File size: 23,326 Bytes
ccfee12
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# MIT License
#
# Adapted from the official implementation of PRoPE
# "Cameras as Relative Positional Encoding" https://arxiv.org/pdf/2507.10496
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

from functools import partial
from typing import Callable, Optional, Tuple, List

import torch
import torch.nn.functional as F

from einops import rearrange


class PropeDotProductAttention(torch.nn.Module):
    """PRoPE attention with precomputed RoPE coefficients."""

    coeffs_x_0: torch.Tensor
    coeffs_x_1: torch.Tensor
    coeffs_y_0: torch.Tensor
    coeffs_y_1: torch.Tensor

    def __init__(
        self,
        head_dim: int,
        patches_x: int,
        patches_y: int,
        image_width: int,
        image_height: int,
        freq_base: float = 10000.0, # qwen 10000
        freq_scale: float = 1.0,
        dim_arrange = [16, 56, 56], # (frame ,height, width) default for qwen. 
        depth = None,
    ):
        super().__init__()
        self.head_dim = head_dim
        self.patches_x = patches_x
        self.patches_y = patches_y
        self.image_width = image_width
        self.image_height = image_height
        self.freq_base = freq_base
        self.freq_scale = freq_scale

        self.use_PRoPE = False
        self.dim_arrange = dim_arrange

        # fit Qwen scale-rope
        pos_index_x = torch.arange(patches_x)
        neg_index_x = torch.arange(patches_x).flip(0) * -1 - 1
        index_x = torch.cat([neg_index_x[-(patches_x - patches_x // 2) :], pos_index_x[: patches_x // 2]], dim=0)
        # print(index_x)
        ## Qwen rope apply order is frame, height, width!
        # coeffs_x
        coeffs_y: Tuple[torch.Tensor, torch.Tensor] = _rope_precompute_coeffs(  #
            # torch.tile(torch.arange(patches_x), (patches_y,)),
            torch.tile(index_x, (patches_y,)),
            freq_base=freq_base,
            freq_scale=freq_scale,
            # feat_dim=head_dim // 4,
            feat_dim=dim_arrange[2],
        )

        # fit Qwen scale-rope
        pos_index_y = torch.arange(patches_y)
        neg_index_y = torch.arange(patches_y).flip(0) * -1 - 1
        index_y = torch.cat([neg_index_y[-(patches_y - patches_y // 2) :], pos_index_y[: patches_y // 2]], dim=0)
        # print(index_y)
        ## Qwen rope apply order is frame, height, width!
        # coeffs_y
        coeffs_x: Tuple[torch.Tensor, torch.Tensor] = _rope_precompute_coeffs(
            # torch.repeat_interleave(torch.arange(patches_y), patches_x),
            torch.repeat_interleave(index_y, patches_x),
            freq_base=freq_base,
            freq_scale=freq_scale,
            # feat_dim=head_dim // 4,
            feat_dim=dim_arrange[1],
        )

        # Do not save coeffs to checkpoint as `cameras` might change during testing.
        self.register_buffer("coeffs_x_0", coeffs_x[0], persistent=False)
        self.register_buffer("coeffs_x_1", coeffs_x[1], persistent=False)
        self.register_buffer("coeffs_y_0", coeffs_y[0], persistent=False)
        self.register_buffer("coeffs_y_1", coeffs_y[1], persistent=False)




    # override load_state_dict to not load coeffs if they exist (for backward compatibility)
    def load_state_dict(self, state_dict, strict=True):
        # remove coeffs from state_dict
        state_dict.pop("coeffs_x_0", None)
        state_dict.pop("coeffs_x_1", None)
        state_dict.pop("coeffs_y_0", None)
        state_dict.pop("coeffs_y_1", None)
        super().load_state_dict(state_dict, strict)

    def forward(
        self,
        q: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
        k: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
        v: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
        viewmats: torch.Tensor,  # (batch, cameras, 4, 4)
        Ks: Optional[torch.Tensor],  # (batch, cameras, 3, 3)
        **kwargs,
    ) -> torch.Tensor:
        return prope_dot_product_attention(
            q,
            k,
            v,
            viewmats=viewmats,
            Ks=Ks,
            patches_x=self.patches_x,
            patches_y=self.patches_y,
            image_width=self.image_width,
            image_height=self.image_height,
            coeffs_x=(self.coeffs_x_0, self.coeffs_x_1),
            coeffs_y=(self.coeffs_y_0, self.coeffs_y_1),
            **kwargs,
        )

    def _precompute_and_cache_apply_fns(
        self, 
        viewmats: torch.Tensor, 
        Ks: Optional[torch.Tensor],
        depth = None,
    ):
        (batch, cameras, _, _) = viewmats.shape
        assert viewmats.shape == (batch, cameras, 4, 4)
        assert Ks is None or Ks.shape == (batch, cameras, 3, 3)
        self.cameras = cameras
        self.use_PRoPE = True

        self.apply_fn_q, self.apply_fn_kv, self.apply_fn_o = _prepare_apply_fns(
            head_dim=self.head_dim,
            viewmats=viewmats,
            Ks=Ks,
            patches_x=self.patches_x,
            patches_y=self.patches_y,
            image_width=self.image_width,
            image_height=self.image_height,
            coeffs_x=(self.coeffs_x_0, self.coeffs_x_1),
            coeffs_y=(self.coeffs_y_0, self.coeffs_y_1),
            dim_arrange=self.dim_arrange,
            freq_base=self.freq_base,
            freq_scale=self.freq_scale,
            depth=depth
        )

    def _apply_to_q(self, q: torch.Tensor) -> torch.Tensor:
        (batch, num_heads, seqlen, head_dim) = q.shape
        # print("!!!", q.shape)
        # print(self.cameras, self.patches_x, self.patches_y)
        assert seqlen == self.cameras * self.patches_x * self.patches_y, f"seqlen:{seqlen}, {self.cameras}, {self.patches_x}, {self.patches_y}"
        assert head_dim == self.head_dim
        assert q.shape == (batch, num_heads, seqlen, head_dim)
        assert self.apply_fn_q is not None
        return self.apply_fn_q(q)

    def _apply_to_kv(self, kv: torch.Tensor) -> torch.Tensor:
        (batch, num_heads, seqlen, head_dim) = kv.shape
        assert seqlen == self.cameras * self.patches_x * self.patches_y, f"seqlen:{seqlen}, {self.cameras}, {self.patches_x}, {self.patches_y}"
        assert head_dim == self.head_dim
        assert kv.shape == (batch, num_heads, seqlen, head_dim)
        assert self.apply_fn_kv is not None
        return self.apply_fn_kv(kv)

    def _apply_to_o(self, o: torch.Tensor) -> torch.Tensor:
        (batch, num_heads, seqlen, head_dim) = o.shape
        assert seqlen == self.cameras * self.patches_x * self.patches_y
        assert head_dim == self.head_dim
        assert o.shape == (batch, num_heads, seqlen, head_dim)
        assert self.apply_fn_o is not None
        return self.apply_fn_o(o)


def prope_dot_product_attention(
    q: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
    k: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
    v: torch.Tensor,  # (batch, num_heads, seqlen, head_dim)
    *,
    viewmats: torch.Tensor,  # (batch, cameras, 4, 4)
    Ks: Optional[torch.Tensor],  # (batch, cameras, 3, 3)
    patches_x: int,  # How many patches wide is each image?
    patches_y: int,  # How many patches tall is each image?
    image_width: int,  # Width of the image. Used to normalize intrinsics.
    image_height: int,  # Height of the image. Used to normalize intrinsics.
    coeffs_x: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
    coeffs_y: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
    **kwargs,
) -> torch.Tensor:
    """Similar to torch.nn.functional.scaled_dot_product_attention, but applies PRoPE-style
    positional encoding.

    Currently, we assume that the sequence length is equal to:

        cameras * patches_x * patches_y

    And token ordering allows the `(seqlen,)` axis to be reshaped into
    `(cameras, patches_x, patches_y)`.
    """
    # We're going to assume self-attention: all inputs are the same shape.
    (batch, num_heads, seqlen, head_dim) = q.shape
    cameras = viewmats.shape[1]
    assert q.shape == k.shape == v.shape
    assert viewmats.shape == (batch, cameras, 4, 4)
    assert Ks is None or Ks.shape == (batch, cameras, 3, 3)
    assert seqlen == cameras * patches_x * patches_y

    apply_fn_q, apply_fn_kv, apply_fn_o = _prepare_apply_fns(
        head_dim=head_dim,
        viewmats=viewmats,
        Ks=Ks,
        patches_x=patches_x,
        patches_y=patches_y,
        image_width=image_width,
        image_height=image_height,
        coeffs_x=coeffs_x,
        coeffs_y=coeffs_y,
    )

    out = F.scaled_dot_product_attention(
        query=apply_fn_q(q),
        key=apply_fn_kv(k),
        value=apply_fn_kv(v),
        **kwargs,
    )
    out = apply_fn_o(out)
    assert out.shape == (batch, num_heads, seqlen, head_dim)
    return out


def _prepare_apply_fns(
    head_dim: int,  # Q/K/V will have this last dimension
    viewmats: torch.Tensor,  # (batch, cameras, 4, 4)
    Ks: Optional[torch.Tensor],  # (batch, cameras, 3, 3)
    patches_x: int,  # How many patches wide is each image?
    patches_y: int,  # How many patches tall is each image?
    image_width: int,  # Width of the image. Used to normalize intrinsics.
    image_height: int,  # Height of the image. Used to normalize intrinsics.
    coeffs_x: Optional[torch.Tensor] = None,
    coeffs_y: Optional[torch.Tensor] = None,
    coeffs_z: Optional[torch.Tensor] = None,
    dim_arrange = None,
    freq_base = None,
    freq_scale = None,
    depth = None,
) -> Tuple[
    Callable[[torch.Tensor], torch.Tensor],
    Callable[[torch.Tensor], torch.Tensor],
    Callable[[torch.Tensor], torch.Tensor],
]:
    """Prepare transforms for PRoPE-style positional encoding."""
    device = viewmats.device
    (batch, cameras, _, _) = viewmats.shape

    viewmats = viewmats.to(torch.float32)
    Ks = Ks.to(torch.float32)

    # Normalize camera intrinsics.
    if Ks is not None:  
        # Ks has been normalized in the dataset getitem !!
        Ks_norm = Ks
        # Compute the camera projection matrices we use in PRoPE.
        # - K is an `image<-camera` transform.
        # - viewmats is a `camera<-world` transform.
        # - P = lift(K) @ viewmats is an `image<-world` transform.
        P = torch.einsum("...ij,...jk->...ik", _lift_K(Ks_norm), viewmats)
        P_T = P.transpose(-1, -2)
        P_inv = torch.einsum(
            "...ij,...jk->...ik",
            _invert_SE3(viewmats),
            _lift_K(_invert_K(Ks_norm)),
        )

    else:
        # GTA formula. P is `camera<-world` transform.
        P = viewmats
        P_T = P.transpose(-1, -2)
        P_inv = _invert_SE3(viewmats)

    assert P.shape == P_inv.shape == (batch, cameras, 4, 4)

    # Precompute cos/sin terms for RoPE. We use tiles/repeats for 'row-major'
    # broadcasting.

    assert coeffs_x is not None
    if coeffs_x is None:
        coeffs_x = _rope_precompute_coeffs(
            torch.tile(torch.arange(patches_x, device=device), (patches_y * cameras,)),
            freq_base=100.0,
            freq_scale=1.0,
            # feat_dim=head_dim // 4,
            feat_dim=dim_arrange[1],
        )
    assert coeffs_y is not None
    if coeffs_y is None:
        coeffs_y = _rope_precompute_coeffs(
            torch.tile(
                torch.repeat_interleave(
                    torch.arange(patches_y, device=device), patches_x
                ),
                (cameras,),
            ),
            freq_base=100.0,
            freq_scale=1.0,
            # feat_dim=head_dim // 4,
            feat_dim=dim_arrange[2],
        )

    if torch.isnan(P_inv).any():
        print("!!P_inv has NaN!!!")
        exit(0)
    if torch.isnan(P_T).any():
        print("!!P_T has NaN!!!")
        exit(0)
    if torch.isnan(coeffs_x[0]).any() or torch.isnan(coeffs_x[1]).any():
        print("!!coeffs_x has NaN!!!")
        exit(0)
    if torch.isnan(coeffs_y[0]).any() or torch.isnan(coeffs_y[1]).any():
        print("!!coeffs_y has NaN!!!")
        exit(0)


    # Block-diagonal transforms to the inputs and outputs of the attention operator.
    assert head_dim % 4 == 0

    transforms_q = [
        (partial(_apply_tiled_projmat, matrix=P_T), dim_arrange[0]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_x), dim_arrange[1]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_y), dim_arrange[2]),
    ]
    transforms_kv = [
        (partial(_apply_tiled_projmat, matrix=P_inv), dim_arrange[0]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_x), dim_arrange[1]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_y), dim_arrange[2]),
    ]
    transforms_o = [
        (partial(_apply_tiled_projmat, matrix=P), dim_arrange[0]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_x, inverse=True), dim_arrange[1]),
        (partial(_rope_apply_coeffs, coeffs=coeffs_y, inverse=True), dim_arrange[2]),
    ]

    if len(dim_arrange) == 4:
        index_z = rearrange(depth, 'b n h w -> b n (h w)') # (batch, frame, seq_len)
        coeffs_z: Tuple[torch.Tensor, torch.Tensor] = _rope_precompute_coeffs_z(
            index_z,
            freq_base=freq_base,
            freq_scale=freq_scale,
            feat_dim=dim_arrange[3],
        )
        coeffs_z_0 = coeffs_z[0]
        coeffs_z_1 = coeffs_z[1]
        coeffs_z = (coeffs_z_0, coeffs_z_1)

        transforms_q += [(partial(_rope_apply_coeffs_z, coeffs=coeffs_z), dim_arrange[3])]
        transforms_kv += [(partial(_rope_apply_coeffs_z, coeffs=coeffs_z), dim_arrange[3])]
        transforms_o += [(partial(_rope_apply_coeffs_z, coeffs=coeffs_z, inverse=True), dim_arrange[3])]

    apply_fn_q = partial(_apply_block_diagonal, func_size_pairs=transforms_q)
    apply_fn_kv = partial(_apply_block_diagonal, func_size_pairs=transforms_kv)
    apply_fn_o = partial(_apply_block_diagonal, func_size_pairs=transforms_o)
    return apply_fn_q, apply_fn_kv, apply_fn_o


def _apply_tiled_projmat(
    feats: torch.Tensor,  # (batch, num_heads, seqlen, feat_dim)
    matrix: torch.Tensor,  # (batch, cameras, D, D)
) -> torch.Tensor:
    """Apply projection matrix to features."""
    # - seqlen => (cameras, patches_x * patches_y)
    # - feat_dim => (feat_dim // 4, 4)

    matrix = matrix.to(feats.dtype)

    (batch, num_heads, seqlen, feat_dim) = feats.shape
    cameras = matrix.shape[1]
    assert seqlen > cameras and seqlen % cameras == 0
    D = matrix.shape[-1]
    assert matrix.shape == (batch, cameras, D, D)
    assert feat_dim % D == 0
    # print(matrix.device, feats.device)
    return torch.einsum(
        "bcij,bncpkj->bncpki",
        matrix,
        feats.reshape((batch, num_heads, cameras, -1, feat_dim // D, D)),
    ).reshape(feats.shape)


def _rope_apply_coeffs_z(
    feats: torch.Tensor,  # (batch, num_heads, seqlen_total, feat_dim)
    coeffs: Tuple[torch.Tensor, torch.Tensor],  # (batch, 1, frame, seqlen_img, num_freqs)
    inverse: bool = False,
) -> torch.Tensor:
    """Apply RoPE coefficients to features. We adopt a 'split' ordering
    convention. (in contrast to 'interleaved')"""

    # print("Inject z rope!!")
    #TODO change to interleaved same as Qwen?
    cos, sin = coeffs 

    batch, num_heads, total_seq_len, feat_dim = feats.shape
    _, __, frames, seq_len_per_img, num_freqs = cos.shape

    cos = cos.to(feats.dtype)   
    sin = sin.to(feats.dtype)
    # We allow (cos, sin) to be either with shape (1, 1, seqlen, feat_dim // 2),
    # or (1, 1, seqlen_per_image, feat_dim // 2) and we repeat it to
    # match the shape of feats.
    feats = feats.reshape((batch, num_heads, frames, seq_len_per_img, feat_dim))
    assert feats.shape[3] * frames == total_seq_len
    # if cos.shape[2] != feats.shape[2]:
    #     n_repeats = feats.shape[2] // cos.shape[2]
    #     cos = cos.repeat(1, 1, n_repeats, 1)
    #     sin = sin.repeat(1, 1, n_repeats, 1)
    assert len(cos.shape) == len(sin.shape) == len(feats.shape) == 5 
    assert cos.shape[-1] == sin.shape[-1] == feats.shape[-1] // 2

    # cos (batch, 1, frame, seqlen_img, feat_dim)
    x_in = feats[..., ::2]  # even  # (batch, num_heads, frames, seqlen_img, feat_dim)
    y_in = feats[..., 1::2]

    if inverse == False: # for qkv
        x_out = cos * x_in - sin * y_in # broadcast on "num_heads"
        y_out = sin * x_in + cos * y_in
    else:   # for out 
        x_out = cos * x_in + sin * y_in
        y_out = -sin * x_in + cos * y_in

    res = torch.stack((x_out, y_out), dim=-1).flatten(start_dim=-2)
    res = rearrange(res, 'b n f s d -> b n (f s) d')
    # print(res.shape)
    return res

def _rope_precompute_coeffs_z(
    positions: torch.Tensor,  # (batch, frame, seq_len)
    freq_base: float,
    freq_scale: float,
    feat_dim: int,
) -> Tuple[torch.Tensor, torch.Tensor]:
    """Precompute RoPE coefficients."""
    assert len(positions.shape) == 3
    assert feat_dim % 2 == 0
    num_freqs = feat_dim // 2
    freqs = freq_scale * (
        freq_base
        ** (
            -torch.arange(num_freqs, device=positions.device)[None, None, None, :]
            / num_freqs
        )
    )
    # print(freqs.shape)
    # print(positions[:128])
    angles = positions[:, None, :, :, None] * freqs
    # Shape should be: `(batch, num_heads, frame, seqlen, num_freqs)`; we're
    # broadcasting across `num_heads`.
    assert angles.shape == (positions.shape[0], 1, positions.shape[1], positions.shape[2], num_freqs)
    return torch.cos(angles), torch.sin(angles)


def _rope_precompute_coeffs(
    positions: torch.Tensor,  # (seqlen,)
    freq_base: float,
    freq_scale: float,
    feat_dim: int,
) -> Tuple[torch.Tensor, torch.Tensor]:
    """Precompute RoPE coefficients."""
    assert len(positions.shape) == 1
    assert feat_dim % 2 == 0
    num_freqs = feat_dim // 2
    freqs = freq_scale * (
        freq_base
        ** (
            -torch.arange(num_freqs, device=positions.device)[None, None, None, :]
            / num_freqs
        )
    )
    # print(freqs.shape)
    # print(positions[:128])
    angles = positions[None, None, :, None] * freqs
    # Shape should be: `(batch, num_heads, seqlen, num_freqs)`; we're
    # broadcasting across `batch` and `num_heads`.
    assert angles.shape == (1, 1, positions.shape[0], num_freqs)
    return torch.cos(angles), torch.sin(angles)


if __name__ == '__main__':

    patches_x = 64
    patches_y = 64
    freq_base = 1
    freq_scale = 10000
    head_dim = 128

    pos_index = torch.arange(patches_x)
    neg_index = torch.arange(patches_x).flip(0) * -1 - 1
    index = torch.cat([neg_index[-(patches_x - patches_x // 2) :], pos_index[: patches_x // 2]], dim=0)
    print(index)
    print(torch.arange(patches_x))
    coeffs_x: Tuple[torch.Tensor, torch.Tensor] = _rope_precompute_coeffs(
            # torch.tile(torch.arange(patches_x), (patches_y,)),
            torch.tile(index, (patches_y,)),
            freq_base=freq_base,
            freq_scale=freq_scale,
            # feat_dim=head_dim // 4,
            feat_dim=56,
        )

    coeffs_y: Tuple[torch.Tensor, torch.Tensor] = _rope_precompute_coeffs(
            torch.repeat_interleave(torch.arange(patches_y), patches_x),
            freq_base=freq_base,
            freq_scale=freq_scale,
            # feat_dim=head_dim // 4,
            feat_dim=56,
        )

def _rope_apply_coeffs(
    feats: torch.Tensor,  # (batch, num_heads, seqlen, feat_dim)
    coeffs: Tuple[torch.Tensor, torch.Tensor],
    inverse: bool = False,
) -> torch.Tensor:
    """Apply RoPE coefficients to features. We adopt a 'split' ordering
    convention. (in contrast to 'interleaved')"""

    #TODO change to interleaved same as Qwen?
    cos, sin = coeffs

    cos = cos.to(feats.dtype)
    sin = sin.to(feats.dtype)
    # We allow (cos, sin) to be either with shape (1, 1, seqlen, feat_dim // 2),
    # or (1, 1, seqlen_per_image, feat_dim // 2) and we repeat it to
    # match the shape of feats.
    if cos.shape[2] != feats.shape[2]:
        n_repeats = feats.shape[2] // cos.shape[2]
        cos = cos.repeat(1, 1, n_repeats, 1)
        sin = sin.repeat(1, 1, n_repeats, 1)
    assert len(feats.shape) == len(cos.shape) == len(sin.shape) == 4
    assert cos.shape[-1] == sin.shape[-1] == feats.shape[-1] // 2

    x_in = feats[..., ::2]  # even  # (batch, num_heads, seqlen, feat_dim)
    y_in = feats[..., 1::2]

    if inverse == False: # for qkv
        x_out = cos * x_in - sin * y_in
        y_out = sin * x_in + cos * y_in
    else:   # for out 
        x_out = cos * x_in + sin * y_in
        y_out = -sin * x_in + cos * y_in

    res = torch.stack((x_out, y_out), dim=-1).flatten(start_dim=-2)
    # print(res.shape)
    return res



def _apply_block_diagonal(
    feats: torch.Tensor,  # (..., dim)
    func_size_pairs: List[Tuple[Callable[[torch.Tensor], torch.Tensor], int]],
) -> torch.Tensor:
    """Apply a block-diagonal function to an input array.

    Each function is specified as a tuple with form:

        ((Tensor) -> Tensor, int)

    Where the integer is the size of the input to the function.
    """
    funcs, block_sizes = zip(*func_size_pairs)
    assert feats.shape[-1] == sum(block_sizes)
    x_blocks = torch.split(feats, block_sizes, dim=-1)
    out = torch.cat(
        [f(x_block) for f, x_block in zip(funcs, x_blocks)],
        dim=-1,
    )
    assert out.shape == feats.shape, "Input/output shapes should match."
    return out


def _invert_SE3(transforms: torch.Tensor) -> torch.Tensor:
    """Invert a 4x4 SE(3) matrix."""
    assert transforms.shape[-2:] == (4, 4)
    Rinv = transforms[..., :3, :3].transpose(-1, -2)
    out = torch.zeros_like(transforms)
    out[..., :3, :3] = Rinv
    out[..., :3, 3] = -torch.einsum("...ij,...j->...i", Rinv, transforms[..., :3, 3])
    out[..., 3, 3] = 1.0
    return out


def _lift_K(Ks: torch.Tensor) -> torch.Tensor:
    """Lift 3x3 matrices to homogeneous 4x4 matrices."""
    assert Ks.shape[-2:] == (3, 3)
    out = torch.zeros(Ks.shape[:-2] + (4, 4), device=Ks.device)
    out[..., :3, :3] = Ks
    out[..., 3, 3] = 1.0
    return out


def _invert_K(Ks: torch.Tensor) -> torch.Tensor:
    """Invert 3x3 intrinsics matrices. Assumes no skew."""
    assert Ks.shape[-2:] == (3, 3)
    out = torch.zeros_like(Ks)
    out[..., 0, 0] = 1.0 / Ks[..., 0, 0]
    out[..., 1, 1] = 1.0 / Ks[..., 1, 1]
    out[..., 0, 2] = -Ks[..., 0, 2] / Ks[..., 0, 0]
    out[..., 1, 2] = -Ks[..., 1, 2] / Ks[..., 1, 1]
    out[..., 2, 2] = 1.0
    return out