# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved. import math import torch import torch.cuda.amp as amp import torch.nn as nn from diffusers.configuration_utils import ConfigMixin, register_to_config from diffusers.models.modeling_utils import ModelMixin import numpy as np from typing import Union,Optional from .attention import pay_attention __all__ = ['WanModel'] def sinusoidal_embedding_1d(dim, position): # preprocess assert dim % 2 == 0 half = dim // 2 position = position.type(torch.float32) # calculation 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 def identify_k( b: float, d: int, N: int): """ This function identifies the index of the intrinsic frequency component in a RoPE-based pre-trained diffusion transformer. Args: b (`float`): The base frequency for RoPE. d (`int`): Dimension of the frequency tensor N (`int`): the first observed repetition frame in latent space Returns: k (`int`): the index of intrinsic frequency component N_k (`int`): the period of intrinsic frequency component in latent space Example: In HunyuanVideo, b=256 and d=16, the repetition occurs approximately 8s (N=48 in latent space). k, N_k = identify_k(b=256, d=16, N=48) In this case, the intrinsic frequency index k is 4, and the period N_k is 50. """ # Compute the period of each frequency in RoPE according to Eq.(4) periods = [] for j in range(1, d // 2 + 1): theta_j = 1.0 / (b ** (2 * (j - 1) / d)) N_j = round(2 * torch.pi / theta_j) periods.append(N_j) # Identify the intrinsic frequency whose period is closed to N（see Eq.(7)） diffs = [abs(N_j - N) for N_j in periods] k = diffs.index(min(diffs)) + 1 N_k = periods[k-1] return k, N_k def rope_params_riflex(max_seq_len, dim, theta=10000, L_test=30, k=6): assert dim % 2 == 0 exponents = torch.arange(0, dim, 2, dtype=torch.float64).div(dim) inv_theta_pow = 1.0 / torch.pow(theta, exponents) inv_theta_pow[k-1] = 0.9 * 2 * torch.pi / L_test freqs = torch.outer(torch.arange(max_seq_len), inv_theta_pow) if True: freqs_cos = freqs.cos().repeat_interleave(2, dim=1).float() # [S, D] freqs_sin = freqs.sin().repeat_interleave(2, dim=1).float() # [S, D] return (freqs_cos, freqs_sin) else: freqs = torch.polar(torch.ones_like(freqs), freqs) # complex64 # [S, D/2] return freqs def rope_apply_(x, grid_sizes, freqs): assert x.shape[0]==1 n, c = x.size(2), x.size(3) // 2 # split freqs freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1) f, h, w = grid_sizes[0] seq_len = f * h * w x_i = x[0, :seq_len, :, :] x_i = x_i.to(torch.float32) x_i = x_i.reshape(seq_len, n, -1, 2) x_i = torch.view_as_complex(x_i) freqs_i = torch.cat([ freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1), freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1), freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1) ], dim=-1) freqs_i= freqs_i.reshape(seq_len, 1, -1) # apply rotary embedding x_i *= freqs_i x_i = torch.view_as_real(x_i).flatten(2) x[0, :seq_len, :, :] = x_i.to(torch.bfloat16) # x_i = torch.cat([x_i, x[0, seq_len:]]) return x # @amp.autocast(enabled=False) def rope_apply(x, grid_sizes, freqs): n, c = x.size(2), x.size(3) // 2 # split freqs freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1) # loop over samples output = [] for i, (f, h, w) in enumerate(grid_sizes): seq_len = f * h * w # precompute multipliers # x_i = x[i, :seq_len] x_i = x[i] x_i = x_i[:seq_len, :, :] x_i = x_i.to(torch.float32) x_i = x_i.reshape(seq_len, n, -1, 2) x_i = torch.view_as_complex(x_i) freqs_i = torch.cat([ freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1), freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1), freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1) ], dim=-1).reshape(seq_len, 