import torch
import torch.nn as nn
import gc
from typing import Any, Dict, List, Optional, Tuple, Union
from dataclasses import dataclass
from diffusers.utils import BaseOutput, logging, scale_lora_layers, unscale_lora_layers, deprecate, USE_PEFT_BACKEND, is_torch_version
from diffusers.utils.torch_utils import apply_freeu


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name


def seed_everything(seed=42):
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = False


@dataclass
class UNet2DConditionOutput(BaseOutput):
    """
    The output of [`UNet2DConditionModel`].

    Args:
        sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
            The hidden states output conditioned on `encoder_hidden_states` input. Output of last layer of model.
    """

    sample: torch.FloatTensor = None


def forward_unet(
    self,
    sample: torch.Tensor,
    timestep: Union[torch.Tensor, float, int],
    encoder_hidden_states: torch.Tensor,
    class_labels: Optional[torch.Tensor] = None,
    timestep_cond: Optional[torch.Tensor] = None,
    attention_mask: Optional[torch.Tensor] = None,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
    down_block_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
    mid_block_additional_residual: Optional[torch.Tensor] = None,
    down_intrablock_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
    encoder_attention_mask: Optional[torch.Tensor] = None
) -> Union[UNet2DConditionOutput, Tuple]:
    r"""
    The [`UNet2DConditionModel`] forward method.

    Args:
        sample (`torch.Tensor`):
            The noisy input tensor with the following shape `(batch, channel, height, width)`.
        timestep (`torch.Tensor` or `float` or `int`): The number of timesteps to denoise an input.
        encoder_hidden_states (`torch.Tensor`):
            The encoder hidden states with shape `(batch, sequence_length, feature_dim)`.
        class_labels (`torch.Tensor`, *optional*, defaults to `None`):
            Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
        timestep_cond: (`torch.Tensor`, *optional*, defaults to `None`):
            Conditional embeddings for timestep. If provided, the embeddings will be summed with the samples passed
            through the `self.time_embedding` layer to obtain the timestep embeddings.
        attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
            An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask
            is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large
            negative values to the attention scores corresponding to "discard" tokens.
        cross_attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
            `self.processor` in
            [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        added_cond_kwargs: (`dict`, *optional*):
            A kwargs dictionary containing additional embeddings that if specified are added to the embeddings that
            are passed along to the UNet blocks.
        down_block_additional_residuals: (`tuple` of `torch.Tensor`, *optional*):
            A tuple of tensors that if specified are added to the residuals of down unet blocks.
        mid_block_additional_residual: (`torch.Tensor`, *optional*):
            A tensor that if specified is added to the residual of the middle unet block.
        down_intrablock_additional_residuals (`tuple` of `torch.Tensor`, *optional*):
            additional residuals to be added within UNet down blocks, for example from T2I-Adapter side model(s)
        encoder_attention_mask (`torch.Tensor`):
            A cross-attention mask of shape `(batch, sequence_length)` is applied to `encoder_hidden_states`. If
            `True` the mask is kept, otherwise if `False` it is discarded. Mask will be converted into a bias,
            which adds large negative values to the attention scores corresponding to "discard" tokens.
        return_dict (`bool`, *optional*, defaults to `True`):
            Whether or not to return a [`~models.unets.unet_2d_condition.UNet2DConditionOutput`] instead of a plain
            tuple.

    Returns:
        [`~models.unets.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
            If `return_dict` is True, an [`~models.unets.unet_2d_condition.UNet2DConditionOutput`] is returned,
            otherwise a `tuple` is returned where the first element is the sample tensor.
    """
    # By default samples have to be AT least a multiple of the overall upsampling factor.
    # The overall upsampling factor is equal to 2 ** (# num of upsampling layers).
    # However, the upsampling interpolation output size can be forced to fit any upsampling size
    # on the fly if necessary.
    torch.cuda.empty_cache()
    gc.collect()
                
    default_overall_up_factor = 2**self.num_upsamplers

    # upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
    forward_upsample_size = False
    upsample_size = None

    for dim in sample.shape[-2:]:
        if dim % default_overall_up_factor != 0:
            # Forward upsample size to force interpolation output size.
            forward_upsample_size = True
            break

