paddlepaddle--paddlenlp/paddlenlp/transformers/layoutlmv2/modeling.py

# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
# Copyright 2021 Microsoft Research and The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Modeling classes for LayoutLMv2 model."""

import copy
import math

import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from paddle.nn import CrossEntropyLoss, Layer

from paddlenlp.utils.log import logger

from ...utils.env import CONFIG_NAME
from .. import PretrainedModel, register_base_model
from ..layoutxlm.visual_backbone import build_resnet_fpn_backbone, read_config
from .configuration import (
    LAYOUTLMV2_PRETRAINED_INIT_CONFIGURATION,
    LAYOUTLMV2_PRETRAINED_RESOURCE_FILES_MAP,
    LayoutLMv2Config,
)

__all__ = [
    "LayoutLMv2Model",
    "LayoutLMv2PretrainedModel",
    "LayoutLMv2ForTokenClassification",
    "LayoutLMv2ForPretraining",
    "LayoutLMv2ForRelationExtraction",
]


def relative_position_bucket(relative_position, bidirectional=True, num_buckets=32, max_distance=128):
    ret = 0
    if bidirectional:
        num_buckets //= 2
        ret += (relative_position > 0).astype(paddle.int64) * num_buckets
        n = paddle.abs(relative_position)
    else:
        n = paddle.max(-relative_position, paddle.zeros_like(relative_position))
    # now n is in the range [0, inf)
    # half of the buckets are for exact increments in positions
    max_exact = num_buckets // 2
    is_small = n < max_exact

    # The other half of the buckets are for logarithmically bigger bins in positions up to max_distance
    val_if_large = max_exact + (
        paddle.log(n.astype(paddle.float32) / max_exact)
        / math.log(max_distance / max_exact)
        * (num_buckets - max_exact)
    ).astype(paddle.int64)

    val_if_large = paddle.minimum(val_if_large, paddle.full_like(val_if_large, num_buckets - 1))

    ret += paddle.where(is_small, n, val_if_large)
    return ret


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMPooler with XLM->LMv2
class LayoutLMv2Pooler(Layer):
    def __init__(self, hidden_size, with_pool):
        super(LayoutLMv2Pooler, self).__init__()
        self.dense = nn.Linear(hidden_size, hidden_size)
        self.activation = nn.Tanh()
        self.with_pool = with_pool

    def forward(self, hidden_states):
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        if self.with_pool == "tanh":
            pooled_output = self.activation(pooled_output)
        return pooled_output


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMEmbeddings with XLM->LMv2
class LayoutLMv2Embeddings(Layer):
    """
    Include embeddings from word, position and token_type embeddings
    """

    def __init__(self, config):
        super(LayoutLMv2Embeddings, self).__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=0)
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
        self.x_position_embeddings = nn.Embedding(config.max_2d_position_embeddings, config.coordinate_size)
        self.y_position_embeddings = nn.Embedding(config.max_2d_position_embeddings, config.coordinate_size)
        self.h_position_embeddings = nn.Embedding(config.max_2d_position_embeddings, config.coordinate_size)
        self.w_position_embeddings = nn.Embedding(config.max_2d_position_embeddings, config.coordinate_size)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, epsilon=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

        self.register_buffer(
            "position_ids", paddle.arange(config.max_position_embeddings, dtype="int64").expand((1, -1))
        )

    def _cal_spatial_position_embeddings(self, bbox):
        try:
            left_position_embeddings = self.x_position_embeddings(bbox[:, :, 0])
            upper_position_embeddings = self.y_position_embeddings(bbox[:, :, 1])
            right_position_embeddings = self.x_position_embeddings(bbox[:, :, 2])
            lower_position_embeddings = self.y_position_embeddings(bbox[:, :, 3])
        except IndexError as e:
            raise IndexError("The :obj:`bbox`coordinate values should be within 0-1000 range.") from e

        h_position_embeddings = self.h_position_embeddings(bbox[:, :, 3] - bbox[:, :, 1])
        w_position_embeddings = self.w_position_embeddings(bbox[:, :, 2] - bbox[:, :, 0])

        spatial_position_embeddings = paddle.concat(
            [
                left_position_embeddings,
                upper_position_embeddings,
                right_position_embeddings,
                lower_position_embeddings,
                h_position_embeddings,
                w_position_embeddings,
            ],
            axis=-1,
        )
        return spatial_position_embeddings

    def forward(self, input_ids, bbox=None, token_type_ids=None, position_ids=None):
        if position_ids is None:
            ones = paddle.ones_like(input_ids, dtype="int64")
            seq_length = paddle.cumsum(ones, axis=-1)

            position_ids = seq_length - ones
            position_ids.stop_gradient = True
        if token_type_ids is None:
            token_type_ids = paddle.zeros_like(input_ids, dtype="int64")

        input_embedings = self.word_embeddings(input_ids)
        position_embeddings = self.position_embeddings(position_ids)

        try:
            left_position_embeddings = self.x_position_embeddings(bbox[:, :, 0])
            upper_position_embeddings = self.y_position_embeddings(bbox[:, :, 1])
            right_position_embeddings = self.x_position_embeddings(bbox[:, :, 2])
            lower_position_embeddings = self.y_position_embeddings(bbox[:, :, 3])
        except IndexError as e:
            raise IndexError("The :obj:`bbox`coordinate values should be within 0-1000 range.") from e
        h_position_embeddings = self.h_position_embeddings(bbox[:, :, 3] - bbox[:, :, 1])
        w_position_embeddings = self.w_position_embeddings(bbox[:, :, 2] - bbox[:, :, 0])

