FazBrowse GitHub Viewer
|
Trending
|
URL:
|
Home
Tools:
[Download Repo ZIP]
[View Raw Code]
[Original HTTPS Page]
diffusers_plus_plus/tests/lora/test_lora_layers_sdxl.py at utils · ModelsLab/diffusers_plus_plus · GitHub
Uh oh!
There was an error while loading.
Please reload this page
.
ModelsLab
/
diffusers_plus_plus
Public
forked from
huggingface/diffusers
Notifications
You must be signed in to change notification settings
Fork
2
Star
14
Code
Issues
13
Pull requests
8
Discussions
Actions
Projects
Security and quality
0
Insights
Additional navigation options
Code
Issues
Pull requests
Discussions
Actions
Projects
Security and quality
Insights
Expand file tree
Breadcrumbs
diffusers_plus_plus
/
tests
/
lora
/
test_lora_layers_sdxl.py
Copy path
More file actions
More file actions
Latest commit
History
History
History
689 lines (543 loc) · 26.3 KB
Breadcrumbs
diffusers_plus_plus
/
tests
/
lora
/
test_lora_layers_sdxl.py
Copy path
File metadata and controls
689 lines (543 loc) · 26.3 KB
Raw
Copy raw file
Download raw file
Open symbols panel
Edit and raw actions
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
# coding=utf-8
# Copyright 2024 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
copy
import
gc
import
importlib
import
sys
import
time
import
unittest
import
numpy
as
np
import
torch
from
packaging
import
version
from
diffusers
import
(
ControlNetModel
,
EulerDiscreteScheduler
,
LCMScheduler
,
StableDiffusionXLAdapterPipeline
,
StableDiffusionXLControlNetPipeline
,
StableDiffusionXLPipeline
,
T2IAdapter
,
)
from
diffusers
.
utils
.
import_utils
import
is_accelerate_available
from
diffusers
.
utils
.
testing_utils
import
(
load_image
,
nightly
,
numpy_cosine_similarity_distance
,
require_peft_backend
,
require_torch_gpu
,
slow
,
torch_device
,
)
sys
.
path
.
append
(
"."
)
from
utils
import
PeftLoraLoaderMixinTests
,
check_if_lora_correctly_set
,
state_dicts_almost_equal
# noqa: E402
if
is_accelerate_available
():
from
accelerate
.
utils
import
release_memory
class
StableDiffusionXLLoRATests
(
PeftLoraLoaderMixinTests
,
unittest
.
TestCase
):
has_two_text_encoders
=
True
pipeline_class
=
StableDiffusionXLPipeline
scheduler_cls
=
EulerDiscreteScheduler
scheduler_kwargs
=
{
"beta_start"
:
0.00085
,
"beta_end"
:
0.012
,
"beta_schedule"
:
"scaled_linear"
,
"timestep_spacing"
:
"leading"
,
"steps_offset"
:
1
,
}
unet_kwargs
=
{
"block_out_channels"
: (
32
,
64
),
"layers_per_block"
:
2
,
"sample_size"
:
32
,
"in_channels"
:
4
,
"out_channels"
:
4
,
"down_block_types"
: (
"DownBlock2D"
,
"CrossAttnDownBlock2D"
),
"up_block_types"
: (
"CrossAttnUpBlock2D"
,
"UpBlock2D"
),
"attention_head_dim"
: (
2
,
4
),
"use_linear_projection"
:
True
,
"addition_embed_type"
:
"text_time"
,
"addition_time_embed_dim"
:
8
,
"transformer_layers_per_block"
: (
1
,
2
),
"projection_class_embeddings_input_dim"
:
80
,
# 6 * 8 + 32
"cross_attention_dim"
:
64
,
}
vae_kwargs
=
{
"block_out_channels"
: [
32
,
64
],
"in_channels"
:
3
,
"out_channels"
:
3
,
"down_block_types"
: [
"DownEncoderBlock2D"
,
"DownEncoderBlock2D"
],
"up_block_types"
: [
"UpDecoderBlock2D"
,
"UpDecoderBlock2D"
],
"latent_channels"
:
4
,
"sample_size"
:
128
,
}
def
setUp
(
self
):
super
().
setUp
()
gc
.
collect
()
torch
.
cuda
.
empty_cache
()
def
tearDown
(
self
):
super
().
tearDown
()
gc
.
collect
()
torch
.
cuda
.
empty_cache
()
@
slow
@
require_torch_gpu
@
require_peft_backend
class
LoraSDXLIntegrationTests
(
unittest
.
TestCase
):
def
setUp
(
self
):
super
().
setUp
()
gc
.
collect
()
torch
.
cuda
.
empty_cache
()
def
tearDown
(
self
):
super
().
tearDown
()
gc
.
collect
()
torch
.
cuda
.
empty_cache
()
def
test_sdxl_0_9_lora_one
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-0.9"
)
lora_model_id
=
"hf-internal-testing/sdxl-0.9-daiton-lora"
lora_filename
=
"daiton-xl-lora-test.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
enable_model_cpu_offload
()
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.3838
,
0.3482
,
0.3588
,
0.3162
,
0.319
,
0.3369
,
0.338
,
0.3366
,
0.3213
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_0_9_lora_two
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-0.9"
)
lora_model_id
=
"hf-internal-testing/sdxl-0.9-costumes-lora"
lora_filename
=
"saijo.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
enable_model_cpu_offload
()
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.3137
,
0.3269
,
0.3355
,
0.255
,
0.2577
,
0.2563
,
0.2679
,
0.2758
,
0.2626
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_0_9_lora_three
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-0.9"
)
lora_model_id
=
"hf-internal-testing/sdxl-0.9-kamepan-lora"
lora_filename
=
"kame_sdxl_v2-000020-16rank.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
enable_model_cpu_offload
()
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.4015
,
0.3761
,
0.3616
,
0.3745
,
0.3462
,
0.3337
,
0.3564
,
0.3649
,
0.3468
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
5e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_1_0_lora
(
self
):
generator
=
torch
.
Generator
(
"cpu"
).
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipe
.
enable_model_cpu_offload
()
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.4468
,
0.4061
,
0.4134
,
0.3637
,
0.3202
,
0.365
,
0.3786
,
0.3725
,
0.3535
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-4
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_1_0_blockwise_lora
(
self
):
generator
=
torch
.
Generator
(
"cpu"
).
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipe
.
enable_model_cpu_offload
()
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
,
adapter_name
=
"offset"
)
scales
=
{
"unet"
: {
"down"
: {
"block_1"
: [
1.0
,
1.0
],
"block_2"
: [
1.0
,
1.0
]},
"mid"
:
1.0
,
"up"
: {
"block_0"
: [
1.0
,
1.0
,
1.0
],
"block_1"
: [
1.0
,
1.0
,
1.0
]},
},
}
pipe
.
set_adapters
([
"offset"
], [
scales
])
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
00.4468
,
0.4061
,
0.4134
,
0.3637
,
0.3202
,
0.365
,
0.3786
,
0.3725
,
0.3535
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-4
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_lcm_lora
(
self
):
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
torch_dtype
=
torch
.
float16
)
pipe
.
scheduler
=
LCMScheduler
.
from_config
(
pipe
.
scheduler
.
config
)
pipe
.
enable_model_cpu_offload
()
generator
=
torch
.
Generator
(
"cpu"
).
manual_seed
(
0
)
lora_model_id
=
"latent-consistency/lcm-lora-sdxl"
pipe
.
load_lora_weights
(
lora_model_id
)
image
=
pipe
(
"masterpiece, best quality, mountain"
,
generator
=
generator
,
num_inference_steps
=
4
,
guidance_scale
=
0.5
).
images
[
0
]
expected_image
=
load_image
(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/lcm_lora/sdxl_lcm_lora.png"
)
image_np
=
pipe
.
image_processor
.
pil_to_numpy
(
image
)
expected_image_np
=
pipe
.
image_processor
.
pil_to_numpy
(
expected_image
)
max_diff
=
numpy_cosine_similarity_distance
(
image_np
.
flatten
(),
expected_image_np
.
flatten
())
assert
max_diff
<
1e-4
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_1_0_lora_fusion
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
fuse_lora
()
# We need to unload the lora weights since in the previous API `fuse_lora` led to lora weights being
# silently deleted - otherwise this will CPU OOM
pipe
.
unload_lora_weights
()
pipe
.
enable_model_cpu_offload
()
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
# This way we also test equivalence between LoRA fusion and the non-fusion behaviour.
expected
=
np
.
array
([
0.4468
,
0.4061
,
0.4134
,
0.3637
,
0.3202
,
0.365
,
0.3786
,
0.3725
,
0.3535
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-4
release_memory
(
pipe
)
def
test_sdxl_1_0_lora_unfusion
(
self
):
generator
=
torch
.
Generator
(
"cpu"
).
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
fuse_lora
()
pipe
.
enable_model_cpu_offload
()
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
3
).
images
images_with_fusion
=
images
.
flatten
()
pipe
.
unfuse_lora
()
generator
=
torch
.
Generator
(
"cpu"
).
manual_seed
(
0
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
3
).
images
images_without_fusion
=
images
.
flatten
()
max_diff
=
numpy_cosine_similarity_distance
(
images_with_fusion
,
images_without_fusion
)
assert
max_diff
<
1e-4
release_memory
(
pipe
)
def
test_sdxl_1_0_lora_unfusion_effectivity
(
self
):
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipe
.
enable_model_cpu_offload
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
original_image_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
pipe
.
fuse_lora
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
_
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
pipe
.
unfuse_lora
()
# We need to unload the lora weights - in the old API unfuse led to unloading the adapter weights
pipe
.
unload_lora_weights
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images_without_fusion_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
max_diff
=
numpy_cosine_similarity_distance
(
images_without_fusion_slice
,
original_image_slice
)
assert
max_diff
<
1e-3
release_memory
(
pipe
)
def
test_sdxl_1_0_lora_fusion_efficiency
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
torch_dtype
=
torch
.
float16
)
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
,
torch_dtype
=
torch
.
float16
)
pipe
.
enable_model_cpu_offload
()
start_time
=
time
.
time
()
for
_
in
range
(
3
):
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
end_time
=
time
.
time
()
elapsed_time_non_fusion
=
end_time
-
start_time
del
pipe
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
torch_dtype
=
torch
.
float16
)
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
,
torch_dtype
=
torch
.
float16
)
pipe
.
fuse_lora
()
# We need to unload the lora weights since in the previous API `fuse_lora` led to lora weights being
# silently deleted - otherwise this will CPU OOM
pipe
.
unload_lora_weights
()
pipe
.
enable_model_cpu_offload
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
start_time
=
time
.
time
()
for
_
in
range
(
3
):
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
end_time
=
time
.
time
()
elapsed_time_fusion
=
end_time
-
start_time
self
.
assertTrue
(
elapsed_time_fusion
<
elapsed_time_non_fusion
)
release_memory
(
pipe
)
def
test_sdxl_1_0_last_ben
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipe
.
enable_model_cpu_offload
()
lora_model_id
=
"TheLastBen/Papercut_SDXL"
lora_filename
=
"papercut.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
images
=
pipe
(
"papercut.safetensors"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.5244
,
0.4347
,
0.4312
,
0.4246
,
0.4398
,
0.4409
,
0.4884
,
0.4938
,
0.4094
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_1_0_fuse_unfuse_all
(
self
):
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
torch_dtype
=
torch
.
float16
)
text_encoder_1_sd
=
copy
.
deepcopy
(
pipe
.
text_encoder
.
state_dict
())
text_encoder_2_sd
=
copy
.
deepcopy
(
pipe
.
text_encoder_2
.
state_dict
())
unet_sd
=
copy
.
deepcopy
(
pipe
.
unet
.
state_dict
())
pipe
.
load_lora_weights
(
"davizca87/sun-flower"
,
weight_name
=
"snfw3rXL-000004.safetensors"
,
torch_dtype
=
torch
.
float16
)
fused_te_state_dict
=
pipe
.
text_encoder
.
state_dict
()
fused_te_2_state_dict
=
pipe
.
text_encoder_2
.
state_dict
()
unet_state_dict
=
pipe
.
unet
.
state_dict
()
peft_ge_070
=
version
.
parse
(
importlib
.
metadata
.
version
(
"peft"
))
>=
version
.
parse
(
"0.7.0"
)
def
remap_key
(
key
,
sd
):
# some keys have moved around for PEFT >= 0.7.0, but they should still be loaded correctly
if
(
key
in
sd
)
or
(
not
peft_ge_070
):
return
key
# instead of linear.weight, we now have linear.base_layer.weight, etc.
if
key
.
endswith
(
".weight"
):
key
=
key
[:
-
7
]
+
".base_layer.weight"
elif
key
.
endswith
(
".bias"
):
key
=
key
[:
-
5
]
+
".base_layer.bias"
return
key
for
key
,
value
in
text_encoder_1_sd
.
items
():
key
=
remap_key
(
key
,
fused_te_state_dict
)
self
.
assertTrue
(
torch
.
allclose
(
fused_te_state_dict
[
key
],
value
))
for
key
,
value
in
text_encoder_2_sd
.
items
():
key
=
remap_key
(
key
,
fused_te_2_state_dict
)
self
.
assertTrue
(
torch
.
allclose
(
fused_te_2_state_dict
[
key
],
value
))
for
key
,
value
in
unet_state_dict
.
items
():
self
.
assertTrue
(
torch
.
allclose
(
unet_state_dict
[
key
],
value
))
pipe
.
fuse_lora
()
pipe
.
unload_lora_weights
()
assert
not
state_dicts_almost_equal
(
text_encoder_1_sd
,
pipe
.
text_encoder
.
state_dict
())
assert
not
state_dicts_almost_equal
(
text_encoder_2_sd
,
pipe
.
text_encoder_2
.
state_dict
())
assert
not
state_dicts_almost_equal
(
unet_sd
,
pipe
.
unet
.
state_dict
())
release_memory
(
pipe
)
del
unet_sd
,
text_encoder_1_sd
,
text_encoder_2_sd
def
test_sdxl_1_0_lora_with_sequential_cpu_offloading
(
self
):
generator
=
torch
.
Generator
().
manual_seed
(
0
)
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipe
.
enable_sequential_cpu_offload
()
lora_model_id
=
"hf-internal-testing/sdxl-1.0-lora"
lora_filename
=
"sd_xl_offset_example-lora_1.0.safetensors"
pipe
.
load_lora_weights
(
lora_model_id
,
weight_name
=
lora_filename
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
images
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected
=
np
.
array
([
0.4468
,
0.4087
,
0.4134
,
0.366
,
0.3202
,
0.3505
,
0.3786
,
0.387
,
0.3535
])
max_diff
=
numpy_cosine_similarity_distance
(
expected
,
images
)
assert
max_diff
<
1e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_controlnet_canny_lora
(
self
):
controlnet
=
ControlNetModel
.
from_pretrained
(
"diffusers/controlnet-canny-sdxl-1.0"
)
pipe
=
StableDiffusionXLControlNetPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
controlnet
=
controlnet
)
pipe
.
load_lora_weights
(
"nerijs/pixel-art-xl"
,
weight_name
=
"pixel-art-xl.safetensors"
)
pipe
.
enable_sequential_cpu_offload
()
generator
=
torch
.
Generator
(
device
=
"cpu"
).
manual_seed
(
0
)
prompt
=
"corgi"
image
=
load_image
(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png"
)
images
=
pipe
(
prompt
,
image
=
image
,
generator
=
generator
,
output_type
=
"np"
,
num_inference_steps
=
3
).
images
assert
images
[
0
].
shape
==
(
768
,
512
,
3
)
original_image
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_image
=
np
.
