AutoAndroidController/py/venv/Lib/site-packages/transformers/models/git/modeling_git.py

# coding=utf-8
# Copyright 2022 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.
"""PyTorch GIT model."""

import math
from dataclasses import dataclass
from typing import Callable, Optional, Union

import torch
from torch import nn

from ...activations import ACT2FN
from ...cache_utils import Cache, DynamicCache
from ...generation import GenerationMixin
from ...modeling_attn_mask_utils import _prepare_4d_attention_mask
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPast,
    BaseModelOutputWithPooling,
    CausalLMOutputWithPast,
)
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import (
    ModelOutput,
    auto_docstring,
    can_return_tuple,
    logging,
    torch_int,
)
from ...utils.deprecation import deprecate_kwarg
from .configuration_git import GitConfig, GitVisionConfig


logger = logging.get_logger(__name__)


@dataclass
@auto_docstring(
    custom_intro="""
    Base class for vision model's outputs that also contains image embeddings of the pooling of the last hidden states.
    """
)
# Copied from transformers.models.clip.modeling_clip.CLIPVisionModelOutput with CLIP->Git
class GitVisionModelOutput(ModelOutput):
    r"""
    image_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
        The image embeddings obtained by applying the projection layer to the pooler_output.
    """

    image_embeds: Optional[torch.FloatTensor] = None
    last_hidden_state: Optional[torch.FloatTensor] = None
    hidden_states: Optional[tuple[torch.FloatTensor, ...]] = None
    attentions: Optional[tuple[torch.FloatTensor, ...]] = None


class GitEmbeddings(nn.Module):
    """Construct the embeddings from word and position embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
        self.register_buffer(
            "position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)), persistent=False
        )

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        past_key_values_length: int = 0,
    ) -> torch.Tensor:
        if input_ids is not None:
            input_shape = input_ids.size()
        else:
            input_shape = inputs_embeds.size()[:-1]

        seq_length = input_shape[1]

        if position_ids is None:
            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]

        if inputs_embeds is None:
            embeddings = self.word_embeddings(input_ids)
        else:
            embeddings = inputs_embeds

        if self.position_embedding_type == "absolute":
            position_embeddings = self.position_embeddings(position_ids)
            embeddings += position_embeddings
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class GitSelfAttention(nn.Module):
    def __init__(self, config, position_embedding_type=None, layer_idx=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.layer_idx = layer_idx
        if layer_idx is None:
            logger.warning_once(
                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
                "when creating this class."
            )

        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.image_patch_tokens = int((config.vision_config.image_size / config.vision_config.patch_size) ** 2 + 1)
        if config.num_image_with_embedding is not None:
            self.image_patch_tokens *= config.num_image_with_embedding

        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)
        self.position_embedding_type = position_embedding_type or getattr(
            config, "position_embedding_type", "absolute"
        )
        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            self.max_position_embeddings = config.max_position_embeddings
            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        pixel_values_present: Optional[bool] = False,
    ) -> tuple[torch.Tensor]:
        batch_size, seq_length, _ = hidden_states.shape
        query_layer = (
            self.query(hidden_states)
            .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
            .transpose(1, 2)
        )

        cutoff = self.image_patch_tokens if pixel_values_present else 0
        key_layer = (
            self.key(hidden_states)
            .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
            .transpose(1, 2)
        )
        value_layer = (
            self.value(hidden_states)
            .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
            .transpose(1, 2)
        )
        if past_key_values is not None:
            # NOTE: like in other caches, we store the text component. In GIT it means we discard the image component.
            key_layer_past, value_layer_past = past_key_values.update(
                key_layer[:, :, cutoff:, :], value_layer[:, :, cutoff:, :], self.layer_idx
            )
            key_layer = torch.cat([key_layer[:, :, :cutoff, :], key_layer_past], dim=2)
            value_layer = torch.cat([value_layer[:, :, :cutoff, :], value_layer_past], dim=2)

        # 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))

        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            query_length, key_length = query_layer.shape[2], key_layer.shape[2]
            if past_key_values is not None:
                position_ids_l = torch.tensor(key_length - 1, dtype=torch.long, device=hidden_states.device).view(
                    -1, 1
                )
            else:
                position_ids_l = torch.arange(query_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
            position_ids_r = torch.arange(key_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
            distance = position_ids_l - position_ids_r