    # ensure attention_mask is a bias, and give it a singleton query_tokens dimension
    # expects mask of shape:
    #   [batch, key_tokens]
    # adds singleton query_tokens dimension:
    #   [batch,                    1, key_tokens]
    # this helps to broadcast it as a bias over attention scores, which will be in one of the following shapes:
    #   [batch,  heads, query_tokens, key_tokens] (e.g. torch sdp attn)
    #   [batch * heads, query_tokens, key_tokens] (e.g. xformers or classic attn)
    if attention_mask is not None:
        # assume that mask is expressed as:
        #   (1 = keep,      0 = discard)
        # convert mask into a bias that can be added to attention scores:
        #       (keep = +0,     discard = -10000.0)
        attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
        attention_mask = attention_mask.unsqueeze(1)

    # convert encoder_attention_mask to a bias the same way we do for attention_mask
    if encoder_attention_mask is not None:
        encoder_attention_mask = (1 - encoder_attention_mask.to(sample.dtype)) * -10000.0
        encoder_attention_mask = encoder_attention_mask.unsqueeze(1)

    # 0. center input if necessary
    if self.config.center_input_sample:
        sample = 2 * sample - 1.0

    # 1. time
    t_emb = self.get_time_embed(sample=sample, timestep=timestep)
    emb = self.time_embedding(t_emb, timestep_cond)
    aug_emb = None

    class_emb = self.get_class_embed(sample=sample, class_labels=class_labels)
    if class_emb is not None:
        if self.config.class_embeddings_concat:
            emb = torch.cat([emb, class_emb], dim=-1)
        else:
            emb = emb + class_emb

    aug_emb = self.get_aug_embed(
        emb=emb, encoder_hidden_states=encoder_hidden_states, added_cond_kwargs=added_cond_kwargs
    )
    if self.config.addition_embed_type == "image_hint":
        aug_emb, hint = aug_emb
        sample = torch.cat([sample, hint], dim=1)

    emb = emb + aug_emb if aug_emb is not None else emb

    if self.time_embed_act is not None:
        emb = self.time_embed_act(emb)

    encoder_hidden_states = self.process_encoder_hidden_states(
        encoder_hidden_states=encoder_hidden_states, added_cond_kwargs=added_cond_kwargs
    )

    # 2. pre-process
    sample = self.conv_in(sample)

    # 2.5 GLIGEN position net
    if cross_attention_kwargs is not None and cross_attention_kwargs.get("gligen", None) is not None:
        cross_attention_kwargs = cross_attention_kwargs.copy()
        gligen_args = cross_attention_kwargs.pop("gligen")
        cross_attention_kwargs["gligen"] = {"objs": self.position_net(**gligen_args)}

    # 3. down
    # we're popping the `scale` instead of getting it because otherwise `scale` will be propagated
    # to the internal blocks and will raise deprecation warnings. this will be confusing for our users.
    if cross_attention_kwargs is not None:
        cross_attention_kwargs = cross_attention_kwargs.copy()
        lora_scale = cross_attention_kwargs.pop("scale", 1.0)
    else:
        lora_scale = 1.0

    if USE_PEFT_BACKEND:
        # weight the lora layers by setting `lora_scale` for each PEFT layer
        scale_lora_layers(self, lora_scale)

    is_controlnet = mid_block_additional_residual is not None and down_block_additional_residuals is not None
    # using new arg down_intrablock_additional_residuals for T2I-Adapters, to distinguish from controlnets
    is_adapter = down_intrablock_additional_residuals is not None
    # maintain backward compatibility for legacy usage, where
    #       T2I-Adapter and ControlNet both use down_block_additional_residuals arg
    #       but can only use one or the other
    if not is_adapter and mid_block_additional_residual is None and down_block_additional_residuals is not None:
        deprecate(
            "T2I should not use down_block_additional_residuals",
            "1.3.0",
            "Passing intrablock residual connections with `down_block_additional_residuals` is deprecated \
                    and will be removed in diffusers 1.3.0.  `down_block_additional_residuals` should only be used \
                    for ControlNet. Please make sure use `down_intrablock_additional_residuals` instead. ",
            standard_warn=False,
        )
        down_intrablock_additional_residuals = down_block_additional_residuals
        is_adapter = True

    down_block_res_samples = (sample,)
    for downsample_block in self.down_blocks:
        if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
            # For t2i-adapter CrossAttnDownBlock2D
            additional_residuals = {}
            if is_adapter and len(down_intrablock_additional_residuals) > 0:
                additional_residuals["additional_residuals"] = down_intrablock_additional_residuals.pop(0)

            sample, res_samples = downsample_block(
                hidden_states=sample,
                temb=emb,
                encoder_hidden_states=encoder_hidden_states,
                attention_mask=attention_mask,
                cross_attention_kwargs=cross_attention_kwargs,
                encoder_attention_mask=encoder_attention_mask,
                **additional_residuals,
            )
        else:
            sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
            if is_adapter and len(down_intrablock_additional_residuals) > 0:
                sample += down_intrablock_additional_residuals.pop(0)

        down_block_res_samples += res_samples

    if is_controlnet:
        new_down_block_res_samples = ()

        for down_block_res_sample, down_block_additional_residual in zip(
            down_block_res_samples, down_block_additional_residuals
        ):
            down_block_res_sample = down_block_res_sample + down_block_additional_residual
            new_down_block_res_samples = new_down_block_res_samples + (down_block_res_sample,)

        down_block_res_samples = new_down_block_res_samples

    # 4. mid
    if self.mid_block is not None:
        if hasattr(self.mid_block, "has_cross_attention") and self.mid_block.has_cross_attention:
            sample = self.mid_block(
                sample,
                emb,
                encoder_hidden_states=encoder_hidden_states,
                attention_mask=attention_mask,
                cross_attention_kwargs=cross_attention_kwargs,
                encoder_attention_mask=encoder_attention_mask,
            )
        else:
            sample = self.mid_block(sample, emb)

        # To support T2I-Adapter-XL
        if (
            is_adapter
            and len(down_intrablock_additional_residuals) > 0
            and sample.shape == down_intrablock_additional_residuals[0].shape
        ):
            sample += down_intrablock_additional_residuals.pop(0)

    if is_controlnet:
        sample = sample + mid_block_additional_residual

    # 5. up
    for i, upsample_block in enumerate(self.up_blocks):
        is_final_block = i == len(self.up_blocks) - 1

        res_samples = down_block_res_samples[-len(upsample_block.resnets):]
        down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]

        # if we have not reached the final block and need to forward the
        # upsample size, we do it here
        if not is_final_block and forward_upsample_size:
            upsample_size = down_block_res_samples[-1].shape[2:]

        if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
            sample = upsample_block(
                hidden_states=sample,
                temb=emb,
                res_hidden_states_tuple=res_samples,
                encoder_hidden_states=encoder_hidden_states,
                cross_attention_kwargs=cross_attention_kwargs,
                upsample_size=upsample_size,
                attention_mask=attention_mask,
                encoder_attention_mask=encoder_attention_mask,
            )
        else:
            sample = upsample_block(
                hidden_states=sample,
                temb=emb,
                res_hidden_states_tuple=res_samples,
                upsample_size=upsample_size,
            )

    # 6. post-process
    if self.conv_norm_out:
        sample = self.conv_norm_out(sample)
        sample = self.conv_act(sample)
    sample = self.conv_out(sample)

    if USE_PEFT_BACKEND:
        # remove `lora_scale` from each PEFT layer
        unscale_lora_layers(self, lora_scale)

    return sample


def chunked_forward_unet(
    self,
    sample: torch.Tensor,
    timestep: Union[torch.Tensor, float, int],
    encoder_hidden_states: torch.Tensor,
    class_labels: Optional[torch.Tensor] = None,
    timestep_cond: Optional[torch.Tensor] = None,
    attention_mask: Optional[torch.Tensor] = None,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
    down_block_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
    mid_block_additional_residual: Optional[torch.Tensor] = None,
    down_intrablock_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
    encoder_attention_mask: Optional[torch.Tensor] = None,
    return_dict: bool = True,
) -> Union[UNet2DConditionOutput, Tuple]:
    sample = torch.cat([forward_unet(
        self,
        sample[idx:idx + 1],
        timestep if not isinstance(timestep, torch.Tensor) or timestep.ndim == 0 else timestep[idx:idx + 1],
        encoder_hidden_states[idx:idx + 1],
        None if class_labels is None else class_labels[idx:idx + 1],
        None if timestep_cond is None else timestep_cond[idx:idx + 1],
        None if attention_mask is None else attention_mask[idx:idx + 1],
        None if cross_attention_kwargs is None else {k: v[idx:idx + 1] 
                                                     for k, v in cross_attention_kwargs.items()},
        None if added_cond_kwargs is None else {
            k: v[idx:idx + 1] for k, v in added_cond_kwargs.items()},
        None if down_block_additional_residuals is None else tuple(
            [item[idx:idx + 1] for item in down_block_additional_residuals]),
        None if mid_block_additional_residual is None else mid_block_additional_residual[idx:idx + 1],
        None if down_intrablock_additional_residuals is None else tuple(
            [item[idx:idx + 1] for item in down_intrablock_additional_residuals]),
        None if encoder_attention_mask is None else encoder_attention_mask[idx:idx + 1]
    ).cpu() for idx in range(len(sample))], dim=0).to(encoder_hidden_states.device)
    if not return_dict:
        return (sample,)

    return UNet2DConditionOutput(sample=sample)


def forward_unet_wrapper(self):
    def _forward(*args, **kwargs):
        return chunked_forward_unet(self, *args, **kwargs)
    return _forward


def forward_resnet(self, input_tensor: torch.Tensor, temb: torch.Tensor, *args, **kwargs) -> torch.Tensor:
    if len(args) > 0 or kwargs.get("scale", None) is not None:
        deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
        deprecate("scale", "1.0.0", deprecation_message)

    if self.time_emb_proj is not None:
        if not self.skip_time_act:
            inplace = self.nonlinearity.inplace
            self.nonlinearity.inplace = False
            temb = self.nonlinearity(temb)
            self.nonlinearity.inplace = inplace
        temb = self.time_emb_proj(temb)[:, :, None, None]

    if self.upsample is not None:
        if temb is not None:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                self.upsample(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor)))) + temb))))
        else:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                self.upsample(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor))))))))
        if self.conv_shortcut is not None:
            input_tensor = self.conv_shortcut(self.upsample(input_tensor))
        else:
            input_tensor = self.upsample(input_tensor)
        
    elif self.downsample is not None:
        if temb is not None:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                self.downsample(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor)))) + temb))))
        else:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                self.downsample(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor))))))))
        if self.conv_shortcut is not None:
            input_tensor = self.conv_shortcut(self.downsample(input_tensor))
        else:
            input_tensor = self.downsample(input_tensor)
        
    else:
        if temb is not None:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor))) + temb))))
        else:
            hidden_states = self.conv2(
                self.dropout(
                    self.nonlinearity(
                        self.norm2(
                            self.conv1(
                                    self.nonlinearity(
                                        self.norm1(
                                            input_tensor)))))))
            
        if self.conv_shortcut is not None:
            input_tensor = self.conv_shortcut(input_tensor)

    return (input_tensor + hidden_states) / self.output_scale_factor


def forward_resnet_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_resnet(self, *args, **kwargs)
    return _forward


def forward_upblock2d(
    self,
    hidden_states: torch.Tensor,
    res_hidden_states_tuple: Tuple[torch.Tensor, ...],
    temb: Optional[torch.Tensor] = None,
    upsample_size: Optional[int] = None,
    *args,
    **kwargs,
) -> torch.Tensor:
    if len(args) > 0 or kwargs.get("scale", None) is not None:
        deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
        deprecate("scale", "1.0.0", deprecation_message)

    is_freeu_enabled = (
        getattr(self, "s1", None)
        and getattr(self, "s2", None)
        and getattr(self, "b1", None)
        and getattr(self, "b2", None)
    )

    for resnet in self.resnets:
        # pop res hidden states
        res_hidden_states = res_hidden_states_tuple[-1].to(hidden_states.device)
        res_hidden_states_tuple = res_hidden_states_tuple[:-1]

        # FreeU: Only operate on the first two stages
        if is_freeu_enabled:
            hidden_states, res_hidden_states = apply_freeu(
                self.resolution_idx,
                hidden_states,
                res_hidden_states,
                s1=self.s1,
                s2=self.s2,
                b1=self.b1,
                b2=self.b2,
            )

        hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
        del res_hidden_states

        if self.training and self.gradient_checkpointing:

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

                return custom_forward

            if is_torch_version(">=", "1.11.0"):
                hidden_states = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(resnet), hidden_states, temb, use_reentrant=False
                )
            else:
                hidden_states = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(resnet), hidden_states, temb
                )
        else:
            hidden_states = resnet(hidden_states, temb)

    if self.upsamplers is not None:
        for upsampler in self.upsamplers:
            hidden_states = upsampler(hidden_states, upsample_size)

    return hidden_states


def forward_upblock2d_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_upblock2d(self, *args, **kwargs)
    return _forward


def forward_crossattnupblock2d(
    self,
    hidden_states: torch.Tensor,
    res_hidden_states_tuple: Tuple[torch.Tensor, ...],
    temb: Optional[torch.Tensor] = None,
    encoder_hidden_states: Optional[torch.Tensor] = None,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    upsample_size: Optional[int] = None,
    attention_mask: Optional[torch.Tensor] = None,
    encoder_attention_mask: Optional[torch.Tensor] = None,
) -> torch.Tensor:
    if cross_attention_kwargs is not None:
        if cross_attention_kwargs.get("scale", None) is not None:
            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")

    is_freeu_enabled = (
        getattr(self, "s1", None)
        and getattr(self, "s2", None)
        and getattr(self, "b1", None)
        and getattr(self, "b2", None)
    )

    for resnet, attn in zip(self.resnets, self.attentions):
        # pop res hidden states
        res_hidden_states = res_hidden_states_tuple[-1].to(hidden_states.device)
        res_hidden_states_tuple = res_hidden_states_tuple[:-1]

        # FreeU: Only operate on the first two stages
        if is_freeu_enabled:
            hidden_states, res_hidden_states = apply_freeu(
                self.resolution_idx,
                hidden_states,
                res_hidden_states,
                s1=self.s1,
                s2=self.s2,
                b1=self.b1,
                b2=self.b2,
            )

        hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
        del res_hidden_states

        if self.training and self.gradient_checkpointing:

            def create_custom_forward(module, return_dict=None):
                def custom_forward(*inputs):
                    if return_dict is not None:
                        return module(*inputs, return_dict=return_dict)
                    else:
                        return module(*inputs)

                return custom_forward

            ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
            hidden_states = torch.utils.checkpoint.checkpoint(
                create_custom_forward(resnet),
                hidden_states,
                temb,
                **ckpt_kwargs,
            )
            hidden_states = attn(
                hidden_states,
                encoder_hidden_states=encoder_hidden_states,
                cross_attention_kwargs=cross_attention_kwargs,
                attention_mask=attention_mask,
                encoder_attention_mask=encoder_attention_mask,
                return_dict=False,
            )[0]
        else:
            hidden_states = resnet(hidden_states, temb)
            hidden_states = attn(
                hidden_states,
                encoder_hidden_states=encoder_hidden_states,
                cross_attention_kwargs=cross_attention_kwargs,
                attention_mask=attention_mask,
                encoder_attention_mask=encoder_attention_mask,
                return_dict=False,
            )[0]

    if self.upsamplers is not None:
        for upsampler in self.upsamplers:
            hidden_states = upsampler(hidden_states, upsample_size)

    return hidden_states


def forward_crossattnupblock2d_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_crossattnupblock2d(self, *args, **kwargs)
    return _forward


def forward_downblock2d(
    self, hidden_states: torch.Tensor, temb: Optional[torch.Tensor] = None, *args, **kwargs
) -> Tuple[torch.Tensor, Tuple[torch.Tensor, ...]]:
    if len(args) > 0 or kwargs.get("scale", None) is not None:
        deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
        deprecate("scale", "1.0.0", deprecation_message)

    output_states = ()

    for resnet in self.resnets:
        if self.training and self.gradient_checkpointing:

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

                return custom_forward

            if is_torch_version(">=", "1.11.0"):
                hidden_states = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(resnet), hidden_states, temb, use_reentrant=False
                )
            else:
                hidden_states = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(resnet), hidden_states, temb
                )
        else:
            hidden_states = resnet(hidden_states, temb)

        output_states = output_states + (hidden_states.cpu(),)

    if self.downsamplers is not None:
        for downsampler in self.downsamplers:
            hidden_states = downsampler(hidden_states)

        output_states = output_states + (hidden_states.cpu(),)

    return hidden_states, output_states


def forward_downblock2d_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_downblock2d(self, *args, **kwargs)
    return _forward


def forward_crossattndownblock2d(
    self,
    hidden_states: torch.Tensor,
    temb: Optional[torch.Tensor] = None,
    encoder_hidden_states: Optional[torch.Tensor] = None,
    attention_mask: Optional[torch.Tensor] = None,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    encoder_attention_mask: Optional[torch.Tensor] = None,
    additional_residuals: Optional[torch.Tensor] = None,
) -> Tuple[torch.Tensor, Tuple[torch.Tensor, ...]]:
    if cross_attention_kwargs is not None:
        if cross_attention_kwargs.get("scale", None) is not None:
            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")

    output_states = ()

    blocks = list(zip(self.resnets, self.attentions))

    for i, (resnet, attn) in enumerate(blocks):
        if self.training and self.gradient_checkpointing:

            def create_custom_forward(module, return_dict=None):
                def custom_forward(*inputs):
                    if return_dict is not None:
                        return module(*inputs, return_dict=return_dict)
                    else:
                        return module(*inputs)

                return custom_forward

            ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
            hidden_states = torch.utils.checkpoint.checkpoint(
                create_custom_forward(resnet),
                hidden_states,
                temb,
                **ckpt_kwargs,
            )
            hidden_states = attn(
                hidden_states,
                encoder_hidden_states=encoder_hidden_states,
                cross_attention_kwargs=cross_attention_kwargs,
                attention_mask=attention_mask,
                encoder_attention_mask=encoder_attention_mask,
                return_dict=False,
            )[0]
        else:
            hidden_states = resnet(hidden_states, temb)
            hidden_states = attn(
                hidden_states,
                encoder_hidden_states=encoder_hidden_states,
                cross_attention_kwargs=cross_attention_kwargs,
                attention_mask=attention_mask,
                encoder_attention_mask=encoder_attention_mask,
                return_dict=False,
            )[0]

        # apply additional residuals to the output of the last pair of resnet and attention blocks
        if i == len(blocks) - 1 and additional_residuals is not None:
            hidden_states = hidden_states + additional_residuals

        output_states = output_states + (hidden_states.cpu(),)

    if self.downsamplers is not None:
        for downsampler in self.downsamplers:
            hidden_states = downsampler(hidden_states)

        output_states = output_states + (hidden_states.cpu(),)

    return hidden_states, output_states


def forward_crossattndownblock2d_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_crossattndownblock2d(self, *args, **kwargs)
    return _forward


def _chunked_feed_forward(ff: nn.Module, hidden_states: torch.Tensor, chunk_dim: int, chunk_size: int):
    # "feed_forward_chunk_size" can be used to save memory
    if hidden_states.shape[chunk_dim] % chunk_size != 0:
        raise ValueError(
            f"`hidden_states` dimension to be chunked: {hidden_states.shape[chunk_dim]} has to be divisible by chunk size: {chunk_size}. Make sure to set an appropriate `chunk_size` when calling `unet.enable_forward_chunking`."
        )

    num_chunks = chunk_size
    ff_output = torch.cat(
        [ff(hid_slice) for hid_slice in hidden_states.chunk(num_chunks, dim=chunk_dim)],
        dim=chunk_dim,
    )
    return ff_output


def forward_transformer_block(
    self,
    hidden_states: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    encoder_hidden_states: Optional[torch.Tensor] = None,
    encoder_attention_mask: Optional[torch.Tensor] = None,
    timestep: Optional[torch.LongTensor] = None,
    cross_attention_kwargs: Dict[str, Any] = None,
    class_labels: Optional[torch.LongTensor] = None,
    added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
) -> torch.Tensor:
    if cross_attention_kwargs is not None:
        if cross_attention_kwargs.get("scale", None) is not None:
            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")

    # Notice that normalization is always applied before the real computation in the following blocks.
    # 0. Self-Attention
    batch_size = hidden_states.shape[0]

    if self.norm_type == "ada_norm":
        norm_hidden_states = self.norm1(hidden_states, timestep)
    elif self.norm_type == "ada_norm_zero":
        norm_hidden_states, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.norm1(
            hidden_states, timestep, class_labels, hidden_dtype=hidden_states.dtype
        )
    elif self.norm_type in ["layer_norm", "layer_norm_i2vgen"]:
        norm_hidden_states = self.norm1(hidden_states)
    elif self.norm_type == "ada_norm_continuous":
        norm_hidden_states = self.norm1(hidden_states, added_cond_kwargs["pooled_text_emb"])
    elif self.norm_type == "ada_norm_single":
        shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
            self.scale_shift_table[None] + timestep.reshape(batch_size, 6, -1)
        ).chunk(6, dim=1)
        norm_hidden_states = self.norm1(hidden_states)
        norm_hidden_states = norm_hidden_states * (1 + scale_msa) + shift_msa
        norm_hidden_states = norm_hidden_states.squeeze(1)
    else:
        raise ValueError("Incorrect norm used")

    if self.pos_embed is not None:
        norm_hidden_states = self.pos_embed(norm_hidden_states)

    # 1. Prepare GLIGEN inputs
    cross_attention_kwargs = cross_attention_kwargs.copy() if cross_attention_kwargs is not None else {}
    gligen_kwargs = cross_attention_kwargs.pop("gligen", None)

    attn_output = self.attn1(
        norm_hidden_states,
        encoder_hidden_states=encoder_hidden_states if self.only_cross_attention else None,
        attention_mask=attention_mask,
        **cross_attention_kwargs,
    )
    if self.norm_type == "ada_norm_zero":
        attn_output = gate_msa.unsqueeze(1) * attn_output
    elif self.norm_type == "ada_norm_single":
        attn_output = gate_msa * attn_output

    hidden_states = attn_output + hidden_states
    if hidden_states.ndim == 4:
        hidden_states = hidden_states.squeeze(1)

    # 1.2 GLIGEN Control
    if gligen_kwargs is not None:
        hidden_states = self.fuser(hidden_states, gligen_kwargs["objs"])

    # 3. Cross-Attention
    if self.attn2 is not None:
        if self.norm_type == "ada_norm":
            norm_hidden_states = self.norm2(hidden_states, timestep)
        elif self.norm_type in ["ada_norm_zero", "layer_norm", "layer_norm_i2vgen"]:
            norm_hidden_states = self.norm2(hidden_states)
        elif self.norm_type == "ada_norm_single":
            # For PixArt norm2 isn't applied here:
            # https://github.com/PixArt-alpha/PixArt-alpha/blob/0f55e922376d8b797edd44d25d0e7464b260dcab/diffusion/model/nets/PixArtMS.py#L70C1-L76C103
            norm_hidden_states = hidden_states
        elif self.norm_type == "ada_norm_continuous":
            norm_hidden_states = self.norm2(hidden_states, added_cond_kwargs["pooled_text_emb"])
        else:
            raise ValueError("Incorrect norm")

        if self.pos_embed is not None and self.norm_type != "ada_norm_single":
            norm_hidden_states = self.pos_embed(norm_hidden_states)

        attn_output = self.attn2(
            norm_hidden_states,
            encoder_hidden_states=encoder_hidden_states,
            attention_mask=encoder_attention_mask,
            **cross_attention_kwargs,
        )
        hidden_states = attn_output + hidden_states

    # 4. Feed-forward
    # i2vgen doesn't have this norm 🤷‍♂️
    if self.norm_type == "ada_norm_continuous":
        norm_hidden_states = self.norm3(hidden_states, added_cond_kwargs["pooled_text_emb"])
    elif not self.norm_type == "ada_norm_single":
        norm_hidden_states = self.norm3(hidden_states)
    
    if self.norm_type == "ada_norm_zero":
        norm_hidden_states = norm_hidden_states * (1 + scale_mlp[:, None]) + shift_mlp[:, None]

    if self.norm_type == "ada_norm_single":
        norm_hidden_states = self.norm2(hidden_states)
        norm_hidden_states = norm_hidden_states * (1 + scale_mlp) + shift_mlp

    if self._chunk_size is not None:
        # "feed_forward_chunk_size" can be used to save memory
        ff_output = _chunked_feed_forward(self.ff, norm_hidden_states, self._chunk_dim, self._chunk_size)
    else:
        ff_output = self.ff(norm_hidden_states)
    
    if self.norm_type == "ada_norm_zero":
        ff_output = gate_mlp.unsqueeze(1) * ff_output
    elif self.norm_type == "ada_norm_single":
        ff_output = gate_mlp * ff_output

    hidden_states = ff_output + hidden_states
    if hidden_states.ndim == 4:
        hidden_states = hidden_states.squeeze(1)

    return hidden_states


def forward_transformer_block_wrapper(self):
    def _forward(*args, **kwargs):
        return forward_transformer_block(self, *args, **kwargs)
    return _forward