aipracticecafe commited on
Commit
5681ebb
·
verified ·
1 Parent(s): 712aa35

Upload folder using huggingface_hub

Browse files
Files changed (25) hide show
  1. dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/config.json +93 -0
  2. dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/ema_model.safetensors +3 -0
  3. dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/model.safetensors +3 -0
  4. dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/optimizer.pt +3 -0
  5. dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/scheduler.pt +3 -0
  6. dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/config.json +93 -0
  7. dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/ema_model.safetensors +3 -0
  8. dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/model.safetensors +3 -0
  9. dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/optimizer.pt +3 -0
  10. dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/scheduler.pt +3 -0
  11. dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/config.json +93 -0
  12. dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/ema_model.safetensors +3 -0
  13. dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/model.safetensors +3 -0
  14. dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/optimizer.pt +3 -0
  15. dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/scheduler.pt +3 -0
  16. dual_stream_768_12_trans_2_llm_te_t5/epoch_800/config.json +93 -0
  17. dual_stream_768_12_trans_2_llm_te_t5/epoch_800/ema_model.safetensors +3 -0
  18. dual_stream_768_12_trans_2_llm_te_t5/epoch_800/model.safetensors +3 -0
  19. dual_stream_768_12_trans_2_llm_te_t5/epoch_800/optimizer.pt +3 -0
  20. dual_stream_768_12_trans_2_llm_te_t5/epoch_800/scheduler.pt +3 -0
  21. dual_stream_768_12_trans_2_llm_te_t5/epoch_900/config.json +93 -0
  22. dual_stream_768_12_trans_2_llm_te_t5/epoch_900/ema_model.safetensors +3 -0
  23. dual_stream_768_12_trans_2_llm_te_t5/epoch_900/model.safetensors +3 -0
  24. dual_stream_768_12_trans_2_llm_te_t5/epoch_900/optimizer.pt +3 -0
  25. dual_stream_768_12_trans_2_llm_te_t5/epoch_900/scheduler.pt +3 -0
dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/config.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 32,
5
+ "out_channels": 3,
6
+ "hidden_dim": 768,
7
+ "num_layers": 3,
8
+ "ch_mult": 2,
9
+ "cross_attention_dim": 768,
10
+ "is_conditional": true,
11
+ "is_latents": true,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "v",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 72,
19
+ 123
20
+ ],
21
+ "max_seq_len": 123,
22
+ "train_output_only": false,
23
+ "name": "dual_stream_768_13_llm_te_t5_random_drop",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": "data/danbooru_eva_faces_v2_flux",
28
+ "image_path": "asuka_ref_clean.jpg",
29
+ "n_samples": 100000,
30
+ "batch_size": 128,
31
+ "num_workers": 4,
32
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
33
+ "compute_normalization": true,
34
+ "is_finetune": false,
35
+ "finetune_orig_ratio": 0.3,
36
+ "use_short_prompts": true,
37
+ "load_into_ram": true,
38
+ "epochs": 1200,
39
+ "lr": 0.0004,
40
+ "use_gradient_checkpointing": true,
41
+ "use_cpu_accumulator": false,
42
+ "device": "cuda",
43
+ "save_dir": "results",
44
+ "log_interval": 1,
45
+ "sample_interval": 10,
46
+ "perturb_t": 0.5,
47
+ "perturb_scale": 0.4,
48
+ "start_lr": 1e-06,
49
+ "end_lr": 0.0001,
50
+ "input_perturbation": 0.0,
51
+ "use_ema": true,
52
+ "ema_decay": 0.999,
53
+ "ema_start_epoch": 300,
54
+ "save_interval": 100,
55
+ "resume_from_checkpoint": "results/checkpoints/20261101-075507/epoch_750",
56
+ "use_bitsandbytes": true,
57
+ "cfg_dropout_prob": 0.1,
58
+ "shuffle_tags": true,
59
+ "tag_dropout_prob": 0.1,
60
+ "wd": 0.02,
61
+ "gradient_accumulation_steps": 2,
62
