Adding all project files

venv/Lib/site-packages/transformers/models/xlm/__init__.py

@@ -0,0 +1,29 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# 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.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_xlm import *
+    from .modeling_tf_xlm import *
+    from .modeling_xlm import *
+    from .tokenization_xlm import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)

venv/Lib/site-packages/transformers/models/xlm/configuration_xlm.py

@@ -0,0 +1,241 @@
+# coding=utf-8
+# Copyright 2019-present, Facebook, Inc 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.
+"""XLM configuration"""
+
+from collections import OrderedDict
+from collections.abc import Mapping
+
+from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class XLMConfig(PretrainedConfig):
+    """
+    This is the configuration class to store the configuration of a [`XLMModel`] or a [`TFXLMModel`]. It is used to
+    instantiate a XLM model according to the specified arguments, defining the model architecture. Instantiating a
+    configuration with the defaults will yield a similar configuration to that of the
+    [FacebookAI/xlm-mlm-en-2048](https://huggingface.co/FacebookAI/xlm-mlm-en-2048) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 30145):
+            Vocabulary size of the BERT model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`XLMModel`] or [`TFXLMModel`].
+        emb_dim (`int`, *optional*, defaults to 2048):
+            Dimensionality of the encoder layers and the pooler layer.
+        n_layer (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        n_head (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        dropout (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout probability for the attention mechanism
+        gelu_activation (`bool`, *optional*, defaults to `True`):
+            Whether or not to use *gelu* for the activations instead of *relu*.
+        sinusoidal_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether or not to use sinusoidal positional embeddings instead of absolute positional embeddings.
+        causal (`bool`, *optional*, defaults to `False`):
+            Whether or not the model should behave in a causal manner. Causal models use a triangular attention mask in
+            order to only attend to the left-side context instead if a bidirectional context.
+        asm (`bool`, *optional*, defaults to `False`):
+            Whether or not to use an adaptive log softmax projection layer instead of a linear layer for the prediction
+            layer.
+        n_langs (`int`, *optional*, defaults to 1):
+            The number of languages the model handles. Set to 1 for monolingual models.
+        use_lang_emb (`bool`, *optional*, defaults to `True`)
+            Whether to use language embeddings. Some models use additional language embeddings, see [the multilingual
+            models page](http://huggingface.co/transformers/multilingual.html#xlm-language-embeddings) for information
+            on how to use them.
+        max_position_embeddings (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        embed_init_std (`float`, *optional*, defaults to 2048^-0.5):
+            The standard deviation of the truncated_normal_initializer for initializing the embedding matrices.
+        init_std (`int`, *optional*, defaults to 50257):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices except the
+            embedding matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        bos_index (`int`, *optional*, defaults to 0):
+            The index of the beginning of sentence token in the vocabulary.
+        eos_index (`int`, *optional*, defaults to 1):
+            The index of the end of sentence token in the vocabulary.
+        pad_index (`int`, *optional*, defaults to 2):
+            The index of the padding token in the vocabulary.
+        unk_index (`int`, *optional*, defaults to 3):
+            The index of the unknown token in the vocabulary.
+        mask_index (`int`, *optional*, defaults to 5):
+            The index of the masking token in the vocabulary.
+        is_encoder(`bool`, *optional*, defaults to `True`):
+            Whether or not the initialized model should be a transformer encoder or decoder as seen in Vaswani et al.
+        summary_type (`string`, *optional*, defaults to "first"):
+            Argument used when doing sequence summary. Used in the sequence classification and multiple choice models.
+
+            Has to be one of the following options:
+
+                - `"last"`: Take the last token hidden state (like XLNet).
+                - `"first"`: Take the first token hidden state (like BERT).
+                - `"mean"`: Take the mean of all tokens hidden states.
+                - `"cls_index"`: Supply a Tensor of classification token position (like GPT/GPT-2).
+                - `"attn"`: Not implemented now, use multi-head attention.
+        summary_use_proj (`bool`, *optional*, defaults to `True`):
+            Argument used when doing sequence summary. Used in the sequence classification and multiple choice models.
+
+            Whether or not to add a projection after the vector extraction.
+        summary_activation (`str`, *optional*):
+            Argument used when doing sequence summary. Used in the sequence classification and multiple choice models.
+
+            Pass `"tanh"` for a tanh activation to the output, any other value will result in no activation.
+        summary_proj_to_labels (`bool`, *optional*, defaults to `True`):
+            Used in the sequence classification and multiple choice models.
+
+            Whether the projection outputs should have `config.num_labels` or `config.hidden_size` classes.
+        summary_first_dropout (`float`, *optional*, defaults to 0.1):
+            Used in the sequence classification and multiple choice models.
+
+            The dropout ratio to be used after the projection and activation.
+        start_n_top (`int`, *optional*, defaults to 5):
+            Used in the SQuAD evaluation script.
+        end_n_top (`int`, *optional*, defaults to 5):
+            Used in the SQuAD evaluation script.
+        mask_token_id (`int`, *optional*, defaults to 0):
+            Model agnostic parameter to identify masked tokens when generating text in an MLM context.
+        lang_id (`int`, *optional*, defaults to 1):
+            The ID of the language used by the model. This parameter is used when generating text in a given language.
+
+    Examples:
+
+    ```python
+    >>> from transformers import XLMConfig, XLMModel
+
+    >>> # Initializing a XLM configuration
+    >>> configuration = XLMConfig()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = XLMModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "xlm"
+    attribute_map = {
+        "hidden_size": "emb_dim",
+        "num_attention_heads": "n_heads",
+        "num_hidden_layers": "n_layers",
+        "n_words": "vocab_size",  # For backward compatibility
+    }
+
+    def __init__(
+        self,
+        vocab_size=30145,
+        emb_dim=2048,
+        n_layers=12,
+        n_heads=16,
+        dropout=0.1,
+        attention_dropout=0.1,
+        gelu_activation=True,
+        sinusoidal_embeddings=False,
+        causal=False,
+        asm=False,
+        n_langs=1,
+        use_lang_emb=True,
+        max_position_embeddings=512,
+        embed_init_std=2048**-0.5,
+        layer_norm_eps=1e-12,
+        init_std=0.02,
+        bos_index=0,
+        eos_index=1,
+        pad_index=2,
+        unk_index=3,
+        mask_index=5,
+        is_encoder=True,
+        summary_type="first",
+        summary_use_proj=True,
+        summary_activation=None,
+        summary_proj_to_labels=True,
+        summary_first_dropout=0.1,
+        start_n_top=5,
+        end_n_top=5,
+        mask_token_id=0,
+        lang_id=0,
+        pad_token_id=2,
+        bos_token_id=0,
+        **kwargs,
+    ):
+        """Constructs XLMConfig."""
