Hugging Face's logo

napalm
/
qwen_image21

Model card Files
xet
 Community
napalm commited on
Commit
0875e55
·
verified ·
1 Parent(s): 30dd8d4

Upload 2 files

Browse files
Files changed (2) hide show
  1. extracted/redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8-r128-comfy.safetensors +3 -0
  2. extracted/redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8-r128-comfy.safetensors.REPORT.txt +255 -0
extracted/redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8-r128-comfy.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0857550da18015759884010f040ae0737d65c6a300ed0292def14b5da1f2ab55
3
+ size 418510664
extracted/redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8-r128-comfy.safetensors.REPORT.txt ADDED
@@ -0,0 +1,255 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================================
2
+ KREA2 LORA EXTRACTOR - STRUCTURE & DIFFERENCE REPORT
3
+ ==============================================================================
4
+ BASE : qwen_image_2.1_int8_convrot.safetensors (6.76 GB)
5
+ FINETUNE : redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8.safetensors (6.76 GB)
6
+ FORMAT : base=safetensors finetune=safetensors
7
+
8
+ ------------------------------------------------------------------------------
9
+ DIFFERENCES DETECTED (structure matching):
10
+ ------------------------------------------------------------------------------
11
+ tensors base=649 finetune=745 delta=+96
12
+ .weight tensors base=265 finetune=297
13
+ .weight_scale tensors base=192 finetune=224
14
+ .comfy_quant meta tensors base=192 finetune=224
15
+ quant descriptors base=192 finetune=224
16
+ metadata tags base=['format'] finetune=[]
17
+
18
+ detected finetune key prefix : (none - same root)
19
+ [both share the same root]
20
+
21
+ ------------------------------------------------------------------------------
22
+ PER-LAYER COMPATIBILITY (after prefix strip):
23
+ ------------------------------------------------------------------------------
24
+ common .weight keys : 233 / 265 (87.9%)
25
+ keys only in base : 32
26
+ transformer_blocks.0.img_mlp.gate_up.weight, transformer_blocks.1.img_mlp.gate_up.weight, transformer_blocks.10.img_mlp.gate_up.weight, transformer_blocks.11.img_mlp.gate_up.weight, transformer_blocks.12.img_mlp.gate_up.weight, transformer_blocks.13.img_mlp.gate_up.weight, transformer_blocks.14.img_mlp.gate_up.weight, transformer_blocks.15.img_mlp.gate_up.weight ...
27
+ keys only in finetune : 64
28
+ transformer_blocks.0.img_mlp.gate_layer.weight, transformer_blocks.0.img_mlp.proj.weight, transformer_blocks.1.img_mlp.gate_layer.weight, transformer_blocks.1.img_mlp.proj.weight, transformer_blocks.10.img_mlp.gate_layer.weight, transformer_blocks.10.img_mlp.proj.weight, transformer_blocks.11.img_mlp.gate_layer.weight, transformer_blocks.11.img_mlp.proj.weight ...
29
+ dtype mismatches : 0
30
+ shape mismatches : 0
31
+ bytes mismatches : 0
32
+
33
+ ------------------------------------------------------------------------------
34
+ PRECISION AUDIT - BASE:
35
+ ------------------------------------------------------------------------------
36
+ dtype counts : {'F32': 192, 'BF16': 73, 'I8': 192, 'U8': 192}
37
+ descriptors: 192
38
+ formats detected: int8_tensorwise=192
39
+ with full_precision_matrix_mult: 0
40
+ with convrot: 192
41
+ with per_row: 0
42
+ groupsize_distribution: {'256': 192}
43
+
44
+ ------------------------------------------------------------------------------
45
+ PRECISION AUDIT - FINETUNE:
46
+ ------------------------------------------------------------------------------
47
+ dtype counts : {'F32': 224, 'BF16': 73, 'I8': 224, 'U8': 224}
48
+ descriptors: 224
49
+ formats detected: int8_tensorwise=224
50
+ with full_precision_matrix_mult: 0
51
+ with convrot: 224
52
+ with per_row: 0
53
+ groupsize_distribution: {'256': 224}
54
+
55
+ ------------------------------------------------------------------------------
56
+ EXTRACTION PARAMETERS:
57
+ ------------------------------------------------------------------------------
58
+ rank : 128
59
+ precision : bf16
60
+ format : comfy
61
+ key prefix : 'diffusion_model.'
62
+ device : cuda
63
+
64
+ ------------------------------------------------------------------------------
65
+ PER-LAYER PROGRESS (168 layers):
66
+ ------------------------------------------------------------------------------
67
+ [ 1/233] img_in shape=(4096, 64) R= 64 energy= 0.0% relerr=1.000
68
+ [ 2/233] modulation.1 shape=(16384, 4096) R=128 energy= 84.3% relerr=0.397
69
+ [ 3/233] norm_out.linear shape=(4096, 4096) R=128 energy= 0.0% relerr=1.000
70
+ [ 4/233] proj_out shape=(64, 4096) R= 64 energy= 0.0% relerr=1.000
71
+ [ 5/233] time_text_embed.timestep_embedder.linear_1 shape=(4096, 256) R=128 energy= 99.4% relerr=0.078
72
+ [ 6/233] time_text_embed.timestep_embedder.linear_2 shape=(4096, 4096) R=128 energy= 98.2% relerr=0.133
73
+ [ 9/233] transformer_blocks.0.attn.to_k shape=(4096, 4096) R=128 energy= 81.3% relerr=0.433
74
+ [ 10/233] transformer_blocks.0.attn.to_out.0 shape=(4096, 4096) R=128 energy= 90.8% relerr=0.304
75
+ [ 11/233] transformer_blocks.0.attn.to_q shape=(4096, 4096) R=128 energy= 81.7% relerr=0.428
76
+ [ 12/233] transformer_blocks.0.attn.to_v shape=(4096, 4096) R=128 energy= 82.9% relerr=0.413
77
+ [ 13/233] transformer_blocks.0.img_mlp.out shape=(4096, 12288) R=128 energy= 73.9% relerr=0.511
78
+ [ 16/233] transformer_blocks.1.attn.to_k shape=(4096, 4096) R=128 energy= 73.1% relerr=0.518
79
+ [ 17/233] transformer_blocks.1.attn.to_out.0 shape=(4096, 4096) R=128 energy= 73.4% relerr=0.515
80
+ [ 18/233] transformer_blocks.1.attn.to_q shape=(4096, 4096) R=128 energy= 82.0% relerr=0.424
81
+ [ 19/233] transformer_blocks.1.attn.to_v shape=(4096, 4096) R=128 energy= 78.9% relerr=0.459
82
+ [ 20/233] transformer_blocks.1.img_mlp.out shape=(4096, 12288) R=128 energy= 72.9% relerr=0.521
83
+ [ 23/233] transformer_blocks.10.attn.to_k shape=(4096, 4096) R=128 energy= 74.0% relerr=0.509
84
+ [ 24/233] transformer_blocks.10.attn.to_out.0 shape=(4096, 4096) R=128 energy= 68.1% relerr=0.564
85
+ [ 25/233] transformer_blocks.10.attn.to_q shape=(4096, 4096) R=128 energy= 73.2% relerr=0.518
86
+ [ 26/233] transformer_blocks.10.attn.to_v shape=(4096, 4096) R=128 energy= 65.0% relerr=0.592
87
+ [ 27/233] transformer_blocks.10.img_mlp.out shape=(4096, 12288) R=128 energy= 71.6% relerr=0.533
88
+ [ 30/233] transformer_blocks.11.attn.to_k shape=(4096, 4096) R=128 energy= 75.1% relerr=0.499
89
+ [ 31/233] transformer_blocks.11.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.3% relerr=0.571
90
+ [ 32/233] transformer_blocks.11.attn.to_q shape=(4096, 4096) R=128 energy= 74.1% relerr=0.509
91
+ [ 33/233] transformer_blocks.11.attn.to_v shape=(4096, 4096) R=128 energy= 65.3% relerr=0.589
92
+ [ 34/233] transformer_blocks.11.img_mlp.out shape=(4096, 12288) R=128 energy= 70.7% relerr=0.541
93
+ [ 37/233] transformer_blocks.12.attn.to_k shape=(4096, 4096) R=128 energy= 74.9% relerr=0.501
94
+ [ 38/233] transformer_blocks.12.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.4% relerr=0.571
95
+ [ 39/233] transformer_blocks.12.attn.to_q shape=(4096, 4096) R=128 energy= 73.9% relerr=0.511
96
