SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add mBART-50 (#10154)

Browse Source 
* add tokenizer for mBART-50

* update tokenizers

* make src_lang and tgt_lang optional

* update tokenizer test

* add setter

* update docs

* update conversion script

* update docs

* update conversion script

* update tokenizer

* update test

* update docs

* doc

* address Sylvain's suggestions

* fix test

* fix formatting

* nits
This commit is contained in:
Suraj Patil
2021-02-15 20:58:54 +05:30
committed by GitHub
parent 570218878a
commit 6fc940ed09
13 changed files with 1008 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/transformers/models/mbart/__init__.py
View File

@@ -32,9 +32,11 @@ _import_structure = {
if is_sentencepiece_available():
_import_structure["tokenization_mbart"] = ["MBartTokenizer"]
_import_structure["tokenization_mbart50"] = ["MBart50Tokenizer"]
if is_tokenizers_available():
_import_structure["tokenization_mbart_fast"] = ["MBartTokenizerFast"]
_import_structure["tokenization_mbart50_fast"] = ["MBart50TokenizerFast"]
if is_torch_available():
_import_structure["modeling_mbart"] = [
@@ -56,8 +58,10 @@ if TYPE_CHECKING:
if is_sentencepiece_available():
from .tokenization_mbart import MBartTokenizer
from .tokenization_mbart50 import MBart50Tokenizer
if is_tokenizers_available():
from .tokenization_mbart50_fast import MBart50TokenizerFast
from .tokenization_mbart_fast import MBartTokenizerFast
if is_torch_available():

46
src/transformers/models/mbart/convert_mbart_original_checkpoint_to_pytorch.py
View File

@@ -15,19 +15,49 @@
import argparse
import torch
from torch import nn
from transformers import BartForConditionalGeneration, MBartConfig
from transformers.models.bart.convert_bart_original_pytorch_checkpoint_to_pytorch import remove_ignore_keys_
from transformers import MBartConfig, MBartForConditionalGeneration
def convert_fairseq_mbart_checkpoint_from_disk(checkpoint_path, hf_config_path="facebook/mbart-large-en-ro"):
def remove_ignore_keys_(state_dict):
ignore_keys = [
"encoder.version",
"decoder.version",
"model.encoder.version",
"model.decoder.version",
"_float_tensor",
"decoder.output_projection.weight",
]
for k in ignore_keys:
state_dict.pop(k, None)
def make_linear_from_emb(emb):
vocab_size, emb_size = emb.weight.shape
lin_layer = nn.Linear(vocab_size, emb_size, bias=False)
lin_layer.weight.data = emb.weight.data
return lin_layer
def convert_fairseq_mbart_checkpoint_from_disk(
checkpoint_path, hf_config_path="facebook/mbart-large-en-ro", finetuned=False, mbart_50=False
):
state_dict = torch.load(checkpoint_path, map_location="cpu")["model"]
remove_ignore_keys_(state_dict)
vocab_size = state_dict["encoder.embed_tokens.weight"].shape[0]
mbart_config = MBartConfig.from_pretrained(hf_config_path, vocab_size=vocab_size)
if mbart_50 and finetuned:
mbart_config.activation_function = "relu"
state_dict["shared.weight"] = state_dict["decoder.embed_tokens.weight"]
model = BartForConditionalGeneration(mbart_config)
model = MBartForConditionalGeneration(mbart_config)
model.model.load_state_dict(state_dict)
if finetuned:
model.lm_head = make_linear_from_emb(model.model.shared)
return model
@@ -42,8 +72,12 @@ if __name__ == "__main__":
"--hf_config",
default="facebook/mbart-large-cc25",
type=str,
help="Which huggingface architecture to use: bart-large-xsum",
help="Which huggingface architecture to use: mbart-large",
)
parser.add_argument("--mbart_50", action="store_true", help="whether the model is mMART-50 checkpoint")
parser.add_argument("--finetuned", action="store_true", help="whether the model is a fine-tuned checkpoint")
args = parser.parse_args()
model = convert_fairseq_mbart_checkpoint_from_disk(args.fairseq_path, hf_config_path=args.hf_config)
model = convert_fairseq_mbart_checkpoint_from_disk(
args.fairseq_path, hf_config_path=args.hf_config, finetuned=args.finetuned, mbart_50=args.mbart_50
)
model.save_pretrained(args.pytorch_dump_folder_path)

