Source code for transformers.models.beit.modeling_beit

# coding=utf-8
# Copyright 2021 Google AI, Ross Wightman, 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.
""" PyTorch BEiT model. """


import collections.abc
import math

import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss

from ...activations import ACT2FN
from ...file_utils import add_start_docstrings, add_start_docstrings_to_model_forward, replace_return_docstrings
from ...modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling, MaskedLMOutput, SequenceClassifierOutput
from ...modeling_utils import PreTrainedModel, find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import logging
from .configuration_beit import BeitConfig


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "BeitConfig"
_CHECKPOINT_FOR_DOC = "microsoft/beit-base-patch16-224"

BEIT_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "microsoft/beit-base-patch16-224",
    # See all BEiT models at https://huggingface.co/models?filter=beit
]


# Inspired by
# https://github.com/rwightman/pytorch-image-models/blob/b9bd960a032c75ca6b808ddeed76bee5f3ed4972/timm/models/layers/helpers.py
# From PyTorch internals
def to_2tuple(x):
    if isinstance(x, collections.abc.Iterable):
        return x
    return (x, x)


# Based on https://github.com/rwightman/pytorch-image-models/blob/a2727c1bf78ba0d7b5727f5f95e37fb7f8866b1f/timm/models/layers/drop.py
def drop_path(x, drop_prob: float = 0.0, training: bool = False):
    """
    Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

    Comment by Ross Wightman: This is the same as the DropConnect impl I created for EfficientNet, etc networks,
    however, the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for changing the
    layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use 'survival rate' as the
    argument.
    """
    if drop_prob == 0.0 or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks)."""

    def __init__(self, drop_prob=None):
        super().__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)

    def extra_repr(self) -> str:
        return "p={}".format(self.drop_prob)


# Based on timm implementation, which can be found here:
# https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
class BeitEmbeddings(nn.Module):
    """
    Construct the CLS token, position and patch embeddings. Optionally, also the mask token.

    """

    def __init__(self, config):
        super().__init__()

        self.cls_token = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
        if config.use_mask_token:
            self.mask_token = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
        else:
            self.mask_token = None
        self.patch_embeddings = PatchEmbeddings(
            image_size=config.image_size,
            patch_size=config.patch_size,
            num_channels=config.num_channels,
            embed_dim=config.hidden_size,
        )
        num_patches = self.patch_embeddings.num_patches
        if config.use_absolute_position_embeddings:
            self.position_embeddings = nn.Parameter(torch.zeros(1, num_patches + 1, config.hidden_size))
        else:
            self.position_embeddings = None
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, pixel_values, bool_masked_pos=None):

        embeddings = self.patch_embeddings(pixel_values)
        batch_size, seq_len, _ = embeddings.size()

        cls_tokens = self.cls_token.expand(batch_size, -1, -1)
        if bool_masked_pos is not None:
            mask_tokens = self.mask_token.expand(batch_size, seq_len, -1)
            # replace the masked visual tokens by mask_tokens
            w = bool_masked_pos.unsqueeze(-1).type_as(mask_tokens)
            embeddings = embeddings * (1 - w) + mask_tokens * w

        embeddings = torch.cat((cls_tokens, embeddings), dim=1)
        if self.position_embeddings is not None:
            embeddings = embeddings + self.position_embeddings
        embeddings = self.dropout(embeddings)

        return embeddings


# Based on timm implementation, which can be found here:
# https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
class PatchEmbeddings(nn.Module):
    """
    Image to Patch Embedding.
    """

    def __init__(self, image_size=224, patch_size=16, num_channels=3, embed_dim=768):
        super().__init__()
        image_size = to_2tuple(image_size)
        patch_size = to_2tuple(patch_size)
        num_patches = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
        patch_shape = (image_size[0] // patch_size[0], image_size[1] // patch_size[1])
        self.image_size = image_size
        self.patch_size = patch_size
        self.num_patches = num_patches
        self.patch_shape = patch_shape

