Add source files directly to repo (#1)

- Update config.json (09ee222c2eb842e3ce6a7ecbb883ce3a6d566601)
- Create configuration_character_bert.py (30397d827839963ebdd7260716f2e92308bdf1f5)
- Create modeling_character_bert.py (925f56a222d4dd223bb007b83955e29b2ffa2782)
- Create tokenization_character_bert.py (9e9c634911af026b3fc12570d763850db70ec353)
- Update tokenizer_config.json (15160b7632e7b425feb1caea96abd3abe6df05d1)

config.json

@@ -1,7 +1,13 @@
 {
+  "_name_or_path": "helboukkouri/character-bert-medical",
   "architectures": [
     "CharacterBertForPreTraining"
   ],
+  "auto_map": {
+    "AutoConfig": "configuration_character_bert.CharacterBertConfig",
+    "AutoModel": "modeling_character_bert.CharacterBertForPreTraining",
+    "AutoModelForMaskedLM": "modeling_character_bert.CharacterBertForMaskedLM"
+  },
   "attention_probs_dropout_prob": 0.1,
   "character_embeddings_dim": 16,
   "cnn_activation": "relu",
@@ -52,4 +58,4 @@
   "transformers_version": "4.7.0.dev0",
   "type_vocab_size": 2,
   "use_cache": true
-}
+}

configuration_character_bert.py

@@ -0,0 +1,156 @@
+# coding=utf-8
+# Copyright Hicham EL BOUKKOURI, Olivier FERRET, Thomas LAVERGNE, Hiroshi NOJI,
+# Pierre ZWEIGENBAUM, Junichi TSUJII and The HuggingFace Inc. 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.
+""" CharacterBERT model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+CHARACTER_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "helboukkouri/character-bert": "https://huggingface.co/helboukkouri/character-bert/resolve/main/config.json",
+    "helboukkouri/character-bert-medical": "https://huggingface.co/helboukkouri/character-bert-medical/resolve/main/config.json",
+    # See all CharacterBERT models at https://huggingface.co/models?filter=character_bert
+}
+
+
+class CharacterBertConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`CharacterBertModel`]. It is
+    used to instantiate an CharacterBERT 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 CharacterBERT
+    [helboukkouri/character-bert](https://huggingface.co/helboukkouri/character-bert) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model
+    outputs. Read the documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        character_embeddings_dim (`int`, *optional*, defaults to `16`):
+            The size of the character embeddings.
+        cnn_activation (`str`, *optional*, defaults to `"relu"`):
+            The activation function to apply to the cnn representations.
+        cnn_filters (:
+            obj:*list(list(int))*, *optional*, defaults to `[[1, 32], [2, 32], [3, 64], [4, 128], [5, 256], [6, 512], [7, 1024]]`): The list of CNN filters to use in the CharacterCNN module.
+        num_highway_layers (`int`, *optional*, defaults to `2`):
+            The number of Highway layers to apply to the CNNs output.
+        max_word_length (`int`, *optional*, defaults to `50`):
+            The maximum token length in characters (actually, in bytes as any non-ascii characters will be converted to
+            a sequence of utf-8 bytes).
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (`int`, *optional*, defaults to 3072):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string,
+            `"gelu"`, `"relu"`, `"selu"` and `"gelu_new"` are supported.
+        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        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).
+        type_vocab_size (`int`, *optional*, defaults to 2):
+            The vocabulary size of the `token_type_ids` passed when calling
+            [`CharacterBertModel`] or [`TFCharacterBertModel`].
+        mlm_vocab_size (`int`, *optional*, defaults to 100000):
+            Size of the output vocabulary for MLM.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+
+    Example:
+
+    ```python
+
+    ```
+
+            >>> from transformers import CharacterBertModel, CharacterBertConfig
+
+            >>> # Initializing a CharacterBERT helboukkouri/character-bert style configuration
+            >>> configuration = CharacterBertConfig()
+
+            >>> # Initializing a model from the helboukkouri/character-bert style configuration
+            >>> model = CharacterBertModel(configuration)
+
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
+    """
+    model_type = "character_bert"
+
+    def __init__(
+        self,
+        character_embeddings_dim=16,
+        cnn_activation="relu",
+        cnn_filters=None,
+        num_highway_layers=2,
+        max_word_length=50,
+        hidden_size=768,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        intermediate_size=3072,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=2,
+        mlm_vocab_size=100000,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        is_encoder_decoder=False,
+        use_cache=True,
+        **kwargs
+    ):
+        tie_word_embeddings = kwargs.pop("tie_word_embeddings", False)
+        if tie_word_embeddings:
+            raise ValueError(
+                "Cannot tie word embeddings in CharacterBERT. Please set " "`config.tie_word_embeddings=False`."
+            )
+        super().__init__(
+            type_vocab_size=type_vocab_size,
+            layer_norm_eps=layer_norm_eps,
+            use_cache=use_cache,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+        if cnn_filters is None:
+            cnn_filters = [[1, 32], [2, 32], [3, 64], [4, 128], [5, 256], [6, 512], [7, 1024]]
+        self.character_embeddings_dim = character_embeddings_dim
+        self.cnn_activation = cnn_activation
+        self.cnn_filters = cnn_filters
+        self.num_highway_layers = num_highway_layers
+        self.max_word_length = max_word_length
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.mlm_vocab_size = mlm_vocab_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.initializer_range = initializer_range

modeling_character_bert.py

@@ -0,0 +1,1954 @@
+# coding=utf-8
+# Copyright Hicham EL BOUKKOURI, Olivier FERRET, Thomas LAVERGNE, Hiroshi NOJI,
+# Pierre ZWEIGENBAUM, Junichi TSUJII, The HuggingFace Inc. and AllenNLP teams.
+# 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.
+"""
+PyTorch CharacterBERT model: this is a variant of BERT that uses the CharacterCNN module from ELMo instead of a
+WordPiece embedding matrix. See: “CharacterBERT: Reconciling ELMo and BERT for Word-Level Open-Vocabulary
+Representations From Characters“ https://www.aclweb.org/anthology/2020.coling-main.609/
+"""
+
+import math
+import warnings
+from dataclasses import dataclass
+from typing import Callable, Optional, Tuple
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss, MSELoss
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from .configuration_character_bert import CharacterBertConfig
+from .tokenization_character_bert import CharacterMapper
+
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "helboukkouri/character-bert"
+_CONFIG_FOR_DOC = "CharacterBertConfig"
+_TOKENIZER_FOR_DOC = "CharacterBertTokenizer"
+
+CHARACTER_BERT_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "helboukkouri/character-bert",
+    "helboukkouri/character-bert-medical",
+    # See all CharacterBERT models at https://huggingface.co/models?filter=character_bert
+]
+
+
+# NOTE: the following class is taken from:
+# https://github.com/allenai/allennlp/blob/main/allennlp/modules/highway.py
+class Highway(torch.nn.Module):
+    """
+    A `Highway layer <https://arxiv.org/abs/1505.00387)>`__ does a gated combination of a linear transformation and a
+    non-linear transformation of its input. :math:`y = g * x + (1 - g) * f(A(x))`, where :math:`A` is a linear
+    transformation, :math:`f` is an element-wise non-linearity, and :math:`g` is an element-wise gate, computed as
+    :math:`sigmoid(B(x))`.
+
+    This module will apply a fixed number of highway layers to its input, returning the final result.
+
+    # Parameters
+
+    input_dim : `int`, required The dimensionality of :math:`x`. We assume the input has shape `(batch_size, ...,
+    input_dim)`. num_layers : `int`, optional (default=`1`) The number of highway layers to apply to the input.
+    activation : `Callable[[torch.Tensor], torch.Tensor]`, optional (default=`torch.nn.functional.relu`) The
+    non-linearity to use in the highway layers.
+    """
+
+    def __init__(
+        self,
+        input_dim: int,
+        num_layers: int = 1,
+        activation: Callable[[torch.Tensor], torch.Tensor] = torch.nn.functional.relu,
+    ) -> None:
+        super().__init__()
+        self._input_dim = input_dim
+        self._layers = torch.nn.ModuleList([torch.nn.Linear(input_dim, input_dim * 2) for _ in range(num_layers)])
+        self._activation = activation
+        for layer in self._layers:
+            # We should bias the highway layer to just carry its input forward.  We do that by
+            # setting the bias on `B(x)` to be positive, because that means `g` will be biased to
+            # be high, so we will carry the input forward.  The bias on `B(x)` is the second half
+            # of the bias vector in each Linear layer.
+            layer.bias[input_dim:].data.fill_(1)
+
+    def forward(self, inputs: torch.Tensor) -> torch.Tensor:
+        current_input = inputs
+        for layer in self._layers:
+            projected_input = layer(current_input)
+            linear_part = current_input
+            # NOTE: if you modify this, think about whether you should modify the initialization
+            # above, too.
