Upload 2 files

configuration_vit.py

@@ -0,0 +1,138 @@
+# coding=utf-8
+# Copyright 2021 Google AI 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.
+"""ViT model configuration"""
+
+from collections import OrderedDict
+from typing import Mapping
+
+from packaging import version
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.onnx import OnnxConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class ViTConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`ViTModel`]. It is used to instantiate an ViT
+    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 ViT
+    [google/vit-base-patch16-224](https://huggingface.co/google/vit-base-patch16-224) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimensionality 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):
+            Dimensionality 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 0.0):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        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.
+        image_size (`int`, *optional*, defaults to 224):
+            The size (resolution) of each image.
+        patch_size (`int`, *optional*, defaults to 16):
+            The size (resolution) of each patch.
+        num_channels (`int`, *optional*, defaults to 3):
+            The number of input channels.
+        qkv_bias (`bool`, *optional*, defaults to `True`):
+            Whether to add a bias to the queries, keys and values.
+        encoder_stride (`int`, *optional*, defaults to 16):
+           Factor to increase the spatial resolution by in the decoder head for masked image modeling.
+
+    Example:
+
+    ```python
+    >>> from transformers import ViTConfig, ViTModel
+
+    >>> # Initializing a ViT vit-base-patch16-224 style configuration
+    >>> configuration = ViTConfig()
+
+    >>> # Initializing a model (with random weights) from the vit-base-patch16-224 style configuration
+    >>> model = ViTModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "vit"
+
+    def __init__(
+        self,
+        hidden_size=768,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        intermediate_size=3072,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.0,
+        attention_probs_dropout_prob=0.0,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        image_size=224,
+        patch_size=16,
+        num_channels=3,
+        qkv_bias=True,
+        encoder_stride=16,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.qkv_bias = qkv_bias
+        self.encoder_stride = encoder_stride
+
+
+class ViTOnnxConfig(OnnxConfig):
+    torch_onnx_minimum_version = version.parse("1.11")
+
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        return OrderedDict(
+            [
+                ("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
+            ]
+        )
+
+    @property
+    def atol_for_validation(self) -> float:
+        return 1e-4

modeling_vit.py

@@ -0,0 +1,954 @@
+# 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 ViT model."""
+
+import collections.abc
+import math
+from typing import Dict, List, Optional, Set, Tuple, Union
+from functools import partial
+from enum import Flag, auto
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (
+    BaseModelOutput,
+    BaseModelOutputWithPooling,
+    ImageClassifierOutput,
+    MaskedImageModelingOutput,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.pytorch_utils import find_pruneable_heads_and_indices, prune_linear_layer
+from transformers.utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from .configuration_vit import ViTConfig
+
+
+logger = logging.get_logger(__name__)
+
+# General docstring
+_CONFIG_FOR_DOC = "ViTConfig"
+
+# Base docstring
+_CHECKPOINT_FOR_DOC = "google/vit-base-patch16-224-in21k"
+_EXPECTED_OUTPUT_SHAPE = [1, 197, 768]
+
+# Image classification docstring
+_IMAGE_CLASS_CHECKPOINT = "google/vit-base-patch16-224"
+_IMAGE_CLASS_EXPECTED_OUTPUT = "Egyptian cat"
+
+
+
+
+class BaseEnumOptions(Flag):
+    def __str__(self):
+        return self.name
+
+    @classmethod
+    def list_names(cls):
+        return [m.name for m in cls]
+class AttentionGateType(BaseEnumOptions):
+    none = 0
+    unconditional_per_head = 1
+    conditional_per_head = 2
+    conditional_per_token = 3
+
+def softmax_n_shifted_zeros(input: torch.Tensor, n: int, dim=-1) -> torch.Tensor:
+    """
+    $\text(softmax)_n(x_i) = exp(x_i) / (n + \sum_j exp(x_j))$
+
+    Note: softmax_n, with fixed input, is _not_ shift-symmetric when n != 0
+    """
+    # compute the maxes along the last dimension
+    input_maxes = input.max(dim=dim, keepdim=True).values
+    # shift the input to prevent overflow (and underflow in the denominator)
+    shifted_inputs = torch.subtract(input, input_maxes)
+    # compute the numerator and softmax_0 denominator using the shifted input
+    numerator = torch.exp(shifted_inputs)
