transformers documentation

LayoutLMV2

# LayoutLMV2

## Overview

The LayoutLMV2 model was proposed in LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou. LayoutLMV2 improves LayoutLM to obtain state-of-the-art results across several document image understanding benchmarks:

• information extraction from scanned documents: the FUNSD dataset (a collection of 199 annotated forms comprising more than 30,000 words), the CORD dataset (a collection of 800 receipts for training, 100 for validation and 100 for testing), the SROIE dataset (a collection of 626 receipts for training and 347 receipts for testing) and the Kleister-NDA dataset (a collection of non-disclosure agreements from the EDGAR database, including 254 documents for training, 83 documents for validation, and 203 documents for testing).
• document image classification: the RVL-CDIP dataset (a collection of 400,000 images belonging to one of 16 classes).
• document visual question answering: the DocVQA dataset (a collection of 50,000 questions defined on 12,000+ document images).

The abstract from the paper is the following:

Pre-training of text and layout has proved effective in a variety of visually-rich document understanding tasks due to its effective model architecture and the advantage of large-scale unlabeled scanned/digital-born documents. In this paper, we present LayoutLMv2 by pre-training text, layout and image in a multi-modal framework, where new model architectures and pre-training tasks are leveraged. Specifically, LayoutLMv2 not only uses the existing masked visual-language modeling task but also the new text-image alignment and text-image matching tasks in the pre-training stage, where cross-modality interaction is better learned. Meanwhile, it also integrates a spatial-aware self-attention mechanism into the Transformer architecture, so that the model can fully understand the relative positional relationship among different text blocks. Experiment results show that LayoutLMv2 outperforms strong baselines and achieves new state-of-the-art results on a wide variety of downstream visually-rich document understanding tasks, including FUNSD (0.7895 -> 0.8420), CORD (0.9493 -> 0.9601), SROIE (0.9524 -> 0.9781), Kleister-NDA (0.834 -> 0.852), RVL-CDIP (0.9443 -> 0.9564), and DocVQA (0.7295 -> 0.8672). The pre-trained LayoutLMv2 model is publicly available at this https URL.

Tips:

• The main difference between LayoutLMv1 and LayoutLMv2 is that the latter incorporates visual embeddings during pre-training (while LayoutLMv1 only adds visual embeddings during fine-tuning).
• LayoutLMv2 adds both a relative 1D attention bias as well as a spatial 2D attention bias to the attention scores in the self-attention layers. Details can be found on page 5 of the paper.
• Demo notebooks on how to use the LayoutLMv2 model on RVL-CDIP, FUNSD, DocVQA, CORD can be found here.
• LayoutLMv2 uses Facebook AI’s Detectron2 package for its visual backbone. See this link for installation instructions.
• In addition to input_ids, forward() expects 2 additional inputs, namely image and bbox. The image input corresponds to the original document image in which the text tokens occur. The model expects each document image to be of size 224x224. This means that if you have a batch of document images, image should be a tensor of shape (batch_size, 3, 224, 224). This can be either a torch.Tensor or a Detectron2.structures.ImageList. You don’t need to normalize the channels, as this is done by the model. Important to note is that the visual backbone expects BGR channels instead of RGB, as all models in Detectron2 are pre-trained using the BGR format. The bbox input are the bounding boxes (i.e. 2D-positions) of the input text tokens. This is identical to LayoutLMModel. These can be obtained using an external OCR engine such as Google’s Tesseract (there’s a Python wrapper available). Each bounding box should be in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. Note that one first needs to normalize the bounding boxes to be on a 0-1000 scale. To normalize, you can use the following function:
def normalize_bbox(bbox, width, height):
return [
int(1000 * (bbox[0] / width)),
int(1000 * (bbox[1] / height)),
int(1000 * (bbox[2] / width)),
int(1000 * (bbox[3] / height)),
]

Here, width and height correspond to the width and height of the original document in which the token occurs (before resizing the image). Those can be obtained using the Python Image Library (PIL) library for example, as follows:

from PIL import Image

image = Image.open("name_of_your_document - can be a png file, pdf, etc.")

width, height = image.size

However, this model includes a brand new LayoutLMv2Processor which can be used to directly prepare data for the model (including applying OCR under the hood). More information can be found in the “Usage” section below.

• Internally, LayoutLMv2Model will send the image input through its visual backbone to obtain a lower-resolution feature map, whose shape is equal to the image_feature_pool_shape attribute of LayoutLMv2Config. This feature map is then flattened to obtain a sequence of image tokens. As the size of the feature map is 7x7 by default, one obtains 49 image tokens. These are then concatenated with the text tokens, and send through the Transformer encoder. This means that the last hidden states of the model will have a length of 512 + 49 = 561, if you pad the text tokens up to the max length. More generally, the last hidden states will have a shape of seq_length + image_feature_pool_shape[0] * config.image_feature_pool_shape[1].
• When calling from_pretrained(), a warning will be printed with a long list of parameter names that are not initialized. This is not a problem, as these parameters are batch normalization statistics, which are going to have values when fine-tuning on a custom dataset.
• If you want to train the model in a distributed environment, make sure to call synchronize_batch_norm on the model in order to properly synchronize the batch normalization layers of the visual backbone.

