Add LayoutLM Model (#7064)

* first version

* finish test docs readme model/config/tokenization class

* apply make style and make quality

* fix layoutlm GitHub link

* fix conflict in index.rst and add layoutlm to pretrained_models.rst

* fix bug in test_parents_and_children_in_mappings

* reformat modeling_auto.py and tokenization_auto.py

* fix bug in test_modeling_layoutlm.py

* Update docs/source/model_doc/layoutlm.rst
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update docs/source/model_doc/layoutlm.rst
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* remove inh, add tokenizer fast, and update some doc

* copy and rename necessary class from modeling_bert to modeling_layoutlm

* Update src/transformers/configuration_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/configuration_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/configuration_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/configuration_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/modeling_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/modeling_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update src/transformers/modeling_layoutlm.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* add mish to activations.py, import ACT2FN and import logging from utils
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

README.md

@@ -183,8 +183,9 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 24. **[MBart](https://github.com/pytorch/fairseq/tree/master/examples/mbart)** (from Facebook) released with the paper  [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.  
 25. **[LXMERT](https://github.com/airsplay/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
 26. **[Funnel Transformer](https://github.com/laiguokun/Funnel-Transformer)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-27. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
-28. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
+27. **[LayoutLM](https://github.com/microsoft/unilm/tree/master/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+28. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
+29. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
 These implementations have been tested on several datasets (see the example scripts) and should match the performances of the original implementations. You can find more details on the performances in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
 

docs/source/index.rst

@@ -137,7 +137,10 @@ conversion utilities for the following models:
 27. `Bert For Sequence Generation <https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder>`_ (from Google) released with the paper
     `Leveraging Pre-trained Checkpoints for Sequence Generation Tasks
     <https://arxiv.org/abs/1907.12461>`_ by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-28. `Other community models <https://huggingface.co/models>`_, contributed by the `community
+28. `LayoutLM <https://github.com/microsoft/unilm/tree/master/layoutlm>`_ (from Microsoft Research Asia) released with the paper
+    `LayoutLM: Pre-training of Text and Layout for Document Image Understanding
+    <https://arxiv.org/abs/1912.13318>`_ by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+29. `Other community models <https://huggingface.co/models>`_, contributed by the `community
     <https://huggingface.co/users>`_.
 
 .. toctree::
@@ -227,6 +230,7 @@ conversion utilities for the following models:
     model_doc/funnel
     model_doc/lxmert
     model_doc/bertgeneration
+    model_doc/layoutlm
     internal/modeling_utils
     internal/tokenization_utils
     internal/pipelines_utils

docs/source/model_doc/layoutlm.rst

+LayoutLM
+----------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~
+
+The LayoutLM model was proposed in `LayoutLM: Pre-training of Text and Layout for Document Image Understanding <https://arxiv.org/abs/1912.13318>`__
+by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, and Ming Zhou. It's a simple but effective pre-training method 
+of text and layout for document image understanding and information extraction tasks, such as form understanding and receipt understanding.
+
+The abstract from the paper is the following:
+
+*Pre-training techniques have been verified successfully in a variety of NLP tasks in recent years. Despite the widespread use of pre-training models for NLP applications, they almost exclusively focus on text-level manipulation, while neglecting layout and style information that is vital for document image understanding. In this paper, we propose the \textbf{LayoutLM} to jointly model interactions between text and layout information across scanned document images, which is beneficial for a great number of real-world document image understanding tasks such as information extraction from scanned documents. Furthermore, we also leverage image features to incorporate words' visual information into LayoutLM. To the best of our knowledge, this is the first time that text and layout are jointly learned in a single framework for document-level pre-training. It achieves new state-of-the-art results in several downstream tasks, including form understanding (from 70.72 to 79.27), receipt understanding (from 94.02 to 95.24) and document image classification (from 93.07 to 94.42).*
+
+Tips:
+
+- LayoutLM has an extra input called :obj:`bbox`, which is the bounding boxes of the input tokens.
+- The :obj:`bbox` requires the data that on 0-1000 scale, which means you should normalize the bounding box before passing them into model.
+
+The original code can be found `here <https://github.com/microsoft/unilm/tree/master/layoutlm>`_.
+
+
+LayoutLMConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMConfig
+    :members:
+
+
+LayoutLMTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMTokenizer
+    :members:
+
+
+LayoutLMModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMModel
+    :members:
+
+
+LayoutLMForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMForMaskedLM
+    :members:
+
+
+LayoutLMForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMForTokenClassification
+    :members:

docs/source/pretrained_models.rst

@@ -408,3 +408,11 @@ For a list that includes community-uploaded models, refer to `https://huggingfac
 |                    |                                                            |                                                                                                                                       |
 |                    |                                                            | (see `details <https://github.com/laiguokun/Funnel-Transformer>`__)                                                                   |
 +--------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+| LayoutLM           | ``microsoft/layoutlm-base-uncased``                        | | 12 layers, 768-hidden, 12-heads, 113M parameters                                                                                    |
+|                    |                                                            |                                                                                                                                       |
+|                    |                                                            | (see `details <https://github.com/microsoft/unilm/tree/master/layoutlm>`__)                                                           |
++                    +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                    | ``microsoft/layoutlm-large-uncased``                       | | 24 layers, 1024-hidden, 16-heads, 343M parameters                                                                                   |
+|                    |                                                            |                                                                                                                                       |
+|                    |                                                            | (see `details <https://github.com/microsoft/unilm/tree/master/layoutlm>`__)                                                           |
++--------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
\ No newline at end of file

src/transformers/__init__.py

@@ -33,6 +33,7 @@ from .configuration_flaubert import FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, Flau
 from .configuration_fsmt import FSMT_PRETRAINED_CONFIG_ARCHIVE_MAP, FSMTConfig
 from .configuration_funnel import FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP, FunnelConfig
 from .configuration_gpt2 import GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP, GPT2Config
+from .configuration_layoutlm import LAYOUTLM_PRETRAINED_CONFIG_ARCHIVE_MAP, LayoutLMConfig
 from .configuration_longformer import LONGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, LongformerConfig
 from .configuration_lxmert import LXMERT_PRETRAINED_CONFIG_ARCHIVE_MAP, LxmertConfig
 from .configuration_marian import MarianConfig
@@ -163,6 +164,7 @@ from .tokenization_flaubert import FlaubertTokenizer
 from .tokenization_fsmt import FSMTTokenizer
 from .tokenization_funnel import FunnelTokenizer, FunnelTokenizerFast
 from .tokenization_gpt2 import GPT2Tokenizer, GPT2TokenizerFast
+from .tokenization_layoutlm import LayoutLMTokenizer, LayoutLMTokenizerFast
 from .tokenization_longformer import LongformerTokenizer, LongformerTokenizerFast
 from .tokenization_lxmert import LxmertTokenizer, LxmertTokenizerFast
 from .tokenization_mbart import MBartTokenizer
@@ -363,6 +365,12 @@ if is_torch_available():
         GPT2PreTrainedModel,
         load_tf_weights_in_gpt2,
     )
+    from .modeling_layoutlm import (
+        LAYOUTLM_PRETRAINED_MODEL_ARCHIVE_LIST,
+        LayoutLMForMaskedLM,
+        LayoutLMForTokenClassification,
+        LayoutLMModel,
+    )
     from .modeling_longformer import (
         LONGFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
         LongformerForMaskedLM,

src/transformers/activations.py

@@ -55,3 +55,7 @@ def get_activation(activation_string):
         return ACT2FN[activation_string]
     else:
         raise KeyError("function {} not found in ACT2FN mapping {}".format(activation_string, list(ACT2FN.keys())))
+
+
+def mish(x):
+    return x * torch.tanh(torch.nn.functional.softplus(x))

src/transformers/configuration_auto.py

@@ -30,6 +30,7 @@ from .configuration_flaubert import FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, Flau
 from .configuration_fsmt import FSMT_PRETRAINED_CONFIG_ARCHIVE_MAP, FSMTConfig
 from .configuration_funnel import FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP, FunnelConfig
 from .configuration_gpt2 import GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP, GPT2Config
+from .configuration_layoutlm import LAYOUTLM_PRETRAINED_CONFIG_ARCHIVE_MAP, LayoutLMConfig
 from .configuration_longformer import LONGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, LongformerConfig
 from .configuration_lxmert import LXMERT_PRETRAINED_CONFIG_ARCHIVE_MAP, LxmertConfig
 from .configuration_marian import MarianConfig
@@ -73,6 +74,7 @@ ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
         RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
         FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP,
         LXMERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        LAYOUTLM_PRETRAINED_CONFIG_ARCHIVE_MAP,
     ]
     for key, value, in pretrained_map.items()
 )
@@ -108,6 +110,7 @@ CONFIG_MAPPING = OrderedDict(
         ("encoder-decoder", EncoderDecoderConfig),
         ("funnel", FunnelConfig),
         ("lxmert", LxmertConfig),
+        ("layoutlm", LayoutLMConfig),
     ]
 )
 
@@ -141,6 +144,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("encoder-decoder", "Encoder decoder"),
         ("funnel", "Funnel Transformer"),
         ("lxmert", "LXMERT"),
+        ("layoutlm", "LayoutLM"),
     ]
 )
 

src/transformers/configuration_layoutlm.py

+# coding=utf-8
+# Copyright 2010, The Microsoft Research Asia LayoutLM Team authors
+#
+# 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.
+""" LayoutLM model configuration """
+
+
+from .configuration_bert import BertConfig
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+LAYOUTLM_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "layoutlm-base-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/microsoft/layoutlm-base-uncased/config.json",
+    "layoutlm-large-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/microsoft/layoutlm-large-uncased/config.json",
+}
+
+
+class LayoutLMConfig(BertConfig):
+    r"""
+    This is the configuration class to store the configuration of a :class:`~transformers.LayoutLMModel`.
+    It is used to instantiate a LayoutLM 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 LayoutLM `layoutlm-base-uncased <https://huggingface.co/microsoft/layoutlm-base-uncased>`__ architecture.
+
+    Configuration objects inherit from :class:`~transformers.BertConfig` and can be used
+    to control the model outputs. Read the documentation from :class:`~transformers.BertConfig`
+    for more information.
+
+
+    Args:
+        vocab_size (:obj:`int`, optional, defaults to 30522):
+            Vocabulary size of the LayoutLM model. Defines the different tokens that
+            can be represented by the `inputs_ids` passed to the forward method of :class:`~transformers.LayoutLMModel`.
+        hidden_size (:obj:`int`, optional, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_hidden_layers (:obj:`int`, optional, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (:obj:`int`, optional, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (:obj:`int`, optional, defaults to 3072):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        hidden_act (:obj:`str` or :obj:`function`, optional, defaults to "gelu"):
+            The non-linear activation function (function or string) in the encoder and pooler.
+            If string, "gelu", "relu", "swish" and "gelu_new" are supported.
+        hidden_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+            The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings (:obj:`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 (:obj:`int`, optional, defaults to 2):
+            The vocabulary size of the `token_type_ids` passed into :class:`~transformers.BertModel`.
+        initializer_range (:obj:`float`, optional, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (:obj:`float`, optional, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        gradient_checkpointing (:obj:`bool`, optional, defaults to :obj:`False`):
+            If True, use gradient checkpointing to save memory at the expense of slower backward pass.
+        max_2d_position_embeddings (:obj:`int`, optional, defaults to 1024):
+            The maximum value that the 2D position embedding might ever used.
+            Typically set this to something large just in case (e.g., 1024).
+
+    Example::
+
+        >>> from transformers import LayoutLMModel, LayoutLMConfig
+
+        >>> # Initializing a LayoutLM configuration
+        >>> configuration = LayoutLMConfig()
+
+        >>> # Initializing a model from the configuration
+        >>> model = LayoutLMModel(configuration)
+
+        >>> # Accessing the model configuration
+        >>> configuration = model.config
+
+    """
+    model_type = "layoutlm"
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        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,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=0,
+        gradient_checkpointing=False,
+        max_2d_position_embeddings=1024,
+        **kwargs
+    ):
+        super().__init__(
+            vocab_size=vocab_size,
+            hidden_size=hidden_size,
+            num_hidden_layers=num_hidden_layers,
+            num_attention_heads=num_attention_heads,
+            intermediate_size=intermediate_size,
+            hidden_act=hidden_act,
+            hidden_dropout_prob=hidden_dropout_prob,
+            attention_probs_dropout_prob=attention_probs_dropout_prob,
+            max_position_embeddings=max_position_embeddings,
+            type_vocab_size=type_vocab_size,
+            initializer_range=initializer_range,
+            layer_norm_eps=layer_norm_eps,
+            pad_token_id=pad_token_id,
+            gradient_checkpointing=gradient_checkpointing,
+            **kwargs,
+        )
+        self.max_2d_position_embeddings = max_2d_position_embeddings

