{{ message }}
Repository navigation
Expand file tree
/
Copy pathtest_scheduler.py
More file actions
executable file
·2885 lines (2173 loc) · 113 KB
/
Copy pathtest_scheduler.py
File metadata and controls
executable file
·2885 lines (2173 loc) · 113 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
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
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# coding=utf-8
# Copyright 2022 HuggingFace Inc.
#
# 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
import json
import os
import tempfile
import unittest
from typing import Dict, List, Tuple
import numpy as np
import torch
import torch.nn.functional as F
import diffusers
from diffusers import (
DDIMScheduler,
DDPMScheduler,
DEISMultistepScheduler,
DPMSolverMultistepScheduler,
DPMSolverSinglestepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
HeunDiscreteScheduler,
IPNDMScheduler,
KDPM2AncestralDiscreteScheduler,
KDPM2DiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
ScoreSdeVeScheduler,
UnCLIPScheduler,
VQDiffusionScheduler,
logging,
)
from diffusers.configuration_utils import ConfigMixin, register_to_config
from diffusers.schedulers.scheduling_utils import SchedulerMixin
from diffusers.utils import torch_device
from diffusers.utils.testing_utils import CaptureLogger
torch.backends.cuda.matmul.allow_tf32 = False
class SchedulerObject(SchedulerMixin, ConfigMixin):
config_name = "config.json"
@register_to_config
def __init__(
self,
a=2,
b=5,
c=(2, 5),
d="for diffusion",
e=[1, 3],
):
pass
class SchedulerObject2(SchedulerMixin, ConfigMixin):
config_name = "config.json"
@register_to_config
def __init__(
self,
a=2,
b=5,
c=(2, 5),
d="for diffusion",
f=[1, 3],
):
pass
class SchedulerObject3(SchedulerMixin, ConfigMixin):
config_name = "config.json"
@register_to_config
def __init__(
self,
a=2,
b=5,
c=(2, 5),
d="for diffusion",
e=[1, 3],
f=[1, 3],
):
pass
class SchedulerBaseTests(unittest.TestCase):
def test_save_load_from_different_config(self):
obj = SchedulerObject()
# mock add obj class to `diffusers`
setattr(diffusers, "SchedulerObject", SchedulerObject)
logger = logging.get_logger("diffusers.configuration_utils")
with tempfile.TemporaryDirectory() as tmpdirname:
obj.save_config(tmpdirname)
with CaptureLogger(logger) as cap_logger_1:
config = SchedulerObject2.load_config(tmpdirname)
new_obj_1 = SchedulerObject2.from_config(config)
# now save a config parameter that is not expected
with open(os.path.join(tmpdirname, SchedulerObject.config_name), "r") as f:
data = json.load(f)
data["unexpected"] = True
with open(os.path.join(tmpdirname, SchedulerObject.config_name), "w") as f:
json.dump(data, f)
with CaptureLogger(logger) as cap_logger_2:
config = SchedulerObject.load_config(tmpdirname)
new_obj_2 = SchedulerObject.from_config(config)
with CaptureLogger(logger) as cap_logger_3:
config = SchedulerObject2.load_config(tmpdirname)
new_obj_3 = SchedulerObject2.from_config(config)
assert new_obj_1.__class__ == SchedulerObject2
assert new_obj_2.__class__ == SchedulerObject
assert new_obj_3.__class__ == SchedulerObject2
assert cap_logger_1.out == ""
assert (
cap_logger_2.out
== "The config attributes {'unexpected': True} were passed to SchedulerObject, but are not expected and"
" will"
" be ignored. Please verify your config.json configuration file.\n"
)
assert cap_logger_2.out.replace("SchedulerObject", "SchedulerObject2") == cap_logger_3.out
def test_save_load_compatible_schedulers(self):
SchedulerObject2._compatibles = ["SchedulerObject"]
SchedulerObject._compatibles = ["SchedulerObject2"]
obj = SchedulerObject()
# mock add obj class to `diffusers`
setattr(diffusers, "SchedulerObject", SchedulerObject)
setattr(diffusers, "SchedulerObject2", SchedulerObject2)
logger = logging.get_logger("diffusers.configuration_utils")
with tempfile.TemporaryDirectory() as tmpdirname:
obj.save_config(tmpdirname)
# now save a config parameter that is expected by another class, but not origin class
with open(os.path.join(tmpdirname, SchedulerObject.config_name), "r") as f:
data = json.load(f)
data["f"] = [0, 0]
data["unexpected"] = True
with open(os.path.join(tmpdirname, SchedulerObject.config_name), "w") as f:
json.dump(data, f)
with CaptureLogger(logger) as cap_logger:
config = SchedulerObject.load_config(tmpdirname)
new_obj = SchedulerObject.from_config(config)
assert new_obj.__class__ == SchedulerObject
assert (
cap_logger.out
== "The config attributes {'unexpected': True} were passed to SchedulerObject, but are not expected and"
" will"