+        self.vocab_size = vocab_size
+        self.emb_dim = emb_dim
+        self.n_layers = n_layers
+        self.n_heads = n_heads
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.gelu_activation = gelu_activation
+        self.sinusoidal_embeddings = sinusoidal_embeddings
+        self.causal = causal
+        self.asm = asm
+        self.n_langs = n_langs
+        self.use_lang_emb = use_lang_emb
+        self.layer_norm_eps = layer_norm_eps
+        self.bos_index = bos_index
+        self.eos_index = eos_index
+        self.pad_index = pad_index
+        self.unk_index = unk_index
+        self.mask_index = mask_index
+        self.is_encoder = is_encoder
+        self.max_position_embeddings = max_position_embeddings
+        self.embed_init_std = embed_init_std
+        self.init_std = init_std
+        self.summary_type = summary_type
+        self.summary_use_proj = summary_use_proj
+        self.summary_activation = summary_activation
+        self.summary_proj_to_labels = summary_proj_to_labels
+        self.summary_first_dropout = summary_first_dropout
+        self.start_n_top = start_n_top
+        self.end_n_top = end_n_top
+        self.mask_token_id = mask_token_id
+        self.lang_id = lang_id
+
+        if "n_words" in kwargs:
+            self.n_words = kwargs["n_words"]
+
+        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, **kwargs)
+
+
+# Copied from transformers.models.bert.configuration_bert.BertOnnxConfig
+class XLMOnnxConfig(OnnxConfig):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "multiple-choice":
+            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
+        else:
+            dynamic_axis = {0: "batch", 1: "sequence"}
+        return OrderedDict(
+            [
+                ("input_ids", dynamic_axis),
+                ("attention_mask", dynamic_axis),
+                ("token_type_ids", dynamic_axis),
+            ]
+        )
+
+
+__all__ = ["XLMConfig", "XLMOnnxConfig"]

venv/Lib/site-packages/transformers/models/xlm/tokenization_xlm.py

@@ -0,0 +1,577 @@
+# coding=utf-8
+# Copyright 2019 The Open AI 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.
+"""Tokenization classes for XLM."""
+
+import json
+import os
+import re
+import sys
+import unicodedata
+from typing import Optional
+
+from ...tokenization_utils import PreTrainedTokenizer
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {
+    "vocab_file": "vocab.json",
+    "merges_file": "merges.txt",
+}
+
+
+def get_pairs(word):
+    """
+    Return set of symbol pairs in a word. word is represented as tuple of symbols (symbols being variable-length
+    strings)
+    """
+    pairs = set()
+    prev_char = word[0]
+    for char in word[1:]:
+        pairs.add((prev_char, char))
+        prev_char = char
+    return pairs
+
+
+def lowercase_and_remove_accent(text):
+    """
+    Lowercase and strips accents from a piece of text based on
+    https://github.com/facebookresearch/XLM/blob/master/tools/lowercase_and_remove_accent.py
+    """
+    text = " ".join(text)
+    text = text.lower()
+    text = unicodedata.normalize("NFD", text)
+    output = []
+    for char in text:
+        cat = unicodedata.category(char)
+        if cat == "Mn":
+            continue
+        output.append(char)
+    return "".join(output).lower().split(" ")
+
+
+def replace_unicode_punct(text):
+    """
+    Port of https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/replace-unicode-punctuation.perl
+    """
+    text = text.replace("，", ",")
+    text = re.sub(r"。\s*", ". ", text)
+    text = text.replace("、", ",")
+    text = text.replace("”", '"')
+    text = text.replace("“", '"')
+    text = text.replace("∶", ":")
+    text = text.replace("：", ":")
+    text = text.replace("？", "?")
+    text = text.replace("《", '"')
+    text = text.replace("》", '"')
+    text = text.replace("）", ")")
+    text = text.replace("！", "!")
+    text = text.replace("（", "(")
+    text = text.replace("；", ";")
+    text = text.replace("１", "1")
+    text = text.replace("」", '"')
+    text = text.replace("「", '"')
+    text = text.replace("０", "0")
+    text = text.replace("３", "3")
+    text = text.replace("２", "2")
+    text = text.replace("５", "5")
+    text = text.replace("６", "6")
+    text = text.replace("９", "9")
+    text = text.replace("７", "7")
+    text = text.replace("８", "8")
+    text = text.replace("４", "4")
+    text = re.sub(r"．\s*", ". ", text)
+    text = text.replace("～", "~")
+    text = text.replace("’", "'")
+    text = text.replace("…", "...")