+            nonlinear_part, gate = projected_input.chunk(2, dim=-1)
+            nonlinear_part = self._activation(nonlinear_part)
+            gate = torch.sigmoid(gate)
+            current_input = gate * linear_part + (1 - gate) * nonlinear_part
+        return current_input
+
+
+# NOTE: The CharacterCnn was adapted from `_ElmoCharacterEncoder`:
+# https://github.com/allenai/allennlp/blob/main/allennlp/modules/elmo.py#L254
+class CharacterCnn(torch.nn.Module):
+    """
+    Computes context insensitive token representation using multiple CNNs. This embedder has input character ids of
+    size (batch_size, sequence_length, 50) and returns (batch_size, sequence_length, hidden_size), where hidden_size is
+    typically 768.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.character_embeddings_dim = config.character_embeddings_dim
+        self.cnn_activation = config.cnn_activation
+        self.cnn_filters = config.cnn_filters
+        self.num_highway_layers = config.num_highway_layers
+        self.max_word_length = config.max_word_length
+        self.hidden_size = config.hidden_size
+        # NOTE: this is the 256 possible utf-8 bytes + special slots for the
+        # [CLS]/[SEP]/[PAD]/[MASK] characters as well as beginning/end of
+        # word symbols and character padding for short words -> total of 263
+        self.character_vocab_size = 263
+        self._init_weights()
+
+    def get_output_dim(self):
+        return self.hidden_size
+
+    def _init_weights(self):
+        self._init_char_embedding()
+        self._init_cnn_weights()
+        self._init_highway()
+        self._init_projection()
+
+    def _init_char_embedding(self):
+        weights = torch.empty((self.character_vocab_size, self.character_embeddings_dim))
+        nn.init.normal_(weights)
+        weights[0].fill_(0.0)  # token padding
+        weights[CharacterMapper.padding_character + 1].fill_(0.0)  # character padding
+        self._char_embedding_weights = torch.nn.Parameter(torch.FloatTensor(weights), requires_grad=True)
+
+    def _init_cnn_weights(self):
+        convolutions = []
+        for i, (width, num) in enumerate(self.cnn_filters):
+            conv = torch.nn.Conv1d(
+                in_channels=self.character_embeddings_dim, out_channels=num, kernel_size=width, bias=True
+            )
+            conv.weight.requires_grad = True
+            conv.bias.requires_grad = True
+            convolutions.append(conv)
+            self.add_module(f"char_conv_{i}", conv)
+        self._convolutions = convolutions
+
+    def _init_highway(self):
+        # the highway layers have same dimensionality as the number of cnn filters
+        n_filters = sum(f[1] for f in self.cnn_filters)
+        self._highways = Highway(n_filters, self.num_highway_layers, activation=nn.functional.relu)
+        for k in range(self.num_highway_layers):
+            # The AllenNLP highway is one matrix multplication with concatenation of
+            # transform and carry weights.
+            self._highways._layers[k].weight.requires_grad = True
+            self._highways._layers[k].bias.requires_grad = True
+
+    def _init_projection(self):
+        n_filters = sum(f[1] for f in self.cnn_filters)
+        self._projection = torch.nn.Linear(n_filters, self.hidden_size, bias=True)
+        self._projection.weight.requires_grad = True
+        self._projection.bias.requires_grad = True
+
+    def forward(self, inputs: torch.Tensor) -> torch.Tensor:
+        """
+        Compute context insensitive token embeddings from characters. # Parameters inputs : `torch.Tensor` Shape
+        `(batch_size, sequence_length, 50)` of character ids representing the current batch. # Returns output:
+        `torch.Tensor` Shape `(batch_size, sequence_length, embedding_dim)` tensor with context insensitive token
+        representations.
+        """
+
+        # character embeddings
+        # (batch_size * sequence_length, max_word_length, embed_dim)
+        character_embedding = torch.nn.functional.embedding(
+            inputs.view(-1, self.max_word_length), self._char_embedding_weights
+        )
+
+        # CNN representations
+        if self.cnn_activation == "tanh":
+            activation = torch.tanh
+        elif self.cnn_activation == "relu":
+            activation = torch.nn.functional.relu
+        else:
+            raise Exception("ConfigurationError: Unknown activation")
+
+        # (batch_size * sequence_length, embed_dim, max_word_length)
+        character_embedding = torch.transpose(character_embedding, 1, 2)
+        convs = []
+        for i in range(len(self._convolutions)):
+            conv = getattr(self, "char_conv_{}".format(i))
+            convolved = conv(character_embedding)
+            # (batch_size * sequence_length, n_filters for this width)
+            convolved, _ = torch.max(convolved, dim=-1)
+            convolved = activation(convolved)
+            convs.append(convolved)
+
+        # (batch_size * sequence_length, n_filters)
+        token_embedding = torch.cat(convs, dim=-1)
+
+        # apply the highway layers (batch_size * sequence_length, n_filters)
+        token_embedding = self._highways(token_embedding)
+
+        # final projection  (batch_size * sequence_length, embedding_dim)
+        token_embedding = self._projection(token_embedding)
+
+        # reshape to (batch_size, sequence_length, embedding_dim)
+        batch_size, sequence_length, _ = inputs.size()
+        output = token_embedding.view(batch_size, sequence_length, -1)
+
+        return output
+
+
+class CharacterBertEmbeddings(nn.Module):
+    """Construct the embeddings from word, position and token_type embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = CharacterCnn(config)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+
+    def forward(
+        self, input_ids=None, token_type_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids[:, :, 0].size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if token_type_ids is None:
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+        token_type_embeddings = self.token_type_embeddings(token_type_ids)
+
+        embeddings = inputs_embeds + token_type_embeddings
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+# Copied from transformers.models.bert.modeling_bert.BertSelfAttention with Bert->CharacterBert
+class CharacterBertSelfAttention(nn.Module):
+    def __init__(self, config, position_embedding_type=None):
+        super().__init__()
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
+                f"heads ({config.num_attention_heads})"
+            )
+
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        self.key = nn.Linear(config.hidden_size, self.all_head_size)
+        self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = position_embedding_type or getattr(
+            config, "position_embedding_type", "absolute"
+        )
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+
+        self.is_decoder = config.is_decoder
+
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention and past_key_value is not None:
+            # reuse k,v, cross_attentions
+            key_layer = past_key_value[0]
+            value_layer = past_key_value[1]
+            attention_mask = encoder_attention_mask
+        elif is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in CharacterBertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.functional.softmax(attention_scores, dim=-1)
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs = attention_probs * head_mask
+
+        context_layer = torch.matmul(attention_probs, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        if self.is_decoder:
+            outputs = outputs + (past_key_value,)
+        return outputs
+
+
+# Copied from transformers.models.bert.modeling_bert.BertSelfOutput with Bert->CharacterBert
+class CharacterBertSelfOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertAttention with Bert->CharacterBert
+class CharacterBertAttention(nn.Module):
+    def __init__(self, config, position_embedding_type=None):
+        super().__init__()
+        self.self = CharacterBertSelfAttention(config, position_embedding_type=position_embedding_type)
+        self.output = CharacterBertSelfOutput(config)
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        self_outputs = self.self(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+# Copied from transformers.models.bert.modeling_bert.BertIntermediate with Bert->CharacterBert
+class CharacterBertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertOutput with Bert->CharacterBert
+class CharacterBertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertLayer with Bert->CharacterBert
+class CharacterBertLayer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = CharacterBertAttention(config)
+        self.is_decoder = config.is_decoder
+        self.add_cross_attention = config.add_cross_attention
+        if self.add_cross_attention:
+            if not self.is_decoder:
+                raise ValueError(f"{self} should be used as a decoder model if cross attention is added")
+            self.crossattention = CharacterBertAttention(config, position_embedding_type="absolute")
+        self.intermediate = CharacterBertIntermediate(config)
+        self.output = CharacterBertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        # if decoder, the last output is tuple of self-attn cache
+        if self.is_decoder:
+            outputs = self_attention_outputs[1:-1]
+            present_key_value = self_attention_outputs[-1]
+        else:
+            outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
+
+        cross_attn_present_key_value = None
+        if self.is_decoder and encoder_hidden_states is not None:
+            if not hasattr(self, "crossattention"):
+                raise ValueError(
+                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`"
+                )
+
+            # cross_attn cached key/values tuple is at positions 3,4 of past_key_value tuple
+            cross_attn_past_key_value = past_key_value[-2:] if past_key_value is not None else None
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                cross_attn_past_key_value,
+                output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights
+
+            # add cross-attn cache to positions 3,4 of present_key_value tuple
+            cross_attn_present_key_value = cross_attention_outputs[-1]
+            present_key_value = present_key_value + cross_attn_present_key_value
+
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        # if decoder, return the attn key/values as the last output
+        if self.is_decoder:
+            outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+# Copied from transformers.models.bert.modeling_bert.BertEncoder with Bert->CharacterBert
+class CharacterBertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([CharacterBertLayer(config) for _ in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+        for i, layer_module in enumerate(self.layer):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warning(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+                if self.config.add_cross_attention:
+                    all_cross_attentions = all_cross_attentions + (layer_outputs[2],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPooler with Bert->CharacterBert
+class CharacterBertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPredictionHeadTransform with Bert->CharacterBert
+class CharacterBertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class CharacterBertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = CharacterBertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.mlm_vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.mlm_vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertOnlyMLMHead with Bert->CharacterBert
+class CharacterBertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = CharacterBertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+# Copied from transformers.models.bert.modeling_bert.BertOnlyNSPHead with Bert->CharacterBert
+class CharacterBertOnlyNSPHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.seq_relationship = nn.Linear(config.hidden_size, 2)
+
+    def forward(self, pooled_output):
+        seq_relationship_score = self.seq_relationship(pooled_output)
+        return seq_relationship_score
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPreTrainingHeads with Bert->CharacterBert
+class CharacterBertPreTrainingHeads(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = CharacterBertLMPredictionHead(config)
+        self.seq_relationship = nn.Linear(config.hidden_size, 2)
+
+    def forward(self, sequence_output, pooled_output):
+        prediction_scores = self.predictions(sequence_output)
+        seq_relationship_score = self.seq_relationship(pooled_output)
+        return prediction_scores, seq_relationship_score
+
+
+class CharacterBertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = CharacterBertConfig
+    load_tf_weights = None
+    base_model_prefix = "character_bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        if isinstance(module, CharacterCnn):
+            # We need to handle the case of these parameters since it is not an actual module
+            module._char_embedding_weights.data.normal_()
+            # token padding
+            module._char_embedding_weights.data[0].fill_(0.0)
+            # character padding
+            module._char_embedding_weights.data[CharacterMapper.padding_character + 1].fill_(0.0)
+        if isinstance(module, nn.Linear):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+
+@dataclass
+# Copied from transformers.models.bert.modeling_bert.BertForPreTrainingOutput with Bert->CharacterBert
+class CharacterBertForPreTrainingOutput(ModelOutput):
+    """
+    Output type of [`CharacterBertForPreTraining`].
+
+    Args:
+        loss (*optional*, returned when `labels` is provided, `torch.FloatTensor` of shape `(1,)`):
+            Total loss as the sum of the masked language modeling loss and the next sequence prediction
+            (classification) loss.
+        prediction_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        seq_relationship_logits (`torch.FloatTensor` of shape `(batch_size, 2)`):
+            Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation
+            before SoftMax).
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
+            shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    prediction_logits: torch.FloatTensor = None
+    seq_relationship_logits: torch.FloatTensor = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+CHARACTER_BERT_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config (:
+            class:*~transformers.CharacterBertConfig*): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model
+            weights.
+"""
+
+CHARACTER_BERT_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `{0}`):
+            Indices of input sequence tokens.
+
+            Indices can be obtained using [`CharacterBertTokenizer`]. See
+            [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for
+            details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `{1}`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `{1}`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `{1}`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (:
+            obj:*torch.FloatTensor* of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): Mask
+            to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (:
+            obj:*torch.FloatTensor* of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert *input_ids* indices into associated vectors
+            than the model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare CharacterBERT Model transformer outputting raw hidden-states without any specific head on top.",
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertModel(CharacterBertPreTrainedModel):
+    """
+
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in [Attention is
+    all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+
+    To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration
+    set to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder`
+    argument and `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = CharacterBertEmbeddings(config)
+        self.encoder = CharacterBertEncoder(config)
+
+        self.pooler = CharacterBertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def resize_token_embeddings(self, *args, **kwargs):
+        raise NotImplementedError("Cannot resize CharacterBERT's token embeddings.")
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPoolingAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        encoder_hidden_states  (:
+            obj:*torch.FloatTensor* of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence
+            of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model
+            is configured as a decoder.
+        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:
+            obj:*tuple(tuple(torch.FloatTensor))* of length `config.n_layers` with each tuple having 4 tensors of
+            shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key
+            and value hidden states of the attention blocks. Can be used to speed up decoding. If
+            `past_key_values` are used, the user can optionally input only the last `decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape `(batch_size, 1)`
+            instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up
+            decoding (see `past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if self.config.is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()[:-1]
+            batch_size, seq_length = input_shape
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+        if token_type_ids is None:
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.is_decoder and encoder_hidden_states is not None:
+            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+
+        embedding_output = self.embeddings(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            inputs_embeds=inputs_embeds,
+            past_key_values_length=past_key_values_length,
+        )
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CharacterBert Model with two heads on top as done during the pretraining: a `masked language modeling` head and a
+    `next sentence prediction (classification)` head.
+    """,
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForPreTraining(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.character_bert = CharacterBertModel(config)
+        self.cls = CharacterBertPreTrainingHeads(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @replace_return_docstrings(output_type=CharacterBertForPreTrainingOutput, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        next_sentence_label=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.mlm_vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.mlm_vocab_size]`
+        next_sentence_label (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair
+            (see `input_ids` docstring) Indices should be in `[0, 1]`:
+
+            - 0 indicates sequence B is a continuation of sequence A,
+            - 1 indicates sequence B is a random sequence.
+        kwargs (`Dict[str, any]`, optional, defaults to *{}*):
+            Used to hide legacy arguments that have been deprecated.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import CharacterBertTokenizer, CharacterBertForPreTraining >>> import torch
+
+        >>> tokenizer = CharacterBertTokenizer.from_pretrained('helboukkouri/character-bert') >>> model =
+        CharacterBertForPreTraining.from_pretrained('helboukkouri/character-bert')
+
+        >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt") >>> outputs = model(**inputs)
+
+        >>> prediction_logits = outputs.prediction_logits >>> seq_relationship_logits =
+        outputs.seq_relationship_logits
+        ```
+"""
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output, pooled_output = outputs[:2]
+        prediction_scores, seq_relationship_score = self.cls(sequence_output, pooled_output)
+
+        total_loss = None
+        if labels is not None and next_sentence_label is not None:
+            loss_fct = CrossEntropyLoss()
+            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.mlm_vocab_size), labels.view(-1))
+            next_sentence_loss = loss_fct(seq_relationship_score.view(-1, 2), next_sentence_label.view(-1))
+            total_loss = masked_lm_loss + next_sentence_loss
+
+        if not return_dict:
+            output = (prediction_scores, seq_relationship_score) + outputs[2:]
+            return ((total_loss,) + output) if total_loss is not None else output
+
+        return CharacterBertForPreTrainingOutput(
+            loss=total_loss,
+            prediction_logits=prediction_scores,
+            seq_relationship_logits=seq_relationship_score,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """CharacterBert Model with a `language modeling` head on top for CLM fine-tuning.""",
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertLMHeadModel(CharacterBertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        if not config.is_decoder:
+            logger.warning("If you want to use `CharacterBertLMHeadModel` as a standalone, add `is_decoder=True.`")
+
+        self.character_bert = CharacterBertModel(config, add_pooling_layer=False)
+        self.cls = CharacterBertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @replace_return_docstrings(output_type=CausalLMOutputWithCrossAttentions, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        labels=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        encoder_hidden_states  (:
+            obj:*torch.FloatTensor* of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence
+            of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model
+            is configured as a decoder.
+        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
+            `[-100, 0, ..., config.mlm_vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100`
+            are ignored (masked), the loss is only computed for the tokens with labels n `[0, ..., config.mlm_vocab_size]`
+        past_key_values (:
+            obj:*tuple(tuple(torch.FloatTensor))* of length `config.n_layers` with each tuple having 4 tensors of
+            shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key
+            and value hidden states of the attention blocks. Can be used to speed up decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape `(batch_size, 1)`
+            instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up
+            decoding (see `past_key_values`).
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import CharacterBertTokenizer, CharacterBertLMHeadModel, CharacterBertConfig >>>
+        import torch
+
+        >>> tokenizer = CharacterBertTokenizer.from_pretrained('helboukkouri/character-bert') >>> config =
+        CharacterBertConfig.from_pretrained("helboukkouri/character-bert") >>> config.is_decoder = True >>> model =
+        CharacterBertLMHeadModel.from_pretrained('helboukkouri/character-bert', config=config)
+
+        >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt") >>> outputs = model(**inputs)
+
+        >>> prediction_logits = outputs.logits
+        ```
+"""
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if labels is not None:
+            use_cache = False
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+
+        lm_loss = None
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss()
+            lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.mlm_vocab_size), labels.view(-1))
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((lm_loss,) + output) if lm_loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=lm_loss,
+            logits=prediction_scores,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        # cut decoder_input_ids if past is used
+        if past is not None:
+            input_ids = input_ids[:, -1:]
+
+        return {"input_ids": input_ids, "attention_mask": attention_mask, "past_key_values": past}
+
+    def _reorder_cache(self, past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past
+
+
+@add_start_docstrings(
+    """CharacterBert Model with a `language modeling` head on top.""", CHARACTER_BERT_START_DOCSTRING
+)
+class CharacterBertForMaskedLM(CharacterBertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        if config.is_decoder:
+            logger.warning(
+                "If you want to use `CharacterBertForMaskedLM` make sure `config.is_decoder=False` for "
+                "bi-directional self-attention."