+ [ 40/233] transformer_blocks.12.attn.to_v shape=(4096, 4096) R=128 energy= 66.3% relerr=0.581
97
+ [ 41/233] transformer_blocks.12.img_mlp.out shape=(4096, 12288) R=128 energy= 72.0% relerr=0.529
98
+ [ 44/233] transformer_blocks.13.attn.to_k shape=(4096, 4096) R=128 energy= 78.0% relerr=0.469
99
+ [ 45/233] transformer_blocks.13.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.8% relerr=0.567
100
+ [ 46/233] transformer_blocks.13.attn.to_q shape=(4096, 4096) R=128 energy= 77.2% relerr=0.477
101
+ [ 47/233] transformer_blocks.13.attn.to_v shape=(4096, 4096) R=128 energy= 65.9% relerr=0.584
102
+ [ 48/233] transformer_blocks.13.img_mlp.out shape=(4096, 12288) R=128 energy= 72.7% relerr=0.523
103
+ [ 51/233] transformer_blocks.14.attn.to_k shape=(4096, 4096) R=128 energy= 78.8% relerr=0.460
104
+ [ 52/233] transformer_blocks.14.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.7% relerr=0.568
105
+ [ 53/233] transformer_blocks.14.attn.to_q shape=(4096, 4096) R=128 energy= 76.3% relerr=0.486
106
+ [ 54/233] transformer_blocks.14.attn.to_v shape=(4096, 4096) R=128 energy= 64.5% relerr=0.596
107
+ [ 55/233] transformer_blocks.14.img_mlp.out shape=(4096, 12288) R=128 energy= 73.6% relerr=0.513
108
+ [ 58/233] transformer_blocks.15.attn.to_k shape=(4096, 4096) R=128 energy= 79.6% relerr=0.452
109
+ [ 59/233] transformer_blocks.15.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.3% relerr=0.571
110
+ [ 60/233] transformer_blocks.15.attn.to_q shape=(4096, 4096) R=128 energy= 77.7% relerr=0.472
111
+ [ 61/233] transformer_blocks.15.attn.to_v shape=(4096, 4096) R=128 energy= 67.1% relerr=0.573
112
+ [ 62/233] transformer_blocks.15.img_mlp.out shape=(4096, 12288) R=128 energy= 74.6% relerr=0.504
113
+ [ 65/233] transformer_blocks.16.attn.to_k shape=(4096, 4096) R=128 energy= 80.0% relerr=0.447
114
+ [ 66/233] transformer_blocks.16.attn.to_out.0 shape=(4096, 4096) R=128 energy= 68.6% relerr=0.561
115
+ [ 67/233] transformer_blocks.16.attn.to_q shape=(4096, 4096) R=128 energy= 80.2% relerr=0.444
116
+ [ 68/233] transformer_blocks.16.attn.to_v shape=(4096, 4096) R=128 energy= 68.3% relerr=0.563
117
+ [ 69/233] transformer_blocks.16.img_mlp.out shape=(4096, 12288) R=128 energy= 74.1% relerr=0.509
118
+ [ 72/233] transformer_blocks.17.attn.to_k shape=(4096, 4096) R=128 energy= 82.5% relerr=0.419
119
+ [ 73/233] transformer_blocks.17.attn.to_out.0 shape=(4096, 4096) R=128 energy= 66.9% relerr=0.576
120
+ [ 74/233] transformer_blocks.17.attn.to_q shape=(4096, 4096) R=128 energy= 82.5% relerr=0.418
121
+ [ 75/233] transformer_blocks.17.attn.to_v shape=(4096, 4096) R=128 energy= 66.5% relerr=0.579
122
+ [ 76/233] transformer_blocks.17.img_mlp.out shape=(4096, 12288) R=128 energy= 75.3% relerr=0.497
123
+ [ 79/233] transformer_blocks.18.attn.to_k shape=(4096, 4096) R=128 energy= 82.1% relerr=0.423
124
+ [ 80/233] transformer_blocks.18.attn.to_out.0 shape=(4096, 4096) R=128 energy= 68.8% relerr=0.558
125
+ [ 81/233] transformer_blocks.18.attn.to_q shape=(4096, 4096) R=128 energy= 80.5% relerr=0.442
126
+ [ 82/233] transformer_blocks.18.attn.to_v shape=(4096, 4096) R=128 energy= 67.6% relerr=0.569