In addition, there’s LayoutXLM, which is a multilingual version of LayoutLMv2. More information can be found on LayoutXLM’s documentation page.

## Usage: LayoutLMv2Processor

The easiest way to prepare data for the model is to use LayoutLMv2Processor, which internally combines a feature extractor (LayoutLMv2FeatureExtractor) and a tokenizer (LayoutLMv2Tokenizer or LayoutLMv2TokenizerFast). The feature extractor handles the image modality, while the tokenizer handles the text modality. A processor combines both, which is ideal for a multi-modal model like LayoutLMv2. Note that you can still use both separately, if you only want to handle one modality.

from transformers import LayoutLMv2FeatureExtractor, LayoutLMv2TokenizerFast, LayoutLMv2Processor

feature_extractor = LayoutLMv2FeatureExtractor()  # apply_ocr is set to True by default
tokenizer = LayoutLMv2TokenizerFast.from_pretrained("microsoft/layoutlmv2-base-uncased")
processor = LayoutLMv2Processor(feature_extractor, tokenizer)

In short, one can provide a document image (and possibly additional data) to LayoutLMv2Processor, and it will create the inputs expected by the model. Internally, the processor first uses LayoutLMv2FeatureExtractor to apply OCR on the image to get a list of words and normalized bounding boxes, as well to resize the image to a given size in order to get the image input. The words and normalized bounding boxes are then provided to LayoutLMv2Tokenizer or LayoutLMv2TokenizerFast, which converts them to token-level input_ids, attention_mask, token_type_ids, bbox. Optionally, one can provide word labels to the processor, which are turned into token-level labels.

LayoutLMv2Processor uses PyTesseract, a Python wrapper around Google’s Tesseract OCR engine, under the hood. Note that you can still use your own OCR engine of choice, and provide the words and normalized boxes yourself. This requires initializing LayoutLMv2FeatureExtractor with apply_ocr set to False.

In total, there are 5 use cases that are supported by the processor. Below, we list them all. Note that each of these use cases work for both batched and non-batched inputs (we illustrate them for non-batched inputs).

Use case 1: document image classification (training, inference) + token classification (inference), apply_ocr = True

This is the simplest case, in which the processor (actually the feature extractor) will perform OCR on the image to get the words and normalized bounding boxes.

from transformers import LayoutLMv2Processor
from PIL import Image

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")

image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
encoding = processor(
image, return_tensors="pt"
)  # you can also add all tokenizer parameters here such as padding, truncation
print(encoding.keys())
# dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])

Use case 2: document image classification (training, inference) + token classification (inference), apply_ocr=False

In case one wants to do OCR themselves, one can initialize the feature extractor with apply_ocr set to False. In that case, one should provide the words and corresponding (normalized) bounding boxes themselves to the processor.

from transformers import LayoutLMv2Processor
from PIL import Image

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")

image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
words = ["hello", "world"]
boxes = [[1, 2, 3, 4], [5, 6, 7, 8]]  # make sure to normalize your bounding boxes
encoding = processor(image, words, boxes=boxes, return_tensors="pt")
print(encoding.keys())
# dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])

Use case 3: token classification (training), apply_ocr=False

For token classification tasks (such as FUNSD, CORD, SROIE, Kleister-NDA), one can also provide the corresponding word labels in order to train a model. The processor will then convert these into token-level labels. By default, it will only label the first wordpiece of a word, and label the remaining wordpieces with -100, which is the ignore_index of PyTorch’s CrossEntropyLoss. In case you want all wordpieces of a word to be labeled, you can initialize the tokenizer with only_label_first_subword set to False.

from transformers import LayoutLMv2Processor
from PIL import Image

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")

image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
words = ["hello", "world"]
boxes = [[1, 2, 3, 4], [5, 6, 7, 8]]  # make sure to normalize your bounding boxes
word_labels = [1, 2]
encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt")
print(encoding.keys())
# dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'labels', 'image'])

Use case 4: visual question answering (inference), apply_ocr=True

For visual question answering tasks (such as DocVQA), you can provide a question to the processor. By default, the processor will apply OCR on the image, and create [CLS] question tokens [SEP] word tokens [SEP].

from transformers import LayoutLMv2Processor
from PIL import Image

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")

image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
question = "What's his name?"
encoding = processor(image, question, return_tensors="pt")
print(encoding.keys())
# dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])

Use case 5: visual question answering (inference), apply_ocr=False

For visual question answering tasks (such as DocVQA), you can provide a question to the processor. If you want to perform OCR yourself, you can provide your own words and (normalized) bounding boxes to the processor.

from transformers import LayoutLMv2Processor
from PIL import Image

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")

image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
question = "What's his name?"
words = ["hello", "world"]
boxes = [[1, 2, 3, 4], [5, 6, 7, 8]]  # make sure to normalize your bounding boxes
encoding = processor(image, question, words, boxes=boxes, return_tensors="pt")
print(encoding.keys())
# dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])