1, -1) # apply rotary embedding x_i *= freqs_i x_i = torch.view_as_real(x_i).flatten(2) x_i = x_i.to(torch.bfloat16) x_i = torch.cat([x_i, x[i, seq_len:]]) # append to collection output.append(x_i) return torch.stack(output) #.float() def relative_l1_distance(last_tensor, current_tensor): l1_distance = torch.abs(last_tensor - current_tensor).mean() norm = torch.abs(last_tensor).mean() relative_l1_distance = l1_distance / norm return relative_l1_distance.to(torch.float32) 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] """ y = x.float() y.pow_(2) y = y.mean(dim=-1, keepdim=True) y += self.eps y.rsqrt_() x *= y x *= self.weight return x # return self._norm(x).type_as(x) * self.weight def _norm(self, x): return x * torch.rsqrt(x.pow(2).mean(dim=-1, keepdim=True) + self.eps) def my_LayerNorm(norm, x): y = x.float() y_m = y.mean(dim=-1, keepdim=True) y -= y_m del y_m y.pow_(2) y = y.mean(dim=-1, keepdim=True) y += norm.eps y.rsqrt_() x = x * y return x 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 F.layer_norm( # input, self.normalized_shape, self.weight, self.bias, self.eps # ) y = super().forward(x) x = y.type_as(x) return x # return super().forward(x).type_as(x) from wan.modules.posemb_layers import apply_rotary_emb 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 forward(self, xlist, seq_lens, grid_sizes, freqs): r""" Args: x(Tensor): Shape [B, L, num_heads, C / num_heads] seq_lens(Tensor): Shape [B] grid_sizes(Tensor): Shape [B, 3], the second dimension contains (F, H, W) freqs(Tensor): Rope freqs, shape [1024, C / num_heads / 2] """ x = xlist[0] xlist.clear() b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim # query, key, value function q = self.q(x) self.norm_q(q) q = q.view(b, s, n, d) # !!! k = self.k(x) self.norm_k(k) k = k.view(b, s, n, d) v = self.v(x).view(b, s, n, d) del x # rope_apply_(q, grid_sizes, freqs) # rope_apply_(k, grid_sizes, freqs) qklist = [q,k] del q,k q,k = apply_rotary_emb(qklist, freqs, head_first=False) qkv_list = [q,k,v] del q,k,v x = pay_attention( qkv_list, # q=q, # k=k, # v=v, # k_lens=seq_lens, window_size=self.window_size) # output x = x.flatten(2) x = self.o(x) return x class WanT2VCrossAttention(WanSelfAttention): def forward(self, xlist, context, context_lens): r""" Args: x(Tensor): Shape [B, L1, C] context(Tensor): Shape [B, L2, C] context_lens(Tensor): Shape [B] """ x = xlist[0] xlist.clear() b, n, d = x.size(0), self.num_heads, self.head_dim # compute query, key, value q = self.q(x) del x self.norm_q(q) q= q.view(b, -1, n, d) k = self.k(context) self.norm_k(k) k = k.view(b, -1, n, d) v = self.v(context).view(b, -1, n, d) # compute attention qvl_list=[q, k, v] del q, k, v x = pay_attention(qvl_list, k_lens=context_lens) # output 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): 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.alpha = nn.Parameter(torch.zeros((1, ))) self.norm_k_img = WanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity() def forward(self, xlist, context, context_lens): r""" Args: x(Tensor): Shape [B, L1, C] context(Tensor): Shape [B, L2, C] context_lens(Tensor): Shape [B] """ ##### Enjoy this spagheti VRAM optimizations done by DeepBeepMeep ! # I am sure you are a nice person and as you copy this code, you will give me officially proper credits: # Please link to https://github.com/deepbeepmeep/Wan2GP and @deepbeepmeep on twitter x = xlist[0] xlist.clear() context_img = context[:, :257] context = context[:, 257:] b, n, d = x.size(0), self.num_heads, self.head_dim # compute query, key, value q = self.q(x) del x self.norm_q(q) q= q.view(b, -1, n, d) k = self.k(context) self.norm_k(k) k = k.view(b, -1, n, d) v = self.v(context).view(b, -1, n, d) qkv_list = [q, k, v] del k,v x = pay_attention(qkv_list, k_lens=context_lens) k_img = self.k_img(context_img) self.norm_k_img(k_img) k_img = k_img.view(b, -1, n, d) v_img = self.v_img(context_img).view(b, -1, n, d) qkv_list = [q, k_img, v_img] del q, k_img, v_img img_x = pay_attention(qkv_list, k_lens=None) # compute attention # output x = x.flatten(2) img_x = img_x.flatten(2) x += img_x del img_x x = self.o(x) return x WAN_CROSSATTENTION_CLASSES = { 't2v_cross_attn': WanT2VCrossAttention, 'i2v_cross_attn': WanI2VCrossAttention, } 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): 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 # 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 = WAN_CROSSATTENTION_CLASSES[cross_attn_type](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) def forward( self, x, e, seq_lens, grid_sizes, freqs, context, context_lens, ): r""" Args: x(Tensor): Shape [B, L, C] e(Tensor): Shape [B, 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] """ e = (self.modulation + e).chunk(6, dim=1) # self-attention x_mod = self.norm1(x) x_mod *= 1 + e[1] x_mod += e[0] xlist = [x_mod] del x_mod y = self.self_attn( xlist, seq_lens, grid_sizes,freqs) x.addcmul_(y, e[2]) del y y = self.norm3(x) ylist= [y] del y x += self.cross_attn(ylist, context, context_lens) y = self.norm2(x) y *= 1 + e[4] y += e[3] ffn = self.ffn[0] gelu = self.ffn[1] ffn2= self.ffn[2] y_shape = y.shape y = y.view(-1, y_shape[-1]) chunk_size = int(y_shape[1]/2.7) chunks =torch.split(y, chunk_size) for y_chunk in chunks: mlp_chunk = ffn(y_chunk) mlp_chunk = gelu(mlp_chunk) y_chunk[...] = ffn2(mlp_chunk) del mlp_chunk y = y.view(y_shape) x.addcmul_(y, e[5]) return x 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 # layers out_dim = math.prod(patch_size) * out_dim self.norm = WanLayerNorm(dim, eps) self.head = nn.Linear(dim, out_dim) # modulation 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, C] """ # assert e.dtype == torch.float32 e = (self.modulation + e.unsqueeze(1)).chunk(2, dim=1) x = self.norm(x).to(torch.bfloat16) x *= (1 + e[1]) x += e[0] x = self.head(x) 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. """ 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, out_dim=16, num_heads=16, num_layers=32, window_size=(-1, -1), qk_norm=True, cross_attn_norm=True, 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'] self.model_type = model_type 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 # embeddings 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)) # blocks cross_attn_type = 't2v_cross_attn' if model_type == 't2v' else 'i2v_cross_attn' self.blocks = nn.ModuleList([ WanAttentionBlock(cross_attn_type, dim, ffn_dim, num_heads, window_size, qk_norm, cross_attn_norm, eps) for _ in range(num_layers) ]) # head self.head = Head(dim, out_dim, patch_size, eps) # buffers (don't use register_buffer otherwise dtype will be changed in to()) if model_type == 'i2v': self.img_emb = MLPProj(1280, dim) # initialize weights self.init_weights() # self.freqs = torch.cat([ # rope_params(1024, d - 4 * (d // 6)), #44 # rope_params(1024, 2 * (d // 6)), #42 # rope_params(1024, 2 * (d // 6)) #42 # ],dim=1) def get_rope_freqs(self, nb_latent_frames, RIFLEx_k = None, device = "cuda"): dim = self.dim num_heads = self.num_heads d = dim // num_heads assert (dim % num_heads) == 0 and (dim // num_heads) % 2 == 0 c1, s1 = rope_params_riflex(1024, dim= d - 4 * (d // 6), L_test=nb_latent_frames, k = RIFLEx_k ) if RIFLEx_k != None else rope_params(1024, dim= d - 4 * (d // 6)) #44 c2, s2 = rope_params(1024, 2 * (d // 6)) #42 c3, s3 = rope_params(1024, 2 * (d // 6)) #42 return (torch.cat([c1,c2,c3],dim=1).to(device) , torch.cat([s1,s2,s3],dim=1).to(device)) def forward( self, x, t, context, seq_len, clip_fea=None, y=None, freqs = None, pipeline = None, current_step = 0, is_uncond=False ): 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 Returns: List[Tensor]: List of denoised video tensors with original input shapes [C_out, F, H / 8, W / 8] """ if self.model_type == 'i2v': assert clip_fea is not None and y is not None # params device = self.patch_embedding.weight.device if torch.is_tensor(freqs) and freqs.device != device: freqs = freqs.to(device) if y is not None: x = [torch.cat([u, v], dim=0) for u, v in zip(x, y)] # embeddings 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]) grid_sizes = [ list(u.shape[2:]) 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) 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)) e0 = self.time_projection(e).unflatten(1, (6, self.dim)).to(torch.bfloat16) # context 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) # bs x 257 x dim context = torch.concat([context_clip, context], dim=1) # deepbeepmeep optimization of kijai's implementation (https://github.com/kijai/ComfyUI-WanVideoWrapper/) of teacache (https://github.com/ali-vilab/TeaCache) should_calc = True if self.enable_teacache and current_step >= self.teacache_start_step: if current_step == self.teacache_start_step: self.accumulated_rel_l1_distance_cond = 0 self.accumulated_rel_l1_distance_uncond = 0 self.teacache_skipped_cond_steps = 0 self.teacache_skipped_uncond_steps = 0 else: prev_input = self.previous_modulated_input_uncond if is_uncond else self.previous_modulated_input_cond acc_distance_attr = 'accumulated_rel_l1_distance_uncond' if is_uncond else 'accumulated_rel_l1_distance_cond' temb_relative_l1 = relative_l1_distance(prev_input, e0) setattr(self, acc_distance_attr, getattr(self, acc_distance_attr) + temb_relative_l1) if getattr(self, acc_distance_attr) < self.rel_l1_thresh: should_calc = False self.teacache_counter += 1 else: should_calc = True setattr(self, acc_distance_attr, 0) if is_uncond: self.previous_modulated_input_uncond = e0.clone() if should_calc: self.previous_residual_uncond = None else: x += self.previous_residual_uncond self.teacache_skipped_cond_steps += 1 # print(f"Skipped uncond:{self.teacache_skipped_cond_steps}/{current_step}" ) else: self.previous_modulated_input_cond = e0.clone() if should_calc: self.previous_residual_cond = None else: x += self.previous_residual_cond self.teacache_skipped_uncond_steps += 1 # print(f"Skipped uncond:{self.teacache_skipped_uncond_steps}/{current_step}" ) if should_calc: if self.enable_teacache: ori_hidden_states = x.clone() # arguments kwargs = dict( e=e0, seq_lens=seq_lens, grid_sizes=grid_sizes, freqs=freqs, context=context, context_lens=context_lens) for block in self.blocks: if pipeline._interrupt: return [None] x = block(x, **kwargs) if self.enable_teacache: residual = ori_hidden_states # just to have a readable code torch.sub(x, ori_hidden_states, out=residual) if is_uncond: self.previous_residual_uncond = residual else: self.previous_residual_cond = residual del residual, ori_hidden_states # head x = self.head(x, e) # unpatchify x = self.unpatchify(x, grid_sizes) return [u.float() for u in x] 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): 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. """ # basic init 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) # init embeddings 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) # init output layer nn.init.zeros_(self.head.head.weight)