array
([
0.4574
,
0.4487
,
0.4435
,
0.5163
,
0.4396
,
0.4411
,
0.518
,
0.4465
,
0.4333
])
max_diff
=
numpy_cosine_similarity_distance
(
expected_image
,
original_image
)
assert
max_diff
<
1e-4
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
def
test_sdxl_t2i_adapter_canny_lora
(
self
):
adapter
=
T2IAdapter
.
from_pretrained
(
"TencentARC/t2i-adapter-lineart-sdxl-1.0"
,
torch_dtype
=
torch
.
float16
).
to
(
"cpu"
)
pipe
=
StableDiffusionXLAdapterPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
adapter
=
adapter
,
torch_dtype
=
torch
.
float16
,
variant
=
"fp16"
,
)
pipe
.
load_lora_weights
(
"CiroN2022/toy-face"
,
weight_name
=
"toy_face_sdxl.safetensors"
)
pipe
.
enable_model_cpu_offload
()
pipe
.
set_progress_bar_config
(
disable
=
None
)
generator
=
torch
.
Generator
(
device
=
"cpu"
).
manual_seed
(
0
)
prompt
=
"toy"
image
=
load_image
(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/t2i_adapter/toy_canny.png"
)
images
=
pipe
(
prompt
,
image
=
image
,
generator
=
generator
,
output_type
=
"np"
,
num_inference_steps
=
3
).
images
assert
images
[
0
].
shape
==
(
768
,
512
,
3
)
image_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_slice
=
np
.
array
([
0.4284
,
0.4337
,
0.4319
,
0.4255
,
0.4329
,
0.4280
,
0.4338
,
0.4420
,
0.4226
])
assert
numpy_cosine_similarity_distance
(
image_slice
,
expected_slice
)
<
1e-4
@
nightly
def
test_sequential_fuse_unfuse
(
self
):
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
,
torch_dtype
=
torch
.
float16
)
# 1. round
pipe
.
load_lora_weights
(
"Pclanglais/TintinIA"
,
torch_dtype
=
torch
.
float16
)
pipe
.
to
(
torch_device
)
pipe
.
fuse_lora
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
images
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
image_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
pipe
.
unfuse_lora
()
# 2. round
pipe
.
load_lora_weights
(
"ProomptEngineer/pe-balloon-diffusion-style"
,
torch_dtype
=
torch
.
float16
)
pipe
.
fuse_lora
()
pipe
.
unfuse_lora
()
# 3. round
pipe
.
load_lora_weights
(
"ostris/crayon_style_lora_sdxl"
,
torch_dtype
=
torch
.
float16
)
pipe
.
fuse_lora
()
pipe
.
unfuse_lora
()
# 4. back to 1st round
pipe
.
load_lora_weights
(
"Pclanglais/TintinIA"
,
torch_dtype
=
torch
.
float16
)
pipe
.
fuse_lora
()
generator
=
torch
.
Generator
().
manual_seed
(
0
)
images_2
=
pipe
(
"masterpiece, best quality, mountain"
,
output_type
=
"np"
,
generator
=
generator
,
num_inference_steps
=
2
).
images
image_slice_2
=
images_2
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
max_diff
=
numpy_cosine_similarity_distance
(
image_slice
,
image_slice_2
)
assert
max_diff
<
1e-3
pipe
.
unload_lora_weights
()
release_memory
(
pipe
)
@
nightly
def
test_integration_logits_multi_adapter
(
self
):
path
=
"stabilityai/stable-diffusion-xl-base-1.0"
lora_id
=
"CiroN2022/toy-face"
pipe
=
StableDiffusionXLPipeline
.
from_pretrained
(
path
,
torch_dtype
=
torch
.
float16
)
pipe
.
load_lora_weights
(
lora_id
,
weight_name
=
"toy_face_sdxl.safetensors"
,
adapter_name
=
"toy"
)
pipe
=
pipe
.
to
(
torch_device
)
self
.
assertTrue
(
check_if_lora_correctly_set
(
pipe
.
unet
),
"Lora not correctly set in Unet"
)
prompt
=
"toy_face of a hacker with a hoodie"
lora_scale
=
0.9
images
=
pipe
(
prompt
=
prompt
,
num_inference_steps
=
30
,
generator
=
torch
.