## LayoutLMv2Config

class transformers.LayoutLMv2Config < >

( 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 max_2d_position_embeddings = 1024 max_rel_pos = 128 rel_pos_bins = 32 fast_qkv = True max_rel_2d_pos = 256 rel_2d_pos_bins = 64 convert_sync_batchnorm = True image_feature_pool_shape = [7, 7, 256] coordinate_size = 128 shape_size = 128 has_relative_attention_bias = True has_spatial_attention_bias = True has_visual_segment_embedding = False detectron2_config_args = None **kwargs )

Parameters

• vocab_size (int, optional, defaults to 30522) — Vocabulary size of the LayoutLMv2 model. Defines the number of different tokens that can be represented by the inputs_ids passed when calling LayoutLMv2Model or TFLayoutLMv2Model.
• hidden_size (int, optional, defaults to 768) — Dimension 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) — Dimension 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 LayoutLMv2Model or TFLayoutLMv2Model.
• 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.
• max_2d_position_embeddings (int, optional, defaults to 1024) — The maximum value that the 2D position embedding might ever be used with. Typically set this to something large just in case (e.g., 1024).
• max_rel_pos (int, optional, defaults to 128) — The maximum number of relative positions to be used in the self-attention mechanism.
• rel_pos_bins (int, optional, defaults to 32) — The number of relative position bins to be used in the self-attention mechanism.
• fast_qkv (bool, optional, defaults to True) — Whether or not to use a single matrix for the queries, keys, values in the self-attention layers.
• max_rel_2d_pos (int, optional, defaults to 256) — The maximum number of relative 2D positions in the self-attention mechanism.
• rel_2d_pos_bins (int, optional, defaults to 64) — The number of 2D relative position bins in the self-attention mechanism.
• image_feature_pool_shape (List[int], optional, defaults to [7, 7, 256]) — The shape of the average-pooled feature map.
• coordinate_size (int, optional, defaults to 128) — Dimension of the coordinate embeddings.
• shape_size (int, optional, defaults to 128) — Dimension of the width and height embeddings.
• has_relative_attention_bias (bool, optional, defaults to True) — Whether or not to use a relative attention bias in the self-attention mechanism.
• has_spatial_attention_bias (bool, optional, defaults to True) — Whether or not to use a spatial attention bias in the self-attention mechanism.
• has_visual_segment_embedding (bool, optional, defaults to False) — Whether or not to add visual segment embeddings.
• detectron2_config_args (dict, optional) — Dictionary containing the configuration arguments of the Detectron2 visual backbone. Refer to this file for details regarding default values.

This is the configuration class to store the configuration of a LayoutLMv2Model. It is used to instantiate an LayoutLMv2 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 LayoutLMv2 microsoft/layoutlmv2-base-uncased architecture.

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

Example:

>>> from transformers import LayoutLMv2Model, LayoutLMv2Config

>>> # Initializing a LayoutLMv2 microsoft/layoutlmv2-base-uncased style configuration
>>> configuration = LayoutLMv2Config()

>>> # Initializing a model from the microsoft/layoutlmv2-base-uncased style configuration
>>> model = LayoutLMv2Model(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

## LayoutLMv2FeatureExtractor

class transformers.LayoutLMv2FeatureExtractor < >

( do_resize = True size = 224 resample = 2 apply_ocr = True ocr_lang = None **kwargs )

Parameters

• do_resize (bool, optional, defaults to True) — Whether to resize the input to a certain size.
• size (int or Tuple(int), optional, defaults to 224) — Resize the input to the given size. If a tuple is provided, it should be (width, height). If only an integer is provided, then the input will be resized to (size, size). Only has an effect if do_resize is set to True.
• resample (int, optional, defaults to PIL.Image.BILINEAR) — An optional resampling filter. This can be one of PIL.Image.NEAREST, PIL.Image.BOX, PIL.Image.BILINEAR, PIL.Image.HAMMING, PIL.Image.BICUBIC or PIL.Image.LANCZOS. Only has an effect if do_resize is set to True.
• apply_ocr (bool, optional, defaults to True) — Whether to apply the Tesseract OCR engine to get words + normalized bounding boxes.
• ocr_lang (Optional[str], optional) — The language, specified by its ISO code, to be used by the Tesseract OCR engine. By default, English is used.

LayoutLMv2FeatureExtractor uses Google’s Tesseract OCR engine under the hood.

Constructs a LayoutLMv2 feature extractor. This can be used to resize document images to the same size, as well as to apply OCR on them in order to get a list of words and normalized bounding boxes.