+    text = text.replace("━", "-")
+    text = text.replace("〈", "<")
+    text = text.replace("〉", ">")
+    text = text.replace("【", "[")
+    text = text.replace("】", "]")
+    text = text.replace("％", "%")
+    return text
+
+
+def remove_non_printing_char(text):
+    """
+    Port of https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/remove-non-printing-char.perl
+    """
+    output = []
+    for char in text:
+        cat = unicodedata.category(char)
+        if cat.startswith("C"):
+            continue
+        output.append(char)
+    return "".join(output)
+
+
+def romanian_preprocessing(text):
+    """Sennrich's WMT16 scripts for Romanian preprocessing, used by model `FacebookAI/xlm-mlm-enro-1024`"""
+    # https://github.com/rsennrich/wmt16-scripts/blob/master/preprocess/normalise-romanian.py
+    text = text.replace("\u015e", "\u0218").replace("\u015f", "\u0219")
+    text = text.replace("\u0162", "\u021a").replace("\u0163", "\u021b")
+    # https://github.com/rsennrich/wmt16-scripts/blob/master/preprocess/remove-diacritics.py
+    text = text.replace("\u0218", "S").replace("\u0219", "s")  # s-comma
+    text = text.replace("\u021a", "T").replace("\u021b", "t")  # t-comma
+    text = text.replace("\u0102", "A").replace("\u0103", "a")
+    text = text.replace("\u00c2", "A").replace("\u00e2", "a")
+    text = text.replace("\u00ce", "I").replace("\u00ee", "i")
+    return text
+
+
+class XLMTokenizer(PreTrainedTokenizer):
+    """
+    Construct an XLM tokenizer. Based on Byte-Pair Encoding. The tokenization process is the following:
+
+    - Moses preprocessing and tokenization for most supported languages.
+    - Language specific tokenization for Chinese (Jieba), Japanese (KyTea) and Thai (PyThaiNLP).
+    - Optionally lowercases and normalizes all inputs text.
+    - The arguments `special_tokens` and the function `set_special_tokens`, can be used to add additional symbols (like
+      "__classify__") to a vocabulary.
+    - The `lang2id` attribute maps the languages supported by the model with their IDs if provided (automatically set
+      for pretrained vocabularies).
+    - The `id2lang` attributes does reverse mapping if provided (automatically set for pretrained vocabularies).
+
+    This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to
+    this superclass for more information regarding those methods.
+
+    Args:
+        vocab_file (`str`):
+            Vocabulary file.
+        merges_file (`str`):
+            Merges file.
+        unk_token (`str`, *optional*, defaults to `"<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.
+        bos_token (`str`, *optional*, defaults to `"<s>"`):
+            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
+
+            <Tip>
+
+            When building a sequence using special tokens, this is not the token that is used for the beginning of
+            sequence. The token used is the `cls_token`.
+
+            </Tip>
+
+        sep_token (`str`, *optional*, defaults to `"</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.
+        pad_token (`str`, *optional*, defaults to `"<pad>"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        cls_token (`str`, *optional*, defaults to `"</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.
+        mask_token (`str`, *optional*, defaults to `"<special1>"`):
+            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.
+        additional_special_tokens (`List[str]`, *optional*, defaults to `['<special0>', '<special1>', '<special2>', '<special3>', '<special4>', '<special5>', '<special6>', '<special7>', '<special8>', '<special9>']`):
+            List of additional special tokens.
+        lang2id (`Dict[str, int]`, *optional*):
+            Dictionary mapping languages string identifiers to their IDs.
+        id2lang (`Dict[int, str]`, *optional*):
+            Dictionary mapping language IDs to their string identifiers.
+        do_lowercase_and_remove_accent (`bool`, *optional*, defaults to `True`):
+            Whether to lowercase and remove accents when tokenizing.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+
+    def __init__(
+        self,
+        vocab_file,
+        merges_file,
+        unk_token="<unk>",
+        bos_token="<s>",
+        sep_token="</s>",
+        pad_token="<pad>",
+        cls_token="</s>",
+        mask_token="<special1>",
+        additional_special_tokens=[
+            "<special0>",
+            "<special1>",
+            "<special2>",
+            "<special3>",
+            "<special4>",
+            "<special5>",
+            "<special6>",
+            "<special7>",
+            "<special8>",
+            "<special9>",
+        ],
+        lang2id=None,
+        id2lang=None,
+        do_lowercase_and_remove_accent=True,
+        **kwargs,
+    ):
+        try:
+            import sacremoses
+        except ImportError:
+            raise ImportError(
+                "You need to install sacremoses to use XLMTokenizer. "
+                "See https://pypi.org/project/sacremoses/ for installation."
