mkt3 commited on
Commit
2110e9c
1 Parent(s): c361cda

Add models

Browse files
README.md CHANGED
@@ -1,3 +1,128 @@
1
  ---
 
 
 
 
 
 
 
2
  license: mit
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language: ja
3
+ tags:
4
+ - ja
5
+ - japanese
6
+ - bart
7
+ - lm
8
+ - nlp
9
  license: mit
10
  ---
11
+
12
+ # bart-base-japanese-news(base-sized model)
13
+ This repository provides a Japanese BART model. The model was trained by [Stockmark Inc.](https://stockmark.co.jp)
14
+
15
+
16
+ ## Model description
17
+
18
+ BART is a transformer encoder-decoder (seq2seq) model with a bidirectional (BERT-like) encoder and an autoregressive (GPT-like) decoder. BART is pre-trained by (1) corrupting text with an arbitrary noising function, and (2) learning a model to reconstruct the original text.
19
+
20
+ BART is particularly effective when fine-tuned for text generation (e.g. summarization, translation) but also works well for comprehension tasks (e.g. text classification, question answering).
21
+
22
+ ## Intended uses & limitations
23
+
24
+ You can use the raw model for text infilling. However, the model is mostly meant to be fine-tuned on a supervised dataset.
25
+
26
+ # How to use the model
27
+
28
+ *NOTE:* Use `trust_remote_code=True` to initiate the tokenizer.
29
+
30
+ ## Simple use
31
+
32
+ ```python
33
+ from transformers import AutoTokenizer, BartModel
34
+
35
+ model_name = "stockmark/bart-base-japanese-news"
36
+
37
+ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
38
+ model = BartModel.from_pretrained(model_name)
39
+
40
+ inputs = tokenizer("今日は良い天気です。", return_tensors="pt")
41
+ outputs = model(**inputs)
42
+
43
+ last_hidden_states = outputs.last_hidden_state
44
+ ```
45
+
46
+ ## Sentence Permutation
47
+ ```python
48
+ import torch
49
+ from transformers import AutoTokenizer, BartForConditionalGeneration
50
+
51
+ model_name = "stockmark/bart-base-japanese-news"
52
+
53
+ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
54
+ model = BartForConditionalGeneration.from_pretrained(model_name)
55
+
56
+ if torch.cuda.is_available():
57
+ model = model.to("cuda")
58
+
59
+ # correct order text is "明日は大雨です。電車は止まる可能性があります。ですから、自宅から働きます。"
60
+ text = "電車は止まる可能性があります。ですから、自宅から働きます。明日は大雨です。"
61
+
62
+ inputs = tokenizer([text], max_length=128, return_tensors="pt", truncation=True)
63
+ text_ids = model.generate(inputs["input_ids"].to(model.device), num_beams=3, max_length=128)
64
+ output = tokenizer.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
65
+
66
+ print(output)
67
+ # sample output: 明日は大雨です。電車は止まる可能性があります。ですから、自宅から働きます。
68
+ ```
69
+ ## Mask filing
70
+ ```python
71
+ import torch
72
+ from transformers import AutoTokenizer, BartForConditionalGeneration
73
+
74
+ model_name = "stockmark/bart-base-japanese-news"
75
+
76
+ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
77
+ model = BartForConditionalGeneration.from_pretrained(model_name)
78
+
79
+ if torch.cuda.is_available():
80
+ model = model.to("cuda")
81
+
82
+ text = "今日の天気は<mask>のため、傘が必要でしょう。"
83
+
84
+ inputs = tokenizer([text], max_length=128, return_tensors="pt", truncation=True)
85
+ text_ids = model.generate(inputs["input_ids"].to(model.device), num_beams=3, max_length=128)
86
+ output = tokenizer.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
87
+
88
+ print(output)
89
+ # sample output: 今日の天気は、雨のため、傘が必要でしょう。
90
+ ```
91
+
92
+ ## Text generation
93
+
94
+ *NOTE:* You can use the raw model for text generation. However, the model is mostly meant to be fine-tuned on a supervised dataset.