This feature extractor inherits from PreTrainedFeatureExtractor() which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

__call__ < >

( images: typing.Union[PIL.Image.Image, numpy.ndarray, ForwardRef('torch.Tensor'), typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[ForwardRef('torch.Tensor')]] return_tensors: typing.Union[str, transformers.file_utils.TensorType, NoneType] = None **kwargs ) BatchFeature

Parameters

• images (PIL.Image.Image, np.ndarray, torch.Tensor, List[PIL.Image.Image], List[np.ndarray], List[torch.Tensor]) — The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch tensor. In case of a NumPy array/PyTorch tensor, each image should be of shape (C, H, W), where C is a number of channels, H and W are image height and width.
• return_tensors (str or TensorType, optional, defaults to 'np') — If set, will return tensors of a particular framework. Acceptable values are:

• 'tf': Return TensorFlow tf.constant objects.
• 'pt': Return PyTorch torch.Tensor objects.
• 'np': Return NumPy np.ndarray objects.
• 'jax': Return JAX jnp.ndarray objects.

Returns

BatchFeature

A BatchFeature with the following fields:

• pixel_values — Pixel values to be fed to a model, of shape (batch_size, num_channels, height, width).
• words — Optional words as identified by Tesseract OCR (only when LayoutLMv2FeatureExtractor was initialized with apply_ocr set to True).
• boxes — Optional bounding boxes as identified by Tesseract OCR, normalized based on the image size (only when LayoutLMv2FeatureExtractor was initialized with apply_ocr set to True).

Main method to prepare for the model one or several image(s).

Examples:

>>> from transformers import LayoutLMv2FeatureExtractor
>>> from PIL import Image

>>> image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")

>>> # option 1: with apply_ocr=True (default)
>>> feature_extractor = LayoutLMv2FeatureExtractor()
>>> encoding = feature_extractor(image, return_tensors="pt")
>>> print(encoding.keys())
>>> # dict_keys(['pixel_values', 'words', 'boxes'])

>>> # option 2: with apply_ocr=False
>>> feature_extractor = LayoutLMv2FeatureExtractor(apply_ocr=False)
>>> encoding = feature_extractor(image, return_tensors="pt")
>>> print(encoding.keys())
>>> # dict_keys(['pixel_values'])

## LayoutLMv2Tokenizer

class transformers.LayoutLMv2Tokenizer < >

( vocab_file 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]' cls_token_box = [0, 0, 0, 0] sep_token_box = [1000, 1000, 1000, 1000] pad_token_box = [0, 0, 0, 0] pad_token_label = -100 only_label_first_subword = True tokenize_chinese_chars = True strip_accents = None model_max_length: int = 512 additional_special_tokens: typing.Optional[typing.List[str]] = None **kwargs )

Construct a LayoutLMv2 tokenizer. Based on WordPiece. LayoutLMv2Tokenizer can be used to turn words, word-level bounding boxes and optional word labels to token-level input_ids, attention_mask, token_type_ids, bbox, and optional labels (for token classification).

This tokenizer inherits from PreTrainedTokenizer which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

LayoutLMv2Tokenizer runs end-to-end tokenization: punctuation splitting and wordpiece. It also turns the word-level bounding boxes into token-level bounding boxes.

__call__ < >

( text: typing.Union[str, typing.List[str], typing.List[typing.List[str]]] text_pair: typing.Union[typing.List[str], typing.List[typing.List[str]], NoneType] = None boxes: typing.Union[typing.List[typing.List[int]], typing.List[typing.List[typing.List[int]]]] = None word_labels: typing.Union[typing.List[int], typing.List[typing.List[int]], NoneType] = None add_special_tokens: bool = True padding: typing.Union[bool, str, transformers.file_utils.PaddingStrategy] = False truncation: typing.Union[bool, str, transformers.tokenization_utils_base.TruncationStrategy] = False max_length: typing.Optional[int] = None stride: int = 0 pad_to_multiple_of: typing.Optional[int] = None return_tensors: typing.Union[str, transformers.file_utils.TensorType, NoneType] = None return_token_type_ids: typing.Optional[bool] = None return_attention_mask: typing.Optional[bool] = None return_overflowing_tokens: bool = False return_special_tokens_mask: bool = False return_offsets_mapping: bool = False return_length: bool = False verbose: bool = True **kwargs )

Parameters

• text (str, List[str], List[List[str]]) — The sequence or batch of sequences to be encoded. Each sequence can be a string, a list of strings (words of a single example or questions of a batch of examples) or a list of list of strings (batch of words).
• text_pair (List[str], List[List[str]]) — The sequence or batch of sequences to be encoded. Each sequence should be a list of strings (pretokenized string).
• boxes (List[List[int]], List[List[List[int]]]) — Word-level bounding boxes. Each bounding box should be normalized to be on a 0-1000 scale.
• word_labels (List[int], List[List[int]], optional) — Word-level integer labels (for token classification tasks such as FUNSD, CORD).
• add_special_tokens (bool, optional, defaults to True) — Whether or not to encode the sequences with the special tokens relative to their model.
• padding (bool, str or PaddingStrategy, optional, defaults to False) — Activates and controls padding. Accepts the following values:

• 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).
• truncation (bool, str or TruncationStrategy, optional, defaults to False) — Activates and controls truncation. Accepts the following values:

• True or 'longest_first': Truncate 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. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_first': Truncate 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. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_second': Truncate 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. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• False or 'do_not_truncate' (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size).
• max_length (int, optional) — Controls the maximum length to use by one of the truncation/padding parameters.