        self.projection = nn.Conv2d(num_channels, embed_dim, kernel_size=patch_size, stride=patch_size)

    def forward(self, pixel_values):
        batch_size, num_channels, height, width = pixel_values.shape
        # FIXME look at relaxing size constraints
        if height != self.image_size[0] or width != self.image_size[1]:
            raise ValueError(
                f"Input image size ({height}*{width}) doesn't match model ({self.image_size[0]}*{self.image_size[1]})."
            )
        x = self.projection(pixel_values).flatten(2).transpose(1, 2)
        return x


class BeitSelfAttention(nn.Module):
    def __init__(self, config, window_size=None):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                f"The hidden size {config.hidden_size,} is not a multiple of the number of attention "
                f"heads {config.num_attention_heads}."
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size, bias=False)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)

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

        if window_size:
            self.relative_position_bias = BeitRelativePositionBias(config, window_size=window_size)
        else:
            self.relative_position_bias = None

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, hidden_states, head_mask=None, output_attentions=False, relative_position_bias=None):
        mixed_query_layer = self.query(hidden_states)

        key_layer = self.transpose_for_scores(self.key(hidden_states))
        value_layer = self.transpose_for_scores(self.value(hidden_states))
        query_layer = self.transpose_for_scores(mixed_query_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)

        # Add relative position bias if present.
        if self.relative_position_bias is not None:
            attention_scores = attention_scores + self.relative_position_bias().unsqueeze(0)

        # Add shared relative position bias if provided.
        if relative_position_bias is not None:
            attention_scores = attention_scores + relative_position_bias

        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs = attention_probs * head_mask

        context_layer = torch.matmul(attention_probs, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)

        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)

        return outputs


class BeitSelfOutput(nn.Module):
    """
    The residual connection is defined in BeitLayer instead of here (as is the case with other models), due to the
    layernorm applied before each block.
    """

    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor, gamma=None):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)

        return hidden_states


class BeitAttention(nn.Module):
    def __init__(self, config, window_size=None):
        super().__init__()
        self.attention = BeitSelfAttention(config, window_size=window_size)
        self.output = BeitSelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.attention.num_attention_heads, self.attention.attention_head_size, self.pruned_heads
        )

        # Prune linear layers
        self.attention.query = prune_linear_layer(self.attention.query, index)
        self.attention.key = prune_linear_layer(self.attention.key, index)
        self.attention.value = prune_linear_layer(self.attention.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.attention.num_attention_heads = self.attention.num_attention_heads - len(heads)
        self.attention.all_head_size = self.attention.attention_head_size * self.attention.num_attention_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    def forward(self, hidden_states, head_mask=None, output_attentions=False, relative_position_bias=None):
        self_outputs = self.attention(hidden_states, head_mask, output_attentions, relative_position_bias)

        attention_output = self.output(self_outputs[0], hidden_states)

        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs


class BeitIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

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

        return hidden_states


class BeitOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)

        return hidden_states


class BeitLayer(nn.Module):
    """This corresponds to the Block class in the timm implementation."""

    def __init__(self, config, window_size=None, drop_path_rate=0.0):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = BeitAttention(config, window_size=window_size)
        self.intermediate = BeitIntermediate(config)
        self.output = BeitOutput(config)
        self.layernorm_before = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.drop_path = DropPath(drop_path_rate) if drop_path_rate > 0.0 else nn.Identity()
        self.layernorm_after = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

        init_values = config.layer_scale_init_value
        if init_values > 0:
            self.lambda_1 = nn.Parameter(init_values * torch.ones((config.hidden_size)), requires_grad=True)
            self.lambda_2 = nn.Parameter(init_values * torch.ones((config.hidden_size)), requires_grad=True)
        else:
            self.lambda_1, self.lambda_2 = None, None

    def forward(self, hidden_states, head_mask=None, output_attentions=False, relative_position_bias=None):
        self_attention_outputs = self.attention(
            self.layernorm_before(hidden_states),  # in BEiT, layernorm is applied before self-attention
            head_mask,
            output_attentions=output_attentions,
            relative_position_bias=relative_position_bias,
        )
        attention_output = self_attention_outputs[0]
        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights

        # apply lambda_1 if present
        if self.lambda_1 is not None:
            attention_output = self.lambda_1 * attention_output

        # first residual connection
        hidden_states = self.drop_path(attention_output) + hidden_states

        # in BEiT, layernorm is also applied after self-attention
        layer_output = self.layernorm_after(hidden_states)

        layer_output = self.intermediate(layer_output)
        layer_output = self.output(layer_output)

        if self.lambda_2 is not None:
            layer_output = self.lambda_2 * layer_output

        # second residual connection
        layer_output = self.drop_path(layer_output) + hidden_states

        outputs = (layer_output,) + outputs

        return outputs


class BeitRelativePositionBias(nn.Module):
    def __init__(self, config, window_size):
        super().__init__()
        self.window_size = window_size
        self.num_relative_distance = (2 * window_size[0] - 1) * (2 * window_size[1] - 1) + 3
        self.relative_position_bias_table = nn.Parameter(
            torch.zeros(self.num_relative_distance, config.num_attention_heads)
        )  # 2*Wh-1 * 2*Ww-1, nH
        # cls to token & token 2 cls & cls to cls

        # get pair-wise relative position index for each token inside the window
        coords_h = torch.arange(window_size[0])
        coords_w = torch.arange(window_size[1])
        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
        relative_coords[:, :, 0] += window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += window_size[1] - 1
        relative_coords[:, :, 0] *= 2 * window_size[1] - 1
        relative_position_index = torch.zeros(
            size=(window_size[0] * window_size[1] + 1,) * 2, dtype=relative_coords.dtype
        )
        relative_position_index[1:, 1:] = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
        relative_position_index[0, 0:] = self.num_relative_distance - 3
        relative_position_index[0:, 0] = self.num_relative_distance - 2
        relative_position_index[0, 0] = self.num_relative_distance - 1

        self.register_buffer("relative_position_index", relative_position_index)

    def forward(self):
        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
            self.window_size[0] * self.window_size[1] + 1, self.window_size[0] * self.window_size[1] + 1, -1
        )  # Wh*Ww,Wh*Ww,nH

        return relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww


class BeitEncoder(nn.Module):
    def __init__(self, config, window_size=None):
        super().__init__()
        self.config = config
        if config.use_shared_relative_position_bias:
            self.relative_position_bias = BeitRelativePositionBias(config, window_size=window_size)
        else:
            self.relative_position_bias = None

        # stochastic depth decay rule
        dpr = [x.item() for x in torch.linspace(0, config.drop_path_rate, config.num_hidden_layers)]
        self.layer = nn.ModuleList(
            [
                BeitLayer(
                    config,
                    window_size=window_size if config.use_relative_position_bias else None,
                    drop_path_rate=dpr[i],
                )
                for i in range(config.num_hidden_layers)
            ]
        )
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states,
        head_mask=None,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
    ):
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None

        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

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, output_attentions)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    layer_head_mask,
                )
            else:
                relative_position_bias = (
                    self.relative_position_bias() if self.relative_position_bias is not None else None
                )
                layer_outputs = layer_module(hidden_states, layer_head_mask, output_attentions, relative_position_bias)

            hidden_states = layer_outputs[0]

            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, all_hidden_states, all_self_attentions] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


class BeitPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = BeitConfig
    base_model_prefix = "beit"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, (nn.Linear, nn.Conv2d)):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, BeitEncoder):
            module.gradient_checkpointing = value


BEIT_START_DOCSTRING = r"""
    This model is a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`_ subclass. Use
    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
    behavior.