95
+
96
+ ```python
97
+ import torch
98
+ from transformers import AutoTokenizer, BartForConditionalGeneration
99
+
100
+ model_name = "stockmark/bart-base-japanese-news"
101
+
102
+ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
103
+ model = BartForConditionalGeneration.from_pretrained(model_name)
104
+
105
+ if torch.cuda.is_available():
106
+ model = model.to("cuda")
107
+
108
+ text = "自然言語処理(しぜんげんごしょり、略称:NLP)は、人間が日常的に使っている自然言語をコンピュータに処理させる一連の技術であり、人工知能と言語学の一分野である。「計算言語学」(computational linguistics)との類似もあるが、自然言語処理は工学的な視点からの言語処理をさすのに対して、計算言語学は言語学的視点を重視する手法をさす事が多い。"
109
+
110
+ inputs = tokenizer([text], max_length=512, return_tensors="pt", truncation=True)
111
+ text_ids = model.generate(inputs["input_ids"].to(model.device), num_beams=3, min_length=0, max_length=40)
112
+ output = tokenizer.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
113
+
114
+ print(output)
115
+ # sample output: 自然言語処理(しぜんげんごしょり、略称:NLP)は、人間が日常的に使っている自然言語をコンピュータに処理させる一連の技術であり、言語学の一分野である。
116
+ ```
117
+
118
+ # Training
119
+ The model was trained on Japanese News Articles.
120
+
121
+ # Tokenization
122
+ The model uses a [sentencepiece](https://github.com/google/sentencepiece)-based tokenizer. The vocabulary was first trained on a selected subset from the training data using the official sentencepiece training script.
123
+
124
+ # Licenses
125
+ The pretrained models are distributed under the terms of the [MIT License](https://opensource.org/licenses/mit-license.php).
126
+
127
+ # Acknowledgement
128
+ This comparison study supported with Cloud TPUs from Google’s TensorFlow Research Cloud (TFRC).
config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "bart-base-japanese-news",
3
+ "activation_dropout": 0.1,
4
+ "activation_function": "gelu",
5
+ "architectures": [
6
+ "BartForConditionalGeneration"
7
+ ],
8
+ "attention_dropout": 0.1,
9
+ "bos_token_id": 1,
10
+ "classifier_dropout": 0.0,
11
+ "d_model": 768,
12
+ "decoder_attention_heads": 12,
13
+ "decoder_ffn_dim": 3072,
14
+ "decoder_layerdrop": 0.0,
15
+ "decoder_layers": 6,
16
+ "decoder_start_token_id": 1,
17
+ "dropout": 0.1,
18
+ "encoder_attention_heads": 12,
19
+ "encoder_ffn_dim": 3072,
20
+ "encoder_layerdrop": 0.0,
21
+ "encoder_layers": 6,
22
+ "eos_token_id": 2,
23
+ "forced_eos_token_id": 2,
24
+ "gradient_checkpointing": false,
25
+ "id2label": {
26
+ "0": "LABEL_0",
27
+ "1": "LABEL_1",
28
+ "2": "LABEL_2"
29
+ },
30
+ "init_std": 0.02,
31
+ "is_encoder_decoder": true,
32
+ "label2id": {
33
+ "LABEL_0": 0,
34
+ "LABEL_1": 1,
35
+ "LABEL_2": 2
36
+ },
37
+ "max_position_embeddings": 512,
38
+ "model_type": "bart",
39
+ "num_hidden_layers": 6,
40
+ "pad_token_id": 3,
41
+ "scale_embedding": false,
42
+ "torch_dtype": "float32",
43
+ "transformers_version": "4.21.3",
44
+ "use_cache": false,
45
+ "vocab_size": 32000
46
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73278760f95038d267440a1c703247ad8c19ba0a4a1dbb41923e046ab6a2df19
3
+ size 498636537
special_tokens_map.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": "<s>",
3
+ "cls_token": "<s>",
4
+ "eos_token": "</s>",
5
+ "mask_token": {
6
+ "content": "<mask>",
7
+ "lstrip": true,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false
11
+ },
12
+ "pad_token": "<pad>",
13
+ "sep_token": "</s>",
14
+ "unk_token": "<unk>"
15
+ }
spiece.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ef4fcd7e2f55b4d3601f0827ca79efdfc5bfea11129c554e0991491f73ba7a0
3
+ size 828378
tf_model.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2c1df518fdd0baa436241ec54e81c21eef9af3a55d5de6af86a7f6d9d6c09ff
3
+ size 498843504
tokenization_bart_japanese_news.py ADDED
@@ -0,0 +1,294 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2023 Stockmark Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+
16
+ import os
17
+ import unicodedata
18
+ from shutil import copyfile
19
+ from typing import Any, Dict, List, Optional, Tuple
20
+
21
+ import regex
22
+ import sentencepiece as spm
23
+ from transformers import AddedToken, BartTokenizer, logging
24
+
25
+ logger = logging.get_logger(__name__)
26
+ VOCAB_FILES_NAMES = {"vocab_file": "spiece.model"}
27
+
28
+ PRETRAINED_VOCAB_FILES_MAP = {
29
+ "vocab_file": {
30
+ "bart-base-japanese-news": "https://huggingface.co/stockmark/bart-base-japanese-news/resolve/main/spiece.model",
31
+ }}
32
+
33
+ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
34
+ "bart-base-japanese-news": 512,
35
+ }
36
+
37
+ SPIECE_UNDERLINE = "▁"
38
+
39
+
40
+ class BartJapaneseNewsTokenizer(BartTokenizer):
41
+ """
42
+ Construct an Bart tokenizer. Based on [SentencePiece](https://github.com/google/sentencepiece).
43
+
44
+ This tokenizer inherits from [`BartTokenizer`] which contains most of the main methods. Users should refer to
45
+ this superclass for more information regarding those methods.
46
+
47
+ Args:
48
+ vocab_file (`str`):
49
+ [SentencePiece](https://github.com/google/sentencepiece) file (generally has a *.spm* extension) that
50
+ contains the vocabulary necessary to instantiate a tokenizer.
51
+ do_lower_case (`bool`, *optional*, defaults to `False`):
52
+ Whether or not to lowercase the input when tokenizing.
53
+ remove_space (`bool`, *optional*, defaults to `False`):
54
+ Whether or not to strip the text when tokenizing (removing excess spaces before and after the string).
55
+ clean_text (`bool`, *optional*, defaults to `False`):
56
+ Whether or not to clean input text
57
+
58
+ bos_token (`str`, *optional*, defaults to `"<s>"`):
59
+ The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
60
+
61
+ <Tip>
62
+
63
+ When building a sequence using special tokens, this is not the token that is used for the beginning of
64
+ sequence. The token used is the `cls_token`.
65
+
66
+ </Tip>
67
+
68
+ eos_token (`str`, *optional*, defaults to `"</s>"`):
69
+ The end of sequence token.
70
+
71
+ <Tip>
72
+
73
+ When building a sequence using special tokens, this is not the token that is used for the end of sequence.
74
+ The token used is the `sep_token`.
75
+
76
+ </Tip>
77
+
78
+ sep_token (`str`, *optional*, defaults to `"</s>"`):
79
+ The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
80
+ sequence classification or for a text and a question for question answering. It is also used as the last
81
+ token of a sequence built with special tokens.
82
+ cls_token (`str`, *optional*, defaults to `"<s>"`):
83
+ The classifier token which is used when doing sequence classification (classification of the whole sequence
84
+ instead of per-token classification). It is the first token of the sequence when built with special tokens.
85
+ unk_token (`str`, *optional*, defaults to `"<unk>"`):
86
+ The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
87
+ token instead.
88
+ pad_token (`str`, *optional*, defaults to `"<pad>"`):
89
+ The token used for padding, for example when batching sequences of different lengths.
90
+ mask_token (`str`, *optional*, defaults to `"<mask>"`):
91
+ The token used for masking values. This is the token used when training this model with masked language
92
+ modeling. This is the token which the model will try to predict.
93
+
94
+ sp_model_kwargs (`dict`, *optional*):
95
+ Will be passed to the `SentencePieceProcessor.__init__()` method. The [Python wrapper for
96
+ SentencePiece](https://github.com/google/sentencepiece/tree/master/python) can be used, among other things,
97
+ to set:
98
+
99
+ - `enable_sampling`: Enable subword regularization.
100
+ - `nbest_size`: Sampling parameters for unigram. Invalid for BPE-Dropout.
101
+
102
+ - `nbest_size = {0,1}`: No sampling is performed.
103
+ - `nbest_size > 1`: samples from the nbest_size results.
104
+ - `nbest_size < 0`: assuming that nbest_size is infinite and samples from the all hypothesis (lattice)
105
+ using forward-filtering-and-backward-sampling algorithm.