            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility

            if self.position_embedding_type == "relative_key":
                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores
            elif self.position_embedding_type == "relative_key_query":
                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in GitModel forward() function)
            attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.functional.softmax(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)

        return context_layer, attention_probs


# Copied from transformers.models.bert.modeling_bert.BertSelfOutput
class GitSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        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 = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


GIT_SELF_ATTENTION_CLASSES = {
    "eager": GitSelfAttention,
}


class GitAttention(nn.Module):
    def __init__(self, config, position_embedding_type=None, layer_idx=None):
        super().__init__()
        self.self = GIT_SELF_ATTENTION_CLASSES[config._attn_implementation](
            config, position_embedding_type=position_embedding_type, layer_idx=layer_idx
        )
        self.output = GitSelfOutput(config)
        self.pruned_heads = set()

    # Copied from transformers.models.bert.modeling_bert.BertAttention.prune_heads
    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
        )

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

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

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        pixel_values_present: Optional[bool] = False,
    ) -> tuple[torch.Tensor]:
        attn_output, self_attn_weights = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            past_key_values,
            output_attentions,
            pixel_values_present,
        )
        attention_output = self.output(attn_output, hidden_states)
        return attention_output, self_attn_weights


# Copied from transformers.models.bert.modeling_bert.BertIntermediate
class GitIntermediate(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: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


# Copied from transformers.models.bert.modeling_bert.BertOutput
class GitOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        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 = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class GitLayer(GradientCheckpointingLayer):
    def __init__(self, config, layer_idx=None):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = GitAttention(config, layer_idx=layer_idx)
        self.intermediate = GitIntermediate(config)
        self.output = GitOutput(config)

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        pixel_values_present: Optional[bool] = False,
    ) -> tuple[torch.Tensor]:
        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
        attention_output, self_attention_weights = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions=output_attentions,
            past_key_values=past_key_values,
            pixel_values_present=pixel_values_present,
        )

        layer_output = apply_chunking_to_forward(
            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
        )
        return layer_output, self_attention_weights

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


class GitEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList([GitLayer(config, i) for i in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Union[Cache, tuple[tuple[torch.FloatTensor]]]] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = False,
        output_hidden_states: Optional[bool] = False,
        pixel_values_present: Optional[bool] = False,
        return_dict: Optional[bool] = True,
    ) -> Union[tuple[torch.Tensor], BaseModelOutputWithPast]:
        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        if use_cache and past_key_values is None:
            past_key_values = DynamicCache(config=self.config)

        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

            layer_outputs = layer_module(
                hidden_states,
                attention_mask,
                layer_head_mask,
                past_key_values,
                output_attentions,
                pixel_values_present,
            )

            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,
                    past_key_values,
                    all_hidden_states,
                    all_self_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


@auto_docstring
class GitPreTrainedModel(PreTrainedModel):
    config: GitConfig
    base_model_prefix = "git"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, GitVisionEmbeddings):
            nn.init.normal_(module.class_embedding, mean=0.0, std=self.config.initializer_range)
            nn.init.normal_(module.patch_embedding.weight, std=self.config.initializer_range)
            nn.init.normal_(module.position_embedding.weight, std=self.config.initializer_range)
        if isinstance(module, nn.Linear):
            # 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)


# Copied from transformers.models.clip.modeling_clip.CLIPVisionEmbeddings with CLIP->Git
class GitVisionEmbeddings(nn.Module):
    def __init__(self, config: GitVisionConfig):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.image_size = config.image_size
        self.patch_size = config.patch_size

        self.class_embedding = nn.Parameter(torch.randn(self.embed_dim))

        self.patch_embedding = nn.Conv2d(
            in_channels=config.num_channels,
            out_channels=self.embed_dim,
            kernel_size=self.patch_size,
            stride=self.patch_size,
            bias=False,
        )

        self.num_patches = (self.image_size // self.patch_size) ** 2
        self.num_positions = self.num_patches + 1
        self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
        self.register_buffer("position_ids", torch.arange(self.num_positions).expand((1, -1)), persistent=False)

    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. This method is also adapted to support torch.jit tracing.

        Adapted from:
        - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and
        - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211
        """

        num_patches = embeddings.shape[1] - 1
        position_embedding = self.position_embedding.weight.unsqueeze(0)
        num_positions = position_embedding.shape[1] - 1

        # always interpolate when tracing to ensure the exported model works for dynamic input shapes
        if not torch.jit.is_tracing() and num_patches == num_positions and height == width:
            return self.position_embedding(self.position_ids)

        class_pos_embed = position_embedding[:, :1]
        patch_pos_embed = position_embedding[:, 1:]

        dim = embeddings.shape[-1]

        new_height = height // self.patch_size
        new_width = width // self.patch_size

        sqrt_num_positions = torch_int(num_positions**0.5)
        patch_pos_embed = patch_pos_embed.reshape(1, sqrt_num_positions, sqrt_num_positions, dim)
        patch_pos_embed = patch_pos_embed.permute(0, 3, 1, 2)

        patch_pos_embed = nn.functional.interpolate(
            patch_pos_embed,
            size=(new_height, new_width),
            mode="bicubic",
            align_corners=False,
        )

        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)

        return torch.cat((class_pos_embed, patch_pos_embed), dim=1)

    def forward(self, pixel_values: torch.FloatTensor, interpolate_pos_encoding=False) -> torch.Tensor:
        batch_size, _, height, width = pixel_values.shape
        if not interpolate_pos_encoding and (height != self.image_size or width != self.image_size):
            raise ValueError(
                f"Input image size ({height}*{width}) doesn't match model ({self.image_size}*{self.image_size})."
            )
        target_dtype = self.patch_embedding.weight.dtype
        patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))  # shape = [*, width, grid, grid]
        patch_embeds = patch_embeds.flatten(2).transpose(1, 2)

        class_embeds = self.class_embedding.expand(batch_size, 1, -1)
        embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
        if interpolate_pos_encoding:
            embeddings = embeddings + self.interpolate_pos_encoding(embeddings, height, width)
        else:
            embeddings = embeddings + self.position_embedding(self.position_ids)
        return embeddings


class GitVisionMLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.activation_fn = ACT2FN[config.hidden_act]
        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.fc1(hidden_states)
        hidden_states = self.activation_fn(hidden_states)
        hidden_states = self.fc2(hidden_states)
        return hidden_states


# Copied from transformers.models.siglip.modeling_siglip.eager_attention_forward
def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: Optional[torch.Tensor],
    scaling: float,
    dropout: float = 0.0,
    **kwargs,
):
    attn_weights = torch.matmul(query, key.transpose(-1, -2)) * scaling
    if attention_mask is not None:
        attn_weights = attn_weights + attention_mask

    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)

    attn_output = torch.matmul(attn_weights, value)
    attn_output = attn_output.transpose(1, 2).contiguous()

    return attn_output, attn_weights


class GitVisionAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(self, config):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.embed_dim // self.num_heads
        if self.head_dim * self.num_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
                f" {self.num_heads})."
            )
        self.scale = self.head_dim**-0.5
        self.dropout = config.attention_dropout
        self.is_causal = False

        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        causal_attention_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = False,
    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
        """Input shape: Batch x Time x Channel"""

        batch_size, seq_length, embed_dim = hidden_states.shape

        queries = self.q_proj(hidden_states)
        keys = self.k_proj(hidden_states)
        values = self.v_proj(hidden_states)

        queries = queries.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)
        keys = keys.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)
        values = values.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)
        # CLIP text model uses both `causal_attention_mask` and `attention_mask`
        # in case FA2 kernel is called, `is_causal` should be inferred from `causal_attention_mask`
        if self.config._attn_implementation != "flash_attention_2":
            if attention_mask is not None and causal_attention_mask is not None:
                attention_mask = attention_mask + causal_attention_mask
            elif causal_attention_mask is not None:
                attention_mask = causal_attention_mask
        else:
            self.is_causal = causal_attention_mask is not None

        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        attn_output, attn_weights = attention_interface(
            self,
            queries,
            keys,
            values,
            attention_mask,
            is_causal=self.is_causal,
            scaling=self.scale,
            dropout=0.0 if not self.training else self.dropout,
        )

        attn_output = attn_output.reshape(batch_size, seq_length, embed_dim).contiguous()
        attn_output = self.out_proj(attn_output)
        if not output_attentions:
            attn_weights = None
        return attn_output, attn_weights


# Copied from transformers.models.altclip.modeling_altclip.AltCLIPEncoderLayer with AltCLIP->GitVision
class GitVisionEncoderLayer(GradientCheckpointingLayer):
    def __init__(self, config: GitVisionConfig):
        super().__init__()
        self.embed_dim = config.hidden_size
        self.self_attn = GitVisionAttention(config)
        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
        self.mlp = GitVisionMLP(config)
        self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: torch.Tensor,
        causal_attention_mask: torch.Tensor,
        output_attentions: Optional[bool] = False,
    ) -> tuple[torch.FloatTensor]:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`): attention mask of size
                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
                `(config.encoder_attention_heads,)`.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
        """
        residual = hidden_states

        hidden_states = self.layer_norm1(hidden_states)
        hidden_states, attn_weights = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            causal_attention_mask=causal_attention_mask,
            output_attentions=output_attentions,
        )
        hidden_states = residual + hidden_states

        residual = hidden_states
        hidden_states = self.layer_norm2(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (attn_weights,)

        return outputs


# Copied from transformers.models.altclip.modeling_altclip.AltCLIPEncoder with AltCLIP->GitVision, CLIPConfig
class GitVisionEncoder(nn.Module):
    """
    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
    [`GitVisionEncoderLayer`].

    Args:
        config: GitVisionConfig
    """

    def __init__(self, config: GitVisionConfig):
        super().__init__()
        self.config = config
        self.layers = nn.ModuleList([GitVisionEncoderLayer(config) for _ in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    @can_return_tuple
    def forward(
        self,
        inputs_embeds,
        attention_mask: Optional[torch.Tensor] = None,
        causal_attention_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[tuple, BaseModelOutput]:
        r"""
        Args:
            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
                than the model's internal embedding lookup matrix.
            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)
            causal_attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Causal mask for the text model. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            output_hidden_states (`bool`, *optional*):
                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
                for more detail.
            return_dict (`bool`, *optional*):
                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        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

        encoder_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None

        hidden_states = inputs_embeds
        for idx, encoder_layer in enumerate(self.layers):
            if output_hidden_states:
                encoder_states = encoder_states + (hidden_states,)
            layer_outputs = encoder_layer(
                hidden_states,
                attention_mask,
                causal_attention_mask,
                output_attentions=output_attentions,
            )

            hidden_states = layer_outputs[0]

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

        if output_hidden_states:
            encoder_states = encoder_states + (hidden_states,)

        return BaseModelOutput(
            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
        )


class GitVisionTransformer(nn.Module):
    # Copied from transformers.models.altclip.modeling_altclip.AltCLIPVisionTransformer.__init__ with AltCLIPEncoder->GitVisionEncoder, AltCLIP->Git
    def __init__(self, config: GitVisionConfig):
        super().__init__()
        self.config = config
        embed_dim = config.hidden_size

        self.embeddings = GitVisionEmbeddings(config)
        self.pre_layrnorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)
        self.encoder = GitVisionEncoder(config)
        self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)

    @auto_docstring
    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: Optional[bool] = False,
        return_dict: Optional[bool] = None,
    ) -> Union[tuple, BaseModelOutput]:
        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")

        hidden_states = self.embeddings(pixel_values, interpolate_pos_encoding=interpolate_pos_encoding)
        hidden_states = self.pre_layrnorm(hidden_states)

        encoder_outputs = self.encoder(
            inputs_embeds=hidden_states,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        last_hidden_state = encoder_outputs[0]

        last_hidden_state = self.post_layernorm(last_hidden_state)

        if not return_dict:
            return (last_hidden_state,) + encoder_outputs[1:]

        return BaseModelOutput(
            last_hidden_state=last_hidden_state,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


@auto_docstring(
    custom_intro="""
    The vision model from CLIP, used in GIT, without any head or projection on top.
    """
)
class GitVisionModel(GitPreTrainedModel):
    config: GitVisionConfig
    main_input_name = "pixel_values"

    # Copied from transformers.models.clip.modeling_clip.CLIPVisionModel.__init__ with CLIP->Git
    def __init__(self, config: GitVisionConfig):
        super().__init__(config)
        self.vision_model = GitVisionTransformer(config)
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self) -> nn.Module:
        return self.vision_model.embeddings.patch_embedding

    @auto_docstring
    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
        return_dict: Optional[bool] = None,
    ) -> Union[tuple, BaseModelOutput]:
        r"""
        Examples:

        ```python
        >>> from PIL import Image
        >>> import requests
        >>> from transformers import AutoProcessor, GitVisionModel

        >>> processor = AutoProcessor.from_pretrained("microsoft/git-base")
        >>> model = GitVisionModel.from_pretrained("microsoft/git-base")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> inputs = processor(images=image, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> last_hidden_state = outputs.last_hidden_state
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        return self.vision_model(
            pixel_values=pixel_values,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            interpolate_pos_encoding=interpolate_pos_encoding,
            return_dict=return_dict,
        )


class GitProjection(nn.Module):
    def __init__(self, config: GitConfig):
        super().__init__()
        self.config = config
        self.visual_projection = nn.Sequential(
            nn.Linear(config.vision_config.hidden_size, config.hidden_size),
            nn.LayerNorm(config.hidden_size, eps=config.vision_config.layer_norm_eps),
        )

    def forward(self, embeddings: torch.Tensor) -> torch.Tensor:
        return self.visual_projection(embeddings)


@auto_docstring(
    custom_intro="""
    The bare GIT Model transformer consisting of a CLIP image encoder and text decoder outputting raw hidden-states
    """
)
class GitModel(GitPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config

        self.embeddings = GitEmbeddings(config)
        self.image_encoder = GitVisionModel(config.vision_config)
        self.encoder = GitEncoder(config)

        self.visual_projection = GitProjection(config)

        if config.num_image_with_embedding is not None:
            self.img_temporal_embedding = nn.ParameterList(
                nn.Parameter(torch.zeros(1, 1, config.vision_config.hidden_size))
                for _ in range(config.num_image_with_embedding)
            )

        # Initialize weights and apply final processing
        self.post_init()

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

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    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)

    def _generate_future_mask(self, size: int, dtype: torch.dtype, device: torch.device) -> torch.Tensor:
        # Default mask is for forward direction. Flip for backward direction.
        mask = torch.triu(torch.ones(size, size, device=device, dtype=dtype), diagonal=1)
        mask = mask.masked_fill(mask == 1, float("-inf"))
        return mask

    def create_attention_mask(self, tgt, memory, tgt_mask, past_key_values_length, memory_key_padding_mask=None):
        num_tgt = tgt.shape[1]
        num_memory = memory.shape[1]
        device = tgt.device
        dtype = tgt.dtype
        top_left = torch.zeros((num_memory, num_memory), device=device, dtype=dtype)
        top_right = torch.full(
            (num_memory, num_tgt + past_key_values_length),
            float("-inf"),
            device=tgt.device,
            dtype=dtype,
        )
        bottom_left = torch.zeros(
            (num_tgt, num_memory),
            dtype=dtype,
            device=tgt_mask.device,
        )

        if past_key_values_length > 0:
            tgt_mask = torch.zeros(
                (tgt_mask.shape[0], tgt_mask.shape[0] + past_key_values_length),
                dtype=dtype,
                device=tgt_mask.device,
            )

        left = torch.cat((top_left, bottom_left), dim=0)
        right = torch.cat((top_right, tgt_mask.to(dtype)), dim=0)

        full_attention_mask = torch.cat((left, right), dim=1)[None, :]

        if memory_key_padding_mask is None:
            memory_key_padding_mask = torch.full((memory.shape[0], memory.shape[1]), fill_value=False, device=device)
        # if it is False, it means valid. That is, it is not a padding
        if memory_key_padding_mask.dtype != torch.bool:
            raise ValueError("Memory key padding mask must be a boolean tensor.")
        zero_negative_infinity = torch.zeros_like(memory_key_padding_mask, dtype=tgt.dtype)
        zero_negative_infinity[memory_key_padding_mask] = float("-inf")
        full_attention_mask = full_attention_mask.expand(
            (memory_key_padding_mask.shape[0], num_memory + num_tgt, num_memory + past_key_values_length + num_tgt)
        )
        full_attention_mask = full_attention_mask.clone()
        origin_left = full_attention_mask[:, :, :num_memory]
        update = zero_negative_infinity[:, None, :]
        full_attention_mask[:, :, :num_memory] = origin_left + update

        # add axis for multi-head
        full_attention_mask = full_attention_mask[:, None, :, :]

        return full_attention_mask

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        pixel_values: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        past_key_values: Optional[Union[Cache, list[torch.FloatTensor]]] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
        return_dict: Optional[bool] = None,
    ) -> Union[tuple[torch.Tensor], BaseModelOutputWithPooling]:
        r"""
        Examples:

        ```python
        >>> from transformers import AutoProcessor, AutoModel
        >>> import requests
        >>> from PIL import Image

        >>> processor = AutoProcessor.from_pretrained("microsoft/git-base")
        >>> model = AutoModel.from_pretrained("microsoft/git-base")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> text = "this is an image of two cats"

        >>> inputs = processor(images=image, text=text, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> last_hidden_state = 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
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
            input_shape = input_ids.size()
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        seq_length = input_shape[1]

        # past_key_values_length
        past_key_values_length = 0
        if past_key_values is not None:
            past_key_values_length = (
                past_key_values.get_seq_length()
                if not isinstance(past_key_values, Cache)
                else past_key_values.get_seq_length()
            )

        # 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)

        projected_visual_features = None
        if pixel_values is not None:
            if pixel_values.ndim == 4:
                # here we assume pixel_values is of shape (batch_size, num_channels, height, width)
                visual_features = self.image_encoder(
                    pixel_values, interpolate_pos_encoding=interpolate_pos_encoding
                ).last_hidden_state

            elif pixel_values.ndim == 5:
                # here we assume pixel_values is of shape (batch_size, num_frames, num_channels, height, width)
                visual_features = []
                for frame_idx in range(pixel_values.shape[1]):
                    visual_features_frame = self.image_encoder(
                        pixel_values[:, frame_idx, :, :], interpolate_pos_encoding=interpolate_pos_encoding
                    ).last_hidden_state
                    visual_features_frame += self.img_temporal_embedding[frame_idx]
                    visual_features.append(visual_features_frame)

                # finally, concatenate all features along sequence dimension
                visual_features = torch.cat(visual_features, dim=1)

            else:
                raise ValueError("pixel_values must be of rank 4 or 5")

            projected_visual_features = self.visual_projection(visual_features)

        embedding_output = self.embeddings(
            input_ids=input_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            past_key_values_length=past_key_values_length,
        )

        if projected_visual_features is None:
            projected_visual_features = torch.zeros(
                (embedding_output.shape[0], 0, embedding_output.shape[2]),
                dtype=embedding_output.dtype,
                device=embedding_output.device,
            )

        # Repeat visual features to match embedding batch size.
        projected_visual_features = projected_visual_features.repeat(
            embedding_output.size(0) // projected_visual_features.size(0), 1, 1
        )

        # concatenate patch token and text token embeddings
        hidden_states = torch.cat((projected_visual_features, embedding_output), dim=1)

        # By default, an additive causal mask is created
        # for masking the future (one direction).
        tgt_mask = self._generate_future_mask(seq_length, embedding_output.dtype, embedding_output.device)

        # Create an attention mask of shape (batch_size, 1, tgt_seq_len, src_seq_len)
        combined_attention_mask = self.create_attention_mask(
            tgt=embedding_output,
            memory=projected_visual_features,
            tgt_mask=tgt_mask,
            past_key_values_length=past_key_values_length,
        )

        if attention_mask is not None:
            # if the user provides an attention mask, we add it to the default one
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            expanded_attn_mask = _prepare_4d_attention_mask(
                attention_mask, embedding_output.dtype, tgt_len=input_shape[-1]
            ).to(embedding_output.device)
            if past_key_values_length > 0:
                expanded_attn_mask = expanded_attn_mask[:, :, -past_key_values_length:, :]
            else:
                combined_attention_mask[:, :, -input_shape[1] :, -input_shape[1] :] += expanded_attn_mask

        encoder_outputs = self.encoder(
            hidden_states,
            attention_mask=combined_attention_mask,
            head_mask=head_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            pixel_values_present=pixel_values is not None,
        )
        sequence_output = encoder_outputs[0]

        if not return_dict:
            return (sequence_output,) + encoder_outputs[1:]

        return BaseModelOutputWithPast(
            last_hidden_state=sequence_output,
            past_key_values=encoder_outputs.past_key_values,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


@auto_docstring(
    custom_intro="""
    GIT Model with a `language modeling` head on top for autoregressive language modeling.
    """
)
class GitForCausalLM(GitPreTrainedModel, GenerationMixin):
    _tied_weights_keys = ["output.weight"]

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

        self.git = GitModel(config)
        self.output = nn.Linear(config.hidden_size, config.vocab_size)

        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        return self.output

    def set_output_embeddings(self, new_embeddings):
        self.output = new_embeddings

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        pixel_values: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        past_key_values: Optional[Union[Cache, list[torch.Tensor]]] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
        return_dict: Optional[bool] = None,
        **kwargs,
    ) -> Union[tuple[torch.Tensor], CausalLMOutputWithPast]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
            `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are
            ignored (masked), the loss is only computed for the tokens with labels n `[0, ..., config.vocab_size]`

        Examples:

        Image captioning example:

        ```python
        >>> from transformers import AutoProcessor, AutoModelForCausalLM
        >>> import requests
        >>> from PIL import Image

        >>> processor = AutoProcessor.from_pretrained("microsoft/git-base-coco")
        >>> model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-coco")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> pixel_values = processor(images=image, return_tensors="pt").pixel_values

        >>> generated_ids = model.generate(pixel_values=pixel_values, max_length=50)
        >>> generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
        >>> print(generated_caption)
        two cats sleeping on a pink blanket next to remotes.
        ```

        Visual question answering (VQA) example:

        ```python
        >>> from transformers import AutoProcessor, AutoModelForCausalLM
        >>> from huggingface_hub import hf_hub_download
        >>> from PIL import Image

        >>> processor = AutoProcessor.from_pretrained("microsoft/git-base-textvqa")
        >>> model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-textvqa")

        >>> file_path = hf_hub_download(repo_id="nielsr/textvqa-sample", filename="bus.png", repo_type="dataset")
        >>> image = Image.open(file_path).convert("RGB")

        >>> pixel_values = processor(images=image, return_tensors="pt").pixel_values

        >>> question = "what does the front of the bus say at the top?"

        >>> input_ids = processor(text=question, add_special_tokens=False).input_ids
        >>> input_ids = [processor.tokenizer.cls_token_id] + input_ids
        >>> input_ids = torch.tensor(input_ids).unsqueeze(0)

        >>> generated_ids = model.generate(pixel_values=pixel_values, input_ids=input_ids, max_length=50)
        >>> print(processor.batch_decode(generated_ids, skip_special_tokens=True))
        ['what does the front of the bus say at the top? special']
        ```

        Video captioning example:

        ```python
        >>> import av
        >>> import numpy as np
        >>> from PIL import Image
        >>> from huggingface_hub import hf_hub_download
        >>> from transformers import AutoProcessor, AutoModelForCausalLM

        >>> processor = AutoProcessor.from_pretrained("microsoft/git-base-vatex")
        >>> model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-vatex")

        >>> # set seed for reproducibility
        >>> np.random.seed(45)