        token_type_embeddings = self.token_type_embeddings(token_type_ids)

        embeddings = (
            input_embedings
            + position_embeddings
            + left_position_embeddings
            + upper_position_embeddings
            + right_position_embeddings
            + lower_position_embeddings
            + h_position_embeddings
            + w_position_embeddings
            + token_type_embeddings
        )

        embeddings = self.layer_norm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class LayoutLMv2PretrainedModel(PretrainedModel):
    model_config_file = CONFIG_NAME
    config_class = LayoutLMv2Config
    resource_files_names = {"model_state": "model_state.pdparams"}

    base_model_prefix = "layoutlmv2"

    pretrained_init_configuration = LAYOUTLMV2_PRETRAINED_INIT_CONFIGURATION
    pretrained_resource_files_map = LAYOUTLMV2_PRETRAINED_RESOURCE_FILES_MAP

    def _init_weights(self, layer):
        """Initialization hook"""
        if isinstance(layer, (nn.Linear, nn.Embedding)):
            if isinstance(layer.weight, paddle.Tensor):
                layer.weight.set_value(
                    paddle.tensor.normal(
                        mean=0.0,
                        std=self.config.initializer_range,
                        shape=layer.weight.shape,
                    )
                )


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMSelfOutput with XLM->LMv2
class LayoutLMv2SelfOutput(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2SelfOutput, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, epsilon=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMSelfAttention with XLM->LMv2
class LayoutLMv2SelfAttention(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2SelfAttention, self).__init__()
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                "The hidden size {} is not a multiple of the number of attention "
                "heads {}".format(config.hidden_size, config.num_attention_heads)
            )
        self.fast_qkv = config.fast_qkv
        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.has_relative_attention_bias = config.has_relative_attention_bias
        self.has_spatial_attention_bias = config.has_spatial_attention_bias

        if self.fast_qkv:
            self.qkv_linear = nn.Linear(config.hidden_size, 3 * self.all_head_size, bias_attr=False)
            self.q_bias = self.create_parameter(
                shape=[1, 1, self.all_head_size], default_initializer=nn.initializer.Constant(0.0)
            )
            self.v_bias = self.create_parameter(
                shape=[1, 1, self.all_head_size], default_initializer=nn.initializer.Constant(0.0)
            )
        else:
            self.query = nn.Linear(config.hidden_size, self.all_head_size)
            self.key = nn.Linear(config.hidden_size, self.all_head_size)
            self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)

    def transpose_for_scores(self, x):
        new_x_shape = list(x.shape[:-1]) + [self.num_attention_heads, self.attention_head_size]
        x = x.reshape(new_x_shape)
        return x.transpose([0, 2, 1, 3])

    def compute_qkv(self, hidden_states):
        if self.fast_qkv:
            qkv = self.qkv_linear(hidden_states)
            q, k, v = paddle.chunk(qkv, 3, axis=-1)
            if q.ndimension() == self.q_bias.ndimension():
                q = q + self.q_bias
                v = v + self.v_bias
            else:
                _sz = (1,) * (q.ndimension() - 1) + (-1,)
                q = q + self.q_bias.reshape(_sz)
                v = v + self.v_bias.vreshape(_sz)
        else:
            q = self.query(hidden_states)
            k = self.key(hidden_states)
            v = self.value(hidden_states)
        return q, k, v

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
        rel_pos=None,
        rel_2d_pos=None,
    ):
        q, k, v = self.compute_qkv(hidden_states)

        # (B, L, H*D) -> (B, H, L, D)
        query_layer = self.transpose_for_scores(q)
        key_layer = self.transpose_for_scores(k)
        value_layer = self.transpose_for_scores(v)

        query_layer = query_layer / math.sqrt(self.attention_head_size)
        # [BSZ, NAT, L, L]
        attention_scores = paddle.matmul(query_layer, key_layer.transpose([0, 1, 3, 2]))
        if self.has_relative_attention_bias:
            attention_scores += rel_pos
        if self.has_spatial_attention_bias:
            attention_scores += rel_2d_pos

        bool_attention_mask = attention_mask.astype(paddle.bool)
        bool_attention_mask.stop_gradient = True
        attention_scores_shape = attention_scores.shape
        attention_scores = paddle.where(
            bool_attention_mask.expand(attention_scores_shape),
            paddle.ones(attention_scores_shape) * float("-1e10"),
            attention_scores,
        )

        attention_probs = F.softmax(attention_scores, axis=-1)
        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs = self.dropout(attention_probs)
        context_layer = paddle.matmul(attention_probs, value_layer)
        context_layer = context_layer.transpose([0, 2, 1, 3])
        new_context_layer_shape = list(context_layer.shape[:-2]) + [self.all_head_size]
        context_layer = context_layer.reshape(new_context_layer_shape)

        if output_attentions:
            outputs = [context_layer, attention_probs]
        else:
            outputs = [context_layer]
        return outputs