" be ignored. Please verify your config.json configuration file.\n"
)
def test_save_load_from_different_config_comp_schedulers(self):
SchedulerObject3._compatibles = ["SchedulerObject", "SchedulerObject2"]
SchedulerObject2._compatibles = ["SchedulerObject", "SchedulerObject3"]
SchedulerObject._compatibles = ["SchedulerObject2", "SchedulerObject3"]
obj = SchedulerObject()
# mock add obj class to `diffusers`
setattr(diffusers, "SchedulerObject", SchedulerObject)
setattr(diffusers, "SchedulerObject2", SchedulerObject2)
setattr(diffusers, "SchedulerObject3", SchedulerObject3)
logger = logging.get_logger("diffusers.configuration_utils")
logger.setLevel(diffusers.logging.INFO)
with tempfile.TemporaryDirectory() as tmpdirname:
obj.save_config(tmpdirname)
with CaptureLogger(logger) as cap_logger_1:
config = SchedulerObject.load_config(tmpdirname)
new_obj_1 = SchedulerObject.from_config(config)
with CaptureLogger(logger) as cap_logger_2:
config = SchedulerObject2.load_config(tmpdirname)
new_obj_2 = SchedulerObject2.from_config(config)
with CaptureLogger(logger) as cap_logger_3:
config = SchedulerObject3.load_config(tmpdirname)
new_obj_3 = SchedulerObject3.from_config(config)
assert new_obj_1.__class__ == SchedulerObject
assert new_obj_2.__class__ == SchedulerObject2
assert new_obj_3.__class__ == SchedulerObject3
assert cap_logger_1.out == ""
assert cap_logger_2.out == "{'f'} was not found in config. Values will be initialized to default values.\n"
assert cap_logger_3.out == "{'f'} was not found in config. Values will be initialized to default values.\n"
class SchedulerCommonTest(unittest.TestCase):
scheduler_classes = ()
forward_default_kwargs = ()
@property
def dummy_sample(self):
batch_size = 4
num_channels = 3
height = 8
width = 8
sample = torch.rand((batch_size, num_channels, height, width))
return sample
@property
def dummy_sample_deter(self):
batch_size = 4
num_channels = 3
height = 8
width = 8
num_elems = batch_size * num_channels * height * width
sample = torch.arange(num_elems)
sample = sample.reshape(num_channels, height, width, batch_size)
sample = sample / num_elems
sample = sample.permute(3, 0, 1, 2)
return sample
def get_scheduler_config(self):
raise NotImplementedError
def dummy_model(self):
def model(sample, t, *args):
return sample * t / (t + 1)
return model
def check_over_configs(self, time_step=0, **config):
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
for scheduler_class in self.scheduler_classes:
# TODO(Suraj) - delete the following two lines once DDPM, DDIM, and PNDM have timesteps casted to float by default
if scheduler_class in (EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, LMSDiscreteScheduler):
time_step = float(time_step)
scheduler_config = self.get_scheduler_config(**config)
scheduler = scheduler_class(**scheduler_config)
if scheduler_class == VQDiffusionScheduler:
num_vec_classes = scheduler_config["num_vec_classes"]
sample = self.dummy_sample(num_vec_classes)
model = self.dummy_model(num_vec_classes)
residual = model(sample, time_step)
else:
sample = self.dummy_sample
residual = 0.1 * sample
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_config(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
new_scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
# Set the seed before step() as some schedulers are stochastic like EulerAncestralDiscreteScheduler, EulerDiscreteScheduler
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
output = scheduler.step(residual, time_step, sample, **kwargs).prev_sample
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
new_output = new_scheduler.step(residual, time_step, sample, **kwargs).prev_sample
assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical"
def check_over_forward(self, time_step=0, **forward_kwargs):
kwargs = dict(self.forward_default_kwargs)
kwargs.update(forward_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
for scheduler_class in self.scheduler_classes:
if scheduler_class in (EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, LMSDiscreteScheduler):
time_step = float(time_step)
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
if scheduler_class == VQDiffusionScheduler:
num_vec_classes = scheduler_config["num_vec_classes"]
sample = self.dummy_sample(num_vec_classes)
model = self.dummy_model(num_vec_classes)
residual = model(sample, time_step)
else:
sample = self.dummy_sample
residual = 0.1 * sample
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_config(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
