import pdb

import torch
from typing import Optional
import torch.nn.functional as F
from diffusers.utils import USE_PEFT_BACKEND
import torch.nn as nn
from diffusers.models.attention_processor import Attention


class AttnProcessor(nn.Module):
    r"""
    Default processor for performing attention-related computations.
    """

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

    def __call__(
            self,
            attn: Attention,
            hidden_states: torch.FloatTensor,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            temb: Optional[torch.FloatTensor] = None,
            scale: float = 1.0,
            attn_store=None,
            do_classifier_free_guidance=None,
            enable_cloth_guidance=None
    ) -> torch.Tensor:
        residual = hidden_states

        args = () if USE_PEFT_BACKEND else (scale,)

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states, *args)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states, *args)
        value = attn.to_v(encoder_hidden_states, *args)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states, *args)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states


class REFAttnProcessor(nn.Module):
    def __init__(self, name, type="read"):
        super().__init__()
        self.name = name
        self.type = type

    def __call__(
            self,
            attn: Attention,
            hidden_states: torch.FloatTensor,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            temb: Optional[torch.FloatTensor] = None,
            scale: float = 1.0,
            attn_store=None,
            do_classifier_free_guidance=None,
            enable_cloth_guidance=None
    ) -> torch.Tensor:
        if self.type == "read":
            attn_store[self.name] = hidden_states
        elif self.type == "write":
            ref_hidden_states = attn_store[self.name]
            if do_classifier_free_guidance:
                empty_copy = torch.zeros_like(ref_hidden_states)
                if enable_cloth_guidance:
                    ref_hidden_states = torch.cat([empty_copy, ref_hidden_states, ref_hidden_states])
                else:
                    ref_hidden_states = torch.cat([empty_copy, ref_hidden_states])
            hidden_states = torch.cat([hidden_states, ref_hidden_states], dim=1)
        else:
            raise ValueError("unsupport type")
        residual = hidden_states

        args = () if USE_PEFT_BACKEND else (scale,)

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states, *args)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states, *args)
        value = attn.to_v(encoder_hidden_states, *args)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        if self.type == "write":
            hidden_states, _ = torch.chunk(hidden_states, 2, dim=1)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states, *args)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states


class AttnProcessor2_0(nn.Module):
    r"""
    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
    """

    def __init__(self):
        super().__init__()
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")

    def __call__(
            self,
            attn: Attention,
            hidden_states: torch.FloatTensor,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            temb: Optional[torch.FloatTensor] = None,
            scale: float = 1.0,
            attn_store=None,
            do_classifier_free_guidance=None,
            enable_cloth_guidance=None
    ) -> torch.FloatTensor:
        residual = hidden_states
        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        args = () if USE_PEFT_BACKEND else (scale,)
        query = attn.to_q(hidden_states, *args)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states, *args)
        value = attn.to_v(encoder_hidden_states, *args)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states, *args)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states


class REFAttnProcessor2_0(nn.Module):
    def __init__(self, name, type="read"):
        super().__init__()
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
        self.name = name
        self.type = type

    def __call__(
            self,
            attn: Attention,
            hidden_states: torch.FloatTensor,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            temb: Optional[torch.FloatTensor] = None,
            scale: float = 1.0,
            attn_store=None,
            do_classifier_free_guidance=False,
            enable_cloth_guidance=True
    ) -> torch.FloatTensor:
        if self.type == "read":
            attn_store[self.name] = hidden_states
        elif self.type == "write":
            ref_hidden_states = attn_store[self.name]
            if do_classifier_free_guidance:
                empty_copy = torch.zeros_like(ref_hidden_states)
                if enable_cloth_guidance:
                    ref_hidden_states = torch.cat([empty_copy, ref_hidden_states, ref_hidden_states])
                else:
                    ref_hidden_states = torch.cat([empty_copy, ref_hidden_states])
            hidden_states = torch.cat([hidden_states, ref_hidden_states], dim=1)
        else:
            raise ValueError("unsupport type")
        residual = hidden_states
        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        args = () if USE_PEFT_BACKEND else (scale,)
        query = attn.to_q(hidden_states, *args)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states, *args)
        value = attn.to_v(encoder_hidden_states, *args)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        if self.type == "write":
            hidden_states, _ = torch.chunk(hidden_states, 2, dim=1)
        # linear proj
        hidden_states = attn.to_out[0](hidden_states, *args)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor
        return hidden_states


class REFAnimateDiffAttnProcessor2_0(nn.Module):
    def __init__(self, name, type="read"):
        super().__init__()
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
        self.name = name
        self.type = type

    def __call__(
            self,
            attn: Attention,
            hidden_states: torch.FloatTensor,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            temb: Optional[torch.FloatTensor] = None,
            scale: float = 1.0,
            attn_store=None,
            do_classifier_free_guidance=False,
    ) -> torch.FloatTensor:
        if self.type == "read":
            attn_store[self.name] = hidden_states
        elif self.type == "write":
            ref_hidden_states = attn_store[self.name]
            if do_classifier_free_guidance:
                empty_copy = torch.zeros_like(ref_hidden_states)
                ref_hidden_states = torch.cat([empty_copy, ref_hidden_states, ref_hidden_states])
            if hidden_states.shape[0] % ref_hidden_states.shape[0] != 0:
                raise ValueError("not evenly divisible")
            # ref_hidden_states = ref_hidden_states*1.05
            hidden_states = torch.cat([hidden_states, ref_hidden_states.repeat(hidden_states.shape[0] // ref_hidden_states.shape[0], 1, 1)], dim=1)
        else:
            raise ValueError("unsupport type")
        residual = hidden_states
        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        args = () if USE_PEFT_BACKEND else (scale,)
        query = attn.to_q(hidden_states, *args)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states, *args)
        value = attn.to_v(encoder_hidden_states, *args)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        if self.type == "write":
            hidden_states, _ = torch.chunk(hidden_states, 2, dim=1)
        # linear proj
        hidden_states = attn.to_out[0](hidden_states, *args)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor
        return hidden_states


class IPAttnProcessor(nn.Module):

    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
        super().__init__()

        self.hidden_size = hidden_size
        self.cross_attention_dim = cross_attention_dim
        self.scale = scale
        self.num_tokens = num_tokens

        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)

    def __call__(
            self,
            attn,
            hidden_states,
            encoder_hidden_states=None,
            attention_mask=None,
            temb=None,
            attn_store=None,
            do_classifier_free_guidance=None,
            enable_cloth_guidance=None
    ):
        residual = hidden_states

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        else:
            # get encoder_hidden_states, ip_hidden_states
            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
            encoder_hidden_states, ip_hidden_states = (
                encoder_hidden_states[:, :end_pos, :],
                encoder_hidden_states[:, end_pos:, :],
            )
            if attn.norm_cross:
                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        # for ip-adapter
        ip_key = self.to_k_ip(ip_hidden_states)
        ip_value = self.to_v_ip(ip_hidden_states)

        ip_key = attn.head_to_batch_dim(ip_key)
        ip_value = attn.head_to_batch_dim(ip_value)

        ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
        self.attn_map = ip_attention_probs
        ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
        ip_hidden_states = attn.batch_to_head_dim(ip_hidden_states)

        hidden_states = hidden_states + self.scale * ip_hidden_states

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states


class IPAttnProcessor2_0(torch.nn.Module):

    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
        super().__init__()

        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")

        self.hidden_size = hidden_size
        self.cross_attention_dim = cross_attention_dim
        self.scale = scale
        self.num_tokens = num_tokens

        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)

    def __call__(
            self,
            attn,
            hidden_states,
            encoder_hidden_states=None,
            attention_mask=None,
            temb=None,
            attn_store=None,
            do_classifier_free_guidance=None,
            enable_cloth_guidance=None
    ):
        residual = hidden_states

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        else:
            # get encoder_hidden_states, ip_hidden_states
            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
            encoder_hidden_states, ip_hidden_states = (
                encoder_hidden_states[:, :end_pos, :],
                encoder_hidden_states[:, end_pos:, :],
            )
            if attn.norm_cross:
                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        # for ip-adapter
        ip_key = self.to_k_ip(ip_hidden_states)
        ip_value = self.to_v_ip(ip_hidden_states)

        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        ip_hidden_states = F.scaled_dot_product_attention(
            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
        )
        with torch.no_grad():
            self.attn_map = query @ ip_key.transpose(-2, -1).softmax(dim=-1)
            # print(self.attn_map.shape)

        ip_hidden_states = ip_hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        ip_hidden_states = ip_hidden_states.to(query.dtype)

        hidden_states = hidden_states + self.scale * ip_hidden_states

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states