motion_module.py

from typing import Optional

import torch
import torch.nn.functional as F
from torch import nn

from ldm.modules.attention import FeedForward

from einops import rearrange, repeat
import math


def zero_module(module):
    # Zero out the parameters of a module and return it.
    for p in module.parameters():
        p.detach().zero_()
    return module


class MotionWrapper(nn.Module):
    def __init__(self, mm_type="mm_sd_v15.ckpt"):
        super().__init__()
        self.down_blocks = nn.ModuleList([])
        self.up_blocks = nn.ModuleList([])
        for c in (320, 640, 1280, 1280):
            self.down_blocks.append(MotionModule(c))
        for c in (1280, 1280, 640, 320):
            self.up_blocks.append(MotionModule(c, is_up=True))
        self.mm_type = mm_type


class MotionModule(nn.Module):
    def __init__(self, in_channels, is_up=False):
        super().__init__()
        self.motion_modules = nn.ModuleList([get_motion_module(in_channels), get_motion_module(in_channels)])
        if is_up:
            self.motion_modules.append(get_motion_module(in_channels))


def get_motion_module(in_channels):
    return VanillaTemporalModule(in_channels=in_channels)


class VanillaTemporalModule(nn.Module):
    def __init__(
        self,
        in_channels,
        num_attention_heads                = 8,
        num_transformer_block              = 1,
        attention_block_types              =( "Temporal_Self", "Temporal_Self" ),
        cross_frame_attention_mode         = None,
        temporal_position_encoding         = True,
        temporal_position_encoding_max_len = 24,
        temporal_attention_dim_div         = 1,
        zero_initialize                    = True,
    ):
        super().__init__()
        
        self.temporal_transformer = TemporalTransformer3DModel(
            in_channels=in_channels,
            num_attention_heads=num_attention_heads,
            attention_head_dim=in_channels // num_attention_heads // temporal_attention_dim_div,
            num_layers=num_transformer_block,
            attention_block_types=attention_block_types,
            cross_frame_attention_mode=cross_frame_attention_mode,
            temporal_position_encoding=temporal_position_encoding,
            temporal_position_encoding_max_len=temporal_position_encoding_max_len,
        )
        
        if zero_initialize:
            self.temporal_transformer.proj_out = zero_module(self.temporal_transformer.proj_out)


    def forward(self, input_tensor, encoder_hidden_states, attention_mask=None):
        return self.temporal_transformer(input_tensor, encoder_hidden_states, attention_mask)


class TemporalTransformer3DModel(nn.Module):
    def __init__(
        self,
        in_channels,
        num_attention_heads,
        attention_head_dim,

        num_layers,
        attention_block_types              = ( "Temporal_Self", "Temporal_Self", ),        
        dropout                            = 0.0,
        norm_num_groups                    = 32,
        cross_attention_dim                = 768,
        activation_fn                      = "geglu",
        attention_bias                     = False,
        upcast_attention                   = False,
        
        cross_frame_attention_mode         = None,
        temporal_position_encoding         = False,
        temporal_position_encoding_max_len = 24,
    ):
        super().__init__()

        inner_dim = num_attention_heads * attention_head_dim

        self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
        self.proj_in = nn.Linear(in_channels, inner_dim)

        self.transformer_blocks = nn.ModuleList(
            [
                TemporalTransformerBlock(
                    dim=inner_dim,
                    num_attention_heads=num_attention_heads,
                    attention_head_dim=attention_head_dim,
                    attention_block_types=attention_block_types,
                    dropout=dropout,
                    norm_num_groups=norm_num_groups,
                    cross_attention_dim=cross_attention_dim,
                    activation_fn=activation_fn,
                    attention_bias=attention_bias,
                    upcast_attention=upcast_attention,
                    cross_frame_attention_mode=cross_frame_attention_mode,
                    temporal_position_encoding=temporal_position_encoding,
                    temporal_position_encoding_max_len=temporal_position_encoding_max_len,
                )
                for d in range(num_layers)
            ]
        )
        self.proj_out = nn.Linear(inner_dim, in_channels)    
    
    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None):
        video_length = hidden_states.shape[0] // 2 # TODO: config this value in scripts
        batch, channel, height, weight = hidden_states.shape
        residual = hidden_states

        hidden_states = self.norm(hidden_states)
        inner_dim = hidden_states.shape[1]
        hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * weight, inner_dim)
        hidden_states = self.proj_in(hidden_states)

        # Transformer Blocks
        for block in self.transformer_blocks:
            hidden_states = block(hidden_states, encoder_hidden_states=encoder_hidden_states, video_length=video_length)
        
        # output
        hidden_states = self.proj_out(hidden_states)
        hidden_states = hidden_states.reshape(batch, height, weight, inner_dim).permute(0, 3, 1, 2).contiguous()

        output = hidden_states + residual
        return output


class TemporalTransformerBlock(nn.Module):
    def __init__(
        self,
        dim,
        num_attention_heads,
        attention_head_dim,
        attention_block_types              = ( "Temporal_Self", "Temporal_Self", ),
        dropout                            = 0.0,
        norm_num_groups                    = 32,
        cross_attention_dim                = 768,
        activation_fn                      = "geglu",
        attention_bias                     = False,
        upcast_attention                   = False,
        cross_frame_attention_mode         = None,
        temporal_position_encoding         = False,
        temporal_position_encoding_max_len = 24,
    ):
        super().__init__()

        attention_blocks = []
        norms = []
        
        for block_name in attention_block_types:
            attention_blocks.append(
                VersatileAttention(
                    attention_mode=block_name.split("_")[0],
                    cross_attention_dim=cross_attention_dim if block_name.endswith("_Cross") else None,
                    
                    query_dim=dim,
                    heads=num_attention_heads,
                    dim_head=attention_head_dim,
                    dropout=dropout,
                    bias=attention_bias,
                    upcast_attention=upcast_attention,
        
                    cross_frame_attention_mode=cross_frame_attention_mode,
                    temporal_position_encoding=temporal_position_encoding,
                    temporal_position_encoding_max_len=temporal_position_encoding_max_len,
                )
            )
            norms.append(nn.LayerNorm(dim))
            
        self.attention_blocks = nn.ModuleList(attention_blocks)
        self.norms = nn.ModuleList(norms)

        self.ff = FeedForward(dim, dropout=dropout, glu=(activation_fn=='geglu'))
        self.ff_norm = nn.LayerNorm(dim)


    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, video_length=None):
        for attention_block, norm in zip(self.attention_blocks, self.norms):
            norm_hidden_states = norm(hidden_states)
            hidden_states = attention_block(
                norm_hidden_states,
                encoder_hidden_states=encoder_hidden_states if attention_block.is_cross_attention else None,
                video_length=video_length,
            ) + hidden_states
            
        hidden_states = self.ff(self.ff_norm(hidden_states)) + hidden_states
        
        output = hidden_states  
        return output


class PositionalEncoding(nn.Module):
    def __init__(
        self, 
        d_model, 
        dropout = 0., 
        max_len = 24
    ):
        super().__init__()
        self.dropout = nn.Dropout(p=dropout)
        position = torch.arange(max_len).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))
        pe = torch.zeros(1, max_len, d_model)
        pe[0, :, 0::2] = torch.sin(position * div_term)
        pe[0, :, 1::2] = torch.cos(position * div_term)
        self.register_buffer('pe', pe)

    def forward(self, x):
        x = x + self.pe[:, :x.size(1)]
        return self.dropout(x)


class CrossAttention(nn.Module):
    r"""
    A cross attention layer.

    Parameters:
        query_dim (`int`): The number of channels in the query.
        cross_attention_dim (`int`, *optional*):
            The number of channels in the encoder_hidden_states. If not given, defaults to `query_dim`.
        heads (`int`,  *optional*, defaults to 8): The number of heads to use for multi-head attention.
        dim_head (`int`,  *optional*, defaults to 64): The number of channels in each head.
        dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
        bias (`bool`, *optional*, defaults to False):
            Set to `True` for the query, key, and value linear layers to contain a bias parameter.
    """

    def __init__(
        self,
        query_dim: int,
        cross_attention_dim: Optional[int] = None,
        heads: int = 8,
        dim_head: int = 64,
        dropout: float = 0.0,
        bias=False,
        upcast_attention: bool = False,
        upcast_softmax: bool = False,
        added_kv_proj_dim: Optional[int] = None,
        norm_num_groups: Optional[int] = None,
    ):
        super().__init__()
        inner_dim = dim_head * heads
        cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim
        self.upcast_attention = upcast_attention
        self.upcast_softmax = upcast_softmax

        self.scale = dim_head**-0.5

        self.heads = heads
        # for slice_size > 0 the attention score computation
        # is split across the batch axis to save memory
        # You can set slice_size with `set_attention_slice`
        self.sliceable_head_dim = heads
        self._slice_size = None
        # self._use_memory_efficient_attention_xformers = shared.xformers_available
        self._use_memory_efficient_attention_xformers = False
        self.added_kv_proj_dim = added_kv_proj_dim

        if norm_num_groups is not None:
            self.group_norm = nn.GroupNorm(num_channels=inner_dim, num_groups=norm_num_groups, eps=1e-5, affine=True)
        else:
            self.group_norm = None

        self.to_q = nn.Linear(query_dim, inner_dim, bias=bias)
        self.to_k = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
        self.to_v = nn.Linear(cross_attention_dim, inner_dim, bias=bias)

        if self.added_kv_proj_dim is not None:
            self.add_k_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)
            self.add_v_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)

        self.to_out = nn.ModuleList([])
        self.to_out.append(nn.Linear(inner_dim, query_dim))
        self.to_out.append(nn.Dropout(dropout))

    def reshape_heads_to_batch_dim(self, tensor):
        batch_size, seq_len, dim = tensor.shape
        head_size = self.heads
        tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
        tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size * head_size, seq_len, dim // head_size)
        return tensor

    def reshape_batch_dim_to_heads(self, tensor):
        batch_size, seq_len, dim = tensor.shape
        head_size = self.heads
        tensor = tensor.reshape(batch_size // head_size, head_size, seq_len, dim)
        tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size // head_size, seq_len, dim * head_size)
        return tensor

    def set_attention_slice(self, slice_size):
        if slice_size is not None and slice_size > self.sliceable_head_dim:
            raise ValueError(f"slice_size {slice_size} has to be smaller or equal to {self.sliceable_head_dim}.")

        self._slice_size = slice_size

    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None):
        batch_size, sequence_length, _ = hidden_states.shape

        encoder_hidden_states = encoder_hidden_states

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

        query = self.to_q(hidden_states)
        dim = query.shape[-1]
        query = self.reshape_heads_to_batch_dim(query)

        if self.added_kv_proj_dim is not None:
            key = self.to_k(hidden_states)
            value = self.to_v(hidden_states)
            encoder_hidden_states_key_proj = self.add_k_proj(encoder_hidden_states)
            encoder_hidden_states_value_proj = self.add_v_proj(encoder_hidden_states)

            key = self.reshape_heads_to_batch_dim(key)
            value = self.reshape_heads_to_batch_dim(value)
            encoder_hidden_states_key_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_key_proj)
            encoder_hidden_states_value_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_value_proj)

            key = torch.concat([encoder_hidden_states_key_proj, key], dim=1)
            value = torch.concat([encoder_hidden_states_value_proj, value], dim=1)
        else:
            encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
            key = self.to_k(encoder_hidden_states)
            value = self.to_v(encoder_hidden_states)

            key = self.reshape_heads_to_batch_dim(key)
            value = self.reshape_heads_to_batch_dim(value)

        if attention_mask is not None:
            if attention_mask.shape[-1] != query.shape[1]:
                target_length = query.shape[1]
                attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
                attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)

        # attention, what we cannot get enough of
        if self._use_memory_efficient_attention_xformers:
            hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
            # Some versions of xformers return output in fp32, cast it back to the dtype of the input
            hidden_states = hidden_states.to(query.dtype)
        else:
            if self._slice_size is None or query.shape[0] // self._slice_size == 1:
                hidden_states = self._attention(query, key, value, attention_mask)
            else:
                hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)

        # linear proj
        hidden_states = self.to_out[0](hidden_states)

        # dropout
        hidden_states = self.to_out[1](hidden_states)
        return hidden_states

    def _attention(self, query, key, value, attention_mask=None):
        if self.upcast_attention:
            query = query.float()
            key = key.float()

        attention_scores = torch.baddbmm(
            torch.empty(query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device),
            query,
            key.transpose(-1, -2),
            beta=0,
            alpha=self.scale,
        )

        if attention_mask is not None:
            attention_scores = attention_scores + attention_mask

        if self.upcast_softmax:
            attention_scores = attention_scores.float()

        attention_probs = attention_scores.softmax(dim=-1)

        # cast back to the original dtype
        attention_probs = attention_probs.to(value.dtype)

        # compute attention output
        hidden_states = torch.bmm(attention_probs, value)

        # reshape hidden_states
        hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
        return hidden_states

    def _sliced_attention(self, query, key, value, sequence_length, dim, attention_mask):
        batch_size_attention = query.shape[0]
        hidden_states = torch.zeros(
            (batch_size_attention, sequence_length, dim // self.heads), device=query.device, dtype=query.dtype
        )
        slice_size = self._slice_size if self._slice_size is not None else hidden_states.shape[0]
        for i in range(hidden_states.shape[0] // slice_size):
            start_idx = i * slice_size
            end_idx = (i + 1) * slice_size

            query_slice = query[start_idx:end_idx]
            key_slice = key[start_idx:end_idx]

            if self.upcast_attention:
                query_slice = query_slice.float()
                key_slice = key_slice.float()

            attn_slice = torch.baddbmm(
                torch.empty(slice_size, query.shape[1], key.shape[1], dtype=query_slice.dtype, device=query.device),
                query_slice,
                key_slice.transpose(-1, -2),
                beta=0,
                alpha=self.scale,
            )

            if attention_mask is not None:
                attn_slice = attn_slice + attention_mask[start_idx:end_idx]

            if self.upcast_softmax:
                attn_slice = attn_slice.float()

            attn_slice = attn_slice.softmax(dim=-1)

            # cast back to the original dtype
            attn_slice = attn_slice.to(value.dtype)
            attn_slice = torch.bmm(attn_slice, value[start_idx:end_idx])

            hidden_states[start_idx:end_idx] = attn_slice

        # reshape hidden_states
        hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
        return hidden_states

    def _memory_efficient_attention_xformers(self, query, key, value, attention_mask):
        # TODO attention_mask
        query = query.contiguous()
        key = key.contiguous()
        value = value.contiguous()
        import xformers.ops
        from modules.sd_hijack_optimizations import get_xformers_flash_attention_op
        hidden_states = xformers.ops.memory_efficient_attention(query, key, value, attn_bias=attention_mask, 
                                                                op=get_xformers_flash_attention_op(query, key, value))
        hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
        return hidden_states


class VersatileAttention(CrossAttention):
    def __init__(
            self,
            attention_mode                     = None,
            cross_frame_attention_mode         = None,
            temporal_position_encoding         = False,
            temporal_position_encoding_max_len = 24,            
            *args, **kwargs
        ):
        super().__init__(*args, **kwargs)
        assert attention_mode == "Temporal"

        self.attention_mode = attention_mode
        self.is_cross_attention = kwargs["cross_attention_dim"] is not None
        
        self.pos_encoder = PositionalEncoding(
            kwargs["query_dim"],
            dropout=0., 
            max_len=temporal_position_encoding_max_len
        ) if (temporal_position_encoding and attention_mode == "Temporal") else None

    def extra_repr(self):
        return f"(Module Info) Attention_Mode: {self.attention_mode}, Is_Cross_Attention: {self.is_cross_attention}"

    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, video_length=None):
        batch_size, sequence_length, _ = hidden_states.shape

        if self.attention_mode == "Temporal":
            d = hidden_states.shape[1]
            hidden_states = rearrange(hidden_states, "(b f) d c -> (b d) f c", f=video_length)
            
            if self.pos_encoder is not None:
                hidden_states = self.pos_encoder(hidden_states)
            
            encoder_hidden_states = repeat(encoder_hidden_states, "b n c -> (b d) n c", d=d) if encoder_hidden_states is not None else encoder_hidden_states
        else:
            raise NotImplementedError

        encoder_hidden_states = encoder_hidden_states

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

        query = self.to_q(hidden_states)
        dim = query.shape[-1]
        query = self.reshape_heads_to_batch_dim(query)

        if self.added_kv_proj_dim is not None:
            raise NotImplementedError

        encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
        key = self.to_k(encoder_hidden_states)
        value = self.to_v(encoder_hidden_states)

        key = self.reshape_heads_to_batch_dim(key)
        value = self.reshape_heads_to_batch_dim(value)

        if attention_mask is not None:
            if attention_mask.shape[-1] != query.shape[1]:
                target_length = query.shape[1]
                attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
                attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)

        # attention, what we cannot get enough of
        if self._use_memory_efficient_attention_xformers:
            hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
            # Some versions of xformers return output in fp32, cast it back to the dtype of the input
            hidden_states = hidden_states.to(query.dtype)
        else:
            if self._slice_size is None or query.shape[0] // self._slice_size == 1:
                hidden_states = self._attention(query, key, value, attention_mask)
            else:
                hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)

        # linear proj
        hidden_states = self.to_out[0](hidden_states)

        # dropout
        hidden_states = self.to_out[1](hidden_states)

        if self.attention_mode == "Temporal":
            hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d)

        return hidden_states