{{ message }}
Repository navigation
Expand file tree
/
Copy path_helpers.py
More file actions
431 lines (381 loc) · 15.5 KB
/
Copy path_helpers.py
File metadata and controls
431 lines (381 loc) · 15.5 KB
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
# Copyright 2026 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import inspect
from dataclasses import dataclass
from typing import Any, Callable, Type
@dataclass
class AttentionProcessorMetadata:
skip_processor_output_fn: Callable[[Any], Any]
@dataclass
class TransformerBlockMetadata:
return_hidden_states_index: int = None
return_encoder_hidden_states_index: int = None
hidden_states_argument_name: str = "hidden_states"
encoder_hidden_states_argument_name: str = "encoder_hidden_states"
_cls: Type = None
_cached_parameter_indices: dict[str, int] = None
def _get_parameter_from_args_kwargs(self, identifier: str, args=(), kwargs=None):
kwargs = kwargs or {}
if identifier in kwargs:
return kwargs[identifier]
if self._cached_parameter_indices is not None:
return args[self._cached_parameter_indices[identifier]]
if self._cls is None:
raise ValueError("Model class is not set for metadata.")
parameters = list(inspect.signature(self._cls.forward).parameters.keys())
parameters = parameters[1:] # skip `self`
self._cached_parameter_indices = {param: i for i, param in enumerate(parameters)}
if identifier not in self._cached_parameter_indices:
raise ValueError(f"Parameter '{identifier}' not found in function signature but was requested.")
index = self._cached_parameter_indices[identifier]
if index >= len(args):
raise ValueError(f"Expected {index} arguments but got {len(args)}.")
return args[index]
class AttentionProcessorRegistry:
_registry = {}
# TODO(aryan): this is only required for the time being because we need to do the registrations
# for classes. If we do it eagerly, i.e. call the functions in global scope, we will get circular
# import errors because of the models imported in this file.
_is_registered = False
@classmethod
def register(cls, model_class: Type, metadata: AttentionProcessorMetadata):
cls._register()
cls._registry[model_class] = metadata
@classmethod
def get(cls, model_class: Type) -> AttentionProcessorMetadata:
cls._register()
if model_class not in cls._registry:
raise ValueError(f"Model class {model_class} not registered.")
return cls._registry[model_class]
@classmethod
def _register(cls):
if cls._is_registered:
return
cls._is_registered = True
_register_attention_processors_metadata()
class TransformerBlockRegistry:
_registry = {}
# TODO(aryan): this is only required for the time being because we need to do the registrations
# for classes. If we do it eagerly, i.e. call the functions in global scope, we will get circular
# import errors because of the models imported in this file.
_is_registered = False
@classmethod
def register(cls, model_class: Type, metadata: TransformerBlockMetadata):
cls._register()
metadata._cls = model_class
cls._registry[model_class] = metadata
@classmethod
def get(cls, model_class: Type) -> TransformerBlockMetadata:
cls._register()
if model_class not in cls._registry:
raise ValueError(f"Model class {model_class} not registered.")
return cls._registry[model_class]
@classmethod
def _register(cls):
if cls._is_registered:
return
cls._is_registered = True
_register_transformer_blocks_metadata()
def _register_attention_processors_metadata():
from ..models.attention_processor import AttnProcessor2_0
from ..models.transformers.transformer_cogview4 import CogView4AttnProcessor
from ..models.transformers.transformer_flux import FluxAttnProcessor
from ..models.transformers.transformer_hunyuanimage import HunyuanImageAttnProcessor
from ..models.transformers.transformer_qwenimage import QwenDoubleStreamAttnProcessor2_0
from ..models.transformers.transformer_wan import WanAttnProcessor2_0
from ..models.transformers.transformer_z_image import ZSingleStreamAttnProcessor
# AttnProcessor2_0
AttentionProcessorRegistry.register(
model_class=AttnProcessor2_0,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_AttnProcessor2_0,
),
)
# CogView4AttnProcessor
AttentionProcessorRegistry.register(
model_class=CogView4AttnProcessor,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_CogView4AttnProcessor,
),
)
# WanAttnProcessor2_0
AttentionProcessorRegistry.register(
model_class=WanAttnProcessor2_0,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_WanAttnProcessor2_0,
),
)
# FluxAttnProcessor
AttentionProcessorRegistry.register(
model_class=FluxAttnProcessor,
metadata=AttentionProcessorMetadata(skip_processor_output_fn=_skip_proc_output_fn_Attention_FluxAttnProcessor),
)
# QwenDoubleStreamAttnProcessor2
AttentionProcessorRegistry.register(
model_class=QwenDoubleStreamAttnProcessor2_0,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_QwenDoubleStreamAttnProcessor2_0
),
)
# HunyuanImageAttnProcessor
AttentionProcessorRegistry.register(
model_class=HunyuanImageAttnProcessor,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_HunyuanImageAttnProcessor,
),
)
# ZSingleStreamAttnProcessor
AttentionProcessorRegistry.register(
model_class=ZSingleStreamAttnProcessor,
metadata=AttentionProcessorMetadata(
skip_processor_output_fn=_skip_proc_output_fn_Attention_ZSingleStreamAttnProcessor,
),
)
def _register_transformer_blocks_metadata():
from ..models.attention import BasicTransformerBlock, JointTransformerBlock
from ..models.transformers.cogvideox_transformer_3d import CogVideoXBlock
from ..models.transformers.transformer_bria import BriaTransformerBlock
from ..models.transformers.transformer_cogview4 import CogView4TransformerBlock
from ..models.transformers.transformer_cosmos3 import Cosmos3VLTextMoTDecoderLayer