127
+ [ 83/233] transformer_blocks.18.img_mlp.out shape=(4096, 12288) R=128 energy= 77.0% relerr=0.479
128
+ [ 86/233] transformer_blocks.19.attn.to_k shape=(4096, 4096) R=128 energy= 83.3% relerr=0.408
129
+ [ 87/233] transformer_blocks.19.attn.to_out.0 shape=(4096, 4096) R=128 energy= 71.5% relerr=0.534
130
+ [ 88/233] transformer_blocks.19.attn.to_q shape=(4096, 4096) R=128 energy= 83.0% relerr=0.413
131
+ [ 89/233] transformer_blocks.19.attn.to_v shape=(4096, 4096) R=128 energy= 72.2% relerr=0.528
132
+ [ 90/233] transformer_blocks.19.img_mlp.out shape=(4096, 12288) R=128 energy= 78.9% relerr=0.459
133
+ [ 93/233] transformer_blocks.2.attn.to_k shape=(4096, 4096) R=128 energy= 77.4% relerr=0.475
134
+ [ 94/233] transformer_blocks.2.attn.to_out.0 shape=(4096, 4096) R=128 energy= 75.6% relerr=0.494
135
+ [ 95/233] transformer_blocks.2.attn.to_q shape=(4096, 4096) R=128 energy= 80.3% relerr=0.444
136
+ [ 96/233] transformer_blocks.2.attn.to_v shape=(4096, 4096) R=128 energy= 84.7% relerr=0.391
137
+ [ 97/233] transformer_blocks.2.img_mlp.out shape=(4096, 12288) R=128 energy= 67.9% relerr=0.566
138
+ [100/233] transformer_blocks.20.attn.to_k shape=(4096, 4096) R=128 energy= 87.9% relerr=0.348
139
+ [101/233] transformer_blocks.20.attn.to_out.0 shape=(4096, 4096) R=128 energy= 74.7% relerr=0.503
140
+ [102/233] transformer_blocks.20.attn.to_q shape=(4096, 4096) R=128 energy= 87.6% relerr=0.352
141
+ [103/233] transformer_blocks.20.attn.to_v shape=(4096, 4096) R=128 energy= 73.8% relerr=0.512
142
+ [104/233] transformer_blocks.20.img_mlp.out shape=(4096, 12288) R=128 energy= 80.1% relerr=0.447
143
+ [107/233] transformer_blocks.21.attn.to_k shape=(4096, 4096) R=128 energy= 91.2% relerr=0.297
144
+ [108/233] transformer_blocks.21.attn.to_out.0 shape=(4096, 4096) R=128 energy= 75.5% relerr=0.495
145
+ [109/233] transformer_blocks.21.attn.to_q shape=(4096, 4096) R=128 energy= 90.5% relerr=0.308
146
+ [110/233] transformer_blocks.21.attn.to_v shape=(4096, 4096) R=128 energy= 76.0% relerr=0.490
147
+ [111/233] transformer_blocks.21.img_mlp.out shape=(4096, 12288) R=128 energy= 81.7% relerr=0.428
148
+ [114/233] transformer_blocks.22.attn.to_k shape=(4096, 4096) R=128 energy= 89.0% relerr=0.331
149
+ [115/233] transformer_blocks.22.attn.to_out.0 shape=(4096, 4096) R=128 energy= 66.9% relerr=0.576
150
+ [116/233] transformer_blocks.22.attn.to_q shape=(4096, 4096) R=128 energy= 88.5% relerr=0.340
151
+ [117/233] transformer_blocks.22.attn.to_v shape=(4096, 4096) R=128 energy= 67.3% relerr=0.572
152
+ [118/233] transformer_blocks.22.img_mlp.out shape=(4096, 12288) R=128 energy= 77.7% relerr=0.472
153
+ [121/233] transformer_blocks.23.attn.to_k shape=(4096, 4096) R=128 energy= 91.3% relerr=0.295
154
+ [122/233] transformer_blocks.23.attn.to_out.0 shape=(4096, 4096) R=128 energy= 68.1% relerr=0.564
155
+ [123/233] transformer_blocks.23.attn.to_q shape=(4096, 4096) R=128 energy= 90.7% relerr=0.304
156
+ [124/233] transformer_blocks.23.attn.to_v shape=(4096, 4096) R=128 energy= 69.3% relerr=0.554
157
+ [125/233] transformer_blocks.23.img_mlp.out shape=(4096, 12288) R=128 energy= 80.1% relerr=0.446
158
+ [128/233] transformer_blocks.24.attn.to_k shape=(4096, 4096) R=128 energy= 89.1% relerr=0.330
159
+ [129/233] transformer_blocks.24.attn.to_out.0 shape=(4096, 4096) R=128 energy= 64.4% relerr=0.597
160
+ [130/233] transformer_blocks.24.attn.to_q shape=(4096, 4096) R=128 energy= 89.5% relerr=0.324
161
+ [131/233] transformer_blocks.24.attn.to_v shape=(4096, 4096) R=128 energy= 61.0% relerr=0.624
162
+ [132/233] transformer_blocks.24.img_mlp.out shape=(4096, 12288) R=128 energy= 76.1% relerr=0.489
163
+ [135/233] transformer_blocks.25.attn.to_k shape=(4096, 4096) R=128 energy= 88.8% relerr=0.335
164
+ [136/233] transformer_blocks.25.attn.to_out.0 shape=(4096, 4096) R=128 energy= 61.9% relerr=0.617
165
+ [137/233] transformer_blocks.25.attn.to_q shape=(4096, 4096) R=128 energy= 88.2% relerr=0.344
166
+ [138/233] transformer_blocks.25.attn.to_v shape=(4096, 4096) R=128 energy= 62.7% relerr=0.611
167
+ [139/233] transformer_blocks.25.img_mlp.out shape=(4096, 12288) R=128 energy= 78.8% relerr=0.460
168
+ [142/233] transformer_blocks.26.attn.to_k shape=(4096, 4096) R=128 energy= 84.9% relerr=0.389
169
+ [143/233] transformer_blocks.26.attn.to_out.0 shape=(4096, 4096) R=128 energy= 62.5% relerr=0.612
170
+ [144/233] transformer_blocks.26.attn.to_q shape=(4096, 4096) R=128 energy= 84.7% relerr=0.391
171
+ [145/233] transformer_blocks.26.attn.to_v shape=(4096, 4096) R=128 energy= 58.7% relerr=0.643
172
+ [146/233] transformer_blocks.26.img_mlp.out shape=(4096, 12288) R=128 energy= 79.6% relerr=0.451
173
+ [149/233] transformer_blocks.27.attn.to_k shape=(4096, 4096) R=128 energy= 83.2% relerr=0.410
174
+ [150/233] transformer_blocks.27.attn.to_out.0 shape=(4096, 4096) R=128 energy= 62.5% relerr=0.612
175
+ [151/233] transformer_blocks.27.attn.to_q shape=(4096, 4096) R=128 energy= 86.3% relerr=0.371
176
+ [152/233] transformer_blocks.27.attn.to_v shape=(4096, 4096) R=128 energy= 55.4% relerr=0.668
177
+ [153/233] transformer_blocks.27.img_mlp.out shape=(4096, 12288) R=128 energy= 82.5% relerr=0.419
178
+ [156/233] transformer_blocks.28.attn.to_k shape=(4096, 4096) R=128 energy= 87.1% relerr=0.360
179
+ [157/233] transformer_blocks.28.attn.to_out.0 shape=(4096, 4096) R=128 energy= 65.7% relerr=0.585
180
+ [158/233] transformer_blocks.28.attn.to_q shape=(4096, 4096) R=128 energy= 85.8% relerr=0.377
181
+ [159/233] transformer_blocks.28.attn.to_v shape=(4096, 4096) R=128 energy= 59.5% relerr=0.636
182
+ [160/233] transformer_blocks.28.img_mlp.out shape=(4096, 12288) R=128 energy= 84.6% relerr=0.392
183
+ [163/233] transformer_blocks.29.attn.to_k shape=(4096, 4096) R=128 energy= 88.4% relerr=0.341
184
+ [164/233] transformer_blocks.29.attn.to_out.0 shape=(4096, 4096) R=128 energy= 65.5% relerr=0.587
185
+ [165/233] transformer_blocks.29.attn.to_q shape=(4096, 4096) R=128 energy= 88.8% relerr=0.335
186
+ [166/233] transformer_blocks.29.attn.to_v shape=(4096, 4096) R=128 energy= 63.2% relerr=0.607
187
+ [167/233] transformer_blocks.29.img_mlp.out shape=(4096, 12288) R=128 energy= 86.4% relerr=0.369
188
+ [170/233] transformer_blocks.3.attn.to_k shape=(4096, 4096) R=128 energy= 75.8% relerr=0.492
189
+ [171/233] transformer_blocks.3.attn.to_out.0 shape=(4096, 4096) R=128 energy= 73.7% relerr=0.513
190
+ [172/233] transformer_blocks.3.attn.to_q shape=(4096, 4096) R=128 energy= 77.1% relerr=0.478
191
+ [173/233] transformer_blocks.3.attn.to_v shape=(4096, 4096) R=128 energy= 74.4% relerr=0.506
192
+ [174/233] transformer_blocks.3.img_mlp.out shape=(4096, 12288) R=128 energy= 70.5% relerr=0.543
193
+ [177/233] transformer_blocks.30.attn.to_k shape=(4096, 4096) R=128 energy= 89.7% relerr=0.321
194
+ [178/233] transformer_blocks.30.attn.to_out.0 shape=(4096, 4096) R=128 energy= 66.9% relerr=0.576
195
+ [179/233] transformer_blocks.30.attn.to_q shape=(4096, 4096) R=128 energy= 89.0% relerr=0.331
196
+ [180/233] transformer_blocks.30.attn.to_v shape=(4096, 4096) R=128 energy= 66.6% relerr=0.578
197
+ [181/233] transformer_blocks.30.img_mlp.out shape=(4096, 12288) R=128 energy= 90.1% relerr=0.315
198
+ [184/233] transformer_blocks.31.attn.to_k shape=(4096, 4096) R=128 energy= 90.0% relerr=0.317
199
+ [185/233] transformer_blocks.31.attn.to_out.0 shape=(4096, 4096) R=128 energy= 59.6% relerr=0.635
200
+ [186/233] transformer_blocks.31.attn.to_q shape=(4096, 4096) R=128 energy= 90.9% relerr=0.302
201
+ [187/233] transformer_blocks.31.attn.to_v shape=(4096, 4096) R=128 energy= 60.2% relerr=0.630
202
+ [188/233] transformer_blocks.31.img_mlp.out shape=(4096, 12288) R=128 energy= 97.0% relerr=0.174
203
+ [191/233] transformer_blocks.4.attn.to_k shape=(4096, 4096) R=128 energy= 79.5% relerr=0.452
204
+ [192/233] transformer_blocks.4.attn.to_out.0 shape=(4096, 4096) R=128 energy= 74.1% relerr=0.509
205
+ [193/233] transformer_blocks.4.attn.to_q shape=(4096, 4096) R=128 energy= 82.6% relerr=0.417
206
+ [194/233] transformer_blocks.4.attn.to_v shape=(4096, 4096) R=128 energy= 82.5% relerr=0.418
207
+ [195/233] transformer_blocks.4.img_mlp.out shape=(4096, 12288) R=128 energy= 68.0% relerr=0.566
208
+ [198/233] transformer_blocks.5.attn.to_k shape=(4096, 4096) R=128 energy= 75.9% relerr=0.491
209
+ [199/233] transformer_blocks.5.attn.to_out.0 shape=(4096, 4096) R=128 energy= 71.4% relerr=0.534
210
+ [200/233] transformer_blocks.5.attn.to_q shape=(4096, 4096) R=128 energy= 76.0% relerr=0.489
211
+ [201/233] transformer_blocks.5.attn.to_v shape=(4096, 4096) R=128 energy= 75.6% relerr=0.494
212
+ [202/233] transformer_blocks.5.img_mlp.out shape=(4096, 12288) R=128 energy= 71.6% relerr=0.533
213
+ [205/233] transformer_blocks.6.attn.to_k shape=(4096, 4096) R=128 energy= 81.9% relerr=0.426
214
+ [206/233] transformer_blocks.6.attn.to_out.0 shape=(4096, 4096) R=128 energy= 69.5% relerr=0.553
215
+ [207/233] transformer_blocks.6.attn.to_q shape=(4096, 4096) R=128 energy= 73.0% relerr=0.520
216
+ [208/233] transformer_blocks.6.attn.to_v shape=(4096, 4096) R=128 energy= 69.5% relerr=0.552
217
+ [209/233] transformer_blocks.6.img_mlp.out shape=(4096, 12288) R=128 energy= 70.8% relerr=0.541