+            )
+        self.character_bert = CharacterBertModel(config, add_pooling_layer=False)
+        self.cls = CharacterBertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=MaskedLMOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.mlm_vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.mlm_vocab_size]`
+        """
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+
+        masked_lm_loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()  # -100 index = padding token
+            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.mlm_vocab_size), labels.view(-1))
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
+
+        return MaskedLMOutput(
+            loss=masked_lm_loss,
+            logits=prediction_scores,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+        effective_batch_size = input_shape[0]
+
+        #  add a dummy token
+        assert self.config.pad_token_id is not None, "The PAD token should be defined for generation"
+        attention_mask = torch.cat([attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1)
+        dummy_token = torch.full(
+            (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device
+        )
+        input_ids = torch.cat([input_ids, dummy_token], dim=1)
+
+        return {"input_ids": input_ids, "attention_mask": attention_mask}
+
+
+@add_start_docstrings(
+    """CharacterBert Model with a `next sentence prediction (classification)` head on top.""",
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForNextSentencePrediction(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.character_bert = CharacterBertModel(config)
+        self.cls = CharacterBertOnlyNSPHead(config)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @replace_return_docstrings(output_type=NextSentencePredictorOutput, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        **kwargs
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair
+            (see `input_ids` docstring). Indices should be in `[0, 1]`:
+
+            - 0 indicates sequence B is a continuation of sequence A,
+            - 1 indicates sequence B is a random sequence.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import CharacterBertTokenizer, CharacterBertForNextSentencePrediction >>> import
+        torch
+
+        >>> tokenizer = CharacterBertTokenizer.from_pretrained('helboukkouri/character-bert') >>> model =
+        CharacterBertForNextSentencePrediction.from_pretrained('helboukkouri/character-bert')
+
+        >>> prompt = "In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced."
+        >>> next_sentence = "The sky is blue due to the shorter wavelength of blue light." >>> encoding =
+        tokenizer(prompt, next_sentence, return_tensors='pt')
+
+        >>> outputs = model(**encoding, labels=torch.LongTensor([1])) >>> logits = outputs.logits >>> assert
+        logits[0, 0] < logits[0, 1] # next sentence was random
+        ```
+"""
+
+        if "next_sentence_label" in kwargs:
+            warnings.warn(
+                "The `next_sentence_label` argument is deprecated and will be removed in a future version, use `labels` instead.",
+                FutureWarning,
+            )
+            labels = kwargs.pop("next_sentence_label")
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        pooled_output = outputs[1]
+
+        seq_relationship_scores = self.cls(pooled_output)
+
+        next_sentence_loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            next_sentence_loss = loss_fct(seq_relationship_scores.view(-1, 2), labels.view(-1))
+
+        if not return_dict:
+            output = (seq_relationship_scores,) + outputs[2:]
+            return ((next_sentence_loss,) + output) if next_sentence_loss is not None else output
+
+        return NextSentencePredictorOutput(
+            loss=next_sentence_loss,
+            logits=seq_relationship_scores,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CharacterBERT Model transformer with a sequence classification/regression head on top (a linear layer on top of the
+    pooled output) e.g. for GLUE tasks.
+    """,
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForSequenceClassification(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.character_bert = CharacterBertModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=SequenceClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss),
+            If `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        pooled_output = outputs[1]
+
+        pooled_output = self.dropout(pooled_output)
+        logits = self.classifier(pooled_output)
+
+        loss = None
+        if labels is not None:
+            if self.num_labels == 1:
+                #  We are doing regression
+                loss_fct = MSELoss()
+                loss = loss_fct(logits.view(-1), labels.view(-1))
+            else:
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CharacterBERT Model with a multiple choice classification head on top (a linear layer on top of the pooled output
+    and a softmax) e.g. for RocStories/SWAG tasks.
+    """,
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForMultipleChoice(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.character_bert = CharacterBertModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, 1)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=MultipleChoiceModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
+            `input_ids` above)
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]
+
+        input_ids = input_ids.view(-1, input_ids.size(-2), input_ids.size(-1)) if input_ids is not None else None
+        attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
+        token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
+        position_ids = position_ids.view(-1, position_ids.size(-1)) if position_ids is not None else None
+        inputs_embeds = (
+            inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
+            if inputs_embeds is not None
+            else None
+        )
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        pooled_output = outputs[1]
+
+        pooled_output = self.dropout(pooled_output)
+        logits = self.classifier(pooled_output)
+        reshaped_logits = logits.view(-1, num_choices)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(reshaped_logits, labels)
+
+        if not return_dict:
+            output = (reshaped_logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return MultipleChoiceModelOutput(
+            loss=loss,
+            logits=reshaped_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CharacterBERT Model with a token classification head on top (a linear layer on top of the hidden-states output)
+    e.g. for Named-Entity-Recognition (NER) tasks.
+    """,
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForTokenClassification(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.character_bert = CharacterBertModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        sequence_output = self.dropout(sequence_output)
+        logits = self.classifier(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            # Only keep active parts of the loss
+            if attention_mask is not None:
+                active_loss = attention_mask.view(-1) == 1
+                active_logits = logits.view(-1, self.num_labels)
+                active_labels = torch.where(
+                    active_loss, labels.view(-1), torch.tensor(loss_fct.ignore_index).type_as(labels)
+                )
+                loss = loss_fct(active_logits, active_labels)
+            else:
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CharacterBERT Model with a span classification head on top for extractive question-answering tasks like SQuAD (a
+    linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`).
+    """,
+    CHARACTER_BERT_START_DOCSTRING,
+)
+class CharacterBertForQuestionAnswering(CharacterBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        config.num_labels = 2
+        self.num_labels = config.num_labels
+
+        self.character_bert = CharacterBertModel(config)
+        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(
+        CHARACTER_BERT_INPUTS_DOCSTRING.format(
+            "(batch_size, sequence_length, maximum_token_length)", "(batch_size, sequence_length)"
+        )
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=QuestionAnsweringModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        start_positions=None,
+        end_positions=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for position (index) of the start of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the
+            sequence are not taken into account for computing the loss.
+        end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for position (index) of the end of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the
+            sequence are not taken into account for computing the loss.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.character_bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        logits = self.qa_outputs(sequence_output)
+        start_logits, end_logits = logits.split(1, dim=-1)
+        start_logits = start_logits.squeeze(-1)
+        end_logits = end_logits.squeeze(-1)
+
+        total_loss = None
+        if start_positions is not None and end_positions is not None:
+            # If we are on multi-GPU, split add a dimension
+            if len(start_positions.size()) > 1:
+                start_positions = start_positions.squeeze(-1)
+            if len(end_positions.size()) > 1:
+                end_positions = end_positions.squeeze(-1)
+            # sometimes the start/end positions are outside our model inputs, we ignore these terms
+            ignored_index = start_logits.size(1)
+            start_positions.clamp_(0, ignored_index)
+            end_positions.clamp_(0, ignored_index)
+
+            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
+            start_loss = loss_fct(start_logits, start_positions)
+            end_loss = loss_fct(end_logits, end_positions)
+            total_loss = (start_loss + end_loss) / 2
+
+        if not return_dict:
+            output = (start_logits, end_logits) + outputs[2:]
+            return ((total_loss,) + output) if total_loss is not None else output
+
+        return QuestionAnsweringModelOutput(
+            loss=total_loss,
+            start_logits=start_logits,
+            end_logits=end_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

tokenization_character_bert.py

@@ -0,0 +1,930 @@
+# coding=utf-8
+# Copyright Hicham EL BOUKKOURI, Olivier FERRET, Thomas LAVERGNE, Hiroshi NOJI,
+# Pierre ZWEIGENBAUM, Junichi TSUJII and The HuggingFace Inc. 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.
+"""Tokenization classes for CharacterBERT."""
