File size: 10,266 Bytes
556467f |
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 |
from abc import ABC
from abc import abstractmethod
class AbstractTokenizer(ABC):
"""Abstract class for tokenizer."""
def __init__(self, name):
self.name = name
super().__init__()
@property
@abstractmethod
def vocab_size(self):
pass
@property
@abstractmethod
def vocab(self):
"""Dictionary from vocab text token to id token."""
pass
@property
@abstractmethod
def inv_vocab(self):
"""Dictionary from vocab id token to text token."""
pass
@abstractmethod
def tokenize(self, text):
pass
def detokenize(self, token_ids):
raise NotImplementedError('detokenizer is not implemented for {} '
'tokenizer'.format(self.name))
@property
def cls(self):
raise NotImplementedError('CLS is not provided for {} '
'tokenizer'.format(self.name))
@property
def sep(self):
raise NotImplementedError('SEP is not provided for {} '
'tokenizer'.format(self.name))
@property
def pad(self):
raise NotImplementedError('PAD is not provided for {} '
'tokenizer'.format(self.name))
@property
def eod(self):
raise NotImplementedError('EOD is not provided for {} '
'tokenizer'.format(self.name))
@property
def mask(self):
raise NotImplementedError('MASK is not provided for {} '
'tokenizer'.format(self.name))
class _SentencePieceTokenizer(AbstractTokenizer):
"""SentencePieceTokenizer-Megatron wrapper"""
def __init__(self, model_file, vocab_extra_ids=0):
name = 'SentencePieceTokenizer'
super().__init__(name)
import sentencepiece
self.tokenizer = sentencepiece.SentencePieceProcessor(model_file=model_file)
self._initalize(vocab_extra_ids)
def _populate_vocab(self):
self._vocab = {}
self._inv_vocab = {}
for i in range(len(self.tokenizer)):
t = self.tokenizer.id_to_piece(i)
self._inv_vocab[i] = t
self._vocab[t] = i
def _initalize(self, vocab_extra_ids):
self._populate_vocab()
self._special_tokens = {}
self._inv_special_tokens = {}
self._t5_tokens = []
def _add_special_token(t):
if t not in self._vocab:
next_id = len(self._vocab)
self._vocab[t] = next_id
self._inv_vocab[next_id] = t
self._special_tokens[t] = self._vocab[t]
self._inv_special_tokens[self._vocab[t]] = t
_add_special_token('<CLS>')
self._cls_id = self._vocab['<CLS>']
_add_special_token('<SEP>')
self._sep_id = self._vocab['<SEP>']
_add_special_token('<EOD>')
self._eod_id = self._vocab['<EOD>']
_add_special_token('<MASK>')
self._mask_id = self._vocab['<MASK>']
pad_id = self.tokenizer.pad_id()
try:
pad_token = self.tokenizer.id_to_piece(pad_id)
except IndexError:
pad_token = '<PAD>'
_add_special_token(pad_token)
self._pad_id = self._vocab[pad_token]
bos_id = self.tokenizer.bos_id()
try:
bos_token = self.tokenizer.id_to_piece(bos_id)
except IndexError:
bos_token = '<BOS>'
_add_special_token(bos_token)
self._bos_id = self._vocab[bos_token]
eos_id = self.tokenizer.eos_id()
try:
eos_token = self.tokenizer.id_to_piece(eos_id)
except IndexError:
eos_token = '<EOS>'
_add_special_token(eos_token)
self._eos_id = self._vocab[eos_token]
for i in range(vocab_extra_ids):
t = "<extra_id_{}>".format(i)
_add_special_token(t)
self._t5_tokens += [t]
@property
def vocab_size(self):
return len(self._vocab)
@property
def vocab(self):
return self._vocab
@property
def inv_vocab(self):
return self._inv_vocab
@property
def decoder(self):
return self._inv_vocab
@property
def encoder(self):
return self._vocab
# From:
# https://github.com/NVIDIA/NeMo/blob/c8fa217e811d60d11d014827c7f3845ff6c99ae7/nemo/collections/common/tokenizers/sentencepiece_tokenizer.py#L89
def tokenize(self, text):
ids = []
idx = 0
while 1:
indices = {}
for token in self._special_tokens:
try:
indices[token] = text[idx:].index(token)
except ValueError:
continue
if len(indices) == 0:
break
next_token = min(indices, key=indices.get)
next_idx = idx + indices[next_token]
ids.extend(self.tokenizer.encode_as_ids(text[idx:next_idx]))
ids.append(self._special_tokens[next_token])
idx = next_idx + len(next_token)
ids.extend(self.tokenizer.encode_as_ids(text[idx:]))