If left unset or set to None, this will use the predefined model maximum length if a maximum length is required by one of the truncation/padding parameters. If the model has no specific maximum input length (like XLNet) truncation/padding to a maximum length will be deactivated.

• stride (int, optional, defaults to 0) — If set to a number along with max_length, the overflowing tokens returned when return_overflowing_tokens=True will contain some tokens from the end of the truncated sequence returned to provide some overlap between truncated and overflowing sequences. The value of this argument defines the number of overlapping tokens.
• is_split_into_words (bool, optional, defaults to False) — Whether or not the input is already pre-tokenized (e.g., split into words). If set to True, the tokenizer assumes the input is already split into words (for instance, by splitting it on whitespace) which it will tokenize. This is useful for NER or token classification.
• 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_tensors (str or 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.
• add_special_tokens (bool, optional, defaults to True) — Whether or not to encode the sequences with the special tokens relative to their model.
• padding (bool, str or PaddingStrategy, optional, defaults to False) — Activates and controls padding. Accepts the following values:

• 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).
• truncation (bool, str or TruncationStrategy, optional, defaults to False) — Activates and controls truncation. Accepts the following values:

• True or 'longest_first': Truncate 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. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_first': Truncate 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. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_second': Truncate 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. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• False or 'do_not_truncate' (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size).
• max_length (int, optional) — Controls the maximum length to use by one of the truncation/padding parameters. If left unset or set to None, this will use the predefined model maximum length if a maximum length is required by one of the truncation/padding parameters. If the model has no specific maximum input length (like XLNet) truncation/padding to a maximum length will be deactivated.
• stride (int, optional, defaults to 0) — If set to a number along with max_length, the overflowing tokens returned when return_overflowing_tokens=True will contain some tokens from the end of the truncated sequence returned to provide some overlap between truncated and overflowing sequences. The value of this argument defines the number of overlapping tokens.
• 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_tensors (str or 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.

Main method to tokenize and prepare for the model one or several sequence(s) or one or several pair(s) of sequences with word-level normalized bounding boxes and optional labels.

## LayoutLMv2TokenizerFast

class transformers.LayoutLMv2TokenizerFast < >

( vocab_file = None tokenizer_file = None do_lower_case = True unk_token = '[UNK]' sep_token = '[SEP]' pad_token = '[PAD]' cls_token = '[CLS]' mask_token = '[MASK]' cls_token_box = [0, 0, 0, 0] sep_token_box = [1000, 1000, 1000, 1000] pad_token_box = [0, 0, 0, 0] pad_token_label = -100 only_label_first_subword = True tokenize_chinese_chars = True strip_accents = None **kwargs )

Parameters

• vocab_file (str) — File containing the vocabulary.
• do_lower_case (bool, optional, defaults to True) — Whether or not to lowercase the input when tokenizing.
• 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.
• cls_token_box (List[int], optional, defaults to [0, 0, 0, 0]) — The bounding box to use for the special [CLS] token.
• sep_token_box (List[int], optional, defaults to [1000, 1000, 1000, 1000]) — The bounding box to use for the special [SEP] token.
• pad_token_box (List[int], optional, defaults to [0, 0, 0, 0]) — The bounding box to use for the special [PAD] token.
• pad_token_label (int, optional, defaults to -100) — The label to use for padding tokens. Defaults to -100, which is the ignore_index of PyTorch’s CrossEntropyLoss.
• only_label_first_subword (bool, optional, defaults to True) — Whether or not to only label the first subword, in case word labels are provided.
• 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). 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 LayoutLMv2).

Construct a “fast” LayoutLMv2 tokenizer (backed by HuggingFace’s tokenizers library). Based on WordPiece.

This tokenizer inherits from PreTrainedTokenizerFast which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

__call__ < >

( text: typing.Union[str, typing.List[str], typing.List[typing.List[str]]] text_pair: typing.Union[typing.List[str], typing.List[typing.List[str]], NoneType] = None boxes: typing.Union[typing.List[typing.List[int]], typing.List[typing.List[typing.List[int]]]] = None word_labels: typing.Union[typing.List[int], typing.List[typing.List[int]], NoneType] = None add_special_tokens: bool = True padding: typing.Union[bool, str, transformers.file_utils.PaddingStrategy] = False truncation: typing.Union[bool, str, transformers.tokenization_utils_base.TruncationStrategy] = False max_length: typing.Optional[int] = None stride: int = 0 pad_to_multiple_of: typing.Optional[int] = None return_tensors: typing.Union[str, transformers.file_utils.TensorType, NoneType] = None return_token_type_ids: typing.Optional[bool] = None return_attention_mask: typing.Optional[bool] = None return_overflowing_tokens: bool = False return_special_tokens_mask: bool = False return_offsets_mapping: bool = False return_length: bool = False verbose: bool = True **kwargs )

