Hugging Face's logo

Spaces:
ujin-song
/
ortha
Running

App Files Community
ujin-song commited on
Commit
50900f7
•
1 Parent(s): 2df2a2b

joy, linguini, raya

Browse files
Files changed (9) hide show
  1. experiments/single-concept/joy/8339_joy_ortho.yml +98 -0
  2. experiments/single-concept/joy/models/edlora_model-latest.pth +3 -0
  3. experiments/single-concept/joy/train_8339_joy_ortho_20240615_121643.log +0 -0
  4. experiments/single-concept/linguini/6931_linguini_ortho.yml +98 -0
  5. experiments/single-concept/linguini/models/edlora_model-latest.pth +3 -0
  6. experiments/single-concept/linguini/train_6931_linguini_ortho_20240615_132945.log +192 -0
  7. experiments/single-concept/raya/7914_raya_ortho.yml +98 -0
  8. experiments/single-concept/raya/models/edlora_model-latest.pth +3 -0
  9. experiments/single-concept/raya/train_7914_raya_ortho_20240615_125726.log +0 -0
experiments/single-concept/joy/8339_joy_ortho.yml ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # GENERATE TIME: Sat Jun 15 12:16:43 2024
2
+ # CMD:
3
+ # train_edlora.py -opt single-concept/train_configs/8339_joy_ortho.yml
4
+
5
+ name: 8339_joy_ortho
6
+ manual_seed: 8339
7
+ mixed_precision: fp16
8
+ gradient_accumulation_steps: 1
9
+ datasets:
10
+ train:
11
+ name: LoraDataset
12
+ concept_list: single-concept/data_configs/joy.json
13
+ use_caption: true
14
+ use_mask: true
15
+ instance_transform:
16
+ - type: HumanResizeCropFinalV3
17
+ size: 512
18
+ crop_p: 0.5
19
+ - type: ToTensor
20
+ - type: Normalize
21
+ mean:
22
+ - 0.5
23
+ std:
24
+ - 0.5
25
+ - type: ShuffleCaption
26
+ keep_token_num: 1
27
+ - type: EnhanceText
28
+ enhance_type: human
29
+ replace_mapping:
30
+ <TOK>: <joy1> <joy2>
31
+ batch_size_per_gpu: 2
32
+ dataset_enlarge_ratio: 500
33
+ val_vis:
34
+ name: PromptDataset
35
+ prompts: single-concept/validation_prompts/characters/test_girl_disney.txt
36
+ num_samples_per_prompt: 8
37
+ latent_size:
38
+ - 4
39
+ - 64
40
+ - 64
41
+ replace_mapping:
42
+ <TOK>: <joy1> <joy2>
43
+ batch_size_per_gpu: 4
44
+ models:
45
+ pretrained_path: nitrosocke/mo-di-diffusion
46
+ enable_edlora: true
47
+ finetune_cfg:
48
+ text_embedding:
49
+ enable_tuning: true
50
+ lr: 0.001
51
+ text_encoder:
52
+ enable_tuning: true
53
+ lora_cfg:
54
+ rank: 5
55
+ alpha: 1.0
56
+ where: CLIPAttention
57
+ lr: 1.0e-05
58
+ unet:
59
+ enable_tuning: true
60
+ lora_cfg:
61
+ rank: 5
62
+ alpha: 1.0
63
+ where: Attention
64
+ lr: 0.0001
65
+ new_concept_token: <joy1>+<joy2>
66
+ initializer_token: <rand-0.013>+girl
67
+ noise_offset: 0.01
68
+ attn_reg_weight: 0.01
69
+ reg_full_identity: false
70
+ use_mask_loss: true
71
+ gradient_checkpoint: false
72
+ enable_xformers: true
73
+ path:
74
+ pretrain_network: null
75
+ train:
76
+ optim_g:
77
+ type: AdamW
78
+ lr: 0.0
79
+ weight_decay: 0.01
80
+ betas:
81
+ - 0.9
82
+ - 0.999
83
+ unet_kv_drop_rate: 0
84
+ scheduler: linear
85
+ emb_norm_threshold: 0.55
86
+ val:
87