+    original_denominator = numerator.sum(dim=dim, keepdim=True)
+    # we need to shift the zeros in the same way we shifted the inputs
+    shifted_zeros = torch.multiply(input_maxes, -1)
+    # and then add this contribution to the denominator
+    denominator = torch.add(original_denominator,
+                            torch.multiply(torch.exp(shifted_zeros), n))
+    return torch.divide(numerator, denominator)
+
+
+def softmax_1(input: torch.Tensor, dim=-1) -> torch.Tensor:
+    """
+    $\text(softmax)_n(x_i) = exp(x_i) / (1 + \sum_j exp(x_j))$
+    """
+    return softmax_n_shifted_zeros(input, 1, dim=dim)
+
+
+def clipped_softmax(data, dim=1, eta=1.1, gamma=-0.1, **kw):
+    sm_out = torch.nn.functional.softmax(data, dim=dim, **kw)
+    stretched_out = sm_out * (eta - gamma) + gamma
+    return torch.clip(stretched_out, 0, 1)
+def clipped_softmax1(data, dim=1, eta=1.1, gamma=-0.1, **kw):
+    sm_out = softmax_1(data, dim=dim, **kw)
+    stretched_out = sm_out * (eta - gamma) + gamma
+    return torch.clip(stretched_out, 0, 1)
+
+class ViTEmbeddings(nn.Module):
+    """
+    Construct the CLS token, position and patch embeddings. Optionally, also the mask token.
+    """
+
+    def __init__(self, config: ViTConfig, use_mask_token: bool = False) -> None:
+        super().__init__()
+
+        self.cls_token = nn.Parameter(torch.randn(1, 1, config.hidden_size))
+        self.mask_token = nn.Parameter(torch.zeros(1, 1, config.hidden_size)) if use_mask_token else None
+        self.patch_embeddings = ViTPatchEmbeddings(config)
+        num_patches = self.patch_embeddings.num_patches
+        self.position_embeddings = nn.Parameter(torch.randn(1, num_patches + 1, config.hidden_size))
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.config = config
+
+    def interpolate_pos_encoding(self, embeddings: torch.Tensor, height: int, width: int) -> torch.Tensor:
+        """
+        This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher
+        resolution images.
+
+        Source:
+        https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174
+        """
+
+        num_patches = embeddings.shape[1] - 1
+        num_positions = self.position_embeddings.shape[1] - 1
+        if num_patches == num_positions and height == width:
+            return self.position_embeddings
+        class_pos_embed = self.position_embeddings[:, 0]
+        patch_pos_embed = self.position_embeddings[:, 1:]
+        dim = embeddings.shape[-1]
+        h0 = height // self.config.patch_size
+        w0 = width // self.config.patch_size
+        # we add a small number to avoid floating point error in the interpolation
+        # see discussion at https://github.com/facebookresearch/dino/issues/8
+        h0, w0 = h0 + 0.1, w0 + 0.1
+        patch_pos_embed = patch_pos_embed.reshape(1, int(math.sqrt(num_positions)), int(math.sqrt(num_positions)), dim)
+        patch_pos_embed = patch_pos_embed.permute(0, 3, 1, 2)
+        patch_pos_embed = nn.functional.interpolate(
+            patch_pos_embed,
+            scale_factor=(h0 / math.sqrt(num_positions), w0 / math.sqrt(num_positions)),
+            mode="bicubic",
+            align_corners=False,
+        )
+        assert int(h0) == patch_pos_embed.shape[-2] and int(w0) == patch_pos_embed.shape[-1]
+        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)
+        return torch.cat((class_pos_embed.unsqueeze(0), patch_pos_embed), dim=1)
+
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        bool_masked_pos: Optional[torch.BoolTensor] = None,
+        interpolate_pos_encoding: bool = False,
+    ) -> torch.Tensor:
+        batch_size, num_channels, height, width = pixel_values.shape
+        embeddings = self.patch_embeddings(pixel_values, interpolate_pos_encoding=interpolate_pos_encoding)
+
+        if bool_masked_pos is not None:
+            seq_length = embeddings.shape[1]
+            mask_tokens = self.mask_token.expand(batch_size, seq_length, -1)
+            # replace the masked visual tokens by mask_tokens
+            mask = bool_masked_pos.unsqueeze(-1).type_as(mask_tokens)
+            embeddings = embeddings * (1.0 - mask) + mask_tokens * mask
+
+        # add the [CLS] token to the embedded patch tokens
+        cls_tokens = self.cls_token.expand(batch_size, -1, -1)
+        embeddings = torch.cat((cls_tokens, embeddings), dim=1)
+
+        # add positional encoding to each token
+        if interpolate_pos_encoding:
+            embeddings = embeddings + self.interpolate_pos_encoding(embeddings, height, width)
+        else:
+            embeddings = embeddings + self.position_embeddings
+
+        embeddings = self.dropout(embeddings)
+
+        return embeddings
+
+
+class ViTPatchEmbeddings(nn.Module):
+    """
+    This class turns `pixel_values` of shape `(batch_size, num_channels, height, width)` into the initial