Parameters

• text (str, List[str], List[List[str]]) — The sequence or batch of sequences to be encoded. Each sequence can be a string, a list of strings (words of a single example or questions of a batch of examples) or a list of list of strings (batch of words).
• text_pair (List[str], List[List[str]]) — The sequence or batch of sequences to be encoded. Each sequence should be a list of strings (pretokenized string).
• boxes (List[List[int]], List[List[List[int]]]) — Word-level bounding boxes. Each bounding box should be normalized to be on a 0-1000 scale.
• word_labels (List[int], List[List[int]], optional) — Word-level integer labels (for token classification tasks such as FUNSD, CORD).
• add_special_tokens (bool, optional, defaults to True) — Whether or not to encode the sequences with the special tokens relative to their model.
• padding (bool, str or PaddingStrategy, optional, defaults to False) — Activates and controls padding. Accepts the following values:

• 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).
• truncation (bool, str or TruncationStrategy, optional, defaults to False) — Activates and controls truncation. Accepts the following values:

• True or 'longest_first': Truncate 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. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_first': Truncate 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. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_second': Truncate 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. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• False or 'do_not_truncate' (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size).
• max_length (int, optional) — Controls the maximum length to use by one of the truncation/padding parameters.

If left unset or set to None, this will use the predefined model maximum length if a maximum length is required by one of the truncation/padding parameters. If the model has no specific maximum input length (like XLNet) truncation/padding to a maximum length will be deactivated.

• stride (int, optional, defaults to 0) — If set to a number along with max_length, the overflowing tokens returned when return_overflowing_tokens=True will contain some tokens from the end of the truncated sequence returned to provide some overlap between truncated and overflowing sequences. The value of this argument defines the number of overlapping tokens.
• is_split_into_words (bool, optional, defaults to False) — Whether or not the input is already pre-tokenized (e.g., split into words). If set to True, the tokenizer assumes the input is already split into words (for instance, by splitting it on whitespace) which it will tokenize. This is useful for NER or token classification.
• 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_tensors (str or 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.
• add_special_tokens (bool, optional, defaults to True) — Whether or not to encode the sequences with the special tokens relative to their model.
• padding (bool, str or PaddingStrategy, optional, defaults to False) — Activates and controls padding. Accepts the following values:

• 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).
• truncation (bool, str or TruncationStrategy, optional, defaults to False) — Activates and controls truncation. Accepts the following values:

• True or 'longest_first': Truncate 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. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_first': Truncate 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. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• 'only_second': Truncate 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. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
• False or 'do_not_truncate' (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size).
• max_length (int, optional) — Controls the maximum length to use by one of the truncation/padding parameters. If left unset or set to None, this will use the predefined model maximum length if a maximum length is required by one of the truncation/padding parameters. If the model has no specific maximum input length (like XLNet) truncation/padding to a maximum length will be deactivated.
• stride (int, optional, defaults to 0) — If set to a number along with max_length, the overflowing tokens returned when return_overflowing_tokens=True will contain some tokens from the end of the truncated sequence returned to provide some overlap between truncated and overflowing sequences. The value of this argument defines the number of overlapping tokens.
• 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_tensors (str or 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.

Main method to tokenize and prepare for the model one or several sequence(s) or one or several pair(s) of sequences with word-level normalized bounding boxes and optional labels.

## LayoutLMv2Processor

class transformers.LayoutLMv2Processor < >

( feature_extractor tokenizer )

Parameters

• feature_extractor (LayoutLMv2FeatureExtractor) — An instance of LayoutLMv2FeatureExtractor. The feature extractor is a required input.
• tokenizer (LayoutLMv2Tokenizer or LayoutLMv2TokenizerFast) — An instance of LayoutLMv2Tokenizer or LayoutLMv2TokenizerFast. The tokenizer is a required input.

Constructs a LayoutLMv2 processor which combines a LayoutLMv2 feature extractor and a LayoutLMv2 tokenizer into a single processor.

LayoutLMv2Processor offers all the functionalities you need to prepare data for the model.

It first uses LayoutLMv2FeatureExtractor to resize document images to a fixed size, and optionally applies OCR to get words and normalized bounding boxes. These are then provided to LayoutLMv2Tokenizer or LayoutLMv2TokenizerFast, which turns the words and bounding boxes into token-level input_ids, attention_mask, token_type_ids, bbox. Optionally, one can provide integer word_labels, which are turned into token-level labels for token classification tasks (such as FUNSD, CORD).