+ val_during_save: false
88
+ compose_visualize: false
89
+ alpha_list:
90
+ - 0
91
+ - 0.7
92
+ - 1.0
93
+ sample:
94
+ num_inference_steps: 50
95
+ guidance_scale: 7.5
96
+ logger:
97
+ print_freq: 10
98
+ save_checkpoint_freq: 10000.0
experiments/single-concept/joy/models/edlora_model-latest.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cace7f5bae7cd1acc54842f33cb94471941a493c17aabcffd72c63d2404d2b9
3
+ size 35173046
experiments/single-concept/joy/train_8339_joy_ortho_20240615_121643.log ADDED
Binary file (14.6 kB). View file
 
experiments/single-concept/linguini/6931_linguini_ortho.yml ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # GENERATE TIME: Sat Jun 15 13:29:45 2024
2
+ # CMD:
3
+ # train_edlora.py -opt single-concept/train_configs/6931_linguini_ortho.yml
4
+
5
+ name: 6931_linguini_ortho
6
+ manual_seed: 6931
7
+ mixed_precision: fp16
8
+ gradient_accumulation_steps: 1
9
+ datasets:
10
+ train:
11
+ name: LoraDataset
12
+ concept_list: single-concept/data_configs/linguini.json
13
+ use_caption: true
14
+ use_mask: true
15
+ instance_transform:
16
+ - type: HumanResizeCropFinalV3
17
+ size: 512
18
+ crop_p: 0.5
19
+ - type: ToTensor
20
+ - type: Normalize
21
+ mean:
22
+ - 0.5
23
+ std:
24
+ - 0.5
25
+ - type: ShuffleCaption
26
+ keep_token_num: 1
27
+ - type: EnhanceText
28
+ enhance_type: human
29
+ replace_mapping:
30
+ <TOK>: <linguini1> <linguini2>
31
+ batch_size_per_gpu: 2
32
+ dataset_enlarge_ratio: 500
33
+ val_vis:
34
+ name: PromptDataset
35
+ prompts: single-concept/validation_prompts/characters/test_man.txt
36
+ num_samples_per_prompt: 8
37
+ latent_size:
38
+ - 4
39
+ - 64
40
+ - 64
41
+ replace_mapping:
42
+ <TOK>: <linguini1> <linguini2>
43
+ batch_size_per_gpu: 4
44
+ models:
45
+ pretrained_path: nitrosocke/mo-di-diffusion
46
+ enable_edlora: true
47
+ finetune_cfg:
48
+ text_embedding:
49
+ enable_tuning: true
50
+ lr: 0.001
51
+ text_encoder:
52
+ enable_tuning: true
53
+ lora_cfg:
54
+ rank: 5
55
+ alpha: 1.0
56
+ where: CLIPAttention
57
+ lr: 1.0e-05
58
+ unet:
59
+ enable_tuning: true
60
+ lora_cfg:
61
+ rank: 5
62
+ alpha: 1.0
63
+ where: Attention
64
+ lr: 0.0001
65
+ new_concept_token: <linguini1>+<linguini2>
66
+ initializer_token: <rand-0.013>+man
67
+ noise_offset: 0.01
68
+ attn_reg_weight: 0.01
69
+ reg_full_identity: false
70
+ use_mask_loss: true
71
+ gradient_checkpoint: false
72
+ enable_xformers: true
73
+ path:
74
+ pretrain_network: null
75
+ train:
76
+ optim_g:
77
+ type: AdamW
78
+ lr: 0.0
79
+ weight_decay: 0.01
80
+ betas:
81
+ - 0.9
82
+ - 0.999
83
+ unet_kv_drop_rate: 0
84
+ scheduler: linear
85
+ emb_norm_threshold: 0.55
86
+ val:
87
+ val_during_save: false
88
+ compose_visualize: false
89
+ alpha_list:
90
+ - 0
91
+ - 0.7
92
+ - 1.0
93
+ sample:
94
+ num_inference_steps: 50
95
+ guidance_scale: 7.5
96
+ logger:
97
+ print_freq: 10
98
+ save_checkpoint_freq: 10000.0
experiments/single-concept/linguini/models/edlora_model-latest.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec394fe552cbe0f1128f1c6b7ceda9517f71d48a08ebf317e4ad879db171bdab
3
+ size 35173046
experiments/single-concept/linguini/train_6931_linguini_ortho_20240615_132945.log ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2024-06-15 13:29:45,503 INFO: Distributed environment: MULTI_GPU Backend: nccl
2
+ Num processes: 2
3
+ Process index: 0
4
+ Local process index: 0
5
+ Device: cuda:0
6
+
7
+ Mixed precision type: fp16
8
+
9
+ 2024-06-15 13:29:45,505 INFO:
10
+ name: 6931_linguini_ortho
11
+ manual_seed: 6931
12
+ mixed_precision: fp16
13
+ gradient_accumulation_steps: 1
14
+ datasets:[
15
+ train:[
16
+ name: LoraDataset
17
+ concept_list: single-concept/data_configs/linguini.json
18
+ use_caption: True
19
+ use_mask: True
20
+ instance_transform: [{'type': 'HumanResizeCropFinalV3', 'size': 512, 'crop_p': 0.5}, {'type': 'ToTensor'}, {'type': 'Normalize', 'mean': [0.5], 'std': [0.5]}, {'type': 'ShuffleCaption', 'keep_token_num': 1}, {'type': 'EnhanceText', 'enhance_type': 'human'}]
21
+ replace_mapping:[
22
+ <TOK>: <linguini1> <linguini2>
23
+ ]
24
+ batch_size_per_gpu: 2
25
+ dataset_enlarge_ratio: 500
26
+ ]
27
+ val_vis:[
28
+ name: PromptDataset
29
+ prompts: single-concept/validation_prompts/characters/test_man.txt
30
+ num_samples_per_prompt: 8
31
+ latent_size: [4, 64, 64]
32
+ replace_mapping:[
33
+ <TOK>: <linguini1> <linguini2>
34
+ ]
35
+ batch_size_per_gpu: 4
36
+ ]
37
+ ]
38
+ models:[
39
+ pretrained_path: nitrosocke/mo-di-diffusion
40
+ enable_edlora: True
41
+ finetune_cfg:[
42
+ text_embedding:[
43
+ enable_tuning: True
44
+ lr: 0.001
45
+ ]
46
+ text_encoder:[
47
+ enable_tuning: True
48
+ lora_cfg:[
49
+ rank: 5
50
+ alpha: 1.0
51
+ where: CLIPAttention
52
+ ]
53
+ lr: 1e-05
54
+ ]
55
+ unet:[
56
+ enable_tuning: True
57
+ lora_cfg:[
58
+ rank: 5
59
+ alpha: 1.0
60
+ where: Attention
61
+ ]
62
+ lr: 0.0001
63
+ ]
64
+ ]
65
+ new_concept_token: <linguini1>+<linguini2>
66
+ initializer_token: <rand-0.013>+man
67
+ noise_offset: 0.01
68
+ attn_reg_weight: 0.01
69
+ reg_full_identity: False
70
+ use_mask_loss: True
71
+ gradient_checkpoint: False
72
+ enable_xformers: True
73
+ ]
74
+ path:[
75
+ pretrain_network: None
76
+ experiments_root: /home/ujinsong/workspace/ortha/experiments/6931_linguini_ortho
77
+ models: /home/ujinsong/workspace/ortha/experiments/6931_linguini_ortho/models
78
+ log: /home/ujinsong/workspace/ortha/experiments/6931_linguini_ortho
79
+ visualization: /home/ujinsong/workspace/ortha/experiments/6931_linguini_ortho/visualization
80
+ ]
81
+ train:[
82
+ optim_g:[
83
+ type: AdamW
84
+ lr: 0.0
85
+ weight_decay: 0.01
86
+ betas: [0.9, 0.999]
87
+ ]
88
+ unet_kv_drop_rate: 0
89
+ scheduler: linear
90
+ emb_norm_threshold: 0.55
91
+ ]
92
+ val:[
93
+ val_during_save: False
94
+ compose_visualize: False
95
+ alpha_list: [0, 0.7, 1.0]
96
+ sample:[
97
+ num_inference_steps: 50
98
+ guidance_scale: 7.5
99
+ ]
100
+ ]
101
+ logger:[