new_scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
output = scheduler.step(residual, time_step, sample, **kwargs).prev_sample
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
new_output = new_scheduler.step(residual, time_step, sample, **kwargs).prev_sample
assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical"
def test_from_save_pretrained(self):
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
for scheduler_class in self.scheduler_classes:
timestep = 1
if scheduler_class in (EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, LMSDiscreteScheduler):
timestep = float(timestep)
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
if scheduler_class == VQDiffusionScheduler:
num_vec_classes = scheduler_config["num_vec_classes"]
sample = self.dummy_sample(num_vec_classes)
model = self.dummy_model(num_vec_classes)
residual = model(sample, timestep)
else:
sample = self.dummy_sample
residual = 0.1 * sample
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_config(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
new_scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
output = scheduler.step(residual, timestep, sample, **kwargs).prev_sample
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
new_output = new_scheduler.step(residual, timestep, sample, **kwargs).prev_sample
assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical"
def test_compatibles(self):
for scheduler_class in self.scheduler_classes:
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
assert all(c is not None for c in scheduler.compatibles)
for comp_scheduler_cls in scheduler.compatibles:
comp_scheduler = comp_scheduler_cls.from_config(scheduler.config)
assert comp_scheduler is not None
new_scheduler = scheduler_class.from_config(comp_scheduler.config)
new_scheduler_config = {k: v for k, v in new_scheduler.config.items() if k in scheduler.config}
scheduler_diff = {k: v for k, v in new_scheduler.config.items() if k not in scheduler.config}
# make sure that configs are essentially identical
assert new_scheduler_config == dict(scheduler.config)
# make sure that only differences are for configs that are not in init
init_keys = inspect.signature(scheduler_class.__init__).parameters.keys()
assert set(scheduler_diff.keys()).intersection(set(init_keys)) == set()
def test_from_pretrained(self):
for scheduler_class in self.scheduler_classes:
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_pretrained(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
assert scheduler.config == new_scheduler.config
def test_step_shape(self):
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
timestep_0 = 0
timestep_1 = 1
for scheduler_class in self.scheduler_classes:
if scheduler_class in (EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, LMSDiscreteScheduler):
timestep_0 = float(timestep_0)
timestep_1 = float(timestep_1)
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
if scheduler_class == VQDiffusionScheduler:
num_vec_classes = scheduler_config["num_vec_classes"]
sample = self.dummy_sample(num_vec_classes)
model = self.dummy_model(num_vec_classes)
residual = model(sample, timestep_0)
else:
sample = self.dummy_sample
residual = 0.1 * sample
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
output_0 = scheduler.step(residual, timestep_0, sample, **kwargs).prev_sample
output_1 = scheduler.step(residual, timestep_1, sample, **kwargs).prev_sample
self.assertEqual(output_0.shape, sample.shape)
self.assertEqual(output_0.shape, output_1.shape)
def test_scheduler_outputs_equivalence(self):
def set_nan_tensor_to_zero(t):
t[t != t] = 0
return t
def recursive_check(tuple_object, dict_object):
if isinstance(tuple_object, (List, Tuple)):
for tuple_iterable_value, dict_iterable_value in zip(tuple_object, dict_object.values()):
recursive_check(tuple_iterable_value, dict_iterable_value)
elif isinstance(tuple_object, Dict):
for tuple_iterable_value, dict_iterable_value in zip(tuple_object.values(), dict_object.values()):
recursive_check(tuple_iterable_value, dict_iterable_value)
elif tuple_object is None:
return
else:
self.assertTrue(
torch.allclose(
set_nan_tensor_to_zero(tuple_object), set_nan_tensor_to_zero(dict_object), atol=1e-5
),
msg=(
"Tuple and dict output are not equal. Difference:"
f" {torch.max(torch.abs(tuple_object - dict_object))}. Tuple has `nan`:"
f" {torch.isnan(tuple_object).any()} and `inf`: {torch.isinf(tuple_object)}. Dict has"
f" `nan`: {torch.isnan(dict_object).any()} and `inf`: {torch.isinf(dict_object)}."