+import json
+import os
+import unicodedata
+from collections import OrderedDict
+from typing import Dict, List, Optional, Tuple, Union
+
+import numpy as np
+
+from transformers.file_utils import is_tf_available, is_torch_available, to_py_obj
+from transformers.tokenization_utils import (
+    BatchEncoding,
+    EncodedInput,
+    PaddingStrategy,
+    PreTrainedTokenizer,
+    TensorType,
+    _is_control,
+    _is_punctuation,
+    _is_whitespace,
+)
+from transformers.tokenization_utils_base import ADDED_TOKENS_FILE
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {
+    "mlm_vocab_file": "mlm_vocab.txt",
+}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "mlm_vocab_file": {
+        "helboukkouri/character-bert": "https://huggingface.co/helboukkouri/character-bert/resolve/main/mlm_vocab.txt",
+        "helboukkouri/character-bert-medical": "https://huggingface.co/helboukkouri/character-bert-medical/resolve/main/mlm_vocab.txt",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "helboukkouri/character-bert": 512,
+    "helboukkouri/character-bert-medical": 512,
+}
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "helboukkouri/character-bert": {"max_word_length": 50, "do_lower_case": True},
+    "helboukkouri/character-bert-medical": {"max_word_length": 50, "do_lower_case": True},
+}
+
+PAD_TOKEN_CHAR_ID = 0
+
+
+def whitespace_tokenize(text):
+    """Runs basic whitespace cleaning and splitting on a piece of text."""
+    text = text.strip()
+    if not text:
+        return []
+    tokens = text.split()
+    return tokens
+
+
+def build_mlm_ids_to_tokens_mapping(mlm_vocab_file):
+    """Builds a Masked Language Modeling ids to masked tokens mapping."""
+    vocabulary = []
+    with open(mlm_vocab_file, "r", encoding="utf-8") as reader:
+        for line in reader:
+            line = line.strip()
+            if line:
+                vocabulary.append(line)
+    return OrderedDict(list(enumerate(vocabulary)))
+
+
+class CharacterBertTokenizer(PreTrainedTokenizer):
+    """
+    Construct a CharacterBERT tokenizer. Based on characters.
+
+    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:
+        mlm_vocab_file (`str`, *optional*, defaults to `None`):
+            Path to the Masked Language Modeling vocabulary. This is used for converting the output (token ids) of the
+            MLM model into tokens.
+        max_word_length (`int`, *optional*, defaults to `50`):
+            The maximum token length in characters (actually, in bytes as any non-ascii characters will be converted to
+            a sequence of utf-8 bytes).
+        do_lower_case (`bool`, *optional*, defaults to `True`):
+            Whether or not to lowercase the input when tokenizing.
+        do_basic_tokenize (`bool`, *optional*, defaults to `True`):
+            Whether or not to do basic tokenization before WordPiece.
+        never_split (`Iterable`, *optional*):
+            Collection of tokens which will never be split during tokenization. Only has an effect when
+            `do_basic_tokenize=True`
+        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.
+        sep_token (`str`, *optional*, defaults to `"[SEP]"`):
+            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 `"[CLS]"`):
+            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 `"[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.
+        tokenize_chinese_chars (`bool`, *optional*, defaults to `True`):
+            Whether or not to tokenize Chinese characters.
+        strip_accents: (`bool`, *optional*):
+            Whether or not to strip all accents. If this option is not specified, then it will be determined by the
+            value for `lowercase` (as in the original BERT).
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(
+        self,
+        mlm_vocab_file=None,
+        max_word_length=50,
+        do_lower_case=True,
+        do_basic_tokenize=True,
+        never_split=None,
+        unk_token="[UNK]",
+        sep_token="[SEP]",
+        pad_token="[PAD]",
+        cls_token="[CLS]",
+        mask_token="[MASK]",
+        tokenize_chinese_chars=True,
+        strip_accents=None,
+        **kwargs
+    ):
+        super().__init__(
+            max_word_length=max_word_length,
+            do_lower_case=do_lower_case,
+            do_basic_tokenize=do_basic_tokenize,
+            never_split=never_split,
+            unk_token=unk_token,
+            sep_token=sep_token,
+            pad_token=pad_token,
+            cls_token=cls_token,
+            mask_token=mask_token,
+            tokenize_chinese_chars=tokenize_chinese_chars,
+            strip_accents=strip_accents,
+            **kwargs,
+        )
+        # This prevents splitting special tokens during tokenization
+        self.unique_no_split_tokens = [self.cls_token, self.mask_token, self.pad_token, self.sep_token, self.unk_token]
+        # This is used for converting MLM ids into tokens
+        if mlm_vocab_file is None:
+            self.ids_to_tokens = None
+        else:
+            if not os.path.isfile(mlm_vocab_file):
+                raise ValueError(
+                    f"Can't find a vocabulary file at path '{mlm_vocab_file}'. "
+                    "To load the vocabulary from a pretrained model use "
+                    "`tokenizer = CharacterBertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`"
+                )
+            self.ids_to_tokens = build_mlm_ids_to_tokens_mapping(mlm_vocab_file)
+        # Tokenization is handled by BasicTokenizer
+        self.do_basic_tokenize = do_basic_tokenize
+        if do_basic_tokenize:
+            self.basic_tokenizer = BasicTokenizer(
+                do_lower_case=do_lower_case,
+                never_split=never_split,
+                tokenize_chinese_chars=tokenize_chinese_chars,
+                strip_accents=strip_accents,
+            )
+        # Then, a CharacterMapper is responsible for converting tokens into character ids
+        self.max_word_length = max_word_length
+        self._mapper = CharacterMapper(max_word_length=max_word_length)
+
+    def __repr__(self) -> str:
+        # NOTE: we overwrite this because CharacterBERT does not have self.vocab_size
+        return (
+            f"CharacterBertTokenizer(name_or_path='{self.name_or_path}', "
+            + (f"mlm_vocab_size={self.mlm_vocab_size}, " if self.ids_to_tokens else "")
+            + f"model_max_len={self.model_max_length}, is_fast={self.is_fast}, "
+            + f"padding_side='{self.padding_side}', special_tokens={self.special_tokens_map_extended})"
+        )
+
+    def __len__(self):
+        """
+        Size of the full vocabulary with the added tokens.
+        """
+        # return self.vocab_size + len(self.added_tokens_encoder)
+        return 0 + len(self.added_tokens_encoder)
+
+    @property
+    def do_lower_case(self):
+        return self.basic_tokenizer.do_lower_case
+
+    @property
+    def vocab_size(self):
+        raise NotImplementedError("CharacterBERT does not use a token vocabulary.")
+
+    @property
+    def mlm_vocab_size(self):
+        if self.ids_to_tokens is None:
+            raise ValueError(
+                "CharacterBertTokenizer was initialized without a MLM "
+                "vocabulary. You can either pass one manually or load a "
+                "pre-trained tokenizer using: "
+                "`tokenizer = CharacterBertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`"
+            )
+        return len(self.ids_to_tokens)
+
+    def add_special_tokens(self, *args, **kwargs):
+        raise NotImplementedError("Adding special tokens is not supported for now.")
+
+    def add_tokens(self, *args, **kwargs):
+        # We don't raise an Exception here to allow for ignoring this step.
+        # Otherwise, many inherited methods would need to be re-implemented...
+        pass
+
+    def get_vocab(self):
+        raise NotImplementedError("CharacterBERT does not have a token vocabulary.")
+
+    def get_mlm_vocab(self):
+        return {token: i for i, token in self.ids_to_tokens.items()}
+
+    def _tokenize(self, text):
+        split_tokens = []
+        if self.do_basic_tokenize:
+            split_tokens = self.basic_tokenizer.tokenize(text=text, never_split=self.all_special_tokens)
+        else:
+            split_tokens = whitespace_tokenize(text)  # Default to whitespace tokenization
+        return split_tokens
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        out_string = " ".join(tokens).strip()
+        return out_string
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) into a sequence of character ids."""
+        return self._mapper.convert_word_to_char_ids(token)
+
+    def _convert_id_to_token(self, index: List[int]):
+        # NOTE: keeping the same variable name `ìndex` although this will
+        # always be a sequence of indices.
+        """Converts an index (actually, a list of indices) in a token (str)."""
+        return self._mapper.convert_char_ids_to_word(index)
+
+    def convert_ids_to_tokens(
+        self, ids: Union[List[int], List[List[int]]], skip_special_tokens: bool = False
+    ) -> Union[str, List[str]]:
+        """
+        Converts a single sequence of character indices or a sequence of character id sequences in a token or a
+        sequence of tokens.
+
+        Args:
+            ids (`int` or `List[int]`):
+                The token id (or token ids) to convert to tokens.
+            skip_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not to remove special tokens in the decoding.
+
+        Returns:
+            `str` or `List[str]`: The decoded token(s).
