ValueError("word_ids() is not available when using Python based tokenizers")
return self._encodings[batch_index].word_ids
def token_to_sequence(self, batch_or_token_index: int, token_index: int | None = None) -> int:
"""
Get the index of the sequence represented by the given token. In the general use case, this method returns `0`
for a single sequence or the first sequence of a pair, and `1` for the second sequence of a pair
Can be called as:
- `self.token_to_sequence(token_index)` if batch size is 1
- `self.token_to_sequence(batch_index, token_index)` if batch size is greater than 1
This method is particularly suited when the input sequences are provided as pre-tokenized sequences (i.e.,
words are defined by the user). In this case it allows to easily associate encoded tokens with provided
tokenized words.
Args:
batch_or_token_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the token in the sequence.
token_index (`int`, *optional*):
If a batch index is provided in *batch_or_token_index*, this can be the index of the token in the
sequence.
Returns:
`int`: Index of the input sequence containing the token (`0` for the first sequence or `1` for the second sequence of a pair).
"""
if not self._encodings:
raise ValueError("token_to_sequence() is not available when using Python based tokenizers")
if token_index is not None:
batch_index = batch_or_token_index
else:
batch_index = 0
token_index = batch_or_token_index
if batch_index < 0:
batch_index = self._batch_size + batch_index
if token_index < 0:
token_index = self._seq_len + token_index
return self._encodings[batch_index].token_to_sequence(token_index)
def token_to_word(self, batch_or_token_index: int, token_index: int | None = None) -> int:
"""
Get the index of the word corresponding (i.e. comprising) to an encoded token in a sequence of the batch.
Can be called as:
- `self.token_to_word(token_index)` if batch size is 1
- `self.token_to_word(batch_index, token_index)` if batch size is greater than 1
This method is particularly suited when the input sequences are provided as pre-tokenized sequences (i.e.,
words are defined by the user). In this case it allows to easily associate encoded tokens with provided
tokenized words.
Args:
batch_or_token_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the token in the sequence.
token_index (`int`, *optional*):
If a batch index is provided in *batch_or_token_index*, this can be the index of the token in the
sequence.
Returns:
`int`: Index of the word in the input sequence.
"""
if not self._encodings:
raise ValueError("token_to_word() is not available when using Python based tokenizers")
if token_index is not None:
batch_index = batch_or_token_index
else:
batch_index = 0
token_index = batch_or_token_index
if batch_index < 0:
batch_index = self._batch_size + batch_index
if token_index < 0:
token_index = self._seq_len + token_index
return self._encodings[batch_index].token_to_word(token_index)
def word_to_tokens(
self, batch_or_word_index: int, word_index: int | None = None, sequence_index: int = 0
) -> TokenSpan | None:
"""
Get the encoded token span corresponding to a word in a sequence of the batch.
Token spans are returned as a [`~tokenization_utils_base.TokenSpan`] with:
- **start** -- Index of the first token.
- **end** -- Index of the token following the last token.
Can be called as:
- `self.word_to_tokens(word_index, sequence_index: int = 0)` if batch size is 1
- `self.word_to_tokens(batch_index, word_index, sequence_index: int = 0)` if batch size is greater or equal to
1
This method is particularly suited when the input sequences are provided as pre-tokenized sequences (i.e. words
are defined by the user). In this case it allows to easily associate encoded tokens with provided tokenized
words.
Args:
batch_or_word_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the word in the sequence.
word_index (`int`, *optional*):
If a batch index is provided in *batch_or_word_index*, this can be the index of the word in the
sequence.
sequence_index (`int`, *optional*, defaults to 0):
If a pair of sequences is encoded in the batch this can be used to specify which sequence in the pair (0
or 1) the provided word index belongs to.
Returns:
([`~tokenization_utils_base.TokenSpan`], *optional*): Span of tokens in the encoded sequence. Returns
`None` if no tokens correspond to the word. This can happen especially when the token is a special token
that has been used to format the tokenization. For example when we add a class token at the very beginning
of the tokenization.
"""
if not self._encodings:
raise ValueError("word_to_tokens() is not available when using Python based tokenizers")
if word_index is not None:
batch_index = batch_or_word_index
else:
batch_index = 0
word_index = batch_or_word_index
if batch_index < 0:
batch_index = self._batch_size + batch_index
if word_index < 0:
word_index = self._seq_len + word_index
span = self._encodings[batch_index].word_to_tokens(word_index, sequence_index)
return TokenSpan(*span) if span is not None else None
def token_to_chars(self, batch_or_token_index: int, token_index: int | None = None) -> CharSpan | None:
"""
Get the character span corresponding to an encoded token in a sequence of the batch.
Character spans are returned as a [`~tokenization_utils_base.CharSpan`] with:
- **start** -- Index of the first character in the original string associated to the token.
- **end** -- Index of the character following the last character in the original string associated to the
token.
Can be called as:
- `self.token_to_chars(token_index)` if batch size is 1
- `self.token_to_chars(batch_index, token_index)` if batch size is greater or equal to 1
Args:
batch_or_token_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the token in the sequence.
token_index (`int`, *optional*):
If a batch index is provided in *batch_or_token_index*, this can be the index of the token or tokens in
the sequence.
Returns:
[`~tokenization_utils_base.CharSpan`]: Span of characters in the original string, or None, if the token
(e.g. <s>, </s>) doesn't correspond to any chars in the origin string.
"""
if not self._encodings:
raise ValueError("token_to_chars() is not available when using Python based tokenizers")
if token_index is not None:
batch_index = batch_or_token_index
else:
batch_index = 0
token_index = batch_or_token_index
span_indices = self._encodings[batch_index].token_to_chars(token_index)
return CharSpan(*span_indices) if span_indices is not None else None
def char_to_token(self, batch_or_char_index: int, char_index: int | None = None, sequence_index: int = 0) -> int:
"""
Get the index of the token in the encoded output comprising a character in the original string for a sequence
of the batch.
Can be called as:
- `self.char_to_token(char_index)` if batch size is 1
- `self.char_to_token(batch_index, char_index)` if batch size is greater or equal to 1
This method is particularly suited when the input sequences are provided as pre-tokenized sequences (i.e. words
are defined by the user). In this case it allows to easily associate encoded tokens with provided tokenized
words.
Args:
batch_or_char_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the character in the sequence
char_index (`int`, *optional*):
If a batch index is provided in *batch_or_char_index*, this can be the index of the character in the
sequence.
sequence_index (`int`, *optional*, defaults to 0):
If a pair of sequences is encoded in the batch this can be used to specify which sequence in the pair (0
or 1) the provided character index belongs to.
Returns:
`int`: Index of the token, or None if the char index refers to a whitespace only token and whitespace is
trimmed with `trim_offsets=True`.
"""
if not self._encodings:
raise ValueError("char_to_token() is not available when using Python based tokenizers")
if char_index is not None:
batch_index = batch_or_char_index
else:
batch_index = 0
char_index = batch_or_char_index
return self._encodings[batch_index].char_to_token(char_index, sequence_index)
def word_to_chars(
self, batch_or_word_index: int, word_index: int | None = None, sequence_index: int = 0
) -> CharSpan:
"""
Get the character span in the original string corresponding to given word in a sequence of the batch.
Character spans are returned as a CharSpan NamedTuple with:
- start: index of the first character in the original string
- end: index of the character following the last character in the original string
Can be called as:
- `self.word_to_chars(word_index)` if batch size is 1
- `self.word_to_chars(batch_index, word_index)` if batch size is greater or equal to 1
Args:
batch_or_word_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the word in the sequence
word_index (`int`, *optional*):
If a batch index is provided in *batch_or_word_index*, this can be the index of the word in the
sequence.
sequence_index (`int`, *optional*, defaults to 0):
If a pair of sequences is encoded in the batch this can be used to specify which sequence in the pair (0
or 1) the provided word index belongs to.
Returns:
`CharSpan`: Span of the associated character or characters in the string. CharSpan
is a NamedTuple with:
- start: index of the first character associated to the word in the original string
- end: index of the character following the last character associated to the word in the original
string
"""
if not self._encodings:
raise ValueError("word_to_chars() is not available when using Python based tokenizers")
if word_index is not None:
batch_index = batch_or_word_index
else:
batch_index = 0
word_index = batch_or_word_index
return CharSpan(*(self._encodings[batch_index].word_to_chars(word_index, sequence_index)))
def char_to_word(self, batch_or_char_index: int, char_index: int | None = None, sequence_index: int = 0) -> int:
"""
Get the word in the original string corresponding to a character in the original string of a sequence of the
batch.
Can be called as:
- `self.char_to_word(char_index)` if batch size is 1
- `self.char_to_word(batch_index, char_index)` if batch size is greater than 1
This method is particularly suited when the input sequences are provided as pre-tokenized sequences (i.e. words
are defined by the user). In this case it allows to easily associate encoded tokens with provided tokenized
words.
Args:
batch_or_char_index (`int`):
Index of the sequence in the batch. If the batch only comprises one sequence, this can be the index of
the character in the original string.
char_index (`int`, *optional*):
If a batch index is provided in *batch_or_char_index*, this can be the index of the character in the
original string.
sequence_index (`int`, *optional*, defaults to 0):
If a pair of sequences is encoded in the batch this can be used to specify which sequence in the pair (0
or 1) the provided character index belongs to.
Returns:
`int`: Index of the word containing the character.
"""
if not self._encodings:
raise ValueError("char_to_word() is not available when using Python based tokenizers")
if char_index is not None:
batch_index = batch_or_char_index
else:
batch_index = 0
char_index = batch_or_char_index
return self._encodings[batch_index].char_to_word(char_index, sequence_index)
def convert_to_tensors(self, tensor_type: str | TensorType | None = None, prepend_batch_axis: bool = False):
"""
Convert the inner content to tensors.
Args:
tensor_type (`str` or [`~utils.TensorType`], *optional*):
The type of tensors to use. If `str`, should be one of the values of the enum [`~utils.TensorType`]. If
`None`, no modification is done.
prepend_batch_axis (`bool`, *optional*, defaults to `False`):
Whether or not to add the batch dimension during the conversion.
"""
if tensor_type is None:
return self
# Convert to TensorType
if not isinstance(tensor_type, TensorType):
tensor_type = TensorType(tensor_type)
if tensor_type == TensorType.PYTORCH:
if not is_torch_available():
raise ImportError("Unable to convert output to PyTorch tensors format, PyTorch is not installed.")
import torch
def as_tensor(value, dtype=None):
if isinstance(value, list) and len(value) > 0 and isinstance(value[0], np.ndarray):
return torch.from_numpy(np.array(value))
if len(flatten(value)) == 0 and dtype is None:
dtype = torch.int64
return torch.tensor(value, dtype=dtype)
is_tensor = torch.is_tensor
elif tensor_type == TensorType.MLX:
if not is_mlx_available():
raise ImportError("Unable to convert output to MLX tensors format, MLX is not installed.")
import mlx.core as mx
def as_tensor(value, dtype=None):
if len(flatten(value)) == 0 and dtype is None:
dtype = mx.int32
return mx.array(value, dtype=dtype)
def is_tensor(obj):
return isinstance(obj, mx.array)
else:
def as_tensor(value, dtype=None):
if (
isinstance(value, (list, tuple))
and len(value) > 0
and isinstance(value[0], (list, tuple, np.ndarray))
):
value_lens = [len(val) for val in value]
if len(set(value_lens)) > 1 and dtype is None:
# we have a ragged list so handle explicitly
value = as_tensor([np.asarray(val) for val in value], dtype=object)
if len(flatten(value)) == 0 and dtype is None:
dtype = np.int64
return np.asarray(value, dtype=dtype)
is_tensor = is_numpy_array
# Do the tensor conversion in batch
for key, value in self.items():
try:
if prepend_batch_axis:
value = [value]
if not is_tensor(value):
tensor = as_tensor(value)
# Removing this for now in favor of controlling the shape with `prepend_batch_axis`
# # at-least2d
# if tensor.ndim > 2:
# tensor = tensor.squeeze(0)
# elif tensor.ndim < 2:
# tensor = tensor[None, :]
self[key] = tensor
except Exception as e:
if key == "overflowing_tokens":
raise ValueError(
"Unable to create tensor returning overflowing tokens of different lengths. "
"Please see if a fast version of this tokenizer is available to have this feature available."
