File size: 27,988 Bytes
1146a67
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
import torch
import torch.nn as nn
import torch.nn.functional as F
import math
from typing import Tuple, Optional
from einops import rearrange
from .wan_video_camera_controller import SimpleAdapter
try:
    import flash_attn_interface
    FLASH_ATTN_3_AVAILABLE = True
except ModuleNotFoundError:
    FLASH_ATTN_3_AVAILABLE = False

try:
    import flash_attn
    FLASH_ATTN_2_AVAILABLE = True
except ModuleNotFoundError:
    FLASH_ATTN_2_AVAILABLE = False

try:
    from sageattention import sageattn
    SAGE_ATTN_AVAILABLE = True
except ModuleNotFoundError:
    SAGE_ATTN_AVAILABLE = False
    
    
def flash_attention(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, num_heads: int, compatibility_mode=False):
    if compatibility_mode:
        q = rearrange(q, "b s (n d) -> b n s d", n=num_heads)
        k = rearrange(k, "b s (n d) -> b n s d", n=num_heads)
        v = rearrange(v, "b s (n d) -> b n s d", n=num_heads)
        x = F.scaled_dot_product_attention(q, k, v)
        x = rearrange(x, "b n s d -> b s (n d)", n=num_heads)
    elif FLASH_ATTN_3_AVAILABLE:
        q = rearrange(q, "b s (n d) -> b s n d", n=num_heads)
        k = rearrange(k, "b s (n d) -> b s n d", n=num_heads)
        v = rearrange(v, "b s (n d) -> b s n d", n=num_heads)
        x = flash_attn_interface.flash_attn_func(q, k, v)
        if isinstance(x,tuple):
            x = x[0]
        x = rearrange(x, "b s n d -> b s (n d)", n=num_heads)
    elif FLASH_ATTN_2_AVAILABLE:
        q = rearrange(q, "b s (n d) -> b s n d", n=num_heads)
        k = rearrange(k, "b s (n d) -> b s n d", n=num_heads)
        v = rearrange(v, "b s (n d) -> b s n d", n=num_heads)
        x = flash_attn.flash_attn_func(q, k, v)
        x = rearrange(x, "b s n d -> b s (n d)", n=num_heads)
    elif SAGE_ATTN_AVAILABLE:
        q = rearrange(q, "b s (n d) -> b n s d", n=num_heads)
        k = rearrange(k, "b s (n d) -> b n s d", n=num_heads)
        v = rearrange(v, "b s (n d) -> b n s d", n=num_heads)
        x = sageattn(q, k, v)
        x = rearrange(x, "b n s d -> b s (n d)", n=num_heads)
    else:
        q = rearrange(q, "b s (n d) -> b n s d", n=num_heads)
        k = rearrange(k, "b s (n d) -> b n s d", n=num_heads)
        v = rearrange(v, "b s (n d) -> b n s d", n=num_heads)
        x = F.scaled_dot_product_attention(q, k, v)
        x = rearrange(x, "b n s d -> b s (n d)", n=num_heads)
    return x

def scaled_dot_product_attention_with_mask(
    q: torch.Tensor,
    k: torch.Tensor,
    v: torch.Tensor,
    num_heads: int,
    attn_mask: Optional[torch.Tensor],
):
    """Always uses PyTorch SDPA because FlashAttention variants may not support arbitrary masks.

    Args:
        q,k,v: (B, S, D)
        attn_mask: float mask broadcastable to (B, num_heads, Sq, Sk) with 0 for allowed, -inf for disallowed
                  or bool mask broadcastable where False indicates disallowed.
    """
    q = rearrange(q, "b s (n d) -> b n s d", n=num_heads)
    k = rearrange(k, "b s (n d) -> b n s d", n=num_heads)
    v = rearrange(v, "b s (n d) -> b n s d", n=num_heads)
    if attn_mask is not None:
        # Make it broadcastable to (B, n, Sq, Sk)
        if attn_mask.dtype == torch.bool:
            mask = attn_mask
        else:
            mask = attn_mask
        if attn_mask.dim() == 3:
            mask = mask.unsqueeze(1)
        elif attn_mask.dim() != 4:
            raise ValueError(f"attn_mask must be 3D or 4D, got shape={attn_mask.shape}")
        x = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)
    else:
        x = F.scaled_dot_product_attention(q, k, v)
    x = rearrange(x, "b n s d -> b s (n d)", n=num_heads)
    return x


class MaskedCrossAttention(nn.Module):
    """Cross-attention with explicit attention mask support (used by IMCA)."""

