# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo

from transformers import (
    Cache,
    DynamicCache,
    PretrainedConfig,
    Qwen2_5_VLTextConfig,
    Qwen2RMSNorm,
)
from transformers.masking_utils import (
    create_causal_mask,
    create_sliding_window_causal_mask,
)
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from transformers.modeling_outputs import BaseModelOutputWithPast
from transformers.utils import TransformersKwargs, is_torchdynamo_compiling

from sglang.multimodal_gen.configs.models.encoders.qwen_image import Qwen2_5VLConfig
from sglang.multimodal_gen.runtime.layers.attention import LocalAttention
from sglang.multimodal_gen.runtime.layers.linear import (
    MergedColumnParallelLinear,
    RowParallelLinear,
)
from sglang.multimodal_gen.runtime.layers.quantization import QuantizationConfig
from sglang.multimodal_gen.runtime.loader.weight_utils import default_weight_loader
from sglang.multimodal_gen.runtime.models.encoders.base import TextEncoder
from sglang.multimodal_gen.runtime.platforms import AttentionBackendEnum
from sglang.multimodal_gen.runtime.utils.common import add_prefix

# coding=utf-8
# Adapted from
# https://github.com/huggingface/transformers/blob/19e6e80e10118f855137b90740936c0b11ac397f/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py
# Copyright 2024 The Qwen team.
# Copyright 2023 The vLLM team.
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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.
"""Inference-only Qwen2-VL model compatible with HuggingFace weights."""
import logging
from typing import Callable, Iterable, Optional, Tuple, Union

try:
    from typing import Unpack  # type: ignore[attr-defined]
except ImportError:
    # Python 3.10 and below
    from typing_extensions import Unpack

import torch
import torch.nn as nn
from transformers.activations import ACT2FN
from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import (
    Qwen2_5_VisionTransformerPretrainedModel,
    Qwen2_5_VLAttention,
    Qwen2_5_VLCausalLMOutputWithPast,
    Qwen2_5_VLModelOutputWithPast,
    Qwen2_5_VLRotaryEmbedding,
    Qwen2MLP,
    apply_multimodal_rotary_pos_emb,
    eager_attention_forward,
)

logger = logging.getLogger(__name__)


class Qwen2_5_VLAttention(nn.Module):
    """
    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
    and "Generating Long Sequences with Sparse Transformers".
    """

    def __init__(self, config: Qwen2_5_VLTextConfig, layer_idx: Optional[int] = None):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        if layer_idx is None:
            logger.warn(
                f"Instantiating {self.__class__.__name__} without passing `layer_idx` is not recommended and will "
                "to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` "
                "when creating this class."
            )

        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.hidden_size // self.num_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
        self.is_causal = True
        self.attention_dropout = config.attention_dropout
        self.rope_scaling = config.rope_scaling
        self.scaling = self.head_dim**-0.5

        if (self.head_dim * self.num_heads) != self.hidden_size:
            raise ValueError(
                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                f" and `num_heads`: {self.num_heads})."
            )
        self.q_proj = nn.Linear(
            self.hidden_size, self.num_heads * self.head_dim, bias=True
        )
        self.k_proj = nn.Linear(
            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True
        )
        self.v_proj = nn.Linear(
            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True
        )
        self.o_proj = nn.Linear(
            self.num_heads * self.head_dim, self.hidden_size, bias=False
        )
        self.sliding_window = (
            config.sliding_window
            if config.layer_types[layer_idx] == "sliding_attention"
            else None
        )

        self.rotary_emb = Qwen2_5_VLRotaryEmbedding(config=config)
        self.attn = LocalAttention(
            num_heads=self.num_heads,
            head_size=self.head_dim,
            num_kv_heads=self.num_key_value_heads,
            softmax_scale=self.scaling,
            causal=True,
            supported_attention_backends=(
                AttentionBackendEnum.FA,
                AttentionBackendEnum.TORCH_SDPA,
            ),
        )

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[
            tuple[torch.Tensor, torch.Tensor]
        ] = None,  # necessary, but kept here for BC
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()

        query_states = self.q_proj(hidden_states)
        key_states = self.k_proj(hidden_states)
        value_states = self.v_proj(hidden_states)

        query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = apply_multimodal_rotary_pos_emb(
            query_states, key_states, cos, sin, self.rope_scaling["mrope_section"]
        )

        if past_key_values is not None:
            cache_kwargs = {
                "sin": sin,
                "cos": cos,
                "cache_position": cache_position,
            }  # Specific to RoPE models
            key_states, value_states = past_key_values.update(
                key_states, value_states, self.layer_idx, cache_kwargs
            )

        attention_interface: Callable = eager_attention_forward
        # if self.config._attn_implementation != "eager":
        # attention_interface = ALL_ATTENTION_FUNCTIONS["sdpa"]
        query_states = query_states.transpose(1, 2)
        key_states = key_states.transpose(1, 2)
        value_states = value_states.transpose(1, 2)
        attn_output = self.attn(query_states, key_states, value_states)
        #
        # attn_output, attn_weights = attention_interface(
        #     self,
        #     query_states,
        #     key_states,
        #     value_states,
        #     attention_mask,
        #     dropout=0.0 if not self.training else self.attention_dropout,
        #     scaling=self.scaling,
        #     sliding_window=self.sliding_window,
        #     position_ids=position_ids,  # pass positions for FA2
        #     **kwargs,
        # )

        attn_output = attn_output.reshape(bsz, q_len, -1).contiguous()
        attn_output = self.o_proj(attn_output)
        return attn_output


class Qwen2_5_VLDecoderLayer(nn.Module):
    def __init__(self, config: Qwen2_5_VLTextConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size

        if (
            config.use_sliding_window
            and config._attn_implementation != "flash_attention_2"
        ):
            logger.warning(
                f"Sliding Window Attention is enabled but not implemented for `{config._attn_implementation}`; "
                "unexpected results may be encountered."
            )
        self.self_attn = Qwen2_5_VLAttention(config, layer_idx)

        self.mlp = Qwen2MLP(config)
        self.input_layernorm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = Qwen2RMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )
        self.attention_type = config.layer_types[layer_idx]

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[
            tuple[torch.Tensor, torch.Tensor]
        ] = None,  # necessary, but kept here for BC
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> tuple[
        torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]
    ]:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
                `(batch, sequence_length)` where padding elements are indicated by 0.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            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`).
            past_key_values (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
                Indices depicting the position of the input sequence tokens in the sequence.
            position_embeddings (`tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
                Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`,
                with `head_dim` being the embedding dimension of each attention head.
            kwargs (`dict`, *optional*):
                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
                into the model
        """

        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            output_attentions=output_attentions,
            use_cache=use_cache,
            cache_position=cache_position,
            position_embeddings=position_embeddings,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        return hidden_states


class Qwen2_5_VLMLP(nn.Module):
    def __init__(
        self,
        in_features: int,
        hidden_features: int = None,
        bias: bool = True,
        hidden_act="silu",
        quant_config: Optional[QuantizationConfig] = None,
        prefix: str = "",
    ):
        super().__init__()
        self.gate_up_proj = MergedColumnParallelLinear(
            input_size=in_features,
            output_sizes=[hidden_features] * 2,  # [gate_proj, up_proj]
            bias=bias,
            quant_config=quant_config,
            prefix=add_prefix("gate_up_proj", prefix),
        )
        self.down_proj = RowParallelLinear(
            hidden_features,
            in_features,
            bias=bias,
            quant_config=quant_config,
            prefix=add_prefix("down_proj", prefix),
        )
        self.act = ACT2FN[hidden_act]

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        gate_up, _ = self.gate_up_proj(x)
        gate, up = gate_up.chunk(2, dim=-1)
        x = self.act(gate) * up
        x_down, _ = self.down_proj(x)
        return x_down


class Qwen2_5_VLTextModel(nn.Module):
    def __init__(self, config: PretrainedConfig):
        super().__init__()
        self.config = config
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(
            config.vocab_size, config.hidden_size, self.padding_idx
        )
        self.layers = nn.ModuleList(
            [
                Qwen2_5_VLDecoderLayer(config, layer_idx)
                for layer_idx in range(config.num_hidden_layers)
            ]
        )
        self._attn_implementation = config._attn_implementation
        self.norm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = Qwen2_5_VLRotaryEmbedding(config=config)
        self.has_sliding_layers = "sliding_attention" in self.config.layer_types

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

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Union[tuple, BaseModelOutputWithPast]:
        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 None) ^ (inputs_embeds is not None):
            raise ValueError(
                "You must specify exactly one of input_ids or inputs_embeds"
            )

        # torch.jit.trace() doesn't support cache objects in the output
        if use_cache and past_key_values is None and not torch.jit.is_tracing():
            past_key_values = DynamicCache(config=self.config)

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)

        if cache_position is None:
            past_seen_tokens = (
                past_key_values.get_seq_length() if past_key_values is not None else 0
            )
            cache_position = torch.arange(
                past_seen_tokens,
                past_seen_tokens + inputs_embeds.shape[1],
                device=inputs_embeds.device,
            )

        # the hard coded `3` is for temporal, height and width.
        if position_ids is None:
            position_ids = cache_position.view(1, 1, -1).expand(
                3, inputs_embeds.shape[0], -1
            )
        elif position_ids.ndim == 2:
            position_ids = position_ids[None, ...].expand(3, position_ids.shape[0], -1)

        # NOTE: we need to pass text position ids for packing. Qwen2-VL uses 3D positions
        # where each dim indicates visual spatial positions for temporal/height/width grids.
        # There are two scenarios when FA2-like packed masking might be activated.
        # 1. User specifically passed packed `position_ids` and no attention mask.
        #    In this case we expect the user to create correct position ids for all 3 grids
        #    and prepend text-only position ids to it. The final tensor will be [4, bs, seq-len]
        # 2. User runs forward with no attention mask and no position ids. In this case, position ids
        #    are prepared by the model (`get_rope_index`) as `[4, bs, seq-len]` tensor. Text-only positions are
        #    prepended by us when creating positions so that the mask is constructed correctly. NOTE: failing to pass
        #    text-only positions will cause incorrect mask construction, do not change `prepare_input_for_generation`
        if position_ids.ndim == 3 and position_ids.shape[0] == 4:
            text_position_ids = position_ids[0]
            position_ids = position_ids[1:]
        else:
            text_position_ids = position_ids[0]

        # It may already have been prepared by e.g. `generate`
        if not isinstance(causal_mask_mapping := attention_mask, dict):
            # Prepare mask arguments
            mask_kwargs = {
                "config": self.config,
                "input_embeds": inputs_embeds,
                "attention_mask": attention_mask,
                "cache_position": cache_position,
                "past_key_values": past_key_values,
                "position_ids": text_position_ids,
            }
            # Create the masks
            causal_mask_mapping = {
                "full_attention": create_causal_mask(**mask_kwargs),
            }
            # The sliding window alternating layers are not always activated depending on the config
            if self.has_sliding_layers:
                causal_mask_mapping["sliding_attention"] = (
                    create_sliding_window_causal_mask(**mask_kwargs)
                )

        hidden_states = inputs_embeds

        # create position embeddings to be shared across the decoder layers
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None

        for decoder_layer in self.layers:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            hidden_states = decoder_layer(
                hidden_states,
                attention_mask=causal_mask_mapping[decoder_layer.attention_type],
                position_ids=text_position_ids,
                past_key_values=past_key_values,
                output_attentions=output_attentions,
                use_cache=use_cache,
                cache_position=cache_position,
                position_embeddings=position_embeddings,
                **kwargs,
            )

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_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_attns,
                ]
                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_attns,
        )


class Qwen2_5_VLModel(nn.Module):
    base_model_prefix = ""
    _checkpoint_conversion_mapping = {"^model": "language_model"}
    # Reference: fix gemma3 grad acc #37208
    accepts_loss_kwargs = False
    _no_split_modules = ["Qwen2_5_VLDecoderLayer", "Qwen2_5_VLVisionBlock"]

    def __init__(self, config, enable_image_understanding: bool = False):
        super().__init__()
        self.language_model = Qwen2_5_VLTextModel(config.text_config)

        if enable_image_understanding:
            self.visual = Qwen2_5_VisionTransformerPretrainedModel._from_config(
                config.vision_config
            )
            self.visual.to(torch.get_default_dtype())
        self.rope_deltas = None  # cache rope_deltas here
        self.config = config
        # Initialize weights and apply final processing
        # self.post_init()

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

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

    def set_decoder(self, decoder):
        self.language_model = decoder

    def get_decoder(self):
        return self.language_model

    def get_rope_index(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        image_grid_thw: Optional[torch.LongTensor] = None,
        video_grid_thw: Optional[torch.LongTensor] = None,
        second_per_grid_ts: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        """
        Calculate the 3D rope index based on image and video's temporal, height and width in LLM.

        Explanation:
            Each embedding sequence contains vision embedding and text embedding or just contains text embedding.

            For pure text embedding sequence, the rotary position embedding has no difference with modern LLMs.
            Examples:
                input_ids: [T T T T T], here T is for text.
                temporal position_ids: [0, 1, 2, 3, 4]
                height position_ids: [0, 1, 2, 3, 4]
                width position_ids: [0, 1, 2, 3, 4]

            For vision and text embedding sequence, we calculate 3D rotary position embedding for vision part
            and 1D rotary position embedding for text part.
            Examples:
                Temporal (Time): 3 patches, representing different segments of the video in time.
                Height: 2 patches, dividing each frame vertically.
                Width: 2 patches, dividing each frame horizontally.
                We also have some important parameters:
                fps (Frames Per Second): The video's frame rate, set to 1. This means one frame is processed each second.
                tokens_per_second: This is a crucial parameter. It dictates how many "time-steps" or "temporal tokens" are conceptually packed into a one-second interval of the video. In this case, we have 25 tokens per second. So each second of the video will be represented with 25 separate time points. It essentially defines the temporal granularity.
                temporal_patch_size: The number of frames that compose one temporal patch. Here, it's 2 frames.
                interval: The step size for the temporal position IDs, calculated as tokens_per_second * temporal_patch_size / fps. In this case, 25 * 2 / 1 = 50. This means that each temporal patch will be have a difference of 50 in the temporal position IDs.
                input_ids: [V V V V V V V V V V V V T T T T T], here V is for vision.
                vision temporal position_ids: [0, 0, 0, 0, 50, 50, 50, 50, 100, 100, 100, 100]
                vision height position_ids: [0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1]
                vision width position_ids: [0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1]
                text temporal position_ids: [101, 102, 103, 104, 105]
                text height position_ids: [101, 102, 103, 104, 105]
                text width position_ids: [101, 102, 103, 104, 105]
                Here we calculate the text start position_ids as the max vision position_ids plus 1.

        Args:
            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
                Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
                it.
            image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
                The temporal, height and width of feature shape of each image in LLM.
            video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
                The temporal, height and width of feature shape of each video in LLM.
            second_per_grid_ts (`torch.Tensor` of shape `(num_videos)`, *optional*):
                The time interval (in seconds) for each grid along the temporal dimension in the 3D position 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**.

        Returns:
            position_ids (`torch.LongTensor` of shape `(3, batch_size, sequence_length)`)
            mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
        """
        spatial_merge_size = self.config.vision_config.spatial_merge_size
        image_token_id = self.config.image_token_id
        video_token_id = self.config.video_token_id
        vision_start_token_id = self.config.vision_start_token_id
        mrope_position_deltas = []
        if input_ids is not None and (
            image_grid_thw is not None or video_grid_thw is not None
        ):
            total_input_ids = input_ids
            if attention_mask is None:
                attention_mask = torch.ones_like(total_input_ids)
            position_ids = torch.ones(
                3,
                input_ids.shape[0],
                input_ids.shape[1],
                dtype=input_ids.dtype,
                device=input_ids.device,
            )
            image_index, video_index = 0, 0
            attention_mask = attention_mask.to(total_input_ids.device)
            for i, input_ids in enumerate(total_input_ids):
                input_ids = input_ids[attention_mask[i] == 1]
                image_nums, video_nums = 0, 0
                vision_start_indices = torch.argwhere(
                    input_ids == vision_start_token_id
                ).squeeze(1)
                vision_tokens = input_ids[vision_start_indices + 1]
                image_nums = (vision_tokens == image_token_id).sum()
                video_nums = (vision_tokens == video_token_id).sum()
                input_tokens = input_ids.tolist()
                llm_pos_ids_list: list = []
                st = 0
                remain_images, remain_videos = image_nums, video_nums
                for _ in range(image_nums + video_nums):
                    if image_token_id in input_tokens and remain_images > 0:
                        ed_image = input_tokens.index(image_token_id, st)
                    else:
                        ed_image = len(input_tokens) + 1
                    if video_token_id in input_tokens and remain_videos > 0:
                        ed_video = input_tokens.index(video_token_id, st)
                    else:
                        ed_video = len(input_tokens) + 1
                    if ed_image < ed_video:
                        t, h, w = (
                            image_grid_thw[image_index][0],
                            image_grid_thw[image_index][1],
                            image_grid_thw[image_index][2],
                        )
                        second_per_grid_t = 0
                        image_index += 1
                        remain_images -= 1
                        ed = ed_image

                    else:
                        t, h, w = (
                            video_grid_thw[video_index][0],
                            video_grid_thw[video_index][1],
                            video_grid_thw[video_index][2],
                        )
                        if second_per_grid_ts is not None:
                            second_per_grid_t = second_per_grid_ts[video_index]
                        else:
                            second_per_grid_t = 1.0
                        video_index += 1
                        remain_videos -= 1
                        ed = ed_video
                    llm_grid_t, llm_grid_h, llm_grid_w = (
                        t.item(),
                        h.item() // spatial_merge_size,
                        w.item() // spatial_merge_size,
                    )
                    text_len = ed - st

                    st_idx = (
                        llm_pos_ids_list[-1].max() + 1
                        if len(llm_pos_ids_list) > 0
                        else 0
                    )
                    llm_pos_ids_list.append(
                        torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx
                    )

                    range_tensor = torch.arange(llm_grid_t).view(-1, 1)
                    expanded_range = range_tensor.expand(-1, llm_grid_h * llm_grid_w)

                    ## normalize type, send to device.
                    second_per_grid_t = torch.as_tensor(
                        second_per_grid_t,
                        dtype=range_tensor.dtype,
                        device=range_tensor.device,
                    )

                    time_tensor = (
                        expanded_range
                        * second_per_grid_t
                        * self.config.vision_config.tokens_per_second
                    )

                    time_tensor_long = time_tensor.long()
                    t_index = time_tensor_long.flatten()

                    h_index = (
                        torch.arange(llm_grid_h)
                        .view(1, -1, 1)
                        .expand(llm_grid_t, -1, llm_grid_w)
                        .flatten()
                    )
                    w_index = (
                        torch.arange(llm_grid_w)
                        .view(1, 1, -1)
                        .expand(llm_grid_t, llm_grid_h, -1)
                        .flatten()
                    )
                    llm_pos_ids_list.append(
                        torch.stack([t_index, h_index, w_index]) + text_len + st_idx
                    )
                    st = ed + llm_grid_t * llm_grid_h * llm_grid_w

                if st < len(input_tokens):
                    st_idx = (
                        llm_pos_ids_list[-1].max() + 1
                        if len(llm_pos_ids_list) > 0
                        else 0
                    )
                    text_len = len(input_tokens) - st
                    llm_pos_ids_list.append(
                        torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx
                    )

                llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
                position_ids[..., i, attention_mask[i] == 1] = llm_positions.to(
                    position_ids.device
                )
                mrope_position_deltas.append(
                    llm_positions.max() + 1 - len(total_input_ids[i])
                )
            mrope_position_deltas = torch.tensor(
                mrope_position_deltas, device=input_ids.device
            ).unsqueeze(1)
            return position_ids, mrope_position_deltas
        else:
            if attention_mask is not None:
                position_ids = attention_mask.long().cumsum(-1) - 1
                position_ids.masked_fill_(attention_mask == 0, 1)
                position_ids = (
                    position_ids.unsqueeze(0)
                    .expand(3, -1, -1)
                    .to(attention_mask.device)
                )
                max_position_ids = position_ids.max(0, keepdim=False)[0].max(
                    -1, keepdim=True
                )[0]
                mrope_position_deltas = max_position_ids + 1 - attention_mask.shape[-1]
            else:
                position_ids = (
                    torch.arange(input_ids.shape[1], device=input_ids.device)
                    .view(1, 1, -1)
                    .expand(3, input_ids.shape[0], -1)
                )
                mrope_position_deltas = torch.zeros(
                    [input_ids.shape[0], 1],
                    device=input_ids.device,
                    dtype=input_ids.dtype,
                )

            return position_ids, mrope_position_deltas

    def get_video_features(
        self,
        pixel_values_videos: torch.FloatTensor,
        video_grid_thw: Optional[torch.LongTensor] = None,
    ):
        """
        Encodes videos into continuous embeddings that can be forwarded to the language model.

        Args:
            pixel_values_videos (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)`):
                The tensors corresponding to the input videos.
            video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
                The temporal, height and width of feature shape of each video in LLM.
        """
        pixel_values_videos = pixel_values_videos.type(self.visual.dtype)
        video_embeds = self.visual(pixel_values_videos, grid_thw=video_grid_thw)
        split_sizes = (
            video_grid_thw.prod(-1) // self.visual.spatial_merge_size**2
        ).tolist()
        video_embeds = torch.split(video_embeds, split_sizes)
        return video_embeds

    def get_image_features(
        self,
        pixel_values: torch.FloatTensor,
        image_grid_thw: Optional[torch.LongTensor] = None,
    ):
        """
        Encodes images into continuous embeddings that can be forwarded to the language model.

        Args:
            pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)`):
                The tensors corresponding to the input images.
            image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
                The temporal, height and width of feature shape of each image in LLM.
        """
        pixel_values = pixel_values.type(self.visual.dtype)
        image_embeds = self.visual(pixel_values, grid_thw=image_grid_thw)
        split_sizes = (
            image_grid_thw.prod(-1) // self.visual.spatial_merge_size**2
        ).tolist()
        image_embeds = torch.split(image_embeds, split_sizes)
        return image_embeds

    def get_placeholder_mask(
        self,
        input_ids: torch.LongTensor,
        inputs_embeds: torch.FloatTensor,
        image_features: torch.FloatTensor = None,
        video_features: torch.FloatTensor = None,
    ):
        """
        Obtains multimodal placeholder mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
        equal to the length of multimodal features. If the lengths are different, an error is raised.
        """
        if input_ids is None:
            special_image_mask = inputs_embeds == self.get_input_embeddings()(
                torch.tensor(
                    self.config.image_token_id,
                    dtype=torch.long,
                    device=inputs_embeds.device,
                )
            )
            special_image_mask = special_image_mask.all(-1)
            special_video_mask = inputs_embeds == self.get_input_embeddings()(
                torch.tensor(
                    self.config.video_token_id,
                    dtype=torch.long,
                    device=inputs_embeds.device,
                )
            )
            special_video_mask = special_video_mask.all(-1)
        else:
            special_image_mask = input_ids == self.config.image_token_id
            special_video_mask = input_ids == self.config.video_token_id

        n_image_tokens = special_image_mask.sum()
        special_image_mask = (
            special_image_mask.unsqueeze(-1)
            .expand_as(inputs_embeds)
            .to(inputs_embeds.device)
        )
        if (
            image_features is not None
            and inputs_embeds[special_image_mask].numel() != image_features.numel()
        ):
            raise ValueError(
                f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {image_features.shape[0]}"
            )

        n_video_tokens = special_video_mask.sum()
        special_video_mask = (
            special_video_mask.unsqueeze(-1)
            .expand_as(inputs_embeds)
            .to(inputs_embeds.device)
        )
        if (
            video_features is not None
            and inputs_embeds[special_video_mask].numel() != video_features.numel()
        ):
            raise ValueError(
                f"Videos features and video tokens do not match: tokens: {n_video_tokens}, features {video_features.shape[0]}"
            )

        return special_image_mask, special_video_mask

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        pixel_values: Optional[torch.Tensor] = None,
        pixel_values_videos: Optional[torch.FloatTensor] = None,
        image_grid_thw: Optional[torch.LongTensor] = None,
        video_grid_thw: Optional[torch.LongTensor] = None,
        rope_deltas: Optional[torch.LongTensor] = None,
        cache_position: Optional[torch.LongTensor] = None,
        second_per_grid_ts: Optional[torch.Tensor] = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[tuple, Qwen2_5_VLModelOutputWithPast]:
        r"""
        image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
            The temporal, height and width of feature shape of each image in LLM.
        video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
            The temporal, height and width of feature shape of each video in LLM.
        rope_deltas (`torch.LongTensor` of shape `(batch_size, )`, *optional*):
            The rope index difference between sequence length and multimodal rope.
        second_per_grid_ts (`torch.Tensor` of shape `(num_videos)`, *optional*):
            The time interval (in seconds) for each grid along the temporal dimension in the 3D position IDs.
        """

        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 inputs_embeds is None:
            inputs_embeds = self.get_input_embeddings()(input_ids)

        if pixel_values is not None:
            image_embeds = self.get_image_features(pixel_values, image_grid_thw)
            image_embeds = torch.cat(image_embeds, dim=0).to(
                inputs_embeds.device, inputs_embeds.dtype
            )
            image_mask, _ = self.get_placeholder_mask(
                input_ids, inputs_embeds=inputs_embeds, image_features=image_embeds
            )
            inputs_embeds = inputs_embeds.masked_scatter(image_mask, image_embeds)

        if pixel_values_videos is not None:
            video_embeds = self.get_video_features(pixel_values_videos, video_grid_thw)
            video_embeds = torch.cat(video_embeds, dim=0).to(
                inputs_embeds.device, inputs_embeds.dtype
            )
            _, video_mask = self.get_placeholder_mask(
                input_ids, inputs_embeds=inputs_embeds, video_features=video_embeds
            )
            inputs_embeds = inputs_embeds.masked_scatter(video_mask, video_embeds)

        if position_ids is None:
            # Calculate RoPE index once per generation in the pre-fill stage only.
            # When compiling, we can't check tensor values thus we check only input length
            # It is safe to assume that `length!=1` means we're in pre-fill because compiled
            # models currently cannot do asssisted decoding
            prefill_compiled_stage = is_torchdynamo_compiling() and (
                (input_ids is not None and input_ids.shape[1] != 1)
                or (inputs_embeds is not None and inputs_embeds.shape[1] != 1)
            )
            prefill_noncompiled_stage = not is_torchdynamo_compiling() and (
                (cache_position is not None and cache_position[0] == 0)
                or (past_key_values is None or past_key_values.get_seq_length() == 0)
            )
            if (
                prefill_compiled_stage or prefill_noncompiled_stage
            ) or self.rope_deltas is None:
                position_ids, rope_deltas = self.get_rope_index(
                    input_ids,
                    image_grid_thw,
                    video_grid_thw,
                    second_per_grid_ts=second_per_grid_ts,
                    attention_mask=attention_mask,
                )
                self.rope_deltas = rope_deltas
            else:
                batch_size, seq_length, _ = inputs_embeds.shape
                position_ids = torch.arange(seq_length, device=inputs_embeds.device)
                position_ids = position_ids.view(1, 1, -1).expand(3, batch_size, -1)
                if cache_position is not None:
                    delta = (cache_position[0] + self.rope_deltas).to(
                        inputs_embeds.device
                    )
                else:
                    delta = torch.zeros(
                        (batch_size, seq_length), device=inputs_embeds.device
                    )
                delta = delta.repeat_interleave(batch_size // delta.shape[0], dim=1)
                position_ids += delta.to(position_ids.device)

        outputs = self.language_model(
            input_ids=None,
            position_ids=position_ids,
            attention_mask=attention_mask,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=True,
            cache_position=cache_position,
            **kwargs,
        )

        output = Qwen2_5_VLModelOutputWithPast(
            last_hidden_state=outputs.last_hidden_state,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            rope_deltas=self.rope_deltas,
        )
        return output if return_dict else output.to_tuple()


class Qwen2_5_VLForConditionalGeneration(TextEncoder):
    # BitandBytes specific attributes
    default_bitsandbytes_target_modules = [
        ".gate_up_proj.",
        ".down_proj.",
        ".q_proj.",
        ".k_proj.",
        ".v_proj.",
        ".o_proj.",
    ]
    bitsandbytes_stacked_params_mapping = {
        # shard_name, weight_name, index
        "q_proj": ("qkv_proj", 0),
        "k_proj": ("qkv_proj", 1),
        "v_proj": ("qkv_proj", 2),
        "gate_proj": ("gate_up_proj", 0),
        "up_proj": ("gate_up_proj", 1),
    }

    def __init__(
        self,
        config: Qwen2_5VLConfig,
        quant_config: Optional[QuantizationConfig] = None,
        prefix: str = "",
    ) -> None:
        super().__init__(config)
        enable_image_understanding = config.enable_image_understanding
        config = config.arch_config
        self.model = Qwen2_5_VLModel(
            config, enable_image_understanding=enable_image_understanding
        )
        self.lm_head = nn.Linear(
            config.text_config.hidden_size, config.text_config.vocab_size, bias=False
        )

        self.enable_image_understanding = enable_image_understanding

        self.config = config

    def get_input_embeddings(self):
        return self.model.embed_tokens

    @torch.no_grad()
    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        pixel_values: Optional[torch.Tensor] = None,
        pixel_values_videos: Optional[torch.FloatTensor] = None,
        image_grid_thw: Optional[torch.LongTensor] = None,
        video_grid_thw: Optional[torch.LongTensor] = None,
        rope_deltas: Optional[torch.LongTensor] = None,
        cache_position: Optional[torch.LongTensor] = None,
        second_per_grid_ts: Optional[torch.Tensor] = None,
        logits_to_keep: Union[int, torch.Tensor] = 0,
        **kwargs: Unpack[TransformersKwargs],
    ):
        """Run forward pass for Qwen2_5-VL.

        Args:
            input_ids: Flattened (concatenated) input_ids corresponding to a
                batch.
            positions: Flattened (concatenated) position ids corresponding to a
                batch.
                **NOTE**: If mrope is enabled (default setting for Qwen2-VL
                opensource models), the shape will be `(3, seq_len)`,
                otherwise it will be `(seq_len,).
                (Use input_metadata.mrope_positions to replace it)
        """
        output_attentions = False
        output_hidden_states = (
            output_hidden_states
            if output_hidden_states is not None
            else self.config.output_hidden_states
        )

        outputs = self.model(
            input_ids=input_ids,
            pixel_values=pixel_values,
            pixel_values_videos=pixel_values_videos,
            image_grid_thw=image_grid_thw,
            video_grid_thw=video_grid_thw,
            second_per_grid_ts=second_per_grid_ts,
            position_ids=position_ids,
            attention_mask=attention_mask,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=True,
            cache_position=cache_position,
            **kwargs,
        )

        hidden_states = outputs[0]

        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        slice_indices = (
            slice(-logits_to_keep, None)
            if isinstance(logits_to_keep, int)
            else logits_to_keep
        )
        logits = self.lm_head(hidden_states[:, slice_indices, :])
        return Qwen2_5_VLCausalLMOutputWithPast(
            loss=None,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            rope_deltas=outputs.rope_deltas,
        )

    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
        loaded_params: set[str] = set()

        params_dict = dict(self.named_parameters(remove_duplicate=False))
        for name, loaded_weight in weights:
            if "rotary_emb.inv_freq" in name:
                continue

            name = name.replace("model.", "model.language_model.")
            if "visual." in name:
                if not self.enable_image_understanding:
                    continue
                name = name.replace("visual.", "model.visual.")
            try:
                # Skip loading extra bias for GPTQ models.
                if name.endswith(".bias") and name not in params_dict:
                    continue
                param = params_dict[name]
            except KeyError:
                raise

            weight_loader = getattr(param, "weight_loader", default_weight_loader)
            loaded_weight = loaded_weight.to(param.dtype)
            weight_loader(param, loaded_weight)
            loaded_params.add(name)
        return loaded_params

    def get_embed_and_head(self):
        return self.model.embed_tokens.weight, self.lm_head.weight


EntryClass = Qwen2_5_VLForConditionalGeneration