) from e
raise ValueError(
"Unable to create tensor, you should probably activate truncation and/or padding with"
" 'padding=True' 'truncation=True' to have batched tensors with the same length. Perhaps your"
f" features (`{key}` in this case) have excessive nesting (inputs type `list` where type `int` is"
" expected)."
) from e
return self
def to(self, device: str | torch.device | int, *, non_blocking: bool = False) -> BatchEncoding[torch.Tensor]:
"""
Send all values to device by calling `v.to(device, non_blocking=non_blocking)` (PyTorch only).
Args:
device (`str` or `torch.device` or `int`): The device to put the tensors on.
non_blocking (`bool`): Whether to perform the copy asynchronously.
Returns:
[`BatchEncoding`]: The same instance after modification.
"""
requires_backends(self, ["torch"])
# This check catches things like APEX blindly calling "to" on all inputs to a module
# Otherwise it passes the casts down and casts the LongTensor containing the token idxs
# into a HalfTensor
if isinstance(device, str) or is_torch_device(device) or isinstance(device, int):
self.data = {
k: v.to(device=device, non_blocking=non_blocking) if hasattr(v, "to") and callable(v.to) else v
for k, v in self.data.items()
}
else:
logger.warning(f"Attempting to cast a BatchEncoding to type {str(device)}. This is not supported.")
return self
ENCODE_KWARGS_DOCSTRING = r"""
add_special_tokens (`bool`, *optional*, defaults to `True`):
Whether or not to add special tokens when encoding the sequences. This will use the underlying
`PretrainedTokenizerBase.build_inputs_with_special_tokens` function, which defines which tokens are
automatically added to the input ids. This is useful if you want to add `bos` or `eos` tokens
automatically.
padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `False`):
Activates and controls padding. Accepts the following values:
- `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
sequence is provided).
- `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
acceptable input length for the model if that argument is not provided.
- `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different
lengths).
truncation (`bool`, `str` or [`~tokenization_utils_base.TruncationStrategy`], *optional*, defaults to `False`):
Activates and controls truncation. Accepts the following values:
- `True` or `'longest_first'`: Truncate to a maximum length specified with the argument `max_length` or
to the maximum acceptable input length for the model if that argument is not provided. This will
truncate token by token, removing a token from the longest sequence in the pair if a pair of
sequences (or a batch of pairs) is provided.
- `'only_first'`: Truncate to a maximum length specified with the argument `max_length` or to the
maximum acceptable input length for the model if that argument is not provided. This will only
truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
- `'only_second'`: Truncate to a maximum length specified with the argument `max_length` or to the
maximum acceptable input length for the model if that argument is not provided. This will only
truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
- `False` or `'do_not_truncate'` (default): No truncation (i.e., can output batch with sequence lengths
greater than the model maximum admissible input size).
max_length (`int`, *optional*):
Controls the maximum length to use by one of the truncation/padding parameters.
If left unset or set to `None`, this will use the predefined model maximum length if a maximum length
is required by one of the truncation/padding parameters. If the model has no specific maximum input
length (like XLNet) truncation/padding to a maximum length will be deactivated.
stride (`int`, *optional*, defaults to 0):
If set to a number along with `max_length`, the overflowing tokens returned when
`return_overflowing_tokens=True` will contain some tokens from the end of the truncated sequence
returned to provide some overlap between truncated and overflowing sequences. The value of this
argument defines the number of overlapping tokens.
is_split_into_words (`bool`, *optional*, defaults to `False`):
Whether or not the input is already pre-tokenized (e.g., split into words). If set to `True`, the
tokenizer assumes the input is already split into words (for instance, by splitting it on whitespace)
which it will tokenize. This is useful for NER or token classification.
pad_to_multiple_of (`int`, *optional*):
If set will pad the sequence to a multiple of the provided value. Requires `padding` to be activated.