+    `hidden_states` (patch embeddings) of shape `(batch_size, seq_length, hidden_size)` to be consumed by a
+    Transformer.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        image_size, patch_size = config.image_size, config.patch_size
+        num_channels, hidden_size = config.num_channels, config.hidden_size
+
+        image_size = image_size if isinstance(image_size, collections.abc.Iterable) else (image_size, image_size)
+        patch_size = patch_size if isinstance(patch_size, collections.abc.Iterable) else (patch_size, patch_size)
+        num_patches = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.num_patches = num_patches
+
+        self.projection = nn.Conv2d(num_channels, hidden_size, kernel_size=patch_size, stride=patch_size)
+
+    def forward(self, pixel_values: torch.Tensor, interpolate_pos_encoding: bool = False) -> torch.Tensor:
+        batch_size, num_channels, height, width = pixel_values.shape
+        if num_channels != self.num_channels:
+            raise ValueError(
+                "Make sure that the channel dimension of the pixel values match with the one set in the configuration."
+                f" Expected {self.num_channels} but got {num_channels}."
+            )
+        if not interpolate_pos_encoding:
+            if height != self.image_size[0] or width != self.image_size[1]:
+                raise ValueError(
+                    f"Input image size ({height}*{width}) doesn't match model"
+                    f" ({self.image_size[0]}*{self.image_size[1]})."
+                )
+        embeddings = self.projection(pixel_values).flatten(2).transpose(1, 2)
+        return embeddings
+
+
+class ViTSelfAttention(nn.Module):
+    def __init__(self, config: ViTConfig) -> 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.softmax_fn = softmax_1
+        self.query = nn.Linear(config.hidden_size, self.all_head_size, bias=config.qkv_bias)
+        self.key = nn.Linear(config.hidden_size, self.all_head_size, bias=config.qkv_bias)
+        self.value = nn.Linear(config.hidden_size, self.all_head_size, bias=config.qkv_bias)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+
+    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
+        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: Optional[torch.Tensor] = None, output_attentions: bool = False
+    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
+        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)
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = self.softmax_fn(attention_scores, dim=-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)
+
+        # 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
+
+def scaled_dot_product_attention(query, key, value, softmax_fn, attn_mask=None, dropout_p=0.0, is_causal=False, scale=None) -> torch.Tensor:
+    # Efficient implementation equivalent to the following:
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    L, S = query.size(-2), key.size(-2)
+    scale_factor = 1 / math.sqrt(query.size(-1)) if scale is None else scale
+    attn_bias = torch.zeros(L, S, dtype=query.dtype, device=query.device)
+    if is_causal:
+        assert attn_mask is None
+        temp_mask = torch.ones(L, S, dtype=torch.bool).tril(diagonal=0)
+        attn_bias.masked_fill_(temp_mask.logical_not(), float("-inf"))
+        attn_bias.to(query.dtype)
+
+    if attn_mask is not None:
+        if attn_mask.dtype == torch.bool:
+            attn_mask.masked_fill_(attn_mask.logical_not(), float("-inf"))
+        else:
+            attn_bias += attn_mask
+    attn_weight = query @ key.transpose(-2, -1) * scale_factor
+    attn_weight += attn_bias
+    attn_weight = softmax_fn(attn_weight, dim=-1)
+    attn_weight = torch.dropout(attn_weight, dropout_p, train=True)
+    return attn_weight @ value
+
+class ViTSdpaSelfAttention(ViTSelfAttention):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__(config)
+        self.attention_probs_dropout_prob = config.attention_probs_dropout_prob
+
+    def forward(
+        self, hidden_states, head_mask: Optional[torch.Tensor] = None, output_attentions: bool = False
+    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
+        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)
+
+        context_layer = scaled_dot_product_attention(
+            query_layer,
+            key_layer,
+            value_layer,
+            dropout_p=self.attention_probs_dropout_prob if self.training else 0.0,
+            attn_mask=head_mask,
+            softmax_fn = self.softmax_fn,
+            is_causal=False,
+            scale=None,
+        )
+
+        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)
+
+        return context_layer, None
+
+
+class ViTSelfOutput(nn.Module):
+    """
+    The residual connection is defined in ViTLayer instead of here (as is the case with other models), due to the
+    layernorm applied before each block.