    Parameters:
        config (:class:`~transformers.BeitConfig`): 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 :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model
            weights.
"""

BEIT_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using :class:`~transformers.BeitFeatureExtractor`. See
            :meth:`transformers.BeitFeatureExtractor.__call__` for details.

        head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`):
            Mask to nullify selected heads of the self-attention modules. Mask values selected in ``[0, 1]``:

            - 1 indicates the head is **not masked**,
            - 0 indicates the head is **masked**.

        output_attentions (:obj:`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 (:obj:`bool`, `optional`):
            Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
            more detail.
        return_dict (:obj:`bool`, `optional`):
            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
"""


[docs]@add_start_docstrings( "The bare Beit Model transformer outputting raw hidden-states without any specific head on top.", BEIT_START_DOCSTRING, ) class BeitModel(BeitPreTrainedModel): def __init__(self, config, add_pooling_layer=True): super().__init__(config) self.config = config self.embeddings = BeitEmbeddings(config) self.encoder = BeitEncoder(config, window_size=self.embeddings.patch_embeddings.patch_shape) self.layernorm = ( nn.Identity() if config.use_mean_pooling else nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) ) self.pooler = BeitPooler(config) if add_pooling_layer else None self.init_weights() def get_input_embeddings(self): return self.embeddings.patch_embeddings def _prune_heads(self, heads_to_prune): """ Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base class PreTrainedModel """ for layer, heads in heads_to_prune.items(): self.encoder.layer[layer].attention.prune_heads(heads)
[docs] @add_start_docstrings_to_model_forward(BEIT_INPUTS_DOCSTRING) @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=_CONFIG_FOR_DOC) def forward( self, pixel_values=None, bool_masked_pos=None, head_mask=None, output_attentions=None, output_hidden_states=None, return_dict=None, ): r""" Returns: Examples:: >>> from transformers import BeitFeatureExtractor, BeitModel >>> from PIL import Image >>> import requests >>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg' >>> image = Image.open(requests.get(url, stream=True).raw) >>> feature_extractor = BeitFeatureExtractor.from_pretrained('microsoft/beit-base-patch16-224-pt22k-ft22k') >>> model = BeitModel.from_pretrained('microsoft/beit-base-patch16-224-pt22k-ft22k') >>> inputs = feature_extractor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> last_hidden_states = outputs.last_hidden_state """ output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions output_hidden_states = ( output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states ) return_dict = return_dict if return_dict is not None else self.config.use_return_dict if pixel_values is None: raise ValueError("You have to specify pixel_values") # Prepare head mask if needed # 1.0 in head_mask indicate we keep the head # attention_probs has shape bsz x n_heads x N x N # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads] # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length] head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers) embedding_output = self.embeddings(pixel_values, bool_masked_pos) encoder_outputs = self.encoder( embedding_output, head_mask=head_mask, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) sequence_output = encoder_outputs[0] sequence_output = self.layernorm(sequence_output) pooled_output = self.pooler(sequence_output) if self.pooler is not None else None if not return_dict: return (sequence_output, pooled_output) + encoder_outputs[1:] return BaseModelOutputWithPooling( last_hidden_state=sequence_output, pooler_output=pooled_output, hidden_states=encoder_outputs.hidden_states, attentions=encoder_outputs.attentions, )
class BeitPooler(nn.Module): def __init__(self, config): super().__init__() self.layernorm = ( nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) if config.use_mean_pooling else None ) def forward(self, hidden_states): if self.layernorm is not None: # Mean pool the final hidden states of the patch tokens patch_tokens = hidden_states[:, 1:, :] pooled_output = self.layernorm(patch_tokens.mean(1)) else: # Pool by simply taking the final hidden state of the [CLS] token pooled_output = hidden_states[:, 0] return pooled_output