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMAttention with XLM->LMv2
class LayoutLMv2Attention(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2Attention, self).__init__()
        self.self = LayoutLMv2SelfAttention(config)
        self.output = LayoutLMv2SelfOutput(config)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
        rel_pos=None,
        rel_2d_pos=None,
    ):

        self_outputs = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            encoder_hidden_states,
            encoder_attention_mask,
            past_key_value,
            output_attentions,
            rel_pos=rel_pos,
            rel_2d_pos=rel_2d_pos,
        )
        attention_output = self.output(self_outputs[0], hidden_states)
        if output_attentions:
            outputs = [
                attention_output,
            ] + self_outputs[1:]
        else:
            outputs = [attention_output]
        return outputs


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMEncoder with XLM->LMv2
class LayoutLMv2Encoder(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2Encoder, self).__init__()
        self.config = config
        self.layer = nn.LayerList([LayoutLMv2Layer(config) for _ in range(config.num_hidden_layers)])

        self.has_relative_attention_bias = config.has_relative_attention_bias
        self.has_spatial_attention_bias = config.has_spatial_attention_bias

        if self.has_relative_attention_bias:
            self.rel_pos_bins = config.rel_pos_bins
            self.max_rel_pos = config.max_rel_pos
            self.rel_pos_onehot_size = config.rel_pos_bins
            self.rel_pos_bias = nn.Linear(self.rel_pos_onehot_size, config.num_attention_heads, bias_attr=False)

        if self.has_spatial_attention_bias:
            self.max_rel_2d_pos = config.max_rel_2d_pos
            self.rel_2d_pos_bins = config.rel_2d_pos_bins
            self.rel_2d_pos_onehot_size = config.rel_2d_pos_bins
            self.rel_pos_x_bias = nn.Linear(self.rel_2d_pos_onehot_size, config.num_attention_heads, bias_attr=False)
            self.rel_pos_y_bias = nn.Linear(self.rel_2d_pos_onehot_size, config.num_attention_heads, bias_attr=False)

    def _cal_1d_pos_emb(self, hidden_states, position_ids):
        rel_pos_mat = position_ids.unsqueeze(-2) - position_ids.unsqueeze(-1)
        rel_pos = relative_position_bucket(
            rel_pos_mat,
            num_buckets=self.rel_pos_bins,
            max_distance=self.max_rel_pos,
        )
        rel_pos = paddle.nn.functional.one_hot(rel_pos, num_classes=self.rel_pos_onehot_size).astype(
            hidden_states.dtype
        )
        rel_pos = self.rel_pos_bias(rel_pos).transpose([0, 3, 1, 2])
        return rel_pos

    def _cal_2d_pos_emb(self, hidden_states, bbox):
        position_coord_x = bbox[:, :, 0]
        position_coord_y = bbox[:, :, 3]
        rel_pos_x_2d_mat = position_coord_x.unsqueeze(-2) - position_coord_x.unsqueeze(-1)
        rel_pos_y_2d_mat = position_coord_y.unsqueeze(-2) - position_coord_y.unsqueeze(-1)
        rel_pos_x = relative_position_bucket(
            rel_pos_x_2d_mat,
            num_buckets=self.rel_2d_pos_bins,
            max_distance=self.max_rel_2d_pos,
        )
        rel_pos_y = relative_position_bucket(
            rel_pos_y_2d_mat,
            num_buckets=self.rel_2d_pos_bins,
            max_distance=self.max_rel_2d_pos,
        )
        rel_pos_x = F.one_hot(rel_pos_x, num_classes=self.rel_2d_pos_onehot_size).astype(hidden_states.dtype)
        rel_pos_y = F.one_hot(rel_pos_y, num_classes=self.rel_2d_pos_onehot_size).astype(hidden_states.dtype)
        rel_pos_x = self.rel_pos_x_bias(rel_pos_x).transpose([0, 3, 1, 2])
        rel_pos_y = self.rel_pos_y_bias(rel_pos_y).transpose([0, 3, 1, 2])
        rel_2d_pos = rel_pos_x + rel_pos_y
        return rel_2d_pos

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        output_attentions=False,
        output_hidden_states=False,
        bbox=None,
        position_ids=None,
    ):
        all_hidden_states = () if output_hidden_states else None

        rel_pos = self._cal_1d_pos_emb(hidden_states, position_ids) if self.has_relative_attention_bias else None
        rel_2d_pos = self._cal_2d_pos_emb(hidden_states, bbox) if self.has_spatial_attention_bias else None

        hidden_save = dict()
        hidden_save["input_hidden_states"] = hidden_states

        for i, layer_module in enumerate(self.layer):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None
            past_key_value = past_key_values[i] if past_key_values is not None else None

            # gradient_checkpointing is set as False here so we remove some codes here
            hidden_save["input_attention_mask"] = attention_mask
            hidden_save["input_layer_head_mask"] = layer_head_mask
            layer_outputs = layer_module(
                hidden_states,
                attention_mask,
                layer_head_mask,
                encoder_hidden_states,
                encoder_attention_mask,
                past_key_value,
                output_attentions,
                rel_pos=rel_pos,
                rel_2d_pos=rel_2d_pos,
            )

            hidden_states = layer_outputs[0]

            hidden_save["{}_data".format(i)] = hidden_states

        return hidden_states, hidden_save


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMIntermediate with XLM->LMv2
class LayoutLMv2Intermediate(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2Intermediate, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if config.hidden_act == "gelu":
            self.intermediate_act_fn = nn.GELU()
        else:
            assert False, "hidden_act is set as: {}, please check it..".format(config.hidden_act)