),
)
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", 50)
timestep = 0
if len(self.scheduler_classes) > 0 and self.scheduler_classes[0] == IPNDMScheduler:
timestep = 1
for scheduler_class in self.scheduler_classes:
if scheduler_class in (EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, LMSDiscreteScheduler):
timestep = float(timestep)
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
if scheduler_class == VQDiffusionScheduler:
num_vec_classes = scheduler_config["num_vec_classes"]
sample = self.dummy_sample(num_vec_classes)
model = self.dummy_model(num_vec_classes)
residual = model(sample, timestep)
else:
sample = self.dummy_sample
residual = 0.1 * sample
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
# Set the seed before state as some schedulers are stochastic like EulerAncestralDiscreteScheduler, EulerDiscreteScheduler
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
outputs_dict = scheduler.step(residual, timestep, sample, **kwargs)
if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
scheduler.set_timesteps(num_inference_steps)
elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
kwargs["num_inference_steps"] = num_inference_steps
# Set the seed before state as some schedulers are stochastic like EulerAncestralDiscreteScheduler, EulerDiscreteScheduler
if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
kwargs["generator"] = torch.manual_seed(0)
outputs_tuple = scheduler.step(residual, timestep, sample, return_dict=False, **kwargs)
recursive_check(outputs_tuple, outputs_dict)
def test_scheduler_public_api(self):
for scheduler_class in self.scheduler_classes:
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
if scheduler_class != VQDiffusionScheduler:
self.assertTrue(
hasattr(scheduler, "init_noise_sigma"),
f"{scheduler_class} does not implement a required attribute `init_noise_sigma`",
)
self.assertTrue(
hasattr(scheduler, "scale_model_input"),
(
f"{scheduler_class} does not implement a required class method `scale_model_input(sample,"
" timestep)`"
),
)
self.assertTrue(
hasattr(scheduler, "step"),
f"{scheduler_class} does not implement a required class method `step(...)`",
)
if scheduler_class != VQDiffusionScheduler:
sample = self.dummy_sample
scaled_sample = scheduler.scale_model_input(sample, 0.0)
self.assertEqual(sample.shape, scaled_sample.shape)
def test_add_noise_device(self):
for scheduler_class in self.scheduler_classes:
if scheduler_class == IPNDMScheduler:
continue
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
scheduler.set_timesteps(100)
sample = self.dummy_sample.to(torch_device)
scaled_sample = scheduler.scale_model_input(sample, 0.0)
self.assertEqual(sample.shape, scaled_sample.shape)
noise = torch.randn_like(scaled_sample).to(torch_device)
t = scheduler.timesteps[5][None]
noised = scheduler.add_noise(scaled_sample, noise, t)
self.assertEqual(noised.shape, scaled_sample.shape)
def test_deprecated_kwargs(self):
for scheduler_class in self.scheduler_classes:
has_kwarg_in_model_class = "kwargs" in inspect.signature(scheduler_class.__init__).parameters
has_deprecated_kwarg = len(scheduler_class._deprecated_kwargs) > 0
if has_kwarg_in_model_class and not has_deprecated_kwarg:
raise ValueError(
f"{scheduler_class} has `**kwargs` in its __init__ method but has not defined any deprecated"
" kwargs under the `_deprecated_kwargs` class attribute. Make sure to either remove `**kwargs` if"
" there are no deprecated arguments or add the deprecated argument with `_deprecated_kwargs ="
" [<deprecated_argument>]`"
)
if not has_kwarg_in_model_class and has_deprecated_kwarg:
raise ValueError(
f"{scheduler_class} doesn't have `**kwargs` in its __init__ method but has defined deprecated"
" kwargs under the `_deprecated_kwargs` class attribute. Make sure to either add the `**kwargs`"
f" argument to {self.model_class}.__init__ if there are deprecated arguments or remove the"
" deprecated argument from `_deprecated_kwargs = [<deprecated_argument>]`"
)
def test_trained_betas(self):
for scheduler_class in self.scheduler_classes:
if scheduler_class == VQDiffusionScheduler:
continue
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config, trained_betas=np.array([0.0, 0.1]))
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_pretrained(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
assert scheduler.betas.tolist() == new_scheduler.betas.tolist()
class DDPMSchedulerTest(SchedulerCommonTest):
scheduler_classes = (DDPMScheduler,)
def get_scheduler_config(self, **kwargs):
config = {
"num_train_timesteps": 1000,
"beta_start": 0.0001,
"beta_end": 0.02,
"beta_schedule": "linear",
"variance_type": "fixed_small",
"clip_sample": True,
}
config.update(**kwargs)
return config
def test_timesteps(self):
for timesteps in [1, 5, 100, 1000]:
self.check_over_configs(num_train_timesteps=timesteps)
def test_betas(self):
for beta_start, beta_end in zip([0.0001, 0.001, 0.01, 0.1], [0.002, 0.02, 0.2, 2]):
self.check_over_configs(beta_start=beta_start, beta_end=beta_end)
def test_schedules(self):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=schedule)
def test_variance_type(self):
for variance in ["fixed_small", "fixed_large", "other"]:
self.check_over_configs(variance_type=variance)
def test_clip_sample(self):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=clip_sample)
def test_prediction_type(self):
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(prediction_type=prediction_type)
def test_time_indices(self):
for t in [0, 500, 999]:
self.check_over_forward(time_step=t)
def test_variance(self):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
assert torch.sum(torch.abs(scheduler._get_variance(0) - 0.0)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(487) - 0.00979)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(999) - 0.02)) < 1e-5
def test_full_loop_no_noise(self):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
num_trained_timesteps = len(scheduler)
model = self.dummy_model()
sample = self.dummy_sample_deter
generator = torch.manual_seed(0)
for t in reversed(range(num_trained_timesteps)):
# 1. predict noise residual
residual = model(sample, t)
# 2. predict previous mean of sample x_t-1
pred_prev_sample = scheduler.step(residual, t, sample, generator=generator).prev_sample
# if t > 0:
# noise = self.dummy_sample_deter
# variance = scheduler.get_variance(t) ** (0.5) * noise
#
# sample = pred_prev_sample + variance
sample = pred_prev_sample
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 258.9606) < 1e-2
assert abs(result_mean.item() - 0.3372) < 1e-3
def test_full_loop_with_v_prediction(self):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config(prediction_type="v_prediction")
scheduler = scheduler_class(**scheduler_config)
num_trained_timesteps = len(scheduler)
model = self.dummy_model()
sample = self.dummy_sample_deter
generator = torch.manual_seed(0)
for t in reversed(range(num_trained_timesteps)):
# 1. predict noise residual
residual = model(sample, t)
# 2. predict previous mean of sample x_t-1
pred_prev_sample = scheduler.step(residual, t, sample, generator=generator).prev_sample
# if t > 0:
# noise = self.dummy_sample_deter
# variance = scheduler.get_variance(t) ** (0.5) * noise
#
# sample = pred_prev_sample + variance
sample = pred_prev_sample
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 202.0296) < 1e-2
assert abs(result_mean.item() - 0.2631) < 1e-3
class DDIMSchedulerTest(SchedulerCommonTest):
scheduler_classes = (DDIMScheduler,)
forward_default_kwargs = (("eta", 0.0), ("num_inference_steps", 50))
def get_scheduler_config(self, **kwargs):
config = {
"num_train_timesteps": 1000,
"beta_start": 0.0001,
"beta_end": 0.02,
"beta_schedule": "linear",
"clip_sample": True,
}
config.update(**kwargs)
return config
def full_loop(self, **config):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config(**config)
scheduler = scheduler_class(**scheduler_config)
num_inference_steps, eta = 10, 0.0
model = self.dummy_model()
sample = self.dummy_sample_deter
scheduler.set_timesteps(num_inference_steps)
for t in scheduler.timesteps:
residual = model(sample, t)
sample = scheduler.step(residual, t, sample, eta).prev_sample
return sample
def test_timesteps(self):
for timesteps in [100, 500, 1000]:
self.check_over_configs(num_train_timesteps=timesteps)
def test_steps_offset(self):
for steps_offset in [0, 1]:
self.check_over_configs(steps_offset=steps_offset)
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config(steps_offset=1)
scheduler = scheduler_class(**scheduler_config)
scheduler.set_timesteps(5)
assert torch.equal(scheduler.timesteps, torch.LongTensor([801, 601, 401, 201, 1]))
def test_betas(self):
for beta_start, beta_end in zip([0.0001, 0.001, 0.01, 0.1], [0.002, 0.02, 0.2, 2]):
self.check_over_configs(beta_start=beta_start, beta_end=beta_end)
def test_schedules(self):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=schedule)
def test_prediction_type(self):
for prediction_type in ["epsilon", "v_prediction"]:
self.check_over_configs(prediction_type=prediction_type)
def test_clip_sample(self):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=clip_sample)
def test_time_indices(self):
for t in [1, 10, 49]:
self.check_over_forward(time_step=t)
def test_inference_steps(self):
for t, num_inference_steps in zip([1, 10, 50], [10, 50, 500]):
self.check_over_forward(time_step=t, num_inference_steps=num_inference_steps)
def test_eta(self):
for t, eta in zip([1, 10, 49], [0.0, 0.5, 1.0]):
self.check_over_forward(time_step=t, eta=eta)
def test_variance(self):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
assert torch.sum(torch.abs(scheduler._get_variance(0, 0) - 0.0)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(420, 400) - 0.14771)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(980, 960) - 0.32460)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(0, 0) - 0.0)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(487, 486) - 0.00979)) < 1e-5
assert torch.sum(torch.abs(scheduler._get_variance(999, 998) - 0.02)) < 1e-5
def test_full_loop_no_noise(self):
sample = self.full_loop()
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 172.0067) < 1e-2
assert abs(result_mean.item() - 0.223967) < 1e-3
def test_full_loop_with_v_prediction(self):
sample = self.full_loop(prediction_type="v_prediction")
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 52.5302) < 1e-2
assert abs(result_mean.item() - 0.0684) < 1e-3
def test_full_loop_with_set_alpha_to_one(self):
# We specify different beta, so that the first alpha is 0.99
sample = self.full_loop(set_alpha_to_one=True, beta_start=0.01)
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 149.8295) < 1e-2
assert abs(result_mean.item() - 0.1951) < 1e-3
def test_full_loop_with_no_set_alpha_to_one(self):
# We specify different beta, so that the first alpha is 0.99
sample = self.full_loop(set_alpha_to_one=False, beta_start=0.01)
result_sum = torch.sum(torch.abs(sample))
result_mean = torch.mean(torch.abs(sample))
assert abs(result_sum.item() - 149.0784) < 1e-2
assert abs(result_mean.item() - 0.1941) < 1e-3
class DPMSolverSinglestepSchedulerTest(SchedulerCommonTest):
scheduler_classes = (DPMSolverSinglestepScheduler,)
forward_default_kwargs = (("num_inference_steps", 25),)
def get_scheduler_config(self, **kwargs):
config = {
"num_train_timesteps": 1000,
"beta_start": 0.0001,
"beta_end": 0.02,
"beta_schedule": "linear",
"solver_order": 2,
"prediction_type": "epsilon",
"thresholding": False,
"sample_max_value": 1.0,
"algorithm_type": "dpmsolver++",
"solver_type": "midpoint",
}
config.update(**kwargs)
return config
def check_over_configs(self, time_step=0, **config):
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
sample = self.dummy_sample
residual = 0.1 * sample
dummy_past_residuals = [residual + 0.2, residual + 0.15, residual + 0.10]
for scheduler_class in self.scheduler_classes:
scheduler_config = self.get_scheduler_config(**config)
scheduler = scheduler_class(**scheduler_config)
scheduler.set_timesteps(num_inference_steps)
# copy over dummy past residuals
scheduler.model_outputs = dummy_past_residuals[: scheduler.config.solver_order]