    def __init__(self, dim: int, num_heads: int, eps: float = 1e-6):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.head_dim = dim // num_heads

        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 = RMSNorm(dim, eps=eps)
        self.norm_k = RMSNorm(dim, eps=eps)

    def forward(self, x_q: torch.Tensor, x_kv: torch.Tensor, attn_mask: Optional[torch.Tensor]):
        q = self.norm_q(self.q(x_q))
        k = self.norm_k(self.k(x_kv))
        v = self.v(x_kv)
        x = scaled_dot_product_attention_with_mask(q, k, v, num_heads=self.num_heads, attn_mask=attn_mask)
        return self.o(x)


class SharedTimestepAdaptivePromptEnhancement(nn.Module):
    """STAPE: I = I + (m_t + alpha1) * CrossAttn(I, T)."""

    def __init__(self, dim: int, num_heads: int, eps: float = 1e-6):
        super().__init__()
        self.attn = CrossAttention(dim, num_heads, eps=eps, has_image_input=False)
        # Dins-dimensional learnable residual gate (initialized to 0 for stability)
        self.mt = nn.Parameter(torch.zeros(1, dim))

    def forward(self, instance_tokens: torch.Tensor, caption_tokens: torch.Tensor, alpha1: torch.Tensor):
        # instance_tokens: (B, F, Nins, D)  caption_tokens: (B, Nctx, D) alpha1: (B, D)
        B, F_, Nins, D = instance_tokens.shape
        I = instance_tokens.reshape(B, F_ * Nins, D)
        delta = self.attn(I, caption_tokens)  # (B, F*Nins, D)
        gate = (self.mt.to(dtype=I.dtype, device=I.device) + alpha1).unsqueeze(1)  # (B, 1, D)
        I = I + gate * delta
        return I.reshape(B, F_, Nins, D)


class InstanceAwareMaskedCrossAttention(nn.Module):
    """IMCA: masked cross-attention from visual tokens to instance prompt tokens."""

    def __init__(self, dim: int, num_heads: int, eps: float = 1e-6):
        super().__init__()
        self.attn = MaskedCrossAttention(dim, num_heads, eps=eps)

    def forward(self, visual_tokens: torch.Tensor, instance_tokens: torch.Tensor, attn_mask: torch.Tensor):
        """Args:
            visual_tokens: (B, F*HW, D)
            instance_tokens: (B, F, Nins, D)
            attn_mask: (B, F, Nins, HW) bool OR float, where True/1 means instance-token attends this visual token.
        Returns:
            (B, F*HW, D)
        """
        B, Nv, D = visual_tokens.shape
        _, F_, Nins, _ = instance_tokens.shape
        HW = Nv // F_
        V = visual_tokens.reshape(B, F_, HW, D)
        I = instance_tokens
        # Convert mask to (B*F, HW, Nins) with 0 / -inf
        M = attn_mask
        if M.shape[-1] != HW:
            raise ValueError(f"attn_mask last dim must be HW={HW}, got {M.shape[-1]}")
        # (B,F,Nins,HW) -> (B,F,HW,Nins)
        M = M.permute(0, 1, 3, 2).contiguous()
        # 使用与 visual_tokens 相同的 dtype(通常是 bfloat16)
        target_dtype = visual_tokens.dtype
        if M.dtype == torch.bool:
            sdpa_mask = torch.where(M, torch.zeros((), device=M.device, dtype=target_dtype),
                                    torch.full((), float("-inf"), device=M.device, dtype=target_dtype))
        else:
            # assume already 0/-inf or similar
            sdpa_mask = M.to(dtype=target_dtype)
        # Merge batch and frame
        V_bf = V.reshape(B * F_, HW, D)
        I_bf = I.reshape(B * F_, Nins, D)
        sdpa_mask_bf = sdpa_mask.reshape(B * F_, HW, Nins)
        out = self.attn(V_bf, I_bf, sdpa_mask_bf)
        return out.reshape(B, F_ * HW, D)




def modulate(x: torch.Tensor, shift: torch.Tensor, scale: torch.Tensor):
    return (x * (1 + scale) + shift)


def sinusoidal_embedding_1d(dim, position):
    sinusoid = torch.outer(position.type(torch.float64), torch.pow(
        10000, -torch.arange(dim//2, dtype=torch.float64, device=position.device).div(dim//2)))
    x = torch.cat([torch.cos(sinusoid), torch.sin(sinusoid)], dim=1)
    return x.to(position.dtype)


def precompute_freqs_cis_3d(dim: int, end: int = 1024, theta: float = 10000.0):
    # 3d rope precompute
    f_freqs_cis = precompute_freqs_cis(dim - 2 * (dim // 3), end, theta)
    h_freqs_cis = precompute_freqs_cis(dim // 3, end, theta)
    w_freqs_cis = precompute_freqs_cis(dim // 3, end, theta)
    return f_freqs_cis, h_freqs_cis, w_freqs_cis


def precompute_freqs_cis(dim: int, end: int = 1024, theta: float = 10000.0):
    # 1d rope precompute
    freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)
                   [: (dim // 2)].double() / dim))
    freqs = torch.outer(torch.arange(end, device=freqs.device), freqs)
    freqs_cis = torch.polar(torch.ones_like(freqs), freqs)  # complex64
    return freqs_cis


def rope_apply(x, freqs, num_heads):
    x = rearrange(x, "b s (n d) -> b s n d", n=num_heads)
    x_out = torch.view_as_complex(x.to(torch.float64).reshape(
        x.shape[0], x.shape[1], x.shape[2], -1, 2))
    x_out = torch.view_as_real(x_out * freqs).flatten(2)
    return x_out.to(x.dtype)


class RMSNorm(nn.Module):
    def __init__(self, dim, eps=1e-5):
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(dim))

    def norm(self, x):
        return x * torch.rsqrt(x.pow(2).mean(dim=-1, keepdim=True) + self.eps)

    def forward(self, x):
        dtype = x.dtype
        return self.norm(x.float()).to(dtype) * self.weight


class AttentionModule(nn.Module):
    def __init__(self, num_heads):
        super().__init__()
        self.num_heads = num_heads
        
    def forward(self, q, k, v):
        x = flash_attention(q=q, k=k, v=v, num_heads=self.num_heads)
        return x


class SelfAttention(nn.Module):
    def __init__(self, dim: int, num_heads: int, eps: float = 1e-6):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.head_dim = dim // num_heads

        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 = RMSNorm(dim, eps=eps)
        self.norm_k = RMSNorm(dim, eps=eps)
        
        self.attn = AttentionModule(self.num_heads)

    def forward(self, x, freqs):
        q = self.norm_q(self.q(x))
        k = self.norm_k(self.k(x))
        v = self.v(x)
        q = rope_apply(q, freqs, self.num_heads)
        k = rope_apply(k, freqs, self.num_heads)
        x = self.attn(q, k, v)
        return self.o(x)


class CrossAttention(nn.Module):
    def __init__(self, dim: int, num_heads: int, eps: float = 1e-6, has_image_input: bool = False):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.head_dim = dim // num_heads

        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 = RMSNorm(dim, eps=eps)
        self.norm_k = RMSNorm(dim, eps=eps)
        self.has_image_input = has_image_input
        if has_image_input:
            self.k_img = nn.Linear(dim, dim)
            self.v_img = nn.Linear(dim, dim)
            self.norm_k_img = RMSNorm(dim, eps=eps)
            
        self.attn = AttentionModule(self.num_heads)

    def forward(self, x: torch.Tensor, y: torch.Tensor):
        if self.has_image_input:
            img = y[:, :257]
            ctx = y[:, 257:]
        else:
            ctx = y
        q = self.norm_q(self.q(x))
        k = self.norm_k(self.k(ctx))
        v = self.v(ctx)
        x = self.attn(q, k, v)

        if self.has_image_input:
            k_img = self.norm_k_img(self.k_img(img))
            v_img = self.v_img(img)
            y = flash_attention(q, k_img, v_img, num_heads=self.num_heads)
            x = x + y
        return self.o(x)


class GateModule(nn.Module):
    def __init__(self,):
        super().__init__()

    def forward(self, x, gate, residual):
        return x + gate * residual


class DiTBlock(nn.Module):
    def __init__(
        self,
        has_image_input: bool,
        dim: int,
        num_heads: int,
        ffn_dim: int,
        eps: float = 1e-6,
        enable_instancev: bool = False,
        stape: Optional[SharedTimestepAdaptivePromptEnhancement] = None,
    ):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.ffn_dim = ffn_dim
        self.enable_instancev = enable_instancev
        self.stape = stape

        self.self_attn = SelfAttention(dim, num_heads, eps)
        self.cross_attn = CrossAttention(dim, num_heads, eps, has_image_input=has_image_input)

        # IMCA is inserted between self-attention and cross-attention as a residual branch
        if enable_instancev:
            self.imca = InstanceAwareMaskedCrossAttention(dim, num_heads, eps=eps)
            # zero-initialized gated parameter m_v (paper Eq. 4)
            self.mv = nn.Parameter(torch.zeros(1))
            self.norm_imca = nn.LayerNorm(dim, eps=eps, elementwise_affine=False)
        else:
            self.imca = None
            self.mv = None
            self.norm_imca = None

        self.norm1 = nn.LayerNorm(dim, eps=eps, elementwise_affine=False)
        self.norm2 = nn.LayerNorm(dim, eps=eps, elementwise_affine=False)
        self.norm3 = nn.LayerNorm(dim, eps=eps)
        self.ffn = nn.Sequential(
            nn.Linear(dim, ffn_dim),
            nn.GELU(approximate='tanh'),
            nn.Linear(ffn_dim, dim),
        )
        self.modulation = nn.Parameter(torch.randn(1, 6, dim) / dim**0.5)
        self.gate = GateModule()

        # Better initialization for IMCA: copy weights from the native cross-attention (paper discussion)
        if enable_instancev and self.imca is not None:
            self._init_imca_from_cross_attention()

    def _init_imca_from_cross_attention(self):
        # copy q,k,v,o and norms
        try:
            self.imca.attn.q.load_state_dict(self.cross_attn.q.state_dict())
            self.imca.attn.k.load_state_dict(self.cross_attn.k.state_dict())
            self.imca.attn.v.load_state_dict(self.cross_attn.v.state_dict())
            self.imca.attn.o.load_state_dict(self.cross_attn.o.state_dict())
            self.imca.attn.norm_q.load_state_dict(self.cross_attn.norm_q.state_dict())
            self.imca.attn.norm_k.load_state_dict(self.cross_attn.norm_k.state_dict())
        except Exception:
            # if anything mismatches, skip silently (keeps compatibility)
            pass

    def forward(
        self,
        x: torch.Tensor,
        context: torch.Tensor,
        t_mod: torch.Tensor,
        freqs: torch.Tensor,
        instance_tokens: Optional[torch.Tensor] = None,
        instance_attn_mask: Optional[torch.Tensor] = None,
        empty_instance_tokens: Optional[torch.Tensor] = None,
        saug_drop_prob: float = 0.0,
    ):
        """Args:
            x: (B, F*H*W, D)
            context: global caption tokens T after embedding (B, Nctx, D)
            instance_tokens: I (B, F, Nins, D) after embedding
            instance_attn_mask: M (B, F, Nins, H*W) bool/float
            empty_instance_tokens: used for SAUG unconditional branch (same shape as instance_tokens)
        """
        has_seq = len(t_mod.shape) == 4
        chunk_dim = 2 if has_seq else 1

        shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
            self.modulation.to(dtype=t_mod.dtype, device=t_mod.device) + t_mod
        ).chunk(6, dim=chunk_dim)

        if has_seq:
            shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
                shift_msa.squeeze(2), scale_msa.squeeze(2), gate_msa.squeeze(2),
                shift_mlp.squeeze(2), scale_mlp.squeeze(2), gate_mlp.squeeze(2),
            )

        # 1) Self-attention (paper Eq. 3)
        input_x = modulate(self.norm1(x), shift_msa, scale_msa)
        x = self.gate(x, gate_msa, self.self_attn(input_x, freqs))

        # 2) IMCA (paper Eq. 4) + STAPE (paper Eq. 6)
        if self.enable_instancev and (self.imca is not None) and (instance_tokens is not None) and (instance_attn_mask is not None):
            # SAUG training-time drop: keep spatial masks but empty the instance prompts with probability p
            if isinstance(saug_drop_prob, torch.Tensor):
                saug_p = float(saug_drop_prob.detach().cpu().item())
            else:
                saug_p = float(saug_drop_prob)

            if self.training and saug_p > 0.0 and empty_instance_tokens is not None:
                if torch.rand((), device=x.device) < saug_p:
                    instance_tokens_use = empty_instance_tokens
                else:
                    instance_tokens_use = instance_tokens
            else:
                instance_tokens_use = instance_tokens

            # STAPE is shared across blocks (paper)
            if self.stape is not None:
                # reuse one AdaLN modulation vector as alpha1 (paper)
                alpha1 = gate_msa  # (B, D)
                instance_tokens_use = self.stape(instance_tokens_use, context, alpha1=alpha1)

            # IMCA: masked cross-attn from visual tokens to instance tokens
            imca_out = self.imca(self.norm_imca(x), instance_tokens_use, instance_attn_mask)
            x = x + self.mv.to(dtype=x.dtype, device=x.device) * imca_out

        # 3) Native cross-attention with global caption tokens (paper Eq. 5)
        x = x + self.cross_attn(self.norm3(x), context)

        # 4) FFN
        input_x = modulate(self.norm2(x), shift_mlp, scale_mlp)
        x = self.gate(x, gate_mlp, self.ffn(input_x))
        return x


class MLP(torch.nn.Module):
    def __init__(self, in_dim, out_dim, has_pos_emb=False):
        super().__init__()
        self.proj = torch.nn.Sequential(
            nn.LayerNorm(in_dim),
            nn.Linear(in_dim, in_dim),
            nn.GELU(),
            nn.Linear(in_dim, out_dim),
            nn.LayerNorm(out_dim)
        )
        self.has_pos_emb = has_pos_emb
        if has_pos_emb:
            self.emb_pos = torch.nn.Parameter(torch.zeros((1, 514, 1280)))

    def forward(self, x):
        if self.has_pos_emb:
            x = x + self.emb_pos.to(dtype=x.dtype, device=x.device)
        return self.proj(x)


class Head(nn.Module):
    def __init__(self, dim: int, out_dim: int, patch_size: Tuple[int, int, int], eps: float):
        super().__init__()
        self.dim = dim
        self.patch_size = patch_size
        self.norm = nn.LayerNorm(dim, eps=eps, elementwise_affine=False)
        self.head = nn.Linear(dim, out_dim * math.prod(patch_size))
        self.modulation = nn.Parameter(torch.randn(1, 2, dim) / dim**0.5)

    def forward(self, x, t_mod):
        if len(t_mod.shape) == 3:
            shift, scale = (self.modulation.unsqueeze(0).to(dtype=t_mod.dtype, device=t_mod.device) + t_mod.unsqueeze(2)).chunk(2, dim=2)
            x = (self.head(self.norm(x) * (1 + scale.squeeze(2)) + shift.squeeze(2)))
        else:
            shift, scale = (self.modulation.to(dtype=t_mod.dtype, device=t_mod.device) + t_mod).chunk(2, dim=1)
            x = (self.head(self.norm(x) * (1 + scale) + shift))
        return x


class WanModel(torch.nn.Module):
    def __init__(
        self,
        dim: int,
        in_dim: int,
        ffn_dim: int,
        out_dim: int,
        text_dim: int,
        freq_dim: int,
        eps: float,
        patch_size: Tuple[int, int, int],
        num_heads: int,
        num_layers: int,
        has_image_input: bool,
        enable_instancev: bool = False,
        has_image_pos_emb: bool = False,
        has_ref_conv: bool = False,
        add_control_adapter: bool = False,
        in_dim_control_adapter: int = 24,
        seperated_timestep: bool = False,
        require_vae_embedding: bool = True,
        require_clip_embedding: bool = True,
        fuse_vae_embedding_in_latents: bool = False,
    ):
        super().__init__()
        self.dim = dim
        self.in_dim = in_dim
        self.freq_dim = freq_dim
        self.has_image_input = has_image_input
        self.patch_size = patch_size
        self.seperated_timestep = seperated_timestep
        self.require_vae_embedding = require_vae_embedding
        self.require_clip_embedding = require_clip_embedding
        self.fuse_vae_embedding_in_latents = fuse_vae_embedding_in_latents

        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.enable_instancev = enable_instancev
        if enable_instancev:
            # STAPE is shared across all DiT blocks (paper Section 4.2)
            self.stape = SharedTimestepAdaptivePromptEnhancement(dim=dim, num_heads=num_heads, eps=eps)
        else:
            self.stape = None

        self.blocks = nn.ModuleList([
            DiTBlock(
                has_image_input=has_image_input,
                dim=dim,
                num_heads=num_heads,
                ffn_dim=ffn_dim,
                eps=eps,
                enable_instancev=enable_instancev,
                stape=self.stape,
            )
            for _ in range(num_layers)
        ])
        self.head = Head(dim, out_dim, patch_size, eps)
        head_dim = dim // num_heads
        self.freqs = precompute_freqs_cis_3d(head_dim)

        if has_image_input:
            self.img_emb = MLP(1280, dim, has_pos_emb=has_image_pos_emb)  # clip_feature_dim = 1280
        if has_ref_conv:
            self.ref_conv = nn.Conv2d(16, dim, kernel_size=(2, 2), stride=(2, 2))
        self.has_image_pos_emb = has_image_pos_emb
        self.has_ref_conv = has_ref_conv
        if add_control_adapter:
            self.control_adapter = SimpleAdapter(in_dim_control_adapter, dim, kernel_size=patch_size[1:], stride=patch_size[1:])
        else:
            self.control_adapter = None

    def patchify(self, x: torch.Tensor, control_camera_latents_input: Optional[torch.Tensor] = None):
        x = self.patch_embedding(x)
        if self.control_adapter is not None and control_camera_latents_input is not None:
            y_camera = self.control_adapter(control_camera_latents_input)
            x = [u + v for u, v in zip(x, y_camera)]
            x = x[0].unsqueeze(0)
        return x

    def unpatchify(self, x: torch.Tensor, grid_size: torch.Tensor):
        return rearrange(
            x, 'b (f h w) (x y z c) -> b c (f x) (h y) (w z)',
            f=grid_size[0], h=grid_size[1], w=grid_size[2], 
            x=self.patch_size[0], y=self.patch_size[1], z=self.patch_size[2]
        )

    def forward(self,
                x: torch.Tensor,
                timestep: torch.Tensor,
                context: torch.Tensor,
                instance_prompt_tokens: Optional[torch.Tensor] = None,
                instance_attn_mask: Optional[torch.Tensor] = None,
                empty_instance_prompt_tokens: Optional[torch.Tensor] = None,
                saug_drop_prob: float = 0.0,
                clip_feature: Optional[torch.Tensor] = None,
                y: Optional[torch.Tensor] = None,
                use_gradient_checkpointing: bool = False,
                use_gradient_checkpointing_offload: bool = False,
                **kwargs,
                ):
        t = self.time_embedding(
            sinusoidal_embedding_1d(self.freq_dim, timestep).to(x.dtype))
        t_mod = self.time_projection(t).unflatten(1, (6, self.dim))
        context = self.text_embedding(context)

        # Instance prompt tokens (paper Section 4.1): encode each instance prompt with the same text embedding layer
        if instance_prompt_tokens is not None:
            instance_tokens = self.text_embedding(instance_prompt_tokens)
        else:
            instance_tokens = None

        if empty_instance_prompt_tokens is not None:
            empty_instance_tokens = self.text_embedding(empty_instance_prompt_tokens)
        else:
            empty_instance_tokens = None
        

        # If SAUG unconditional tokens are not provided but InstanceV is enabled, default to zeros.
        # (For best results, provide the pretrained <extra_id> tokens as described in the paper.)
        if self.enable_instancev and (instance_tokens is not None) and (empty_instance_tokens is None):
            empty_instance_tokens = torch.zeros_like(instance_tokens)

        if self.has_image_input:
            x = torch.cat([x, y], dim=1)  # (b, c_x + c_y, f, h, w)
            clip_embdding = self.img_emb(clip_feature)
            context = torch.cat([clip_embdding, context], dim=1)
        
        x, (f, h, w) = self.patchify(x)
        
        freqs = torch.cat([
            self.freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1),
            self.freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1),
            self.freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1)
        ], dim=-1).reshape(f * h * w, 1, -1).to(x.device)
        
        def create_custom_forward(module):
            def custom_forward(*inputs):
                return module(*inputs)
            return custom_forward

        
        for block in self.blocks:
            use_instancev_inputs = (self.enable_instancev and (instance_tokens is not None) and (instance_attn_mask is not None))
            if self.training and use_gradient_checkpointing:
                if use_instancev_inputs:
                    if use_gradient_checkpointing_offload:
                        with torch.autograd.graph.save_on_cpu():
                            x = torch.utils.checkpoint.checkpoint(
                                create_custom_forward(block),
                                x, context, t_mod, freqs,
                                instance_tokens, instance_attn_mask, empty_instance_tokens,
                                torch.tensor(float(saug_drop_prob), device=x.device, dtype=x.dtype),
                                use_reentrant=False,
                            )
                    else:
                        x = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(block),
                            x, context, t_mod, freqs,
                            instance_tokens, instance_attn_mask, empty_instance_tokens,
                            torch.tensor(float(saug_drop_prob), device=x.device, dtype=x.dtype),
                            use_reentrant=False,
                        )
                else:
                    if use_gradient_checkpointing_offload:
                        with torch.autograd.graph.save_on_cpu():
                            x = torch.utils.checkpoint.checkpoint(
                                create_custom_forward(block),
                                x, context, t_mod, freqs,
                                use_reentrant=False,
                            )
                    else:
                        x = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(block),
                            x, context, t_mod, freqs,
                            use_reentrant=False,
                        )
            else:
                if use_instancev_inputs:
                    x = block(x, context, t_mod, freqs, instance_tokens, instance_attn_mask, empty_instance_tokens, saug_drop_prob)
                else:
                    x = block(x, context, t_mod, freqs)

        x = self.head(x, t)
        x = self.unpatchify(x, (f, h, w))
        return x


def apply_saug(eps_cond: torch.Tensor, eps_uncond: torch.Tensor, w: float) -> torch.Tensor:
    """Spatially-Aware Unconditional Guidance (SAUG), paper Eq. (7):
    eps_tilde = (1 + w) * eps_cond - w * eps_uncond
    where eps_uncond is predicted with *empty instance prompts* but the same spatial masks.
    """
    return (1.0 + w) * eps_cond - w * eps_uncond