+            )
+
+        self.sm = sacremoses
+
+        # cache of sm.MosesPunctNormalizer instance
+        self.cache_moses_punct_normalizer = {}
+        # cache of sm.MosesTokenizer instance
+        self.cache_moses_tokenizer = {}
+        self.lang_with_custom_tokenizer = {"zh", "th", "ja"}
+        # True for current supported model (v1.2.0), False for XLM-17 & 100
+        self.do_lowercase_and_remove_accent = do_lowercase_and_remove_accent
+        self.lang2id = lang2id
+        self.id2lang = id2lang
+        if lang2id is not None and id2lang is not None:
+            assert len(lang2id) == len(id2lang)
+
+        self.ja_word_tokenizer = None
+        self.zh_word_tokenizer = None
+
+        with open(vocab_file, encoding="utf-8") as vocab_handle:
+            self.encoder = json.load(vocab_handle)
+        self.decoder = {v: k for k, v in self.encoder.items()}
+        with open(merges_file, encoding="utf-8") as merges_handle:
+            merges = merges_handle.read().split("\n")[:-1]
+        merges = [tuple(merge.split()[:2]) for merge in merges]
+        self.bpe_ranks = dict(zip(merges, range(len(merges))))
+        self.cache = {}
+        super().__init__(
+            unk_token=unk_token,
+            bos_token=bos_token,
+            sep_token=sep_token,
+            pad_token=pad_token,
+            cls_token=cls_token,
+            mask_token=mask_token,
+            additional_special_tokens=additional_special_tokens,
+            lang2id=lang2id,
+            id2lang=id2lang,
+            do_lowercase_and_remove_accent=do_lowercase_and_remove_accent,
+            **kwargs,
+        )
+
+    @property
+    def do_lower_case(self):
+        return self.do_lowercase_and_remove_accent
+
+    def moses_punct_norm(self, text, lang):
+        if lang not in self.cache_moses_punct_normalizer:
+            punct_normalizer = self.sm.MosesPunctNormalizer(lang=lang)
+            self.cache_moses_punct_normalizer[lang] = punct_normalizer
+        else:
+            punct_normalizer = self.cache_moses_punct_normalizer[lang]
+        return punct_normalizer.normalize(text)
+
+    def moses_tokenize(self, text, lang):
+        if lang not in self.cache_moses_tokenizer:
+            moses_tokenizer = self.sm.MosesTokenizer(lang=lang)
+            self.cache_moses_tokenizer[lang] = moses_tokenizer
+        else:
+            moses_tokenizer = self.cache_moses_tokenizer[lang]
+        return moses_tokenizer.tokenize(text, return_str=False, escape=False)
+
+    def moses_pipeline(self, text, lang):
+        text = replace_unicode_punct(text)
+        text = self.moses_punct_norm(text, lang)
+        text = remove_non_printing_char(text)
+        return text
+
+    def ja_tokenize(self, text):
+        if self.ja_word_tokenizer is None:
+            try:
+                import Mykytea
+
+                self.ja_word_tokenizer = Mykytea.Mykytea(
+                    f"-model {os.path.expanduser('~')}/local/share/kytea/model.bin"
+                )
+            except (AttributeError, ImportError):
+                logger.error(
+                    "Make sure you install KyTea (https://github.com/neubig/kytea) and it's python wrapper"
+                    " (https://github.com/chezou/Mykytea-python) with the following steps"
+                )
+                logger.error("1. git clone git@github.com:neubig/kytea.git && cd kytea")
+                logger.error("2. autoreconf -i")
+                logger.error("3. ./configure --prefix=$HOME/local")
+                logger.error("4. make && make install")
+                logger.error("5. pip install kytea")
+                raise
+        return list(self.ja_word_tokenizer.getWS(text))
+
+    @property
+    def vocab_size(self):
+        return len(self.encoder)
+
+    def get_vocab(self):
+        return dict(self.encoder, **self.added_tokens_encoder)