[docs]@add_start_docstrings( "Beit Model transformer with a 'language' modeling head on top (to predict visual tokens).", BEIT_START_DOCSTRING, ) class BeitForMaskedImageModeling(BeitPreTrainedModel): def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.beit = BeitModel(config, add_pooling_layer=False) # Classifier head self.layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) self.lm_head = nn.Linear(config.hidden_size, config.vocab_size) self.init_weights()
[docs] @add_start_docstrings_to_model_forward(BEIT_INPUTS_DOCSTRING) @replace_return_docstrings(output_type=MaskedLMOutput, config_class=_CONFIG_FOR_DOC) def forward( self, pixel_values=None, bool_masked_pos=None, head_mask=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None, ): r""" bool_masked_pos (:obj:`torch.BoolTensor` of shape :obj:`(batch_size, num_patches)`): Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`): Labels for computing the image classification/regression loss. Indices should be in :obj:`[0, ..., config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss), If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy). Returns: Examples:: >>> from transformers import BeitFeatureExtractor, BeitForMaskedImageModeling >>> from PIL import Image >>> import requests >>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg' >>> image = Image.open(requests.get(url, stream=True).raw) >>> feature_extractor = BeitFeatureExtractor.from_pretrained('microsoft/beit-base-patch16-224-pt22k') >>> model = BeitForMaskedImageModeling.from_pretrained('microsoft/beit-base-patch16-224-pt22k') >>> inputs = feature_extractor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict outputs = self.beit( pixel_values, bool_masked_pos=bool_masked_pos, head_mask=head_mask, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) sequence_output = outputs[0] sequence_output = self.layernorm(sequence_output) prediction_scores = self.lm_head(sequence_output[:, 1:]) masked_lm_loss = None if labels is not None: loss_fct = CrossEntropyLoss() # -100 index = padding token masked_lm_loss = loss_fct(prediction_scores[bool_masked_pos], labels) if not return_dict: output = (prediction_scores,) + outputs[2:] return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output return MaskedLMOutput( loss=masked_lm_loss, logits=prediction_scores, hidden_states=outputs.hidden_states, attentions=outputs.attentions, )
[docs]@add_start_docstrings( """ Beit Model transformer with an image classification head on top (a linear layer on top of the average of the final hidden states of the patch tokens) e.g. for ImageNet. """, BEIT_START_DOCSTRING, ) class BeitForImageClassification(BeitPreTrainedModel): def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.beit = BeitModel(config, add_pooling_layer=True) # Classifier head self.classifier = nn.Linear(config.hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity() self.init_weights()
[docs] @add_start_docstrings_to_model_forward(BEIT_INPUTS_DOCSTRING) @replace_return_docstrings(output_type=SequenceClassifierOutput, config_class=_CONFIG_FOR_DOC) def forward( self, pixel_values=None, head_mask=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None, ): r""" labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`): Labels for computing the image classification/regression loss. Indices should be in :obj:`[0, ..., config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss), If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy). Returns: Examples:: >>> from transformers import BeitFeatureExtractor, BeitForImageClassification >>> from PIL import Image >>> import requests >>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg' >>> image = Image.open(requests.get(url, stream=True).raw) >>> feature_extractor = BeitFeatureExtractor.from_pretrained('microsoft/beit-base-patch16-224') >>> model = BeitForImageClassification.from_pretrained('microsoft/beit-base-patch16-224') >>> inputs = feature_extractor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits >>> # model predicts one of the 1000 ImageNet classes >>> predicted_class_idx = logits.argmax(-1).item() >>> print("Predicted class:", model.config.id2label[predicted_class_idx]) """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict outputs = self.beit( pixel_values, head_mask=head_mask, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) pooled_output = outputs.pooler_output if return_dict else outputs[1] logits = self.classifier(pooled_output) loss = None if labels is not None: if self.num_labels == 1: # We are doing regression loss_fct = MSELoss() loss = loss_fct(logits.view(-1), labels.view(-1)) else: loss_fct = CrossEntropyLoss() loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) if not return_dict: output = (logits,) + outputs[2:] return ((loss,) + output) if loss is not None else output return SequenceClassifierOutput( loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions, )