308
src/transformers/models/mbart/tokenization_mbart50.py Normal file
View File

@@ -0,0 +1,308 @@
# coding=utf-8
# Copyright 2021 The Facebook AI Research Team Authors and 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.
import os
from contextlib import contextmanager
from shutil import copyfile
from typing import Dict, List, Optional, Tuple
import sentencepiece as spm
from ...tokenization_utils import AddedToken, BatchEncoding, PreTrainedTokenizer
from ...utils import logging
logger = logging.get_logger(__name__)
SPIECE_UNDERLINE = "▁"
VOCAB_FILES_NAMES = {"vocab_file": "sentencepiece.bpe.model"}
_all_mbart50_models = ["facebook/mbart-large-50-one-to-many-mmt"]
SPM_URL = "https://huggingface.co/facebook/mbart-large-50-one-to-many-mmt/resolve/main/sentencepiece.bpe.model"
# fmt: off
FAIRSEQ_LANGUAGE_CODES = ["ar_AR", "cs_CZ", "de_DE", "en_XX", "es_XX", "et_EE", "fi_FI", "fr_XX", "gu_IN", "hi_IN", "it_IT", "ja_XX", "kk_KZ", "ko_KR", "lt_LT", "lv_LV", "my_MM", "ne_NP", "nl_XX", "ro_RO", "ru_RU", "si_LK", "tr_TR", "vi_VN", "zh_CN", "af_ZA", "az_AZ", "bn_IN", "fa_IR", "he_IL", "hr_HR", "id_ID", "ka_GE", "km_KH", "mk_MK", "ml_IN", "mn_MN", "mr_IN", "pl_PL", "ps_AF", "pt_XX", "sv_SE", "sw_KE", "ta_IN", "te_IN", "th_TH", "tl_XX", "uk_UA", "ur_PK", "xh_ZA", "gl_ES", "sl_SI"]
# fmt: on
class MBart50Tokenizer(PreTrainedTokenizer):
"""
Construct a MBart50 tokenizer. Based on `SentencePiece <https://github.com/google/sentencepiece>`__.
This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains most of the main methods.
Users should refer to this superclass for more information regarding those methods.
Args:
vocab_file (:obj:`str`):
Path to the vocabulary file.
src_lang (:obj:`str`, `optional`):
A string representing the source language.
tgt_lang (:obj:`str`, `optional`):
A string representing the target language.
eos_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
The end of sequence token.
sep_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
sequence classification or for a text and a question for question answering. It is also used as the last
token of a sequence built with special tokens.
cls_token (:obj:`str`, `optional`, defaults to :obj:`"<s>"`):
The classifier token which is used when doing sequence classification (classification of the whole sequence
instead of per-token classification). It is the first token of the sequence when built with special tokens.
unk_token (:obj:`str`, `optional`, defaults to :obj:`"<unk>"`):
The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
token instead.
pad_token (:obj:`str`, `optional`, defaults to :obj:`"<pad>"`):
The token used for padding, for example when batching sequences of different lengths.
mask_token (:obj:`str`, `optional`, defaults to :obj:`"<mask>"`):
The token used for masking values. This is the token used when training this model with masked language
modeling. This is the token which the model will try to predict.
Examples::
>>> from transformers import MBart50Tokenizer
>>> tokenizer = MBart50Tokenizer.from_pretrained("facebook/mbart-large-50", src_lang="en_XX", tgt_lang="ro_RO")
>>> src_text = " UN Chief Says There Is No Military Solution in Syria"
>>> tgt_text = "Şeful ONU declară că nu există o soluţie militară în Siria"
>>> model_inputs = tokenizer(src_text, return_tensors="pt")
>>> with tokenizer.as_target_tokenizer():
... labels = tokenizer(tgt_text, return_tensors="pt").input_ids
>>> # model(**model_inputs, labels=labels) should work
"""
vocab_files_names = VOCAB_FILES_NAMES
max_model_input_sizes = {m: 1024 for m in _all_mbart50_models}
pretrained_vocab_files_map = {"vocab_file": {m: SPM_URL for m in _all_mbart50_models}}
model_input_names = ["input_ids", "attention_mask"]