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_config(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
new_scheduler.set_timesteps(num_inference_steps)
# copy over dummy past residuals
new_scheduler.model_outputs = dummy_past_residuals[: new_scheduler.config.solver_order]
output, new_output = sample, sample
for t in range(time_step, time_step + scheduler.config.solver_order + 1):
output = scheduler.step(residual, t, output, **kwargs).prev_sample
new_output = new_scheduler.step(residual, t, new_output, **kwargs).prev_sample
assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical"
def test_from_save_pretrained(self):
pass
def check_over_forward(self, time_step=0, **forward_kwargs):
kwargs = dict(self.forward_default_kwargs)
num_inference_steps = kwargs.pop("num_inference_steps", None)
sample = self.dummy_sample
residual = 0.1 * sample
dummy_past_residuals = [residual + 0.2, residual + 0.15, residual + 0.10]
for scheduler_class in self.scheduler_classes:
scheduler_config = self.get_scheduler_config()
scheduler = scheduler_class(**scheduler_config)
scheduler.set_timesteps(num_inference_steps)
# copy over dummy past residuals (must be after setting timesteps)
scheduler.model_outputs = dummy_past_residuals[: scheduler.config.solver_order]
with tempfile.TemporaryDirectory() as tmpdirname:
scheduler.save_config(tmpdirname)
new_scheduler = scheduler_class.from_pretrained(tmpdirname)
# copy over dummy past residuals
new_scheduler.set_timesteps(num_inference_steps)
# copy over dummy past residual (must be after setting timesteps)
new_scheduler.model_outputs = dummy_past_residuals[: new_scheduler.config.solver_order]
output = scheduler.step(residual, time_step, sample, **kwargs).prev_sample
new_output = new_scheduler.step(residual, time_step, sample, **kwargs).prev_sample
assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical"
def full_loop(self, **config):
scheduler_class = self.scheduler_classes[0]
scheduler_config = self.get_scheduler_config(**config)
scheduler = scheduler_class(**scheduler_config)
num_inference_steps = 10
model = self.dummy_model()
sample = self.dummy_sample_deter
scheduler.set_timesteps(num_inference_steps)
for i, t in enumerate(scheduler.timesteps):
residual = model(sample, t)
sample = scheduler.step(residual, t, sample).prev_sample
return sample
def test_timesteps(self):
for timesteps in [25, 50, 100, 999, 1000]:
self.check_over_configs(num_train_timesteps=timesteps)
def test_thresholding(self):
self.check_over_configs(thresholding=False)
for order in [1, 2, 3]:
for solver_type in ["midpoint", "heun"]:
for threshold in [0.5, 1.0, 2.0]:
for prediction_type in ["epsilon", "sample"]:
self.check_over_configs(
thresholding=True,
prediction_type=prediction_type,
sample_max_value=threshold,
algorithm_type="dpmsolver++",
solver_order=order,
solver_type=solver_type,
)
def test_prediction_type(self):
for prediction_type in ["epsilon", "v_prediction"]:
self.check_over_configs(prediction_type=prediction_type)
def test_solver_order_and_type(self):
for algorithm_type in ["dpmsolver", "dpmsolver++"]:
for solver_type in ["midpoint", "heun"]:
for order in [1, 2, 3]:
for prediction_type in ["epsilon", "sample"]:
self.check_over_configs(
solver_order=order,
solver_type=solver_type,
prediction_type=prediction_type,
algorithm_type=algorithm_type,
)
sample = self.full_loop(
solver_order=order,
solver_type=solver_type,
prediction_type=prediction_type,
algorithm_type=algorithm_type,
)
assert not torch.isnan(sample).any(), "Samples have nan numbers"
def test_lower_order_final(self):
self.check_over_configs(lower_order_final=True)
self.check_over_configs(lower_order_final=False)
def test_inference_steps(self):
for num_inference_steps in [1, 2, 3, 5, 10, 50, 100, 999, 1000]:
self.check_over_forward(num_inference_steps=num_inference_steps, time_step=0)
def test_full_loop_no_noise(self):
sample = self.full_loop()
result_mean = torch.mean(torch.abs(sample))
assert abs(result_mean.item() - 0.2791) < 1e-3
You can’t perform that action at this time.