This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
`>= 7.5` (Volta).
padding_side (`str`, *optional*):
The side on which the model should have padding applied. Should be selected between ['right', 'left'].
Default value is picked from the class attribute of the same name.
return_tensors (`str` or [`~utils.TensorType`], *optional*):
If set, will return tensors instead of list of python integers. Acceptable values are:
- `'pt'`: Return PyTorch `torch.Tensor` objects.
- `'np'`: Return Numpy `np.ndarray` objects.
"""
ENCODE_PLUS_ADDITIONAL_KWARGS_DOCSTRING = r"""
return_token_type_ids (`bool`, *optional*):
Whether to return token type IDs. If left to the default, will return the token type IDs according to
the specific tokenizer's default, defined by the `return_outputs` attribute.
[What are token type IDs?](../glossary#token-type-ids)
return_attention_mask (`bool`, *optional*):
Whether to return the attention mask. If left to the default, will return the attention mask according
to the specific tokenizer's default, defined by the `return_outputs` attribute.
[What are attention masks?](../glossary#attention-mask)
return_overflowing_tokens (`bool`, *optional*, defaults to `False`):
Whether or not to return overflowing token sequences. If a pair of sequences of input ids (or a batch
of pairs) is provided with `truncation_strategy = longest_first` or `True`, an error is raised instead
of returning overflowing tokens.
return_special_tokens_mask (`bool`, *optional*, defaults to `False`):
Whether or not to return special tokens mask information.
return_offsets_mapping (`bool`, *optional*, defaults to `False`):
Whether or not to return `(char_start, char_end)` for each token.
This is only available on fast tokenizers inheriting from [`PreTrainedTokenizerFast`], if using
Python's tokenizer, this method will raise `NotImplementedError`.
return_length (`bool`, *optional*, defaults to `False`):
Whether or not to return the lengths of the encoded inputs.
verbose (`bool`, *optional*, defaults to `True`):
Whether or not to print more information and warnings.
**kwargs: passed to the `self.tokenize()` method
Return:
[`BatchEncoding`]: A [`BatchEncoding`] with the following fields:
- **input_ids** -- List of token ids to be fed to a model.
[What are input IDs?](../glossary#input-ids)
- **token_type_ids** -- List of token type ids to be fed to a model (when `return_token_type_ids=True` or
if *"token_type_ids"* is in `self.model_input_names`).
[What are token type IDs?](../glossary#token-type-ids)
- **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
`return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names`).
[What are attention masks?](../glossary#attention-mask)
- **overflowing_tokens** -- List of overflowing tokens sequences (when a `max_length` is specified and
`return_overflowing_tokens=True`).
- **num_truncated_tokens** -- Number of tokens truncated (when a `max_length` is specified and
`return_overflowing_tokens=True`).
- **special_tokens_mask** -- List of 0s and 1s, with 1 specifying added special tokens and 0 specifying
regular sequence tokens (when `add_special_tokens=True` and `return_special_tokens_mask=True`).
- **length** -- The length of the inputs (when `return_length=True`)
"""
INIT_TOKENIZER_DOCSTRING = r"""
Class attributes (overridden by derived classes)
- **vocab_files_names** (`dict[str, str]`) -- A dictionary with, as keys, the `__init__` keyword name of each
vocabulary file required by the model, and as associated values, the filename for saving the associated file
(string).
- **pretrained_vocab_files_map** (`dict[str, dict[str, str]]`) -- A dictionary of dictionaries, with the
high-level keys being the `__init__` keyword name of each vocabulary file required by the model, the
low-level being the `short-cut-names` of the pretrained models with, as associated values, the `url` to the
associated pretrained vocabulary file.
- **model_input_names** (`list[str]`) -- A list of inputs expected in the forward pass of the model.
- **padding_side** (`str`) -- The default value for the side on which the model should have padding applied.