prefix_tokens: List[int] = []
suffix_tokens: List[int] = []
def __init__(
self,
vocab_file,
src_lang=None,
tgt_lang=None,
eos_token="</s>",
sep_token="</s>",
cls_token="<s>",
unk_token="<unk>",
pad_token="<pad>",
mask_token="<mask>",
**kwargs
):
# Mask token behave like a normal word, i.e. include the space before it
mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token
super().__init__(
src_lang=src_lang,
tgt_lang=tgt_lang,
eos_token=eos_token,
unk_token=unk_token,
sep_token=sep_token,
cls_token=cls_token,
pad_token=pad_token,
mask_token=mask_token,
**kwargs,
)
self.sp_model = spm.SentencePieceProcessor()
self.sp_model.Load(str(vocab_file))
self.vocab_file = vocab_file
# Original fairseq vocab and spm vocab must be "aligned":
# Vocab | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9
# -------- | ------- | ------- | ------ | ------- | --- | --- | --- | ----- | ----- | ----
# fairseq | '<s>' | '<pad>' | '</s>' | '<unk>' | ',' | '.' | '▁' | 's' | '▁de' | '-'
# spm | '<unk>' | '<s>' | '</s>' | ',' | '.' | '▁' | 's' | '▁de' | '-' | '▁a'
# Mimic fairseq token-to-id alignment for the first 4 token
self.fairseq_tokens_to_ids = {"<s>": 0, "<pad>": 1, "</s>": 2, "<unk>": 3}
# The first "real" token "," has position 4 in the original fairseq vocab and position 3 in the spm vocab
self.fairseq_offset = 1
self.sp_model_size = len(self.sp_model)
self.lang_code_to_id = {
code: self.sp_model_size + i + self.fairseq_offset for i, code in enumerate(FAIRSEQ_LANGUAGE_CODES)
}
self.id_to_lang_code = {v: k for k, v in self.lang_code_to_id.items()}
self.fairseq_tokens_to_ids["<mask>"] = len(self.sp_model) + len(self.lang_code_to_id) + self.fairseq_offset
self.fairseq_tokens_to_ids.update(self.lang_code_to_id)
self.fairseq_ids_to_tokens = {v: k for k, v in self.fairseq_tokens_to_ids.items()}
self._additional_special_tokens = list(self.lang_code_to_id.keys())
self._src_lang = src_lang if src_lang is not None else "en_XX"
self.cur_lang_code_id = self.lang_code_to_id[self._src_lang]
self.tgt_lang = tgt_lang
self.set_src_lang_special_tokens(self._src_lang)
@property
def vocab_size(self) -> int:
return len(self.sp_model) + len(self.lang_code_to_id) + self.fairseq_offset + 1 # Plus 1 for the mask token
@property
def src_lang(self) -> str:
return self._src_lang
@src_lang.setter
def src_lang(self, new_src_lang: str) -> None:
self._src_lang = new_src_lang
self.set_src_lang_special_tokens(self._src_lang)
def __getstate__(self) -> Dict:
state = self.__dict__.copy()
state["sp_model"] = None
return state
def __setstate__(self, d: Dict) -> None:
self.__dict__ = d
self.sp_model = spm.SentencePieceProcessor()
self.sp_model.Load(self.vocab_file)
def get_vocab(self) -> Dict:
vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
vocab.update(self.added_tokens_encoder)
return vocab
def _tokenize(self, text: str) -> List[str]:
return self.sp_model.EncodeAsPieces(text)
def _convert_token_to_id(self, token: str) -> int:
""" Converts a token (str) in an id using the vocab. """
if token in self.fairseq_tokens_to_ids:
return self.fairseq_tokens_to_ids[token]
spm_id = self.sp_model.PieceToId(token)
# Need to return unknown token if the SP model returned 0
return spm_id + self.fairseq_offset if spm_id else self.unk_token_id
def _convert_id_to_token(self, index: int) -> str:
"""Converts an index (integer) in a token (str) using the vocab."""
if index in self.fairseq_ids_to_tokens:
return self.fairseq_ids_to_tokens[index]
return self.sp_model.IdToPiece(index - self.fairseq_offset)
def convert_tokens_to_string(self, tokens: List[str]) -> str:
"""Converts a sequence of tokens (strings for sub-words) in a single string."""
out_string = "".join(tokens).replace(SPIECE_UNDERLINE, " ").strip()