+ "use_cos_scheduler": false,
63
+ "ignore_checkpoint_scheduler": false,
64
+ "ignore_checkpoint_optimizer": false,
65
+ "skip_warmup": false,
66
+ "final_lr_frac": 0.1,
67
+ "is_pixel_training": false,
68
+ "sampler_type": "dpm-solver",
69
+ "timestep_fn": "logit-normal",
70
+ "use_ot": true,
71
+ "train_shift": 1.0,
72
+ "sample_shift": 3.0,
73
+ "clip_prediction": false,
74
+ "use_cfm": false,
75
+ "depth": 12,
76
+ "num_heads": 12,
77
+ "dropout": 0.1,
78
+ "compile_model": true,
79
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
80
+ "use_pos": true,
81
+ "cfg_mask_prob": 0.0,
82
+ "drop_ratio": 0.0,
83
+ "use_i1": true,
84
+ "use_rope_text_adapter": false,
85
+ "hf_text_encoder": "google/t5gemma-2-270m-270m",
86
+ "skip_checkpointing_layers": 6,
87
+ "patch_size": 2,
88
+ "use_pixel_decoder": false,
89
+ "sample_file": "sample.toml",
90
+ "cfg_scale": 6.0,
91
+ "dynamic_thresholding": false,
92
+ "projection_dim": 2
93
+ }
dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/ema_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02b1da0b23198599c15d5aff42b4fbacc7abd97364aacbb5e4e0c934fd68015f
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66b4c6def765e49cc1758bc991667d0006e67c7c9bbd7b69c73a5858bd21b9b5
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25ea6c0f56073b1c86eca643ae14980ddfee8414b7e10160b022ea7d24c41d56
3
+ size 436537805
dual_stream_768_12_trans_2_llm_te_t5/epoch_1000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b1ee4c067b2adb22410d8c459698a38a6313d0e5e7c16b1ae952d44d40a11c9
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/config.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 32,
5
+ "out_channels": 3,
6
+ "hidden_dim": 768,
7
+ "num_layers": 3,
8
+ "ch_mult": 2,
9
+ "cross_attention_dim": 768,
10
+ "is_conditional": true,
11
+ "is_latents": true,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "v",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 72,
19
+ 123
20
+ ],
21
+ "max_seq_len": 123,
22
+ "train_output_only": false,
23
+ "name": "dual_stream_768_13_llm_te_t5_random_drop",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": "data/danbooru_eva_faces_v2_flux",
28
+ "image_path": "asuka_ref_clean.jpg",
29
+ "n_samples": 100000,
30
+ "batch_size": 128,
31
+ "num_workers": 4,
32
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
33
+ "compute_normalization": true,
34
+ "is_finetune": false,
35
+ "finetune_orig_ratio": 0.3,
36
+ "use_short_prompts": true,
37
+ "load_into_ram": true,
38
+ "epochs": 1200,
39
+ "lr": 0.0004,
40
+ "use_gradient_checkpointing": true,
41
+ "use_cpu_accumulator": false,
42
+ "device": "cuda",
43
+ "save_dir": "results",
44
+ "log_interval": 1,
45
+ "sample_interval": 10,
46
+ "perturb_t": 0.5,
47
+ "perturb_scale": 0.4,
48
+ "start_lr": 1e-06,
49
+ "end_lr": 0.0001,
50
+ "input_perturbation": 0.0,
51
+ "use_ema": true,
52
+ "ema_decay": 0.999,
53
+ "ema_start_epoch": 300,
54
+ "save_interval": 100,
55
+ "resume_from_checkpoint": "results/checkpoints/20261101-075507/epoch_750",
56
+ "use_bitsandbytes": true,
57
+ "cfg_dropout_prob": 0.1,
58
+ "shuffle_tags": true,
59
+ "tag_dropout_prob": 0.1,
60
+ "wd": 0.02,
61
+ "gradient_accumulation_steps": 2,
62
+ "use_cos_scheduler": false,
63
+ "ignore_checkpoint_scheduler": false,
64
+ "ignore_checkpoint_optimizer": false,
65
+ "skip_warmup": false,
66
+ "final_lr_frac": 0.1,
67
+ "is_pixel_training": false,
68
+ "sampler_type": "dpm-solver",
69
+ "timestep_fn": "logit-normal",
70
+ "use_ot": true,
71
+ "train_shift": 1.0,
72
+ "sample_shift": 3.0,
73
+ "clip_prediction": false,
74
+ "use_cfm": false,
75
+ "depth": 12,
76
+ "num_heads": 12,
77
+ "dropout": 0.1,
78
+ "compile_model": true,
79
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
80
+ "use_pos": true,
81
+ "cfg_mask_prob": 0.0,
82