+        """
+        if isinstance(ids, list) and isinstance(ids[0], int):
+            if tuple(ids) in self.added_tokens_decoder:
+                return self.added_tokens_decoder[tuple(ids)]
+            else:
+                return self._convert_id_to_token(ids)
+        tokens = []
+        for indices in ids:
+            indices = list(map(int, indices))
+            if skip_special_tokens and tuple(indices) in self.all_special_ids:
+                continue
+            if tuple(indices) in self.added_tokens_decoder:
+                tokens.append(self.added_tokens_decoder[tuple(indices)])
+            else:
+                tokens.append(self._convert_id_to_token(indices))
+        return tokens
+
+    def convert_mlm_id_to_token(self, mlm_id):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        if self.ids_to_tokens is None:
+            raise ValueError(
+                "CharacterBertTokenizer was initialized without a MLM "
+                "vocabulary. You can either pass one manually or load a "
+                "pre-trained tokenizer using: "
+                "`tokenizer = CharacterBertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`"
+            )
+        assert (
+            mlm_id < self.mlm_vocab_size
+        ), "Attempting to convert a MLM id that is greater than the MLM vocabulary size."
+        return self.ids_to_tokens[mlm_id]
+
+    def build_inputs_with_special_tokens(
+        self, token_ids_0: List[List[int]], token_ids_1: Optional[List[List[int]]] = None
+    ) -> List[List[int]]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. A CharacterBERT sequence has the following format:
+
+        - single sequence: `[CLS] X [SEP]`
+        - pair of sequences: `[CLS] A [SEP] B [SEP]`
+
+        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.
+        """
+        if token_ids_1 is None:
+            return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
+        cls = [self.cls_token_id]
+        sep = [self.sep_token_id]
+        return cls + token_ids_0 + sep + token_ids_1 + sep
+
+    def get_special_tokens_mask(
+        self,
+        token_ids_0: List[List[int]],
+        token_ids_1: Optional[List[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:
+            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))
+
+        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 create_token_type_ids_from_sequences(
+        self, token_ids_0: List[List[int]], token_ids_1: Optional[List[List[int]]] = None
+    ) -> List[int]:
+        """
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task. A CharacterBERT
+        sequence pair mask has the following format:
+
+        ```
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 | first sequence | second sequence |
+        ```
+
+                If `token_ids_1` is `None`, this method only returns the first portion of the mask (0s).
+
+                Args:
+                    token_ids_0 (`List[int]`):
+                        List of IDs.
+                    token_ids_1 (`List[int]`, *optional*):
+                        Optional second list of IDs for sequence pairs.
+
+                Returns:
+                    `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given
+                    sequence(s).
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
+        return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
+
+    # def pad(
+    #     self,
+    #     encoded_inputs: Union[
+    #         BatchEncoding,
+    #         List[BatchEncoding],
+    #         Dict[str, EncodedInput],
+    #         Dict[str, List[EncodedInput]],
+    #         List[Dict[str, EncodedInput]],
+    #     ],
+    #     padding: Union[bool, str, PaddingStrategy] = True,
+    #     max_length: Optional[int] = None,
+    #     pad_to_multiple_of: Optional[int] = None,
+    #     return_attention_mask: Optional[bool] = None,
+    #     return_tensors: Optional[Union[str, TensorType]] = None,
+    #     verbose: bool = True,
+    # ) -> BatchEncoding:
+    #     """
+    #     Pad a single encoded input or a batch of encoded inputs up to predefined length or to the max sequence length
+    #     in the batch.
+
+    #     Padding side (left/right) padding token ids are defined at the tokenizer level (with `self.padding_side`,
+    #     `self.pad_token_id` and `self.pad_token_type_id`)
+
+    #     <Tip>
+
+    #     If the `encoded_inputs` passed are dictionary of numpy arrays, PyTorch tensors or TensorFlow tensors, the
+    #     result will use the same type unless you provide a different tensor type with `return_tensors`. In the
+    #     case of PyTorch tensors, you will lose the specific device of your tensors however.
+
+    #     </Tip>
+
+    #     Args:
+    #         encoded_inputs (:
+    #             class:*~transformers.BatchEncoding*, list of [`BatchEncoding`], `Dict[str, List[int]]`, `Dict[str, List[List[int]]` or `List[Dict[str, List[int]]]`): Tokenized inputs.
+    #             Can represent one input ([`BatchEncoding`] or `Dict[str, List[int]]`) or a
+    #             batch of tokenized inputs (list of [`BatchEncoding`], *Dict[str, List[List[int]]]*
+    #             or *List[Dict[str, List[int]]]*) so you can use this method during preprocessing as well as in a
+    #             PyTorch Dataloader collate function.
+
+    #             Instead of `List[int]` you can have tensors (numpy arrays, PyTorch tensors or TensorFlow tensors),
+    #             see the note above for the return type.
+    #         padding (:
+    #              obj:*bool*, `str` or [`~file_utils.PaddingStrategy`], *optional*, defaults to
+    #              `True`): Select a strategy to pad the returned sequences (according to the model's padding side
+    #              and padding index) among:
+
+    #             - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a
+    #               single sequence if provided).
+    #             - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the
+    #               maximum acceptable input length for the model if that argument is not provided.
+    #             - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+    #               different lengths).
+    #         max_length (`int`, *optional*):
+    #             Maximum length of the returned list and optionally padding length (see above).
+    #         pad_to_multiple_of (`int`, *optional*):
+    #             If set will pad the sequence to a multiple of the provided value.
+
+    #             This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
+    #             >= 7.5 (Volta).
+    #         return_attention_mask (`bool`, *optional*):
+    #             Whether to return the attention mask. If left to the default, will return the attention mask according
+    #             to the specific tokenizer's default, defined by the `return_outputs` attribute.
+
+    #             [What are attention masks?](../glossary#attention-mask)
+    #         return_tensors (`str` or [`~file_utils.TensorType`], *optional*):
+    #             If set, will return tensors instead of list of python integers. Acceptable values are:
+
+    #             - `'tf'`: Return TensorFlow `tf.constant` objects.
+    #             - `'pt'`: Return PyTorch `torch.Tensor` objects.
+    #             - `'np'`: Return Numpy `np.ndarray` objects.
+    #         verbose (`bool`, *optional*, defaults to `True`):
+    #             Whether or not to print more information and warnings.
+    #     """
+    #     # If we have a list of dicts, let's convert it in a dict of lists
+    #     # We do this to allow using this method as a collate_fn function in PyTorch Dataloader
+    #     if isinstance(encoded_inputs, (list, tuple)) and isinstance(encoded_inputs[0], (dict, BatchEncoding)):
+    #         encoded_inputs = {key: [example[key] for example in encoded_inputs] for key in encoded_inputs[0].keys()}
+
+    #     # The model's main input name, usually `input_ids`, has be passed for padding
+    #     if self.model_input_names[0] not in encoded_inputs:
+    #         raise ValueError(
+    #             "You should supply an encoding or a list of encodings to this method "
+    #             f"that includes {self.model_input_names[0]}, but you provided {list(encoded_inputs.keys())}"
+    #         )
+
+    #     required_input = encoded_inputs[self.model_input_names[0]]
+
+    #     if not required_input:
+    #         if return_attention_mask:
+    #             encoded_inputs["attention_mask"] = []
+    #         return encoded_inputs
+
+    #     # If we have PyTorch/TF/NumPy tensors/arrays as inputs, we cast them as python objects
+    #     # and rebuild them afterwards if no return_tensors is specified
+    #     # Note that we lose the specific device the tensor may be on for PyTorch
+
+    #     first_element = required_input[0]
+    #     if isinstance(first_element, (list, tuple)):
+    #         # first_element might be an empty list/tuple in some edge cases so we grab the first non empty element.
+    #         index = 0
+    #         while len(required_input[index]) == 0:
+    #             index += 1
+    #         if index < len(required_input):
+    #             first_element = required_input[index][0]
+    #     # At this state, if `first_element` is still a list/tuple, it's an empty one so there is nothing to do.
+    #     if not isinstance(first_element, (int, list, tuple)):
+    #         if is_tf_available() and _is_tensorflow(first_element):
+    #             return_tensors = "tf" if return_tensors is None else return_tensors
+    #         elif is_torch_available() and _is_torch(first_element):
+    #             return_tensors = "pt" if return_tensors is None else return_tensors
+    #         elif isinstance(first_element, np.ndarray):
+    #             return_tensors = "np" if return_tensors is None else return_tensors
+    #         else:
+    #             raise ValueError(
+    #                 f"type of {first_element} unknown: {type(first_element)}. "
+    #                 f"Should be one of a python, numpy, pytorch or tensorflow object."
