File size: 29,672 Bytes
048b8bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
"""
TinyFlux-Deep with Expert Predictor

Integrates a distillation pathway for SD1.5-flow timestep expertise.
During training: learns to predict expert features from (timestep, CLIP).
During inference: runs standalone, no expert needed.

Based on TinyFlux-Deep: 15 double + 25 single blocks.
"""

import torch
import torch.nn as nn
import torch.nn.functional as F
import math
from dataclasses import dataclass
from typing import Optional, Tuple, Dict


@dataclass
class TinyFluxDeepConfig:
    """Configuration for TinyFlux-Deep model."""
    hidden_size: int = 512
    num_attention_heads: int = 4
    attention_head_dim: int = 128

    in_channels: int = 16
    patch_size: int = 1

    joint_attention_dim: int = 768
    pooled_projection_dim: int = 768

    num_double_layers: int = 15
    num_single_layers: int = 25

    mlp_ratio: float = 4.0
    axes_dims_rope: Tuple[int, int, int] = (16, 56, 56)
    
    # Expert predictor config
    use_expert_predictor: bool = True
    expert_dim: int = 1280  # SD1.5 mid-block dimension
    expert_hidden_dim: int = 512
    expert_dropout: float = 0.1  # Dropout during training for robustness
    
    # Legacy guidance (disabled when using expert)
    guidance_embeds: bool = False

    def __post_init__(self):
        assert self.num_attention_heads * self.attention_head_dim == self.hidden_size
        assert sum(self.axes_dims_rope) == self.attention_head_dim


# =============================================================================
# Normalization
# =============================================================================

class RMSNorm(nn.Module):
    """Root Mean Square Layer Normalization."""

    def __init__(self, dim: int, eps: float = 1e-6, elementwise_affine: bool = True):
        super().__init__()
        self.eps = eps
        self.elementwise_affine = elementwise_affine
        if elementwise_affine:
            self.weight = nn.Parameter(torch.ones(dim))
        else:
            self.register_parameter('weight', None)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        norm = x.float().pow(2).mean(-1, keepdim=True).add(self.eps).rsqrt()
        out = (x * norm).type_as(x)
        if self.weight is not None:
            out = out * self.weight
        return out


# =============================================================================
# RoPE - Old format with cached frequency buffers
# =============================================================================

class EmbedND(nn.Module):
    """Original TinyFlux RoPE with cached frequency buffers."""

    def __init__(self, theta: float = 10000.0, axes_dim: Tuple[int, int, int] = (16, 56, 56)):
        super().__init__()
        self.theta = theta
        self.axes_dim = axes_dim
        
        for i, dim in enumerate(axes_dim):
            freqs = 1.0 / (theta ** (torch.arange(0, dim, 2).float() / dim))
            self.register_buffer(f'freqs_{i}', freqs, persistent=True)

    def forward(self, ids: torch.Tensor) -> torch.Tensor:
        device = ids.device
        n_axes = ids.shape[-1]
        emb_list = []

        for i in range(n_axes):
            freqs = getattr(self, f'freqs_{i}').to(device)
            pos = ids[:, i].float()
            angles = pos.unsqueeze(-1) * freqs.unsqueeze(0)
            cos = angles.cos()
            sin = angles.sin()
            emb = torch.stack([cos, sin], dim=-1).flatten(-2)
            emb_list.append(emb)

        rope = torch.cat(emb_list, dim=-1)
        return rope.unsqueeze(1)


def apply_rotary_emb_old(x: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor:
    """Apply rotary embeddings (old interleaved format)."""
    freqs = freqs_cis.squeeze(1)
    cos = freqs[:, 0::2].repeat_interleave(2, dim=-1)
    sin = freqs[:, 1::2].repeat_interleave(2, dim=-1)
    cos = cos[None, None, :, :].to(x.device)
    sin = sin[None, None, :, :].to(x.device)
    x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1)
    x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(-2)
    return (x.float() * cos + x_rotated.float() * sin).to(x.dtype)


# =============================================================================
# Embeddings
# =============================================================================

class MLPEmbedder(nn.Module):
    """MLP for embedding scalars (timestep)."""

    def __init__(self, hidden_size: int):
        super().__init__()
        self.mlp = nn.Sequential(
            nn.Linear(256, hidden_size),
            nn.SiLU(),
            nn.Linear(hidden_size, hidden_size),
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        half_dim = 128
        emb = math.log(10000) / (half_dim - 1)
        emb = torch.exp(torch.arange(half_dim, device=x.device, dtype=x.dtype) * -emb)
        emb = x.unsqueeze(-1) * emb.unsqueeze(0)
        emb = torch.cat([emb.sin(), emb.cos()], dim=-1)
        return self.mlp(emb)


# =============================================================================
# Expert Predictor
# =============================================================================

class ExpertPredictor(nn.Module):
    """
    Predicts SD1.5-flow expert features from (timestep_emb, CLIP_pooled).
    
    Training: learns to match real expert features via distillation loss.
    Inference: runs standalone, no expert model needed.
    
    The predictor learns:
    - What the expert "sees" at each timestep
    - How text conditioning modulates that view
    - Trajectory shape priors from the expert's knowledge
    """

    def __init__(
        self,
        time_dim: int = 512,
        clip_dim: int = 768,
        expert_dim: int = 1280,
        hidden_dim: int = 512,
        output_dim: int = 512,
        dropout: float = 0.1,
    ):
        super().__init__()
        
        self.expert_dim = expert_dim
        self.dropout = dropout
        
        # Input fusion
        self.input_proj = nn.Linear(time_dim + clip_dim, hidden_dim)
        
        # Predictor core - learns expert behavior
        self.predictor = nn.Sequential(
            nn.SiLU(),
            nn.Linear(hidden_dim, hidden_dim),
            nn.SiLU(),
            nn.Dropout(dropout),
            nn.Linear(hidden_dim, hidden_dim),
            nn.SiLU(),
            nn.Linear(hidden_dim, expert_dim),
        )
        
        # Project predicted expert features to vec dimension
        self.output_proj = nn.Sequential(
            nn.LayerNorm(expert_dim),
            nn.Linear(expert_dim, output_dim),
        )
        
        # Learnable gate for expert influence
        self.expert_gate = nn.Parameter(torch.ones(1) * 0.5)
        
        self._init_weights()
    
    def _init_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Linear):
                nn.init.xavier_uniform_(m.weight, gain=0.5)
                if m.bias is not None:
                    nn.init.zeros_(m.bias)
    
    def forward(
        self,
        time_emb: torch.Tensor,
        clip_pooled: torch.Tensor,
        real_expert_features: Optional[torch.Tensor] = None,
        force_predictor: bool = False,
    ) -> Dict[str, torch.Tensor]:
        """
        Forward pass.
        
        Args:
            time_emb: [B, time_dim] - timestep embedding from time_in
            clip_pooled: [B, clip_dim] - pooled CLIP features
            real_expert_features: [B, expert_dim] - real expert output (training only)
            force_predictor: if True, use predictor even when real features available
        
        Returns:
            dict with:
                - 'expert_signal': [B, output_dim] - signal to add to vec
                - 'expert_pred': [B, expert_dim] - predicted expert features (for loss)
                - 'expert_used': str - 'real' or 'predicted'
        """
        B = time_emb.shape[0]
        device = time_emb.device
        
        # Fuse inputs
        combined = torch.cat([time_emb, clip_pooled], dim=-1)
        hidden = self.input_proj(combined)
        
        # Predict expert features
        expert_pred = self.predictor(hidden)
        
        # Decide which features to use
        use_real = (
            real_expert_features is not None 
            and self.training 
            and not force_predictor
            and torch.rand(1).item() > self.dropout  # Sometimes use predictor even in training
        )
        
        if use_real:
            expert_features = real_expert_features
            expert_used = 'real'
        else:
            expert_features = expert_pred
            expert_used = 'predicted'
        
        # Project to output dimension with gating
        gate = torch.sigmoid(self.expert_gate)
        expert_signal = gate * self.output_proj(expert_features)
        
        return {
            'expert_signal': expert_signal,
            'expert_pred': expert_pred,
            'expert_used': expert_used,
        }
    
    def compute_distillation_loss(
        self,
        expert_pred: torch.Tensor,
        real_expert_features: torch.Tensor,
    ) -> torch.Tensor:
        """MSE loss between predicted and real expert features."""
        return F.mse_loss(expert_pred, real_expert_features)


# =============================================================================
# AdaLayerNorm
# =============================================================================

class AdaLayerNormZero(nn.Module):
    """AdaLN-Zero for double-stream blocks (6 params)."""

    def __init__(self, hidden_size: int):
        super().__init__()
        self.silu = nn.SiLU()
        self.linear = nn.Linear(hidden_size, 6 * hidden_size, bias=True)
        self.norm = RMSNorm(hidden_size)

    def forward(self, x: torch.Tensor, emb: torch.Tensor):
        emb_out = self.linear(self.silu(emb))
        shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = emb_out.chunk(6, dim=-1)
        x = self.norm(x) * (1 + scale_msa.unsqueeze(1)) + shift_msa.unsqueeze(1)
        return x, gate_msa, shift_mlp, scale_mlp, gate_mlp


class AdaLayerNormZeroSingle(nn.Module):
    """AdaLN-Zero for single-stream blocks (3 params)."""

    def __init__(self, hidden_size: int):
        super().__init__()
        self.silu = nn.SiLU()
        self.linear = nn.Linear(hidden_size, 3 * hidden_size, bias=True)
        self.norm = RMSNorm(hidden_size)

    def forward(self, x: torch.Tensor, emb: torch.Tensor):
        emb_out = self.linear(self.silu(emb))
        shift, scale, gate = emb_out.chunk(3, dim=-1)
        x = self.norm(x) * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1)
        return x, gate


# =============================================================================
# Attention
# =============================================================================

class Attention(nn.Module):
    """Multi-head attention."""

    def __init__(self, hidden_size: int, num_heads: int, head_dim: int, use_bias: bool = False):
        super().__init__()
        self.num_heads = num_heads
        self.head_dim = head_dim
        self.scale = head_dim ** -0.5

        self.qkv = nn.Linear(hidden_size, 3 * num_heads * head_dim, bias=use_bias)
        self.out_proj = nn.Linear(num_heads * head_dim, hidden_size, bias=use_bias)

    def forward(self, x: torch.Tensor, rope: Optional[torch.Tensor] = None) -> torch.Tensor:
        B, N, _ = x.shape
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, self.head_dim)
        q, k, v = qkv.permute(2, 0, 3, 1, 4)

        if rope is not None:
            q = apply_rotary_emb_old(q, rope)
            k = apply_rotary_emb_old(k, rope)

        attn = F.scaled_dot_product_attention(q, k, v)
        out = attn.transpose(1, 2).reshape(B, N, -1)
        return self.out_proj(out)


class JointAttention(nn.Module):
    """Joint attention for double-stream blocks."""

    def __init__(self, hidden_size: int, num_heads: int, head_dim: int, use_bias: bool = False):
        super().__init__()
        self.num_heads = num_heads
        self.head_dim = head_dim
        self.scale = head_dim ** -0.5

        self.txt_qkv = nn.Linear(hidden_size, 3 * num_heads * head_dim, bias=use_bias)
        self.img_qkv = nn.Linear(hidden_size, 3 * num_heads * head_dim, bias=use_bias)

        self.txt_out = nn.Linear(num_heads * head_dim, hidden_size, bias=use_bias)
        self.img_out = nn.Linear(num_heads * head_dim, hidden_size, bias=use_bias)

    def forward(
        self,
        txt: torch.Tensor,
        img: torch.Tensor,
        rope: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        B, L, _ = txt.shape
        _, N, _ = img.shape

        txt_qkv = self.txt_qkv(txt).reshape(B, L, 3, self.num_heads, self.head_dim)
        img_qkv = self.img_qkv(img).reshape(B, N, 3, self.num_heads, self.head_dim)

        txt_q, txt_k, txt_v = txt_qkv.permute(2, 0, 3, 1, 4)
        img_q, img_k, img_v = img_qkv.permute(2, 0, 3, 1, 4)

        if rope is not None:
            img_q = apply_rotary_emb_old(img_q, rope)
            img_k = apply_rotary_emb_old(img_k, rope)

        k = torch.cat([txt_k, img_k], dim=2)
        v = torch.cat([txt_v, img_v], dim=2)

        txt_out = F.scaled_dot_product_attention(txt_q, k, v)
        txt_out = txt_out.transpose(1, 2).reshape(B, L, -1)

        img_out = F.scaled_dot_product_attention(img_q, k, v)
        img_out = img_out.transpose(1, 2).reshape(B, N, -1)

        return self.txt_out(txt_out), self.img_out(img_out)


# =============================================================================
# MLP
# =============================================================================

class MLP(nn.Module):
    """Feed-forward network with GELU activation."""

    def __init__(self, hidden_size: int, mlp_ratio: float = 4.0):
        super().__init__()
        mlp_hidden = int(hidden_size * mlp_ratio)
        self.fc1 = nn.Linear(hidden_size, mlp_hidden, bias=True)
        self.act = nn.GELU(approximate='tanh')
        self.fc2 = nn.Linear(mlp_hidden, hidden_size, bias=True)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.fc2(self.act(self.fc1(x)))


# =============================================================================
# Transformer Blocks
# =============================================================================

class DoubleStreamBlock(nn.Module):
    """Double-stream transformer block."""

    def __init__(self, config: TinyFluxDeepConfig):
        super().__init__()
        hidden = config.hidden_size
        heads = config.num_attention_heads
        head_dim = config.attention_head_dim

        self.img_norm1 = AdaLayerNormZero(hidden)
        self.txt_norm1 = AdaLayerNormZero(hidden)
        self.attn = JointAttention(hidden, heads, head_dim, use_bias=False)
        self.img_norm2 = RMSNorm(hidden)
        self.txt_norm2 = RMSNorm(hidden)
        self.img_mlp = MLP(hidden, config.mlp_ratio)
        self.txt_mlp = MLP(hidden, config.mlp_ratio)

    def forward(
        self,
        txt: torch.Tensor,
        img: torch.Tensor,
        vec: torch.Tensor,
        rope: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        img_normed, img_gate_msa, img_shift_mlp, img_scale_mlp, img_gate_mlp = self.img_norm1(img, vec)
        txt_normed, txt_gate_msa, txt_shift_mlp, txt_scale_mlp, txt_gate_mlp = self.txt_norm1(txt, vec)

        txt_attn_out, img_attn_out = self.attn(txt_normed, img_normed, rope)

        txt = txt + txt_gate_msa.unsqueeze(1) * txt_attn_out
        img = img + img_gate_msa.unsqueeze(1) * img_attn_out

        txt_mlp_in = self.txt_norm2(txt) * (1 + txt_scale_mlp.unsqueeze(1)) + txt_shift_mlp.unsqueeze(1)
        img_mlp_in = self.img_norm2(img) * (1 + img_scale_mlp.unsqueeze(1)) + img_shift_mlp.unsqueeze(1)

        txt = txt + txt_gate_mlp.unsqueeze(1) * self.txt_mlp(txt_mlp_in)
        img = img + img_gate_mlp.unsqueeze(1) * self.img_mlp(img_mlp_in)

        return txt, img


class SingleStreamBlock(nn.Module):
    """Single-stream transformer block."""

    def __init__(self, config: TinyFluxDeepConfig):
        super().__init__()
        hidden = config.hidden_size
        heads = config.num_attention_heads
        head_dim = config.attention_head_dim

        self.norm = AdaLayerNormZeroSingle(hidden)
        self.attn = Attention(hidden, heads, head_dim, use_bias=False)
        self.mlp = MLP(hidden, config.mlp_ratio)
        self.norm2 = RMSNorm(hidden)

    def forward(
        self,
        txt: torch.Tensor,
        img: torch.Tensor,
        vec: torch.Tensor,
        rope: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        L = txt.shape[1]
        x = torch.cat([txt, img], dim=1)
        x_normed, gate = self.norm(x, vec)
        x = x + gate.unsqueeze(1) * self.attn(x_normed, rope)
        x = x + self.mlp(self.norm2(x))
        txt, img = x.split([L, x.shape[1] - L], dim=1)
        return txt, img


# =============================================================================
# Main Model
# =============================================================================

class TinyFluxDeep(nn.Module):
    """
    TinyFlux-Deep with Expert Predictor.
    
    The expert predictor learns to emulate SD1.5-flow's timestep expertise,
    allowing the model to benefit from trajectory priors without requiring
    the expert model at inference time.
    """

    def __init__(self, config: Optional[TinyFluxDeepConfig] = None):
        super().__init__()
        self.config = config or TinyFluxDeepConfig()
        cfg = self.config

        # Input projections
        self.img_in = nn.Linear(cfg.in_channels, cfg.hidden_size, bias=True)
        self.txt_in = nn.Linear(cfg.joint_attention_dim, cfg.hidden_size, bias=True)

        # Conditioning
        self.time_in = MLPEmbedder(cfg.hidden_size)
        self.vector_in = nn.Sequential(
            nn.SiLU(),
            nn.Linear(cfg.pooled_projection_dim, cfg.hidden_size, bias=True)
        )
        
        # Expert predictor (replaces guidance_in)
        if cfg.use_expert_predictor:
            self.expert_predictor = ExpertPredictor(
                time_dim=cfg.hidden_size,
                clip_dim=cfg.pooled_projection_dim,
                expert_dim=cfg.expert_dim,
                hidden_dim=cfg.expert_hidden_dim,
                output_dim=cfg.hidden_size,
                dropout=cfg.expert_dropout,
            )
        else:
            self.expert_predictor = None
        
        # Legacy guidance (for backward compat / comparison)
        if cfg.guidance_embeds:
            self.guidance_in = MLPEmbedder(cfg.hidden_size)
        else:
            self.guidance_in = None

        # RoPE
        self.rope = EmbedND(theta=10000.0, axes_dim=cfg.axes_dims_rope)

        # Transformer blocks
        self.double_blocks = nn.ModuleList([
            DoubleStreamBlock(cfg) for _ in range(cfg.num_double_layers)
        ])
        self.single_blocks = nn.ModuleList([
            SingleStreamBlock(cfg) for _ in range(cfg.num_single_layers)
        ])

        # Output
        self.final_norm = RMSNorm(cfg.hidden_size)
        self.final_linear = nn.Linear(cfg.hidden_size, cfg.in_channels, bias=True)

        self._init_weights()

    def _init_weights(self):
        def _init(module):
            if isinstance(module, nn.Linear):
                nn.init.xavier_uniform_(module.weight)
                if module.bias is not None:
                    nn.init.zeros_(module.bias)
        self.apply(_init)
        nn.init.zeros_(self.final_linear.weight)

    def forward(
        self,
        hidden_states: torch.Tensor,
        encoder_hidden_states: torch.Tensor,
        pooled_projections: torch.Tensor,
        timestep: torch.Tensor,
        img_ids: torch.Tensor,
        txt_ids: Optional[torch.Tensor] = None,
        guidance: Optional[torch.Tensor] = None,
        expert_features: Optional[torch.Tensor] = None,
        return_expert_pred: bool = False,
    ) -> torch.Tensor:
        """
        Forward pass.
        
        Args:
            hidden_states: [B, N, C] - image latents
            encoder_hidden_states: [B, L, D] - T5 text embeddings
            pooled_projections: [B, D] - CLIP pooled features
            timestep: [B] - diffusion timestep
            img_ids: [N, 3] or [B, N, 3] - image position IDs
            txt_ids: [L, 3] or [B, L, 3] - text position IDs (optional)
            guidance: [B] - legacy guidance scale (if guidance_embeds=True)
            expert_features: [B, 1280] - real expert features (training only)
            return_expert_pred: if True, return (output, expert_info) tuple
        
        Returns:
            output: [B, N, C] - predicted velocity
            expert_info: dict (if return_expert_pred=True)
        """
        B = hidden_states.shape[0]
        L = encoder_hidden_states.shape[1]
        N = hidden_states.shape[1]

        # Input projections
        img = self.img_in(hidden_states)
        txt = self.txt_in(encoder_hidden_states)

        # Conditioning: time + pooled text
        time_emb = self.time_in(timestep)
        vec = time_emb + self.vector_in(pooled_projections)
        
        # Expert predictor (third stream)
        expert_info = None
        if self.expert_predictor is not None:
            expert_out = self.expert_predictor(
                time_emb=time_emb,
                clip_pooled=pooled_projections,
                real_expert_features=expert_features,
            )
            vec = vec + expert_out['expert_signal']
            expert_info = expert_out
        
        # Legacy guidance (fallback)
        elif self.guidance_in is not None and guidance is not None:
            vec = vec + self.guidance_in(guidance)

        # Handle img_ids shape
        if img_ids.ndim == 3:
            img_ids = img_ids[0]
        img_rope = self.rope(img_ids)

        # Double-stream blocks
        for block in self.double_blocks:
            txt, img = block(txt, img, vec, img_rope)

        # Build full sequence RoPE for single-stream
        if txt_ids is None:
            txt_ids = torch.zeros(L, 3, device=img_ids.device, dtype=img_ids.dtype)
        elif txt_ids.ndim == 3:
            txt_ids = txt_ids[0]

        all_ids = torch.cat([txt_ids, img_ids], dim=0)
        full_rope = self.rope(all_ids)

        # Single-stream blocks
        for block in self.single_blocks:
            txt, img = block(txt, img, vec, full_rope)

        # Output
        img = self.final_norm(img)
        output = self.final_linear(img)

        if return_expert_pred:
            return output, expert_info
        return output

    def compute_loss(
        self,
        output: torch.Tensor,
        target: torch.Tensor,
        expert_pred: Optional[torch.Tensor] = None,
        real_expert_features: Optional[torch.Tensor] = None,
        distill_weight: float = 0.1,
    ) -> Dict[str, torch.Tensor]:
        """
        Compute combined loss.
        
        Args:
            output: model prediction
            target: flow matching target (data - noise)
            expert_pred: predicted expert features
            real_expert_features: real expert features
            distill_weight: weight for distillation loss
        
        Returns:
            dict with 'total', 'main', 'distill' losses
        """
        # Main flow matching loss
        main_loss = F.mse_loss(output, target)
        
        losses = {
            'main': main_loss,
            'distill': torch.tensor(0.0, device=output.device),
            'total': main_loss,
        }
        
        # Distillation loss
        if expert_pred is not None and real_expert_features is not None:
            distill_loss = self.expert_predictor.compute_distillation_loss(
                expert_pred, real_expert_features
            )
            losses['distill'] = distill_loss
            losses['total'] = main_loss + distill_weight * distill_loss
        
        return losses

    @staticmethod
    def create_img_ids(batch_size: int, height: int, width: int, device: torch.device) -> torch.Tensor:
        """Create image position IDs for RoPE."""
        img_ids = torch.zeros(height * width, 3, device=device)
        for i in range(height):
            for j in range(width):
                idx = i * width + j
                img_ids[idx, 0] = 0
                img_ids[idx, 1] = i
                img_ids[idx, 2] = j
        return img_ids

    @staticmethod
    def create_txt_ids(text_len: int, device: torch.device) -> torch.Tensor:
        """Create text position IDs."""
        txt_ids = torch.zeros(text_len, 3, device=device)
        txt_ids[:, 0] = torch.arange(text_len, device=device)
        return txt_ids

    def count_parameters(self) -> Dict[str, int]:
        """Count parameters by component."""
        counts = {}
        counts['img_in'] = sum(p.numel() for p in self.img_in.parameters())
        counts['txt_in'] = sum(p.numel() for p in self.txt_in.parameters())
        counts['time_in'] = sum(p.numel() for p in self.time_in.parameters())
        counts['vector_in'] = sum(p.numel() for p in self.vector_in.parameters())
        
        if self.expert_predictor is not None:
            counts['expert_predictor'] = sum(p.numel() for p in self.expert_predictor.parameters())
        if self.guidance_in is not None:
            counts['guidance_in'] = sum(p.numel() for p in self.guidance_in.parameters())
            
        counts['double_blocks'] = sum(p.numel() for p in self.double_blocks.parameters())
        counts['single_blocks'] = sum(p.numel() for p in self.single_blocks.parameters())
        counts['final'] = sum(p.numel() for p in self.final_norm.parameters()) + \
                          sum(p.numel() for p in self.final_linear.parameters())
        counts['total'] = sum(p.numel() for p in self.parameters())
        return counts


# =============================================================================
# Test
# =============================================================================

def test_model():
    """Test TinyFlux-Deep with Expert Predictor."""
    print("=" * 60)
    print("TinyFlux-Deep + Expert Predictor Test")
    print("=" * 60)

    config = TinyFluxDeepConfig(
        use_expert_predictor=True,
        expert_dim=1280,
        expert_hidden_dim=512,
        guidance_embeds=False,
    )
    model = TinyFluxDeep(config)

    counts = model.count_parameters()
    print(f"\nConfig:")
    print(f"  hidden_size: {config.hidden_size}")
    print(f"  num_double_layers: {config.num_double_layers}")
    print(f"  num_single_layers: {config.num_single_layers}")
    print(f"  expert_dim: {config.expert_dim}")
    print(f"  use_expert_predictor: {config.use_expert_predictor}")
    
    print(f"\nParameters:")
    for name, count in counts.items():
        print(f"  {name}: {count:,}")

    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    model = model.to(device)

    B, H, W = 2, 64, 64
    L = 77

    hidden_states = torch.randn(B, H * W, config.in_channels, device=device)
    encoder_hidden_states = torch.randn(B, L, config.joint_attention_dim, device=device)
    pooled_projections = torch.randn(B, config.pooled_projection_dim, device=device)
    timestep = torch.rand(B, device=device)
    img_ids = TinyFluxDeep.create_img_ids(B, H, W, device)
    txt_ids = TinyFluxDeep.create_txt_ids(L, device)
    
    # Simulated expert features
    expert_features = torch.randn(B, config.expert_dim, device=device)

    print("\n[Test 1: Training mode with expert features]")
    model.train()
    with torch.no_grad():
        output, expert_info = model(
            hidden_states=hidden_states,
            encoder_hidden_states=encoder_hidden_states,
            pooled_projections=pooled_projections,
            timestep=timestep,
            img_ids=img_ids,
            txt_ids=txt_ids,
            expert_features=expert_features,
            return_expert_pred=True,
        )
    print(f"  Output shape: {output.shape}")
    print(f"  Expert used: {expert_info['expert_used']}")
    print(f"  Expert pred shape: {expert_info['expert_pred'].shape}")

    print("\n[Test 2: Inference mode (no expert)]")
    model.eval()
    with torch.no_grad():
        output = model(
            hidden_states=hidden_states,
            encoder_hidden_states=encoder_hidden_states,
            pooled_projections=pooled_projections,
            timestep=timestep,
            img_ids=img_ids,
            txt_ids=txt_ids,
            expert_features=None,  # No expert at inference
        )
    print(f"  Output shape: {output.shape}")
    print(f"  Output range: [{output.min():.4f}, {output.max():.4f}]")

    print("\n[Test 3: Loss computation]")
    target = torch.randn_like(output)
    model.train()
    output, expert_info = model(
        hidden_states=hidden_states,
        encoder_hidden_states=encoder_hidden_states,
        pooled_projections=pooled_projections,
        timestep=timestep,
        img_ids=img_ids,
        txt_ids=txt_ids,
        expert_features=expert_features,
        return_expert_pred=True,
    )
    losses = model.compute_loss(
        output=output,
        target=target,
        expert_pred=expert_info['expert_pred'],
        real_expert_features=expert_features,
        distill_weight=0.1,
    )
    print(f"  Main loss: {losses['main']:.4f}")
    print(f"  Distill loss: {losses['distill']:.4f}")
    print(f"  Total loss: {losses['total']:.4f}")

    print("\n" + "=" * 60)
    print("✓ All tests passed!")
    print("=" * 60)


if __name__ == "__main__":
    test_model()