__call__ < >

( images text: typing.Union[str, typing.List[str], typing.List[typing.List[str]]] = None text_pair: typing.Union[typing.List[str], typing.List[typing.List[str]], NoneType] = None boxes: typing.Union[typing.List[typing.List[int]], typing.List[typing.List[typing.List[int]]]] = None word_labels: typing.Union[typing.List[int], typing.List[typing.List[int]], NoneType] = None add_special_tokens: bool = True padding: typing.Union[bool, str, transformers.file_utils.PaddingStrategy] = False truncation: typing.Union[bool, str, transformers.tokenization_utils_base.TruncationStrategy] = False max_length: typing.Optional[int] = None stride: int = 0 pad_to_multiple_of: typing.Optional[int] = None return_token_type_ids: typing.Optional[bool] = None return_attention_mask: typing.Optional[bool] = None return_overflowing_tokens: bool = False return_special_tokens_mask: bool = False return_offsets_mapping: bool = False return_length: bool = False verbose: bool = True return_tensors: typing.Union[str, transformers.file_utils.TensorType, NoneType] = None **kwargs )

This method first forwards the images argument to call(). In case LayoutLMv2FeatureExtractor was initialized with apply_ocr set to True, it passes the obtained words and bounding boxes along with the additional arguments to call() and returns the output, together with resized images. In case LayoutLMv2FeatureExtractor was initialized with apply_ocr set to False, it passes the words (text/text_pair) and boxes specified by the user along with the additional arguments to [__call__()](/docs/transformers/master/en/model_doc/layoutlmv2#transformers.LayoutLMv2Tokenizer.__call__) and returns the output, together with resized images.

## LayoutLMv2Model

class transformers.LayoutLMv2Model < >

( config )

Parameters

• config (LayoutLMv2Config) — 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 from_pretrained() method to load the model weights.

The bare LayoutLMv2 Model transformer outputting raw hidden-states without any specific head on top. This model is a PyTorch 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.

forward < >

( input_ids = None bbox = None image = None attention_mask = None token_type_ids = None position_ids = None head_mask = None inputs_embeds = None output_attentions = None output_hidden_states = None return_dict = None ) transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)

Parameters

• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.

Indices can be obtained using LayoutLMv2Tokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

• bbox (torch.LongTensor of shape ((batch_size, sequence_length), 4), optional) — Bounding boxes of each input sequence tokens. Selected in the range [0, config.max_2d_position_embeddings-1]. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner.
• image (torch.FloatTensor of shape (batch_size, num_channels, height, width) or detectron.structures.ImageList whose tensors is of shape (batch_size, num_channels, height, width)) — Batch of document images.
• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), 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.
• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), 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.
• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

• head_mask (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]:

• inputs_embeds (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 ModelOutput instead of a plain tuple.

Returns

transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)

A transformers.modeling_outputs.BaseModelOutput or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (LayoutLMv2Config) and inputs.

• last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.

• 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.

The LayoutLMv2Model forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Examples:

>>> from transformers import LayoutLMv2Processor, LayoutLMv2Model
>>> from PIL import Image

>>> processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
>>> model = LayoutLMv2Model.from_pretrained("microsoft/layoutlmv2-base-uncased")

>>> image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")

>>> encoding = processor(image, return_tensors="pt")

>>> outputs = model(**encoding)
>>> last_hidden_states = outputs.last_hidden_state

## LayoutLMv2ForSequenceClassification

class transformers.LayoutLMv2ForSequenceClassification < >

( config )

Parameters

• config (LayoutLMv2Config) — 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 from_pretrained() method to load the model weights.

LayoutLMv2 Model with a sequence classification head on top (a linear layer on top of the concatenation of the final hidden state of the [CLS] token, average-pooled initial visual embeddings and average-pooled final visual embeddings, e.g. for document image classification tasks such as the RVL-CDIP dataset.

This model is a PyTorch 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.

forward < >

( input_ids = None bbox = None image = 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 ) transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)

Parameters

• input_ids (torch.LongTensor of shape batch_size, sequence_length) — Indices of input sequence tokens in the vocabulary.

Indices can be obtained using LayoutLMv2Tokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

• bbox (torch.LongTensor of shape (batch_size, sequence_length, 4), optional) — Bounding boxes of each input sequence tokens. Selected in the range [0, config.max_2d_position_embeddings-1]. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner.
• image (torch.FloatTensor of shape (batch_size, num_channels, height, width) or detectron.structures.ImageList whose tensors is of shape (batch_size, num_channels, height, width)) — Batch of document images.
• attention_mask (torch.FloatTensor of shape batch_size, sequence_length, 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.
• token_type_ids (torch.LongTensor of shape batch_size, sequence_length, 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.
• position_ids (torch.LongTensor of shape batch_size, sequence_length, optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

• head_mask (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]:

• inputs_embeds (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 ModelOutput instead of a plain tuple.
• 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).

Returns

transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)

A transformers.modeling_outputs.SequenceClassifierOutput or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (LayoutLMv2Config) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification (or regression if config.num_labels==1) loss.

• logits (torch.FloatTensor of shape (batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (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.

