Merge 447aa08620 into 7c81b2f27d

generate.py

@@ -4,7 +4,6 @@ import logging
 import os
 import sys
 import warnings
-from time import perf_counter
 from datetime import datetime
 
 warnings.filterwarnings('ignore')
@@ -20,7 +19,7 @@ try:
     import torch_musa
     from torch_musa.core.device import set_device
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 import wan
 from wan.configs import MAX_AREA_CONFIGS, SIZE_CONFIGS, SUPPORTED_SIZES, WAN_CONFIGS
@@ -366,7 +365,6 @@ def generate(args):
             logging.info(f"Extended prompt: {args.prompt}")
 
         logging.info("Creating WanT2V pipeline.")
-        start_time = perf_counter()
         wan_t2v = wan.WanT2V(
             config=cfg,
             checkpoint_dir=args.ckpt_dir,
@@ -377,8 +375,6 @@ def generate(args):
             use_usp=(args.ulysses_size > 1 or args.ring_size > 1),
             t5_cpu=args.t5_cpu,
         )
-        end_time = perf_counter()
-        logging.info(f"Creating WanT2V pipeline took {end_time - start_time:.2f} seconds.")
 
         logging.info(
             f"Generating {'image' if 't2i' in args.task else 'video'} ...")
@@ -392,6 +388,7 @@ def generate(args):
             guide_scale=args.sample_guide_scale,
             seed=args.base_seed,
             offload_model=args.offload_model)
+
     elif "i2v" in args.task:
         if args.prompt is None:
             args.prompt = EXAMPLE_PROMPT[args.task]["prompt"]
@@ -425,7 +422,6 @@ def generate(args):
             logging.info(f"Extended prompt: {args.prompt}")
 
         logging.info("Creating WanI2V pipeline.")
-        start_time = perf_counter()
         wan_i2v = wan.WanI2V(
             config=cfg,
             checkpoint_dir=args.ckpt_dir,
@@ -436,8 +432,6 @@ def generate(args):
             use_usp=(args.ulysses_size > 1 or args.ring_size > 1),
             t5_cpu=args.t5_cpu,
         )
-        end_time = perf_counter()
-        logging.info(f"Creating WanI2V pipeline took {end_time - start_time:.2f} seconds.")
 
         logging.info("Generating video ...")
         video = wan_i2v.generate(
@@ -586,7 +580,6 @@ def generate(args):
                 value_range=(-1, 1))
         else:
             logging.info(f"Saving generated video to {args.save_file}")
-            start_time = perf_counter()
             cache_video(
                 tensor=video[None],
                 save_file=args.save_file,
@@ -594,8 +587,6 @@ def generate(args):
                 nrow=1,
                 normalize=True,
                 value_range=(-1, 1))
-            end_time = perf_counter()
-            logging.info(f"Saving Video took {end_time - start_time:.2f} seconds")
     logging.info("Finished.")
 
 

wan/distributed/fsdp.py

@@ -13,7 +13,7 @@ try:
     import torch_musa
     from torch_musa.core.memory import empty_cache
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 def shard_model(
     model,

wan/distributed/xdit_context_parallel.py

@@ -14,11 +14,24 @@ try:
     import torch_musa.core.amp as amp
     attn_type = AttnType.TORCH
 except ImportError:
-    pass
+    torch_musa = None
 
 from ..modules.model import sinusoidal_embedding_1d
 
 
+def pad_freqs(original_tensor, target_len):
+    seq_len, s1, s2 = original_tensor.shape
+    pad_size = target_len - seq_len
+    padding_tensor = torch.ones(
+        pad_size,
+        s1,
+        s2,
+        dtype=original_tensor.dtype,
+        device=original_tensor.device)
+    padded_tensor = torch.cat([original_tensor, padding_tensor], dim=0)
+    return padded_tensor
+
+
 def pad_tensor(original_tensor, target_len, pad_value=0.0):
     seq_len, s1, s2 = original_tensor.shape
     pad_size = target_len - seq_len
@@ -34,6 +47,47 @@ def pad_tensor(original_tensor, target_len, pad_value=0.0):
 
 @amp.autocast(enabled=False)
 def rope_apply(x, grid_sizes, freqs):
+    """
+    x:          [B, L, N, C].
+    grid_sizes: [B, 3].
+    freqs:      [M, C // 2].
+    """
+    s, n, c = x.size(1), 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.tolist()):
+        seq_len = f * h * w
+
+        # precompute multipliers
+        x_i = torch.view_as_complex(x[i, :s].to(torch.float64).reshape(
+            s, n, -1, 2))
+        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
+        sp_size = get_sequence_parallel_world_size()
+        sp_rank = get_sequence_parallel_rank()
+        freqs_i = pad_freqs(freqs_i, s * sp_size)
+        s_per_rank = s
+        freqs_i_rank = freqs_i[(sp_rank * s_per_rank):((sp_rank + 1) *
+                                                       s_per_rank), :, :]
+        x_i = torch.view_as_real(x_i * freqs_i_rank).flatten(2)
+        x_i = torch.cat([x_i, x[i, s:]])
+
+        # append to collection
+        output.append(x_i)
+    return torch.stack(output).float()
+
+
+@amp.autocast(enabled=False)
+def rope_apply_musa(x, grid_sizes, freqs):
     """
     x:          [B, L, N, C].
     grid_sizes: [B, 3].
@@ -141,11 +195,15 @@ def usp_dit_forward(
     # params
     dtype = self.patch_embedding.weight.dtype
     device = self.patch_embedding.weight.device
-    if self.freqs[0].dtype != dtype or self.freqs[0].device != device:
-        self.freqs = (
-            self.freqs[0].to(dtype=dtype, device=device),
-            self.freqs[-1].to(dtype=dtype, device=device)
-        )
+    if torch_musa is not None:
+        if self.freqs[0].dtype != dtype or self.freqs[0].device != device:
+            self.freqs = (
+                self.freqs[0].to(dtype=dtype, device=device),
+                self.freqs[-1].to(dtype=dtype, device=device)
+            )
+    else:
+        if self.freqs.dtype != dtype or self.freqs.device != device:
+            self.freqs = self.freqs.to(dtype=dtype, device=device)
 
     if self.model_type != 'vace' and y is not None:
         x = [torch.cat([u, v], dim=0) for u, v in zip(x, y)]
@@ -163,9 +221,11 @@ def usp_dit_forward(
     ])
 
     # time embeddings
-    e = self.time_embedding(
-        sinusoidal_embedding_1d(self.freq_dim, t))
-    e0 = self.time_projection(e).unflatten(1, (6, self.dim))
+    with amp.autocast(dtype=torch.float32):
+        e = self.time_embedding(
+            sinusoidal_embedding_1d(self.freq_dim, t).float())
+        e0 = self.time_projection(e).unflatten(1, (6, self.dim))
+        assert e.dtype == torch.float32 and e0.dtype == torch.float32
 
     # context
     context_lens = None
@@ -209,7 +269,7 @@ def usp_dit_forward(
 
     # unpatchify
     x = self.unpatchify(x, grid_sizes)
-    return x
+    return [u.float() for u in x]
 
 
 def usp_attn_forward(self,
@@ -232,8 +292,13 @@ def usp_attn_forward(self,
         return q, k, v
 
     q, k, v = qkv_fn(x)
-    q = rope_apply(q, grid_sizes, freqs)
-    k = rope_apply(k, grid_sizes, freqs)
+        
+    if torch_musa is not None:
+        q = rope_apply_musa(q, grid_sizes, freqs)
+        k = rope_apply_musa(k, grid_sizes, freqs)
+    else:
+        q = rope_apply(q, grid_sizes, freqs)
+        k = rope_apply(k, grid_sizes, freqs)
 
     # TODO: We should use unpaded q,k,v for attention.
     # k_lens = seq_lens // get_sequence_parallel_world_size()

wan/first_last_frame2video.py

@@ -23,7 +23,7 @@ try:
     from torch_musa.core.memory import empty_cache
     from torch_musa.core.device import synchronize
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from .distributed.fsdp import shard_model
 from .modules.clip import CLIPModel

wan/image2video.py

@@ -6,7 +6,6 @@ import os
 import random
 import sys
 import types
-from time import perf_counter
 from contextlib import contextmanager
 from functools import partial
 
@@ -24,7 +23,7 @@ try:
     from torch_musa.core.memory import empty_cache
     from torch_musa.core.device import synchronize
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from .distributed.fsdp import shard_model
 from .modules.clip import CLIPModel
@@ -231,7 +230,6 @@ class WanI2V:
             n_prompt = self.sample_neg_prompt
 