+    """
+
+    def __init__(self, config: ViTConfig) -> None:
+        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: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+
+        return hidden_states
+
+
+class ViTAttention(nn.Module):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__()
+        self.attention = ViTSelfAttention(config)
+        self.output = ViTSelfOutput(config)
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads: Set[int]) -> None:
+        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: torch.Tensor,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
+        self_outputs = self.attention(hidden_states, head_mask, output_attentions)
+
+        attention_output = self.output(self_outputs[0], hidden_states)
+
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class ViTSdpaAttention(ViTAttention):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__(config)
+        self.attention = ViTSdpaSelfAttention(config)
+
+
+class ViTIntermediate(nn.Module):
+    def __init__(self, config: ViTConfig) -> None:
+        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: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+
+        return hidden_states
+
+
+class ViTOutput(nn.Module):
+    def __init__(self, config: ViTConfig) -> None:
+        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: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+
+        hidden_states = hidden_states + input_tensor
+
+        return hidden_states
+
+
+VIT_ATTENTION_CLASSES = {
+    "eager": ViTAttention,
+    "sdpa": ViTSdpaAttention,
+}
+
+
+class ViTLayer(nn.Module):
+    """This corresponds to the Block class in the timm implementation."""
+
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = VIT_ATTENTION_CLASSES[config._attn_implementation](config)
+        self.intermediate = ViTIntermediate(config)
+        self.output = ViTOutput(config)
+        self.layernorm_before = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.layernorm_after = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
+        self_attention_outputs = self.attention(
+            self.layernorm_before(hidden_states),  # in ViT, layernorm is applied before self-attention
+            head_mask,
+            output_attentions=output_attentions,
+        )
+        attention_output = self_attention_outputs[0]
+        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
+
+        # first residual connection
+        hidden_states = attention_output + hidden_states
+
+        # in ViT, layernorm is also applied after self-attention
+        layer_output = self.layernorm_after(hidden_states)
+        layer_output = self.intermediate(layer_output)
+
+        # second residual connection is done here
+        layer_output = self.output(layer_output, hidden_states)
+
+        outputs = (layer_output,) + outputs
+
+        return outputs
+
+
+class ViTEncoder(nn.Module):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([ViTLayer(config) for _ in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+        output_hidden_states: bool = False,
+        return_dict: bool = True,
+    ) -> Union[tuple, BaseModelOutput]:
+        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:
+                layer_outputs = self._gradient_checkpointing_func(
+                    layer_module.__call__,
+                    hidden_states,
+                    layer_head_mask,
+                    output_attentions,
+                )
+            else:
+                layer_outputs = layer_module(hidden_states, layer_head_mask, output_attentions)
+
+            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 ViTPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = ViTConfig
+    base_model_prefix = "vit"
+    main_input_name = "pixel_values"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["ViTEmbeddings", "ViTLayer"]
+    _supports_sdpa = True
+
+    def _init_weights(self, module: Union[nn.Linear, nn.Conv2d, nn.LayerNorm]) -> None:
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            # Upcast the input in `fp32` and cast it back to desired `dtype` to avoid
+            # `trunc_normal_cpu` not implemented in `half` issues
+            module.weight.data = nn.init.trunc_normal_(
+                module.weight.data.to(torch.float32), mean=0.0, std=self.config.initializer_range
+            ).to(module.weight.dtype)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        elif isinstance(module, ViTEmbeddings):
+            module.position_embeddings.data = nn.init.trunc_normal_(
+                module.position_embeddings.data.to(torch.float32),
+                mean=0.0,
+                std=self.config.initializer_range,
+            ).to(module.position_embeddings.dtype)
+
+            module.cls_token.data = nn.init.trunc_normal_(
+                module.cls_token.data.to(torch.float32),
+                mean=0.0,
+                std=self.config.initializer_range,
+            ).to(module.cls_token.dtype)
+
+
+VIT_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 ([`ViTConfig`]): 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 [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+VIT_INPUTS_DOCSTRING = r"""
+    Args:
+        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
+            Pixel values. Pixel values can be obtained using [`AutoImageProcessor`]. See [`ViTImageProcessor.__call__`]
+            for details.