from ..models.transformers.transformer_flux import FluxSingleTransformerBlock, FluxTransformerBlock
from ..models.transformers.transformer_flux2 import Flux2SingleTransformerBlock, Flux2TransformerBlock
from ..models.transformers.transformer_hunyuan_video import (
HunyuanVideoSingleTransformerBlock,
HunyuanVideoTokenReplaceSingleTransformerBlock,
HunyuanVideoTokenReplaceTransformerBlock,
HunyuanVideoTransformerBlock,
)
from ..models.transformers.transformer_hunyuanimage import (
HunyuanImageSingleTransformerBlock,
HunyuanImageTransformerBlock,
)
from ..models.transformers.transformer_kandinsky import Kandinsky5TransformerDecoderBlock
from ..models.transformers.transformer_ltx import LTXVideoTransformerBlock
from ..models.transformers.transformer_mochi import MochiTransformerBlock
from ..models.transformers.transformer_motif_video import (
MotifVideoSingleTransformerBlock,
MotifVideoTransformerBlock,
)
from ..models.transformers.transformer_qwenimage import QwenImageTransformerBlock
from ..models.transformers.transformer_wan import WanTransformerBlock
from ..models.transformers.transformer_z_image import ZImageTransformerBlock
# BasicTransformerBlock
TransformerBlockRegistry.register(
model_class=BasicTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
TransformerBlockRegistry.register(
model_class=BriaTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
# CogVideoX
TransformerBlockRegistry.register(
model_class=CogVideoXBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# CogView4
TransformerBlockRegistry.register(
model_class=CogView4TransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# Cosmos 3
TransformerBlockRegistry.register(
model_class=Cosmos3VLTextMoTDecoderLayer,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
hidden_states_argument_name="gen_seq",
encoder_hidden_states_argument_name="und_seq",
),
)
# Flux
TransformerBlockRegistry.register(
model_class=FluxTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
),
)
TransformerBlockRegistry.register(
model_class=FluxSingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
),
)
# Flux2
TransformerBlockRegistry.register(
model_class=Flux2TransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
),
)
TransformerBlockRegistry.register(
model_class=Flux2SingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
# HunyuanVideo
TransformerBlockRegistry.register(
model_class=HunyuanVideoTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
TransformerBlockRegistry.register(
model_class=HunyuanVideoSingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
TransformerBlockRegistry.register(
model_class=HunyuanVideoTokenReplaceTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
TransformerBlockRegistry.register(
model_class=HunyuanVideoTokenReplaceSingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# LTXVideo
TransformerBlockRegistry.register(
model_class=LTXVideoTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
# Mochi
TransformerBlockRegistry.register(
model_class=MochiTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# MotifVideo
TransformerBlockRegistry.register(
model_class=MotifVideoTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
TransformerBlockRegistry.register(
model_class=MotifVideoSingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# Wan
TransformerBlockRegistry.register(
model_class=WanTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
# QwenImage
TransformerBlockRegistry.register(
model_class=QwenImageTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
),
)
# HunyuanImage2.1
TransformerBlockRegistry.register(
model_class=HunyuanImageTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
TransformerBlockRegistry.register(
model_class=HunyuanImageSingleTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=1,
),
)
# ZImage
TransformerBlockRegistry.register(
model_class=ZImageTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
),
)
TransformerBlockRegistry.register(
model_class=JointTransformerBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=1,
return_encoder_hidden_states_index=0,
),
)
# Kandinsky 5.0 (Kandinsky5TransformerDecoderBlock)
TransformerBlockRegistry.register(
model_class=Kandinsky5TransformerDecoderBlock,
metadata=TransformerBlockMetadata(
return_hidden_states_index=0,
return_encoder_hidden_states_index=None,
hidden_states_argument_name="visual_embed",
),
)
# fmt: off
def _skip_attention___ret___hidden_states(self, *args, **kwargs):
hidden_states = kwargs.get("hidden_states", None)
if hidden_states is None and len(args) > 0:
hidden_states = args[0]
return hidden_states
def _skip_attention___ret___hidden_states___encoder_hidden_states(self, *args, **kwargs):
hidden_states = kwargs.get("hidden_states", None)
encoder_hidden_states = kwargs.get("encoder_hidden_states", None)
if hidden_states is None and len(args) > 0:
hidden_states = args[0]
if encoder_hidden_states is None and len(args) > 1:
encoder_hidden_states = args[1]
return hidden_states, encoder_hidden_states
_skip_proc_output_fn_Attention_AttnProcessor2_0 = _skip_attention___ret___hidden_states
_skip_proc_output_fn_Attention_CogView4AttnProcessor = _skip_attention___ret___hidden_states___encoder_hidden_states
_skip_proc_output_fn_Attention_WanAttnProcessor2_0 = _skip_attention___ret___hidden_states
# not sure what this is yet.
_skip_proc_output_fn_Attention_FluxAttnProcessor = _skip_attention___ret___hidden_states
_skip_proc_output_fn_Attention_QwenDoubleStreamAttnProcessor2_0 = _skip_attention___ret___hidden_states
_skip_proc_output_fn_Attention_HunyuanImageAttnProcessor = _skip_attention___ret___hidden_states
_skip_proc_output_fn_Attention_ZSingleStreamAttnProcessor = _skip_attention___ret___hidden_states
# fmt: on
You can’t perform that action at this time.