manual_seed
(
0
),
cross_attention_kwargs
=
{
"scale"
:
lora_scale
},
output_type
=
"np"
,
).
images
expected_slice_scale
=
np
.
array
([
0.538
,
0.539
,
0.540
,
0.540
,
0.542
,
0.539
,
0.538
,
0.541
,
0.539
])
predicted_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
max_diff
=
numpy_cosine_similarity_distance
(
expected_slice_scale
,
predicted_slice
)
assert
max_diff
<
1e-3
pipe
.
load_lora_weights
(
"nerijs/pixel-art-xl"
,
weight_name
=
"pixel-art-xl.safetensors"
,
adapter_name
=
"pixel"
)
pipe
.
set_adapters
(
"pixel"
)
prompt
=
"pixel art, a hacker with a hoodie, simple, flat colors"
images
=
pipe
(
prompt
,
num_inference_steps
=
30
,
guidance_scale
=
7.5
,
cross_attention_kwargs
=
{
"scale"
:
lora_scale
},
generator
=
torch
.
manual_seed
(
0
),
output_type
=
"np"
,
).
images
predicted_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_slice_scale
=
np
.
array
(
[
0.61973065
,
0.62018543
,
0.62181497
,
0.61933696
,
0.6208608
,
0.620576
,
0.6200281
,
0.62258327
,
0.6259889
]
)
max_diff
=
numpy_cosine_similarity_distance
(
expected_slice_scale
,
predicted_slice
)
assert
max_diff
<
1e-3
# multi-adapter inference
pipe
.
set_adapters
([
"pixel"
,
"toy"
],
adapter_weights
=
[
0.5
,
1.0
])
images
=
pipe
(
prompt
,
num_inference_steps
=
30
,
guidance_scale
=
7.5
,
cross_attention_kwargs
=
{
"scale"
:
1.0
},
generator
=
torch
.
manual_seed
(
0
),
output_type
=
"np"
,
).
images
predicted_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_slice_scale
=
np
.
array
([
0.5888
,
0.5897
,
0.5946
,
0.5888
,
0.5935
,
0.5946
,
0.5857
,
0.5891
,
0.5909
])
max_diff
=
numpy_cosine_similarity_distance
(
expected_slice_scale
,
predicted_slice
)
assert
max_diff
<
1e-3
# Lora disabled
pipe
.
disable_lora
()
images
=
pipe
(
prompt
,
num_inference_steps
=
30
,
guidance_scale
=
7.5
,
cross_attention_kwargs
=
{
"scale"
:
lora_scale
},
generator
=
torch
.
manual_seed
(
0
),
output_type
=
"np"
,
).
images
predicted_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_slice_scale
=
np
.
array
([
0.5456
,
0.5466
,
0.5487
,
0.5458
,
0.5469
,
0.5454
,
0.5446
,
0.5479
,
0.5487
])
max_diff
=
numpy_cosine_similarity_distance
(
expected_slice_scale
,
predicted_slice
)
assert
max_diff
<
1e-3
@
nightly
def
test_integration_logits_for_dora_lora
(
self
):
pipeline
=
StableDiffusionXLPipeline
.
from_pretrained
(
"stabilityai/stable-diffusion-xl-base-1.0"
)
pipeline
.
load_lora_weights
(
"hf-internal-testing/dora-trained-on-kohya"
)
pipeline
.
enable_model_cpu_offload
()
images
=
pipeline
(
"photo of ohwx dog"
,
num_inference_steps
=
10
,
generator
=
torch
.
manual_seed
(
0
),
output_type
=
"np"
,
).
images
predicted_slice
=
images
[
0
,
-
3
:,
-
3
:,
-
1
].
flatten
()
expected_slice_scale
=
np
.
array
([
0.3932
,
0.3742
,
0.4429
,
0.3737
,
0.3504
,
0.433
,
0.3948
,
0.3769
,
0.4516
])
max_diff
=
numpy_cosine_similarity_distance
(
expected_slice_scale
,
predicted_slice
)
assert
max_diff
<
1e-3
Back
|
FazBrowse Home
|
New Git URL