Should be `'right'` or `'left'`.
- **truncation_side** (`str`) -- The default value for the side on which the model should have truncation
applied. Should be `'right'` or `'left'`.
Args:
model_max_length (`int`, *optional*):
The maximum length (in number of tokens) for the inputs to the transformer model. When the tokenizer is
loaded with [`~tokenization_utils_base.PreTrainedTokenizerBase.from_pretrained`], this will be set to the
value stored for the associated model in `max_model_input_sizes` (see above). If no value is provided, will
default to VERY_LARGE_INTEGER (`int(1e30)`).
padding_side (`str`, *optional*):
The side on which the model should have padding applied. Should be selected between ['right', 'left'].
Default value is picked from the class attribute of the same name.
truncation_side (`str`, *optional*):
The side on which the model should have truncation applied. Should be selected between ['right', 'left'].
Default value is picked from the class attribute of the same name.
chat_template (`str`, *optional*):
A Jinja template string that will be used to format lists of chat messages. See
https://huggingface.co/docs/transformers/chat_templating for a full description.
model_input_names (`list[string]`, *optional*):
The list of inputs accepted by the forward pass of the model (like `"token_type_ids"` or
`"attention_mask"`). Default value is picked from the class attribute of the same name.
bos_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token representing the beginning of a sentence.
eos_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token representing the end of a sentence.
unk_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token representing an out-of-vocabulary token.
sep_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token separating two different sentences in the same input (used by BERT for instance).
pad_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token used to make arrays of tokens the same size for batching purpose. Will then be ignored by
attention mechanisms or loss computation.
cls_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token representing the class of the input (used by BERT for instance).
mask_token (`str` or `tokenizers.AddedToken`, *optional*):
A special token representing a masked token (used by masked-language modeling pretraining objectives, like
BERT). Will be associated to `self.mask_token` and `self.mask_token_id`.
extra_special_tokens (list of `str` or `tokenizers.AddedToken`, *optional*):
A list of extra model-specific special tokens. Add them here to ensure they are skipped when decoding with
`skip_special_tokens` is set to True. If they are not part of the vocabulary, they will be added at the end
of the vocabulary.
split_special_tokens (`bool`, *optional*, defaults to `False`):
Whether or not the special tokens should be split during the tokenization process. Passing will affect the
internal state of the tokenizer. The default behavior is to not split special tokens. This means that if
`<s>` is the `bos_token`, then `tokenizer.tokenize("<s>") = ['<s>`]. Otherwise, if
`split_special_tokens=True`, then `tokenizer.tokenize("<s>")` will be give `['<','s', '>']`.
"""
@add_end_docstrings(INIT_TOKENIZER_DOCSTRING)
class PreTrainedTokenizerBase(PushToHubMixin):
"""
Base class for all tokenizer backends.
"""
vocab_files_names: dict[str, str] = {}
pretrained_vocab_files_map: dict[str, dict[str, str]] = {}
_auto_class: str | None = None
# first name has to correspond to main model input name
# to make sure `tokenizer.pad(...)` works correctly
model_input_names: list[str] = ["input_ids", "attention_mask"]
padding_side: str = "right"
truncation_side: str = "right"
slow_tokenizer_class = None
# Special tokens support (moved from SpecialTokensMixin)
# V5: Clean separation of named special tokens from extra special tokens
SPECIAL_TOKENS_ATTRIBUTES = [
"bos_token",
"eos_token",
"unk_token",
"sep_token",
"pad_token",
"cls_token",
"mask_token",
]
def __init__(self, **kwargs):
self.init_inputs = ()
for key in kwargs:
if hasattr(self, key) and callable(getattr(self, key)):
raise AttributeError(f"{key} conflicts with the method {key} in {self.__class__.__name__}")
# V5: Convert deprecated additional_special_tokens to extra_special_tokens before storing init_kwargs
if "additional_special_tokens" in kwargs and "extra_special_tokens" not in kwargs:
kwargs["extra_special_tokens"] = kwargs.pop("additional_special_tokens")
self.init_kwargs = copy.deepcopy(kwargs)