The LayoutLMv2ForSequenceClassification forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Examples:

>>> from transformers import LayoutLMv2Processor, LayoutLMv2ForSequenceClassification
>>> from PIL import Image
>>> import torch

>>> processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
>>> model = LayoutLMv2ForSequenceClassification.from_pretrained("microsoft/layoutlmv2-base-uncased")

>>> image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")

>>> encoding = processor(image, return_tensors="pt")
>>> sequence_label = torch.tensor([1])

>>> outputs = model(**encoding, labels=sequence_label)
>>> loss = outputs.loss
>>> logits = outputs.logits

## LayoutLMv2ForTokenClassification

class transformers.LayoutLMv2ForTokenClassification < >

( config )

Parameters

• config (LayoutLMv2Config) — 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 from_pretrained() method to load the model weights.

LayoutLMv2 Model with a token classification head on top (a linear layer on top of the text part of the hidden states) e.g. for sequence labeling (information extraction) tasks such as FUNSD, SROIE, CORD and Kleister-NDA.

This model is a PyTorch 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.

forward < >

( input_ids = None bbox = None image = 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 ) transformers.modeling_outputs.TokenClassifierOutput or tuple(torch.FloatTensor)

Parameters

• input_ids (torch.LongTensor of shape batch_size, sequence_length) — Indices of input sequence tokens in the vocabulary.

Indices can be obtained using LayoutLMv2Tokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

• bbox (torch.LongTensor of shape (batch_size, sequence_length, 4), optional) — Bounding boxes of each input sequence tokens. Selected in the range [0, config.max_2d_position_embeddings-1]. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner.
• image (torch.FloatTensor of shape (batch_size, num_channels, height, width) or detectron.structures.ImageList whose tensors is of shape (batch_size, num_channels, height, width)) — Batch of document images.
• attention_mask (torch.FloatTensor of shape batch_size, sequence_length, 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.
• token_type_ids (torch.LongTensor of shape batch_size, sequence_length, 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.
• position_ids (torch.LongTensor of shape batch_size, sequence_length, optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

• head_mask (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]:

• inputs_embeds (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 ModelOutput instead of a plain tuple.
• 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].

Returns

transformers.modeling_outputs.TokenClassifierOutput or tuple(torch.FloatTensor)

A transformers.modeling_outputs.TokenClassifierOutput or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (LayoutLMv2Config) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification loss.

• logits (torch.FloatTensor of shape (batch_size, sequence_length, config.num_labels)) — Classification scores (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.

The LayoutLMv2ForTokenClassification forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Examples:

>>> from transformers import LayoutLMv2Processor, LayoutLMv2ForTokenClassification
>>> from PIL import Image

>>> processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")
>>> model = LayoutLMv2ForTokenClassification.from_pretrained("microsoft/layoutlmv2-base-uncased")

>>> image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
>>> words = ["hello", "world"]
>>> boxes = [[1, 2, 3, 4], [5, 6, 7, 8]]  # make sure to normalize your bounding boxes
>>> word_labels = [0, 1]

>>> encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt")

>>> outputs = model(**encoding)
>>> loss = outputs.loss
>>> logits = outputs.logits

( config has_visual_segment_embedding = True )

Parameters

• config (LayoutLMv2Config) — 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 from_pretrained() method to load the model weights.

LayoutLMv2 Model with a span classification head on top for extractive question-answering tasks such as DocVQA (a linear layer on top of the text part of the hidden-states output to compute span start logits and span end logits).

This model is a PyTorch 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.

forward < >

( input_ids = None bbox = None image = 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 ) transformers.modeling_outputs.QuestionAnsweringModelOutput or tuple(torch.FloatTensor)

Parameters

• input_ids (torch.LongTensor of shape batch_size, sequence_length) — Indices of input sequence tokens in the vocabulary.

Indices can be obtained using LayoutLMv2Tokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

• bbox (torch.LongTensor of shape (batch_size, sequence_length, 4), optional) — Bounding boxes of each input sequence tokens. Selected in the range [0, config.max_2d_position_embeddings-1]. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner.
• image (torch.FloatTensor of shape (batch_size, num_channels, height, width) or detectron.structures.ImageList whose tensors is of shape (batch_size, num_channels, height, width)) — Batch of document images.
• attention_mask (torch.FloatTensor of shape batch_size, sequence_length, 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.
• token_type_ids (torch.LongTensor of shape batch_size, sequence_length, 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.
• position_ids (torch.LongTensor of shape batch_size, sequence_length, optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

• head_mask (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]:

• inputs_embeds (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 ModelOutput instead of a plain tuple.
• 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.

Returns

transformers.modeling_outputs.QuestionAnsweringModelOutput or tuple(torch.FloatTensor)

A transformers.modeling_outputs.QuestionAnsweringModelOutput or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (LayoutLMv2Config) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.

• start_logits (torch.FloatTensor of shape (batch_size, sequence_length)) — Span-start scores (before SoftMax).

• end_logits (torch.FloatTensor of shape (batch_size, sequence_length)) — Span-end scores (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.

The LayoutLMv2ForQuestionAnswering forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Examples:

>>> from transformers import LayoutLMv2Processor, LayoutLMv2ForQuestionAnswering
>>> from PIL import Image
>>> import torch

>>> processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
>>> end_scores = outputs.end_logits