        >>> def read_video_pyav(container, indices):
        ...     '''
        ...     Decode the video with PyAV decoder.
        ...     Args:
        ...         container (`av.container.input.InputContainer`): PyAV container.
        ...         indices (`list[int]`): List of frame indices to decode.
        ...     Returns:
        ...         result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
        ...     '''
        ...     frames = []
        ...     container.seek(0)
        ...     start_index = indices[0]
        ...     end_index = indices[-1]
        ...     for i, frame in enumerate(container.decode(video=0)):
        ...         if i > end_index:
        ...             break
        ...         if i >= start_index and i in indices:
        ...             frames.append(frame)
        ...     return np.stack([x.to_ndarray(format="rgb24") for x in frames])


        >>> def sample_frame_indices(clip_len, frame_sample_rate, seg_len):
        ...     '''
        ...     Sample a given number of frame indices from the video.
        ...     Args:
        ...         clip_len (`int`): Total number of frames to sample.
        ...         frame_sample_rate (`int`): Sample every n-th frame.
        ...         seg_len (`int`): Maximum allowed index of sample's last frame.
        ...     Returns:
        ...         indices (`list[int]`): List of sampled frame indices
        ...     '''
        ...     converted_len = int(clip_len * frame_sample_rate)
        ...     end_idx = np.random.randint(converted_len, seg_len)
        ...     start_idx = end_idx - converted_len
        ...     indices = np.linspace(start_idx, end_idx, num=clip_len)
        ...     indices = np.clip(indices, start_idx, end_idx - 1).astype(np.int64)
        ...     return indices


        >>> # load video
        >>> file_path = hf_hub_download(
        ...     repo_id="nielsr/video-demo", filename="eating_spaghetti.mp4", repo_type="dataset"
        ... )
        >>> container = av.open(file_path)

        >>> # sample frames
        >>> num_frames = model.config.num_image_with_embedding
        >>> indices = sample_frame_indices(
        ...     clip_len=num_frames, frame_sample_rate=4, seg_len=container.streams.video[0].frames
        ... )
        >>> frames = read_video_pyav(container, indices)

        >>> pixel_values = processor(images=list(frames), return_tensors="pt").pixel_values

        >>> generated_ids = model.generate(pixel_values=pixel_values, max_length=50)

        >>> print("Generated caption:", processor.batch_decode(generated_ids, skip_special_tokens=True))
        Generated caption: ['a woman is sitting at a table and she is talking about the food she is holding.']
        ```
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        if labels is not None:
            use_cache = False

        outputs = self.git(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            pixel_values=pixel_values,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            past_key_values=past_key_values,
            use_cache=use_cache,
            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.output(sequence_output)

        loss = None
        if labels is not None:
            # we are doing next-token prediction; shift prediction scores and input ids by one
            num_image_tokens = self.git.encoder.layer[0].attention.self.image_patch_tokens
            shifted_logits = logits[:, num_image_tokens:-1, :].contiguous()
            labels = labels[:, 1:].contiguous()
            loss = self.loss_function(
                shifted_logits.view(-1, self.config.vocab_size),
                labels.view(-1),
                vocab_size=self.config.vocab_size,
                **kwargs,
            )

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(
        self, input_ids, past_key_values=None, attention_mask=None, use_cache=None, **kwargs
    ):
        # Overwritten -- `git` has special cache handling and doesn't support generating from `inputs_embeds` atm

        # cut decoder_input_ids if past_key_values is used
        if past_key_values is not None:
            past_length = past_key_values.get_seq_length()

            # Some generation methods already pass only the last input ID
            if input_ids.shape[1] > past_length:
                remove_prefix_length = past_length
            else:
                # Default to old behavior: keep only final ID
                remove_prefix_length = input_ids.shape[1] - 1

            input_ids = input_ids[:, remove_prefix_length:]

        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
        input_shape = input_ids.shape
        if attention_mask is None:
            attention_mask = input_ids.new_ones(input_shape)

        model_inputs = {
            "input_ids": input_ids,
            "attention_mask": attention_mask,
            "pixel_values": kwargs.get("pixel_values"),
            "past_key_values": past_key_values,
            "use_cache": use_cache,
        }

        # Forward ALL kwargs that are uninitialized (e.g. `use_cache`).
        for key, value in kwargs.items():
            if key not in model_inputs:
                model_inputs[key] = value

        return model_inputs


__all__ = ["GitForCausalLM", "GitModel", "GitPreTrainedModel", "GitVisionModel"]