https://github.com/omerbt/TokenFlow/issues/25
https://github.com/omerbt/TokenFlow/issues/31
https://github.com/omerbt/TokenFlow/issues/32
https://github.com/eps696/SDfu

本文主要讲解TokenFlow的Model部分是如何构造的，代码摘自TokenFlow/tokenflow_utils.py。

tokenflow的Model构建逻辑是先加载原始的Stable Diffusion，然后重新注册需要修改的UNet的模块，修改操作的调用首先在run_tokenflow.py中：

self.init_method(conv_injection_t=pnp_f_t, qk_injection_t=pnp_attn_t)

想看懂后面的代码, 首先需要看懂SD的BasicTransformerBlock的源码，最好再看看PnP的源码 pnp-diffusers，因为TokenFlow就是基于PnP改进而来的。

    def init_method(self, conv_injection_t, qk_injection_t):self.qk_injection_timesteps = self.scheduler.timesteps[:qk_injection_t] if qk_injection_t >= 0 else []self.conv_injection_timesteps = self.scheduler.timesteps[:conv_injection_t] if conv_injection_t >= 0 else []register_extended_attention_pnp(self, self.qk_injection_timesteps)register_conv_injection(self, self.conv_injection_timesteps)set_tokenflow(self.unet)

init_method函数完成了3件事：
（1）register_extended_attention_pnp：replace unet 的 self attention（扩展为KV来自多帧，同时完成inject。
（2）register_conv_injection：replace conv 的 conv_injection (UpBlock的第二个resnet block，完成inject。
（3）set_tokenflow：replace unet 的 16 BasicTransformerBlock to TokenFlowBlock。
其中，qk_injection_timesteps 和 conv_injection_timesteps是两个timestep list，用于控制PnP Inject操作只在前几个step执行。

除了这些对UNet Model的修改，源码中batched_denoise_step函数为了先编辑关键帧也进行了register_pivotal设置关键帧id。在编辑每个batch时前进行了register_batch_idx设置batch id。在预测噪声前register_time为UNet的某些layer设置step t。

接下来，我们将按照顺序一个一个解析，tokenflow在推理过程中，对原始Stable Diffusion模型做的修改。

register_extended_attention_pnp

register_extended_attention_pnp函数的作用：为UNet的所有BasicTransformerBlock layer的attn1（16个）重构forward函数为extend attention的sa_forward，但只对其中的部分attn1（8个）注入injection_schedule使用PnP操作。

由BasicTransformerBlock的结果可知：attn1虽然Class实现上是CrossAttention，但推理时不传入context做KV，本质上是SelfAttention。

class BasicTransformerBlock(nn.Module):def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True):super().__init__()self.attn1 = CrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout)  # is a self-attentionself.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)self.attn2 = CrossAttention(query_dim=dim, context_dim=context_dim,heads=n_heads, dim_head=d_head, dropout=dropout)  # cross attentionself.norm1 = nn.LayerNorm(dim)self.norm2 = nn.LayerNorm(dim)self.norm3 = nn.LayerNorm(dim)def forward(self, x, context=None):x = self.attn1(self.norm1(x)) + xx = self.attn2(self.norm2(x), context=context) + xx = self.ff(self.norm3(x)) + xreturn x

入参：register_extended_attention_pnp函数传入的两个必须参数是unet model和injection_schedule。injection_schedule用于控制推理过程中PnP Injection执行的时间步，因为我们希望只在前几个timestep进行PnP操作。

重构：因为原始tokenflow的extend_attention的forward是sa_forward，是对所有帧进行attention矩阵运算，消耗资源太大，于是我为其添加了sa_3frame_forward仅计算相邻3帧的attention。我重构过的register_extended_attention_pnp函数如下图，在原始基础上添加了一个is_3_frame参数用于选择是否使用sa_3frame_forward。

首先我们先跳过sa_forward和sa_3frame_forward这两个函数，看一下如何找到UNet对应的模块并修改其forward方法。

1. 为所有BasicTransformerBlock layer的attn1重构forward

根据 register_forward_fun 判断使用那种sa_forward，然后遍历unet的每个模块，判断改模块是否继承自BasicTransformerBlock ，如果有则为其修改forward，但将其injection_schedule置空（即不执行PnP）。

    module_names = []register_forward_fun = sa_3frame_forward if is_3_frame else sa_forwardfor module_name, module in model.unet.named_modules():if isinstance_str(module, "BasicTransformerBlock"):module_names.append(module_name)# replace BasicTransformerBlock.attn1's forward with sa_forwardmodule.attn1.forward = register_forward_fun(module.attn1)# set injection_schedule empty[] for BasicTransformerBlock.attn1setattr(module.attn1, 'injection_schedule', [])print(f"all change {len(module_names)} layer's BasicTransformerBlock.attn1.forward() for extended_attention_pnp...")print(module_names)  # up_blocks.1.attentions.0.transformer_blocks.0

isinstance_str 判断x的继承的类型列表中是否包含cls_name类：

def isinstance_str(x: object, cls_name: str):for _cls in x.__class__.__mro__:if _cls.__name__ == cls_name:return Truereturn False

第一次重构了unet中如下16层attention的forward：6个down_block的，9个up_block，1个mid_block的。

down_blocks.0.attentions.0.transformer_blocks.0.attn1
down_blocks.0.attentions.1.transformer_blocks.0.attn1
down_blocks.1.attentions.0.transformer_blocks.0.attn1
down_blocks.1.attentions.1.transformer_blocks.0.attn1
down_blocks.2.attentions.0.transformer_blocks.0.attn1
down_blocks.2.attentions.1.transformer_blocks.0.attn1up_blocks.1.attentions.0.transformer_blocks.0.attn1
up_blocks.1.attentions.1.transformer_blocks.0.attn1
up_blocks.1.attentions.2.transformer_blocks.0.attn1
up_blocks.2.attentions.0.transformer_blocks.0.attn1
up_blocks.2.attentions.1.transformer_blocks.0.attn1
up_blocks.2.attentions.2.transformer_blocks.0.attn1
up_blocks.3.attentions.0.transformer_blocks.0.attn1
up_blocks.3.attentions.1.transformer_blocks.0.attn1
up_blocks.3.attentions.2.transformer_blocks.0.attn1mid_block.attentions.0.transformer_blocks.0.attn1

2. 对其中的部分attn1（8个）注入injection_schedule使用PnP操作

在res_dict 的指示下，对具体的attn1修改forward函数，同时为其注册injection_schedule，使用PnP操作。

    res_dict = {1: [1, 2], 2: [0, 1, 2], 3: [0, 1, 2]}  # upblock's self-attention layers# we are injecting attention in blocks 4 - 11 of the Unet UpBlock, so not in the first block of the lowest resolutionfor res in res_dict:  # res = 1for block in res_dict[res]:  # res_dict[res] = [1, 2]module = model.unet.up_blocks[res].attentions[block].transformer_blocks[0].attn1module.forward = sa_forward(module)setattr(module, 'injection_schedule', injection_schedule)

第二次重构8个up_blocks的：

model.unet.up_blocks.1.attentions.1.transformer_blocks.0.attn1
model.unet.up_blocks.1.attentions.2.transformer_blocks.0.attn1
model.unet.up_blocks.2.attentions.0.transformer_blocks.0.attn1
model.unet.up_blocks.2.attentions.1.transformer_blocks.0.attn1
model.unet.up_blocks.2.attentions.2.transformer_blocks.0.attn1
model.unet.up_blocks.3.attentions.0.transformer_blocks.0.attn1
model.unet.up_blocks.3.attentions.1.transformer_blocks.0.attn1
model.unet.up_blocks.3.attentions.2.transformer_blocks.0.attn1

3. sa_forward

PnP的操作：因为TokenFlow的输入同时考虑了PnP和classifer-free guidance，所以原本UNet输入的单个latent变成了3份：source_latents + x + x（其中一个 x 对应edit_prompt，一个 x 对应null_prompt，source_latents对应了source_prompt）。

latent_model_input = torch.cat([source_latents] + ([x] * 2))  

这样就可以在Unet推理的时候，直接从输入的latents x中切片，分为3份将source_latents注入到uncond_latents和cond_latents（PnP的注入就是直接替换），对于self-attention，我们只替换Q和K。

source_latents = x[:n_frames]
uncond_latents = x[n_frames:2*n_frames]
cond_latents = x[2*n_frames:]
# source inject uncond
q[n_frames:2*n_frames] = q[:n_frames]
k[n_frames:2*n_frames] = k[:n_frames]
# source inject cond
q[2*n_frames:] = q[:n_frames]
k[2*n_frames:] = k[:n_frames]

Extend_Attention：tokenflow实现扩展的self-attention使用，因为对于第i帧，计算self attention时，Q是第i帧的特征，KV要来自其他所有帧，所以要repeat一下K和V，方便后面计算。
$$T_{base}=Softmax(\frac{Q^i;[K^{i1},...,K^{ik}]}{\sqrt{d}})\cdot [V^{i1},...,V^{ik}]$$

# KV reshape and repeat for extend_attention: Softmax(Q_i_frame @ K_all_frame) @ V_all_frame
# (n_frames, seq_len, dim) -> (1, n_frames * seq_len, dim) -> (n_frames, n_frames * seq_len, dim)
k_source = k[:n_frames]
k_uncond = k[n_frames:2 * n_frames].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)
k_cond = k[2 * n_frames:].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)
v_source = v[:n_frames]
v_uncond = v[n_frames:2 * n_frames].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)
v_cond = v[2 * n_frames:].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)

因为逐帧进行计算，而且多头注意力逐头进行计算，构造双重for循环计算attention 得到第 i 帧第 j 头的 attention out，最后分别concat帧维度和头维度得到最终atention结果：

Q @ K -> sim：
(b, 1, seq_len, dim//head) @ (b, 1, dim//head, frame*seq_len) -> (b, 1, seq_len, frame*seq_len)sim @ V -> out：
(b, 1, seq_len, frame*seq_len) @ (b, 1, frame*seq_len, dim//head) -> (b, 1, seq_len, dim//head)cat each head's out：
(b->n_frames, 1, seq_len, dim//head) -> (n_frames, 1, seq_len, dim//head)cat each frame's out：
(n_frames, 1, seq_len, dim//head) -> (n_frames, heads, seq_len, dim//heads)

sa_forward完整代码如下:

def sa_forward(self):to_out = self.to_out  # self.to_out = [linear, dropout]if type(to_out) is torch.nn.modules.container.ModuleList:to_out = self.to_out[0]else:to_out = self.to_outdef forward(x, encoder_hidden_states=None, attention_mask=None):  is_cross = encoder_hidden_states is not None  # corss-attention or self-attentionh = self.headsbatch_size, sequence_length, dim = x.shape  # (3*n_frames, seq_len, dim)# batch: 前n_frames个样本为source feature, 中间n_frames个样本为uncond featur, 后n_frames个样本为cond featuren_frames = batch_size // 3# source_latents = x[:n_frames], uncond_latents = x[n_frames:2*n_frames], cond_latents = x[2*n_frames:]encoder_hidden_states = encoder_hidden_states if is_cross else xq = self.to_q(x)k = self.to_k(encoder_hidden_states)v = self.to_v(encoder_hidden_states)# PnP Injection QK：只需要sample过程中的前几个timestep进行injection (判断t是否符合)，且只在UpBlock进行injectif self.injection_schedule is not None and (self.t in self.injection_schedule or self.t == 1000):# source inject into unconditionalq[n_frames:2 * n_frames] = q[:n_frames]k[n_frames:2 * n_frames] = k[:n_frames]# source inject into conditionalq[2 * n_frames:] = q[:n_frames]k[2 * n_frames:] = k[:n_frames]# KV reshape and repeat for extend_attention: Softmax(Q_i_frame @ K_all_frame) @ V_all_frame# (n_frames, seq_len, dim) -> (1, n_frames * seq_len, dim) -> (n_frames, n_frames * seq_len, dim)k_source = k[:n_frames]k_uncond = k[n_frames:2 * n_frames].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)k_cond = k[2 * n_frames:].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)v_source = v[:n_frames]v_uncond = v[n_frames:2 * n_frames].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)v_cond = v[2 * n_frames:].reshape(1, n_frames * sequence_length, -1).repeat(n_frames, 1, 1)# project QKV's source, cond and uncond, respectively q_source = self.reshape_heads_to_batch_dim(q[:n_frames])  # q (n_frames*heads, seq_len, dim//heads)q_uncond = self.reshape_heads_to_batch_dim(q[n_frames:2 * n_frames])q_cond = self.reshape_heads_to_batch_dim(q[2 * n_frames:])k_source = self.reshape_heads_to_batch_dim(k_source)  # kv (n_frames*heads, n_frames * seq_len, dim//heads)k_uncond = self.reshape_heads_to_batch_dim(k_uncond)k_cond = self.reshape_heads_to_batch_dim(k_cond)v_source = self.reshape_heads_to_batch_dim(v_source)v_uncond = self.reshape_heads_to_batch_dim(v_uncond)v_cond = self.reshape_heads_to_batch_dim(v_cond)# split headsq_src = q_source.view(n_frames, h, sequence_length, dim // h)k_src = k_source.view(n_frames, h, sequence_length, dim // h)v_src = v_source.view(n_frames, h, sequence_length, dim // h)q_uncond = q_uncond.view(n_frames, h, sequence_length, dim // h)k_uncond = k_uncond.view(n_frames, h, sequence_length * n_frames, dim // h)v_uncond = v_uncond.view(n_frames, h, sequence_length * n_frames, dim // h)q_cond = q_cond.view(n_frames, h, sequence_length, dim // h)k_cond = k_cond.view(n_frames, h, sequence_length * n_frames, dim // h)v_cond = v_cond.view(n_frames, h, sequence_length * n_frames, dim // h)out_source_all = []out_uncond_all = []out_cond_all = []# each frame or single_batch framessingle_batch = n_frames<=12b = n_frames if single_batch else 1  # b=1# do attention for each frame respectively. frames [frame：frame=b]for frame in range(0, n_frames, b):out_source = []out_uncond = []out_cond = []# do attention for each head respectively. head jfor j in range(h):# do attention for source, cond and uncond respectively, (b, 1, seq_len, dim//head) @ (b, 1, dim//head, frame*seq_len) -> (b, 1, seq_len, frame*seq_len)sim_source_b = torch.bmm(q_src[frame: frame+ b, j], k_src[frame: frame+ b, j].transpose(-1, -2)) * self.scalesim_uncond_b = torch.bmm(q_uncond[frame: frame+ b, j], k_uncond[frame: frame+ b, j].transpose(-1, -2)) * self.scalesim_cond = torch.bmm(q_cond[frame: frame+ b, j], k_cond[frame: frame+ b, j].transpose(-1, -2)) * self.scale# append each head's out, (b, 1, seq_len, frame*seq_len) @ (b, 1, frame*seq_len, dim//head) -> (b, 1, seq_len, dim//head)out_source.append(torch.bmm(sim_source_b.softmax(dim=-1), v_src[frame: frame+ b, j]))out_uncond.append(torch.bmm(sim_uncond_b.softmax(dim=-1), v_uncond[frame: frame+ b, j]))out_cond.append(torch.bmm(sim_cond.softmax(dim=-1), v_cond[frame: frame+ b, j]))# cat each head's out, (b->n_frames, 1, seq_len, dim//head) -> (n_frames, 1, seq_len, dim//head)out_source = torch.cat(out_source, dim=0)out_uncond = torch.cat(out_uncond, dim=0) out_cond = torch.cat(out_cond, dim=0) if single_batch: # if use single_batch, view single_batch frame's outout_source = out_source.view(h, n_frames,sequence_length, dim // h).permute(1, 0, 2, 3).reshape(h * n_frames, sequence_length, -1)out_uncond = out_uncond.view(h, n_frames,sequence_length, dim // h).permute(1, 0, 2, 3).reshape(h * n_frames, sequence_length, -1)out_cond = out_cond.view(h, n_frames,sequence_length, dim // h).permute(1, 0, 2, 3).reshape(h * n_frames, sequence_length, -1)# append each frame's outout_source_all.append(out_source)out_uncond_all.append(out_uncond)out_cond_all.append(out_cond)# cat each frame's out, (n_frames, 1, seq_len, dim//head) -> (n_frames, heads, seq_len, dim//heads)out_source = torch.cat(out_source_all, dim=0)out_uncond = torch.cat(out_uncond_all, dim=0)out_cond = torch.cat(out_cond_all, dim=0)# cat source, cond and uncond's out, (n_frames, heads, seq_len, dim//heads) -> (3*n_frames, heads, seq_len, dim//heads)out = torch.cat([out_source, out_uncond, out_cond], dim=0)out = self.reshape_batch_dim_to_heads(out)return to_out(out)return forward

3. sa_3frame_forward

因为使用了PnP：每次进行self attention不再是像sa_forward一样，对输入repeat重复n_frames，而是 source_latent 进行正常的KV来自单帧的self attention forward_original，uncond_latent 和 cond_latent 进行 KV 来自相邻3帧 的self attention forward_extended。

每次计算第 i 帧的attention时（window_size=3），以第 i 帧为中心 ，取下标=[i-1, i, i+1]的3帧作为KV：

 def sa_3frame_forward(self):  # self attention只是扩展到连续的 3 个关键帧，而不是所有关键帧。to_out = self.to_outif type(to_out) is torch.nn.modules.container.ModuleList:to_out = self.to_out[0]else:to_out = self.to_out# 原始的UNet attention forwarddef forward_original(q, k, v):n_frames, seq_len, dim = q.shapeh = self.headshead_dim = dim // hq = self.head_to_batch_dim(q).reshape(n_frames, h, seq_len, head_dim)k = self.head_to_batch_dim(k).reshape(n_frames, h, seq_len, head_dim)v = self.head_to_batch_dim(v).reshape(n_frames, h, seq_len, head_dim)out_all = []for frame in range(n_frames):out = []for j in range(h):sim = torch.matmul(q[frame, j], k[frame, j].transpose(-1, -2)) * self.scale # (seq_len, seq_len)                                            out.append(torch.matmul(sim.softmax(dim=-1), v[frame, j])) # h * (seq_len, head_dim)out = torch.cat(out, dim=0).reshape(-1, seq_len, head_dim) # (h, seq_len, head_dim)out_all.append(out) # n_frames * (h, seq_len, head_dim)out = torch.cat(out_all, dim=0) # (n_frames * h, seq_len, head_dim)out = self.batch_to_head_dim(out) # (n_frames, seq_len, h * head_dim)return out# extend UNet attention forward(all frames)def forward_extended(q, k, v):n_frames, seq_len, dim = q.shapeh = self.headshead_dim = dim // hq = self.head_to_batch_dim(q).reshape(n_frames, h, seq_len, head_dim)k = self.head_to_batch_dim(k).reshape(n_frames, h, seq_len, head_dim)v = self.head_to_batch_dim(v).reshape(n_frames, h, seq_len, head_dim)out_all = []window_size = 3for frame in range(n_frames):  # frame=32, window_size=3: window=[14, 15, 16, 17, 18]out = []# sliding window to improve speed.  以当前帧frame为中心，取window_size大小的帧，如frame_idx=1时, window: [0, 1, 2]window = range(max(0, frame-window_size // 2), min(n_frames, frame+window_size//2+1))  for j in range(h):sim_all = []  # 存当前帧frame和window内3帧的sim，len(sim_all)=3for kframe in window:  # (1, 1, seq_len, head_dim) @ (1, 1, head_dim, seq_len) -> (1, 1, seq_len, seq_len)# 当前帧frame 依次和window内的帧kframe，计算sim存入sim_allsim_all.append(torch.matmul(q[frame, j], k[kframe, j].transpose(-1, -2)) * self.scale) # window * (seq_len, seq_len)sim_all = torch.cat(sim_all).reshape(len(window), seq_len, seq_len).transpose(0, 1) # (seq_len, window, seq_len)sim_all = sim_all.reshape(seq_len, len(window) * seq_len) # (seq_len, window * seq_len)out.append(torch.matmul(sim_all.softmax(dim=-1), v[window, j].reshape(len(window) * seq_len, head_dim))) # h * (seq_len, head_dim)out = torch.cat(out, dim=0).reshape(-1, seq_len, head_dim) # (h, seq_len, head_dim)out_all.append(out) # n_frames * (h, seq_len, head_dim)out = torch.cat(out_all, dim=0) # (n_frames * h, seq_len, head_dim)out = self.batch_to_head_dim(out) # (n_frames, seq_len, h * head_dim)return outdef forward(x, encoder_hidden_states=None, attention_mask=None):batch_size, sequence_length, dim = x.shapeh = self.headsn_frames = batch_size // 3is_cross = encoder_hidden_states is not Noneencoder_hidden_states = encoder_hidden_states if is_cross else xq = self.to_q(x)k = self.to_k(encoder_hidden_states)v = self.to_v(encoder_hidden_states)if self.injection_schedule is not None and (self.t in self.injection_schedule or self.t == 1000):# inject unconditionalq[n_frames:2 * n_frames] = q[:n_frames]k[n_frames:2 * n_frames] = k[:n_frames]# inject conditionalq[2 * n_frames:] = q[:n_frames]k[2 * n_frames:] = k[:n_frames]# source_latent 正常的self attention, uncond 和 cond进行 KV来自相邻3帧的self attentionout_source = forward_original(q[:n_frames], k[:n_frames], v[:n_frames])  out_uncond = forward_extended(q[n_frames:2 * n_frames], k[n_frames:2 * n_frames], v[n_frames:2 * n_frames])out_cond = forward_extended(q[2 * n_frames:], k[2 * n_frames:], v[2 * n_frames:])out = torch.cat([out_source, out_uncond, out_cond], dim=0) # (3 * n_frames, seq_len, dim)return to_out(out)return forward

register_conv_injection

为UNet注册完SelfAttention的forward后，再来为其UNet的unet.up_blocks[1].resnets[1]的ResnetBlock2D注册新的forward，同时注册injection_schedule控制PnP注入时间步。

在conv_forward中只比普通的ResnetBlock2D的forward多了一步PnP Inject的操作：

if self.injection_schedule is not None and (self.t in self.injection_schedule or self.t == 1000):source_batch_size = int(hidden_states.shape[0] // 3)# inject unconditionalhidden_states[source_batch_size:2 * source_batch_size] = hidden_states[:source_batch_size]# inject conditionalhidden_states[2 * source_batch_size:] = hidden_states[:source_batch_size]

set_tokenflow

__class__就是返回自己的父类，set_tokenflow就是找到UNet中所有父类是BasicTransformerBlock的模块，对他们把BasicTransformerBlock作为父类，外面再套一层TokenFlowBlock类。

def set_tokenflow(model: torch.nn.Module):"""Sets the tokenflow attention blocks in a model."""for _, module in model.named_modules():if isinstance_str(module, "BasicTransformerBlock"):# 16个 module.__class__ = <class 'diffusers.models.attention.BasicTransformerBlock'>make_tokenflow_block_fn = make_tokenflow_attention_block # 将BasicTransformerBlock作为父类，外面再套一层TokenFlowBlock类module.__class__ = make_tokenflow_block_fn(module.__class__)# Something needed for older versions of diffusersif not hasattr(module, "use_ada_layer_norm_zero"):module.use_ada_layer_norm = Falsemodule.use_ada_layer_norm_zero = Falsereturn model

make_tokenflow_attention_block

这个函数就是定义了一个TokenFlowBlock类，然后返回TokenFlowBlock类。TokenFlowBlock就继承自BasicTransformerBlock类，只重写了forward函数。

首先pivotal_pass判断是否是关键帧：

• 关键帧，就存下pivot_hidden_states
• 非关键帧，取非关键帧与关键帧的source_latent，计算其与关键帧的余弦相似度cosine_sim，shape=(n_frames * seq_len, len(batch_idxs) * seq_len)，求得相似度最大的帧下标idx，然后为source、uncond、cond 堆叠3份。
  • 如果当前batch不是第一个batch，len(batch_idxs) =2， 分别保存最相似的帧下标到idx1和idx2
  • 如果是第一个batch，len(batch_idxs) =1，保存最相似的帧下标到idx1

			batch_size, sequence_length, dim = hidden_states.shape  # (batch, seq_len, dim)n_frames = batch_size // 3  # batch = 3 * n_frames: source + uncond + condmid_idx = n_frames // 2hidden_states = hidden_states.view(3, n_frames, sequence_length, dim)  # (source + uncond + cond, n_frames, seq_len, dim)norm_hidden_states = self.norm1(hidden_states).view(3, n_frames, sequence_length, dim)if self.pivotal_pass:  # is_pivotal = True # 关键帧，存下self.pivot_hidden_states = norm_hidden_states  # (3, n_frames, sequence_length, dim) ,关键帧的n_frames=5else:  # is_pivotal = False # 非关键帧，与关键帧计算source_latent的cosine_simidx1 = []idx2 = []batch_idxs = [self.batch_idx]  # 每batch_size帧进行一批处理，batch_idx是第几个batch，如32帧，batch_size=8，batch_idx可以为0或1或2或3或4if self.batch_idx > 0:  # 如果不是第一个batchbatch_idxs.append(self.batch_idx - 1)  # 加入前一个batch的idx，如当前batch_idx=1时，再加入0，则batch_idxs=[1,0]# 取source_latent的非关键帧与关键帧计算cosine_sim，如果batch_idxs=[1,0]，则只拿第0个batch和第1个batch的关键帧和其norm_hidden_states计算simsim = batch_cosine_sim(norm_hidden_states[0].reshape(-1, dim),  # (n_frames*sequence_length, dim)self.pivot_hidden_states[0][batch_idxs].reshape(-1, dim))  # (len(batch_idxs)*sequence_length, dim)if len(batch_idxs) == 2:  # 如果不是第一个batch， 分别保存最相似的帧下标到idx1和idx2# sim: (n_frames * seq_len, len(batch_idxs) * seq_len),  len(batch_idxs)=2sim1, sim2 = sim.chunk(2, dim=1) idx1.append(sim1.argmax(dim=-1))  # (n_frames * seq_len) 个数,每个数在[0,76]之间idx2.append(sim2.argmax(dim=-1))  # (n_frames * seq_len) 个数,每个数在[0,76]之间else:  # 如果是第一个batch，保存最相似的帧下标到idx1idx1.append(sim.argmax(dim=-1))# 为source、uncond、cond 堆叠3份idx1 = torch.stack(idx1 * 3, dim=0) # （3, n_frames * seq_len）idx1 = idx1.squeeze(1)if len(batch_idxs) == 2:idx2 = torch.stack(idx2 * 3, dim=0) # (3, n_frames * seq_len)idx2 = idx2.squeeze(1)

接下来依次进行Self-Attention attn1、Cross-Attention attn2、和Feed-forward ff，其中Cross-Attention和Feed-forward没有任何改变，唯一改变的就是Self-Attention过程 ：

• 对于关键帧，计算 self-attention 结果，并将其保存下来。
• 对于非关键帧，将其与关键帧的 attention 结果进行融合。融合方式为加权平均，权重由帧与关键帧之间的距离决定。如果非关键帧是第一个 batch 中的帧，则直接使用关键帧的 attention 结果。如果非关键帧是第二个 batch 中的帧，则计算与第一个 batch 中的关键帧和第二个 batch 中的关键帧的 attention 结果，然后进行加权平均。权重由帧与两个关键帧之间的距离决定。具体公式如下：
  $$weight=|s-p1|/(|s-p1|+|s-p2|)$$
  其中，s 表示帧的编号，p1 表示第一个关键帧的编号，p2 表示第二个关键帧的编号。

			# 1. Self-Attentioncross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}if self.pivotal_pass:# norm_hidden_states.shape = 3, n_frames * seq_len, dimself.attn_output = self.attn1(norm_hidden_states.view(batch_size, sequence_length, dim),encoder_hidden_states=encoder_hidden_states if self.only_cross_attention else None,**cross_attention_kwargs,)# 3, n_frames * seq_len, dim - > 3 * n_frames, seq_len, dimself.kf_attn_output = self.attn_output else:batch_kf_size, _, _ = self.kf_attn_output.shapeself.attn_output = self.kf_attn_output.view(3, batch_kf_size // 3, sequence_length, dim)[:,batch_idxs]  # 3, n_frames, seq_len, dim --> 3, len(batch_idxs), seq_len, dim# gather values from attn_output, using idx as indices, and get a tensor of shape 3, n_frames, seq_len, dimif not self.pivotal_pass:if len(batch_idxs) == 2:attn_1, attn_2 = self.attn_output[:, 0], self.attn_output[:, 1]attn_output1 = attn_1.gather(dim=1, index=idx1.unsqueeze(-1).repeat(1, 1, dim))attn_output2 = attn_2.gather(dim=1, index=idx2.unsqueeze(-1).repeat(1, 1, dim))s = torch.arange(0, n_frames).to(idx1.device) + batch_idxs[0] * n_frames# distance from the pivotp1 = batch_idxs[0] * n_frames + n_frames // 2p2 = batch_idxs[1] * n_frames + n_frames // 2d1 = torch.abs(s - p1)d2 = torch.abs(s - p2)# weightw1 = d2 / (d1 + d2)w1 = torch.sigmoid(w1)w1 = w1.unsqueeze(0).unsqueeze(-1).unsqueeze(-1).repeat(3, 1, sequence_length, dim)attn_output1 = attn_output1.view(3, n_frames, sequence_length, dim)attn_output2 = attn_output2.view(3, n_frames, sequence_length, dim)attn_output = w1 * attn_output1 + (1 - w1) * attn_output2else:attn_output = self.attn_output[:,0].gather(dim=1, index=idx1.unsqueeze(-1).repeat(1, 1, dim))attn_output = attn_output.reshape(batch_size, sequence_length, dim)  # 3 * n_frames, seq_len, dimelse:attn_output = self.attn_outputhidden_states = hidden_states.reshape(batch_size, sequence_length, dim)  # 3 * n_frames, seq_len, dimhidden_states = attn_output + hidden_states  # res_connect