102
+ print_freq: 10
103
+ save_checkpoint_freq: 10000.0
104
+ ]
105
+ is_train: True
106
+
107
+ 2024-06-15 13:30:00,748 INFO: <linguini1> (49408-49423) is random initialized by: <rand-0.013>
108
+ 2024-06-15 13:30:01,143 INFO: <linguini2> (49424-49439) is random initialized by existing token (man): 786
109
+ 2024-06-15 13:30:01,148 INFO: optimizing embedding using lr: 0.001
110
+ 2024-06-15 13:30:01,366 INFO: optimizing text_encoder (48 LoRAs), using lr: 1e-05
111
+ 2024-06-15 13:30:02,126 INFO: optimizing unet (128 LoRAs), using lr: 0.0001
112
+ 2024-06-15 13:30:24,806 INFO: ***** Running training *****
113
+ 2024-06-15 13:30:24,808 INFO: Num examples = 3000
114
+ 2024-06-15 13:30:24,809 INFO: Instantaneous batch size per device = 2
115
+ 2024-06-15 13:30:24,810 INFO: Total train batch size (w. parallel, distributed & accumulation) = 4
116
+ 2024-06-15 13:30:24,810 INFO: Total optimization steps = 750.0
117
+ 2024-06-15 13:31:49,284 INFO: [6931_..][Iter: 10, lr:(9.867e-04,9.867e-06,9.867e-05,)] [eta: 1:34:35] loss: 1.7287e+00 Norm_mean: 3.7056e-01
118
+ 2024-06-15 13:32:16,449 INFO: [6931_..][Iter: 20, lr:(9.733e-04,9.733e-06,9.733e-05,)] [eta: 1:04:35] loss: 6.2218e-01 Norm_mean: 3.8719e-01
119
+ 2024-06-15 13:32:27,804 INFO: [6931_..][Iter: 30, lr:(9.600e-04,9.600e-06,9.600e-05,)] [eta: 0:47:32] loss: 2.2293e-01 Norm_mean: 3.9956e-01
120
+ 2024-06-15 13:32:39,110 INFO: [6931_..][Iter: 40, lr:(9.467e-04,9.467e-06,9.467e-05,)] [eta: 0:38:42] loss: 3.6020e-01 Norm_mean: 4.0824e-01
121
+ 2024-06-15 13:32:50,393 INFO: [6931_..][Iter: 50, lr:(9.333e-04,9.333e-06,9.333e-05,)] [eta: 0:33:15] loss: 1.9713e-01 Norm_mean: 4.1583e-01
122
+ 2024-06-15 13:33:01,693 INFO: [6931_..][Iter: 60, lr:(9.200e-04,9.200e-06,9.200e-05,)] [eta: 0:29:31] loss: 8.7788e-01 Norm_mean: 4.2170e-01
123
+ 2024-06-15 13:33:13,121 INFO: [6931_..][Iter: 70, lr:(9.067e-04,9.067e-06,9.067e-05,)] [eta: 0:26:49] loss: 9.8421e-01 Norm_mean: 4.2614e-01
124
+ 2024-06-15 13:33:24,563 INFO: [6931_..][Iter: 80, lr:(8.933e-04,8.933e-06,8.933e-05,)] [eta: 0:24:44] loss: 3.0400e-02 Norm_mean: 4.3180e-01
125
+ 2024-06-15 13:33:35,986 INFO: [6931_..][Iter: 90, lr:(8.800e-04,8.800e-06,8.800e-05,)] [eta: 0:23:04] loss: 2.8843e-01 Norm_mean: 4.3760e-01
126
+ 2024-06-15 13:33:47,435 INFO: [6931_..][Iter: 100, lr:(8.667e-04,8.667e-06,8.667e-05,)] [eta: 0:21:42] loss: 2.8393e-01 Norm_mean: 4.4246e-01
127
+ 2024-06-15 13:33:58,752 INFO: [6931_..][Iter: 110, lr:(8.533e-04,8.533e-06,8.533e-05,)] [eta: 0:20:31] loss: 4.6023e-01 Norm_mean: 4.4686e-01
128
+ 2024-06-15 13:34:10,207 INFO: [6931_..][Iter: 120, lr:(8.400e-04,8.400e-06,8.400e-05,)] [eta: 0:19:31] loss: 9.5684e-01 Norm_mean: 4.5143e-01
129
+ 2024-06-15 13:34:21,286 INFO: [6931_..][Iter: 130, lr:(8.267e-04,8.267e-06,8.267e-05,)] [eta: 0:18:37] loss: 2.9791e-01 Norm_mean: 4.5649e-01
130
+ 2024-06-15 13:34:32,602 INFO: [6931_..][Iter: 140, lr:(8.133e-04,8.133e-06,8.133e-05,)] [eta: 0:17:50] loss: 1.8953e-01 Norm_mean: 4.6124e-01