return out_string
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
if not os.path.isdir(save_directory):
logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
return
out_vocab_file = os.path.join(
save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
)
if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
copyfile(self.vocab_file, out_vocab_file)
return (out_vocab_file,)
def get_special_tokens_mask(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
) -> List[int]:
"""
Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer ``prepare_for_model`` method.
Args:
token_ids_0 (:obj:`List[int]`):
List of IDs.
token_ids_1 (:obj:`List[int]`, `optional`):
Optional second list of IDs for sequence pairs.
already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
Whether or not the token list is already formatted with special tokens for the model.
Returns:
:obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
"""
if already_has_special_tokens:
if token_ids_1 is not None:
raise ValueError(
"You should not supply a second sequence if the provided sequence of "
"ids is already formatted with special tokens for the model."
)
return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
prefix_ones = [1] * len(self.prefix_tokens)
suffix_ones = [1] * len(self.suffix_tokens)
if token_ids_1 is None:
return prefix_ones + ([0] * len(token_ids_0)) + suffix_ones
return prefix_ones + ([0] * len(token_ids_0)) + ([0] * len(token_ids_1)) + suffix_ones
def build_inputs_with_special_tokens(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
"""
Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. An MBART-50 sequence has the following format, where ``X`` represents the sequence:
- ``input_ids`` (for encoder) ``[src_lang_code] X [eos]``
- ``labels``: (for decoder) ``[tgt_lang_code] X [eos]``
BOS is never used. Pairs of sequences are not the expected use case, but they will be handled without a
separator.
Args:
token_ids_0 (:obj:`List[int]`):
List of IDs to which the special tokens will be added.
token_ids_1 (:obj:`List[int]`, `optional`):
Optional second list of IDs for sequence pairs.
Returns:
:obj:`List[int]`: List of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
"""
if token_ids_1 is None:
return self.prefix_tokens + token_ids_0 + self.suffix_tokens
# We don't expect to process pairs, but leave the pair logic for API consistency
return self.prefix_tokens + token_ids_0 + token_ids_1 + self.suffix_tokens
def prepare_seq2seq_batch(
self,
src_texts: List[str],
src_lang: str = "en_XX",
tgt_texts: Optional[List[str]] = None,
tgt_lang: str = "ro_RO",
**kwargs,
) -> BatchEncoding:
self.src_lang = src_lang
self.tgt_lang = tgt_lang
return super().prepare_seq2seq_batch(src_texts, tgt_texts, **kwargs)
@contextmanager
def as_target_tokenizer(self):
"""
Temporarily sets the tokenizer for encoding the targets. Useful for tokenizer associated to
sequence-to-sequence models that need a slightly different processing for the labels.
"""
self.set_tgt_lang_special_tokens(self.tgt_lang)
yield
self.set_src_lang_special_tokens(self.src_lang)
def set_src_lang_special_tokens(self, src_lang: str) -> None:
"""Reset the special tokens to the source lang setting. prefix=[src_lang_code] and suffix=[eos]."""
self.cur_lang_code_id = self.lang_code_to_id[src_lang]
self.prefix_tokens = [self.cur_lang_code_id]
self.suffix_tokens = [self.eos_token_id]
def set_tgt_lang_special_tokens(self, tgt_lang: str) -> None:
"""Reset the special tokens to the target language setting. prefix=[tgt_lang_code] and suffix=[eos]."""
self.cur_lang_code_id = self.lang_code_to_id[tgt_lang]
self.prefix_tokens = [self.cur_lang_code_id]
self.suffix_tokens = [self.eos_token_id]