+
+        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]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        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.
+        interpolate_pos_encoding (`bool`, *optional*):
+            Whether to interpolate the pre-trained position encodings.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare ViT Model transformer outputting raw hidden-states without any specific head on top.",
+    VIT_START_DOCSTRING,
+)
+class ViTModel(ViTPreTrainedModel):
+    def __init__(self, config: ViTConfig, add_pooling_layer: bool = True, use_mask_token: bool = False):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = ViTEmbeddings(config, use_mask_token=use_mask_token)
+        self.encoder = ViTEncoder(config)
+
+        self.layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.pooler = ViTPooler(config) if add_pooling_layer else None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self) -> ViTPatchEmbeddings:
+        return self.embeddings.patch_embeddings
+
+    def _prune_heads(self, heads_to_prune: Dict[int, List[int]]) -> None:
+        """
+        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)
+
+    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPooling,
+        config_class=_CONFIG_FOR_DOC,
+        modality="vision",
+        expected_output=_EXPECTED_OUTPUT_SHAPE,
+    )
+    def forward(
+        self,
+        pixel_values: Optional[torch.Tensor] = None,
+        bool_masked_pos: Optional[torch.BoolTensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        interpolate_pos_encoding: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPooling]:
+        r"""
+        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, num_patches)`, *optional*):
+            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
+        """
+        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)
+
+        # TODO: maybe have a cleaner way to cast the input (from `ImageProcessor` side?)
+        expected_dtype = self.embeddings.patch_embeddings.projection.weight.dtype
+        if pixel_values.dtype != expected_dtype:
+            pixel_values = pixel_values.to(expected_dtype)
+
+        embedding_output = self.embeddings(
+            pixel_values, bool_masked_pos=bool_masked_pos, interpolate_pos_encoding=interpolate_pos_encoding
+        )
+
+        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:
+            head_outputs = (sequence_output, pooled_output) if pooled_output is not None else (sequence_output,)
+            return head_outputs + 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 ViTPooler(nn.Module):
+    def __init__(self, config: ViTConfig):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    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)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+@add_start_docstrings(
+    """ViT Model with a decoder on top for masked image modeling, as proposed in [SimMIM](https://arxiv.org/abs/2111.09886).
+
+    <Tip>
+
+    Note that we provide a script to pre-train this model on custom data in our [examples
+    directory](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-pretraining).
+
+    </Tip>
+    """,
+    VIT_START_DOCSTRING,
+)
+class ViTForMaskedImageModeling(ViTPreTrainedModel):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__(config)
+
+        self.vit = ViTModel(config, add_pooling_layer=False, use_mask_token=True)
+
+        self.decoder = nn.Sequential(
+            nn.Conv2d(
+                in_channels=config.hidden_size,
+                out_channels=config.encoder_stride**2 * config.num_channels,
+                kernel_size=1,
+            ),
+            nn.PixelShuffle(config.encoder_stride),
+        )
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=MaskedImageModelingOutput, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        pixel_values: Optional[torch.Tensor] = None,
+        bool_masked_pos: Optional[torch.BoolTensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        interpolate_pos_encoding: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[tuple, MaskedImageModelingOutput]:
+        r"""
+        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, num_patches)`):
+            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
+
+        Returns:
+
+        Examples:
+        ```python
+        >>> from transformers import AutoImageProcessor, ViTForMaskedImageModeling
+        >>> import torch
+        >>> from PIL import Image
+        >>> import requests
+
+        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+        >>> image = Image.open(requests.get(url, stream=True).raw)
+
+        >>> image_processor = AutoImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
+        >>> model = ViTForMaskedImageModeling.from_pretrained("google/vit-base-patch16-224-in21k")
+
+        >>> num_patches = (model.config.image_size // model.config.patch_size) ** 2
+        >>> pixel_values = image_processor(images=image, return_tensors="pt").pixel_values
+        >>> # create random boolean mask of shape (batch_size, num_patches)
+        >>> bool_masked_pos = torch.randint(low=0, high=2, size=(1, num_patches)).bool()
+
+        >>> outputs = model(pixel_values, bool_masked_pos=bool_masked_pos)
+        >>> loss, reconstructed_pixel_values = outputs.loss, outputs.reconstruction
+        >>> list(reconstructed_pixel_values.shape)
+        [1, 3, 224, 224]
+        ```"""
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if bool_masked_pos is not None and (self.config.patch_size != self.config.encoder_stride):
+            raise ValueError(
+                "When `bool_masked_pos` is provided, `patch_size` must be equal to `encoder_stride` to ensure that "
+                "the reconstructed image has the same dimensions as the input. "
+                f"Got `patch_size` = {self.config.patch_size} and `encoder_stride` = {self.config.encoder_stride}."
