How to get the resulting model/MBW? #108

miasik · 2024-01-25T09:15:09Z

miasik
Jan 25, 2024

It might be related to #103
My settings for the run:
As I understand my setting below are almost default

miasik · 2024-01-25T09:19:20Z

miasik
Jan 25, 2024
Author

I often see that AutoMerger uses 0.5 again and again:

Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5).safetensors [c7585966bf]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 0.9s (create model: 0.4s, apply weights to model: 0.4s).
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.09it/s]
Total progress: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.98it/s]
 - Overriden seed = 67██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.19it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  e88eb73e29589e848fa45daadac3a6935b501f263190ebde4e80a84d8347c467
  - mm_use [True, False, False, False]
  - model_a My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - base_model My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - max_models 4
  - models ['Cartoons\\fluffity_v23.safetensors [27a31020bf]']
  - modes ['Add-Diff']
  - calcmodes ['Rebasin']
  - usembws [['BASE', 'INP*', 'M00', 'OUT*', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']]
  - weights ['0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5']
  - alpha [0.5]
  - adjust
  - use elemental [False]
  - elementals ['']
  - Parse elemental merge...
model_a = My_YACAM-SR-16286.fp16
Loading from file E:\SD\automatic1111\models\Stable-diffusion\My\YACAM-SR-16286.fp16.safetensors...
isxl = False , sd2 = False
check affected permutation blocks by rebasin merge...
compact_mode =  True
 - check possible UNet partial update...
 - partial changed blocks =  ['BASE', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'M00', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']
 - UNet partial update mode
 - Dynamic loading rebasin module...
Rebasin mode
 - Calulation device for Rebasin is  cuda
 - LAP library is lap
Loading model Cartoons_fluffity_v23...
Loading from file E:\SD\automatic1111\models\Stable-diffusion\Cartoons\fluffity_v23.safetensors...
mode = Add-Diff, mbw mode, alpha = [0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5]
Stage #1/3: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:12<00:00, 69.85it/s]
Check uninitialized #2/3: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:00<00:00, 219657.46it/s]
Rebasin calc...ized #2/3:   0%|                                                                                                                                             | 0/876 [00:00<?, ?it/s]
maximize weight matching using scipy linear_sum_assignment...
P_bg324: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  7.83it/s]
P_bg324: new - old = 0.0

0 replies

miasik · 2024-01-25T09:33:19Z

miasik
Jan 25, 2024
Author

Finally I have something in the log:

Results: 'hyper_score'
   Best score: 0.7357643595849361
   Best parameter set:
      'model_b.BASE'  : 0.3
      'model_b.IN00'  : 0.49
      'model_b.IN01'  : 0.3
      'model_b.IN02'  : 0.3
      'model_b.IN03'  : 0.3
      'model_b.IN04'  : 0.3
      'model_b.IN05'  : 0.3
      'model_b.IN06'  : 0.49
      'model_b.IN07'  : 0.3
      'model_b.IN08'  : 0.49
      'model_b.IN09'  : 0.49
      'model_b.IN10'  : 0.49
      'model_b.IN11'  : 0.3
      'model_b.M00'   : 0.3
      'model_b.OUT00' : 0.49
      'model_b.OUT01' : 0.49
      'model_b.OUT02' : 0.3
      'model_b.OUT03' : 0.49
      'model_b.OUT04' : 0.49
      'model_b.OUT05' : 0.49
      'model_b.OUT06' : 0.3
      'model_b.OUT07' : 0.3
      'model_b.OUT08' : 0.3
      'model_b.OUT09' : 0.49
      'model_b.OUT10' : 0.3
      'model_b.OUT11' : 0.3
   Best iteration: 0

   Random seed: 1947387721

   Evaluation time   : 652.0190331935883 sec    [100.0 %]
   Optimization time : 0.001996755599975586 sec    [0.0 %]
   Iteration time    : 652.0210299491882 sec    [2.61 sec/iter]

Is it the result? Should I manually copy and paste MBW from the log to the main CMM window?
I'd like to try mergin using 0.1 as minimal value 0.9 as maximum and 0.25 as steps size. Is there a way to set min, max and step values for AutoHelper?
All pictures rendered during an AutoHelper run are slightly different(obviously because #103 has been fixed), but have the same in image info:

ModelMixer simple mbw b: "BASE,INP*,M00,OUT*,IN00,IN01,IN02,IN03,IN04,IN05,IN06,IN07,IN08,IN09,IN10,IN11,OUT00,OUT01,OUT02,OUT03,OUT04,OUT05,OUT06,OUT07,OUT08,OUT09,OUT10,OUT11", ModelMixer mbw weights b: "0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5",

Is there a way to save the full log?

0 replies

wkpark · 2024-01-25T10:38:08Z

wkpark
Jan 25, 2024
Maintainer

the last image you got has its configuration. and from the result image you can restore configuration using PNG Info tab or https://github.com/wkpark/sd-webui-pnginfo-diff (enhanced Png Info)
or Load Settings accordion (recently introduced by PR #106 )

0 replies

wkpark · 2024-01-25T10:40:29Z

wkpark
Jan 25, 2024
Maintainer

When you've selected any MBW blocks (simple mode or advanced mode MBWs), alpha value will be ignored, and MBW weights will be used.

1 reply

miasik Jan 25, 2024
Author

it looks promising, but something wrong again on my side. I left alpha as 0.5 (it's grayeg out) and set MBW as full 0.25.
Now I'm watching it again:

Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25).safetensors [f125e91796]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 1.0s (create model: 0.4s, apply weights to model: 0.4s).
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.90it/s]
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.07it/s]
0: 640x448 1 face, 9.0ms
Speed: 2.0ms preprocess, 9.0ms inference, 2.0ms postprocess per image at shape (1, 3, 640, 448)
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.12it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.13it/s]
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - mm_use [True, False, False, False]
  - model_a My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - base_model My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - max_models 4
  - models ['Cartoons\\fluffity_v23.safetensors [27a31020bf]']
  - modes ['Add-Diff']
  - calcmodes ['Rebasin']
  - usembws [['BASE', 'INP*', 'M00', 'OUT*', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']]
  - weights ['0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25']
  - alpha [0.5]
  - adjust
  - use elemental [False]
  - elementals ['']
  - Parse elemental merge...
model_a = My_YACAM-SR-16286.fp16
Loading from file E:\SD\automatic1111\models\Stable-diffusion\My\YACAM-SR-16286.fp16.safetensors...
isxl = False , sd2 = False
check affected permutation blocks by rebasin merge...
compact_mode =  True
 - check possible UNet partial update...
 - partial changed blocks =  ['BASE', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'M00', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10']
 - UNet partial update mode
 - Dynamic loading rebasin module...
Rebasin mode
 - Calulation device for Rebasin is  cuda
 - LAP library is lap
Loading model Cartoons_fluffity_v23...
Loading from file E:\SD\automatic1111\models\Stable-diffusion\Cartoons\fluffity_v23.safetensors...
mode = Add-Diff, mbw mode, alpha = [0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25]
Stage #1/3: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 838/838 [00:12<00:00, 65.41it/s]
Check uninitialized #2/3: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 838/838 [00:00<00:00, 210066.15it/s]
Rebasin calc...%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▌     | 806/838 [00:12<00:00, 121.51it/s]
maximize weight matching using scipy linear_sum_assignment...                                                                                                               | 0/838 [00:00<?, ?it/s]
P_bg371: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  7.91it/s]
P_bg371: new - old = 0.0

miasik · 2024-01-25T11:36:41Z

miasik
Jan 25, 2024
Author

A tail of my merging with 0.25 MWB. I see the best result and some actions after it. Is it OK?

Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25).safetensors [f125e91796]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 0.9s (create model: 0.4s, apply weights to model: 0.4s).
Total progress:   2%|██▌                                                                                                                                          | 21/1155 [00:25<23:21,  1.24s/it]
 - Overriden seed = 67█▌                                                                                                                                          | 21/1155 [00:25<04:42,  4.01it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - use current mixed model f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.96it/s]
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.16it/s]
0: 640x448 1 face, 9.0ms
Speed: 3.0ms preprocess, 9.0ms inference, 2.0ms postprocess per image at shape (1, 3, 640, 448)
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.08it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.17it/s]
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - use current mixed model f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
Total progress:   2%|██▌                                                                                                                                          | 21/1176 [00:08<08:13,  2.34it/s]
 - Overriden seed = 969613105                                                                                                                                     | 21/1176 [00:08<04:42,  4.08it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - use current mixed model f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.03it/s]
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  4.16it/s]
0: 640x448 1 face, 10.0ms
Speed: 2.0ms preprocess, 10.0ms inference, 2.0ms postprocess per image at shape (1, 3, 640, 448)
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.12it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  4.22it/s]
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - use current mixed model f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
Total progress:   2%|██▍                                                                                                                                          | 21/1197 [00:08<07:45,  2.52it/s]
 - score is 0.7292895009019714                                                                                                                                    | 21/1197 [00:08<04:47,  4.09it/s]
[0] hyper_score (Hill Climbing):   7%| | 18/250 [20:35<4:25:24, 68.64s/it, best_iter=0, best_pos=[ 2 32  3 21 17  2 36 16 29 24 17 21  2 26  3 28 39  4 22  2 17  8 36  0 12  9], best_score=0.77548


Results: 'hyper_score'
   Best score: 0.7754829361784187
   Best parameter set:
      'model_b.BASE'  : 0.07
      'model_b.IN00'  : 0.37
      'model_b.IN01'  : 0.08
      'model_b.IN02'  : 0.26
      'model_b.IN03'  : 0.22
      'model_b.IN04'  : 0.07
      'model_b.IN05'  : 0.41
      'model_b.IN06'  : 0.21
      'model_b.IN07'  : 0.34
      'model_b.IN08'  : 0.29
      'model_b.IN09'  : 0.22
      'model_b.IN10'  : 0.26
      'model_b.IN11'  : 0.07
      'model_b.M00'   : 0.31
      'model_b.OUT00' : 0.08
      'model_b.OUT01' : 0.33
      'model_b.OUT02' : 0.44
      'model_b.OUT03' : 0.09
      'model_b.OUT04' : 0.27
      'model_b.OUT05' : 0.07
      'model_b.OUT06' : 0.22
      'model_b.OUT07' : 0.13
      'model_b.OUT08' : 0.41
      'model_b.OUT09' : 0.05
      'model_b.OUT10' : 0.17
      'model_b.OUT11' : 0.14
   Best iteration: 0

   Random seed: 918024581

   Evaluation time   : 1235.4809987545013 sec    [100.0 %]
   Optimization time : 0.01408839225769043 sec    [0.0 %]
   Iteration time    : 1235.495087146759 sec    [4.94 sec/iter]

 - Best weights para =  ['0.07,0.37,0.08,0.26,0.22,0.07,0.41,0.21,0.34,0.29,0.22,0.26,0.07,0.31,0.08,0.33,0.44,0.09,0.27,0.07,0.22,0.13,0.41,0.05,0.17,0.14'] [True, False, False, False]
 - Best alpha para =  []
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  823e71c2e7e0d47fbef3309faa58402d54aa5446cb5495dec911a5752738b218
  - mm_use [True, False, False, False]
  - model_a My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - base_model My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - max_models 4
  - models ['Cartoons\\fluffity_v23.safetensors [27a31020bf]']
  - modes ['Add-Diff']
  - calcmodes ['Rebasin']
  - usembws [['BASE', 'INP*', 'M00', 'OUT*', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']]
  - weights ['0.07,0.37,0.08,0.26,0.22,0.07,0.41,0.21,0.34,0.29,0.22,0.26,0.07,0.31,0.08,0.33,0.44,0.09,0.27,0.07,0.22,0.13,0.41,0.05,0.17,0.14']
  - alpha [0.5]
  - adjust
  - use elemental [False]
  - elementals ['']
  - Parse elemental merge...
model_a = My_YACAM-SR-16286.fp16
Loading from file E:\SD\automatic1111\models\Stable-diffusion\My\YACAM-SR-16286.fp16.safetensors...
isxl = False , sd2 = False
check affected permutation blocks by rebasin merge...
compact_mode =  True
 - check possible UNet partial update...
 - partial changed blocks =  ['BASE', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'M00', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']
 - UNet partial update mode
 - Dynamic loading rebasin module...
Rebasin mode
 - Calulation device for Rebasin is  cuda
 - LAP library is lap
Loading model Cartoons_fluffity_v23...
Loading from file E:\SD\automatic1111\models\Stable-diffusion\Cartoons\fluffity_v23.safetensors...
mode = Add-Diff, mbw mode, alpha = [0.07, 0.37, 0.08, 0.26, 0.22, 0.07, 0.41, 0.21, 0.34, 0.29, 0.22, 0.26, 0.07, 0.31, 0.08, 0.33, 0.44, 0.09, 0.27, 0.07, 0.22, 0.13, 0.41, 0.05, 0.17, 0.14]
Stage #1/3: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:12<00:00, 71.04it/s]
Check uninitialized #2/3: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:00<00:00, 175597.89it/s]
Rebasin calc...
maximize weight matching using scipy linear_sum_assignment...
P_bg371: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  7.68it/s]
P_bg371: new - old = 0.0
weight order changed layers = ['P_bg324', 'P_bg371']
Save unchanged weights #3/3: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 251/251 [00:00<?, ?it/s]
Clip is fine
 - merge processing in 15.6s (prepare: 0.5s, merging: 15.1s).
 - loading scripts.patches...
 - lora patch
 - Textencoder(BASE) has been successfully updated
 - update UNet block input_blocks.0.
 - update UNet block input_blocks.1.
 - update UNet block input_blocks.2.
 - update UNet block input_blocks.3.
 - update UNet block input_blocks.4.
 - update UNet block input_blocks.5.
 - update UNet block input_blocks.6.
 - update UNet block input_blocks.7.
 - update UNet block input_blocks.8.
 - update UNet block input_blocks.9.
 - update UNet block input_blocks.10.
 - update UNet block input_blocks.11.
 - update UNet block middle_block.
 - update UNet block output_blocks.0.
 - update UNet block output_blocks.1.
 - update UNet block output_blocks.2.
 - update UNet block output_blocks.3.
 - update UNet block output_blocks.4.
 - update UNet block output_blocks.5.
 - update UNet block output_blocks.6.
 - update UNet block output_blocks.7.
 - update UNet block output_blocks.8.
 - update UNet block output_blocks.9.
 - update UNet block output_blocks.10.
 - update UNet block output_blocks.11.
 - UNet partial blocks have been successfully updated
 - Reload full state_dict...
 - remove old checkpointinfo
Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.24,0.26,0.35,0.14,0.28,0.15,0.19,0.1,0.25,0.18,0.31,0.42,0.1,0.44,0.14,0.19,0.17,0.23,0.3,0.17,0.15,0.26,0.07,0.41,0.29,0.43).safetensors [61beeff2f6]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 0.9s (create model: 0.4s, apply weights to model: 0.4s).
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.51it/s]
Total progress: 21it [00:05,  3.68it/s]
0: 640x448 1 face, 18.0ms
Speed: 3.0ms preprocess, 18.0ms inference, 3.0ms postprocess per image at shape (1, 3, 640, 448)
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:02<00:00,  3.66it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  f125e917962490426652c03fa8d20780021efc5a2258009c65e1900f49c60ea5
  - mm_use [True, False, False, False]
  - model_a My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - base_model My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - max_models 4
  - models ['Cartoons\\fluffity_v23.safetensors [27a31020bf]']
  - modes ['Add-Diff']
  - calcmodes ['Rebasin']
  - usembws [['BASE', 'INP*', 'M00', 'OUT*', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']]
  - weights ['0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25']
  - alpha [0.5]
  - adjust
  - use elemental [False]
  - elementals ['']
  - Parse elemental merge...
model_a = My_YACAM-SR-16286.fp16
Loading from file E:\SD\automatic1111\models\Stable-diffusion\My\YACAM-SR-16286.fp16.safetensors...
isxl = False , sd2 = False
check affected permutation blocks by rebasin merge...
compact_mode =  True
 - check possible UNet partial update...
 - partial changed blocks =  ['BASE', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'M00', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11']
 - UNet partial update mode
 - Dynamic loading rebasin module...
Rebasin mode
 - Calulation device for Rebasin is  cuda
 - LAP library is lap
Loading model Cartoons_fluffity_v23...
Loading from file E:\SD\automatic1111\models\Stable-diffusion\Cartoons\fluffity_v23.safetensors...
mode = Add-Diff, mbw mode, alpha = [0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25]
Stage #1/3: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:12<00:00, 68.79it/s]
Check uninitialized #2/3: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 876/876 [00:00<00:00, 175715.46it/s]
Rebasin calc...
maximize weight matching using scipy linear_sum_assignment...
P_bg371: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  7.43it/s]
P_bg371: new - old = 0.0
weight order changed layers = ['P_bg324', 'P_bg371']
Save unchanged weights #3/3: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 251/251 [00:00<00:00, 251678.29it/s]
Clip is fine
 - merge processing in 16.0s (prepare: 0.5s, merging: 15.5s).
 - loading scripts.patches...
 - lora patch
 - Textencoder(BASE) has been successfully updated
 - update UNet block input_blocks.0.
 - update UNet block input_blocks.1.
 - update UNet block input_blocks.2.
 - update UNet block input_blocks.3.
 - update UNet block input_blocks.4.
 - update UNet block input_blocks.5.
 - update UNet block input_blocks.6.
 - update UNet block input_blocks.7.
 - update UNet block input_blocks.8.
 - update UNet block input_blocks.9.
 - update UNet block input_blocks.10.
 - update UNet block input_blocks.11.
 - update UNet block middle_block.
 - update UNet block output_blocks.0.
 - update UNet block output_blocks.1.
 - update UNet block output_blocks.2.
 - update UNet block output_blocks.3.
 - update UNet block output_blocks.4.
 - update UNet block output_blocks.5.
 - update UNet block output_blocks.6.
 - update UNet block output_blocks.7.
 - update UNet block output_blocks.8.
 - update UNet block output_blocks.9.
 - update UNet block output_blocks.10.
 - update UNet block output_blocks.11.
 - UNet partial blocks have been successfully updated
 - Reload full state_dict...
 - remove old checkpointinfo
Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25,0.25).safetensors [f125e91796]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 0.9s (create model: 0.4s, apply weights to model: 0.4s).
Total progress: 21it [00:27,  1.32s/it]
 - Result score = 0.8113161807219965/s]