src/transformers/modeling_auto.py

@@ -33,6 +33,7 @@ from .configuration_auto import (
     FSMTConfig,
     FunnelConfig,
     GPT2Config,
+    LayoutLMConfig,
     LongformerConfig,
     LxmertConfig,
     MBartConfig,
@@ -124,6 +125,7 @@ from .modeling_funnel import (
     FunnelModel,
 )
 from .modeling_gpt2 import GPT2LMHeadModel, GPT2Model
+from .modeling_layoutlm import LayoutLMForMaskedLM, LayoutLMForTokenClassification, LayoutLMModel
 from .modeling_longformer import (
     LongformerForMaskedLM,
     LongformerForMultipleChoice,
@@ -206,6 +208,7 @@ MODEL_MAPPING = OrderedDict(
         (BartConfig, BartModel),
         (LongformerConfig, LongformerModel),
         (RobertaConfig, RobertaModel),
+        (LayoutLMConfig, LayoutLMModel),
         (BertConfig, BertModel),
         (OpenAIGPTConfig, OpenAIGPTModel),
         (GPT2Config, GPT2Model),
@@ -226,6 +229,7 @@ MODEL_MAPPING = OrderedDict(
 
 MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
     [
+        (LayoutLMConfig, LayoutLMForMaskedLM),
         (RetriBertConfig, RetriBertModel),
         (T5Config, T5ForConditionalGeneration),
         (DistilBertConfig, DistilBertForMaskedLM),
@@ -252,6 +256,7 @@ MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
 
 MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
     [
+        (LayoutLMConfig, LayoutLMForMaskedLM),
         (T5Config, T5ForConditionalGeneration),
         (DistilBertConfig, DistilBertForMaskedLM),
         (AlbertConfig, AlbertForMaskedLM),
@@ -300,6 +305,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING = OrderedDict(
 
 MODEL_FOR_MASKED_LM_MAPPING = OrderedDict(
     [
+        (LayoutLMConfig, LayoutLMForMaskedLM),
         (DistilBertConfig, DistilBertForMaskedLM),
         (AlbertConfig, AlbertForMaskedLM),
         (BartConfig, BartForConditionalGeneration),
@@ -370,6 +376,7 @@ MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
 
 MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
     [
+        (LayoutLMConfig, LayoutLMForTokenClassification),
         (DistilBertConfig, DistilBertForTokenClassification),
         (CamembertConfig, CamembertForTokenClassification),
         (FlaubertConfig, FlaubertForTokenClassification),

src/transformers/tokenization_auto.py

@@ -32,6 +32,7 @@ from .configuration_auto import (
     FSMTConfig,
     FunnelConfig,
     GPT2Config,
+    LayoutLMConfig,
     LongformerConfig,
     LxmertConfig,
     MarianConfig,
@@ -64,6 +65,7 @@ from .tokenization_flaubert import FlaubertTokenizer
 from .tokenization_fsmt import FSMTTokenizer
 from .tokenization_funnel import FunnelTokenizer, FunnelTokenizerFast
 from .tokenization_gpt2 import GPT2Tokenizer, GPT2TokenizerFast
+from .tokenization_layoutlm import LayoutLMTokenizer, LayoutLMTokenizerFast
 from .tokenization_longformer import LongformerTokenizer, LongformerTokenizerFast
 from .tokenization_lxmert import LxmertTokenizer, LxmertTokenizerFast
 from .tokenization_marian import MarianTokenizer
@@ -107,6 +109,7 @@ TOKENIZER_MAPPING = OrderedDict(
         (ElectraConfig, (ElectraTokenizer, ElectraTokenizerFast)),
         (FunnelConfig, (FunnelTokenizer, FunnelTokenizerFast)),
         (LxmertConfig, (LxmertTokenizer, LxmertTokenizerFast)),
+        (LayoutLMConfig, (LayoutLMTokenizer, LayoutLMTokenizerFast)),
         (BertConfig, (BertTokenizer, BertTokenizerFast)),
         (OpenAIGPTConfig, (OpenAIGPTTokenizer, OpenAIGPTTokenizerFast)),
         (GPT2Config, (GPT2Tokenizer, GPT2TokenizerFast)),
@@ -117,6 +120,7 @@ TOKENIZER_MAPPING = OrderedDict(
         (CTRLConfig, (CTRLTokenizer, None)),
         (FSMTConfig, (FSMTTokenizer, None)),
         (BertGenerationConfig, (BertGenerationTokenizer, None)),
+        (LayoutLMConfig, (LayoutLMTokenizer, None)),
     ]
 )
 