         # preprocess
-        start_time = perf_counter()
         if not self.t5_cpu:
             self.text_encoder.model.to(self.device)
             context = self.text_encoder([input_prompt], self.device)
@@ -243,18 +241,12 @@ class WanI2V:
             context_null = self.text_encoder([n_prompt], torch.device('cpu'))
             context = [t.to(self.device) for t in context]
             context_null = [t.to(self.device) for t in context_null]
-        end_time = perf_counter()
-        logging.info(f"T5 Encoding took {end_time - start_time:.2f} seconds.")
 
-        start_time = perf_counter()
         self.clip.model.to(self.device)
         clip_context = self.clip.visual([img[:, None, :, :]])
         if offload_model:
             self.clip.model.cpu()
-        end_time = perf_counter()
-        logging.info(f"CLIP took {end_time - start_time:.2f} seconds.")
 
-        start_time = perf_counter()
         y = self.vae.encode([
             torch.concat([
                 torch.nn.functional.interpolate(
@@ -264,9 +256,6 @@ class WanI2V:
             ],
                          dim=1).to(self.device)
         ])[0]
-        end_time = perf_counter()
-        logging.info(f"VAE Encoding took {end_time - start_time:.2f} seconds.")
-
         y = torch.concat([msk, y])
 
         @contextmanager
@@ -319,7 +308,6 @@ class WanI2V:
             if offload_model:
                 empty_cache()
 
-            start_time = perf_counter()
             self.model.to(self.device)
             for _, t in enumerate(tqdm(timesteps)):
                 latent_model_input = [latent.to(self.device)]
@@ -353,18 +341,13 @@ class WanI2V:
 
                 x0 = [latent.to(self.device)]
                 del latent_model_input, timestep
-            end_time = perf_counter()
-            logging.info(f"Sampling took {end_time - start_time:.2f} seconds.")
 
             if offload_model:
                 self.model.cpu()
                 empty_cache()
 
             if self.rank == 0:
-                start_time = perf_counter()
                 videos = self.vae.decode(x0)
-                end_time = perf_counter()
-                logging.info(f"VAE Decoding took {end_time - start_time:.2f} seconds.")
 
         del noise, latent
         del sample_scheduler

wan/modules/attention.py

@@ -20,7 +20,7 @@ try:
     FLASH_ATTN_3_AVAILABLE = False
     FLASH_ATTN_2_AVAILABLE = False
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 
 __all__ = [
@@ -180,9 +180,8 @@ def attention(
         k = k.transpose(1, 2).to(dtype)
         v = v.transpose(1, 2).to(dtype)
 
-        with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False):
-            out = torch.nn.functional.scaled_dot_product_attention(
-                q, k, v, attn_mask=attn_mask, dropout_p=dropout_p, is_causal=causal, scale=softmax_scale)
+        out = torch.nn.functional.scaled_dot_product_attention(
+            q, k, v, attn_mask=attn_mask, dropout_p=dropout_p, is_causal=causal, scale=softmax_scale)
 
         out = out.transpose(1, 2).contiguous()
         return out

wan/modules/clip.py

@@ -18,7 +18,7 @@ try:
     import torch_musa.core.amp as amp
     from .attention import attention as flash_attention
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 __all__ = [
     'XLMRobertaCLIP',
@@ -37,7 +37,7 @@ def pos_interpolate(pos, seq_len):
         return torch.cat([
             pos[:, :n],
             F.interpolate(
-                pos[:, n:].reshape(1, src_grid, src_grid, -1).permute(
+                pos[:, n:].float().reshape(1, src_grid, src_grid, -1).permute(
                     0, 3, 1, 2),
                 size=(tar_grid, tar_grid),
                 mode='bicubic',
@@ -52,6 +52,12 @@ class QuickGELU(nn.Module):
         return x * torch.sigmoid(1.702 * x)
 
 
+class LayerNorm(nn.LayerNorm):
+
+    def forward(self, x):
+        return super().forward(x.float()).type_as(x)
+
+
 class SelfAttention(nn.Module):
 
     def __init__(self,
@@ -84,7 +90,10 @@ class SelfAttention(nn.Module):
 
         # compute attention
         p = self.attn_dropout if self.training else 0.0
-        x = flash_attention(q, k, v, dropout_p=p, causal=self.causal)
+        if torch_musa is not None:
+            x = flash_attention(q, k, v, dropout_p=p, causal=self.causal)
+        else:
+            x = flash_attention(q, k, v, dropout_p=p, causal=self.causal, version=2)
         x = x.reshape(b, s, c)
 
         # output
@@ -133,10 +142,10 @@ class AttentionBlock(nn.Module):
         self.norm_eps = norm_eps
 
         # layers
-        self.norm1 = nn.LayerNorm(dim, eps=norm_eps)
+        self.norm1 = LayerNorm(dim, eps=norm_eps)
         self.attn = SelfAttention(dim, num_heads, causal, attn_dropout,
                                   proj_dropout)
-        self.norm2 = nn.LayerNorm(dim, eps=norm_eps)
+        self.norm2 = LayerNorm(dim, eps=norm_eps)
         if activation == 'swi_glu':
             self.mlp = SwiGLU(dim, int(dim * mlp_ratio))
         else:
@@ -179,7 +188,7 @@ class AttentionPool(nn.Module):
         self.to_q = nn.Linear(dim, dim)
         self.to_kv = nn.Linear(dim, dim * 2)
         self.proj = nn.Linear(dim, dim)
-        self.norm = nn.LayerNorm(dim, eps=norm_eps)
+        self.norm = LayerNorm(dim, eps=norm_eps)
         self.mlp = nn.Sequential(
             nn.Linear(dim, int(dim * mlp_ratio)),
             QuickGELU() if activation == 'quick_gelu' else nn.GELU(),
@@ -196,7 +205,10 @@ class AttentionPool(nn.Module):
         k, v = self.to_kv(x).view(b, s, 2, n, d).unbind(2)
 
         # compute attention
-        x = flash_attention(q, k, v, version=2)
+        if torch_musa is not None:
+            x = flash_attention(q, k, v)
+        else:
+            x = flash_attention(q, k, v, version=2)
         x = x.reshape(b, 1, c)
 
         # output
@@ -261,13 +273,13 @@ class VisionTransformer(nn.Module):
         self.dropout = nn.Dropout(embedding_dropout)
 
         # transformer
-        self.pre_norm = nn.LayerNorm(dim, eps=norm_eps) if pre_norm else None
+        self.pre_norm = LayerNorm(dim, eps=norm_eps) if pre_norm else None
         self.transformer = nn.Sequential(*[
             AttentionBlock(dim, mlp_ratio, num_heads, post_norm, False,
                            activation, attn_dropout, proj_dropout, norm_eps)
             for _ in range(num_layers)
         ])
-        self.post_norm = nn.LayerNorm(dim, eps=norm_eps)
+        self.post_norm = LayerNorm(dim, eps=norm_eps)
 
         # head
         if pool_type == 'token':

wan/modules/model.py

@@ -7,7 +7,6 @@ import torch.nn as nn
 from diffusers.configuration_utils import ConfigMixin, register_to_config
 from diffusers.models.modeling_utils import ModelMixin
 
-from wan.utils.platform import get_device
 from wan.modules.attention import flash_attention
 
 try:
@@ -15,7 +14,7 @@ try:
     import torch_musa.core.amp as amp
     from wan.modules.attention import attention as flash_attention
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 __all__ = ['WanModel']
 
@@ -27,7 +26,7 @@ def sinusoidal_embedding_1d(dim, position):
     # preprocess
     assert dim % 2 == 0
     half = dim // 2
-    position = position.type(torch.bfloat16)
+    position = position.type(torch.float32)
 
     # calculation
     sinusoid = torch.outer(
@@ -36,6 +35,17 @@ def sinusoidal_embedding_1d(dim, position):
     return x
 
 
+@amp.autocast(enabled=False)
+def rope_params(max_seq_len, dim, theta=10000):
+    assert dim % 2 == 0
+    freqs = torch.outer(
+        torch.arange(max_seq_len),
+        1.0 / torch.pow(theta,
+                        torch.arange(0, dim, 2).to(torch.float64).div(dim)))
+    freqs = torch.polar(torch.ones_like(freqs), freqs)
+    return freqs
+
+
 @amp.autocast(enabled=False)
 def rope_params_real(
     max_seq_len, dim, theta=10000, dtype=torch.float32, device=torch.device("cpu")
@@ -69,6 +79,37 @@ def rope_params_imag(
 @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.tolist()):
+        seq_len = f * h * w
+
+        # precompute multipliers
+        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][: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 = torch.view_as_real(x_i * freqs_i).flatten(2)
+        x_i = torch.cat([x_i, x[i, seq_len:]])
+
+        # append to collection
+        output.append(x_i)
+    return torch.stack(output).float()
+
+
+@amp.autocast(enabled=False)
+def rope_apply_musa(x, grid_sizes, freqs):
+    n, c = x.size(2), x.size(3) // 2
     c0 = c - 2 * (c // 3)
     c1 = c // 3
     c2 = c // 3
@@ -134,6 +175,19 @@ class WanRMSNorm(nn.Module):
         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.float()).type_as(x)
+
+
 class WanSelfAttention(nn.Module):
 
     def __init__(self,
@@ -178,12 +232,22 @@ class WanSelfAttention(nn.Module):
 
         q, k, v = qkv_fn(x)
 
-        x = flash_attention(
-            q=rope_apply(q, grid_sizes, freqs),
-            k=rope_apply(k, grid_sizes, freqs),
-            v=v,
-            k_lens=seq_lens,
-            window_size=self.window_size)
+        if torch_musa is not None:
+            x = flash_attention(
+                q=rope_apply_musa(q, grid_sizes, freqs),
+                k=rope_apply_musa(k, grid_sizes, freqs),
+                v=v,
+                k_lens=seq_lens,
+                window_size=self.window_size,
+            )
+        else:
+            x = flash_attention(
+                q=rope_apply(q, grid_sizes, freqs),
+                k=rope_apply(k, grid_sizes, freqs),
+                v=v,
+                k_lens=seq_lens,
+                window_size=self.window_size,
+            )
 
         # output
         x = x.flatten(2)
@@ -288,10 +352,10 @@ class WanAttentionBlock(nn.Module):
         self.eps = eps
 
         # layers
-        self.norm1 = nn.LayerNorm(dim, eps, elementwise_affine=False)
+        self.norm1 = WanLayerNorm(dim, eps)
         self.self_attn = WanSelfAttention(dim, num_heads, window_size, qk_norm,
                                           eps)
-        self.norm3 = nn.LayerNorm(
+        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,
@@ -299,7 +363,7 @@ class WanAttentionBlock(nn.Module):
                                                                       (-1, -1),
                                                                       qk_norm,
                                                                       eps)
-        self.norm2 = nn.LayerNorm(dim, eps, elementwise_affine=False)
+        self.norm2 = WanLayerNorm(dim, eps)
         self.ffn = nn.Sequential(
             nn.Linear(dim, ffn_dim), nn.GELU(approximate='tanh'),
             nn.Linear(ffn_dim, dim))
@@ -325,19 +389,24 @@ class WanAttentionBlock(nn.Module):
             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)
+        assert e.dtype == torch.float32
+        with amp.autocast(dtype=torch.float32):
+            e = (self.modulation + e).chunk(6, dim=1)
+        assert e[0].dtype == torch.float32
 
         # self-attention
         y = self.self_attn(
-            self.norm1(x) * (1 + e[1]) + e[0], seq_lens, grid_sizes,
+            self.norm1(x).float() * (1 + e[1]) + e[0], seq_lens, grid_sizes,
             freqs)
-        x = x + y * e[2]
+        with amp.autocast(dtype=torch.float32):
+            x = x + y * e[2]
 
         # cross-attention & ffn function
         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) * (1 + e[4]) + e[3])
-            x = x + y * e[5]
+            y = self.ffn(self.norm2(x).float() * (1 + e[4]) + e[3])
+            with amp.autocast(dtype=torch.float32):
+                x = x + y * e[5]
             return x
 
         x = cross_attn_ffn(x, context, context_lens, e)
@@ -355,7 +424,7 @@ class Head(nn.Module):
 
         # layers
         out_dim = math.prod(patch_size) * out_dim
-        self.norm = nn.LayerNorm(dim, eps, elementwise_affine=False)
+        self.norm = WanLayerNorm(dim, eps)
         self.head = nn.Linear(dim, out_dim)
 
         # modulation
@@ -367,8 +436,10 @@ class Head(nn.Module):
             x(Tensor): Shape [B, L1, C]
             e(Tensor): Shape [B, C]
         """
-        e = (self.modulation + e.unsqueeze(1)).chunk(2, dim=1)
-        x = self.head(self.norm(x) * (1 + e[1]) + e[0])
+        assert e.dtype == torch.float32
+        with amp.autocast(dtype=torch.float32):
+            e = (self.modulation + e.unsqueeze(1)).chunk(2, dim=1)
+            x = (self.head(self.norm(x) * (1 + e[1]) + e[0]))
         return x
 
 
@@ -502,23 +573,33 @@ class WanModel(ModelMixin, ConfigMixin):
         # buffers (don't use register_buffer otherwise dtype will be changed in to())
         assert (dim % num_heads) == 0 and (dim // num_heads) % 2 == 0
         d = dim // num_heads
-        freqs_real = torch.cat(
-            [
-                rope_params_real(1024, d - 4 * (d // 6)),
-                rope_params_real(1024, 2 * (d // 6)),
-                rope_params_real(1024, 2 * (d // 6)),
-            ],
-            dim=1,
-        )
-        freqs_imag = torch.cat(
-            [
-                rope_params_imag(1024, d - 4 * (d // 6)),
-                rope_params_imag(1024, 2 * (d // 6)),
-                rope_params_imag(1024, 2 * (d // 6)),
-            ],
-            dim=1,
-        )
-        self.freqs = (freqs_real, freqs_imag)
+        if torch_musa is not None:
+            freqs_real = torch.cat(
+                [
+                    rope_params_real(1024, d - 4 * (d // 6)),
+                    rope_params_real(1024, 2 * (d // 6)),
+                    rope_params_real(1024, 2 * (d // 6)),
+                ],
+                dim=1,
+            )
+            freqs_imag = torch.cat(
+                [
+                    rope_params_imag(1024, d - 4 * (d // 6)),
+                    rope_params_imag(1024, 2 * (d // 6)),
+                    rope_params_imag(1024, 2 * (d // 6)),
+                ],
+                dim=1,
+            )
+            self.freqs = (freqs_real, freqs_imag)
+        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,
+            )
 
         if model_type == 'i2v' or model_type == 'flf2v':
             self.img_emb = MLPProj(1280, dim, flf_pos_emb=model_type == 'flf2v')
@@ -561,11 +642,15 @@ class WanModel(ModelMixin, ConfigMixin):
         # params
         dtype = self.patch_embedding.weight.dtype
         device = self.patch_embedding.weight.device
-        if self.freqs[0].dtype != dtype or self.freqs[0].device != device:
-            self.freqs = (
-                self.freqs[0].to(dtype=dtype, device=device),
-                self.freqs[-1].to(dtype=dtype, device=device)
-            )
+        if torch_musa is not None:
+            if self.freqs[0].dtype != dtype or self.freqs[0].device != device:
+                self.freqs = (
+                    self.freqs[0].to(dtype=dtype, device=device),
+                    self.freqs[-1].to(dtype=dtype, device=device)
+                )
+        else:
+            if self.freqs.dtype != dtype or self.freqs.device != device:
+                self.freqs = self.freqs.to(dtype=dtype, device=device)
 
         if y is not None:
             x = [torch.cat([u, v], dim=0) for u, v in zip(x, y)]
@@ -583,9 +668,11 @@ class WanModel(ModelMixin, ConfigMixin):
         ])
 
         # time embeddings
-        e = self.time_embedding(
-            sinusoidal_embedding_1d(self.freq_dim, t))
-        e0 = self.time_projection(e).unflatten(1, (6, self.dim))
+        with amp.autocast(dtype=torch.float32):
+            e = self.time_embedding(
+                sinusoidal_embedding_1d(self.freq_dim, t).float())
+            e0 = self.time_projection(e).unflatten(1, (6, self.dim))
+            assert e.dtype == torch.float32 and e0.dtype == torch.float32
 
         # context
         context_lens = None
@@ -617,7 +704,7 @@ class WanModel(ModelMixin, ConfigMixin):
 
         # unpatchify
         x = self.unpatchify(x, grid_sizes)
-        return x
+        return [u.float() for u in x]
 
     def unpatchify(self, x, grid_sizes):
         r"""

wan/modules/t5.py

@@ -6,15 +6,15 @@ import math
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
+from torch.cuda import current_device
 
 try:
     import torch_musa
     from torch_musa.core.device import current_device
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
-from wan.modules.tokenizers import HuggingfaceTokenizer
-from wan.utils.platform import get_device
+from .tokenizers import HuggingfaceTokenizer
 
 __all__ = [
     'T5Model',
@@ -66,8 +66,10 @@ class T5LayerNorm(nn.Module):
         self.weight = nn.Parameter(torch.ones(dim))
 
     def forward(self, x):
-        x = x * torch.rsqrt(x.pow(2).mean(dim=-1, keepdim=True) +
+        x = x * torch.rsqrt(x.float().pow(2).mean(dim=-1, keepdim=True) +
                             self.eps)
+        if self.weight.dtype in [torch.float16, torch.bfloat16]:
+            x = x.type_as(self.weight)
         return self.weight * x
 
 
@@ -115,7 +117,7 @@ class T5Attention(nn.Module):
 
         # compute attention (T5 does not use scaling)
         attn = torch.einsum('binc,bjnc->bnij', q, k) + attn_bias
-        attn = F.softmax(attn, dim=-1)
+        attn = F.softmax(attn.float(), dim=-1).type_as(attn)
         x = torch.einsum('bnij,bjnc->binc', attn, v)
 
         # output
@@ -260,7 +262,7 @@ class T5RelativeEmbedding(nn.Module):
 
         # embeddings for small and large positions
         max_exact = num_buckets // 2
-        rel_pos_large = max_exact + (torch.log(rel_pos / max_exact) /
+        rel_pos_large = max_exact + (torch.log(rel_pos.float() / max_exact) /
                                      math.log(self.max_dist / max_exact) *
                                      (num_buckets - max_exact)).long()
         rel_pos_large = torch.min(
@@ -480,7 +482,7 @@ class T5EncoderModel:
         self,
         text_len,
         dtype=torch.bfloat16,
-        device=get_device(),
+        device=current_device(),
         checkpoint_path=None,
         tokenizer_path=None,
         shard_fn=None,

wan/modules/vace_model.py

@@ -8,7 +8,7 @@ try:
     import torch_musa
     import torch_musa.core.amp as amp
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from .model import WanAttentionBlock, WanModel, sinusoidal_embedding_1d
 

wan/modules/vae.py

@@ -1,8 +1,8 @@
 # Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
 import logging
-from math import sqrt
 
 import torch
+import torch.cuda.amp as amp
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn import Upsample
@@ -10,8 +10,9 @@ from einops import rearrange
 
 try:
     import torch_musa
+    import torch_musa.core.amp as amp
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from wan.utils.platform import get_device
 
@@ -52,17 +53,23 @@ class RMS_norm(nn.Module):
         shape = (dim, *broadcastable_dims) if channel_first else (dim,)
 
         self.channel_first = channel_first
-        self.scale = sqrt(dim)
+        self.scale = dim**0.5
         self.gamma = nn.Parameter(torch.ones(shape))
         self.bias = nn.Parameter(torch.zeros(shape)) if bias else 0.
 
     def forward(self, x):
-        return (
-            F.normalize(x.float(), dim=(1 if self.channel_first else -1)).type_as(x)
-            * self.scale
-            * self.gamma
-            + self.bias
-        )
+        return F.normalize(
+            x, dim=(1 if self.channel_first else
+                    -1)) * self.scale * self.gamma + self.bias
+
+
+class Upsample(nn.Upsample):
+
+    def forward(self, x):
+        """
+        Fix bfloat16 support for nearest neighbor interpolation.
+        """
+        return super().forward(x.float()).type_as(x)
 
 
 class Resample(nn.Module):
@@ -255,10 +262,6 @@ class AttentionBlock(nn.Module):
             q,
             k,
             v,
-            attn_mask=None,
-            dropout_p=0.0,
-            is_causal=False,
-            scale=None,
         )
         x = x.squeeze(1).permute(0, 2, 1).reshape(b * t, c, h, w)
 
@@ -627,7 +630,7 @@ class WanVAE:
     def __init__(self,
                  z_dim=16,
                  vae_pth='cache/vae_step_411000.pth',
-                 dtype=torch.bfloat16,
+                 dtype=torch.float,
                  device=get_device()):
         self.dtype = dtype
         self.device = device
@@ -654,12 +657,16 @@ class WanVAE:
         """
         videos: A list of videos each with shape [C, T, H, W].
         """
-        return [
-            self.model.encode(u.unsqueeze(0), self.scale).squeeze(0) for u in videos
-        ]
+        with amp.autocast(dtype=self.dtype):
+            return [
+                self.model.encode(u.unsqueeze(0), self.scale).float().squeeze(0)
+                for u in videos
+            ]
 
     def decode(self, zs):
-        return [
-            self.model.decode(u.unsqueeze(0), self.scale).clamp_(-1, 1).squeeze(0)
-            for u in zs
-        ]
+        with amp.autocast(dtype=self.dtype):
+            return [
+                self.model.decode(u.unsqueeze(0),
+                                  self.scale).float().clamp_(-1, 1).squeeze(0)
+                for u in zs
+            ]

wan/text2video.py

@@ -6,7 +6,6 @@ import os
 import random
 import sys
 import types
-from time import perf_counter
 from contextlib import contextmanager
 from functools import partial
 
@@ -22,7 +21,7 @@ try:
     from torch_musa.core.memory import empty_cache
     from torch_musa.core.device import synchronize
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from .distributed.fsdp import shard_model
 from .modules.model import WanModel
@@ -181,7 +180,6 @@ class WanT2V:
         seed_g = torch.Generator(device=self.device)
         seed_g.manual_seed(seed)
 
-        start_time = perf_counter()
         if not self.t5_cpu:
             self.text_encoder.model.to(self.device)
             context = self.text_encoder([input_prompt], self.device)
@@ -193,8 +191,6 @@ class WanT2V:
             context_null = self.text_encoder([n_prompt], torch.device('cpu'))
             context = [t.to(self.device) for t in context]
             context_null = [t.to(self.device) for t in context_null]
-        end_time = perf_counter()
-        logging.info(f"T5 Encoding Context took {end_time - start_time:.2f} seconds.")
 
         noise = [
             torch.randn(
@@ -243,14 +239,13 @@ class WanT2V:
             arg_c = {'context': context, 'seq_len': seq_len}
             arg_null = {'context': context_null, 'seq_len': seq_len}
 
-            start_time = perf_counter()
-            self.model.to(self.device)
             for _, t in enumerate(tqdm(timesteps)):
                 latent_model_input = latents
                 timestep = [t]
 
                 timestep = torch.stack(timestep)
 
+                self.model.to(self.device)
                 noise_pred_cond = self.model(
                     latent_model_input, t=timestep, **arg_c)[0]
                 noise_pred_uncond = self.model(
@@ -266,18 +261,13 @@ class WanT2V:
                     return_dict=False,
                     generator=seed_g)[0]
                 latents = [temp_x0.squeeze(0)]
-            end_time = perf_counter()
-            logging.info(f"Sampling took {end_time - start_time:.2f} seconds.")
 
             x0 = latents
             if offload_model:
                 self.model.cpu()
                 empty_cache()
             if self.rank == 0:
-                start_time = perf_counter()
                 videos = self.vae.decode(x0)
-                end_time = perf_counter()
-                logging.info(f"VAE Decoding took {end_time - start_time:.2f} seconds.")
 
         del noise, latents
         del sample_scheduler

wan/utils/platform.py

@@ -5,7 +5,7 @@ import torch
 try:
     import torch_musa
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 
 def _is_musa():
@@ -31,4 +31,4 @@ def get_torch_distributed_backend() -> str:
     elif _is_musa():
         return "mccl"
     else:
-        raise NotImplementedError("No Accelerators(AMD/NV/MTT GPU, AMD MI instinct accelerators) available")
+        raise NotImplementedError("No Accelerators(NV/MTT GPU) available")

wan/vace.py

@@ -27,7 +27,7 @@ try:
     from torch_musa.core.memory import empty_cache
     from torch_musa.core.device import synchronize
 except ModuleNotFoundError:
-    pass
+    torch_musa = None
 
 from .modules.vace_model import VaceWanModel
 from .text2video import (