+            )
+
+        outputs = self.vit(
+            pixel_values,
+            bool_masked_pos=bool_masked_pos,
+            head_mask=head_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            interpolate_pos_encoding=interpolate_pos_encoding,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        # Reshape to (batch_size, num_channels, height, width)
+        sequence_output = sequence_output[:, 1:]
+        batch_size, sequence_length, num_channels = sequence_output.shape
+        height = width = math.floor(sequence_length**0.5)
+        sequence_output = sequence_output.permute(0, 2, 1).reshape(batch_size, num_channels, height, width)
+
+        # Reconstruct pixel values
+        reconstructed_pixel_values = self.decoder(sequence_output)
+
+        masked_im_loss = None
+        if bool_masked_pos is not None:
+            size = self.config.image_size // self.config.patch_size
+            bool_masked_pos = bool_masked_pos.reshape(-1, size, size)
+            mask = (
+                bool_masked_pos.repeat_interleave(self.config.patch_size, 1)
+                .repeat_interleave(self.config.patch_size, 2)
+                .unsqueeze(1)
+                .contiguous()
+            )
+            reconstruction_loss = nn.functional.l1_loss(pixel_values, reconstructed_pixel_values, reduction="none")
+            masked_im_loss = (reconstruction_loss * mask).sum() / (mask.sum() + 1e-5) / self.config.num_channels
+
+        if not return_dict:
+            output = (reconstructed_pixel_values,) + outputs[1:]
+            return ((masked_im_loss,) + output) if masked_im_loss is not None else output
+
+        return MaskedImageModelingOutput(
+            loss=masked_im_loss,
+            reconstruction=reconstructed_pixel_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    ViT Model transformer with an image classification head on top (a linear layer on top of the final hidden state of
+    the [CLS] token) e.g. for ImageNet.
+
+    <Tip>
+
+        Note that it's possible to fine-tune ViT on higher resolution images than the ones it has been trained on, by
+        setting `interpolate_pos_encoding` to `True` in the forward of the model. This will interpolate the pre-trained
+        position embeddings to the higher resolution.
+
+    </Tip>
+    """,
+    VIT_START_DOCSTRING,
+)
+class ViTForImageClassification(ViTPreTrainedModel):
+    def __init__(self, config: ViTConfig) -> None:
+        super().__init__(config)
+
+        self.num_labels = config.num_labels
+        self.vit = ViTModel(config, add_pooling_layer=False)
+
+        # Classifier head
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity()
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_IMAGE_CLASS_CHECKPOINT,
+        output_type=ImageClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+        expected_output=_IMAGE_CLASS_EXPECTED_OUTPUT,
+    )
+    def forward(
+        self,
+        pixel_values: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        interpolate_pos_encoding: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[tuple, ImageClassifierOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the image 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).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.vit(
+            pixel_values,
+            head_mask=head_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            interpolate_pos_encoding=interpolate_pos_encoding,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        logits = self.classifier(sequence_output[:, 0, :])
+
+        loss = None
+        if labels is not None:
+            # move labels to correct device to enable model parallelism
+            labels = labels.to(logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return ImageClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )