#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/instructblipvideo/modular_instructblipvideo.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_instructblipvideo.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# coding=utf-8
# Copyright 2024 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.

import math
from dataclasses import dataclass
from typing import Any, Optional, Tuple, Union

import torch
from torch import nn
from torch.nn import CrossEntropyLoss

from ...activations import ACT2FN
from ...generation import GenerationMixin
from ...modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPastAndCrossAttentions,
    BaseModelOutputWithPooling,
    BaseModelOutputWithPoolingAndCrossAttentions,
)
from ...modeling_utils import PreTrainedModel
from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import (
    ModelOutput,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
    replace_return_docstrings,
    torch_int,
)
from ..auto import AutoModelForCausalLM, AutoModelForSeq2SeqLM
from .configuration_instructblipvideo import (
    InstructBlipVideoConfig,
    InstructBlipVideoQFormerConfig,
    InstructBlipVideoVisionConfig,
)


logger = logging.get_logger(__name__)


@dataclass
class InstructBlipVideoForConditionalGenerationModelOutput(ModelOutput):
    """
    Class defining the outputs of [`InstructBlipVideoForConditionalGeneration`].

    Args:
        loss (`torch.FloatTensor`, *optional*, returned when `labels` is provided, `torch.FloatTensor` of shape `(1,)`):
            Language modeling loss from the language model.
        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head of the language model.
        vision_outputs (`BaseModelOutputWithPooling`):
            Outputs of the vision encoder.
        qformer_outputs (`BaseModelOutputWithPoolingAndCrossAttentions`):
            Outputs of the Q-Former (Querying Transformer).
        language_model_outputs (`CausalLMOutputWithPast` or `Seq2SeqLMOutput`):
            Outputs of the language model.
    """

    loss: Optional[Tuple[torch.FloatTensor]] = None
    logits: Optional[Tuple[torch.FloatTensor]] = None
    vision_outputs: Optional[torch.FloatTensor] = None
    qformer_outputs: Optional[Tuple[torch.FloatTensor]] = None
    language_model_outputs: Optional[Tuple[torch.FloatTensor]] = None

    def to_tuple(self) -> Tuple[Any]:
        return tuple(
            self[k]
            if k not in ["vision_outputs", "qformer_outputs", "language_model_outputs"]
            else getattr(self, k).to_tuple()
            for k in self.keys()
        )


class InstructBlipVideoVisionEmbeddings(nn.Module):
    def __init__(self, config: InstructBlipVideoVisionConfig):
        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(1, 1, self.embed_dim))

        self.patch_embedding = nn.Conv2d(
            in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size
        )

        self.num_patches = (self.image_size // self.patch_size) ** 2
        self.num_positions = self.num_patches + 1

        self.position_embedding = nn.Parameter(torch.randn(1, self.num_positions, self.embed_dim))

    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
        num_positions = self.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

        class_pos_embed = self.position_embedding[:, :1]
        patch_pos_embed = self.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: bool = False) -> torch.Tensor:
        batch_size, _, height, width = pixel_values.shape
        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).to(target_dtype)
        embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
        if interpolate_pos_encoding:
            position_embedding = self.interpolate_pos_encoding(embeddings, height, width)
        else:
            position_embedding = self.position_embedding
        embeddings = embeddings + position_embedding[:, : embeddings.size(1), :].to(target_dtype)
        return embeddings


class InstructBlipVideoAttention(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 = nn.Dropout(config.attention_dropout)

        # small tweak here compared to CLIP, no bias here
        self.qkv = nn.Linear(self.embed_dim, 3 * self.embed_dim, bias=False)

        if config.qkv_bias:
            q_bias = nn.Parameter(torch.zeros(self.embed_dim))
            v_bias = nn.Parameter(torch.zeros(self.embed_dim))
        else:
            q_bias = None
            v_bias = None

        if q_bias is not None:
            qkv_bias = torch.cat((q_bias, torch.zeros_like(v_bias, requires_grad=False), v_bias))
            self.qkv.bias = nn.Parameter(qkv_bias)

        self.projection = nn.Linear(self.embed_dim, self.embed_dim)

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

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

        bsz, tgt_len, embed_dim = hidden_states.size()

        mixed_qkv = self.qkv(hidden_states)

        mixed_qkv = mixed_qkv.reshape(bsz, tgt_len, 3, self.num_heads, embed_dim // self.num_heads).permute(
            2, 0, 3, 1, 4
        )
        query_states, key_states, value_states = mixed_qkv[0], mixed_qkv[1], mixed_qkv[2]

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

        attention_scores = attention_scores * self.scale

        # 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_states).permute(0, 2, 1, 3)

        new_context_layer_shape = context_layer.size()[:-2] + (self.embed_dim,)
        context_layer = context_layer.reshape(new_context_layer_shape)

        output = self.projection(context_layer)

        outputs = (output, attention_probs) if output_attentions else (output, None)

        return outputs


class InstructBlipVideoMLP(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


class InstructBlipVideoEncoderLayer(nn.Module):
    def __init__(self, config: InstructBlipVideoConfig):
        super().__init__()
        self.embed_dim = config.hidden_size
        self.self_attn = InstructBlipVideoAttention(config)
        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
        self.mlp = InstructBlipVideoMLP(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,
        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,
            head_mask=attention_mask,
            output_attentions=output_attentions,
        )
        hidden_states = hidden_states + residual
        residual = hidden_states
        hidden_states = self.layer_norm2(hidden_states)
        hidden_states = self.mlp(hidden_states)

        hidden_states = hidden_states + residual

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (attn_weights,)

        return outputs


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

    config_class = InstructBlipVideoConfig
    base_model_prefix = "blip"
    supports_gradient_checkpointing = True

    _no_split_modules = [
        "InstructBlipVideoQFormerEmbeddings",
        "InstructBlipVideoAttention",
        "InstructBlipVideoQFormerMultiHeadAttention",
        "InstructBlipVideoQFormerSelfOutput",
    ]

    def _init_weights(self, module):
        """Initialize the weights"""
        factor = self.config.initializer_range
        if isinstance(module, nn.Conv2d) or isinstance(module, nn.Embedding) or isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=factor)
            if hasattr(module, "bias") and module.bias is not None:
                module.bias.data.zero_()

        if isinstance(module, InstructBlipVideoVisionEmbeddings):
            if hasattr(self.config, "vision_config") and not isinstance(self.config, InstructBlipVideoVisionConfig):
                factor = self.config.vision_config.initializer_range
            nn.init.trunc_normal_(module.position_embedding, mean=0.0, std=factor)
            nn.init.trunc_normal_(module.class_embedding, mean=0.0, std=factor)

        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        elif isinstance(module, nn.Linear) and module.bias is not None:
            module.bias.data.zero_()


class InstructBlipVideoEncoder(nn.Module):
    """
    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
    [`InstructBlipVideoEncoderLayer`].

    Args:
        config (`InstructBlipVideoConfig`):
            The corresponding vision configuration for the `InstructBlipVideoEncoder`.
    """

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

    def forward(
        self,
        inputs_embeds,
        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)`):
                Embedded representation of the inputs. Should be float, not int tokens.
            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)
            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,)
            if self.gradient_checkpointing and self.training:
                layer_outputs = self._gradient_checkpointing_func(
                    encoder_layer.__call__,
                    hidden_states,
                    attention_mask,
                    output_attentions,
                )
            else:
                layer_outputs = encoder_layer(
                    hidden_states,
                    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,)

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


INSTRUCTBLIPVIDEO_VISION_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using [`InstructBlipVideoProcessor`]. See
            [`InstructBlipVideoProcessor.__call__`] for details.
        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.
        interpolate_pos_encoding (`bool`, *optional*, defaults to `False`):
            Whether to interpolate the pre-trained position encodings.
"""


class InstructBlipVideoVisionModel(InstructBlipVideoPreTrainedModel):
    main_input_name = "pixel_values"
    config_class = InstructBlipVideoVisionConfig

    def __init__(self, config: InstructBlipVideoVisionConfig):
        super().__init__(config)
        self.config = config
        embed_dim = config.hidden_size

        self.embeddings = InstructBlipVideoVisionEmbeddings(config)
        self.encoder = InstructBlipVideoEncoder(config)
        self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)

        self.post_init()

    @add_start_docstrings_to_model_forward(INSTRUCTBLIPVIDEO_VISION_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=InstructBlipVideoVisionConfig)
    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
    ) -> Union[Tuple, BaseModelOutputWithPooling]:
        r"""
        Returns:

        """
        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)

        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)

        pooled_output = last_hidden_state[:, 0, :]
        pooled_output = self.post_layernorm(pooled_output)

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

        return BaseModelOutputWithPooling(
            last_hidden_state=last_hidden_state,
            pooler_output=pooled_output,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )

    def get_input_embeddings(self):
        return self.embeddings


class InstructBlipVideoQFormerMultiHeadAttention(nn.Module):
    def __init__(self, config, is_cross_attention=False):
        super().__init__()
        self.config = config
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention heads (%d)"
                % (config.hidden_size, config.num_attention_heads)
            )

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

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        if is_cross_attention:
            self.key = nn.Linear(config.encoder_hidden_size, self.all_head_size)
            self.value = nn.Linear(config.encoder_hidden_size, self.all_head_size)
        else:
            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 = 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)
        self.save_attention = False

    def save_attn_gradients(self, attn_gradients):
        self.attn_gradients = attn_gradients

    def get_attn_gradients(self):
        return self.attn_gradients

    def save_attention_map(self, attention_map):
        self.attention_map = attention_map

    def get_attention_map(self):
        return self.attention_map

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

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):
        # If this is instantiated as a cross-attention module, the keys
        # and values come from an encoder; the attention mask needs to be
        # such that the encoder's padding tokens are not attended to.
        is_cross_attention = encoder_hidden_states is not None

        if is_cross_attention:
            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
            attention_mask = encoder_attention_mask
        elif past_key_value is not None:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))
            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
        else:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))

        mixed_query_layer = self.query(hidden_states)

        query_layer = self.transpose_for_scores(mixed_query_layer)

        past_key_value = (key_layer, value_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))

        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            seq_length = hidden_states.size()[1]
            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
            position_ids_r = torch.arange(seq_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)
        attention_scores_dtype = attention_scores.dtype

        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
            attention_scores = attention_scores + attention_mask

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

        if is_cross_attention and self.save_attention:
            self.save_attention_map(attention_probs)
            attention_probs.register_hook(self.save_attn_gradients)

        # 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_dropped = self.dropout(attention_probs)

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

        context_layer = torch.matmul(attention_probs_dropped, 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,)

        outputs = outputs + (past_key_value,)
        return outputs


class InstructBlipVideoQFormerSelfOutput(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


class InstructBlipVideoQFormerAttention(nn.Module):
    def __init__(self, config, is_cross_attention=False):
        super().__init__()
        self.attention = InstructBlipVideoQFormerMultiHeadAttention(config, is_cross_attention)
        self.output = InstructBlipVideoQFormerSelfOutput(config)
        self.pruned_heads = set()

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

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

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

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        encoder_hidden_states: Optional[torch.FloatTensor] = None,
        encoder_attention_mask: Optional[torch.FloatTensor] = None,
        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
        output_attentions: Optional[bool] = False,
    ) -> Tuple[torch.Tensor]:
        self_outputs = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            encoder_hidden_states,
            encoder_attention_mask,
            past_key_value,
            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 InstructBlipVideoQFormerIntermediate(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


class InstructBlipVideoQFormerOutput(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 InstructBlipVideoQFormerLayer(nn.Module):
    def __init__(self, config, layer_idx):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = InstructBlipVideoQFormerAttention(config)

        self.layer_idx = layer_idx

        if layer_idx % config.cross_attention_frequency == 0:
            self.crossattention = InstructBlipVideoQFormerAttention(config, is_cross_attention=True)
            self.has_cross_attention = True
        else:
            self.has_cross_attention = False

        self.intermediate = InstructBlipVideoQFormerIntermediate(config)
        self.output = InstructBlipVideoQFormerOutput(config)

        self.intermediate_query = InstructBlipVideoQFormerIntermediate(config)
        self.output_query = InstructBlipVideoQFormerOutput(config)

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

        present_key_value = self_attention_outputs[-1]

        if query_length > 0:
            query_attention_output = attention_output[:, :query_length, :]

            if self.has_cross_attention:
                if encoder_hidden_states is None:
                    raise ValueError("encoder_hidden_states must be given for cross-attention layers")
                cross_attention_outputs = self.crossattention(
                    query_attention_output,
                    attention_mask,
                    head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    output_attentions=output_attentions,
                )
                query_attention_output = cross_attention_outputs[0]
                # add cross attentions if we output attention weights
                outputs = outputs + cross_attention_outputs[1:-1]

            layer_output = apply_chunking_to_forward(
                self.feed_forward_chunk_query,
                self.chunk_size_feed_forward,
                self.seq_len_dim,
                query_attention_output,
            )

            if attention_output.shape[1] > query_length:
                layer_output_text = apply_chunking_to_forward(
                    self.feed_forward_chunk,
                    self.chunk_size_feed_forward,
                    self.seq_len_dim,
                    attention_output[:, query_length:, :],
                )
                layer_output = torch.cat([layer_output, layer_output_text], dim=1)
        else:
            layer_output = apply_chunking_to_forward(
                self.feed_forward_chunk,
                self.chunk_size_feed_forward,
                self.seq_len_dim,
                attention_output,
            )
        outputs = (layer_output,) + outputs

        outputs = outputs + (present_key_value,)

        return outputs

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

    def feed_forward_chunk_query(self, attention_output):
        intermediate_output = self.intermediate_query(attention_output)
        layer_output = self.output_query(intermediate_output, attention_output)
        return layer_output


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

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
        query_length=0,
    ):
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        all_cross_attentions = () if output_attentions else None

        next_decoder_cache = () if use_cache else None

        for i in range(self.config.num_hidden_layers):
            layer_module = self.layer[i]
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

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

            if getattr(self.config, "gradient_checkpointing", False) and self.training:
                if use_cache:
                    logger.warning(
                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                    )
                    use_cache = False
                layer_outputs = self._gradient_checkpointing_func(
                    layer_module.__call__,
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                )
            else:
                layer_outputs = layer_module(
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    past_key_value,
                    output_attentions,
                    query_length,
                )

            hidden_states = layer_outputs[0]
            if use_cache:
                next_decoder_cache += (layer_outputs[-1],)
            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)
                if layer_module.has_cross_attention:
                    all_cross_attentions = all_cross_attentions + (layer_outputs[2],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(
                v
                for v in [
                    hidden_states,
                    next_decoder_cache,
                    all_hidden_states,
                    all_self_attentions,
                    all_cross_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=next_decoder_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
            cross_attentions=all_cross_attentions,
        )


class InstructBlipVideoQFormerEmbeddings(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 = 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.register_buffer(
            "position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)), persistent=False
        )
        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")

        self.config = config

    def forward(
        self,
        input_ids=None,
        position_ids=None,
        query_embeds=None,
        past_key_values_length=0,
    ):
        if input_ids is not None:
            seq_length = input_ids.size()[1]
        else:
            seq_length = 0

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

        if input_ids is not None:
            embeddings = self.word_embeddings(input_ids)
            if self.position_embedding_type == "absolute":
                position_embeddings = self.position_embeddings(position_ids.to(embeddings.device))
                embeddings = embeddings + position_embeddings

            if query_embeds is not None:
                embeddings = torch.cat((query_embeds, embeddings), dim=1)
        else:
            embeddings = query_embeds

        embeddings = embeddings.to(self.layernorm.weight.dtype)
        embeddings = self.layernorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class InstructBlipVideoQFormerModel(InstructBlipVideoPreTrainedModel):
    """
    Querying Transformer (Q-Former), used in InstructBlipVideo. Slightly modified from BLIP-2 as it also takes the
    instruction as input.
    """

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

        self.embeddings = InstructBlipVideoQFormerEmbeddings(config)

        self.encoder = InstructBlipVideoQFormerEncoder(config)

        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 get_extended_attention_mask(
        self,
        attention_mask: torch.Tensor,
        input_shape: Tuple[int],
        device: torch.device,
        has_query: bool = False,
    ) -> torch.Tensor:
        """
        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.

        Arguments:
            attention_mask (`torch.Tensor`):
                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
            input_shape (`Tuple[int]`):
                The shape of the input to the model.
            device: (`torch.device`):
                The device of the input to the model.

        Returns:
            `torch.Tensor` The extended attention mask, with a the same dtype as `attention_mask.dtype`.
        """
        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        if attention_mask.dim() == 3:
            extended_attention_mask = attention_mask[:, None, :, :]
        elif attention_mask.dim() == 2:
            # Provided a padding mask of dimensions [batch_size, seq_length]
            # - the model is an encoder, so make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
            extended_attention_mask = attention_mask[:, None, None, :]
        else:
            raise ValueError(
                f"Wrong shape for input_ids (shape {input_shape}) or attention_mask (shape {attention_mask.shape})",
            )

        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
        return extended_attention_mask

    def forward(
        self,
        input_ids: torch.LongTensor,
        attention_mask: Optional[torch.FloatTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        query_embeds: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        encoder_hidden_states: Optional[torch.FloatTensor] = None,
        encoder_attention_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple[torch.FloatTensor], BaseModelOutputWithPoolingAndCrossAttentions]:
        r"""
        encoder_hidden_states  (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
            the model is configured as a decoder.
        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.
        past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of:
            shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and
            value hidden states of the attention blocks. Can be used to speed up decoding. If `past_key_values` are
            used, the user can optionally input only the last `decoder_input_ids` (those that don't have their past key
            value states given to this model) of shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape
            `(batch_size, sequence_length)`.
        use_cache (`bool`, *optional*):
            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
            `past_key_values`).
        """
        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 input_ids is None and query_embeds is None:
            raise ValueError("You have to specify query_embeds when input_ids is None")

        # past_key_values_length
        past_key_values_length = (
            past_key_values[0][0].shape[2] - self.config.query_length if past_key_values is not None else 0
        )

        query_length = query_embeds.shape[1] if query_embeds is not None else 0

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

        input_shape = embedding_output.size()[:-1]
        batch_size, seq_length = input_shape
        device = embedding_output.device

        if attention_mask is None:
            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)

        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device)

        # If a 2D or 3D attention mask is provided for the cross-attention
        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
        if encoder_hidden_states is not None:
            if isinstance(encoder_hidden_states, list):
                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
            else:
                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)

            if isinstance(encoder_attention_mask, list):
                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
            elif encoder_attention_mask is None:
                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
            else:
                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
        else:
            encoder_extended_attention_mask = None

        # 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)

        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask=extended_attention_mask,
            head_mask=head_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_extended_attention_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,
            query_length=query_length,
        )
        sequence_output = encoder_outputs[0]
        pooled_output = sequence_output[:, 0, :]

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

        return BaseModelOutputWithPoolingAndCrossAttentions(
            last_hidden_state=sequence_output,
            pooler_output=pooled_output,
            past_key_values=encoder_outputs.past_key_values,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
            cross_attentions=encoder_outputs.cross_attentions,
        )


INSTRUCTBLIPVIDEO_START_DOCSTRING = r"""
    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
    etc.)

    This model is also 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 ([`InstructBlipVideoConfig`]): 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.
"""

INSTRUCTBLIPVIDEO_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using [`InstructBlipVideoProcessor`]. See
            [`InstructBlipVideoProcessor.__call__`] for details.

        qformer_input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Indices of input sequence tokens in the vocabulary of the Q-Former. Input tokens can optionally be provided
            to serve as text prompt, which the Q-Former model will encode.

            Indices can be obtained using [`InstructBlipVideoProcessor`]. See [`InstructBlipVideoProcessor.__call__`] for
            details.

            [What are input IDs?](../glossary#input-ids)

        qformer_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)

        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Indices of input sequence tokens in the vocabulary of the language model. Input tokens can optionally be
            provided to serve as text prompt, which the language model can continue.

            Indices can be obtained using [`InstructBlipVideoProcessor`]. See [`InstructBlipVideoProcessor.__call__`] for
            details.

            [What are input IDs?](../glossary#input-ids)

        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)

        decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
            Indices of decoder input sequence tokens in the vocabulary of the language model. Only relevant in case an
            encoder-decoder language model (like T5) is used.

            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details. [What are decoder input IDs?](../glossary#decoder-input-ids)

        decoder_attention_mask (`torch.BoolTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
            Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
            be used by default.

            Only relevant in case an encoder-decoder language model (like T5) is used.

        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.
        interpolate_pos_encoding (`bool`, *optional*, defaults to `False`):
            Whether to interpolate the pre-trained position encodings.
        use_cache (`bool`, *optional*):
            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
            `past_key_values`).
"""


@add_start_docstrings(
    """
    InstructBlipVideo Model for generating text given an image and an optional text prompt. The model consists of a vision
    encoder, Querying Transformer (Q-Former) and a language model.

    One can optionally pass `input_ids` to the model, which serve as a text prompt, to make the language model continue
    the prompt. Otherwise, the language model starts generating text from the [BOS] (beginning-of-sequence) token.
    """,
    INSTRUCTBLIPVIDEO_START_DOCSTRING,
)
class InstructBlipVideoForConditionalGeneration(InstructBlipVideoPreTrainedModel, GenerationMixin):
    config_class = InstructBlipVideoConfig
    main_input_name = "pixel_values"
    _supports_cache_class = True
    _supports_static_cache = True
    _supports_quantized_cache = False  # not all LM bacbones support (e.g. T5)
    _keep_in_fp32_modules = ["query_tokens"]  # TODO @ArthurZucker I don't know why this is required for FP8

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

        self.vision_model = InstructBlipVideoVisionModel(config.vision_config)

        self.query_tokens = nn.Parameter(torch.zeros(1, config.num_query_tokens, config.qformer_config.hidden_size))
        self.qformer = InstructBlipVideoQFormerModel(config.qformer_config)

        self.language_projection = nn.Linear(config.qformer_config.hidden_size, config.text_config.hidden_size)

        if config.use_decoder_only_language_model:
            language_model = AutoModelForCausalLM.from_config(config.text_config)
        else:
            language_model = AutoModelForSeq2SeqLM.from_config(config.text_config)

        if language_model._no_split_modules is not None:
            self._no_split_modules.extend(language_model._no_split_modules)

        if language_model._keep_in_fp32_modules is not None:
            self._keep_in_fp32_modules.extend(language_model._keep_in_fp32_modules)

        self.language_model = language_model

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

    def get_input_embeddings(self):
        return self.language_model.get_input_embeddings()

    def set_input_embeddings(self, value):
        self.language_model.set_input_embeddings(value)

    def set_output_embeddings(self, new_embeddings):
        self.language_model.set_output_embeddings(new_embeddings)

    def get_output_embeddings(self) -> nn.Module:
        return self.language_model.get_output_embeddings()

    def get_encoder(self):
        return self.language_model.get_encoder()

    def get_decoder(self):
        return self.language_model.get_decoder()

    def _tie_weights(self):
        if not self.config.use_decoder_only_language_model:
            self.language_model.encoder.embed_tokens = self.language_model.shared
            self.language_model.decoder.embed_tokens = self.language_model.shared

    def _preprocess_accelerate(self):
        r"""
        Some pre-processing hacks to make the model `accelerate` compatible. Check
        https://github.com/huggingface/transformers/pull/21707 for more details.
        """
        hf_device_map = self.hf_device_map

        if len(hf_device_map) > 1 and "language_model" not in hf_device_map and torch.cuda.device_count() > 1:
            # warn users about unexpected behavior when using multi-GPU + InstructBlipVideo + `accelerate`.
            logger.warning(
                "The `language_model` is not in the `hf_device_map` dictionary and you are running your script"
                " in a multi-GPU environment. this may lead to unexpected behavior when using `accelerate`."
                " Please pass a `device_map` that contains `language_model` to remove this warning."
                " Please refer to https://github.com/huggingface/blog/blob/main/accelerate-large-models.md for"
                " more details on creating a `device_map` for large models.",
            )

        if hasattr(self.language_model, "_hf_hook"):
            self.language_model._hf_hook.io_same_device = True  # For `generate` compatibility

    @add_start_docstrings_to_model_forward(INSTRUCTBLIPVIDEO_INPUTS_DOCSTRING)
    @replace_return_docstrings(
        output_type=InstructBlipVideoForConditionalGenerationModelOutput, config_class=InstructBlipVideoVisionConfig
    )
    def forward(
        self,
        pixel_values: torch.FloatTensor,
        qformer_input_ids: torch.FloatTensor,
        qformer_attention_mask: Optional[torch.LongTensor] = None,
        input_ids: Optional[torch.FloatTensor] = None,
        attention_mask: Optional[torch.LongTensor] = None,
        decoder_input_ids: Optional[torch.LongTensor] = None,
        decoder_attention_mask: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        labels: Optional[torch.LongTensor] = None,
        return_dict: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
        use_cache: Optional[bool] = None,
    ) -> Union[Tuple, InstructBlipVideoForConditionalGenerationModelOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size -
            1]`. All labels set to `-100` are ignored (masked), the loss is only computed for labels in `[0, ...,
            config.vocab_size]`

        Returns:

        Examples:

        ```python
        >>> from transformers import InstructBlipVideoProcessor, InstructBlipVideoForConditionalGeneration
        >>> import torch
        >>> from huggingface_hub import hf_hub_download
        >>> import av
        >>> import numpy as np

        >>> 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])

        >>> model = InstructBlipVideoForConditionalGeneration.from_pretrained("Salesforce/instructblip-vicuna-7b", device_map="auto")
        >>> processor = InstructBlipVideoProcessor.from_pretrained("Salesforce/instructblip-vicuna-7b")

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

        >>> # sample uniformly 4 frames from the videWhy is this video funny?o
        >>> total_frames = container.streams.video[0].frames
        >>> indices = np.arange(0, total_frames, total_frames / 4).astype(int)
        >>> clip = read_video_pyav(container, indices)

        >>> prompt = "What is happening in the video?"
        >>> inputs = processor(text=prompt, images=clip, return_tensors="pt").to(model.device)

        >>> outputs = model.generate(
        ...     **inputs,
        ...     do_sample=False,
        ...     num_beams=5,
        ...     max_length=256,
        ...     repetition_penalty=1.5,
        ...     length_penalty=1.0,
        ... )
        >>> generated_text = processor.batch_decode(outputs, skip_special_tokens=True)[0].strip()
        >>> print(generated_text)
        "A person is eating a bowl of pasta, and they are using a fork to eat it. The person is sitting at a table, and the plate of pasta is on the table in front"
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # step 1: forward the images through the vision encoder,
        # we process in a batched way, later unbatch it back (video has frames=4 always)
        batch_size, frames, channel, height, width = pixel_values.shape
        pixel_values = pixel_values.reshape(batch_size * frames, channel, height, width)

        vision_outputs = self.vision_model(
            pixel_values=pixel_values,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            interpolate_pos_encoding=interpolate_pos_encoding,
        )
        image_embeds = vision_outputs[0]

        # step 2: forward the query tokens through the QFormer, using the image embeddings for cross-attention
        image_attention_mask = torch.ones(image_embeds.size()[:-1], dtype=torch.long, device=image_embeds.device)

        # difference with BLIP-2 here: we also feed the instruction prompt to the Q-Former
        query_tokens = self.query_tokens.expand(image_embeds.shape[0], -1, -1)
        query_attention_mask = torch.ones(query_tokens.size()[:-1], dtype=torch.long, device=image_embeds.device)

        if qformer_attention_mask is None:
            qformer_attention_mask = torch.ones_like(qformer_input_ids)

        qformer_input_ids = qformer_input_ids.repeat_interleave(frames, dim=0)
        qformer_attention_mask = qformer_attention_mask.repeat_interleave(frames, dim=0)
        qformer_attention_mask = torch.cat([query_attention_mask, qformer_attention_mask], dim=1)
        query_outputs = self.qformer(
            input_ids=qformer_input_ids,
            attention_mask=qformer_attention_mask,
            query_embeds=query_tokens,
            encoder_hidden_states=image_embeds,
            encoder_attention_mask=image_attention_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        query_output = query_outputs[0][:, : query_tokens.size(1), :]

        # step 3: use the language model, conditioned on the query outputs and the prompt
        language_model_inputs = self.language_projection(query_output)

        # unbatch inputs back, each video-frame gets `num_query_tokens` seq length
        language_model_inputs = language_model_inputs.reshape(batch_size, self.config.num_query_tokens * frames, -1)
        language_model_attention_mask = torch.ones(
            language_model_inputs.size()[:-1], dtype=torch.long, device=language_model_inputs.device
        )

        inputs_embeds = self.language_model.get_input_embeddings()(input_ids)
        if attention_mask is None:
            attention_mask = torch.ones_like(input_ids)

        # if the model already has "video_token_index" then the input is expanded to account for image embeds
        # otherwise we expand manually by concatenating
        if getattr(self.config, "video_token_index", None) is not None:
            special_image_mask = (input_ids == self.config.video_token_index).unsqueeze(-1).expand_as(inputs_embeds)
            inputs_embeds[special_image_mask] = language_model_inputs.flatten().to(inputs_embeds.device)
        else:
            logger.warning_once(
                "Expanding inputs for video tokens in InstructBLIPVideo should be done in processing. "
                "Please follow instruction here (https://gist.github.com/zucchini-nlp/65f22892b054dc0d68228af56fbeaac2) to update your InstructBLIPVideo model. "
                "Using processors without these attributes in the config is deprecated and will throw an error in v4.47."
            )
            inputs_embeds = torch.cat([language_model_inputs, inputs_embeds.to(language_model_inputs.device)], dim=1)
            attention_mask = torch.cat(
                [language_model_attention_mask, attention_mask.to(language_model_attention_mask.device)], dim=1
            )

        if self.config.use_decoder_only_language_model:
            outputs = self.language_model(
                inputs_embeds=inputs_embeds,
                attention_mask=attention_mask,
                output_attentions=output_attentions,
                output_hidden_states=output_hidden_states,
                return_dict=return_dict,
                use_cache=use_cache,
            )
            logits = outputs.logits if return_dict else outputs[0]
            loss = None
            # we compute the loss here since we need to take into account the sequence length of the query embeds
            if labels is not None:
                labels = labels.to(logits.device)
                logits = logits[:, -labels.size(1) :, :]
                # Shift so that tokens < n predict n
                shift_logits = logits[..., :-1, :].contiguous()
                shift_labels = labels[..., 1:].contiguous().to(logits.device)

                # Flatten the tokens
                loss_fct = CrossEntropyLoss(reduction="mean")

                loss = loss_fct(shift_logits.view(-1, self.config.text_config.vocab_size), shift_labels.view(-1))
        else:
            outputs = self.language_model(
                inputs_embeds=inputs_embeds,
                attention_mask=attention_mask,
                decoder_input_ids=decoder_input_ids,
                decoder_attention_mask=decoder_attention_mask,
                output_attentions=output_attentions,
                output_hidden_states=output_hidden_states,
                return_dict=return_dict,
                labels=labels,
                use_cache=use_cache,
            )
            loss = outputs.loss if return_dict else outputs[0]
            logits = outputs.logits if return_dict else outputs[1]

        if not return_dict:
            output = (logits, vision_outputs, query_outputs, outputs)
            return ((loss,) + output) if loss is not None else output

        return InstructBlipVideoForConditionalGenerationModelOutput(
            loss=loss,
            logits=logits,
            vision_outputs=vision_outputs,
            qformer_outputs=query_outputs,
            language_model_outputs=outputs,
        )

    @torch.no_grad()
    def generate(
        self,
        pixel_values: torch.FloatTensor,
        qformer_input_ids: Optional[torch.LongTensor] = None,
        qformer_attention_mask: Optional[torch.LongTensor] = None,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.LongTensor] = None,
        interpolate_pos_encoding: bool = False,
        **generate_kwargs,
    ) -> torch.LongTensor:
        r"""
        Overrides `generate` function to be able to use the model as a conditional generator.

        Args:
            pixel_values (`torch.FloatTensor` of shape (batch_size, num_channels, height, width) or
                (batch_size, num_frames, num_channels, height, width)): Input images or videos to be processed.
            qformer_input_ids (`torch.LongTensor` of shape (batch_size, sequence_length), *optional*):
                The sequence used as a prompt to be fed to the Q-Former module.
            qformer_attention_mask (`torch.LongTensor` of shape (batch_size, sequence_length), *optional*):
                Mask to avoid performing attention on padding token indices.
            input_ids (`torch.LongTensor` of shape (batch_size, sequence_length), *optional*):
                The sequence used as a prompt for the generation.
            attention_mask (`torch.LongTensor` of shape (batch_size, sequence_length), *optional*):
                Mask to avoid performing attention on padding token indices.
            interpolate_pos_encoding (`bool`, *optional*, defaults to `False`):
                Whether to interpolate the positional encoding of the image embeddings.