src/transformers/tokenization_layoutlm.py

+# coding=utf-8
+# Copyright 2018 The Microsoft Research Asia LayoutLM Team Authors.
+#
+# 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 class for model LayoutLM."""
+
+
+from .tokenization_bert import BertTokenizer, BertTokenizerFast
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "microsoft/layoutlm-base-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt",
+        "microsoft/layoutlm-large-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-vocab.txt",
+    }
+}
+
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "microsoft/layoutlm-base-uncased": 512,
+    "microsoft/layoutlm-large-uncased": 512,
+}
+
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "microsoft/layoutlm-base-uncased": {"do_lower_case": True},
+    "microsoft/layoutlm-large-uncased": {"do_lower_case": True},
+}
+
+
+class LayoutLMTokenizer(BertTokenizer):
+    r"""
+    Constructs a LayoutLM tokenizer.
+
+    :class:`~transformers.LayoutLMTokenizer is identical to :class:`~transformers.BertTokenizer` and runs end-to-end
+    tokenization: punctuation splitting + wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizer` for usage examples and documentation concerning
+    parameters.
+    """
+
+    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
+
+
+class LayoutLMTokenizerFast(BertTokenizerFast):
+    r"""
+    Constructs a  "Fast" LayoutLMTokenizer.
+
+    :class:`~transformers.LayoutLMTokenizerFast` is identical to :class:`~transformers.BertTokenizerFast` and runs end-to-end
+    tokenization: punctuation splitting + wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizerFast` for usage examples and documentation concerning
+    parameters.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+    model_input_names = ["attention_mask"]