return ids
# From:
# https://github.com/NVIDIA/NeMo/blob/c8fa217e811d60d11d014827c7f3845ff6c99ae7/nemo/collections/common/tokenizers/sentencepiece_tokenizer.py#L125
def detokenize(self, ids):
text = ""
last_i = 0
for i, id in enumerate(ids):
if id in self._inv_special_tokens:
text += self.tokenizer.decode_ids(ids[last_i:i]) + " "
text += self._inv_special_tokens[id] + " "
last_i = i + 1
text += self.tokenizer.decode_ids(ids[last_i:])
return text
@property
def cls(self):
return self._cls_id
@property
def sep(self):
return self._sep_id
@property
def pad(self):
return self._pad_id
@property
def bos_token_id(self):
return self._bos_id
@property
def bos(self):
return self._bos_id
@property
def eod(self):
return self._eod_id
@property
def eos_token_id(self):
return self._eos_id
@property
def eos(self):
return self._eos_id
@property
def mask(self):
return self._mask_id
@property
def additional_special_tokens_ids(self):
return [self.vocab[k] for k in self._t5_tokens]
class _MMSentencePieceTokenizer(_SentencePieceTokenizer):
"""SentencePieceTokenizer-Megatron wrapper"""
def __init__(self, model_file, vocab_extra_ids=0):
super().__init__(model_file, vocab_extra_ids)
def _initalize(self, vocab_extra_ids):
self._populate_vocab()
self._special_tokens = {}
self._inv_special_tokens = {}
self._t5_tokens = []
def _add_special_token(t):
if t not in self._vocab:
next_id = len(self._vocab)
self._vocab[t] = next_id
self._inv_vocab[next_id] = t
self._special_tokens[t] = self._vocab[t]
self._inv_special_tokens[self._vocab[t]] = t
_add_special_token('<CLS>')
self._cls_id = self._vocab['<CLS>']
_add_special_token('<SEP>')
self._sep_id = self._vocab['<SEP>']
_add_special_token('<EOD>')
self._eod_id = self._vocab['<EOD>']
_add_special_token('<MASK>')
self._mask_id = self._vocab['<MASK>']
_add_special_token('<SOA>')
self._soa_id = self._vocab['<SOA>']
_add_special_token('<EOA>')
self._eoa_id = self._vocab['<EOA>']
_add_special_token('<SOV>')
self._sov_id = self._vocab['<SOV>']
_add_special_token('<EOV>')
self._eov_id = self._vocab['<EOV>']
_add_special_token('<SOI>')
self._soi_id = self._vocab['<SOI>']
_add_special_token('<EOI>')
self._eoi_id = self._vocab['<EOI>']
_add_special_token('<s_local>')
self._s_local_id = self._vocab['<s_local>']
_add_special_token('<e_local>')
self._e_local_id = self._vocab['<e_local>']
_add_special_token('<s_global>')
self._s_global_id = self._vocab['<s_global>']
_add_special_token('<e_global>')
self._e_global_id = self._vocab['<e_global>']
_add_special_token('<stage_1>')
self._stage_1_id = self._vocab['<stage_1>']
_add_special_token('<stage_2>')
self._stage_2_id = self._vocab['<stage_2>']
pad_id = self.tokenizer.pad_id()
try:
pad_token = self.tokenizer.id_to_piece(pad_id)
except IndexError:
pad_token = '<PAD>'
_add_special_token(pad_token)
self._pad_id = self._vocab[pad_token]
bos_id = self.tokenizer.bos_id()
try:
bos_token = self.tokenizer.id_to_piece(bos_id)
except IndexError:
bos_token = '<BOS>'
_add_special_token(bos_token)
self._bos_id = self._vocab[bos_token]
eos_id = self.tokenizer.eos_id()
try:
eos_token = self.tokenizer.id_to_piece(eos_id)
except IndexError:
eos_token = '<EOS>'
_add_special_token(eos_token)
self._eos_id = self._vocab[eos_token]
for i in range(vocab_extra_ids):
t = "<extra_id_{}>".format(i)
_add_special_token(t)
self._t5_tokens += [t]
@property
def soa(self):
return self._soa_id
@property
def eoa(self):
return self._eoa_id
@property
def sov(self):
return self._sov_id
@property
def eov(self):
return self._eov_id
@property
def soi(self):
return self._soi_id
@property
def eoi(self):
return self._eoi_id
@property
def s_local(self):
return self._s_local_id
@property
def e_local(self):
return self._e_local_id
@property
def s_global(self):
return self._s_global_id
@property
def e_global(self):
return self._e_global_id
@property
def stage_1(self):
return self._stage_1_id
@property
def stage_2(self):
return self._stage_2_id
|