131
+ 2024-06-15 13:34:43,992 INFO: [6931_..][Iter: 150, lr:(8.000e-04,8.000e-06,8.000e-05,)] [eta: 0:17:08] loss: 9.1619e-02 Norm_mean: 4.6565e-01
132
+ 2024-06-15 13:34:55,375 INFO: [6931_..][Iter: 160, lr:(7.867e-04,7.867e-06,7.867e-05,)] [eta: 0:16:29] loss: 5.2668e-01 Norm_mean: 4.6892e-01
133
+ 2024-06-15 13:35:06,680 INFO: [6931_..][Iter: 170, lr:(7.733e-04,7.733e-06,7.733e-05,)] [eta: 0:15:54] loss: 1.8469e-01 Norm_mean: 4.7256e-01
134
+ 2024-06-15 13:35:18,009 INFO: [6931_..][Iter: 180, lr:(7.600e-04,7.600e-06,7.600e-05,)] [eta: 0:15:21] loss: 8.5297e-01 Norm_mean: 4.7638e-01
135
+ 2024-06-15 13:35:29,336 INFO: [6931_..][Iter: 190, lr:(7.467e-04,7.467e-06,7.467e-05,)] [eta: 0:14:51] loss: 2.5142e-01 Norm_mean: 4.7967e-01
136
+ 2024-06-15 13:35:40,717 INFO: [6931_..][Iter: 200, lr:(7.333e-04,7.333e-06,7.333e-05,)] [eta: 0:14:22] loss: 7.4884e-01 Norm_mean: 4.8255e-01
137
+ 2024-06-15 13:35:51,924 INFO: [6931_..][Iter: 210, lr:(7.200e-04,7.200e-06,7.200e-05,)] [eta: 0:13:55] loss: 1.7828e-01 Norm_mean: 4.8658e-01
138
+ 2024-06-15 13:36:03,290 INFO: [6931_..][Iter: 220, lr:(7.067e-04,7.067e-06,7.067e-05,)] [eta: 0:13:30] loss: 2.2186e-01 Norm_mean: 4.8997e-01
139
+ 2024-06-15 13:36:14,559 INFO: [6931_..][Iter: 230, lr:(6.933e-04,6.933e-06,6.933e-05,)] [eta: 0:13:05] loss: 6.7478e-01 Norm_mean: 4.9278e-01
140
+ 2024-06-15 13:36:25,831 INFO: [6931_..][Iter: 240, lr:(6.800e-04,6.800e-06,6.800e-05,)] [eta: 0:12:42] loss: 1.3032e+00 Norm_mean: 4.9667e-01
141
+ 2024-06-15 13:36:37,152 INFO: [6931_..][Iter: 250, lr:(6.667e-04,6.667e-06,6.667e-05,)] [eta: 0:12:20] loss: 6.4526e-01 Norm_mean: 5.0024e-01
142
+ 2024-06-15 13:36:48,413 INFO: [6931_..][Iter: 260, lr:(6.533e-04,6.533e-06,6.533e-05,)] [eta: 0:11:58] loss: 2.5206e-01 Norm_mean: 5.0346e-01
143
+ 2024-06-15 13:36:59,644 INFO: [6931_..][Iter: 270, lr:(6.400e-04,6.400e-06,6.400e-05,)] [eta: 0:11:37] loss: 7.8783e-01 Norm_mean: 5.0614e-01
144
+ 2024-06-15 13:37:10,854 INFO: [6931_..][Iter: 280, lr:(6.267e-04,6.267e-06,6.267e-05,)] [eta: 0:11:17] loss: 5.4203e-01 Norm_mean: 5.0837e-01
145
+ 2024-06-15 13:37:22,219 INFO: [6931_..][Iter: 290, lr:(6.133e-04,6.133e-06,6.133e-05,)] [eta: 0:10:58] loss: 1.1699e+00 Norm_mean: 5.1131e-01
146
+ 2024-06-15 13:37:33,628 INFO: [6931_..][Iter: 300, lr:(6.000e-04,6.000e-06,6.000e-05,)] [eta: 0:10:39] loss: 5.1344e-01 Norm_mean: 5.1467e-01
147
+ 2024-06-15 13:37:44,756 INFO: [6931_..][Iter: 310, lr:(5.867e-04,5.867e-06,5.867e-05,)] [eta: 0:10:21] loss: 6.6832e-01 Norm_mean: 5.1752e-01
148
+ 2024-06-15 13:37:56,059 INFO: [6931_..][Iter: 320, lr:(5.733e-04,5.733e-06,5.733e-05,)] [eta: 0:10:03] loss: 1.3929e+00 Norm_mean: 5.1997e-01
149
+ 2024-06-15 13:38:07,378 INFO: [6931_..][Iter: 330, lr:(5.600e-04,5.600e-06,5.600e-05,)] [eta: 0:09:45] loss: 3.1073e-01 Norm_mean: 5.2261e-01
150
+ 2024-06-15 13:38:18,699 INFO: [6931_..][Iter: 340, lr:(5.467e-04,5.467e-06,5.467e-05,)] [eta: 0:09:28] loss: 3.0713e-02 Norm_mean: 5.2523e-01