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMOutput with XLM->LMv2
class LayoutLMv2Output(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2Output, self).__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, epsilon=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMLayer with XLM->LMv2
class LayoutLMv2Layer(nn.Layer):
    def __init__(self, config):
        super(LayoutLMv2Layer, self).__init__()
        # since chunk_size_feed_forward is 0 as default, no chunk is needed here.
        self.seq_len_dim = 1
        self.attention = LayoutLMv2Attention(config)
        self.add_cross_attention = False  # default as false
        self.intermediate = LayoutLMv2Intermediate(config)
        self.output = LayoutLMv2Output(config)

    def feed_forward_chunk(self, attention_output):
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
        rel_pos=None,
        rel_2d_pos=None,
    ):
        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
        self_attention_outputs = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions=output_attentions,
            past_key_value=self_attn_past_key_value,
            rel_pos=rel_pos,
            rel_2d_pos=rel_2d_pos,
        )
        attention_output = self_attention_outputs[0]

        layer_output = self.feed_forward_chunk(attention_output)

        if output_attentions:
            outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
            outputs = [
                layer_output,
            ] + list(outputs)
        else:
            outputs = [layer_output]
        return outputs


# Copied from paddlenlp.transformers.layoutxlm.modeling.VisualBackbone
class VisualBackbone(nn.Layer):
    def __init__(self, config):
        super(VisualBackbone, self).__init__()
        self.cfg = read_config()
        self.backbone = build_resnet_fpn_backbone(self.cfg)

        assert len(self.cfg.MODEL.PIXEL_MEAN) == len(self.cfg.MODEL.PIXEL_STD)
        num_channels = len(self.cfg.MODEL.PIXEL_MEAN)
        self.register_buffer("pixel_mean", paddle.to_tensor(self.cfg.MODEL.PIXEL_MEAN).reshape([num_channels, 1, 1]))
        self.register_buffer("pixel_std", paddle.to_tensor(self.cfg.MODEL.PIXEL_STD).reshape([num_channels, 1, 1]))
        self.out_feature_key = "p2"
        # is_deterministic is disabled here.
        self.pool = nn.AdaptiveAvgPool2D(config.image_feature_pool_shape[:2])
        if len(config.image_feature_pool_shape) == 2:
            config.image_feature_pool_shape.append(self.backbone.output_shape()[self.out_feature_key].channels)
        assert self.backbone.output_shape()[self.out_feature_key].channels == config.image_feature_pool_shape[2]

    def forward(self, images):
        images_input = (paddle.to_tensor(images) - self.pixel_mean) / self.pixel_std
        features = self.backbone(images_input)
        features = features[self.out_feature_key]
        features = self.pool(features).flatten(start_axis=2).transpose([0, 2, 1])
        return features


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMModel with XLM->LMv2
@register_base_model
class LayoutLMv2Model(LayoutLMv2PretrainedModel):
    """
    The bare LayoutLMv2 Model outputting raw hidden-states.

    This model inherits from :class:`~paddlenlp.transformers.model_utils.PretrainedModel`.
    Refer to the superclass documentation for the generic methods.

    This model is also a Paddle `paddle.nn.Layer <https://www.paddlepaddle.org.cn/documentation
    /docs/zh/api/paddle/nn/Layer_cn.html>`__ subclass. Use it as a regular Paddle Layer
    and refer to the Paddle documentation for all matter related to general usage and behavior.

    Args:
        vocab_size (`int`):
            Vocabulary size of the XLNet model. Defines the number of different tokens that can
            be represented by the `inputs_ids` passed when calling XLNetModel.
        hidden_size (`int`, optional):
            Dimensionality of the encoder layers and the pooler layer. Defaults to ``768``.
        num_hidden_layers (`int`, optional):
            Number of hidden layers in the Transformer encoder. Defaults to ``12``.
        num_attention_heads (`int`, optional):
            Number of attention heads for each attention layer in the Transformer encoder.
            Defaults to ``12``.
        intermediate_size (`int`, optional):
            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.
            Defaults to ``3072``.
        hidden_act (`str`, optional):
            The non-linear activation function in the feed-forward layer.
            ``"gelu"``, ``"relu"`` and any other paddle supported activation functions
            are supported. Defaults to ``"gelu"``.
        hidden_dropout_prob (`float`, optional):
            The dropout probability for all fully connected layers in the embeddings and encoder.
            Defaults to ``0.1``.
        attention_probs_dropout_prob (`float`, optional):
            The dropout probability for all fully connected layers in the pooler.
            Defaults to ``0.1``.
        initializer_range (`float`, optional):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
            Defaults to ``0.02``.
    """

    def __init__(self, config):
        super(LayoutLMv2Model, self).__init__(config)
        self.use_visual_backbone = config.use_visual_backbone
        self.has_visual_segment_embedding = config.has_visual_segment_embedding
        self.embeddings = LayoutLMv2Embeddings(config)