278
src/transformers/models/mbart/tokenization_mbart50_fast.py Normal file
View File

@@ -0,0 +1,278 @@
# coding=utf-8
# Copyright 2021 The Facebook AI Research Team Authors and 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.
import os
from contextlib import contextmanager
from shutil import copyfile
from typing import List, Optional, Tuple
from tokenizers import processors
from ...file_utils import is_sentencepiece_available
from ...tokenization_utils import AddedToken, BatchEncoding
from ...tokenization_utils_fast import PreTrainedTokenizerFast
from ...utils import logging
if is_sentencepiece_available():
from .tokenization_mbart50 import MBart50Tokenizer
else:
MBart50Tokenizer = None
logger = logging.get_logger(__name__)
VOCAB_FILES_NAMES = {"vocab_file": "sentencepiece.bpe.model", "tokenizer_file": "tokenizer.json"}
_all_mbart50_models = ["facebook/mbart-large-50-one-to-many-mmt"]
SPM_URL = "https://huggingface.co/facebook/mbart-large-50-one-to-many-mmt/resolve/main/sentencepiece.bpe.model"
tokenizer_URL = "https://huggingface.co/facebook/mbart-large-50-one-to-many-mmt/resolve/main/tokenizer.json"
# fmt: off
FAIRSEQ_LANGUAGE_CODES = ["ar_AR", "cs_CZ", "de_DE", "en_XX", "es_XX", "et_EE", "fi_FI", "fr_XX", "gu_IN", "hi_IN", "it_IT", "ja_XX", "kk_KZ", "ko_KR", "lt_LT", "lv_LV", "my_MM", "ne_NP", "nl_XX", "ro_RO", "ru_RU", "si_LK", "tr_TR", "vi_VN", "zh_CN", "af_ZA", "az_AZ", "bn_IN", "fa_IR", "he_IL", "hr_HR", "id_ID", "ka_GE", "km_KH", "mk_MK", "ml_IN", "mn_MN", "mr_IN", "pl_PL", "ps_AF", "pt_XX", "sv_SE", "sw_KE", "ta_IN", "te_IN", "th_TH", "tl_XX", "uk_UA", "ur_PK", "xh_ZA", "gl_ES", "sl_SI"]
# fmt: on
class MBart50TokenizerFast(PreTrainedTokenizerFast):
"""
Construct a "fast" MBART tokenizer for mBART-50 (backed by HuggingFace's `tokenizers` library). Based on `BPE
<https://huggingface.co/docs/tokenizers/python/latest/components.html?highlight=BPE#models>`__.
This tokenizer inherits from :class:`~transformers.PreTrainedTokenizerFast` which contains most of the main
methods. Users should refer to this superclass for more information regarding those methods.
Args:
vocab_file (:obj:`str`):
Path to the vocabulary file.
src_lang (:obj:`str`, `optional`):
A string representing the source language.
tgt_lang (:obj:`str`, `optional`):
A string representing the target language.
eos_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
The end of sequence token.
sep_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
sequence classification or for a text and a question for question answering. It is also used as the last
token of a sequence built with special tokens.
cls_token (:obj:`str`, `optional`, defaults to :obj:`"<s>"`):
The classifier token which is used when doing sequence classification (classification of the whole sequence
instead of per-token classification). It is the first token of the sequence when built with special tokens.
unk_token (:obj:`str`, `optional`, defaults to :obj:`"<unk>"`):
The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
token instead.
pad_token (:obj:`str`, `optional`, defaults to :obj:`"<pad>"`):
The token used for padding, for example when batching sequences of different lengths.
mask_token (:obj:`str`, `optional`, defaults to :obj:`"<mask>"`):
The token used for masking values. This is the token used when training this model with masked language
