File size: 16,454 Bytes
cf812a0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from einops import rearrange, repeat
import torch
import torch.nn as nn
from ..wanvideo.modules.attention import attention

def timestep_transform(
    t,
    shift=5.0,
    num_timesteps=1000,
):
    t = t / num_timesteps
    # shift the timestep based on ratio
    new_t = shift * t / (1 + (shift - 1) * t)
    new_t = new_t * num_timesteps
    return new_t

def add_noise(
    original_samples: torch.FloatTensor,
    noise: torch.FloatTensor,
    timesteps: torch.IntTensor,
) -> torch.FloatTensor:
    """
    compatible with diffusers add_noise()
    """
    timesteps = timesteps.float() / 1000
    timesteps = timesteps.view(timesteps.shape + (1,) * (len(noise.shape)-1))

    return (1 - timesteps) * original_samples + timesteps * noise

def normalize_and_scale(column, source_range, target_range, epsilon=1e-8):

    source_min, source_max = source_range
    new_min, new_max = target_range
 
    normalized = (column - source_min) / (source_max - source_min + epsilon)
    scaled = normalized * (new_max - new_min) + new_min
    return scaled

def rotate_half(x):
    x = rearrange(x, "... (d r) -> ... d r", r=2)
    x1, x2 = x.unbind(dim=-1)
    x = torch.stack((-x2, x1), dim=-1)
    return rearrange(x, "... d r -> ... (d r)")

def calculate_x_ref_attn_map(visual_q, ref_k, ref_target_masks, mode='mean', attn_bias=None):
    
    ref_k = ref_k.to(visual_q.dtype).to(visual_q.device)
    scale = 1.0 / visual_q.shape[-1] ** 0.5
    visual_q = visual_q * scale
    visual_q = visual_q.transpose(1, 2)
    ref_k = ref_k.transpose(1, 2)
    attn = visual_q @ ref_k.transpose(-2, -1)

    if attn_bias is not None:
        attn = attn + attn_bias

    x_ref_attn_map_source = attn.softmax(-1) # B, H, x_seqlens, ref_seqlens


    x_ref_attn_maps = []
    ref_target_masks = ref_target_masks.to(visual_q.dtype)
    x_ref_attn_map_source = x_ref_attn_map_source.to(visual_q.dtype)

    for class_idx, ref_target_mask in enumerate(ref_target_masks):
        ref_target_mask = ref_target_mask[None, None, None, ...]
        x_ref_attnmap = x_ref_attn_map_source * ref_target_mask
        x_ref_attnmap = x_ref_attnmap.sum(-1) / ref_target_mask.sum() # B, H, x_seqlens, ref_seqlens --> B, H, x_seqlens
        x_ref_attnmap = x_ref_attnmap.permute(0, 2, 1) # B, x_seqlens, H
       
        if mode == 'mean':
            x_ref_attnmap = x_ref_attnmap.mean(-1) # B, x_seqlens
        elif mode == 'max':
            x_ref_attnmap = x_ref_attnmap.max(-1) # B, x_seqlens
        
        x_ref_attn_maps.append(x_ref_attnmap)
    
    del attn, x_ref_attn_map_source

    return torch.concat(x_ref_attn_maps, dim=0)

def get_attn_map_with_target(visual_q, ref_k, shape, ref_target_masks=None, split_num=2):
    """Args:
        query (torch.tensor): B M H K
        key (torch.tensor): B M H K
        shape (tuple): (N_t, N_h, N_w)
        ref_target_masks: [B, N_h * N_w]
    """

    N_t, N_h, N_w = shape
    
    x_seqlens = N_h * N_w
    ref_k     = ref_k[:, :x_seqlens]
    _, seq_lens, heads, _ = visual_q.shape
    class_num, _ = ref_target_masks.shape
    x_ref_attn_maps = torch.zeros(class_num, seq_lens).to(visual_q.device).to(visual_q.dtype)

    split_chunk = heads // split_num
    
    for i in range(split_num):
        x_ref_attn_maps_perhead = calculate_x_ref_attn_map(visual_q[:, :, i*split_chunk:(i+1)*split_chunk, :], ref_k[:, :, i*split_chunk:(i+1)*split_chunk, :], ref_target_masks)
        x_ref_attn_maps += x_ref_attn_maps_perhead
    
    return x_ref_attn_maps / split_num

class RotaryPositionalEmbedding1D(nn.Module):

    def __init__(self,
                 head_dim,
                 ):
        super().__init__()
        self.head_dim = head_dim
        self.base = 10000

    def precompute_freqs_cis_1d(self, pos_indices):

        freqs = 1.0 / (self.base ** (torch.arange(0, self.head_dim, 2)[: (self.head_dim // 2)].float() / self.head_dim))
        freqs = freqs.to(pos_indices.device)
        freqs = torch.einsum("..., f -> ... f", pos_indices.float(), freqs)
        freqs = repeat(freqs, "... n -> ... (n r)", r=2)
        return freqs

    def forward(self, x, pos_indices):
        """1D RoPE.