+
+    def bpe(self, token):
+        word = tuple(token[:-1]) + (token[-1] + "</w>",)
+        if token in self.cache:
+            return self.cache[token]
+        pairs = get_pairs(word)
+
+        if not pairs:
+            return token + "</w>"
+
+        while True:
+            bigram = min(pairs, key=lambda pair: self.bpe_ranks.get(pair, float("inf")))
+            if bigram not in self.bpe_ranks:
+                break
+            first, second = bigram
+            new_word = []
+            i = 0
+            while i < len(word):
+                try:
+                    j = word.index(first, i)
+                except ValueError:
+                    new_word.extend(word[i:])
+                    break
+                else:
+                    new_word.extend(word[i:j])
+                    i = j
+
+                if word[i] == first and i < len(word) - 1 and word[i + 1] == second:
+                    new_word.append(first + second)
+                    i += 2
+                else:
+                    new_word.append(word[i])
+                    i += 1
+            new_word = tuple(new_word)
+            word = new_word
+            if len(word) == 1:
+                break
+            else:
+                pairs = get_pairs(word)
+        word = " ".join(word)
+        if word == "\n  </w>":
+            word = "\n</w>"
+        self.cache[token] = word
+        return word
+
+    def _tokenize(self, text, lang="en", bypass_tokenizer=False):
+        """
+        Tokenize a string given language code. For Chinese, Japanese and Thai, we use a language specific tokenizer.
+        Otherwise, we use Moses.
+
+        Details of tokenization:
+
+            - [sacremoses](https://github.com/alvations/sacremoses): port of Moses
+            - Install with `pip install sacremoses`
+            - [pythainlp](https://github.com/PyThaiNLP/pythainlp): Thai tokenizer
+            - Install with `pip install pythainlp`
+            - [kytea](https://github.com/chezou/Mykytea-python): Japanese tokenizer, wrapper of
+              [KyTea](https://github.com/neubig/kytea)
+            - Install with the following steps:
+
+            ::
+
+                git clone git@github.com:neubig/kytea.git && cd kytea autoreconf -i ./configure --prefix=$HOME/local
+                make && make install pip install kytea
+
+            - [jieba](https://github.com/fxsjy/jieba): Chinese tokenizer (*)
+            - Install with `pip install jieba`
+
+        (*) The original XLM used [Stanford
+        Segmenter](https://nlp.stanford.edu/software/stanford-segmenter-2018-10-16.zip). However, the wrapper
+        (`nltk.tokenize.stanford_segmenter`) is slow due to JVM overhead, and it will be deprecated. Jieba is a lot
+        faster and pip-installable. Note there is some mismatch with the Stanford Segmenter. It should be fine if you
+        fine-tune the model with Chinese supervisionself. If you want the same exact behaviour, use the original XLM
+        [preprocessing script](https://github.com/facebookresearch/XLM/tree/master/tools) to tokenize the sentence
+        externally, and set `bypass_tokenizer=True` to bypass the tokenizer.
+
+        Args:
+            - lang: ISO language code (default = 'en') (string). Languages should belong of the model supported
+              languages. However, we don't enforce it.
+            - bypass_tokenizer: Allow users to preprocess and tokenize the sentences externally (default = False)
+              (bool). If True, we only apply BPE.
+
+        Returns:
+            List of tokens.
+        """
+        if lang and self.lang2id and lang not in self.lang2id:
+            logger.error(
+                "Supplied language code not found in lang2id mapping. Please check that your language is supported by"
+                " the loaded pretrained model."