        if self.use_visual_backbone is True:
            self.visual = VisualBackbone(config)
            self.visual.stop_gradient = True
            self.visual_proj = nn.Linear(config.image_feature_pool_shape[-1], config.hidden_size)
        if self.has_visual_segment_embedding:
            self.visual_segment_embedding = self.create_parameter(
                shape=[
                    config.hidden_size,
                ],
                dtype=paddle.float32,
            )
        self.visual_LayerNorm = nn.LayerNorm(config.hidden_size, epsilon=config.layer_norm_eps)
        self.visual_dropout = nn.Dropout(config.hidden_dropout_prob)

        self.encoder = LayoutLMv2Encoder(config)
        self.pooler = LayoutLMv2Pooler(config.hidden_size, config.with_pool)
        self.config = config

    def _calc_text_embeddings(self, input_ids, bbox, position_ids, token_type_ids):
        words_embeddings = self.embeddings.word_embeddings(input_ids)
        position_embeddings = self.embeddings.position_embeddings(position_ids)
        spatial_position_embeddings = self.embeddings._cal_spatial_position_embeddings(bbox)
        token_type_embeddings = self.embeddings.token_type_embeddings(token_type_ids)
        embeddings = words_embeddings + position_embeddings + spatial_position_embeddings + token_type_embeddings
        embeddings = self.embeddings.LayerNorm(embeddings)
        embeddings = self.embeddings.dropout(embeddings)
        return embeddings

    def _calc_img_embeddings(self, image, bbox, position_ids):
        position_embeddings = self.embeddings.position_embeddings(position_ids)
        spatial_position_embeddings = self.embeddings._cal_spatial_position_embeddings(bbox)
        if self.use_visual_backbone is True:
            visual_embeddings = self.visual_proj(self.visual(image.astype(paddle.float32)))
            embeddings = visual_embeddings + position_embeddings + spatial_position_embeddings
        else:
            embeddings = position_embeddings + spatial_position_embeddings
        if self.has_visual_segment_embedding:
            embeddings += self.visual_segment_embedding
        embeddings = self.visual_LayerNorm(embeddings)
        embeddings = self.visual_dropout(embeddings)
        return embeddings

    def resize_position_embeddings(self, new_num_position_embeddings):
        """
        Resizes position embeddings of the model if `new_num_position_embeddings != config["max_position_embeddings"]`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embedding matrix. If position embeddings are learned, increasing the size
                will add newly initialized vectors at the end, whereas reducing the size will remove vectors from the
                end.
        """
        num_position_embeds_diff = new_num_position_embeddings - self.config.max_position_embeddings

        # no resizing needs to be done if the length stays the same
        if num_position_embeds_diff == 0:
            return

        logger.info(f"Setting `config.max_position_embeddings={new_num_position_embeddings}`...")
        self.config.max_position_embeddings = new_num_position_embeddings

        old_position_embeddings_weight = self.embeddings.position_embeddings.weight

        self.embeddings.position_embeddings = nn.Embedding(
            self.config.max_position_embeddings, self.config.hidden_size
        )

        with paddle.no_grad():
            if num_position_embeds_diff > 0:
                self.embeddings.position_embeddings.weight[:-num_position_embeds_diff] = old_position_embeddings_weight
            else:
                self.embeddings.position_embeddings.weight = old_position_embeddings_weight[:num_position_embeds_diff]

    def forward(
        self,
        input_ids=None,
        bbox=None,
        image=None,
        token_type_ids=None,
        position_ids=None,
        attention_mask=None,
        head_mask=None,
        output_hidden_states=False,
        output_attentions=False,
    ):
        input_shape = input_ids.shape

        visual_shape = list(input_shape)
        visual_shape[1] = self.config.image_feature_pool_shape[0] * self.config.image_feature_pool_shape[1]

        visual_bbox_x = (
            paddle.arange(
                0,
                1000 * (self.config.image_feature_pool_shape[1] + 1),
                1000,
                dtype=bbox.dtype,
            )
            // self.config.image_feature_pool_shape[1]
        )
        visual_bbox_y = (
            paddle.arange(
                0,
                1000 * (self.config.image_feature_pool_shape[0] + 1),
                1000,
                dtype=bbox.dtype,
            )
            // self.config.image_feature_pool_shape[0]
        )

        expand_shape = self.config.image_feature_pool_shape[0:2]

        visual_bbox = paddle.stack(
            [
                visual_bbox_x[:-1].expand(expand_shape),
                visual_bbox_y[:-1].expand(expand_shape[::-1]).transpose([1, 0]),
                visual_bbox_x[1:].expand(expand_shape),
                visual_bbox_y[1:].expand(expand_shape[::-1]).transpose([1, 0]),
            ],
            axis=-1,
        ).reshape([expand_shape[0] * expand_shape[1], bbox.shape[-1]])
        visual_bbox = visual_bbox.expand([input_shape[0], visual_bbox.shape[0], visual_bbox.shape[1]])
        final_bbox = paddle.concat([bbox, visual_bbox], axis=1)

        if attention_mask is None:
            attention_mask = paddle.ones(input_shape)

        if self.use_visual_backbone is True:
            visual_attention_mask = paddle.ones(visual_shape)
        else:
            visual_attention_mask = paddle.zeros(visual_shape)