106
+
107
+ - `alpha`: Smoothing parameter for unigram sampling, and dropout probability of merge operations for
108
+ BPE-dropout.
109
+
110
+ Attributes:
111
+ sp_model (`SentencePieceProcessor`):
112
+ The *SentencePiece* processor that is used for every conversion (string, tokens and IDs).
113
+ """
114
+
115
+ vocab_files_names = VOCAB_FILES_NAMES
116
+ pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
117
+ max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
118
+
119
+ def __init__(
120
+ self,
121
+ vocab_file,
122
+ do_lower_case=False,
123
+ remove_space=False,
124
+ clean_text=False,
125
+ bos_token="<s>",
126
+ eos_token="</s>",
127
+ unk_token="<unk>",
128
+ sep_token="</s>",
129
+ pad_token="<pad>",
130
+ cls_token="<s>",
131
+ mask_token="<mask>",
132
+ sp_model_kwargs: Optional[Dict[str, Any]] = None,
133
+ **kwargs
134
+ ) -> None:
135
+ # Mask token behave like a normal word, i.e. include the space before it and
136
+ # is included in the raw text, there should be a match in a non-normalized sentence.
137
+ mask_token = (
138
+ AddedToken(mask_token, lstrip=True, rstrip=True, normalized=False)
139
+ if isinstance(mask_token, str)
140
+ else mask_token
141
+ )
142
+
143
+ self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
144
+
145
+ super(BartTokenizer, self).__init__(
146
+ do_lower_case=do_lower_case,
147
+ remove_space=remove_space,
148
+ clean_text=clean_text,
149
+ bos_token=bos_token,
150
+ eos_token=eos_token,
151
+ unk_token=unk_token,
152
+ sep_token=sep_token,
153
+ pad_token=pad_token,
154
+ cls_token=cls_token,
155
+ mask_token=mask_token,
156
+ sp_model_kwargs=self.sp_model_kwargs,
157
+ **kwargs,
158
+ )
159
+
160
+ self.do_lower_case = do_lower_case
161
+ self.remove_space = remove_space
162
+ self.clean_text = clean_text
163
+ self.vocab_file = vocab_file
164
+
165
+ self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
166
+ self.sp_model.Load(vocab_file)
167
+
168
+ @property
169
+ def vocab_size(self):
170
+ return len(self.sp_model)
171
+
172
+ def get_vocab(self):
173
+ vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
174
+ vocab.update(self.added_tokens_encoder)
175
+ return vocab
176
+
177
+ def __getstate__(self):
178
+ state = self.__dict__.copy()
179
+ state["sp_model"] = None
180
+ return state
181
+
182
+ def __setstate__(self, d):
183
+ self.__dict__ = d
184
+
185
+ # for backward compatibility
186
+ if not hasattr(self, "sp_model_kwargs"):
187
+ self.sp_model_kwargs = {}
188
+
189
+ self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
190
+ self.sp_model.Load(self.vocab_file)
191
+
192
+ def preprocess_text(self, inputs):
193
+ if self.remove_space:
194
+ outputs = " ".join(inputs.strip().split())
195
+ else:
196
+ outputs = inputs
197
+
198
+ outputs = unicodedata.normalize("NFKD", outputs)
199
+ outputs = ''.join([ s for s in outputs if not self._is_control(s) ])
200
+
201
+ if self.clean_text:
202
+ outputs = outputs.replace('‘','\'').replace('’','\'').replace('“','"').replace('”','"')
203
+ outputs = outputs.replace('〈','<').replace('〉','>').replace('《','<').replace('》','>').replace('〔','【').replace('〕','】').replace('『','「').replace('』','」')
204
+ outputs = regex.sub(r'[\p{GeometricShapes}\p{MiscellaneousSymbols}]','*', outputs)
205
+ outputs = ''.join(regex.findall(r'[\p{InHiragana}\p{InKatakana}\p{BasicLatin}\p{Han}、。〃〆「」【】〒〜]',outputs))
206
+
207
+ if self.do_lower_case:
208
+ outputs = outputs.lower()
209
+
210
+ outputs = unicodedata.normalize('NFKC',outputs)
211
+ outputs = outputs.strip()
212
+ return outputs
213
+
214
+ def _tokenize(self, text: str) -> List[str]:
215
+ """Tokenize a string."""