        Args:
            query (torch.tensor): [B, head, seq, head_dim]
            pos_indices (torch.tensor): [seq,]
        Returns:
            query with the same shape as input.
        """
        freqs_cis = self.precompute_freqs_cis_1d(pos_indices)

        x_ = x.float()

        freqs_cis = freqs_cis.float().to(x.device)
        cos, sin = freqs_cis.cos(), freqs_cis.sin()
        cos, sin = rearrange(cos, 'n d -> 1 1 n d'), rearrange(sin, 'n d -> 1 1 n d')
        x_ = (x_ * cos) + (rotate_half(x_) * sin)

        return x_.type_as(x)

class AudioProjModel(nn.Module):
    def __init__(
        self,
        seq_len=5,
        seq_len_vf=12,
        blocks=12,  
        channels=768, 
        intermediate_dim=512,
        output_dim=768,
        context_tokens=32,
        norm_output_audio=False,
    ):
        super().__init__()

        self.seq_len = seq_len
        self.blocks = blocks
        self.channels = channels
        self.input_dim = seq_len * blocks * channels  
        self.input_dim_vf = seq_len_vf * blocks * channels
        self.intermediate_dim = intermediate_dim
        self.context_tokens = context_tokens
        self.output_dim = output_dim

        # define multiple linear layers
        self.proj1 = nn.Linear(self.input_dim, intermediate_dim)
        self.proj1_vf = nn.Linear(self.input_dim_vf, intermediate_dim)
        self.proj2 = nn.Linear(intermediate_dim, intermediate_dim)
        self.proj3 = nn.Linear(intermediate_dim, context_tokens * output_dim)
        self.norm = nn.LayerNorm(output_dim) if norm_output_audio else nn.Identity()

    def forward(self, audio_embeds, audio_embeds_vf):
        video_length = audio_embeds.shape[1] + audio_embeds_vf.shape[1]
        B, _, _, S, C = audio_embeds.shape

        # process audio of first frame
        audio_embeds = rearrange(audio_embeds, "bz f w b c -> (bz f) w b c")
        batch_size, window_size, blocks, channels = audio_embeds.shape
        audio_embeds = audio_embeds.view(batch_size, window_size * blocks * channels)

        # process audio of latter frame
        audio_embeds_vf = rearrange(audio_embeds_vf, "bz f w b c -> (bz f) w b c")
        batch_size_vf, window_size_vf, blocks_vf, channels_vf = audio_embeds_vf.shape
        audio_embeds_vf = audio_embeds_vf.view(batch_size_vf, window_size_vf * blocks_vf * channels_vf)

        # first projection
        audio_embeds = torch.relu(self.proj1(audio_embeds)) 
        audio_embeds_vf = torch.relu(self.proj1_vf(audio_embeds_vf)) 
        audio_embeds = rearrange(audio_embeds, "(bz f) c -> bz f c", bz=B)
        audio_embeds_vf = rearrange(audio_embeds_vf, "(bz f) c -> bz f c", bz=B)
        audio_embeds_c = torch.concat([audio_embeds, audio_embeds_vf], dim=1) 
        batch_size_c, N_t, C_a = audio_embeds_c.shape
        audio_embeds_c = audio_embeds_c.view(batch_size_c*N_t, C_a)

        # second projection
        audio_embeds_c = torch.relu(self.proj2(audio_embeds_c))

        context_tokens = self.proj3(audio_embeds_c).reshape(batch_size_c*N_t, self.context_tokens, self.output_dim)

        # normalization and reshape
        context_tokens = self.norm(context_tokens)
        context_tokens = rearrange(context_tokens, "(bz f) m c -> bz f m c", f=video_length)

        return context_tokens

#@torch.compiler.disable()
class SingleStreamAttention(nn.Module):
    def __init__(
        self,
        dim: int,
        encoder_hidden_states_dim: int,
        num_heads: int,
        qkv_bias: bool,
        attention_mode: str = 'sdpa',
    ) -> None:
        super().__init__()
        assert dim % num_heads == 0, "dim should be divisible by num_heads"
        self.dim = dim
        self.encoder_hidden_states_dim = encoder_hidden_states_dim
        self.num_heads = num_heads
        self.head_dim = dim // num_heads
        self.attention_mode = attention_mode

        self.q_linear = nn.Linear(dim, dim, bias=qkv_bias)
        self.proj = nn.Linear(dim, dim)
        self.kv_linear = nn.Linear(encoder_hidden_states_dim, dim * 2, bias=qkv_bias)

    def forward(self, x: torch.Tensor, encoder_hidden_states: torch.Tensor, shape=None) -> torch.Tensor:
        N_t, N_h, N_w = shape

        expected_tokens = N_t * N_h * N_w
        actual_tokens = x.shape[1]
        x_extra = None

        if actual_tokens != expected_tokens:
            x_extra = x[:, -N_h * N_w:, :]
            x = x[:, :-N_h * N_w, :]
            N_t = N_t - 1

        B = x.shape[0]
        S = N_h * N_w
        x = x.view(B * N_t, S, self.dim)

        # get q for hidden_state
        q = self.q_linear(x).view(B * N_t, S, self.num_heads, self.head_dim)
        
        # get kv from encoder_hidden_states # shape: (B, N, num_heads, head_dim)
        kv = self.kv_linear(encoder_hidden_states)
        encoder_k, encoder_v = kv.view(B * N_t, encoder_hidden_states.shape[1], 2, self.num_heads, self.head_dim).unbind(2)

        x = attention(q, encoder_k, encoder_v, attention_mode=self.attention_mode)

        # linear transform
        x = self.proj(x.reshape(B * N_t, S, self.dim))
        x = x.view(B, N_t * S, self.dim)
    
        if x_extra is not None:
            x = torch.cat([x, torch.zeros_like(x_extra)], dim=1)

        return x

    
class SingleStreamMultiAttention(SingleStreamAttention):
    """Multi-speaker rotary-position cross-attention.

    This implementation generalises the original 2-speaker logic to an arbitrary
    number of voices.  Each speaker is allocated a contiguous *class_interval*
    segment inside a shared *class_range* rotary bucket.  The centre of each
    bucket is applied to that speaker's KV tokens while queries are modulated
    per-token according to which speaker dominates the pixel.
    """

    def __init__(
        self,
        dim: int,
        encoder_hidden_states_dim: int,
        num_heads: int,
        qkv_bias: bool,
        class_range: int = 24,
        class_interval: int = 4,
        attention_mode: str = 'sdpa',
    ) -> None:
        super().__init__(
            dim=dim,
            encoder_hidden_states_dim=encoder_hidden_states_dim,
            num_heads=num_heads,
            qkv_bias=qkv_bias,
            attention_mode=attention_mode,
        )

        # Rotary-embedding layout parameters
        self.class_interval = class_interval
        self.class_range = class_range
        self.max_humans = self.class_range // self.class_interval

        # Constant bucket used for background tokens
        self.rope_bak = int(self.class_range // 2)

        self.rope_1d = RotaryPositionalEmbedding1D(self.head_dim)

        self.attention_mode = attention_mode

    def forward(
        self,
        x: torch.Tensor,
        encoder_hidden_states: torch.Tensor,
        shape=None,
        x_ref_attn_map=None,
        human_num=None,
    ) -> torch.Tensor:
        encoder_hidden_states = encoder_hidden_states.squeeze(0)

        # Single-speaker fall-through
        if human_num is None or human_num <= 1:
            return super().forward(x, encoder_hidden_states, shape)

        N_t, N_h, N_w = shape
        
        x_extra = None
        if x.shape[0] * N_t != encoder_hidden_states.shape[0]:
            x_extra = x[:, -N_h * N_w:, :]
            x = x[:, :-N_h * N_w, :]
            N_t = N_t - 1
        x = rearrange(x, "B (N_t S) C -> (B N_t) S C", N_t=N_t)

        # Query projection
        B, N, C = x.shape
        q = self.q_linear(x)
        q = q.view(B, N, self.num_heads, self.head_dim).permute(0, 2, 1, 3)

        if human_num == 2:
            # Use `class_range` logic for exactly 2 speakers
            rope_h1 = (0, self.class_interval)
            rope_h2 = (self.class_range - self.class_interval, self.class_range)
            rope_bak = int(self.class_range // 2)

            # Normalize and scale attention maps for each speaker
            max_values = x_ref_attn_map.max(1).values[:, None, None]
            min_values = x_ref_attn_map.min(1).values[:, None, None]
            max_min_values = torch.cat([max_values, min_values], dim=2)

            human1_max_value, human1_min_value = max_min_values[0, :, 0].max(), max_min_values[0, :, 1].min()
            human2_max_value, human2_min_value = max_min_values[1, :, 0].max(), max_min_values[1, :, 1].min()

            human1 = normalize_and_scale(x_ref_attn_map[0], (human1_min_value, human1_max_value), rope_h1)
            human2 = normalize_and_scale(x_ref_attn_map[1], (human2_min_value, human2_max_value), rope_h2)
            back = torch.full((x_ref_attn_map.size(1),), rope_bak, dtype=human1.dtype, device=human1.device)

            # Token-wise speaker dominance
            max_indices = x_ref_attn_map.argmax(dim=0)
            normalized_map = torch.stack([human1, human2, back], dim=1)
            normalized_pos = normalized_map[torch.arange(x_ref_attn_map.size(1)), max_indices]
        else:
            # General case for more than 2 speakers
            rope_ranges = [
                (i * self.class_interval, (i + 1) * self.class_interval)
                for i in range(human_num)
            ]

            # Normalize each speaker's attention map into its own bucket
            human_norm_list = []
            for idx in range(human_num):
                attn_map = x_ref_attn_map[idx]
                att_min, att_max = attn_map.min(), attn_map.max()
                human_norm = normalize_and_scale(
                    attn_map, (att_min, att_max), rope_ranges[idx]
                )
                human_norm_list.append(human_norm)

            # Background constant bucket
            back = torch.full(
                (x_ref_attn_map.size(1),),
                self.rope_bak,
                dtype=x_ref_attn_map.dtype,
                device=x_ref_attn_map.device,
            )

            # Token-wise speaker dominance
            max_indices = x_ref_attn_map.argmax(dim=0)
            normalized_map = torch.stack(human_norm_list + [back], dim=1)
            normalized_pos = normalized_map[torch.arange(x_ref_attn_map.size(1)), max_indices]

        # Apply rotary to Q
        q = rearrange(q, "(B N_t) H S C -> B H (N_t S) C", N_t=N_t)
        q = self.rope_1d(q, normalized_pos)
        q = rearrange(q, "B H (N_t S) C -> (B N_t) H S C", N_t=N_t)

        # Keys / Values
        _, N_a, _ = encoder_hidden_states.shape
        encoder_kv = self.kv_linear(encoder_hidden_states)
        encoder_kv = encoder_kv.view(B, N_a, 2, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
        encoder_k, encoder_v = encoder_kv.unbind(0)

        # Rotary for keys – assign centre of each speaker bucket to its context tokens
        if human_num == 2:
            per_frame = torch.zeros(N_a, dtype=encoder_k.dtype, device=encoder_k.device)
            per_frame[: per_frame.size(0) // 2] = (rope_h1[0] + rope_h1[1]) / 2
            per_frame[per_frame.size(0) // 2 :] = (rope_h2[0] + rope_h2[1]) / 2
            encoder_pos = torch.cat([per_frame] * N_t, dim=0)
        else:
            tokens_per_human = N_a // human_num
            encoder_pos_list = []
            for i in range(human_num):
                start, end = rope_ranges[i]
                centre = (start + end) / 2
                encoder_pos_list.append(
                    torch.full(
                        (tokens_per_human,), centre, dtype=encoder_k.dtype, device=encoder_k.device
                    )
                )
            encoder_pos = torch.cat(encoder_pos_list * N_t, dim=0)

        encoder_k = rearrange(encoder_k, "(B N_t) H S C -> B H (N_t S) C", N_t=N_t)
        encoder_k = self.rope_1d(encoder_k, encoder_pos)
        encoder_k = rearrange(encoder_k, "B H (N_t S) C -> (B N_t) H S C", N_t=N_t)

        # Final attention
        q = rearrange(q, "B H M K -> B M H K")
        encoder_k = rearrange(encoder_k, "B H M K -> B M H K")
        encoder_v = rearrange(encoder_v, "B H M K -> B M H K")
        x = attention(
            q, encoder_k, encoder_v, attention_mode=self.attention_mode
        )

        # Linear projection
        x = x.reshape(B, N, C)
        x = self.proj(x)

        # Restore original layout
        x = rearrange(x, "(B N_t) S C -> B (N_t S) C", N_t=N_t)
        if x_extra is not None:
            x = torch.cat([x, torch.zeros_like(x_extra)], dim=1)

        return x