        attention_mask = attention_mask.astype(visual_attention_mask.dtype)

        final_attention_mask = paddle.concat([attention_mask, visual_attention_mask], axis=1)

        if token_type_ids is None:
            token_type_ids = paddle.zeros(input_shape, dtype=paddle.int64)

        if position_ids is None:
            seq_length = input_shape[1]
            position_ids = self.embeddings.position_ids[:, :seq_length]
            position_ids = position_ids.expand(input_shape)

        visual_position_ids = paddle.arange(0, visual_shape[1]).expand([input_shape[0], visual_shape[1]])
        final_position_ids = paddle.concat([position_ids, visual_position_ids], axis=1)

        if bbox is None:
            bbox = paddle.zeros(input_shape + [4])

        text_layout_emb = self._calc_text_embeddings(
            input_ids=input_ids,
            bbox=bbox,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
        )

        visual_emb = self._calc_img_embeddings(
            image=image,
            bbox=visual_bbox,
            position_ids=visual_position_ids,
        )
        final_emb = paddle.concat([text_layout_emb, visual_emb], axis=1)

        extended_attention_mask = final_attention_mask.unsqueeze(1).unsqueeze(2)

        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0

        if head_mask is not None:
            if head_mask.dim() == 1:
                head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)
                head_mask = head_mask.expand(self.config.num_hidden_layers, -1, -1, -1, -1)
            elif head_mask.dim() == 2:
                head_mask = head_mask.unsqueeze(1).unsqueeze(-1).unsqueeze(-1)
            head_mask = head_mask.to(dtype=next(self.parameters()).dtype)
        else:
            head_mask = [None] * self.config.num_hidden_layers

        encoder_outputs = self.encoder(
            final_emb,
            extended_attention_mask,
            bbox=final_bbox,
            position_ids=final_position_ids,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )
        sequence_output = encoder_outputs[0]
        pooled_output = self.pooler(sequence_output)
        # i_data (i in [0, 12) is the key of the hidden states
        return sequence_output, pooled_output, encoder_outputs[1]


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMForTokenClassification with XLM->LMv2
class LayoutLMv2ForTokenClassification(LayoutLMv2PretrainedModel):
    def __init__(self, config):
        super(LayoutLMv2ForTokenClassification, self).__init__(config)
        self.num_labels = config.num_labels
        self.layoutlmv2 = LayoutLMv2Model(config)
        self.dropout = nn.Dropout(
            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
        )
        self.classifier = nn.Linear(config.hidden_size, self.num_labels)
        self.num_hidden_layers = config.num_hidden_layers

    def get_input_embeddings(self):
        return self.layoutlmv2.embeddings.word_embeddings

    def resize_position_embeddings(self, new_num_position_embeddings):
        """
        Resizes position embeddings of the model if `new_num_position_embeddings != config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embedding matrix. If position embeddings are learned, increasing the size
                will add newly initialized vectors at the end, whereas reducing the size will remove vectors from the
                end.
        """
        self.layoutlmv2.resize_position_embeddings(new_num_position_embeddings)

    def forward(
        self,
        input_ids=None,
        bbox=None,
        image=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        labels=None,
    ):
        outputs = self.layoutlmv2(
            input_ids=input_ids,
            bbox=bbox,
            image=image,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            head_mask=head_mask,
        )
        seq_length = input_ids.shape[1]
        # sequence out and image out
        sequence_output, _ = outputs[0][:, :seq_length], outputs[0][:, seq_length:]
        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)

        hidden_states = {f"hidden_states_{idx}": outputs[2][f"{idx}_data"] for idx in range(self.num_hidden_layers)}

        if self.training:
            outputs = logits, hidden_states
        else:
            outputs = (logits,)

        if labels is not None:
            loss_fct = nn.CrossEntropyLoss()

            if attention_mask is not None:
                active_loss = (
                    attention_mask.reshape(
                        [
                            -1,
                        ]
                    )
                    == 1
                )
                active_logits = logits.reshape([-1, self.num_classes])[active_loss]
                active_labels = labels.reshape(
                    [
                        -1,
                    ]
                )[active_loss]
                loss = loss_fct(active_logits, active_labels)
            else:
                loss = loss_fct(
                    logits.reshape([-1, self.num_classes]),
                    labels.reshape(
                        [
                            -1,
                        ]
                    ),
                )

            outputs = (loss,) + outputs

        return outputs


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMPredictionHead with XLM->LMv2
class LayoutLMv2PredictionHead(Layer):
    """
    Bert Model with a `language modeling` head on top for CLM fine-tuning.
    """

    def __init__(self, hidden_size, vocab_size, activation, embedding_weights=None):
        super(LayoutLMv2PredictionHead, self).__init__()
        self.transform = nn.Linear(hidden_size, hidden_size)
        self.activation = getattr(nn.functional, activation)
        self.layer_norm = nn.LayerNorm(hidden_size)
        self.decoder_weight = (
            self.create_parameter(shape=[vocab_size, hidden_size], dtype=self.transform.weight.dtype, is_bias=False)
            if embedding_weights is None
            else embedding_weights
        )
        self.decoder_bias = self.create_parameter(shape=[vocab_size], dtype=self.decoder_weight.dtype, is_bias=True)

    def forward(self, hidden_states, masked_positions=None):
        if masked_positions is not None:
            hidden_states = paddle.reshape(hidden_states, [-1, hidden_states.shape[-1]])
            hidden_states = paddle.tensor.gather(hidden_states, masked_positions)
        # gather masked tokens might be more quick
        hidden_states = self.transform(hidden_states)
        hidden_states = self.activation(hidden_states)
        hidden_states = self.layer_norm(hidden_states)
        hidden_states = paddle.tensor.matmul(hidden_states, self.decoder_weight, transpose_y=True) + self.decoder_bias
        return hidden_states