3 replies

wkpark Jan 25, 2024
Maintainer

OK. I can see the bug in the last confirm step.

optimization loop
last step: print out best paramters. - OK.
last confirm step: generate images with the best parameters. --> has some BUGs.

wkpark Jan 25, 2024
Maintainer

~~I guess, there are some extensions to prevent model-mixer not working properly.~~
I can confirm that the adetailer not working correctly with model-mixer. you can use uddetailer with model-mixer

wkpark Jan 25, 2024
Maintainer

I'm not sure but model-mixer has some bug/mistake to prevent the adetailer work properly.
(I just have updated model-mixer a bit)

you can use model-mixer, anyway.

you don't have to trust auto-merger best result. It just optimize "Scoring" of image classifier.
test image classifiers before auto-merger, selecting the most proper image classifier is the most essential step.

miasik · 2024-01-25T13:26:13Z

miasik
Jan 25, 2024
Author

~~I guess, there are some extensions to prevent model-mixer not working properly.~~ I can confirm that the adetailer not working correctly with model-mixer. you can use uddetailer with model-mixer

I changed Ad to Ud
Cheeses Christ! It works now as expected, I guess. Still have questions related to settings. I saw the wiki but it's hard to understand.

Could you please elaborate on the image below?
Is there a way to set minimum and maximum values for merging? For example - not less than 0.1 but no more than 0.9
How starting MBW influences on the process and the result?

2 replies

wkpark Jan 25, 2024
Maintainer

MBW = 0.2,0.2,0.4,0.5,...
Upper = 0.2, Lower = -0.2 // Max Limit = 0.5 (minimum == 0. maximum == 0.5)
- 0.2 => 0.2 - 0.2 < weight < 0.2 + 0.2 < 0.5 -> 0 < weight < 0.4 (A)
- 0.4 => 0.4 - 0.2 < weight < 0.4 + 0.4 < 0.5 -> 0.2 < weight < 0.5 (B)
Search Steps 5 => (A) weight (0 ~ 0.4 divided by 5 steps) or, Search inc 0.01 (not 0 case) =>0.01, 0.02, ... 0.4

any other options are not supported to make auto-merger clear/simpler

miasik Jan 25, 2024
Author

MBW is full 0.5
Upper 0.45 Lower -0.45 Max limit 1
Does it mean that each block will be evaluated from 0.5-0.45=0.05 to 0.5+0.45=0.95?

miasik · 2024-01-25T14:36:40Z

miasik
Jan 25, 2024
Author

Watched the log more carefully. Saw starting 0.5 MBW again periodically. is it a normal state? Should I increase the models limit to keep the starting MBW model untouched?

Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.25,0.35,0.6,0.1,0.35,0.25,0.85,0,0,0.85,0.85,0.35,0.6,0.25,0.35,0.25,0,0.1,0.25,0.35,0.85,0.35,0.6,0.25,0.25,0.5).safetensors [7c1c950a28]
Creating model from config: E:\SD\automatic1111\configs\v1-inference.yaml
Applying attention optimization: sdp... done.
Model loaded in 1.1s (create model: 0.4s, apply weights to model: 0.6s).
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.64it/s]
Total progress: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.56it/s]
 - Overriden seed = 67██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 21/21 [00:05<00:00,  3.62it/s]
debugs =  ['elemental merge', 'merge', 'adjust']
use_extra_elements =  True
 - mm_max_models =  4
config hash =  f9999c2929fe248a4c620b75f07f040ca02169d3c63352de800da1c7597543a7
  - mm_use [True, False, False, False]
  - model_a My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - base_model My\YACAM-SR-16286.fp16.safetensors [670934b0bd]
  - max_models 4
  - models ['Cartoons\\fluffity_v23.safetensors [27a31020bf]']
  - modes ['Add-Diff']
  - calcmodes ['Rebasin']
  - usembws [['BASE', 'INP*', 'M00', 'OUT*', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN07', 'IN08', 'IN09', 'IN10', 'IN11', 'OUT00', 'OUT01', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10', 'OUT11', 'M00', 'M00']]
  - weights ['0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5']
  - alpha [0.5]
  - adjust
  - use elemental [True]
  - elementals ['']
  - Parse elemental merge...
 Elemental merge wegiths =  None
model_a = My_YACAM-SR-16286.fp16
Loading from file E:\SD\automatic1111\models\Stable-diffusion\My\YACAM-SR-16286.fp16.safetensors...
isxl = False , sd2 = False
check affected permutation blocks by rebasin merge...
compact_mode =  True
 - check possible UNet partial update...
 - partial changed blocks =  ['BASE', 'IN00', 'IN01', 'IN02', 'IN03', 'IN04', 'IN05', 'IN06', 'IN08', 'IN09', 'IN10', 'IN11', 'M00', 'OUT00', 'OUT02', 'OUT03', 'OUT04', 'OUT05', 'OUT06', 'OUT07', 'OUT08', 'OUT09', 'OUT10']
 - UNet partial update mode
 - Dynamic loading rebasin module...
Rebasin mode
 - Calulation device for Rebasin is  cuda
 - LAP library is lap
Loading model Cartoons_fluffity_v23...
Loading from file E:\SD\automatic1111\models\Stable-diffusion\Cartoons\fluffity_v23.safetensors...
mode = Add-Diff, mbw mode, alpha = [0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5]
Stage #1/3: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 788/788 [00:11<00:00, 67.29it/s]
Check uninitialized #2/3: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 788/788 [00:00<00:00, 197532.37it/s]
Rebasin calc...
maximize weight matching using scipy linear_sum_assignment...
P_bg371: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  7.25it/s]
P_bg371: new - old = 0.0
weight order changed layers = ['P_bg324', 'P_bg371']
Save unchanged weights #3/3: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 251/251 [00:00<?, ?it/s]
Clip is fine
 - merge processing in 15.3s (prepare: 0.5s, merging: 14.7s).
 - loading scripts.patches...
 - lora patch
 - Textencoder(BASE) has been successfully updated
 - update UNet block input_blocks.0.
 - update UNet block input_blocks.1.
 - update UNet block input_blocks.2.
 - update UNet block input_blocks.3.
 - update UNet block input_blocks.4.
 - update UNet block input_blocks.5.
 - update UNet block input_blocks.6.
 - update UNet block input_blocks.8.
 - update UNet block input_blocks.9.
 - update UNet block input_blocks.10.
 - update UNet block input_blocks.11.
 - update UNet block middle_block.
 - update UNet block output_blocks.0.
 - update UNet block output_blocks.2.
 - update UNet block output_blocks.3.
 - update UNet block output_blocks.4.
 - update UNet block output_blocks.5.
 - update UNet block output_blocks.6.
 - update UNet block output_blocks.7.
 - update UNet block output_blocks.8.
 - update UNet block output_blocks.9.
 - update UNet block output_blocks.10.
 - UNet partial blocks have been successfully updated
 - Reload full state_dict...
 - remove old checkpointinfo
Unloading model 3 over the limit of 2...
 - model 2: My_YACAM-SR-16286.fp16 + (Cartoons_fluffity_v23 - My\YACAM-SR-16286.fp16.safetensors [670934b0bd]) x alpha_0(0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5).safetensors [f9999c2929]

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

How to get the resulting model/MBW? #108

{{title}}

Replies: 7 comments 6 replies

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

Select a reply

How to get the resulting model/MBW? #108

miasik Jan 25, 2024

Replies: 7 comments · 6 replies

miasik Jan 25, 2024 Author

miasik Jan 25, 2024 Author

wkpark Jan 25, 2024 Maintainer

wkpark Jan 25, 2024 Maintainer

miasik Jan 25, 2024 Author

miasik Jan 25, 2024 Author

wkpark Jan 25, 2024 Maintainer

wkpark Jan 25, 2024 Maintainer

wkpark Jan 25, 2024 Maintainer

miasik Jan 25, 2024 Author

wkpark Jan 25, 2024 Maintainer

miasik Jan 25, 2024 Author

miasik Jan 25, 2024 Author

miasik
Jan 25, 2024

Replies: 7 comments 6 replies

miasik
Jan 25, 2024
Author

miasik
Jan 25, 2024
Author

wkpark
Jan 25, 2024
Maintainer

wkpark
Jan 25, 2024
Maintainer

miasik Jan 25, 2024
Author

miasik
Jan 25, 2024
Author

wkpark Jan 25, 2024
Maintainer

wkpark Jan 25, 2024
Maintainer

wkpark Jan 25, 2024
Maintainer

miasik
Jan 25, 2024
Author

wkpark Jan 25, 2024
Maintainer

miasik Jan 25, 2024
Author

miasik
Jan 25, 2024
Author