File size: 27,281 Bytes
bdd549c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import inspect
import math
from typing import Callable, List, Optional, Union

import torch
import torch.nn.functional as F
from torch import nn

from diffusers.image_processor import IPAdapterMaskProcessor
from diffusers.utils import deprecate, logging
from diffusers.utils.import_utils import is_torch_npu_available, is_xformers_available
from diffusers.utils.torch_utils import maybe_allow_in_graph
from diffusers.models.attention_processor import Attention


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

if is_torch_npu_available():
    import torch_npu

if is_xformers_available():
    import xformers
    import xformers.ops
else:
    xformers = None

class AttnProcessor:
    r"""
    Default processor for performing attention-related computations.
    """

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.Tensor,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        temb: Optional[torch.Tensor] = None,
        *args,
        **kwargs,
    ) -> torch.Tensor:
        if len(args) > 0 or kwargs.get("scale", None) is not None:
            deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
            deprecate("scale", "1.0.0", deprecation_message)

        residual = hidden_states

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states

class AttnProcessor2_0(nn.Module):
    r"""
    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
    """

    def __init__(self):
        super().__init__()
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.Tensor,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        temb: Optional[torch.Tensor] = None,
        ip_adapter_masks: Optional[torch.Tensor] = None,
        *args,
        **kwargs,
    ) -> torch.Tensor:
        if len(args) > 0 or kwargs.get("scale", None) is not None:
            deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
            deprecate("scale", "1.0.0", deprecation_message)

        residual = hidden_states
        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states
    
class IPAdapterAttnProcessor(nn.Module):
    r"""
    Attention processor for Multiple IP-Adapters.

    Args:
        hidden_size (`int`):
            The hidden size of the attention layer.
        cross_attention_dim (`int`):
            The number of channels in the `encoder_hidden_states`.
        num_tokens (`int`, `Tuple[int]` or `List[int]`, defaults to `(4,)`):
            The context length of the image features.
        scale (`float` or List[`float`], defaults to 1.0):
            the weight scale of image prompt.
    """

    def __init__(self, hidden_size, cross_attention_dim=None, num_tokens=(4,), scale=1.0):
        super().__init__()

        self.hidden_size = hidden_size
        self.cross_attention_dim = cross_attention_dim

        if not isinstance(num_tokens, (tuple, list)):
            num_tokens = [num_tokens]
        self.num_tokens = num_tokens

        if not isinstance(scale, list):
            scale = [scale] * len(num_tokens)
        if len(scale) != len(num_tokens):
            raise ValueError("`scale` should be a list of integers with the same length as `num_tokens`.")
        self.scale = scale

        self.to_k_ip = nn.ModuleList(
            [nn.Linear(cross_attention_dim, hidden_size, bias=False) for _ in range(len(num_tokens))]
        )
        self.to_v_ip = nn.ModuleList(
            [nn.Linear(cross_attention_dim, hidden_size, bias=False) for _ in range(len(num_tokens))]
        )

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.Tensor,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        temb: Optional[torch.Tensor] = None,
        scale: float = 1.0,
        ip_adapter_masks: Optional[torch.Tensor] = None,
    ):
        residual = hidden_states

        # separate ip_hidden_states from encoder_hidden_states
        if encoder_hidden_states is not None:
            if isinstance(encoder_hidden_states, tuple):
                encoder_hidden_states, ip_hidden_states = encoder_hidden_states
            else:
                deprecation_message = (
                    "You have passed a tensor as `encoder_hidden_states`. This is deprecated and will be removed in a future release."
                    " Please make sure to update your script to pass `encoder_hidden_states` as a tuple to suppress this warning."
                )
                deprecate("encoder_hidden_states not a tuple", "1.0.0", deprecation_message, standard_warn=False)
                end_pos = encoder_hidden_states.shape[1] - self.num_tokens[0]
                encoder_hidden_states, ip_hidden_states = (
                    encoder_hidden_states[:, :end_pos, :],
                    [encoder_hidden_states[:, end_pos:, :]],
                )