216
+ text = self.preprocess_text(text)
217
+ pieces = self.sp_model.encode(text, out_type=str)
218
+ new_pieces = []
219
+ for piece in pieces:
220
+ if len(piece) > 1 and piece[-1] == str(",") and piece[-2].isdigit():
221
+ cur_pieces = self.sp_model.EncodeAsPieces(piece[:-1].replace(SPIECE_UNDERLINE, ""))
222
+ if piece[0] != SPIECE_UNDERLINE and cur_pieces[0][0] == SPIECE_UNDERLINE:
223
+ if len(cur_pieces[0]) == 1:
224
+ cur_pieces = cur_pieces[1:]
225
+ else:
226
+ cur_pieces[0] = cur_pieces[0][1:]
227
+ cur_pieces.append(piece[-1])
228
+ new_pieces.extend(cur_pieces)
229
+ else:
230
+ new_pieces.append(piece)
231
+
232
+ return new_pieces
233
+
234
+ def _convert_token_to_id(self, token):
235
+ """Converts a token (str) in an id using the vocab."""
236
+ return self.sp_model.PieceToId(token)
237
+
238
+ def _convert_id_to_token(self, index):
239
+ """Converts an index (integer) in a token (str) using the vocab."""
240
+ return self.sp_model.IdToPiece(index)
241
+
242
+ def convert_tokens_to_string(self, tokens):
243
+ """Converts a sequence of tokens (string) in a single string."""
244
+ current_sub_tokens = []
245
+ out_string = ""
246
+ prev_is_special = False
247
+ for token in tokens:
248
+ # make sure that special tokens are not decoded using sentencepiece model
249
+ if token in self.all_special_tokens:
250
+ if not prev_is_special:
251
+ out_string += " "
252
+ out_string += self.sp_model.decode(current_sub_tokens) + token
253
+ prev_is_special = True
254
+ current_sub_tokens = []
255
+ else:
256
+ current_sub_tokens.append(token)
257
+ prev_is_special = False
258
+ out_string += self.sp_model.decode(current_sub_tokens)
259
+ return out_string.strip()
260
+
261
+ def _is_control(self, char):
262
+ '''
263
+ Check control char
264
+ Args:
265
+ char (str):
266
+ Returns:
267
+ bool:
268
+ '''
269
+ if char == "\t" or char == "\n" or char == "\r":
270
+ return False
271
+ cat = unicodedata.category(char)
272
+ if cat.startswith("C") or ord(char)==0 or ord(char)==0xfffd:
273
+ return True
274
+ return False
275
+
276
+ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
277
+ if not os.path.isdir(save_directory):
278
+ logger.error(f"Vocabulary path ({save_directory}) should be a directory")
279
+ return
280
+ out_vocab_file = os.path.join(
281
+ save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
282
+ )
283
+
284
+ if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
285
+ copyfile(self.vocab_file, out_vocab_file)
286
+ elif not os.path.isfile(self.vocab_file):
287
+ with open(out_vocab_file, "wb") as fi:
288
+ content_spiece_model = self.sp_model.serialized_model_proto()
289
+ fi.write(content_spiece_model)
290
+
291
+ return (out_vocab_file,)
292
+
293
+ def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):
294
+ return super(BartTokenizer, self).prepare_for_tokenization(text, is_split_into_words=False, **kwargs)
tokenizer_config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": "<s>",
3
+ "clean_text": false,
4
+ "cls_token": "<s>",
5
+ "do_lower_case": false,
6
+ "eos_token": "</s>",
7
+ "mask_token": {
8
+ "__type": "AddedToken",
9
+ "content": "<mask>",
10
+ "lstrip": true,
11
+ "normalized": false,
12
+ "rstrip": false,
13
+ "single_word": false
14
+ },
15
+ "pad_token": "<pad>",
16
+ "remove_space": false,
17
+ "sep_token": "</s>",
18
+ "sp_model_kwargs": {},
19
+ "tokenizer_class": "BartJapaneseNewsTokenizer",
20
+ "unk_token": "<unk>",
21
+ "auto_map": {
22
+ "AutoTokenizer": [
23
+ "tokenization_bart_japanese_news.BartJapaneseNewsTokenizer",
24
+ null
25
+ ]
26
+ }
27
+ }