| |
| import math |
| from typing import Tuple, Optional |
|
|
| import torch |
| import torch.amp as amp |
| import torch.nn as nn |
| import torch.nn.functional as F |
|
|
| from diffusers.configuration_utils import ConfigMixin, register_to_config |
| from diffusers.models.modeling_utils import ModelMixin |
| from .attention import flash_attention |
| from torch.utils.checkpoint import checkpoint |
| from distributed_comms.communications import all_gather, all_to_all_4D |
| from distributed_comms.parallel_states import nccl_info, get_sequence_parallel_state |
|
|
|
|
| def gradient_checkpointing(module: nn.Module, *args, enabled: bool, **kwargs): |
| if enabled: |
| return checkpoint(module, *args, use_reentrant=False, **kwargs) |
| else: |
| return module(*args, **kwargs) |
|
|
|
|
| def sinusoidal_embedding_1d(dim, position): |
| |
| assert dim % 2 == 0 |
| half = dim // 2 |
| position = position.type(torch.float64) |
|
|
| |
| sinusoid = torch.outer( |
| position, torch.pow(10000, -torch.arange(half).to(position).div(half))) |
| x = torch.cat([torch.cos(sinusoid), torch.sin(sinusoid)], dim=1) |
| return x |
|
|
|
|
| @amp.autocast('cuda', enabled=False) |
| def rope_params(max_seq_len, dim, theta=10000, freqs_scaling=1.0): |
| assert dim % 2 == 0 |
| pos = torch.arange(max_seq_len) |
| freqs = 1.0 / torch.pow(theta, torch.arange(0, dim, 2).to(torch.float64).div(dim)) |
| freqs = freqs_scaling * freqs |
| freqs = torch.outer(pos, freqs) |
| freqs = torch.polar(torch.ones_like(freqs), freqs) |
| return freqs |
|
|
| @amp.autocast('cuda', enabled=False) |
| def rope_apply_1d(x, grid_sizes, freqs, offsets=None): |
| n, c = x.size(2), x.size(3) // 2 |
| c_rope = freqs.shape[1] |
| assert c_rope <= c, "RoPE dimensions cannot exceed half of hidden size" |
| |
| |
| output = [] |
| for i, (l, ) in enumerate(grid_sizes.tolist()): |
| offset = offsets[i] if offsets is not None else 0 |
| seq_len = l |
| |
| x_i = torch.view_as_complex(x[i, :seq_len].to(torch.float64).reshape( |
| seq_len, n, -1, 2)) |
| x_i_rope = x_i[:, :, :c_rope] * freqs[offset:offset+seq_len, None, :] |
| x_i_passthrough = x_i[:, :, c_rope:] |
| x_i = torch.cat([x_i_rope, x_i_passthrough], dim=2) |
|
|
| |
| x_i = torch.view_as_real(x_i).flatten(2) |
| x_i = torch.cat([x_i, x[i, seq_len:]]) |
|
|
| |
| output.append(x_i) |
| return torch.stack(output).bfloat16() |
|
|
| @amp.autocast('cuda', enabled=False) |
| def rope_apply_3d(x, grid_sizes, freqs, offsets=None): |
| n, c = x.size(2), x.size(3) // 2 |
|
|
| |
| freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1) |
| |
| |
| output = [] |
| for i, (f, h, w) in enumerate(grid_sizes.tolist()): |
| seq_len = f * h * w |
| offset = offsets[i].tolist() if offsets is not None else [0, 0, 0] |
| offset_f, offset_h, offset_w = offset |
|
|
| |
| x_i = torch.view_as_complex(x[i, :seq_len].to(torch.float64).reshape(seq_len, n, -1, 2)) |
| freqs_i = torch.cat([ |
| freqs[0][offset_f:offset_f+f].view(f, 1, 1, -1).expand(f, h, w, -1), |
| freqs[1][offset_h:offset_h+h].view(1, h, 1, -1).expand(f, h, w, -1), |
| freqs[2][offset_w:offset_w+w].view(1, 1, w, -1).expand(f, h, w, -1) |
| ], dim=-1).reshape(seq_len, 1, -1) |
|
|
| |
| x_i = torch.view_as_real(x_i * freqs_i).flatten(2) |
| x_i = torch.cat([x_i, x[i, seq_len:]]) |
|
|
| |
| output.append(x_i) |
| return torch.stack(output).bfloat16() |
|
|
| @amp.autocast('cuda', enabled=False) |
| def rope_apply(x, grid_sizes, freqs, offsets=None): |
| x_ndim = grid_sizes.shape[-1] |
| if x_ndim == 3: |
| return rope_apply_3d(x, grid_sizes, freqs, offsets) |
| else: |
| return rope_apply_1d(x, grid_sizes, freqs, offsets) |
|
|
| class ChannelLastConv1d(nn.Conv1d): |
|
|
| def forward(self, x: torch.Tensor) -> torch.Tensor: |
| x = x.permute(0, 2, 1) |
| x = super().forward(x) |
| x = x.permute(0, 2, 1) |
| return x |
|
|
|
|
| class ConvMLP(nn.Module): |
|
|
| def __init__( |
| self, |
| dim: int, |
| hidden_dim: int, |
| multiple_of: int = 256, |
| kernel_size: int = 3, |
| padding: int = 1, |
| ): |
| """ |
| Initialize the FeedForward module. |
| |
| Args: |
| dim (int): Input dimension. |
| hidden_dim (int): Hidden dimension of the feedforward layer. |
| multiple_of (int): Value to ensure hidden dimension is a multiple of this value. |
| |
| Attributes: |
| w1 (ColumnParallelLinear): Linear transformation for the first layer. |
| w2 (RowParallelLinear): Linear transformation for the second layer. |
| w3 (ColumnParallelLinear): Linear transformation for the third layer. |
| |
| """ |
| super().__init__() |
| hidden_dim = int(2 * hidden_dim / 3) |
| hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of) |
|
|
| self.w1 = ChannelLastConv1d(dim, |
| hidden_dim, |
| bias=False, |
| kernel_size=kernel_size, |
| padding=padding) |
| self.w2 = ChannelLastConv1d(hidden_dim, |
| dim, |
| bias=False, |
| kernel_size=kernel_size, |
| padding=padding) |
| self.w3 = ChannelLastConv1d(dim, |
| hidden_dim, |
| bias=False, |
| kernel_size=kernel_size, |
| padding=padding) |
|
|
| def forward(self, x): |
| return self.w2(F.silu(self.w1(x)) * self.w3(x)) |
|
|
| class WanRMSNorm(nn.Module): |
|
|
| def __init__(self, dim, eps=1e-5): |
| super().__init__() |
| self.dim = dim |
| self.eps = eps |
| self.weight = nn.Parameter(torch.ones(dim)) |
|
|
| def forward(self, x): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L, C] |
| """ |
| return self._norm(x.bfloat16()).type_as(x) * self.weight.bfloat16() |
|
|
| def _norm(self, x): |
| return x * torch.rsqrt(x.pow(2).mean(dim=-1, keepdim=True) + self.eps) |
|
|
|
|
| class WanLayerNorm(nn.LayerNorm): |
|
|
| def __init__(self, dim, eps=1e-6, elementwise_affine=False): |
| super().__init__(dim, elementwise_affine=elementwise_affine, eps=eps) |
|
|
| def forward(self, x): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L, C] |
| """ |
| return super().forward(x.bfloat16()).type_as(x) |
|
|
| class LoRALinearLayer(nn.Module): |
| def __init__( |
| self, |
| in_features: int, |
| out_features: int, |
| rank: int = 128, |
| device="cuda", |
| dtype: Optional[torch.dtype] = torch.float32, |
| ): |
| super().__init__() |
| self.down = nn.Linear(in_features, rank, bias=False, device=device, dtype=dtype) |
| self.up = nn.Linear(rank, out_features, bias=False, device=device, dtype=dtype) |
| self.rank = rank |
| self.out_features = out_features |
| self.in_features = in_features |
|
|
| nn.init.normal_(self.down.weight, std=1 / rank) |
| nn.init.zeros_(self.up.weight) |
|
|
| def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: |
| orig_dtype = hidden_states.dtype |
| dtype = self.down.weight.dtype |
|
|
| down_hidden_states = self.down(hidden_states.to(dtype)) |
| up_hidden_states = self.up(down_hidden_states) |
| return up_hidden_states.to(orig_dtype) |
|
|
|
|
| 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 |
|
|
| |
| 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() |
| |
| |
| self.use_sp = get_sequence_parallel_state() |
| if self.use_sp: |
| self.sp_size = nccl_info.sp_size |
| self.sp_rank = nccl_info.rank_within_group |
| assert self.num_heads % self.sp_size == 0, \ |
| f"Num heads {self.num_heads} must be divisible by sp_size {self.sp_size}" |
|
|
| def init_lora(self, self_lora=False, train=False, using_ip_emb=False): |
| dim = self.dim |
| |
| self.q_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.k_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.v_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.o_loras = LoRALinearLayer(dim, dim, rank=128) |
|
|
| requires_grad = train |
| for lora in [self.q_loras, self.k_loras, self.v_loras, self.o_loras]: |
| for param in lora.parameters(): |
| param.requires_grad = requires_grad |
| |
| |
| if using_ip_emb: |
| self.ip_embedding = nn.Linear(dim, dim, bias=True) |
| for param in self.ip_embedding.parameters(): |
| param.requires_grad = requires_grad |
| |
| |
| if self_lora: |
| self.s_q_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.s_k_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.s_v_loras = LoRALinearLayer(dim, dim, rank=128) |
| self.s_o_loras = LoRALinearLayer(dim, dim, rank=128) |
|
|
| requires_grad = train |
| for lora in [self.s_q_loras, self.s_k_loras, self.s_v_loras, self.s_o_loras]: |
| for param in lora.parameters(): |
| param.requires_grad = requires_grad |
| |
| |
| def qkv_fn(self, x): |
| b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim |
|
|
| 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 |
| |
| def qkv_fn_with_lora(self, x): |
| b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim |
|
|
| q = self.norm_q(self.q(x) + self.q_loras(x)).view(b, s, n, d) |
| k = self.norm_k(self.k(x) + self.k_loras(x)).view(b, s, n, d) |
| v = (self.v(x) + self.v_loras(x)).view(b, s, n, d) |
| return q, k, v |
|
|
| def forward(self, x, seq_lens, grid_sizes, freqs, x_ip=None, ip_grid_sizes=None, ip_freqs=None, ip_offsets=None, ip_emb=None): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L, C] |
| seq_lens(Tensor): Shape [B] |
| grid_sizes(Tensor): Shape [B, 3], the second dimension contains (F, H, W) or [B, 1] for (L,) |
| freqs(Tensor): Rope freqs, shape [1024, C / num_heads / 2] |
| """ |
| if ip_emb is not None: |
| |
| x = x + self.ip_embedding(ip_emb) |
| |
| q, k, v = self.qkv_fn(x) |
| if self.use_sp: |
| |
| q = all_to_all_4D(q, scatter_dim=2, gather_dim=1) |
| k = all_to_all_4D(k, scatter_dim=2, gather_dim=1) |
| v = all_to_all_4D(v, scatter_dim=2, gather_dim=1) |
|
|
| |
| if x_ip is not None and ip_grid_sizes is not None and ip_freqs is not None: |
| q_ip, k_ip, v_ip = self.qkv_fn_with_lora(x_ip) |
| if self.use_sp: |
| |
| q_ip = all_to_all_4D(q_ip, scatter_dim=2, gather_dim=1) |
| k_ip = all_to_all_4D(k_ip, scatter_dim=2, gather_dim=1) |
| v_ip = all_to_all_4D(v_ip, scatter_dim=2, gather_dim=1) |
| x_ip = flash_attention( |
| q=rope_apply(q_ip, ip_grid_sizes, ip_freqs, ip_offsets), |
| k=rope_apply(k_ip, ip_grid_sizes, ip_freqs, ip_offsets), |
| v=v_ip, |
| k_lens=ip_grid_sizes.prod(dim=1, keepdim=False)) |
| if self.use_sp: |
| |
| x_ip = all_to_all_4D(x_ip, scatter_dim=1, gather_dim=2) |
| x_ip = x_ip.flatten(2) |
| x_ip = self.o(x_ip) + self.o_loras(x_ip) |
| else: |
| q_ip = k_ip = v_ip = x_ip = None |
|
|
| ip_seq_lens = 0 |
| if k_ip is not None and v_ip is not None: |
| ip_seq_lens = ip_grid_sizes.prod(dim=1, keepdim=False) |
| |
| k = torch.cat([k, k_ip[:,:ip_seq_lens]], dim=1) |
| v = torch.cat([v, v_ip[:,:ip_seq_lens]], dim=1) |
| |
| x = flash_attention( |
| q=rope_apply(q, grid_sizes, freqs), |
| k=rope_apply(k, grid_sizes, freqs), |
| v=v, |
| k_lens=seq_lens + ip_seq_lens, |
| window_size=self.window_size) |
| if self.use_sp: |
| |
| x = all_to_all_4D(x, scatter_dim=1, gather_dim=2) |
| |
| x = x.flatten(2) |
| x = self.o(x) |
| return x, x_ip |
|
|
|
|
| class WanT2VCrossAttention(WanSelfAttention): |
| def qkv_fn(self, x, context): |
| b, n, d = x.size(0), self.num_heads, self.head_dim |
|
|
| |
| q = self.norm_q(self.q(x)).view(b, -1, n, d) |
| k = self.norm_k(self.k(context)).view(b, -1, n, d) |
| v = self.v(context).view(b, -1, n, d) |
|
|
| return q, k, v |
|
|
| def forward(self, x, context, context_lens): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L1, C] |
| context(Tensor): Shape [B, L2, C] |
| context_lens(Tensor): Shape [B] |
| """ |
| q, k, v = self.qkv_fn(x, context) |
|
|
| |
| x = flash_attention(q, k, v, k_lens=context_lens) |
|
|
| |
| x = x.flatten(2) |
| x = self.o(x) |
| return x |
|
|
|
|
| class WanI2VCrossAttention(WanSelfAttention): |
|
|
| def __init__(self, |
| dim, |
| num_heads, |
| window_size=(-1, -1), |
| qk_norm=True, |
| eps=1e-6, |
| additional_emb_length=None): |
| super().__init__(dim, num_heads, window_size, qk_norm, eps) |
|
|
| self.k_img = nn.Linear(dim, dim) |
| self.v_img = nn.Linear(dim, dim) |
| |
| self.norm_k_img = WanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity() |
| self.additional_emb_length = additional_emb_length |
|
|
| def qkv_fn(self, x, context): |
| context_img = context[:, : self.additional_emb_length] |
| context = context[:, self.additional_emb_length :] |
| b, n, d = x.size(0), self.num_heads, self.head_dim |
|
|
| |
| q = self.norm_q(self.q(x)).view(b, -1, n, d) |
| k = self.norm_k(self.k(context)).view(b, -1, n, d) |
| v = self.v(context).view(b, -1, n, d) |
| k_img = self.norm_k_img(self.k_img(context_img)).view(b, -1, n, d) |
| v_img = self.v_img(context_img).view(b, -1, n, d) |
|
|
| return q, k, v, k_img, v_img |
|
|
|
|
| def forward(self, x, context, context_lens): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L1, C] |
| context(Tensor): Shape [B, L2, C] |
| context_lens(Tensor): Shape [B] |
| """ |
| q, k, v, k_img, v_img = self.qkv_fn(x, context) |
|
|
| if self.use_sp: |
| |
| q = all_to_all_4D(q, scatter_dim=2, gather_dim=1) |
| k = torch.chunk(k, self.sp_size, dim=2)[self.sp_rank] |
| v = torch.chunk(v, self.sp_size, dim=2)[self.sp_rank] |
| k_img = torch.chunk(k_img, self.sp_size, dim=2)[self.sp_rank] |
| v_img = torch.chunk(v_img, self.sp_size, dim=2)[self.sp_rank] |
| |
| |
| |
| img_x = flash_attention(q, k_img, v_img, k_lens=None) |
| |
| x = flash_attention(q, k, v, k_lens=context_lens) |
| if self.use_sp: |
| |
| x = all_to_all_4D(x, scatter_dim=1, gather_dim=2) |
| |
| |
| x = x.flatten(2) |
| img_x = img_x.flatten(2) |
| x = x + img_x |
| x = self.o(x) |
| return x |
|
|
|
|
| WAN_CROSSATTENTION_CLASSES = { |
| 't2v_cross_attn': WanT2VCrossAttention, |
| 'i2v_cross_attn': WanI2VCrossAttention, |
| } |
|
|
| class ModulationAdd(nn.Module): |
| def __init__(self, dim, num): |
| super().__init__() |
| self.modulation = nn.Parameter(torch.randn(1, num, dim) / dim**0.5) |
|
|
| def forward(self, e): |
| return self.modulation.bfloat16() + e.bfloat16() |
|
|
| class WanAttentionBlock(nn.Module): |
|
|
| 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, |
| additional_emb_length=None): |
| 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 |
|
|
| |
| 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() |
| if cross_attn_type == 'i2v_cross_attn': |
| assert additional_emb_length is not None, "additional_emb_length should be specified for i2v_cross_attn" |
| self.cross_attn = WanI2VCrossAttention(dim, |
| num_heads, |
| (-1, -1), |
| qk_norm, |
| eps, |
| additional_emb_length) |
| else: |
| assert additional_emb_length is None, "additional_emb_length should be None for t2v_cross_attn" |
| self.cross_attn = WanT2VCrossAttention(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)) |
|
|
| |
| |
| |
| self.modulation = ModulationAdd(dim, 6) |
|
|
| def init_lora(self, self_lora=False, train=False, using_ip_emb=False): |
| self.self_attn.init_lora(self_lora, train, using_ip_emb) |
|
|
| def forward( |
| self, |
| x, |
| x_ip, |
| e, |
| seq_lens, |
| grid_sizes, |
| freqs, |
| context, |
| context_lens, |
| e_ip=None, |
| ip_grid_sizes=None, |
| ip_freqs=None, |
| ip_offsets=None, |
| ip_emb=None |
| ): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L, C] |
| e(Tensor): Shape [B, L1, 6, C] |
| seq_lens(Tensor): Shape [B], length of each sequence in batch |
| grid_sizes(Tensor): Shape [B, 3], the second dimension contains (F, H, W) |
| freqs(Tensor): Rope freqs, shape [1024, C / num_heads / 2] |
| """ |
| assert e.dtype == torch.bfloat16 |
| assert len(e.shape) == 4 and e.size(2) == 6 and e.shape[1] == x.shape[1], f"{e.shape}, {x.shape}" |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| e = self.modulation(e).chunk(6, dim=2) |
| assert e[0].dtype == torch.bfloat16 |
|
|
| if e_ip is not None and x_ip is not None and ip_grid_sizes is not None and ip_freqs is not None: |
| assert e_ip.dtype == torch.bfloat16 |
| assert len(e_ip.shape) == 4 and e_ip.size(2) == 6 and (e_ip.shape[1] == x_ip.shape[1] or e_ip.shape[1] == 1), f"{e_ip.shape}, {x_ip.shape}" |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| e_ip = self.modulation(e_ip).chunk(6, dim=2) |
| assert e_ip[0].dtype == torch.bfloat16 |
| input_x_ip = self.norm1(x_ip).bfloat16() * (1 + e_ip[1].squeeze(2)) + e_ip[0].squeeze(2) |
| else: |
| input_x_ip = None |
|
|
| |
| y, y_ip = self.self_attn( |
| self.norm1(x).bfloat16() * (1 + e[1].squeeze(2)) + e[0].squeeze(2), |
| seq_lens, grid_sizes, freqs, |
| x_ip=input_x_ip, ip_grid_sizes=ip_grid_sizes, |
| ip_freqs=ip_freqs, ip_offsets=ip_offsets, ip_emb=ip_emb |
| ) |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| x = x + y * e[2].squeeze(2) |
| if y_ip is not None: |
| x_ip = x_ip + y_ip * e_ip[2].squeeze(2) |
|
|
| |
| def cross_attn_ffn(x, context, context_lens, e): |
| x = x + self.cross_attn(self.norm3(x), context, context_lens) |
| y = self.ffn( |
| self.norm2(x).bfloat16() * (1 + e[4].squeeze(2)) + e[3].squeeze(2)) |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| x = x + y * e[5].squeeze(2) |
| return x |
|
|
| x = cross_attn_ffn(x, context, context_lens, e) |
|
|
| |
| if x_ip is not None: |
| y_ip = self.ffn( |
| self.norm2(x_ip).bfloat16() * (1 + e_ip[4].squeeze(2)) + e_ip[3].squeeze(2)) |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| x_ip = x_ip + y_ip * e_ip[5].squeeze(2) |
| |
| return x, x_ip |
|
|
|
|
| class Head(nn.Module): |
|
|
| def __init__(self, dim, out_dim, patch_size, eps=1e-6): |
| super().__init__() |
| self.dim = dim |
| self.out_dim = out_dim |
| self.patch_size = patch_size |
| self.eps = eps |
|
|
| |
| out_dim = math.prod(patch_size) * out_dim |
| self.norm = WanLayerNorm(dim, eps) |
| self.head = nn.Linear(dim, out_dim) |
|
|
| |
| self.modulation = nn.Parameter(torch.randn(1, 2, dim) / dim**0.5) |
|
|
| def forward(self, x, e): |
| r""" |
| Args: |
| x(Tensor): Shape [B, L1, C] |
| e(Tensor): Shape [B, L, C] |
| """ |
| assert e.dtype == torch.bfloat16 |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| e = (self.modulation.bfloat16().unsqueeze(0) + e.unsqueeze(2)).chunk(2, dim=2) |
| x = self.head(self.norm(x) * (1 + e[1].squeeze(2)) + e[0].squeeze(2)) |
|
|
| return x |
|
|
|
|
| class MLPProj(torch.nn.Module): |
|
|
| def __init__(self, in_dim, out_dim): |
| super().__init__() |
|
|
| self.proj = torch.nn.Sequential( |
| torch.nn.LayerNorm(in_dim), torch.nn.Linear(in_dim, in_dim), |
| torch.nn.GELU(), torch.nn.Linear(in_dim, out_dim), |
| torch.nn.LayerNorm(out_dim)) |
|
|
| def forward(self, image_embeds): |
| clip_extra_context_tokens = self.proj(image_embeds) |
| return clip_extra_context_tokens |
|
|
|
|
| class WanModel(ModelMixin, ConfigMixin): |
| r""" |
| Wan diffusion backbone supporting both text-to-video and image-to-video, text-to-audio. |
| """ |
|
|
| 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='t2v', |
| patch_size=(1, 2, 2), |
| text_len=512, |
| in_dim=16, |
| dim=2048, |
| ffn_dim=8192, |
| freq_dim=256, |
| text_dim=4096, |
| additional_emb_dim=None, |
| additional_emb_length=None, |
| out_dim=16, |
| num_heads=16, |
| num_layers=32, |
| window_size=(-1, -1), |
| qk_norm=True, |
| cross_attn_norm=True, |
| gradient_checkpointing = False, |
| temporal_rope_scaling_factor=1.0, |
| eps=1e-6): |
| r""" |
| Initialize the diffusion model backbone. |
| |
| Args: |
| model_type (`str`, *optional*, defaults to 't2v'): |
| Model variant - 't2v' (text-to-video) or 'i2v' (image-to-video) |
| patch_size (`tuple`, *optional*, defaults to (1, 2, 2)): |
| 3D patch dimensions for video embedding (t_patch, h_patch, w_patch) |
| text_len (`int`, *optional*, defaults to 512): |
| Fixed length for text embeddings |
| in_dim (`int`, *optional*, defaults to 16): |
| Input video channels (C_in) |
| dim (`int`, *optional*, defaults to 2048): |
| Hidden dimension of the transformer |
| ffn_dim (`int`, *optional*, defaults to 8192): |
| Intermediate dimension in feed-forward network |
| freq_dim (`int`, *optional*, defaults to 256): |
| Dimension for sinusoidal time embeddings |
| text_dim (`int`, *optional*, defaults to 4096): |
| Input dimension for text embeddings |
| out_dim (`int`, *optional*, defaults to 16): |
| Output video channels (C_out) |
| num_heads (`int`, *optional*, defaults to 16): |
| Number of attention heads |
| num_layers (`int`, *optional*, defaults to 32): |
| Number of transformer blocks |
| window_size (`tuple`, *optional*, defaults to (-1, -1)): |
| Window size for local attention (-1 indicates global attention) |
| qk_norm (`bool`, *optional*, defaults to True): |
| Enable query/key normalization |
| cross_attn_norm (`bool`, *optional*, defaults to False): |
| Enable cross-attention normalization |
| eps (`float`, *optional*, defaults to 1e-6): |
| Epsilon value for normalization layers |
| """ |
|
|
| super().__init__() |
|
|
| assert model_type in ['t2v', 'i2v', 't2a', 'tt2a', 'ti2v'] |
| self.model_type = model_type |
| is_audio_type = "a" in self.model_type |
| is_video_type = "v" in self.model_type |
| assert is_audio_type ^ is_video_type, "Either audio or video model should be specified" |
| if is_audio_type: |
| |
| assert len(patch_size) == 1 and patch_size[0] == 1, "Audio model should only accept 1 dimensional input, and we dont do patchify" |
|
|
| self.patch_size = patch_size |
| self.text_len = text_len |
| self.in_dim = in_dim |
| self.dim = dim |
| self.ffn_dim = ffn_dim |
| self.freq_dim = freq_dim |
| self.text_dim = text_dim |
| self.out_dim = out_dim |
| self.num_heads = num_heads |
| self.num_layers = num_layers |
| self.window_size = window_size |
| self.qk_norm = qk_norm |
| self.cross_attn_norm = cross_attn_norm |
| self.eps = eps |
| self.temporal_rope_scaling_factor = temporal_rope_scaling_factor |
| self.is_audio_type = is_audio_type |
| self.is_video_type = is_video_type |
| |
| if is_audio_type: |
| |
| self.patch_embedding = nn.Sequential( |
| ChannelLastConv1d(in_dim, dim, kernel_size=7, padding=3), |
| nn.SiLU(), |
| ConvMLP(dim, dim * 4, kernel_size=7, padding=3), |
| ) |
| else: |
| self.patch_embedding = nn.Conv3d( |
| in_dim, dim, kernel_size=patch_size, stride=patch_size) |
| |
| self.text_embedding = nn.Sequential( |
| nn.Linear(text_dim, dim), nn.GELU(approximate='tanh'), |
| nn.Linear(dim, dim)) |
|
|
| self.time_embedding = nn.Sequential( |
| nn.Linear(freq_dim, dim), nn.SiLU(), nn.Linear(dim, dim)) |
| self.time_projection = nn.Sequential(nn.SiLU(), nn.Linear(dim, dim * 6)) |
| self.use_sp = get_sequence_parallel_state() |
| if self.use_sp: |
| self.sp_size = nccl_info.sp_size |
| self.sp_rank = nccl_info.rank_within_group |
| assert self.num_heads % self.sp_size == 0, \ |
| f"Num heads {self.num_heads} must be divisible by sp_size {self.sp_size}" |
| |
| |
| cross_attn_type = 't2v_cross_attn' if model_type in ['t2v', 't2a', 'ti2v'] else 'i2v_cross_attn' |
|
|
| if cross_attn_type == 't2v_cross_attn': |
| assert additional_emb_dim is None and additional_emb_length is None, "additional_emb_length should be None for t2v and t2a model" |
| else: |
| assert additional_emb_dim is not None and additional_emb_length is not None, "additional_emb_length should be specified for i2v and tt2a model" |
|
|
| self.blocks = nn.ModuleList([ |
| WanAttentionBlock(cross_attn_type, dim, ffn_dim, num_heads, |
| window_size, qk_norm, cross_attn_norm, eps, additional_emb_length) |
| for _ in range(num_layers) |
| ]) |
|
|
| |
| self.head = Head(dim, out_dim, patch_size, eps) |
|
|
| self.set_gradient_checkpointing(enable=gradient_checkpointing) |
| self.set_rope_params() |
|
|
| if model_type in ['i2v', 'tt2a']: |
| self.img_emb = MLPProj(additional_emb_dim, dim) |
|
|
| |
| self.init_weights() |
|
|
| self.gradient_checkpointing = False |
|
|
| def init_lora(self, self_lora=False, train=False, ip_emb_dim=None): |
| |
| if ip_emb_dim is not None: |
| self.ip_projection = nn.Sequential( |
| nn.Linear(ip_emb_dim, self.dim), nn.SiLU(), nn.Linear(self.dim, self.dim) |
| ) |
| |
| requires_grad = train |
| for param in self.ip_projection.parameters(): |
| param.requires_grad = requires_grad |
| |
| for block in self.blocks: |
| block.init_lora(self_lora, train, ip_emb_dim is not None) |
|
|
| def set_rope_params(self): |
| |
| dim = self.dim |
| num_heads = self.num_heads |
| assert (dim % num_heads) == 0 and (dim // num_heads) % 2 == 0 |
| d = dim // num_heads |
|
|
| if self.is_audio_type: |
| |
| |
| self.freqs = rope_params(1024, d - 4 * (d // 6), freqs_scaling=self.temporal_rope_scaling_factor) |
| else: |
| self.freqs = torch.cat([ |
| rope_params(1024, d - 4 * (d // 6)), |
| rope_params(1024, 2 * (d // 6)), |
| rope_params(1024, 2 * (d // 6)) |
| ], dim=1 |
| ) |
|
|
| def set_gradient_checkpointing(self, enable: bool): |
| self.gradient_checkpointing = enable |
|
|
| def prepare_transformer_block_kwargs( |
| self, |
| x, |
| t, |
| context, |
| seq_len, |
| clip_fea=None, |
| y=None, |
| first_frame_is_clean=False, |
| x_ip=None, |
| ip_emb=None |
| ): |
|
|
| |
| |
| if x_ip is not None: |
| assert clip_fea is None and y is None, "clip_fea and y should be None when x_ip is provided" |
|
|
| device = next(self.patch_embedding.parameters()).device |
|
|
| if self.freqs.device != device: |
| self.freqs = self.freqs.to(device) |
|
|
| if y is not None: |
| x = [torch.cat([u, v], dim=0) for u, v in zip(x, y)] |
|
|
| |
| |
| |
| |
| x = [self.patch_embedding(u.unsqueeze(0)) for u in x] |
| if x_ip is not None: |
| x_ip = [self.patch_embedding(u.unsqueeze(0)) for u in x_ip] |
| |
| if self.is_audio_type: |
| |
| grid_sizes = torch.stack( |
| [torch.tensor(u.shape[1:2], dtype=torch.long) for u in x] |
| ) |
| if x_ip is not None: |
| ip_grid_sizes = torch.stack( |
| [torch.tensor(u.shape[1:2], dtype=torch.long) for u in x_ip] |
| ) |
| else: |
| ip_grid_sizes = None |
| else: |
| |
| 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] |
| if x_ip is not None: |
| ip_grid_sizes = torch.stack( |
| [torch.tensor(u.shape[2:], dtype=torch.long) for u in x_ip]) |
| x_ip = [u.flatten(2).transpose(1, 2) for u in x_ip] |
| else: |
| ip_grid_sizes = None |
| |
| |
| if x_ip is not None: |
| ip_offsets = grid_sizes |
| else: |
| ip_offsets = None |
|
|
| seq_lens = torch.tensor([u.size(1) for u in x], dtype=torch.long) |
| assert seq_lens.max() <= seq_len, f"Sequence length {seq_lens.max()} exceeds maximum {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 |
| ]) |
|
|
| if x_ip is not None: |
| ip_seq_lens = torch.tensor([u.size(1) for u in x_ip], dtype=torch.long) |
| ip_seq_len = ip_seq_lens.max() |
| x_ip = torch.cat([ |
| torch.cat([u, u.new_zeros(1, ip_seq_len - u.size(1), u.size(2))], |
| dim=1) for u in x_ip |
| ]) |
|
|
| |
| if t.dim() == 1: |
| if first_frame_is_clean: |
| t = torch.ones((t.size(0), seq_len), device=t.device, dtype=t.dtype) * t.unsqueeze(1) |
| _first_images_seq_len = grid_sizes[:, 1:].prod(-1) |
| for i in range(t.size(0)): |
| t[i, :_first_images_seq_len[i]] = 0 |
| |
| else: |
| t = t.unsqueeze(1).expand(t.size(0), seq_len) |
| with amp.autocast('cuda', dtype=torch.bfloat16): |
| bt = t.size(0) |
| t = t.flatten() |
| e = self.time_embedding( |
| sinusoidal_embedding_1d(self.freq_dim,t).unflatten(0, (bt, seq_len)).bfloat16() |
| ) |
| e0 = self.time_projection(e).unflatten(2, (6, self.dim)) |
| assert e.dtype == torch.bfloat16 and e0.dtype == torch.bfloat16 |
| |
| t_ip = torch.zeros(bt, device=t.device, dtype=t.dtype) |
| e_ip = self.time_embedding( |
| sinusoidal_embedding_1d(self.freq_dim, t_ip).unflatten(0, (bt, 1)).bfloat16() |
| ) |
| e0_ip = self.time_projection(e_ip).unflatten(2, (6, self.dim)) |
| assert e_ip.dtype == torch.bfloat16 and e0_ip.dtype == torch.bfloat16 |
| |
| if self.use_sp: |
| current_len = x.shape[1] |
| |
| pad_size = (-current_len ) % self.sp_size |
|
|
| if pad_size > 0: |
| padding = torch.zeros( |
| x.shape[0], pad_size, x.shape[2], |
| device=x.device, |
| dtype=x.dtype |
| ) |
| x = torch.cat([x, padding], dim=1) |
| e_padding = torch.zeros( |
| e.shape[0], pad_size, e.shape[2], |
| device=e.device, |
| dtype=e.dtype |
| ) |
| e = torch.cat([e, e_padding], dim=1) |
| e0_padding = torch.zeros( |
| e0.shape[0], pad_size, e0.shape[2], e0.shape[3], |
| device=e0.device, |
| dtype=e0.dtype |
| ) |
| e0 = torch.cat([e0, e0_padding], dim=1) |
|
|
| x = torch.chunk(x, self.sp_size, dim=1)[self.sp_rank] |
| e = torch.chunk(e, self.sp_size, dim=1)[self.sp_rank] |
| e0 = torch.chunk(e0, self.sp_size, dim=1)[self.sp_rank] |
|
|
| if x_ip is not None: |
| current_ip_len = x_ip.shape[1] |
| ip_pad_size = (-current_ip_len ) % self.sp_size |
| if ip_pad_size > 0: |
| ip_padding = torch.zeros( |
| x_ip.shape[0], ip_pad_size, x_ip.shape[2], |
| device=x_ip.device, |
| dtype=x_ip.dtype |
| ) |
| x_ip = torch.cat([x_ip, ip_padding], dim=1) |
| |
| x_ip = torch.chunk(x_ip, self.sp_size, dim=1)[self.sp_rank] |
| |
| |
| 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 |
| ])) |
|
|
| if clip_fea is not None: |
| context_clip = self.img_emb(clip_fea) |
| context = torch.concat([context_clip, context], dim=1) |
| |
| |
| if ip_emb is not None: |
| assert ip_emb.dim() == 2 or ip_emb.dim() == 3 |
| ip_emb = self.ip_projection(ip_emb) |
| if ip_emb.dim() == 2: |
| ip_emb = ip_emb.unsqueeze(1) |
|
|
| |
| kwargs = dict( |
| e=e0, |
| seq_lens=seq_lens, |
| grid_sizes=grid_sizes, |
| freqs=self.freqs, |
| context=context, |
| context_lens=context_lens, |
| e_ip=e0_ip, |
| ip_grid_sizes=ip_grid_sizes, |
| ip_freqs=self.freqs, |
| ip_offsets=ip_offsets, |
| ip_emb=ip_emb |
| ) |
|
|
| return x, e, x_ip, e_ip, kwargs |
| |
| def post_transformer_block_out(self, x, grid_sizes, e): |
| |
| x = self.head(x, e) |
| if self.use_sp: |
| x = all_gather(x, dim=1) |
| |
| if self.is_audio_type: |
| |
| |
| grid_sizes = [gs[0] for gs in grid_sizes] |
| |
| assert len(x) == len(grid_sizes) |
| x = [u[:gs] for u, gs in zip(x, grid_sizes)] |
| else: |
| |
| x = self.unpatchify(x, grid_sizes) |
|
|
| return torch.stack([u.bfloat16() for u in x], dim=0) |
|
|
|
|
| def forward( |
| self, |
| x, |
| t, |
| context, |
| seq_len, |
| clip_fea=None, |
| y=None, |
| first_frame_is_clean=False, |
| x_ip=None, |
| ip_emb=None |
| ): |
| r""" |
| Forward pass through the diffusion model |
| |
| Args: |
| x (List[Tensor]): |
| List of input video tensors, each with shape [C_in, F, H, W] |
| OR |
| List of input audio tensors, each with shape [L, C_in] |
| 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 |
| |
| Returns: |
| List[Tensor]: |
| List of denoised video tensors with original input shapes [C_out, F, H / 8, W / 8] |
| OR |
| List of denoised audio tensors with original input shapes [L, C_in] |
| """ |
| x, e, x_ip, e_ip, kwargs = self.prepare_transformer_block_kwargs( |
| x=x, |
| t=t, |
| context=context, |
| seq_len=seq_len, |
| clip_fea=clip_fea, |
| y=y, |
| first_frame_is_clean=first_frame_is_clean, |
| x_ip=x_ip, |
| ip_emb=ip_emb |
| ) |
|
|
| for block in self.blocks: |
| x, x_ip = gradient_checkpointing( |
| enabled=(self.gradient_checkpointing), |
| module=block, |
| x=x, |
| x_ip=x_ip, |
| **kwargs |
| ) |
| |
| x_out = self.post_transformer_block_out(x, kwargs['grid_sizes'], e) |
| x_ip_out = None |
|
|
| return x_out |
|
|
| def unpatchify(self, x, grid_sizes): |
| r""" |
| Reconstruct video tensors from patch embeddings. |
| |
| Args: |
| x (List[Tensor]): |
| List of patchified features, each with shape [L, C_out * prod(patch_size)] |
| grid_sizes (Tensor): |
| Original spatial-temporal grid dimensions before patching, |
| shape [B, 3] (3 dimensions correspond to F_patches, H_patches, W_patches) |
| |
| Returns: |
| List[Tensor]: |
| Reconstructed video tensors with shape [C_out, F, H / 8, W / 8] |
| """ |
|
|
| c = self.out_dim |
| out = [] |
| for u, v in zip(x, grid_sizes.tolist()): |
| |
| u = u[:math.prod(v)].view(*v, *self.patch_size, c) |
| u = torch.einsum('fhwpqrc->cfphqwr', u) |
| u = u.reshape(c, *[i * j for i, j in zip(v, self.patch_size)]) |
| out.append(u) |
| |
| return out |
|
|
| def init_weights(self): |
| r""" |
| Initialize model parameters using Xavier initialization. |
| """ |
|
|
| |
| for m in self.modules(): |
| if isinstance(m, nn.Linear): |
| nn.init.xavier_uniform_(m.weight) |
| if m.bias is not None: |
| nn.init.zeros_(m.bias) |
|
|
| |
| if self.is_video_type: |
| assert isinstance(self.patch_embedding, nn.Conv3d), f"Patch embedding for video should be a Conv3d layer, got {type(self.patch_embedding)}" |
| nn.init.xavier_uniform_(self.patch_embedding.weight.flatten(1)) |
| for m in self.text_embedding.modules(): |
| if isinstance(m, nn.Linear): |
| nn.init.normal_(m.weight, std=.02) |
| for m in self.time_embedding.modules(): |
| if isinstance(m, nn.Linear): |
| nn.init.normal_(m.weight, std=.02) |
|
|
| |
| nn.init.zeros_(self.head.head.weight) |