+    #             )
+
+    #         for key, value in encoded_inputs.items():
+    #             encoded_inputs[key] = to_py_obj(value)
+
+    #     # Convert padding_strategy in PaddingStrategy
+    #     padding_strategy, _, max_length, _ = self._get_padding_truncation_strategies(
+    #         padding=padding, max_length=max_length, verbose=verbose
+    #     )
+
+    #     required_input = encoded_inputs[self.model_input_names[0]]
+    #     if required_input and not isinstance(required_input[0][0], (list, tuple)):
+    #         encoded_inputs = self._pad(
+    #             encoded_inputs,
+    #             max_length=max_length,
+    #             padding_strategy=padding_strategy,
+    #             pad_to_multiple_of=pad_to_multiple_of,
+    #             return_attention_mask=return_attention_mask,
+    #         )
+    #         return BatchEncoding(encoded_inputs, tensor_type=return_tensors)
+
+    #     batch_size = len(required_input)
+    #     assert all(
+    #         len(v) == batch_size for v in encoded_inputs.values()
+    #     ), "Some items in the output dictionary have a different batch size than others."
+
+    #     if padding_strategy == PaddingStrategy.LONGEST:
+    #         max_length = max(len(inputs) for inputs in required_input)
+    #         padding_strategy = PaddingStrategy.MAX_LENGTH
+
+    #     batch_outputs = {}
+    #     for i in range(batch_size):
+    #         inputs = dict((k, v[i]) for k, v in encoded_inputs.items())
+    #         outputs = self._pad(
+    #             inputs,
+    #             max_length=max_length,
+    #             padding_strategy=padding_strategy,
+    #             pad_to_multiple_of=pad_to_multiple_of,
+    #             return_attention_mask=return_attention_mask,
+    #         )
+
+    #         for key, value in outputs.items():
+    #             if key not in batch_outputs:
+    #                 batch_outputs[key] = []
+    #             batch_outputs[key].append(value)
+
+    #     return BatchEncoding(batch_outputs, tensor_type=return_tensors)
+
+    # def _pad(
+    #     self,
+    #     encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
+    #     max_length: Optional[int] = None,
+    #     padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+    #     pad_to_multiple_of: Optional[int] = None,
+    #     return_attention_mask: Optional[bool] = None,
+    # ) -> dict:
+    #     """
+    #     Pad encoded inputs (on left/right and up to predefined length or max length in the batch)
+
+    #     Args:
+    #         encoded_inputs:
+    #             Dictionary of tokenized inputs (`List[int]`) or batch of tokenized inputs (`List[List[int]]`).
+    #         max_length: maximum length of the returned list and optionally padding length (see below).
+    #             Will truncate by taking into account the special tokens.
+    #         padding_strategy: PaddingStrategy to use for padding.
+
+    #             - PaddingStrategy.LONGEST Pad to the longest sequence in the batch
+    #             - PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
+    #             - PaddingStrategy.DO_NOT_PAD: Do not pad
+    #             The tokenizer padding sides are defined in self.padding_side:
+
+    #                 - 'left': pads on the left of the sequences
+    #                 - 'right': pads on the right of the sequences
+    #         pad_to_multiple_of: (optional) Integer if set will pad the sequence to a multiple of the provided value.
+    #             This is especially useful to enable the use of Tensor Core on NVIDIA hardware with compute capability
+    #             >= 7.5 (Volta).
+    #         return_attention_mask:
+    #             (optional) Set to False to avoid returning attention mask (default: set to model specifics)
+    #     """
+    #     # Load from model defaults
+    #     if return_attention_mask is None:
+    #         return_attention_mask = "attention_mask" in self.model_input_names
+
+    #     required_input = encoded_inputs[self.model_input_names[0]]
+
+    #     if padding_strategy == PaddingStrategy.LONGEST:
+    #         max_length = len(required_input)
+
+    #     if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
+    #         max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of
+
+    #     needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length
+
+    #     if needs_to_be_padded:
+    #         difference = max_length - len(required_input)
+    #         if self.padding_side == "right":
+    #             if return_attention_mask:
+    #                 encoded_inputs["attention_mask"] = [1] * len(required_input) + [0] * difference
+    #             if "token_type_ids" in encoded_inputs:
+    #                 encoded_inputs["token_type_ids"] = (
+    #                     encoded_inputs["token_type_ids"] + [self.pad_token_type_id] * difference
+    #                 )
+    #             if "special_tokens_mask" in encoded_inputs:
+    #                 encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"] + [1] * difference
+    #             encoded_inputs[self.model_input_names[0]] = required_input + [self.pad_token_id] * difference
+    #         elif self.padding_side == "left":
+    #             if return_attention_mask:
+    #                 encoded_inputs["attention_mask"] = [0] * difference + [1] * len(required_input)
+    #             if "token_type_ids" in encoded_inputs:
+    #                 encoded_inputs["token_type_ids"] = [self.pad_token_type_id] * difference + encoded_inputs[
+    #                     "token_type_ids"
+    #                 ]
+    #             if "special_tokens_mask" in encoded_inputs:
+    #                 encoded_inputs["special_tokens_mask"] = [1] * difference + encoded_inputs["special_tokens_mask"]
+    #             encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
+    #         else:
+    #             raise ValueError("Invalid padding strategy:" + str(self.padding_side))
+    #     elif return_attention_mask and "attention_mask" not in encoded_inputs:
+    #         if isinstance(encoded_inputs["token_type_ids"], list):
+    #             encoded_inputs["attention_mask"] = [1] * len(required_input)
+    #         else:
+    #             encoded_inputs["attention_mask"] = 1
+
+    #     return encoded_inputs
+
+    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        logger.warning("CharacterBERT does not have a token vocabulary. " "Skipping saving `vocab.txt`.")
+        return ()
+
+    def save_mlm_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        # NOTE: CharacterBERT has no token vocabulary, this is just to allow
+        # saving tokenizer configuration via CharacterBertTokenizer.save_pretrained
+        if os.path.isdir(save_directory):
+            vocab_file = os.path.join(
+                save_directory, (filename_prefix + "-" if filename_prefix else "") + "mlm_vocab.txt"
+            )
+        else:
+            vocab_file = (filename_prefix + "-" if filename_prefix else "") + save_directory
+        with open(vocab_file, "w", encoding="utf-8") as f:
+            for _, token in self.ids_to_tokens.items():
+                f.write(token + "\n")
+        return (vocab_file,)
+
+    def _save_pretrained(
+        self,
+        save_directory: Union[str, os.PathLike],
+        file_names: Tuple[str],
+        legacy_format: Optional[bool] = None,
+        filename_prefix: Optional[str] = None,
+    ) -> Tuple[str]:
+        """
+        Save a tokenizer using the slow-tokenizer/legacy format: vocabulary + added tokens.
+
+        Fast tokenizers can also be saved in a unique JSON file containing {config + vocab + added-tokens} using the
+        specific [`~tokenization_utils_fast.PreTrainedTokenizerFast._save_pretrained`]
+        """
+        if legacy_format is False:
+            raise ValueError(
+                "Only fast tokenizers (instances of PreTrainedTokenizerFast) can be saved in non legacy format."
+            )
+
+        save_directory = str(save_directory)
+
+        added_tokens_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + ADDED_TOKENS_FILE
+        )
+        added_vocab = self.get_added_vocab()
+        if added_vocab:
+            with open(added_tokens_file, "w", encoding="utf-8") as f:
+                out_str = json.dumps(added_vocab, ensure_ascii=False)
+                f.write(out_str)
+                logger.info(f"added tokens file saved in {added_tokens_file}")
+
+        vocab_files = self.save_mlm_vocabulary(save_directory, filename_prefix=filename_prefix)
+
+        return file_names + vocab_files + (added_tokens_file,)
+
+
+class BasicTokenizer(object):
+    """
+    Constructs a BasicTokenizer that will run basic tokenization (punctuation splitting, lower casing, etc.).
+
+    Args:
+        do_lower_case (`bool`, *optional*, defaults to `True`):
+            Whether or not to lowercase the input when tokenizing.
+        never_split (`Iterable`, *optional*):
+            Collection of tokens which will never be split during tokenization. Only has an effect when
+            `do_basic_tokenize=True`
+        tokenize_chinese_chars (`bool`, *optional*, defaults to `True`):
+            Whether or not to tokenize Chinese characters.
+
+            This should likely be deactivated for Japanese (see this [issue](https://github.com/huggingface/transformers/issues/328)).
+        strip_accents: (`bool`, *optional*):
+            Whether or not to strip all accents. If this option is not specified, then it will be determined by the
+            value for `lowercase` (as in the original BERT).
+    """
+
+    def __init__(self, do_lower_case=True, never_split=None, tokenize_chinese_chars=True, strip_accents=None):
+        if never_split is None:
+            never_split = []
+        self.do_lower_case = do_lower_case
+        self.never_split = set(never_split)
+        self.tokenize_chinese_chars = tokenize_chinese_chars
+        self.strip_accents = strip_accents
+
+    def tokenize(self, text, never_split=None):
+        """
+        Basic Tokenization of a piece of text. Split on "white spaces" only, for sub-word tokenization, see
+        WordPieceTokenizer.
