Add LayoutXLMProcessor (and LayoutXLMTokenizer, LayoutXLMTokenizerFast)

docs/source/model_doc/layoutxlm.rst

@@ -40,17 +40,45 @@ One can directly plug in the weights of LayoutXLM into a LayoutLMv2 model, like
 
     model = LayoutLMv2Model.from_pretrained('microsoft/layoutxlm-base') 
 
-Note that LayoutXLM requires a different tokenizer, based on :class:`~transformers.XLMRobertaTokenizer`. You can
-initialize it as follows:
+Note that LayoutXLM has its own tokenizer, based on
+:class:`~transformers.LayoutXLMTokenizer`/:class:`~transformers.LayoutXLMTokenizerFast`. You can initialize it as
+follows:
 
 .. code-block::
 
-    from transformers import AutoTokenizer
+    from transformers import LayoutXLMTokenizer
 
-    tokenizer = AutoTokenizer.from_pretrained('microsoft/layoutxlm-base') 
+    tokenizer = LayoutXLMTokenizer.from_pretrained('microsoft/layoutxlm-base') 
+
+Similar to LayoutLMv2, you can use :class:`~transformers.LayoutXLMProcessor` (which internally applies
+:class:`~transformers.LayoutLMv2FeatureExtractor` and
+:class:`~transformers.LayoutXLMTokenizer`/:class:`~transformers.LayoutXLMTokenizerFast` in sequence) to prepare all
+data for the model.
 
 As LayoutXLM's architecture is equivalent to that of LayoutLMv2, one can refer to :doc:`LayoutLMv2's documentation page
 <layoutlmv2>` for all tips, code examples and notebooks.
 
 This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code can be found `here
 <https://github.com/microsoft/unilm>`__.
+
+
+LayoutXLMTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutXLMTokenizer
+    :members: __call__, build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+
+LayoutXLMTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutXLMTokenizerFast
+    :members: __call__
+
+
+LayoutXLMProcessor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutXLMProcessor
+    :members: __call__

src/transformers/__init__.py

@@ -229,6 +229,7 @@
         "LayoutLMv2Processor",
         "LayoutLMv2Tokenizer",
     ],
+    "models.layoutxlm": ["LayoutXLMProcessor"],
     "models.led": ["LED_PRETRAINED_CONFIG_ARCHIVE_MAP", "LEDConfig", "LEDTokenizer"],
     "models.longformer": ["LONGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongformerConfig", "LongformerTokenizer"],
     "models.luke": ["LUKE_PRETRAINED_CONFIG_ARCHIVE_MAP", "LukeConfig", "LukeTokenizer"],
@@ -365,6 +366,7 @@
     _import_structure["models.big_bird"].append("BigBirdTokenizer")
     _import_structure["models.camembert"].append("CamembertTokenizer")
     _import_structure["models.deberta_v2"].append("DebertaV2Tokenizer")
+    _import_structure["models.layoutxlm"].append("LayoutXLMTokenizer")
     _import_structure["models.m2m_100"].append("M2M100Tokenizer")
     _import_structure["models.marian"].append("MarianTokenizer")
     _import_structure["models.mbart"].append("MBartTokenizer")
@@ -411,6 +413,7 @@
     _import_structure["models.herbert"].append("HerbertTokenizerFast")
     _import_structure["models.layoutlm"].append("LayoutLMTokenizerFast")
     _import_structure["models.layoutlmv2"].append("LayoutLMv2TokenizerFast")
+    _import_structure["models.layoutxlm"].append("LayoutXLMTokenizerFast")
     _import_structure["models.led"].append("LEDTokenizerFast")
     _import_structure["models.longformer"].append("LongformerTokenizerFast")
     _import_structure["models.lxmert"].append("LxmertTokenizerFast")
@@ -477,6 +480,7 @@
     _import_structure["models.detr"].append("DetrFeatureExtractor")
     _import_structure["models.layoutlmv2"].append("LayoutLMv2FeatureExtractor")
     _import_structure["models.layoutlmv2"].append("LayoutLMv2Processor")
+    _import_structure["models.layoutxlm"].append("LayoutXLMProcessor")
     _import_structure["models.segformer"].append("SegformerFeatureExtractor")
     _import_structure["models.vit"].append("ViTFeatureExtractor")
 else:
@@ -2140,6 +2144,7 @@
         LayoutLMv2Processor,
         LayoutLMv2Tokenizer,
     )
+    from .models.layoutxlm import LayoutXLMProcessor
     from .models.led import LED_PRETRAINED_CONFIG_ARCHIVE_MAP, LEDConfig, LEDTokenizer
     from .models.longformer import LONGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, LongformerConfig, LongformerTokenizer
     from .models.luke import LUKE_PRETRAINED_CONFIG_ARCHIVE_MAP, LukeConfig, LukeTokenizer
@@ -2266,6 +2271,7 @@
         from .models.big_bird import BigBirdTokenizer
         from .models.camembert import CamembertTokenizer
         from .models.deberta_v2 import DebertaV2Tokenizer
+        from .models.layoutxlm import LayoutXLMTokenizer
         from .models.m2m_100 import M2M100Tokenizer
         from .models.marian import MarianTokenizer
         from .models.mbart import MBart50Tokenizer, MBartTokenizer
@@ -2302,6 +2308,7 @@
         from .models.herbert import HerbertTokenizerFast
         from .models.layoutlm import LayoutLMTokenizerFast
         from .models.layoutlmv2 import LayoutLMv2TokenizerFast
+        from .models.layoutxlm import LayoutXLMTokenizerFast
         from .models.led import LEDTokenizerFast
         from .models.longformer import LongformerTokenizerFast
         from .models.lxmert import LxmertTokenizerFast
@@ -2349,6 +2356,7 @@
         from .models.deit import DeiTFeatureExtractor
         from .models.detr import DetrFeatureExtractor
         from .models.layoutlmv2 import LayoutLMv2FeatureExtractor, LayoutLMv2Processor
+        from .models.layoutxlm import LayoutXLMProcessor
         from .models.segformer import SegformerFeatureExtractor
         from .models.vit import ViTFeatureExtractor
     else:

src/transformers/convert_slow_tokenizer.py

@@ -944,6 +944,7 @@ def converted(self) -> Tokenizer:
     "HerbertTokenizer": HerbertConverter,
     "LayoutLMTokenizer": BertConverter,
     "LayoutLMv2Tokenizer": BertConverter,
+    "LayoutXLMTokenizer": XLMRobertaConverter,
     "LongformerTokenizer": RobertaConverter,
     "LEDTokenizer": RobertaConverter,
     "LxmertTokenizer": BertConverter,

src/transformers/models/__init__.py

@@ -59,6 +59,7 @@
     ibert,
     layoutlm,
     layoutlmv2,
+    layoutxlm,
     led,
     longformer,
     luke,

src/transformers/models/auto/tokenization_auto.py

@@ -124,6 +124,7 @@
             ("lxmert", ("LxmertTokenizer", "LxmertTokenizerFast" if is_tokenizers_available() else None)),
             ("layoutlm", ("LayoutLMTokenizer", "LayoutLMTokenizerFast" if is_tokenizers_available() else None)),
             ("layoutlmv2", ("LayoutLMv2Tokenizer", "LayoutLMv2TokenizerFast" if is_tokenizers_available() else None)),
+            ("layoutxlm", ("LayoutXLMTokenizer", "LayoutXLMTokenizerFast" if is_tokenizers_available() else None)),
             (
                 "dpr",
                 (

src/transformers/models/layoutxlm/__init__.py

@@ -0,0 +1,54 @@
+# flake8: noqa
+# There's no way to ignore "F401 '...' imported but unused" warnings in this
+# module, but to preserve other warnings. So, don't check this module at all.
+
+# Copyright 2021 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import TYPE_CHECKING
+
+from ...file_utils import (
+    _LazyModule,
+    is_sentencepiece_available,
+    is_tokenizers_available,
+    is_torch_available,
+    is_vision_available,
+)
+
+
+_import_structure = {}
+
+if is_sentencepiece_available():
+    _import_structure["tokenization_layoutxlm"] = ["LayoutXLMTokenizer"]
+
+if is_tokenizers_available():
+    _import_structure["tokenization_layoutxlm_fast"] = ["LayoutXLMTokenizerFast"]
+
+if is_vision_available():
+    _import_structure["processing_layoutxlm"] = ["LayoutXLMProcessor"]
+
+if TYPE_CHECKING:
+    if is_sentencepiece_available():
+        from .tokenization_layoutxlm import LayoutXLMTokenizer
+
+    if is_tokenizers_available():
+        from .tokenization_layoutxlm_fast import LayoutXLMTokenizerFast
+
+    if is_vision_available():
+        from .processing_layoutlmv2 import LayoutXLMProcessor
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure)