1135 lines
46 KiB
Python
1135 lines
46 KiB
Python
# Modified from https://github.com/MeiGen-AI/InfiniteTalk/blob/main/wan/modules/multitalk_model.py
|
|
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
|
|
import math
|
|
import types
|
|
from functools import lru_cache
|
|
from typing import Any, Dict
|
|
|
|
import numpy as np
|
|
import torch
|
|
import torch.cuda.amp as amp
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
from diffusers import ModelMixin
|
|
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
|
from diffusers.utils import is_torch_version
|
|
from einops import rearrange, repeat
|
|
|
|
from ..dist import (get_sequence_parallel_rank,
|
|
get_sequence_parallel_world_size, get_sp_group,
|
|
usp_attn_infinitetalk_forward)
|
|
from .attention_utils import attention
|
|
from .wan_transformer3d import (WanI2VCrossAttention, WanLayerNorm, WanRMSNorm,
|
|
WanTransformer3DModel, rope_apply, rope_params,
|
|
sinusoidal_embedding_1d)
|
|
|
|
|
|
def normalize_and_scale(column, source_range, target_range, epsilon=1e-8):
|
|
source_min, source_max = source_range
|
|
new_min, new_max = target_range
|
|
normalized = (column - source_min) / (source_max - source_min + epsilon)
|
|
scaled = normalized * (new_max - new_min) + new_min
|
|
return scaled
|
|
|
|
|
|
def split_token_counts_and_frame_ids(T, token_frame, world_size, rank):
|
|
S = T * token_frame
|
|
# Compute split sizes per rank
|
|
base = S // world_size
|
|
rem = S % world_size
|
|
split_sizes = torch.full((world_size,), base, dtype=torch.long)
|
|
split_sizes[:rem] += 1
|
|
|
|
start = split_sizes[:rank].sum()
|
|
end = start + split_sizes[rank]
|
|
|
|
# Vectorized mapping: global index -> frame id
|
|
idx = torch.arange(start, end, dtype=torch.long)
|
|
frame_ids = idx // token_frame.to(idx.device)
|
|
|
|
# Unique counts
|
|
unique_frames, counts = torch.unique(frame_ids, return_counts=True)
|
|
|
|
# Return as Python list (optional)
|
|
return counts.tolist(), unique_frames.tolist()
|
|
|
|
|
|
def rotate_half(x):
|
|
x = rearrange(x, "... (d r) -> ... d r", r=2)
|
|
x1, x2 = x.unbind(dim=-1)
|
|
x = torch.stack((-x2, x1), dim=-1)
|
|
return rearrange(x, "... d r -> ... (d r)")
|
|
|
|
|
|
@torch.compile
|
|
def calculate_x_ref_attn_map(visual_q, ref_k, ref_target_masks, mode='mean', attn_bias=None):
|
|
ref_k = ref_k.to(visual_q.dtype).to(visual_q.device)
|
|
scale = 1.0 / visual_q.shape[-1] ** 0.5
|
|
visual_q = visual_q * scale
|
|
visual_q = visual_q.transpose(1, 2)
|
|
ref_k = ref_k.transpose(1, 2)
|
|
attn = visual_q @ ref_k.transpose(-2, -1)
|
|
if attn_bias is not None:
|
|
attn = attn + attn_bias
|
|
x_ref_attn_map_source = attn.softmax(-1)
|
|
x_ref_attn_maps = []
|
|
ref_target_masks = ref_target_masks.to(visual_q.dtype)
|
|
x_ref_attn_map_source = x_ref_attn_map_source.to(visual_q.dtype)
|
|
for class_idx, ref_target_mask in enumerate(ref_target_masks):
|
|
ref_target_mask = ref_target_mask[None, None, None, ...]
|
|
x_ref_attnmap = x_ref_attn_map_source * ref_target_mask
|
|
x_ref_attnmap = x_ref_attnmap.sum(-1) / ref_target_mask.sum()
|
|
x_ref_attnmap = x_ref_attnmap.permute(0, 2, 1)
|
|
if mode == 'mean':
|
|
x_ref_attnmap = x_ref_attnmap.mean(-1)
|
|
elif mode == 'max':
|
|
x_ref_attnmap = x_ref_attnmap.max(-1)
|
|
x_ref_attn_maps.append(x_ref_attnmap)
|
|
del attn
|
|
del x_ref_attn_map_source
|
|
return torch.concat(x_ref_attn_maps, dim=0)
|
|
|
|
|
|
def get_attn_map_with_target(visual_q, ref_k, shape, ref_target_masks=None, split_num=2, enable_sp=False):
|
|
N_t, N_h, N_w = shape
|
|
x_seqlens = N_h * N_w
|
|
ref_k = ref_k[:, :x_seqlens]
|
|
_, seq_lens, heads, _ = visual_q.shape
|
|
class_num, _ = ref_target_masks.shape
|
|
x_ref_attn_maps = torch.zeros(class_num, seq_lens).to(visual_q.device).to(visual_q.dtype)
|
|
split_chunk = heads // split_num
|
|
for i in range(split_num):
|
|
x_ref_attn_maps_perhead = calculate_x_ref_attn_map(
|
|
visual_q[:, :, i*split_chunk:(i+1)*split_chunk, :],
|
|
ref_k[:, :, i*split_chunk:(i+1)*split_chunk, :],
|
|
ref_target_masks
|
|
)
|
|
x_ref_attn_maps += x_ref_attn_maps_perhead
|
|
return x_ref_attn_maps / split_num
|
|
|
|
|
|
class RotaryPositionalEmbedding1D(nn.Module):
|
|
def __init__(self, head_dim):
|
|
super().__init__()
|
|
self.head_dim = head_dim
|
|
self.base = 10000
|
|
|
|
@lru_cache(maxsize=32)
|
|
def precompute_freqs_cis_1d(self, pos_indices):
|
|
freqs = 1.0 / (self.base ** (torch.arange(0, self.head_dim, 2)[: (self.head_dim // 2)].float() / self.head_dim))
|
|
freqs = freqs.to(pos_indices.device)
|
|
freqs = torch.einsum("..., f -> ... f", pos_indices.float(), freqs)
|
|
freqs = repeat(freqs, "... n -> ... (n r)", r=2)
|
|
return freqs
|
|
|
|
def forward(self, x, pos_indices):
|
|
freqs_cis = self.precompute_freqs_cis_1d(pos_indices)
|
|
x_ = x.float()
|
|
freqs_cis = freqs_cis.float().to(x.device)
|
|
cos, sin = freqs_cis.cos(), freqs_cis.sin()
|
|
cos, sin = rearrange(cos, 'n d -> 1 1 n d'), rearrange(sin, 'n d -> 1 1 n d')
|
|
x_ = (x_ * cos) + (rotate_half(x_) * sin)
|
|
return x_.type_as(x)
|
|
|
|
|
|
class SingleStreamAttention(nn.Module):
|
|
def __init__(
|
|
self,
|
|
dim: int,
|
|
encoder_hidden_states_dim: int,
|
|
num_heads: int,
|
|
qkv_bias: bool,
|
|
qk_norm: bool,
|
|
norm_layer: nn.Module,
|
|
attn_drop: float = 0.0,
|
|
proj_drop: float = 0.0,
|
|
eps: float = 1e-6,
|
|
) -> None:
|
|
super().__init__()
|
|
assert dim % num_heads == 0, "dim should be divisible by num_heads"
|
|
self.dim = dim
|
|
self.encoder_hidden_states_dim = encoder_hidden_states_dim
|
|
self.num_heads = num_heads
|
|
self.head_dim = dim // num_heads
|
|
self.scale = self.head_dim**-0.5
|
|
self.qk_norm = qk_norm
|
|
|
|
self.q_linear = nn.Linear(dim, dim, bias=qkv_bias)
|
|
self.q_norm = norm_layer(self.head_dim, eps=eps) if qk_norm else nn.Identity()
|
|
self.k_norm = norm_layer(self.head_dim, eps=eps) if qk_norm else nn.Identity()
|
|
self.attn_drop = nn.Dropout(attn_drop)
|
|
self.proj = nn.Linear(dim, dim)
|
|
self.proj_drop = nn.Dropout(proj_drop)
|
|
self.kv_linear = nn.Linear(encoder_hidden_states_dim, dim * 2, bias=qkv_bias)
|
|
self.add_q_norm = norm_layer(self.head_dim) if qk_norm else nn.Identity()
|
|
self.add_k_norm = norm_layer(self.head_dim) if qk_norm else nn.Identity()
|
|
|
|
def forward(self, x: torch.Tensor, encoder_hidden_states: torch.Tensor, shape=None, enable_sp=False, kv_seq=None, dtype=torch.bfloat16) -> torch.Tensor:
|
|
r"""
|
|
Args:
|
|
x (`Tensor`):
|
|
Input tensor
|
|
encoder_hidden_states (`Tensor`):
|
|
Encoder hidden states
|
|
shape (`tuple`, *optional*):
|
|
Grid shape (N_t, N_h, N_w)
|
|
enable_sp (`bool`, *optional*, defaults to False):
|
|
Whether to enable sequence parallel
|
|
kv_seq (`int`, *optional*):
|
|
KV sequence length
|
|
dtype (`torch.dtype`, *optional*, defaults to torch.bfloat16):
|
|
Output dtype to match transformer precision
|
|
"""
|
|
N_t, N_h, N_w = shape
|
|
if not enable_sp:
|
|
x = rearrange(x, "B (N_t S) C -> (B N_t) S C", N_t=N_t)
|
|
B, N, C = x.shape
|
|
q = self.q_linear(x)
|
|
q_shape = (B, N, self.num_heads, self.head_dim)
|
|
q = q.view(q_shape).permute((0, 2, 1, 3))
|
|
if self.qk_norm:
|
|
q = self.q_norm(q)
|
|
_, N_a, _ = encoder_hidden_states.shape
|
|
encoder_kv = self.kv_linear(encoder_hidden_states)
|
|
encoder_kv_shape = (B, N_a, 2, self.num_heads, self.head_dim)
|
|
encoder_kv = encoder_kv.view(encoder_kv_shape).permute((2, 0, 3, 1, 4))
|
|
encoder_k, encoder_v = encoder_kv.unbind(0)
|
|
if self.qk_norm:
|
|
encoder_k = self.add_k_norm(encoder_k)
|
|
q = rearrange(q, "B H M K -> B M H K")
|
|
encoder_k = rearrange(encoder_k, "B H M K -> B M H K")
|
|
encoder_v = rearrange(encoder_v, "B H M K -> B M H K")
|
|
# Use attention from attention_utils with dtype conversion
|
|
x = attention(q=q.to(dtype), k=encoder_k.to(dtype), v=encoder_v.to(dtype), attention_type="FLASH_ATTENTION")
|
|
x = rearrange(x, "B M H K -> B H M K")
|
|
x_output_shape = (B, N, C)
|
|
x = x.transpose(1, 2)
|
|
x = x.reshape(x_output_shape)
|
|
x = self.proj(x)
|
|
x = self.proj_drop(x)
|
|
if not enable_sp:
|
|
x = rearrange(x, "(B N_t) S C -> B (N_t S) C", N_t=N_t)
|
|
return x
|
|
|
|
|
|
class SingleStreamMutiAttention(SingleStreamAttention):
|
|
def __init__(
|
|
self,
|
|
dim: int,
|
|
encoder_hidden_states_dim: int,
|
|
num_heads: int,
|
|
qkv_bias: bool,
|
|
qk_norm: bool,
|
|
norm_layer: nn.Module,
|
|
attn_drop: float = 0.0,
|
|
proj_drop: float = 0.0,
|
|
eps: float = 1e-6,
|
|
class_range: int = 24,
|
|
class_interval: int = 4,
|
|
) -> None:
|
|
super().__init__(
|
|
dim=dim,
|
|
encoder_hidden_states_dim=encoder_hidden_states_dim,
|
|
num_heads=num_heads,
|
|
qkv_bias=qkv_bias,
|
|
qk_norm=qk_norm,
|
|
norm_layer=norm_layer,
|
|
attn_drop=attn_drop,
|
|
proj_drop=proj_drop,
|
|
eps=eps,
|
|
)
|
|
self.class_interval = class_interval
|
|
self.class_range = class_range
|
|
self.rope_h1 = (0, self.class_interval)
|
|
self.rope_h2 = (self.class_range - self.class_interval, self.class_range)
|
|
self.rope_bak = int(self.class_range // 2)
|
|
self.rope_1d = RotaryPositionalEmbedding1D(self.head_dim)
|
|
|
|
def forward(self, x: torch.Tensor, encoder_hidden_states: torch.Tensor, shape=None, x_ref_attn_map=None, human_num=None, dtype=torch.bfloat16) -> torch.Tensor:
|
|
r"""
|
|
Args:
|
|
x (`Tensor`):
|
|
Input tensor
|
|
encoder_hidden_states (`Tensor`):
|
|
Audio encoder hidden states
|
|
shape (`tuple`, *optional*):
|
|
Grid shape (N_t, N_h, N_w)
|
|
x_ref_attn_map (`Tensor`, *optional*):
|
|
Reference attention map
|
|
human_num (`int`, *optional*):
|
|
Number of humans
|
|
dtype (`torch.dtype`, *optional*, defaults to torch.bfloat16):
|
|
Output dtype to match transformer precision
|
|
"""
|
|
encoder_hidden_states = encoder_hidden_states.squeeze(0)
|
|
if human_num == 1:
|
|
return super().forward(x, encoder_hidden_states, shape, dtype=dtype)
|
|
N_t, _, _ = shape
|
|
x = rearrange(x, "B (N_t S) C -> (B N_t) S C", N_t=N_t)
|
|
B, N, C = x.shape
|
|
q = self.q_linear(x)
|
|
q_shape = (B, N, self.num_heads, self.head_dim)
|
|
q = q.view(q_shape).permute((0, 2, 1, 3))
|
|
if self.qk_norm:
|
|
q = self.q_norm(q)
|
|
max_values = x_ref_attn_map.max(1).values[:, None, None]
|
|
min_values = x_ref_attn_map.min(1).values[:, None, None]
|
|
max_min_values = torch.cat([max_values, min_values], dim=2)
|
|
human1_max_value, human1_min_value = max_min_values[0, :, 0].max(), max_min_values[0, :, 1].min()
|
|
human2_max_value, human2_min_value = max_min_values[1, :, 0].max(), max_min_values[1, :, 1].min()
|
|
human1 = normalize_and_scale(x_ref_attn_map[0], (human1_min_value, human1_max_value), (self.rope_h1[0], self.rope_h1[1]))
|
|
human2 = normalize_and_scale(x_ref_attn_map[1], (human2_min_value, human2_max_value), (self.rope_h2[0], self.rope_h2[1]))
|
|
back = torch.full((x_ref_attn_map.size(1),), self.rope_bak, dtype=human1.dtype).to(human1.device)
|
|
max_indices = x_ref_attn_map.argmax(dim=0)
|
|
normalized_map = torch.stack([human1, human2, back], dim=1)
|
|
normalized_pos = normalized_map[range(x_ref_attn_map.size(1)), max_indices]
|
|
q = rearrange(q, "(B N_t) H S C -> B H (N_t S) C", N_t=N_t)
|
|
q = self.rope_1d(q, normalized_pos)
|
|
q = rearrange(q, "B H (N_t S) C -> (B N_t) H S C", N_t=N_t)
|
|
_, N_a, _ = encoder_hidden_states.shape
|
|
encoder_kv = self.kv_linear(encoder_hidden_states)
|
|
encoder_kv_shape = (B, N_a, 2, self.num_heads, self.head_dim)
|
|
encoder_kv = encoder_kv.view(encoder_kv_shape).permute((2, 0, 3, 1, 4))
|
|
encoder_k, encoder_v = encoder_kv.unbind(0)
|
|
if self.qk_norm:
|
|
encoder_k = self.add_k_norm(encoder_k)
|
|
per_frame = torch.zeros(N_a, dtype=encoder_k.dtype).to(encoder_k.device)
|
|
per_frame[:per_frame.size(0)//2] = (self.rope_h1[0] + self.rope_h1[1]) / 2
|
|
per_frame[per_frame.size(0)//2:] = (self.rope_h2[0] + self.rope_h2[1]) / 2
|
|
encoder_pos = torch.concat([per_frame]*N_t, dim=0)
|
|
encoder_k = rearrange(encoder_k, "(B N_t) H S C -> B H (N_t S) C", N_t=N_t)
|
|
encoder_k = self.rope_1d(encoder_k, encoder_pos)
|
|
encoder_k = rearrange(encoder_k, "B H (N_t S) C -> (B N_t) H S C", N_t=N_t)
|
|
q = rearrange(q, "B H M K -> B M H K")
|
|
encoder_k = rearrange(encoder_k, "B H M K -> B M H K")
|
|
encoder_v = rearrange(encoder_v, "B H M K -> B M H K")
|
|
|
|
# Use attention from attention_utils with dtype conversion
|
|
x = attention(q=q.to(dtype), k=encoder_k.to(dtype), v=encoder_v.to(dtype), attention_type="FLASH_ATTENTION")
|
|
x = rearrange(x, "B M H K -> B H M K")
|
|
x_output_shape = (B, N, C)
|
|
x = x.transpose(1, 2)
|
|
x = x.reshape(x_output_shape)
|
|
x = self.proj(x)
|
|
x = self.proj_drop(x)
|
|
x = rearrange(x, "(B N_t) S C -> B (N_t S) C", N_t=N_t)
|
|
return x
|
|
|
|
|
|
class WanSelfAttention(nn.Module):
|
|
|
|
def __init__(self,
|
|
dim,
|
|
num_heads,
|
|
window_size=(-1, -1),
|
|
qk_norm=True,
|
|
eps=1e-6):
|
|
assert dim % num_heads == 0
|
|
super().__init__()
|
|
self.dim = dim
|
|
self.num_heads = num_heads
|
|
self.head_dim = dim // num_heads
|
|
self.window_size = window_size
|
|
self.qk_norm = qk_norm
|
|
self.eps = eps
|
|
|
|
# Layers
|
|
self.q = nn.Linear(dim, dim)
|
|
self.k = nn.Linear(dim, dim)
|
|
self.v = nn.Linear(dim, dim)
|
|
self.o = nn.Linear(dim, dim)
|
|
self.norm_q = WanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity()
|
|
self.norm_k = WanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity()
|
|
|
|
def _get_attn_map_with_target(self, visual_q, ref_k, shape, ref_target_masks=None, split_num=2):
|
|
"""Helper method for computing attention map, used by xfuser."""
|
|
return get_attn_map_with_target(visual_q, ref_k, shape, ref_target_masks, split_num)
|
|
|
|
def forward(self, x, seq_lens, grid_sizes, freqs, ref_target_masks=None, dtype=torch.bfloat16, t=0):
|
|
r"""
|
|
Args:
|
|
x (`Tensor`):
|
|
Input tensor with shape [B, L, C]
|
|
seq_lens (`Tensor`):
|
|
Sequence lengths
|
|
grid_sizes (`Tensor`):
|
|
Grid sizes [(N_t, N_h, N_w), ...]
|
|
freqs (`Tensor`):
|
|
RoPE frequencies
|
|
ref_target_masks (`Tensor`, *optional*):
|
|
Reference target masks for attention map computation
|
|
dtype (`torch.dtype`, *optional*, defaults to torch.bfloat16):
|
|
Output dtype to match transformer precision
|
|
t (`int`, *optional*, defaults to 0):
|
|
Timestep (unused, kept for API compatibility)
|
|
"""
|
|
b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim
|
|
|
|
# Query, key, value function
|
|
def qkv_fn(x):
|
|
q = self.norm_q(self.q(x)).view(b, s, n, d)
|
|
k = self.norm_k(self.k(x)).view(b, s, n, d)
|
|
v = self.v(x).view(b, s, n, d)
|
|
return q, k, v
|
|
q, k, v = qkv_fn(x)
|
|
|
|
q = rope_apply(q, grid_sizes, freqs)
|
|
k = rope_apply(k, grid_sizes, freqs)
|
|
|
|
x = attention(
|
|
q.to(dtype),
|
|
k.to(dtype),
|
|
v=v.to(dtype),
|
|
k_lens=seq_lens,
|
|
window_size=self.window_size
|
|
)
|
|
|
|
# Output
|
|
x = x.flatten(2)
|
|
x = self.o(x)
|
|
|
|
# Compute attention map for audio cross-attention
|
|
with torch.no_grad():
|
|
x_ref_attn_map = get_attn_map_with_target(q, k, grid_sizes[0],
|
|
ref_target_masks=ref_target_masks)
|
|
|
|
return x, x_ref_attn_map
|
|
|
|
|
|
class WanAttentionBlock(nn.Module):
|
|
r"""
|
|
Attention block with audio cross-attention support.
|
|
"""
|
|
|
|
def __init__(self,
|
|
cross_attn_type,
|
|
dim,
|
|
ffn_dim,
|
|
num_heads,
|
|
window_size=(-1, -1),
|
|
qk_norm=True,
|
|
cross_attn_norm=False,
|
|
eps=1e-6,
|
|
output_dim=768,
|
|
context_tokens=32,
|
|
norm_input_visual=True,
|
|
class_range=24,
|
|
class_interval=4):
|
|
r"""
|
|
Args:
|
|
cross_attn_type (`str`):
|
|
Cross-attention type (unused)
|
|
dim (`int`):
|
|
Transformer dimension
|
|
ffn_dim (`int`):
|
|
Feed-forward network dimension
|
|
num_heads (`int`):
|
|
Number of attention heads
|
|
window_size (`tuple`, *optional*, defaults to (-1, -1)):
|
|
Window size for windowed attention
|
|
qk_norm (`bool`, *optional*, defaults to True):
|
|
Whether to apply QK normalization
|
|
cross_attn_norm (`bool`, *optional*, defaults to False):
|
|
Whether to apply cross-attention normalization
|
|
eps (`float`, *optional*, defaults to 1e-6):
|
|
Epsilon for layer normalization
|
|
output_dim (`int`, *optional*, defaults to 768):
|
|
Output dimension for audio projection
|
|
context_tokens (`int`, *optional*, defaults to 32):
|
|
Number of context tokens for audio
|
|
norm_input_visual (`bool`, *optional*, defaults to True):
|
|
Whether to normalize visual input
|
|
class_range (`int`, *optional*, defaults to 24):
|
|
Range for positional embedding
|
|
class_interval (`int`, *optional*, defaults to 4):
|
|
Interval for positional embedding
|
|
"""
|
|
super().__init__()
|
|
self.dim = dim
|
|
self.ffn_dim = ffn_dim
|
|
self.num_heads = num_heads
|
|
self.window_size = window_size
|
|
self.qk_norm = qk_norm
|
|
self.cross_attn_norm = cross_attn_norm
|
|
self.eps = eps
|
|
|
|
# Multi-GPU sequence parallel attributes
|
|
self.sp_world_size = 1
|
|
self.sp_world_rank = 0
|
|
self.all_gather = None
|
|
|
|
# Layers
|
|
self.norm1 = WanLayerNorm(dim, eps)
|
|
self.self_attn = WanSelfAttention(dim, num_heads, window_size, qk_norm, eps)
|
|
self.norm3 = WanLayerNorm(
|
|
dim, eps,
|
|
elementwise_affine=True) if cross_attn_norm else nn.Identity()
|
|
self.cross_attn = WanI2VCrossAttention(dim,
|
|
num_heads,
|
|
(-1, -1),
|
|
qk_norm,
|
|
eps)
|
|
self.norm2 = WanLayerNorm(dim, eps)
|
|
self.ffn = nn.Sequential(
|
|
nn.Linear(dim, ffn_dim), nn.GELU(approximate='tanh'),
|
|
nn.Linear(ffn_dim, dim))
|
|
|
|
# Modulation
|
|
self.modulation = nn.Parameter(torch.randn(1, 6, dim) / dim**0.5)
|
|
|
|
# Init audio module
|
|
self.audio_cross_attn = SingleStreamMutiAttention(
|
|
dim=dim,
|
|
encoder_hidden_states_dim=output_dim,
|
|
num_heads=num_heads,
|
|
qk_norm=False,
|
|
qkv_bias=True,
|
|
eps=eps,
|
|
norm_layer=WanRMSNorm,
|
|
class_range=class_range,
|
|
class_interval=class_interval
|
|
)
|
|
self.norm_x = WanLayerNorm(dim, eps, elementwise_affine=True) if norm_input_visual else nn.Identity()
|
|
|
|
|
|
def forward(
|
|
self,
|
|
x,
|
|
e,
|
|
seq_lens,
|
|
grid_sizes,
|
|
freqs,
|
|
context,
|
|
context_lens,
|
|
audio_embedding=None,
|
|
ref_target_masks=None,
|
|
human_num=None,
|
|
dtype=torch.bfloat16,
|
|
t=0,
|
|
):
|
|
r"""
|
|
Args:
|
|
x (`Tensor`):
|
|
Input tensor with shape [B, L, C]
|
|
e (`Tensor`):
|
|
Time embedding modulation with shape [B, 6, C]
|
|
seq_lens (`Tensor`):
|
|
Sequence lengths with shape [B]
|
|
grid_sizes (`Tensor`):
|
|
Grid sizes with shape [B, 3], contains (F, H, W)
|
|
freqs (`Tensor`):
|
|
RoPE frequencies
|
|
context (`Tensor`):
|
|
Text context embeddings with shape [B, L_context, C]
|
|
context_lens (`Tensor`, *optional*):
|
|
Context lengths with shape [B]
|
|
audio_embedding (`Tensor`, *optional*):
|
|
Audio embeddings for cross-attention
|
|
ref_target_masks (`Tensor`, *optional*):
|
|
Reference target masks
|
|
human_num (`int`, *optional*):
|
|
Number of humans in the scene
|
|
dtype (`torch.dtype`, *optional*, defaults to torch.bfloat16):
|
|
Output dtype to match transformer precision
|
|
t (`int`, *optional*, defaults to 0):
|
|
Timestep
|
|
"""
|
|
# Modulation
|
|
e = (self.modulation + e).chunk(6, dim=1)
|
|
|
|
# Self-attention with modulation
|
|
temp_x = self.norm1(x) * (1 + e[1]) + e[0]
|
|
temp_x = temp_x.to(dtype)
|
|
|
|
y, x_ref_attn_map = self.self_attn(
|
|
temp_x,
|
|
seq_lens,
|
|
grid_sizes,
|
|
freqs,
|
|
ref_target_masks=ref_target_masks,
|
|
dtype=dtype,
|
|
t=t,
|
|
)
|
|
x = x + y * e[2]
|
|
|
|
# Cross-attention and audio cross-attention with modulation
|
|
def cross_attn_ffn(x, context, context_lens, e):
|
|
# Cross-attention: attend to text context
|
|
x = x + self.cross_attn(self.norm3(x).to(x.dtype), context, context_lens, dtype=dtype)
|
|
|
|
# Cross-attention for audio
|
|
# For multi-GPU: audio_cross_attn expects full sequence, so all_gather x first
|
|
if self.sp_world_size > 1 and self.all_gather is not None:
|
|
# All gather x to get full sequence for audio cross attention
|
|
x_full = self.all_gather(x, dim=1)
|
|
x_a_full = self.audio_cross_attn(
|
|
self.norm_x(x_full),
|
|
encoder_hidden_states=audio_embedding,
|
|
shape=grid_sizes[0],
|
|
x_ref_attn_map=x_ref_attn_map,
|
|
human_num=human_num,
|
|
dtype=dtype,
|
|
)
|
|
# Chunk result back to local rank
|
|
x_a = torch.chunk(x_a_full, self.sp_world_size, dim=1)[self.sp_world_rank]
|
|
else:
|
|
x_a = self.audio_cross_attn(
|
|
self.norm_x(x),
|
|
encoder_hidden_states=audio_embedding,
|
|
shape=grid_sizes[0],
|
|
x_ref_attn_map=x_ref_attn_map,
|
|
human_num=human_num,
|
|
dtype=dtype,
|
|
)
|
|
x = x + x_a
|
|
|
|
# FFN with modulation
|
|
temp_x = self.norm2(x) * (1 + e[4]) + e[3]
|
|
temp_x = temp_x.to(dtype)
|
|
|
|
y = self.ffn(temp_x)
|
|
x = x + y * e[5]
|
|
return x
|
|
|
|
x = cross_attn_ffn(x, context, context_lens, e)
|
|
return x
|
|
|
|
|
|
class AudioProjModel(ModelMixin, ConfigMixin):
|
|
r"""
|
|
Audio projection model to process audio embeddings.
|
|
"""
|
|
def __init__(
|
|
self,
|
|
seq_len=5,
|
|
seq_len_vf=12,
|
|
blocks=12,
|
|
channels=768,
|
|
intermediate_dim=512,
|
|
output_dim=768,
|
|
context_tokens=32,
|
|
norm_output_audio=False,
|
|
):
|
|
r"""
|
|
Args:
|
|
seq_len (`int`, *optional*, defaults to 5):
|
|
Sequence length for first frame audio
|
|
seq_len_vf (`int`, *optional*, defaults to 12):
|
|
Sequence length for subsequent frames audio
|
|
blocks (`int`, *optional*, defaults to 12):
|
|
Number of wav2vec blocks
|
|
channels (`int`, *optional*, defaults to 768):
|
|
Number of channels per block
|
|
intermediate_dim (`int`, *optional*, defaults to 512):
|
|
Intermediate projection dimension
|
|
output_dim (`int`, *optional*, defaults to 768):
|
|
Output dimension
|
|
context_tokens (`int`, *optional*, defaults to 32):
|
|
Number of context tokens
|
|
norm_output_audio (`bool`, *optional*, defaults to False):
|
|
Whether to apply layer normalization to output
|
|
"""
|
|
super().__init__()
|
|
|
|
self.seq_len = seq_len
|
|
self.blocks = blocks
|
|
self.channels = channels
|
|
self.input_dim = seq_len * blocks * channels
|
|
self.input_dim_vf = seq_len_vf * blocks * channels
|
|
self.intermediate_dim = intermediate_dim
|
|
self.context_tokens = context_tokens
|
|
self.output_dim = output_dim
|
|
|
|
# Define multiple linear layers
|
|
self.proj1 = nn.Linear(self.input_dim, intermediate_dim)
|
|
self.proj1_vf = nn.Linear(self.input_dim_vf, intermediate_dim)
|
|
self.proj2 = nn.Linear(intermediate_dim, intermediate_dim)
|
|
self.proj3 = nn.Linear(intermediate_dim, context_tokens * output_dim)
|
|
self.norm = nn.LayerNorm(output_dim) if norm_output_audio else nn.Identity()
|
|
|
|
def forward(self, audio_embeds, audio_embeds_vf, dtype=torch.bfloat16):
|
|
r"""
|
|
Args:
|
|
audio_embeds (`Tensor`):
|
|
First frame audio with shape [B, 1, seq_len, blocks, channels]
|
|
audio_embeds_vf (`Tensor`):
|
|
Subsequent frames audio with shape [B, F-1, seq_len_vf, blocks, channels]
|
|
dtype (`torch.dtype`, *optional*, defaults to torch.bfloat16):
|
|
Output dtype to match transformer precision
|
|
|
|
Returns:
|
|
`Tensor`:
|
|
Context tokens with shape [B, F, context_tokens, output_dim]
|
|
"""
|
|
# Ensure input dtype matches target dtype
|
|
if audio_embeds.dtype != dtype:
|
|
audio_embeds = audio_embeds.to(dtype=dtype)
|
|
if audio_embeds_vf.dtype != dtype:
|
|
audio_embeds_vf = audio_embeds_vf.to(dtype=dtype)
|
|
|
|
video_length = audio_embeds.shape[1] + audio_embeds_vf.shape[1]
|
|
B = audio_embeds.shape[0]
|
|
|
|
# Process first frame audio
|
|
audio_embeds = rearrange(audio_embeds, "bz f w b c -> (bz f) w b c")
|
|
batch_size, window_size, blocks, channels = audio_embeds.shape
|
|
audio_embeds = audio_embeds.reshape(batch_size, window_size * blocks * channels)
|
|
|
|
# Process latter frame audio
|
|
audio_embeds_vf = rearrange(audio_embeds_vf, "bz f w b c -> (bz f) w b c")
|
|
batch_size_vf, window_size_vf, blocks_vf, channels_vf = audio_embeds_vf.shape
|
|
audio_embeds_vf = audio_embeds_vf.reshape(batch_size_vf, window_size_vf * blocks_vf * channels_vf)
|
|
|
|
# First projection
|
|
audio_embeds = torch.relu(self.proj1(audio_embeds))
|
|
audio_embeds_vf = torch.relu(self.proj1_vf(audio_embeds_vf))
|
|
audio_embeds = rearrange(audio_embeds, "(bz f) c -> bz f c", bz=B)
|
|
audio_embeds_vf = rearrange(audio_embeds_vf, "(bz f) c -> bz f c", bz=B)
|
|
audio_embeds_c = torch.concat([audio_embeds, audio_embeds_vf], dim=1)
|
|
batch_size_c, N_t, C_a = audio_embeds_c.shape
|
|
audio_embeds_c = audio_embeds_c.view(batch_size_c*N_t, C_a)
|
|
|
|
# Second projection
|
|
audio_embeds_c = torch.relu(self.proj2(audio_embeds_c))
|
|
|
|
context_tokens = self.proj3(audio_embeds_c).reshape(batch_size_c*N_t, self.context_tokens, self.output_dim)
|
|
|
|
# Normalization and reshape
|
|
context_tokens = self.norm(context_tokens)
|
|
context_tokens = rearrange(context_tokens, "(bz f) m c -> bz f m c", f=video_length)
|
|
|
|
# Ensure output dtype matches transformer precision
|
|
if context_tokens.dtype != dtype:
|
|
context_tokens = context_tokens.to(dtype=dtype)
|
|
|
|
return context_tokens
|
|
|
|
|
|
class InfiniteTalkTransformer3DModel(WanTransformer3DModel):
|
|
r"""
|
|
InfiniteTalk Transformer 3D model with audio integration.
|
|
"""
|
|
ignore_for_config = [
|
|
'patch_size', 'cross_attn_norm', 'qk_norm', 'text_dim', 'window_size'
|
|
]
|
|
_no_split_modules = ['WanAttentionBlock']
|
|
|
|
@register_to_config
|
|
def __init__(
|
|
self,
|
|
model_type='i2v',
|
|
patch_size=(1, 2, 2),
|
|
text_len=512,
|
|
in_dim=16,
|
|
dim=2048,
|
|
ffn_dim=8192,
|
|
freq_dim=256,
|
|
text_dim=4096,
|
|
out_dim=16,
|
|
num_heads=16,
|
|
num_layers=32,
|
|
window_size=(-1, -1),
|
|
qk_norm=True,
|
|
cross_attn_norm=True,
|
|
eps=1e-6,
|
|
# Audio params
|
|
audio_window=5,
|
|
intermediate_dim=512,
|
|
output_dim=768,
|
|
context_tokens=32,
|
|
vae_scale=4, # VAE time downsample scale
|
|
|
|
norm_input_visual=True,
|
|
norm_output_audio=True,
|
|
weight_init=True,
|
|
# Custom block params
|
|
class_range=24,
|
|
class_interval=4,
|
|
):
|
|
r"""
|
|
Initialize the InfiniteTalk diffusion model backbone.
|
|
|
|
Args:
|
|
model_type (`str`, *optional*, defaults to 'i2v'):
|
|
Model type, must be 'i2v' for InfiniteTalk
|
|
patch_size (`tuple`, *optional*, defaults to (1, 2, 2)):
|
|
Patch embedding size (t, h, w)
|
|
text_len (`int`, *optional*, defaults to 512):
|
|
Maximum text sequence length
|
|
in_dim (`int`, *optional*, defaults to 16):
|
|
Input channels
|
|
dim (`int`, *optional*, defaults to 2048):
|
|
Transformer dimension
|
|
ffn_dim (`int`, *optional*, defaults to 8192):
|
|
Feed-forward network dimension
|
|
freq_dim (`int`, *optional*, defaults to 256):
|
|
Frequency dimension for time embedding
|
|
text_dim (`int`, *optional*, defaults to 4096):
|
|
Text embedding dimension
|
|
out_dim (`int`, *optional*, defaults to 16):
|
|
Output channels
|
|
num_heads (`int`, *optional*, defaults to 16):
|
|
Number of attention heads
|
|
num_layers (`int`, *optional*, defaults to 32):
|
|
Number of transformer layers
|
|
window_size (`tuple`, *optional*, defaults to (-1, -1)):
|
|
Window size for windowed attention
|
|
qk_norm (`bool`, *optional*, defaults to True):
|
|
Whether to apply QK normalization
|
|
cross_attn_norm (`bool`, *optional*, defaults to True):
|
|
Whether to apply cross-attention normalization
|
|
eps (`float`, *optional*, defaults to 1e-6):
|
|
Epsilon for layer normalization
|
|
audio_window (`int`, *optional*, defaults to 5):
|
|
Audio window size
|
|
intermediate_dim (`int`, *optional*, defaults to 512):
|
|
Intermediate dimension for audio projection
|
|
output_dim (`int`, *optional*, defaults to 768):
|
|
Output dimension for audio projection
|
|
context_tokens (`int`, *optional*, defaults to 32):
|
|
Number of context tokens for audio
|
|
vae_scale (`int`, *optional*, defaults to 4):
|
|
VAE temporal downsample factor
|
|
norm_input_visual (`bool`, *optional*, defaults to True):
|
|
Whether to normalize visual input
|
|
norm_output_audio (`bool`, *optional*, defaults to True):
|
|
Whether to normalize audio output
|
|
weight_init (`bool`, *optional*, defaults to True):
|
|
Whether to initialize weights
|
|
class_range (`int`, *optional*, defaults to 24):
|
|
Range for positional embedding
|
|
class_interval (`int`, *optional*, defaults to 4):
|
|
Interval for positional embedding
|
|
"""
|
|
# Call parent class initialization
|
|
super().__init__(
|
|
model_type=model_type,
|
|
patch_size=patch_size,
|
|
text_len=text_len,
|
|
in_dim=in_dim,
|
|
dim=dim,
|
|
ffn_dim=ffn_dim,
|
|
freq_dim=freq_dim,
|
|
text_dim=text_dim,
|
|
out_dim=out_dim,
|
|
num_heads=num_heads,
|
|
num_layers=num_layers,
|
|
window_size=window_size,
|
|
qk_norm=qk_norm,
|
|
cross_attn_norm=cross_attn_norm,
|
|
eps=eps,
|
|
cross_attn_type='i2v_cross_attn',
|
|
)
|
|
|
|
assert model_type == 'i2v', 'MultiTalk model requires your model_type is i2v.'
|
|
|
|
# Audio-specific parameters
|
|
self.norm_output_audio = norm_output_audio
|
|
self.audio_window = audio_window
|
|
self.intermediate_dim = intermediate_dim
|
|
self.vae_scale = vae_scale
|
|
|
|
# Replace blocks with audio-enhanced version
|
|
self.blocks = nn.ModuleList([
|
|
WanAttentionBlock(cross_attn_type='i2v_cross_attn', dim=dim, ffn_dim=ffn_dim,
|
|
num_heads=num_heads, window_size=window_size, qk_norm=qk_norm,
|
|
cross_attn_norm=cross_attn_norm, eps=eps,
|
|
output_dim=output_dim, context_tokens=context_tokens,
|
|
norm_input_visual=norm_input_visual,
|
|
class_range=class_range, class_interval=class_interval)
|
|
for _ in range(num_layers)
|
|
])
|
|
|
|
# Initialize audio adapter
|
|
# Blocks should match the number of hidden layers from audio encoder
|
|
# Wav2Vec2 base has 12 transformer layers, so blocks=12
|
|
self.audio_proj = AudioProjModel(
|
|
seq_len=audio_window,
|
|
seq_len_vf=audio_window+vae_scale-1,
|
|
blocks=12, # Match Wav2Vec2 hidden layers
|
|
channels=768,
|
|
intermediate_dim=intermediate_dim,
|
|
output_dim=output_dim,
|
|
context_tokens=context_tokens,
|
|
norm_output_audio=norm_output_audio,
|
|
)
|
|
|
|
# Initialize weights
|
|
if weight_init:
|
|
self.init_weights()
|
|
|
|
def enable_multi_gpus_inference(self):
|
|
r"""
|
|
Enable multi-GPU inference using sequence parallel.
|
|
"""
|
|
self.sp_world_size = get_sequence_parallel_world_size()
|
|
self.sp_world_rank = get_sequence_parallel_rank()
|
|
self.all_gather = get_sp_group().all_gather
|
|
|
|
# Replace self_attn forward with xfuser version for all blocks
|
|
# And pass sp parameters to each block for audio_cross_attn
|
|
for block in self.blocks:
|
|
block.self_attn.forward = types.MethodType(
|
|
usp_attn_infinitetalk_forward, block.self_attn)
|
|
# Pass sp parameters to block for audio_cross_attn multi-GPU support
|
|
block.sp_world_size = self.sp_world_size
|
|
block.sp_world_rank = self.sp_world_rank
|
|
block.all_gather = self.all_gather
|
|
|
|
def forward(
|
|
self,
|
|
x,
|
|
t,
|
|
context,
|
|
seq_len,
|
|
clip_fea=None,
|
|
y=None,
|
|
audio=None,
|
|
ref_target_masks=None,
|
|
cond_flag=True,
|
|
):
|
|
r"""
|
|
Forward pass through the diffusion model.
|
|
|
|
Args:
|
|
x (`List[Tensor]`):
|
|
List of input video tensors, each with shape [C_in, F, H, W]
|
|
t (`Tensor`):
|
|
Diffusion timesteps tensor of shape [B]
|
|
context (`List[Tensor]`):
|
|
List of text embeddings each with shape [L, C]
|
|
seq_len (`int`):
|
|
Maximum sequence length for positional encoding
|
|
clip_fea (`Tensor`, *optional*):
|
|
CLIP image features for image-to-video mode
|
|
y (`List[Tensor]`, *optional*):
|
|
Conditional video inputs for image-to-video mode, same shape as x
|
|
audio (`Tensor`, *optional*):
|
|
Audio wav2vec features
|
|
ref_target_masks (`Tensor`, *optional*):
|
|
Reference target masks for attention
|
|
cond_flag (`bool`, *optional*, defaults to True):
|
|
Whether this is conditional forward pass
|
|
|
|
Returns:
|
|
`List[Tensor]`:
|
|
List of denoised video tensors with original input shapes [C_out, F, H / 8, W / 8]
|
|
"""
|
|
# Get device and dtype
|
|
device = self.patch_embedding.weight.device
|
|
dtype = x.dtype
|
|
if self.freqs.device != device and torch.device(type="meta") != device:
|
|
self.freqs = self.freqs.to(device)
|
|
x[0] = x[0].to(dtype)
|
|
|
|
# Get size
|
|
_, T, H, W = x[0].shape
|
|
N_t = T // self.patch_size[0]
|
|
N_h = H // self.patch_size[1]
|
|
N_w = W // self.patch_size[2]
|
|
|
|
# Concatenate condition video to input (for I2V)
|
|
if y is not None:
|
|
x = [torch.cat([u, v], dim=0) for u, v in zip(x, y)]
|
|
|
|
# Patch embedding: convert video to sequence of patches
|
|
x = [self.patch_embedding(u.unsqueeze(0)) for u in x]
|
|
grid_sizes = torch.stack(
|
|
[torch.tensor(u.shape[2:], dtype=torch.long) for u in x])
|
|
x = [u.flatten(2).transpose(1, 2) for u in x]
|
|
seq_lens = torch.tensor([u.size(1) for u in x], dtype=torch.long)
|
|
|
|
# Padding for multi-GPU inference
|
|
if hasattr(self, 'sp_world_size') and self.sp_world_size > 1:
|
|
seq_len = int(math.ceil(seq_len / self.sp_world_size)) * self.sp_world_size
|
|
assert seq_lens.max() <= seq_len
|
|
x = torch.cat([
|
|
torch.cat([u, u.new_zeros(1, seq_len - u.size(1), u.size(2))],
|
|
dim=1) for u in x
|
|
])
|
|
|
|
# Time embeddings
|
|
e = self.time_embedding(
|
|
sinusoidal_embedding_1d(self.freq_dim, t).float()).to(dtype)
|
|
e0 = self.time_projection(e).unflatten(1, (6, self.dim))
|
|
|
|
# Text embeddings
|
|
context_lens = None
|
|
context = self.text_embedding(
|
|
torch.stack([
|
|
torch.cat(
|
|
[u, u.new_zeros(self.text_len - u.size(0), u.size(1))])
|
|
for u in context
|
|
]))
|
|
|
|
# CLIP embeddings
|
|
if clip_fea is not None:
|
|
context_clip = self.img_emb(clip_fea)
|
|
context = torch.concat([context_clip, context], dim=1)
|
|
|
|
# Audio processing
|
|
audio_cond = audio.to(device=x.device, dtype=x.dtype)
|
|
first_frame_audio_emb_s = audio_cond[:, :1, ...]
|
|
latter_frame_audio_emb = audio_cond[:, 1:, ...]
|
|
latter_frame_audio_emb = rearrange(latter_frame_audio_emb, "b (n_t n) w s c -> b n_t n w s c", n=self.vae_scale)
|
|
middle_index = self.audio_window // 2
|
|
latter_first_frame_audio_emb = latter_frame_audio_emb[:, :, :1, :middle_index+1, ...]
|
|
latter_first_frame_audio_emb = rearrange(latter_first_frame_audio_emb, "b n_t n w s c -> b n_t (n w) s c")
|
|
latter_last_frame_audio_emb = latter_frame_audio_emb[:, :, -1:, middle_index:, ...]
|
|
latter_last_frame_audio_emb = rearrange(latter_last_frame_audio_emb, "b n_t n w s c -> b n_t (n w) s c")
|
|
latter_middle_frame_audio_emb = latter_frame_audio_emb[:, :, 1:-1, middle_index:middle_index+1, ...]
|
|
latter_middle_frame_audio_emb = rearrange(latter_middle_frame_audio_emb, "b n_t n w s c -> b n_t (n w) s c")
|
|
latter_frame_audio_emb_s = torch.concat([latter_first_frame_audio_emb, latter_middle_frame_audio_emb, latter_last_frame_audio_emb], dim=2)
|
|
|
|
# Project audio to context tokens
|
|
audio_embedding = self.audio_proj(first_frame_audio_emb_s, latter_frame_audio_emb_s, dtype=dtype)
|
|
human_num = len(audio_embedding)
|
|
audio_embedding = torch.concat(audio_embedding.split(1), dim=2)
|
|
|
|
# Convert ref_target_masks to token_ref_target_masks
|
|
if ref_target_masks is not None:
|
|
ref_target_masks = ref_target_masks.unsqueeze(0)
|
|
token_ref_target_masks = nn.functional.interpolate(ref_target_masks, size=(N_h, N_w), mode='nearest')
|
|
token_ref_target_masks = token_ref_target_masks.squeeze(0)
|
|
token_ref_target_masks = (token_ref_target_masks > 0)
|
|
token_ref_target_masks = token_ref_target_masks.view(token_ref_target_masks.shape[0], -1)
|
|
token_ref_target_masks = token_ref_target_masks.to(x.dtype)
|
|
|
|
# Context Parallel: split input across GPUs
|
|
if hasattr(self, 'sp_world_size') and self.sp_world_size > 1:
|
|
x = torch.chunk(x, self.sp_world_size, dim=1)[self.sp_world_rank]
|
|
e0 = torch.chunk(e0, self.sp_world_size, dim=1)[self.sp_world_rank]
|
|
e = torch.chunk(e, self.sp_world_size, dim=1)[self.sp_world_rank]
|
|
|
|
# TeaCache
|
|
if self.teacache is not None:
|
|
if cond_flag:
|
|
modulated_inp = e0
|
|
skip_flag = self.teacache.cnt < self.teacache.num_skip_start_steps
|
|
if skip_flag:
|
|
self.should_calc = True
|
|
self.teacache.accumulated_rel_l1_distance = 0
|
|
else:
|
|
if cond_flag:
|
|
rel_l1_distance = self.teacache.compute_rel_l1_distance(self.teacache.previous_modulated_input, modulated_inp)
|
|
self.teacache.accumulated_rel_l1_distance += self.teacache.rescale_func(rel_l1_distance)
|
|
if self.teacache.accumulated_rel_l1_distance < self.teacache.rel_l1_thresh:
|
|
self.should_calc = False
|
|
else:
|
|
self.should_calc = True
|
|
self.teacache.accumulated_rel_l1_distance = 0
|
|
self.teacache.previous_modulated_input = modulated_inp
|
|
self.teacache.should_calc = self.should_calc
|
|
else:
|
|
self.should_calc = self.teacache.should_calc
|
|
|
|
# Prepare checkpointing utilities
|
|
if torch.is_grad_enabled() and self.gradient_checkpointing:
|
|
def create_custom_forward(module):
|
|
def custom_forward(*inputs):
|
|
return module(*inputs)
|
|
return custom_forward
|
|
ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
|
|
|
|
# Main transformer loop
|
|
if self.teacache is not None:
|
|
if not self.should_calc:
|
|
# Skip: use cached residual
|
|
previous_residual = self.teacache.previous_residual_cond if cond_flag else self.teacache.previous_residual_uncond
|
|
x = x + previous_residual.to(x.device)[-x.size()[0]:,]
|
|
else:
|
|
ori_x = x.clone().cpu() if self.teacache.offload else x.clone()
|
|
|
|
for block in self.blocks:
|
|
if torch.is_grad_enabled() and self.gradient_checkpointing:
|
|
x = torch.utils.checkpoint.checkpoint(
|
|
create_custom_forward(block),
|
|
x,
|
|
e0,
|
|
seq_lens,
|
|
grid_sizes,
|
|
self.freqs,
|
|
context,
|
|
context_lens,
|
|
audio_embedding,
|
|
token_ref_target_masks,
|
|
human_num,
|
|
dtype,
|
|
t,
|
|
**ckpt_kwargs,
|
|
)
|
|
else:
|
|
x = block(
|
|
x,
|
|
e=e0,
|
|
seq_lens=seq_lens,
|
|
grid_sizes=grid_sizes,
|
|
freqs=self.freqs,
|
|
context=context,
|
|
context_lens=context_lens,
|
|
audio_embedding=audio_embedding,
|
|
ref_target_masks=token_ref_target_masks,
|
|
human_num=human_num,
|
|
dtype=dtype,
|
|
t=t,
|
|
)
|
|
|
|
if cond_flag:
|
|
self.teacache.previous_residual_cond = x.cpu() - ori_x if self.teacache.offload else x - ori_x
|
|
else:
|
|
self.teacache.previous_residual_uncond = x.cpu() - ori_x if self.teacache.offload else x - ori_x
|
|
else:
|
|
for block in self.blocks:
|
|
if torch.is_grad_enabled() and self.gradient_checkpointing:
|
|
x = torch.utils.checkpoint.checkpoint(
|
|
create_custom_forward(block),
|
|
x,
|
|
e0,
|
|
seq_lens,
|
|
grid_sizes,
|
|
self.freqs,
|
|
context,
|
|
context_lens,
|
|
audio_embedding,
|
|
token_ref_target_masks,
|
|
human_num,
|
|
dtype,
|
|
t,
|
|
**ckpt_kwargs,
|
|
)
|
|
else:
|
|
x = block(
|
|
x,
|
|
e=e0,
|
|
seq_lens=seq_lens,
|
|
grid_sizes=grid_sizes,
|
|
freqs=self.freqs,
|
|
context=context,
|
|
context_lens=context_lens,
|
|
audio_embedding=audio_embedding,
|
|
ref_target_masks=token_ref_target_masks,
|
|
human_num=human_num,
|
|
dtype=dtype,
|
|
t=t,
|
|
)
|
|
|
|
# Head: project to output space
|
|
if torch.is_grad_enabled() and self.gradient_checkpointing:
|
|
x = torch.utils.checkpoint.checkpoint(create_custom_forward(self.head), x, e, **ckpt_kwargs)
|
|
else:
|
|
x = self.head(x, e)
|
|
|
|
# Context Parallel: gather results from all GPUs
|
|
if hasattr(self, 'sp_world_size') and self.sp_world_size > 1:
|
|
x = self.all_gather(x, dim=1)
|
|
|
|
# Unpatchify: reconstruct video from patches
|
|
x = self.unpatchify(x, grid_sizes)
|
|
x = torch.stack(x)
|
|
|
|
# Increment teacache counter and reset if completed
|
|
if self.teacache is not None and cond_flag:
|
|
self.teacache.cnt += 1
|
|
if self.teacache.cnt == self.teacache.num_steps:
|
|
self.teacache.reset()
|
|
return x |