+            )
+        if bypass_tokenizer:
+            text = text.split()
+        elif lang not in self.lang_with_custom_tokenizer:
+            text = self.moses_pipeline(text, lang=lang)
+            # TODO: make sure we are using `FacebookAI/xlm-mlm-enro-1024`, since XLM-100 doesn't have this step
+            if lang == "ro":
+                text = romanian_preprocessing(text)
+            text = self.moses_tokenize(text, lang=lang)
+        elif lang == "th":
+            text = self.moses_pipeline(text, lang=lang)
+            try:
+                if "pythainlp" not in sys.modules:
+                    from pythainlp.tokenize import word_tokenize as th_word_tokenize
+                else:
+                    th_word_tokenize = sys.modules["pythainlp"].word_tokenize
+            except (AttributeError, ImportError):
+                logger.error(
+                    "Make sure you install PyThaiNLP (https://github.com/PyThaiNLP/pythainlp) with the following steps"
+                )
+                logger.error("1. pip install pythainlp")
+                raise
+            text = th_word_tokenize(text)
+        elif lang == "zh":
+            try:
+                if "jieba" not in sys.modules:
+                    import jieba
+                else:
+                    jieba = sys.modules["jieba"]
+            except (AttributeError, ImportError):
+                logger.error("Make sure you install Jieba (https://github.com/fxsjy/jieba) with the following steps")
+                logger.error("1. pip install jieba")
+                raise
+            text = " ".join(jieba.cut(text))
+            text = self.moses_pipeline(text, lang=lang)
+            text = text.split()
+        elif lang == "ja":
+            text = self.moses_pipeline(text, lang=lang)
+            text = self.ja_tokenize(text)
+        else:
+            raise ValueError("It should not reach here")
+
+        if self.do_lowercase_and_remove_accent and not bypass_tokenizer:
+            text = lowercase_and_remove_accent(text)
+
+        split_tokens = []
+        for token in text:
+            if token:
+                split_tokens.extend(list(self.bpe(token).split(" ")))
+
+        return split_tokens
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.encoder.get(token, self.encoder.get(self.unk_token))
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.decoder.get(index, self.unk_token)
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        out_string = "".join(tokens).replace("</w>", " ").strip()
+        return out_string
+
+    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 XLM sequence has the following format:
+
+        - single sequence: `<s> X </s>`
+        - pair of sequences: `<s> A </s> B </s>`
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
+
+        """
+        bos = [self.bos_token_id]
+        sep = [self.sep_token_id]
+
+        if token_ids_1 is None:
+            return bos + token_ids_0 + sep
+        return bos + token_ids_0 + sep + token_ids_1 + sep
+
+    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 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `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:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
+            )
+
+        if token_ids_1 is not None:
+            return [1] + ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1]
+
+    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> tuple[str]:
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
+        )
+        merge_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["merges_file"]
+        )
+
+        with open(vocab_file, "w", encoding="utf-8") as f:
+            f.write(json.dumps(self.encoder, indent=2, sort_keys=True, ensure_ascii=False) + "\n")
+
+        index = 0
+        with open(merge_file, "w", encoding="utf-8") as writer:
+            for bpe_tokens, token_index in sorted(self.bpe_ranks.items(), key=lambda kv: kv[1]):
+                if index != token_index:
+                    logger.warning(
+                        f"Saving vocabulary to {merge_file}: BPE merge indices are not consecutive."
+                        " Please check that the tokenizer is not corrupted!"
+                    )
+                    index = token_index
+                writer.write(" ".join(bpe_tokens) + "\n")
+                index += 1
+
+        return vocab_file, merge_file
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sm"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+
+        try:
+            import sacremoses
+        except ImportError:
+            raise ImportError(
+                "You need to install sacremoses to use XLMTokenizer. "
+                "See https://pypi.org/project/sacremoses/ for installation."
+            )
+
+        self.sm = sacremoses
+
+
+__all__ = ["XLMTokenizer"]