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMPretrainingHeads with XLM->LMv2
class LayoutLMv2PretrainingHeads(Layer):
    def __init__(self, hidden_size, vocab_size, activation, embedding_weights=None):
        super(LayoutLMv2PretrainingHeads, self).__init__()
        self.predictions = LayoutLMv2PredictionHead(hidden_size, vocab_size, activation, embedding_weights)

    def forward(self, sequence_output, masked_positions=None):
        prediction_scores = self.predictions(sequence_output, masked_positions)
        return prediction_scores


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMForPretraining with XLM->LMv2
class LayoutLMv2ForPretraining(LayoutLMv2PretrainedModel):
    def __init__(self, config):
        super(LayoutLMv2ForPretraining, self).__init__(config)
        self.layoutlmv2 = LayoutLMv2Model(config)
        self.cls = LayoutLMv2PretrainingHeads(
            self.layoutlmv2.config.hidden_size,
            self.layoutlmv2.config.vocab_size,
            self.layoutlmv2.config.hidden_act,
            embedding_weights=self.layoutlmv2.embeddings.word_embeddings.weight,
        )

    def resize_position_embeddings(self, new_num_position_embeddings):
        """
        Resizes position embeddings of the model if `new_num_position_embeddings != config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embedding matrix. If position embeddings are learned, increasing the size
                will add newly initialized vectors at the end, whereas reducing the size will remove vectors from the
                end.
        """
        self.layoutlmv2.resize_position_embeddings(new_num_position_embeddings)

    def forward(
        self,
        input_ids=None,
        bbox=None,
        image=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        masked_positions=None,
    ):
        outputs = self.layoutlmv2(
            input_ids=input_ids,
            bbox=bbox,
            image=image,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            head_mask=head_mask,
        )
        sequence_output = outputs[0]
        prediction_scores = self.cls(sequence_output, masked_positions)
        return prediction_scores


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMOutput with XLM->LMv2
class BiaffineAttention(nn.Layer):
    """Implements a biaffine attention operator for binary relation classification."""

    def __init__(self, in_features, out_features):
        super(BiaffineAttention, self).__init__()

        self.in_features = in_features
        self.out_features = out_features

        self.bilinear = nn.Bilinear(in_features, in_features, out_features, bias_attr=False)
        self.linear = nn.Linear(2 * in_features, out_features)

    def forward(self, x_1, x_2):
        return self.bilinear(x_1, x_2) + self.linear(paddle.concat((x_1, x_2), axis=-1))


# Copied from paddlenlp.transformers.layoutxlm.modeling.REDecoder
class REDecoder(nn.Layer):
    def __init__(self, hidden_size=768, hidden_dropout_prob=0.1):
        super(REDecoder, self).__init__()
        self.entity_emb = nn.Embedding(3, hidden_size)
        projection = nn.Sequential(
            nn.Linear(hidden_size * 2, hidden_size),
            nn.ReLU(),
            nn.Dropout(hidden_dropout_prob),
            nn.Linear(hidden_size, hidden_size // 2),
            nn.ReLU(),
            nn.Dropout(hidden_dropout_prob),
        )
        self.ffnn_head = copy.deepcopy(projection)
        self.ffnn_tail = copy.deepcopy(projection)
        self.rel_classifier = BiaffineAttention(hidden_size // 2, 2)
        self.loss_fct = CrossEntropyLoss()

    def build_relation(self, relations, entities):
        batch_size = len(relations)
        new_relations = []
        for b in range(batch_size):
            if len(entities[b]["start"]) <= 2:
                entities[b] = {"end": [1, 1], "label": [0, 0], "start": [0, 0]}
            all_possible_relations = set(
                [
                    (i, j)
                    for i in range(len(entities[b]["label"]))
                    for j in range(len(entities[b]["label"]))
                    if entities[b]["label"][i] == 1 and entities[b]["label"][j] == 2
                ]
            )
            if len(all_possible_relations) == 0:
                all_possible_relations = {(0, 1)}
            if "head" in relations[b]:
                positive_relations = set(list(zip(relations[b]["head"], relations[b]["tail"])))
            else:
                positive_relations = set()
            negative_relations = all_possible_relations - positive_relations
            positive_relations = set([i for i in positive_relations if i in all_possible_relations])
            reordered_relations = list(positive_relations) + list(negative_relations)
            relation_per_doc = {
                "head": [i[0] for i in reordered_relations],
                "tail": [i[1] for i in reordered_relations],
                "label": [1] * len(positive_relations) + [0] * (len(reordered_relations) - len(positive_relations)),
            }
            assert len(relation_per_doc["head"]) != 0
            new_relations.append(relation_per_doc)
        return new_relations, entities