        Returns:
            captions (list): A list of strings of length batch_size * num_captions.
        """
        if hasattr(self, "hf_device_map"):
            # preprocess for `accelerate`
            self._preprocess_accelerate()

        # we process in a batched way, later unbatch it back (video has frames=4)
        batch_size, frames, channel, height, width = pixel_values.shape
        pixel_values = pixel_values.reshape(batch_size * frames, channel, height, width)

        image_embeds = self.vision_model(
            pixel_values,
            return_dict=True,
            interpolate_pos_encoding=interpolate_pos_encoding,
        ).last_hidden_state
        image_attention_mask = torch.ones(image_embeds.size()[:-1], dtype=torch.long, device=image_embeds.device)

        query_tokens = self.query_tokens.expand(image_embeds.shape[0], -1, -1)
        query_attention_mask = torch.ones(query_tokens.size()[:-1], dtype=torch.long, device=image_embeds.device)
        if qformer_attention_mask is None:
            qformer_attention_mask = torch.ones_like(qformer_input_ids)

        qformer_input_ids = qformer_input_ids.repeat_interleave(frames, dim=0)
        qformer_attention_mask = qformer_attention_mask.repeat_interleave(frames, dim=0)
        qformer_attention_mask = torch.cat([query_attention_mask, qformer_attention_mask], dim=1)
        query_outputs = self.qformer(
            input_ids=qformer_input_ids,
            attention_mask=qformer_attention_mask,
            query_embeds=query_tokens,
            encoder_hidden_states=image_embeds,
            encoder_attention_mask=image_attention_mask,
            return_dict=True,
        )
        query_output = query_outputs.last_hidden_state[:, : query_tokens.size(1), :]

        language_model_inputs = self.language_projection(query_output)

        # unbatch the embeddings back by moving frames to seq-len
        language_model_inputs = language_model_inputs.reshape(batch_size, self.config.num_query_tokens * frames, -1)
        language_attention_mask = torch.ones(
            language_model_inputs.size()[:-1], dtype=torch.long, device=language_model_inputs.device
        )

        if input_ids is None:
            start_tokens = [self.config.text_config.bos_token_id]
            if getattr(self.config, "video_token_index", None) is not None:
                start_tokens = [self.config.video_token_index] * self.config.num_query_tokens * 4 + start_tokens
            input_ids = torch.tensor([start_tokens], dtype=torch.long, device=image_embeds.device)
            input_ids = input_ids.repeat(batch_size, 1)

        if attention_mask is None:
            attention_mask = torch.ones_like(input_ids)

        inputs_embeds = self.get_input_embeddings()(input_ids)

        # if the model already has "video_token_index" then the input is expanded to account for image embeds
        # otherwise we expand manually by concatenating
        if getattr(self.config, "video_token_index", None) is not None:
            special_image_mask = (input_ids == self.config.video_token_index).unsqueeze(-1).expand_as(inputs_embeds)
            inputs_embeds[special_image_mask] = language_model_inputs.flatten().to(inputs_embeds.device)
        else:
            logger.warning_once(
                "Expanding inputs for video tokens in InstructBLIPVideo should be done in processing. "
                "Please follow instruction here (https://gist.github.com/zucchini-nlp/65f22892b054dc0d68228af56fbeaac2) to update your InstructBLIPVideo model. "
                "Using processors without these attributes in the config is deprecated and will throw an error in v4.47."
            )
            inputs_embeds = torch.cat([language_model_inputs, inputs_embeds.to(language_model_inputs.device)], dim=1)
            attention_mask = torch.cat(
                [language_attention_mask, attention_mask.to(language_attention_mask.device)], dim=1
            )

            # add image_embeds length to max_length, so that the final max_length in counted only on token embeds
            # -1 is to account for the prepended BOS after `generate.`
            if not self.language_model.config.is_encoder_decoder:
                generate_kwargs["max_length"] = (
                    generate_kwargs.get("max_length", 20) + language_model_inputs.shape[1] - 1
                )
                generate_kwargs["min_length"] = generate_kwargs.get("min_length", 0) + language_model_inputs.shape[1]

        inputs = {"inputs_embeds": inputs_embeds, "attention_mask": attention_mask}
        if not self.language_model.config.is_encoder_decoder:
            inputs["input_ids"] = input_ids

        outputs = self.language_model.generate(**inputs, **generate_kwargs)

        return outputs