+
+        Args:
+            **never_split**: (*optional*) list of str
+                Kept for backward compatibility purposes. Now implemented directly at the base class level (see
+                [`PreTrainedTokenizer.tokenize`]) List of token not to split.
+        """
+        # union() returns a new set by concatenating the two sets.
+        never_split = self.never_split.union(set(never_split)) if never_split else self.never_split
+        text = self._clean_text(text)
+
+        # This was added on November 1st, 2018 for the multilingual and Chinese
+        # models. This is also applied to the English models now, but it doesn't
+        # matter since the English models were not trained on any Chinese data
+        # and generally don't have any Chinese data in them (there are Chinese
+        # characters in the vocabulary because Wikipedia does have some Chinese
+        # words in the English Wikipedia.).
+        if self.tokenize_chinese_chars:
+            text = self._tokenize_chinese_chars(text)
+        orig_tokens = whitespace_tokenize(text)
+        split_tokens = []
+        for token in orig_tokens:
+            if token not in never_split:
+                if self.do_lower_case:
+                    token = token.lower()
+                    if self.strip_accents is not False:
+                        token = self._run_strip_accents(token)
+                elif self.strip_accents:
+                    token = self._run_strip_accents(token)
+            split_tokens.extend(self._run_split_on_punc(token, never_split))
+
+        output_tokens = whitespace_tokenize(" ".join(split_tokens))
+        return output_tokens
+
+    def _run_strip_accents(self, text):
+        """Strips accents from a piece of text."""
+        text = unicodedata.normalize("NFD", text)
+        output = []
+        for char in text:
+            cat = unicodedata.category(char)
+            if cat == "Mn":
+                continue
+            output.append(char)
+        return "".join(output)
+
+    def _run_split_on_punc(self, text, never_split=None):
+        """Splits punctuation on a piece of text."""
+        if never_split is not None and text in never_split:
+            return [text]
+        chars = list(text)
+        i = 0
+        start_new_word = True
+        output = []
+        while i < len(chars):
+            char = chars[i]
+            if _is_punctuation(char):
+                output.append([char])
+                start_new_word = True
+            else:
+                if start_new_word:
+                    output.append([])
+                start_new_word = False
+                output[-1].append(char)
+            i += 1
+
+        return ["".join(x) for x in output]
+
+    def _tokenize_chinese_chars(self, text):
+        """Adds whitespace around any CJK character."""
+        output = []
+        for char in text:
+            cp = ord(char)
+            if self._is_chinese_char(cp):
+                output.append(" ")
+                output.append(char)
+                output.append(" ")
+            else:
+                output.append(char)
+        return "".join(output)
+
+    def _is_chinese_char(self, cp):
+        """Checks whether CP is the codepoint of a CJK character."""
+        # This defines a "chinese character" as anything in the CJK Unicode block:
+        #   https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
+        #
+        # Note that the CJK Unicode block is NOT all Japanese and Korean characters,
+        # despite its name. The modern Korean Hangul alphabet is a different block,
+        # as is Japanese Hiragana and Katakana. Those alphabets are used to write
+        # space-separated words, so they are not treated specially and handled
+        # like the all of the other languages.
+        if (
+            (cp >= 0x4E00 and cp <= 0x9FFF)
+            or (cp >= 0x3400 and cp <= 0x4DBF)  #
+            or (cp >= 0x20000 and cp <= 0x2A6DF)  #
+            or (cp >= 0x2A700 and cp <= 0x2B73F)  #
+            or (cp >= 0x2B740 and cp <= 0x2B81F)  #
+            or (cp >= 0x2B820 and cp <= 0x2CEAF)  #
+            or (cp >= 0xF900 and cp <= 0xFAFF)
+            or (cp >= 0x2F800 and cp <= 0x2FA1F)  #
+        ):  #
+            return True
+
+        return False
+
+    def _clean_text(self, text):
+        """Performs invalid character removal and whitespace cleanup on text."""
+        output = []
+        for char in text:
+            cp = ord(char)
+            if cp == 0 or cp == 0xFFFD or _is_control(char):
+                continue
+            if _is_whitespace(char):
+                output.append(" ")
+            else:
+                output.append(char)
+        return "".join(output)
+
+
+class CharacterMapper:
+    """
+    NOTE: Adapted from ElmoCharacterMapper:
+    https://github.com/allenai/allennlp/blob/main/allennlp/data/token_indexers/elmo_indexer.py Maps individual tokens
+    to sequences of character ids, compatible with CharacterBERT.
+    """
+
+    # char ids 0-255 come from utf-8 encoding bytes
+    # assign 256-300 to special chars
+    beginning_of_sentence_character = 256  # <begin sentence>
+    end_of_sentence_character = 257  # <end sentence>
+    beginning_of_word_character = 258  # <begin word>
+    end_of_word_character = 259  # <end word>
+    padding_character = 260  # <padding> | short tokens are padded using this + 1
+    mask_character = 261  # <mask>
+
+    bos_token = "[CLS]"  # previously: bos_token = "<S>"
+    eos_token = "[SEP]"  # previously: eos_token = "</S>"
+    pad_token = "[PAD]"
+    mask_token = "[MASK]"
+
+    def __init__(
+        self,
+        max_word_length: int = 50,
+    ):
+        self.max_word_length = max_word_length
+        self.beginning_of_sentence_characters = self._make_char_id_sequence(self.beginning_of_sentence_character)
+        self.end_of_sentence_characters = self._make_char_id_sequence(self.end_of_sentence_character)
+        self.mask_characters = self._make_char_id_sequence(self.mask_character)
+        # This is the character id sequence for the pad token (i.e. [PAD]).
+        # We remove 1 because we will add 1 later on and it will be equal to 0.
+        self.pad_characters = [PAD_TOKEN_CHAR_ID - 1] * self.max_word_length
+
+    def _make_char_id_sequence(self, character: int):
+        char_ids = [self.padding_character] * self.max_word_length
+        char_ids[0] = self.beginning_of_word_character
+        char_ids[1] = character
+        char_ids[2] = self.end_of_word_character
+        return char_ids
+
+    def convert_word_to_char_ids(self, word: str) -> List[int]:
+        if word == self.bos_token:
+            char_ids = self.beginning_of_sentence_characters
+        elif word == self.eos_token:
+            char_ids = self.end_of_sentence_characters
+        elif word == self.mask_token:
+            char_ids = self.mask_characters
+        elif word == self.pad_token:
+            char_ids = self.pad_characters
+        else:
+            # Convert characters to indices
+            word_encoded = word.encode("utf-8", "ignore")[: (self.max_word_length - 2)]
+            # Initialize character_ids with padding
+            char_ids = [self.padding_character] * self.max_word_length
+            # First character is BeginningOfWord
+            char_ids[0] = self.beginning_of_word_character
+            # Populate character_ids with computed indices
+            for k, chr_id in enumerate(word_encoded, start=1):
+                char_ids[k] = chr_id
+            # Last character is EndOfWord
+            char_ids[len(word_encoded) + 1] = self.end_of_word_character
+
+        # +1 one for masking so that character padding == 0
+        # char_ids domain is therefore: (1, 256) for actual characters
+        # and (257-262) for special symbols (BOS/EOS/BOW/EOW/padding/MLM Mask)
+        return [c + 1 for c in char_ids]
+
+    def convert_char_ids_to_word(self, char_ids: List[int]) -> str:
+        "Converts a sequence of character ids into its corresponding word."
+
+        assert len(char_ids) <= self.max_word_length, (
+            f"Got character sequence of length {len(char_ids)} while `max_word_length={self.max_word_length}`"
+        )
+
+        char_ids_ = [(i - 1) for i in char_ids]
+        if char_ids_ == self.beginning_of_sentence_characters:
+            return self.bos_token
+        elif char_ids_ == self.end_of_sentence_characters:
+            return self.eos_token
+        elif char_ids_ == self.mask_characters:
+            return self.mask_token
+        elif char_ids_ == self.pad_characters:  # token padding
+            return self.pad_token
+        else:
+            utf8_codes = list(
+                filter(
+                    lambda x: (x != self.padding_character)
+                    and (x != self.beginning_of_word_character)
+                    and (x != self.end_of_word_character),
+                    char_ids_,
+                )
+            )
+            return bytes(utf8_codes).decode("utf-8")

tokenizer_config.json

@@ -1 +1 @@
-{"max_word_length": 50, "do_lower_case": true, "do_basic_tokenize": true, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}
+{"name_or_path": "helboukkouri/character-bert-medical", "tokenizer_class": "CharacterBertTokenizer", "max_word_length": 50, "do_lower_case": true, "do_basic_tokenize": true, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "auto_map": {"AutoTokenizer": ["tokenization_character_bert.CharacterBertTokenizer", null]}}