tests/test_modeling_layoutlm.py

+# coding=utf-8
+# Copyright 2018 The Microsoft Research Asia LayoutLM Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import unittest
+
+from transformers import is_torch_available
+from transformers.file_utils import cached_property
+from transformers.testing_utils import require_torch, require_torch_and_cuda, slow, torch_device
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_common import ModelTesterMixin, ids_tensor
+
+
+if is_torch_available():
+    from transformers import LayoutLMConfig, LayoutLMForMaskedLM, LayoutLMForTokenClassification, LayoutLMModel
+
+
+class LayoutLMModelTester:
+    """You can also import this e.g from .test_modeling_bart import BartModelTester """
+
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        seq_length=7,
+        is_training=True,
+        use_input_mask=True,
+        use_token_type_ids=True,
+        use_labels=True,
+        vocab_size=99,
+        hidden_size=32,
+        num_hidden_layers=5,
+        num_attention_heads=4,
+        intermediate_size=37,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=16,
+        type_sequence_label_size=2,
+        initializer_range=0.02,
+        num_labels=3,
+        num_choices=4,
+        scope=None,
+        range_bbox=1000,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.seq_length = seq_length
+        self.is_training = is_training
+        self.use_input_mask = use_input_mask
+        self.use_token_type_ids = use_token_type_ids
+        self.use_labels = use_labels
+        self.vocab_size = vocab_size
+        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.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.type_vocab_size = type_vocab_size
+        self.type_sequence_label_size = type_sequence_label_size
+        self.initializer_range = initializer_range
+        self.num_labels = num_labels
+        self.num_choices = num_choices
+        self.scope = scope
+        self.range_bbox = range_bbox
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+
+        bbox = ids_tensor([self.batch_size, self.seq_length, 4], self.range_bbox)
+        # Ensure that bbox is legal
+        for i in range(bbox.shape[0]):
+            for j in range(bbox.shape[1]):
+                if bbox[i, j, 3] < bbox[i, j, 1]:
+                    t = bbox[i, j, 3]
+                    bbox[i, j, 3] = bbox[i, j, 1]
+                    bbox[i, j, 1] = t
+                if bbox[i, j, 2] < bbox[i, j, 0]:
+                    t = bbox[i, j, 2]
+                    bbox[i, j, 2] = bbox[i, j, 0]
+                    bbox[i, j, 0] = t
+
+        input_mask = None
+        if self.use_input_mask:
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+        token_type_ids = None
+        if self.use_token_type_ids:
+            token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
+
+        sequence_labels = None
+        token_labels = None
+        choice_labels = None
+        if self.use_labels:
+            sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+            token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
+            choice_labels = ids_tensor([self.batch_size], self.num_choices)
+
+        config = LayoutLMConfig(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            num_hidden_layers=self.num_hidden_layers,
+            num_attention_heads=self.num_attention_heads,
+            intermediate_size=self.intermediate_size,
+            hidden_act=self.hidden_act,
+            hidden_dropout_prob=self.hidden_dropout_prob,
+            attention_probs_dropout_prob=self.attention_probs_dropout_prob,
+            max_position_embeddings=self.max_position_embeddings,
+            type_vocab_size=self.type_vocab_size,
+            initializer_range=self.initializer_range,
+            return_dict=True,
+        )
+
+        return config, input_ids, bbox, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def create_and_check_model(
+        self, config, input_ids, bbox, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = LayoutLMModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, bbox, attention_mask=input_mask, token_type_ids=token_type_ids)
+        result = model(input_ids, bbox, token_type_ids=token_type_ids)
+        result = model(input_ids, bbox)
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+        self.parent.assertEqual(result.pooler_output.shape, (self.batch_size, self.hidden_size))
+
+    def create_and_check_for_masked_lm(
+        self, config, input_ids, bbox, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = LayoutLMForMaskedLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, bbox, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_for_token_classification(
+        self, config, input_ids, bbox, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = LayoutLMForTokenClassification(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, bbox, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            bbox,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {
+            "input_ids": input_ids,
+            "bbox": bbox,
+            "token_type_ids": token_type_ids,
+            "attention_mask": input_mask,
+        }
+        return config, inputs_dict
+
+
+@require_torch
+class LayoutLMModelTest(ModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (LayoutLMModel, LayoutLMForMaskedLM, LayoutLMForTokenClassification) if is_torch_available() else ()
+    )
+
+    def setUp(self):
+        self.model_tester = LayoutLMModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=LayoutLMConfig, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_for_masked_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)
+
+    def test_for_token_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_token_classification(*config_and_inputs)
+
+    @cached_property
+    def big_model(self):
+        """Cached property means this code will only be executed once."""
+        checkpoint_path = "microsoft/layoutlm-large-uncased"
+        model = LayoutLMForMaskedLM.from_pretrained(checkpoint_path).to(
+            torch_device
+        )  # test whether AutoModel can determine your model_class from checkpoint name
+        if torch_device == "cuda":
+            model.half()
+
+    # optional: do more testing! This will save you time later!
+    @slow
+    def test_that_LayoutLM_can_be_used_in_a_pipeline(self):
+        """We can use self.big_model here without calling __init__ again."""
+        pass
+
+    def test_LayoutLM_loss_doesnt_change_if_you_add_padding(self):
+        pass
+
+    def test_LayoutLM_bad_args(self):
+        pass
+
+    def test_LayoutLM_backward_pass_reduces_loss(self):
+        """Test loss/gradients same as reference implementation, for example."""
+        pass
+
+    @require_torch_and_cuda
+    def test_large_inputs_in_fp16_dont_cause_overflow(self):
+        pass

tests/test_tokenization_layoutlm.py

+# coding=utf-8
+# Copyright 2018 The Microsoft Research Asia LayoutLM Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import os
+import unittest
+
+from transformers.tokenization_layoutlm import VOCAB_FILES_NAMES, LayoutLMTokenizer
+
+from .test_tokenization_common import TokenizerTesterMixin
+
+
+class LayoutLMTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+
+    tokenizer_class = LayoutLMTokenizer
+
+    def setUp(self):
+        super().setUp()
+
+        vocab_tokens = [
+            "[UNK]",
+            "[CLS]",
+            "[SEP]",
+            "want",
+            "##want",
+            "##ed",
+            "wa",
+            "un",
+            "runn",
+            "##ing",
+            ",",
+            "low",
+            "lowest",
+        ]
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
+            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
+
+    def get_tokenizer(self, **kwargs):
+        return LayoutLMTokenizer.from_pretrained(self.tmpdirname, **kwargs)
+
+    def get_input_output_texts(self, tokenizer):
+        input_text = "UNwant\u00E9d,running"
+        output_text = "unwanted, running"
+        return input_text, output_text
+
+    def test_full_tokenizer(self):
+        tokenizer = self.tokenizer_class(self.vocab_file)
+
+        tokens = tokenizer.tokenize("UNwant\u00E9d,running")
+        self.assertListEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
+        self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [7, 4, 5, 10, 8, 9])
+
+    def test_special_tokens_as_you_expect(self):
+        """If you are training a seq2seq model that expects a decoder_prefix token make sure it is prepended to decoder_input_ids """
+        pass