218
+ [212/233] transformer_blocks.7.attn.to_k shape=(4096, 4096) R=128 energy= 73.9% relerr=0.511
219
+ [213/233] transformer_blocks.7.attn.to_out.0 shape=(4096, 4096) R=128 energy= 69.5% relerr=0.552
220
+ [214/233] transformer_blocks.7.attn.to_q shape=(4096, 4096) R=128 energy= 76.2% relerr=0.488
221
+ [215/233] transformer_blocks.7.attn.to_v shape=(4096, 4096) R=128 energy= 68.7% relerr=0.560
222
+ [216/233] transformer_blocks.7.img_mlp.out shape=(4096, 12288) R=128 energy= 72.6% relerr=0.523
223
+ [219/233] transformer_blocks.8.attn.to_k shape=(4096, 4096) R=128 energy= 73.9% relerr=0.511
224
+ [220/233] transformer_blocks.8.attn.to_out.0 shape=(4096, 4096) R=128 energy= 68.1% relerr=0.565
225
+ [221/233] transformer_blocks.8.attn.to_q shape=(4096, 4096) R=128 energy= 78.2% relerr=0.467
226
+ [222/233] transformer_blocks.8.attn.to_v shape=(4096, 4096) R=128 energy= 66.5% relerr=0.579
227
+ [223/233] transformer_blocks.8.img_mlp.out shape=(4096, 12288) R=128 energy= 74.1% relerr=0.509
228
+ [226/233] transformer_blocks.9.attn.to_k shape=(4096, 4096) R=128 energy= 74.0% relerr=0.510
229
+ [227/233] transformer_blocks.9.attn.to_out.0 shape=(4096, 4096) R=128 energy= 67.4% relerr=0.571
230
+ [228/233] transformer_blocks.9.attn.to_q shape=(4096, 4096) R=128 energy= 73.0% relerr=0.520
231
+ [229/233] transformer_blocks.9.attn.to_v shape=(4096, 4096) R=128 energy= 64.9% relerr=0.592
232
+ [230/233] transformer_blocks.9.img_mlp.out shape=(4096, 12288) R=128 energy= 72.0% relerr=0.529
233
+ [231/233] txt_in.in_layer shape=(4096, 4096) R=128 energy= 0.0% relerr=1.000
234
+ [232/233] txt_in.out_layer shape=(4096, 4096) R=128 energy= 0.0% relerr=1.000
235
+
236
+ ------------------------------------------------------------------------------
237
+ EXTRACTION RESULT:
238
+ ------------------------------------------------------------------------------
239
+ layers extracted : 168
240
+ bias diffs : 0
241
+ layers skipped : 65
242
+ skipped: transformer_blocks.0.attn.norm_k.weight (shape-mismatch), transformer_blocks.0.attn.norm_q.weight (shape-mismatch), transformer_blocks.1.attn.norm_k.weight (shape-mismatch), transformer_blocks.1.attn.norm_q.weight (shape-mismatch), transformer_blocks.10.attn.norm_k.weight (shape-mismatch), transformer_blocks.10.attn.norm_q.weight (shape-mismatch), transformer_blocks.11.attn.norm_k.weight (shape-mismatch), transformer_blocks.11.attn.norm_q.weight (shape-mismatch)
243
+ layers unsupported: 0
244
+ avg energy captured @ rank 128 : 73.61%
245
+ min / median / max energy : 0.00% / 74.43% / 99.39%
246
+ elapsed : 1185.53 s
247
+
248
+ ------------------------------------------------------------------------------
249
+ OUTPUT:
250
+ ------------------------------------------------------------------------------
251
+ path : e:\python_scripts\2026\comfyui_lora_extractor_max\redqw21UNLOCKEDV2TI2I_redqw21UNLOCKEDV2_int8-r128-comfy.safetensors
252
+ size : 399.12 MB
253
+ tensors : 336
254
+ STATUS : OK - LoRA saved successfully
255
+ ==============================================================================