    def get_predicted_relations(self, logits, relations, entities):
        pred_relations = []
        for i, pred_label in enumerate(logits.argmax(-1)):
            if pred_label != 1:
                continue
            rel = {}
            rel["head_id"] = relations["head"][i]
            rel["head"] = (entities["start"][rel["head_id"]], entities["end"][rel["head_id"]])
            rel["head_type"] = entities["label"][rel["head_id"]]

            rel["tail_id"] = relations["tail"][i]
            rel["tail"] = (entities["start"][rel["tail_id"]], entities["end"][rel["tail_id"]])
            rel["tail_type"] = entities["label"][rel["tail_id"]]
            rel["type"] = 1
            pred_relations.append(rel)
        return pred_relations

    def forward(self, hidden_states, entities, relations):
        batch_size, max_n_words, context_dim = hidden_states.shape
        relations, entities = self.build_relation(relations, entities)
        loss = 0
        all_pred_relations = []
        for b in range(batch_size):
            if "head" not in relations[b]:
                continue
            head_entities = paddle.to_tensor(relations[b]["head"])
            tail_entities = paddle.to_tensor(relations[b]["tail"])
            relation_labels = paddle.to_tensor(relations[b]["label"])
            entities_start_index = paddle.to_tensor(entities[b]["start"])
            entities_labels = paddle.to_tensor(entities[b]["label"])
            head_index = entities_start_index[head_entities]
            head_label = entities_labels[head_entities]
            head_label_repr = self.entity_emb(head_label)

            tail_index = entities_start_index[tail_entities]
            tail_label = entities_labels[tail_entities]
            tail_label_repr = self.entity_emb(tail_label)

            tmp_hidden_states = hidden_states[b][head_index]
            if len(tmp_hidden_states.shape) == 1:
                tmp_hidden_states = paddle.unsqueeze(tmp_hidden_states, axis=0)
            head_repr = paddle.concat((tmp_hidden_states, head_label_repr), axis=-1)

            tmp_hidden_states = hidden_states[b][tail_index]
            if len(tmp_hidden_states.shape) == 1:
                tmp_hidden_states = paddle.unsqueeze(tmp_hidden_states, axis=0)
            tail_repr = paddle.concat((tmp_hidden_states, tail_label_repr), axis=-1)

            heads = self.ffnn_head(head_repr)
            tails = self.ffnn_tail(tail_repr)
            logits = self.rel_classifier(heads, tails)
            loss += self.loss_fct(logits, relation_labels)
            pred_relations = self.get_predicted_relations(logits, relations[b], entities[b])
            all_pred_relations.append(pred_relations)
        return loss, all_pred_relations


# Copied from paddlenlp.transformers.layoutxlm.modeling.LayoutXLMForRelationExtraction with XLM->LMv2
class LayoutLMv2ForRelationExtraction(LayoutLMv2PretrainedModel):
    def __init__(self, config):
        super(LayoutLMv2ForRelationExtraction, self).__init__(config)
        self.layoutlmv2 = LayoutLMv2Model(config)

        self.extractor = REDecoder(config.hidden_size, config.hidden_dropout_prob)

        self.dropout = nn.Dropout(
            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
        )

    def _init_weights(self, layer):
        """Initialize the weights"""
        if isinstance(layer, nn.Linear):
            layer.weight.set_value(paddle.tensor.normal(mean=0.0, std=0.02, shape=layer.weight.shape))
            if layer.bias is not None:
                layer.bias.set_value(paddle.tensor.zeros(shape=layer.bias.shape))
        elif isinstance(layer, nn.Embedding):
            layer.weight.set_value(paddle.tensor.normal(mean=0.0, std=0.02, shape=layer.weight.shape))
            if layer._padding_idx is not None:
                layer.weight[layer._padding_idx].set_value(
                    paddle.tensor.normal(mean=0.0, std=0.02, shape=layer.weight[layer._padding_idx].shape)
                )
        elif isinstance(layer, nn.LayerNorm):
            layer.weight.set_value(paddle.tensor.ones(shape=layer.bias.shape))
            layer.bias.set_value(paddle.tensor.zeros(shape=layer.bias.shape))

    def resize_position_embeddings(self, new_num_position_embeddings):
        """
        Resizes position embeddings of the model if `new_num_position_embeddings != config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embedding matrix. If position embeddings are learned, increasing the size
                will add newly initialized vectors at the end, whereas reducing the size will remove vectors from the
                end.
        """
        self.layoutlmv2.resize_position_embeddings(new_num_position_embeddings)

    def forward(
        self,
        input_ids,
        bbox,
        labels=None,
        image=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        entities=None,
        relations=None,
    ):
        outputs = self.layoutlmv2(
            input_ids=input_ids,
            bbox=bbox,
            image=image,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            head_mask=head_mask,
        )

        seq_length = input_ids.shape[1]
        sequence_output, _ = outputs[0][:, :seq_length], outputs[0][:, seq_length:]
        sequence_output = self.dropout(sequence_output)
        loss, pred_relations = self.extractor(sequence_output, entities, relations)

        return dict(
            loss=loss,
            entities=entities,
            relations=relations,
            pred_relations=pred_relations,
            hidden_states=outputs[0],
        )