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        if ip_adapter_masks is not None:
            if not isinstance(ip_adapter_masks, List):
                # for backward compatibility, we accept `ip_adapter_mask` as a tensor of shape [num_ip_adapter, 1, height, width]
                ip_adapter_masks = list(ip_adapter_masks.unsqueeze(1))
            if not (len(ip_adapter_masks) == len(self.scale) == len(ip_hidden_states)):
                raise ValueError(
                    f"Length of ip_adapter_masks array ({len(ip_adapter_masks)}) must match "
                    f"length of self.scale array ({len(self.scale)}) and number of ip_hidden_states "
                    f"({len(ip_hidden_states)})"
                )
            else:
                for index, (mask, scale, ip_state) in enumerate(zip(ip_adapter_masks, self.scale, ip_hidden_states)):
                    if not isinstance(mask, torch.Tensor) or mask.ndim != 4:
                        raise ValueError(
                            "Each element of the ip_adapter_masks array should be a tensor with shape "
                            "[1, num_images_for_ip_adapter, height, width]."
                            " Please use `IPAdapterMaskProcessor` to preprocess your mask"
                        )
                    if mask.shape[1] != ip_state.shape[1]:
                        raise ValueError(
                            f"Number of masks ({mask.shape[1]}) does not match "
                            f"number of ip images ({ip_state.shape[1]}) at index {index}"
                        )
                    if isinstance(scale, list) and not len(scale) == mask.shape[1]:
                        raise ValueError(
                            f"Number of masks ({mask.shape[1]}) does not match "
                            f"number of scales ({len(scale)}) at index {index}"
                        )
        else:
            ip_adapter_masks = [None] * len(self.scale)

        # for ip-adapter
        for current_ip_hidden_states, scale, to_k_ip, to_v_ip, mask in zip(
            ip_hidden_states, self.scale, self.to_k_ip, self.to_v_ip, ip_adapter_masks
        ):
            skip = False
            if isinstance(scale, list):
                if all(s == 0 for s in scale):
                    skip = True
            elif scale == 0:
                skip = True
            if not skip:
                if mask is not None:
                    if not isinstance(scale, list):
                        scale = [scale] * mask.shape[1]

                    current_num_images = mask.shape[1]
                    for i in range(current_num_images):
                        ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
                        ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])

                        ip_key = attn.head_to_batch_dim(ip_key)
                        ip_value = attn.head_to_batch_dim(ip_value)

                        ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
                        _current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
                        _current_ip_hidden_states = attn.batch_to_head_dim(_current_ip_hidden_states)

                        mask_downsample = IPAdapterMaskProcessor.downsample(
                            mask[:, i, :, :],
                            batch_size,
                            _current_ip_hidden_states.shape[1],
                            _current_ip_hidden_states.shape[2],
                        )

                        mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)

                        hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
                else:
                    ip_key = to_k_ip(current_ip_hidden_states)
                    ip_value = to_v_ip(current_ip_hidden_states)

                    ip_key = attn.head_to_batch_dim(ip_key)
                    ip_value = attn.head_to_batch_dim(ip_value)

                    ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
                    current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
                    current_ip_hidden_states = attn.batch_to_head_dim(current_ip_hidden_states)

                    hidden_states = hidden_states + scale * current_ip_hidden_states

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states


class IPAdapterAttnProcessor2_0(torch.nn.Module):
    r"""
    Attention processor for IP-Adapter for PyTorch 2.0.

    Args:
        hidden_size (`int`):
            The hidden size of the attention layer.
        cross_attention_dim (`int`):
            The number of channels in the `encoder_hidden_states`.
        num_tokens (`int`, `Tuple[int]` or `List[int]`, defaults to `(4,)`):
            The context length of the image features.
        scale (`float` or `List[float]`, defaults to 1.0):
            the weight scale of image prompt.
    """

    def __init__(self, hidden_size, cross_attention_dim=None, num_tokens=(4,), scale=1.0):
        super().__init__()

        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError(
                f"{self.__class__.__name__} requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
            )

        self.hidden_size = hidden_size
        self.cross_attention_dim = cross_attention_dim

        if not isinstance(num_tokens, (tuple, list)):
            num_tokens = [num_tokens]
        self.num_tokens = num_tokens

        if not isinstance(scale, list):
            scale = [scale] * len(num_tokens)
        if len(scale) != len(num_tokens):
            raise ValueError("`scale` should be a list of integers with the same length as `num_tokens`.")
        self.scale = scale

        self.to_k_ip = nn.ModuleList(
            [nn.Linear(cross_attention_dim, hidden_size, bias=False) for _ in range(len(num_tokens))]
        )
        self.to_v_ip = nn.ModuleList(
            [nn.Linear(cross_attention_dim, hidden_size, bias=False) for _ in range(len(num_tokens))]
        )

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.Tensor,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        temb: Optional[torch.Tensor] = None,
        scale: float = 1.0,
        ip_adapter_masks: Optional[torch.Tensor] = None,
    ):
        residual = hidden_states

        # separate ip_hidden_states from encoder_hidden_states
        if encoder_hidden_states is not None:
            if isinstance(encoder_hidden_states, tuple):
                encoder_hidden_states, ip_hidden_states = encoder_hidden_states
                
            else:
                deprecation_message = (
                    "You have passed a tensor as `encoder_hidden_states`. This is deprecated and will be removed in a future release."
                    " Please make sure to update your script to pass `encoder_hidden_states` as a tuple to suppress this warning."
                )
                deprecate("encoder_hidden_states not a tuple", "1.0.0", deprecation_message, standard_warn=False)
                end_pos = encoder_hidden_states.shape[1] - self.num_tokens[0]
                encoder_hidden_states, ip_hidden_states = (
                    encoder_hidden_states[:, :end_pos, :],
                    [encoder_hidden_states[:, end_pos:, :]],
                )

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        if ip_adapter_masks is not None:
            if not isinstance(ip_adapter_masks, List):
                # for backward compatibility, we accept `ip_adapter_mask` as a tensor of shape [num_ip_adapter, 1, height, width]
                ip_adapter_masks = list(ip_adapter_masks.unsqueeze(1))
            if not (len(ip_adapter_masks) == len(self.scale) == len(ip_hidden_states)):
                raise ValueError(
                    f"Length of ip_adapter_masks array ({len(ip_adapter_masks)}) must match "
                    f"length of self.scale array ({len(self.scale)}) and number of ip_hidden_states "
                    f"({len(ip_hidden_states)})"
                )
            else:
                for index, (mask, scale, ip_state) in enumerate(zip(ip_adapter_masks, self.scale, ip_hidden_states)):
                    ip_hidden_states[index] = ip_state = ip_state.unsqueeze(1)
                    if not isinstance(mask, torch.Tensor) or mask.ndim != 4:
                        raise ValueError(
                            "Each element of the ip_adapter_masks array should be a tensor with shape "
                            "[1, num_images_for_ip_adapter, height, width]."
                            " Please use `IPAdapterMaskProcessor` to preprocess your mask"
                        )
                    if mask.shape[1] != ip_state.shape[1]:
                        raise ValueError(
                            f"Number of masks ({mask.shape[1]}) does not match "
                            f"number of ip images ({ip_state.shape[1]}) at index {index}"
                        )
                    if isinstance(scale, list) and not len(scale) == mask.shape[1]:
                        raise ValueError(
                            f"Number of masks ({mask.shape[1]}) does not match "
                            f"number of scales ({len(scale)}) at index {index}"
                        )
        else:
            ip_adapter_masks = [None] * len(self.scale)

        # for ip-adapter
        for current_ip_hidden_states, scale, to_k_ip, to_v_ip, mask in zip(
            ip_hidden_states, self.scale, self.to_k_ip, self.to_v_ip, ip_adapter_masks
        ):
            skip = False
            if isinstance(scale, list):
                if all(s == 0 for s in scale):
                    skip = True
            elif scale == 0:
                skip = True
            if not skip:
                if mask is not None:
                    if not isinstance(scale, list):
                        scale = [scale] * mask.shape[1]

                    current_num_images = mask.shape[1]
                    for i in range(current_num_images):
                        ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
                        ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])

                        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
                        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

                        # the output of sdp = (batch, num_heads, seq_len, head_dim)
                        # TODO: add support for attn.scale when we move to Torch 2.1
                        _current_ip_hidden_states = F.scaled_dot_product_attention(
                            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
                        )

                        _current_ip_hidden_states = _current_ip_hidden_states.transpose(1, 2).reshape(
                            batch_size, -1, attn.heads * head_dim
                        )
                        _current_ip_hidden_states = _current_ip_hidden_states.to(query.dtype)

                        mask_downsample = IPAdapterMaskProcessor.downsample(
                            mask[:, i, :, :],
                            batch_size,
                            _current_ip_hidden_states.shape[1],
                            _current_ip_hidden_states.shape[2],
                        )
                        mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)
                        hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
                        
                else:
                    ip_key = to_k_ip(current_ip_hidden_states)
                    ip_value = to_v_ip(current_ip_hidden_states)

                    ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
                    ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

                    # the output of sdp = (batch, num_heads, seq_len, head_dim)
                    # TODO: add support for attn.scale when we move to Torch 2.1
                    current_ip_hidden_states = F.scaled_dot_product_attention(
                        query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
                    )

                    current_ip_hidden_states = current_ip_hidden_states.transpose(1, 2).reshape(
                        batch_size, -1, attn.heads * head_dim
                    )
                    current_ip_hidden_states = current_ip_hidden_states.to(query.dtype)

                    hidden_states = hidden_states + scale * current_ip_hidden_states


        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states