modeling. This is the token which the model will try to predict.
Examples::
>>> from transformers import MBart50TokenizerFast
>>> tokenizer = MBart50TokenizerFast.from_pretrained("facebook/mbart-large-50", src_lang="en_XX", tgt_lang="ro_RO")
>>> src_text = " UN Chief Says There Is No Military Solution in Syria"
>>> tgt_text = "Şeful ONU declară că nu există o soluţie militară în Siria"
>>> model_inputs = tokenizer(src_text, return_tensors="pt")
>>> with tokenizer.as_target_tokenizer():
... labels = tokenizer(tgt_text, return_tensors="pt").input_ids
>>> # model(**model_inputs, labels=labels) should work
"""
vocab_files_names = VOCAB_FILES_NAMES
max_model_input_sizes = {m: 1024 for m in _all_mbart50_models}
pretrained_vocab_files_map = {"vocab_file": {m: SPM_URL for m in _all_mbart50_models}}
model_input_names = ["input_ids", "attention_mask"]
slow_tokenizer_class = MBart50Tokenizer
prefix_tokens: List[int] = []
suffix_tokens: List[int] = []
def __init__(
self,
vocab_file,
src_lang=None,
tgt_lang=None,
tokenizer_file=None,
eos_token="</s>",
sep_token="</s>",
cls_token="<s>",
unk_token="<unk>",
pad_token="<pad>",
mask_token="<mask>",
**kwargs
):
# Mask token behave like a normal word, i.e. include the space before it
mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token
super().__init__(
vocab_file,
src_lang=src_lang,
tgt_lang=tgt_lang,
tokenizer_file=tokenizer_file,
eos_token=eos_token,
sep_token=sep_token,
cls_token=cls_token,
unk_token=unk_token,
pad_token=pad_token,
mask_token=mask_token,
**kwargs,
)
self.vocab_file = vocab_file
self.add_special_tokens({"additional_special_tokens": FAIRSEQ_LANGUAGE_CODES})
self.lang_code_to_id = {
lang_code: self.convert_tokens_to_ids(lang_code) for lang_code in FAIRSEQ_LANGUAGE_CODES
}
self._src_lang = src_lang if src_lang is not None else "en_XX"
self.tgt_lang = tgt_lang
self.cur_lang_code_id = self.lang_code_to_id[self._src_lang]
self.set_src_lang_special_tokens(self._src_lang)
@property
def src_lang(self) -> str:
return self._src_lang
@src_lang.setter
def src_lang(self, new_src_lang: str) -> None:
self._src_lang = new_src_lang
self.set_src_lang_special_tokens(self._src_lang)
def get_special_tokens_mask(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
) -> List[int]:
"""
Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer ``prepare_for_model`` method.
Args:
token_ids_0 (:obj:`List[int]`):
List of ids.
token_ids_1 (:obj:`List[int]`, `optional`):
Optional second list of IDs for sequence pairs.
already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
Whether or not the token list is already formatted with special tokens for the model.
Returns:
:obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
"""
if already_has_special_tokens:
if token_ids_1 is not None:
raise ValueError(
"You should not supply a second sequence if the provided sequence of "
"ids is already formatted with special tokens for the model."
)
return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
prefix_ones = [1] * len(self.prefix_tokens)
suffix_ones = [1] * len(self.suffix_tokens)
if token_ids_1 is None:
return prefix_ones + ([0] * len(token_ids_0)) + suffix_ones
return prefix_ones + ([0] * len(token_ids_0)) + ([0] * len(token_ids_1)) + suffix_ones
def build_inputs_with_special_tokens(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
"""
Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. The special tokens depend on calling set_lang.
An MBART-50 sequence has the following format, where ``X`` represents the sequence:
- ``input_ids`` (for encoder) ``[src_lang_code] X [eos]``
- ``labels``: (for decoder) ``[tgt_lang_code] X [eos]``
BOS is never used. Pairs of sequences are not the expected use case, but they will be handled without a
separator.
Args:
token_ids_0 (:obj:`List[int]`):
List of IDs to which the special tokens will be added.
token_ids_1 (:obj:`List[int]`, `optional`):
Optional second list of IDs for sequence pairs.
Returns:
:obj:`List[int]`: list of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
"""
if token_ids_1 is None:
return self.prefix_tokens + token_ids_0 + self.suffix_tokens
# We don't expect to process pairs, but leave the pair logic for API consistency
return self.prefix_tokens + token_ids_0 + token_ids_1 + self.suffix_tokens
def prepare_seq2seq_batch(
self,
src_texts: List[str],
src_lang: str = "en_XX",
tgt_texts: Optional[List[str]] = None,
tgt_lang: str = "ro_RO",
**kwargs,
) -> BatchEncoding:
self.src_lang = src_lang
self.tgt_lang = tgt_lang
return super().prepare_seq2seq_batch(src_texts, tgt_texts, **kwargs)
@contextmanager
def as_target_tokenizer(self):
"""
Temporarily sets the tokenizer for encoding the targets. Useful for tokenizer associated to
sequence-to-sequence models that need a slightly different processing for the labels.
"""
self.set_tgt_lang_special_tokens(self.tgt_lang)
yield
self.set_src_lang_special_tokens(self.src_lang)
def set_src_lang_special_tokens(self, src_lang: str) -> None:
"""Reset the special tokens to the source lang setting. prefix=[src_lang_code] and suffix=[eos]."""
self.cur_lang_code_id = self.convert_tokens_to_ids(src_lang)
self.prefix_tokens = [self.cur_lang_code_id]
self.suffix_tokens = [self.eos_token_id]
prefix_tokens_str = self.convert_ids_to_tokens(self.prefix_tokens)
suffix_tokens_str = self.convert_ids_to_tokens(self.suffix_tokens)
self._tokenizer.post_processor = processors.TemplateProcessing(
single=prefix_tokens_str + ["$A"] + suffix_tokens_str,
pair=prefix_tokens_str + ["$A", "$B"] + suffix_tokens_str,
special_tokens=list(zip(prefix_tokens_str + suffix_tokens_str, self.prefix_tokens + self.suffix_tokens)),
)
def set_tgt_lang_special_tokens(self, tgt_lang: str) -> None:
"""Reset the special tokens to the target language setting. prefix=[src_lang_code] and suffix=[eos]."""
self.cur_lang_code_id = self.convert_tokens_to_ids(tgt_lang)
self.prefix_tokens = [self.cur_lang_code_id]
self.suffix_tokens = [self.eos_token_id]
prefix_tokens_str = self.convert_ids_to_tokens(self.prefix_tokens)
suffix_tokens_str = self.convert_ids_to_tokens(self.suffix_tokens)
self._tokenizer.post_processor = processors.TemplateProcessing(
single=prefix_tokens_str + ["$A"] + suffix_tokens_str,
pair=prefix_tokens_str + ["$A", "$B"] + suffix_tokens_str,
special_tokens=list(zip(prefix_tokens_str + suffix_tokens_str, self.prefix_tokens + self.suffix_tokens)),
)
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
if not os.path.isdir(save_directory):
logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
return
out_vocab_file = os.path.join(
save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
)
if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
copyfile(self.vocab_file, out_vocab_file)
return (out_vocab_file,)