151
+ 2024-06-15 13:38:30,001 INFO: [6931_..][Iter: 350, lr:(5.333e-04,5.333e-06,5.333e-05,)] [eta: 0:09:11] loss: 1.3819e-01 Norm_mean: 5.2838e-01
152
+ 2024-06-15 13:38:41,195 INFO: [6931_..][Iter: 360, lr:(5.200e-04,5.200e-06,5.200e-05,)] [eta: 0:08:54] loss: 1.8907e-01 Norm_mean: 5.3182e-01
153
+ 2024-06-15 13:38:52,657 INFO: [6931_..][Iter: 370, lr:(5.067e-04,5.067e-06,5.067e-05,)] [eta: 0:08:38] loss: 1.8054e+00 Norm_mean: 5.3477e-01
154
+ 2024-06-15 13:39:04,174 INFO: [6931_..][Iter: 380, lr:(4.933e-04,4.933e-06,4.933e-05,)] [eta: 0:08:23] loss: 1.0654e-01 Norm_mean: 5.3691e-01
155
+ 2024-06-15 13:39:15,535 INFO: [6931_..][Iter: 390, lr:(4.800e-04,4.800e-06,4.800e-05,)] [eta: 0:08:07] loss: 5.7298e-01 Norm_mean: 5.3854e-01
156
+ 2024-06-15 13:39:26,859 INFO: [6931_..][Iter: 400, lr:(4.667e-04,4.667e-06,4.667e-05,)] [eta: 0:07:51] loss: 1.6251e+00 Norm_mean: 5.3993e-01
157
+ 2024-06-15 13:39:38,127 INFO: [6931_..][Iter: 410, lr:(4.533e-04,4.533e-06,4.533e-05,)] [eta: 0:07:36] loss: 2.7363e-01 Norm_mean: 5.4159e-01
158
+ 2024-06-15 13:39:49,457 INFO: [6931_..][Iter: 420, lr:(4.400e-04,4.400e-06,4.400e-05,)] [eta: 0:07:21] loss: 1.0843e+00 Norm_mean: 5.4319e-01
159
+ 2024-06-15 13:40:00,654 INFO: [6931_..][Iter: 430, lr:(4.267e-04,4.267e-06,4.267e-05,)] [eta: 0:07:06] loss: 1.5981e+00 Norm_mean: 5.4475e-01
160
+ 2024-06-15 13:40:11,922 INFO: [6931_..][Iter: 440, lr:(4.133e-04,4.133e-06,4.133e-05,)] [eta: 0:06:51] loss: 3.5640e-01 Norm_mean: 5.4615e-01
161
+ 2024-06-15 13:40:23,216 INFO: [6931_..][Iter: 450, lr:(4.000e-04,4.000e-06,4.000e-05,)] [eta: 0:06:36] loss: 1.4516e-01 Norm_mean: 5.4760e-01
162
+ 2024-06-15 13:40:34,528 INFO: [6931_..][Iter: 460, lr:(3.867e-04,3.867e-06,3.867e-05,)] [eta: 0:06:22] loss: 2.3089e-02 Norm_mean: 5.4894e-01
163
+ 2024-06-15 13:40:45,727 INFO: [6931_..][Iter: 470, lr:(3.733e-04,3.733e-06,3.733e-05,)] [eta: 0:06:07] loss: 8.1703e-01 Norm_mean: 5.4999e-01
164
+ 2024-06-15 13:40:57,155 INFO: [6931_..][Iter: 480, lr:(3.600e-04,3.600e-06,3.600e-05,)] [eta: 0:05:53] loss: 2.9238e-01 Norm_mean: 5.5008e-01
165
+ 2024-06-15 13:41:08,445 INFO: [6931_..][Iter: 490, lr:(3.467e-04,3.467e-06,3.467e-05,)] [eta: 0:05:39] loss: 2.1595e-01 Norm_mean: 5.5008e-01
166
+ 2024-06-15 13:41:19,674 INFO: [6931_..][Iter: 500, lr:(3.333e-04,3.333e-06,3.333e-05,)] [eta: 0:05:25] loss: 3.1869e-01 Norm_mean: 5.5008e-01
167
+ 2024-06-15 13:41:30,940 INFO: [6931_..][Iter: 510, lr:(3.200e-04,3.200e-06,3.200e-05,)] [eta: 0:05:11] loss: 6.6768e-01 Norm_mean: 5.5008e-01
168
+ 2024-06-15 13:41:42,194 INFO: [6931_..][Iter: 520, lr:(3.067e-04,3.067e-06,3.067e-05,)] [eta: 0:04:57] loss: 4.5661e-01 Norm_mean: 5.5008e-01
169
+ 2024-06-15 13:41:53,545 INFO: [6931_..][Iter: 530, lr:(2.933e-04,2.933e-06,2.933e-05,)] [eta: 0:04:44] loss: 5.0165e-01 Norm_mean: 5.5008e-01
170
+ 2024-06-15 13:42:04,899 INFO: [6931_..][Iter: 540, lr:(2.800e-04,2.800e-06,2.800e-05,)] [eta: 0:04:30] loss: 7.3532e-01 Norm_mean: 5.5008e-01
171
+ 2024-06-15 13:42:16,246 INFO: [6931_..][Iter: 550, lr:(2.667e-04,2.667e-06,2.667e-05,)] [eta: 0:04:16] loss: 3.1165e-01 Norm_mean: 5.5008e-01
172
+ 2024-06-15 13:42:27,595 INFO: [6931_..][Iter: 560, lr:(2.533e-04,2.533e-06,2.533e-05,)] [eta: 0:04:03] loss: 1.0138e+00 Norm_mean: 5.5008e-01
173
+ 2024-06-15 13:42:38,849 INFO: [6931_..][Iter: 570, lr:(2.400e-04,2.400e-06,2.400e-05,)] [eta: 0:03:50] loss: 6.9475e-02 Norm_mean: 5.5008e-01
174
+ 2024-06-15 13:42:50,064 INFO: [6931_..][Iter: 580, lr:(2.267e-04,2.267e-06,2.267e-05,)] [eta: 0:03:36] loss: 7.4352e-02 Norm_mean: 5.5008e-01
175
+ 2024-06-15 13:43:01,460 INFO: [6931_..][Iter: 590, lr:(2.133e-04,2.133e-06,2.133e-05,)] [eta: 0:03:23] loss: 6.7372e-01 Norm_mean: 5.5008e-01
176
+ 2024-06-15 13:43:13,528 INFO: [6931_..][Iter: 600, lr:(2.000e-04,2.000e-06,2.000e-05,)] [eta: 0:03:10] loss: 2.8489e-01 Norm_mean: 5.5008e-01
177
+ 2024-06-15 13:43:26,895 INFO: [6931_..][Iter: 610, lr:(1.867e-04,1.867e-06,1.867e-05,)] [eta: 0:02:57] loss: 7.2095e-01 Norm_mean: 5.5008e-01
178
+ 2024-06-15 13:43:38,258 INFO: [6931_..][Iter: 620, lr:(1.733e-04,1.733e-06,1.733e-05,)] [eta: 0:02:44] loss: 1.7169e+00 Norm_mean: 5.5008e-01
179
+ 2024-06-15 13:43:49,680 INFO: [6931_..][Iter: 630, lr:(1.600e-04,1.600e-06,1.600e-05,)] [eta: 0:02:31] loss: 6.7559e-02 Norm_mean: 5.5008e-01
180
+ 2024-06-15 13:44:01,143 INFO: [6931_..][Iter: 640, lr:(1.467e-04,1.467e-06,1.467e-05,)] [eta: 0:02:18] loss: 2.7285e-01 Norm_mean: 5.5008e-01
181
+ 2024-06-15 13:44:12,596 INFO: [6931_..][Iter: 650, lr:(1.333e-04,1.333e-06,1.333e-05,)] [eta: 0:02:05] loss: 4.9240e-01 Norm_mean: 5.5008e-01
182
+ 2024-06-15 13:44:26,318 INFO: [6931_..][Iter: 660, lr:(1.200e-04,1.200e-06,1.200e-05,)] [eta: 0:01:53] loss: 1.8557e-01 Norm_mean: 5.5008e-01
183
+ 2024-06-15 13:44:37,700 INFO: [6931_..][Iter: 670, lr:(1.067e-04,1.067e-06,1.067e-05,)] [eta: 0:01:40] loss: 1.5337e+00 Norm_mean: 5.5008e-01
184
+ 2024-06-15 13:44:49,155 INFO: [6931_..][Iter: 680, lr:(9.333e-05,9.333e-07,9.333e-06,)] [eta: 0:01:27] loss: 3.5969e-01 Norm_mean: 5.5008e-01
185
+ 2024-06-15 13:45:00,525 INFO: [6931_..][Iter: 690, lr:(8.000e-05,8.000e-07,8.000e-06,)] [eta: 0:01:14] loss: 5.1855e-01 Norm_mean: 5.5008e-01
186
+ 2024-06-15 13:45:11,983 INFO: [6931_..][Iter: 700, lr:(6.667e-05,6.667e-07,6.667e-06,)] [eta: 0:01:02] loss: 4.5046e-01 Norm_mean: 5.5008e-01
187
+ 2024-06-15 13:45:23,196 INFO: [6931_..][Iter: 710, lr:(5.333e-05,5.333e-07,5.333e-06,)] [eta: 0:00:49] loss: 3.2902e-01 Norm_mean: 5.5008e-01
188
+ 2024-06-15 13:45:34,609 INFO: [6931_..][Iter: 720, lr:(4.000e-05,4.000e-07,4.000e-06,)] [eta: 0:00:36] loss: 7.7347e-01 Norm_mean: 5.5008e-01
189
+ 2024-06-15 13:45:45,893 INFO: [6931_..][Iter: 730, lr:(2.667e-05,2.667e-07,2.667e-06,)] [eta: 0:00:23] loss: 4.2478e-01 Norm_mean: 5.5008e-01
190
+ 2024-06-15 13:45:57,289 INFO: [6931_..][Iter: 740, lr:(1.333e-05,1.333e-07,1.333e-06,)] [eta: 0:00:11] loss: 6.1343e-01 Norm_mean: 5.5008e-01
191
+ 2024-06-15 13:46:08,602 INFO: [6931_..][Iter: 750, lr:(0.000e+00,0.000e+00,0.000e+00,)] [eta: -1 day, 23:59:59] loss: 2.7929e-01 Norm_mean: 5.5008e-01
192
+ 2024-06-15 13:46:08,738 INFO: Save state to /home/ujinsong/workspace/ortha/experiments/6931_linguini_ortho/models/edlora_model-latest.pth
experiments/single-concept/raya/7914_raya_ortho.yml ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # GENERATE TIME: Sat Jun 15 12:57:26 2024
2
+ # CMD:
3
+ # train_edlora.py -opt single-concept/train_configs/7914_raya_ortho.yml
4
+
5
+ name: 7914_raya_ortho
6
+ manual_seed: 7914
7
+ mixed_precision: fp16
8
+ gradient_accumulation_steps: 1
9
+ datasets:
10
+ train:
11
+ name: LoraDataset
12
+ concept_list: single-concept/data_configs/raya.json
13
+ use_caption: true
14
+ use_mask: true
15
+ instance_transform:
16
+ - type: HumanResizeCropFinalV3
17
+ size: 512
18
+ crop_p: 0.5
19
+ - type: ToTensor
20
+ - type: Normalize
21
+ mean:
22
+ - 0.5
23
+ std:
24
+ - 0.5
25
+ - type: ShuffleCaption
26
+ keep_token_num: 1
27
+ - type: EnhanceText
28
+ enhance_type: human
29
+ replace_mapping:
30
+ <TOK>: <raya1> <raya2>
31
+ batch_size_per_gpu: 2
32
+ dataset_enlarge_ratio: 500
33
+ val_vis:
34
+ name: PromptDataset
35
+ prompts: single-concept/validation_prompts/characters/test_woman.txt
36
+ num_samples_per_prompt: 8
37
+ latent_size:
38
+ - 4
39
+ - 64
40
+ - 64
41
+ replace_mapping:
42
+ <TOK>: <raya1> <raya2>
43
+ batch_size_per_gpu: 4
44
+ models:
45
+ pretrained_path: nitrosocke/mo-di-diffusion
46
+ enable_edlora: true
47
+ finetune_cfg:
48
+ text_embedding:
49
+ enable_tuning: true
50
+ lr: 0.001
51
+ text_encoder:
52
+ enable_tuning: true
53
+ lora_cfg:
54
+ rank: 5
55
+ alpha: 1.0
56
+ where: CLIPAttention
57
+ lr: 1.0e-05
58
+ unet:
59
+ enable_tuning: true
60
+ lora_cfg:
61
+ rank: 5
62
+ alpha: 1.0
63
+ where: Attention
64
+ lr: 0.0001
65
+ new_concept_token: <raya1>+<raya2>
66
+ initializer_token: <rand-0.013>+woman
67
+ noise_offset: 0.01
68
+ attn_reg_weight: 0.01
69
+ reg_full_identity: false
70
+ use_mask_loss: true
71
+ gradient_checkpoint: false
72
+ enable_xformers: true
73
+ path:
74
+ pretrain_network: null
75
+ train:
76
+ optim_g:
77
+ type: AdamW
78
+ lr: 0.0
79
+ weight_decay: 0.01
80
+ betas:
81
+ - 0.9
82
+ - 0.999
83
+ unet_kv_drop_rate: 0
84
+ scheduler: linear
85
+ emb_norm_threshold: 0.55
86
+ val:
87
+ val_during_save: false
88
+ compose_visualize: false
89
+ alpha_list:
90
+ - 0
91
+ - 0.7
92
+ - 1.0
93
+ sample:
94
+ num_inference_steps: 50
95
+ guidance_scale: 7.5
96
+ logger:
97
+ print_freq: 10
98
+ save_checkpoint_freq: 10000.0
experiments/single-concept/raya/models/edlora_model-latest.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f192a843a1d50c870bb056c4bd4d4a5b4fb54308026f8c027f55b7aa7c65fe16
3
+ size 35173046
experiments/single-concept/raya/train_7914_raya_ortho_20240615_125726.log ADDED
Binary file (14.7 kB). View file