+ "drop_ratio": 0.0,
83
+ "use_i1": true,
84
+ "use_rope_text_adapter": false,
85
+ "hf_text_encoder": "google/t5gemma-2-270m-270m",
86
+ "skip_checkpointing_layers": 6,
87
+ "patch_size": 2,
88
+ "use_pixel_decoder": false,
89
+ "sample_file": "sample.toml",
90
+ "cfg_scale": 6.0,
91
+ "dynamic_thresholding": false,
92
+ "projection_dim": 2
93
+ }
dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/ema_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21b7ea688611fd6afe831bfa6c724064cc991a8916c57725935d2ff3e3219500
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4dd1ed258472334783851563437e45d34bdf9c03800e95e5c10f35aada690ce
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34aa5af7192fc94a83204c3ce4b443247b125a97b7fb6cd0da019e94d1a1e7ab
3
+ size 436537805
dual_stream_768_12_trans_2_llm_te_t5/epoch_1100/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96c5ac2691e57abe36fb294407e06981c519a753d4523328d94e9d374a1e297b
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/config.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 32,
5
+ "out_channels": 3,
6
+ "hidden_dim": 768,
7
+ "num_layers": 3,
8
+ "ch_mult": 2,
9
+ "cross_attention_dim": 768,
10
+ "is_conditional": true,
11
+ "is_latents": true,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "v",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 72,
19
+ 123
20
+ ],
21
+ "max_seq_len": 123,
22
+ "train_output_only": false,
23
+ "name": "dual_stream_768_13_llm_te_t5_random_drop",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": "data/danbooru_eva_faces_v2_flux",
28
+ "image_path": "asuka_ref_clean.jpg",
29
+ "n_samples": 100000,
30
+ "batch_size": 128,
31
+ "num_workers": 4,
32
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
33
+ "compute_normalization": true,
34
+ "is_finetune": false,
35
+ "finetune_orig_ratio": 0.3,
36
+ "use_short_prompts": true,
37
+ "load_into_ram": true,
38
+ "epochs": 1200,
39
+ "lr": 0.0004,
40
+ "use_gradient_checkpointing": true,
41
+ "use_cpu_accumulator": false,
42
+ "device": "cuda",
43
+ "save_dir": "results",
44
+ "log_interval": 1,
45
+ "sample_interval": 10,
46
+ "perturb_t": 0.5,
47
+ "perturb_scale": 0.4,
48
+ "start_lr": 1e-06,
49
+ "end_lr": 0.0001,
50
+ "input_perturbation": 0.0,
51
+ "use_ema": true,
52
+ "ema_decay": 0.999,
53
+ "ema_start_epoch": 300,
54
+ "save_interval": 100,
55
+ "resume_from_checkpoint": "results/checkpoints/20261101-075507/epoch_750",
56
+ "use_bitsandbytes": true,
57
+ "cfg_dropout_prob": 0.1,
58
+ "shuffle_tags": true,
59
+ "tag_dropout_prob": 0.1,
60
+ "wd": 0.02,
61
+ "gradient_accumulation_steps": 2,
62
+ "use_cos_scheduler": false,
63
+ "ignore_checkpoint_scheduler": false,
64
+ "ignore_checkpoint_optimizer": false,
65
+ "skip_warmup": false,
66
+ "final_lr_frac": 0.1,
67
+ "is_pixel_training": false,
68
+ "sampler_type": "dpm-solver",
69
+ "timestep_fn": "logit-normal",
70
+ "use_ot": true,
71
+ "train_shift": 1.0,
72
+ "sample_shift": 3.0,
73
+ "clip_prediction": false,
74
+ "use_cfm": false,
75
+ "depth": 12,
76
+ "num_heads": 12,
77
+ "dropout": 0.1,
78
+ "compile_model": true,
79
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
80
+ "use_pos": true,
81
+ "cfg_mask_prob": 0.0,
82
+ "drop_ratio": 0.0,
83
+ "use_i1": true,
84
+ "use_rope_text_adapter": false,
85
+ "hf_text_encoder": "google/t5gemma-2-270m-270m",
86
+ "skip_checkpointing_layers": 6,
87
+ "patch_size": 2,
88
+ "use_pixel_decoder": false,
89
+ "sample_file": "sample.toml",
90
+ "cfg_scale": 6.0,
91
+ "dynamic_thresholding": false,
92
+ "projection_dim": 2
93
+ }
dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/ema_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05ca5ab2ab4c581593605903227a227c51b67e9fe89b1016333474d76f6d469e
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c5d452d2e9ec527bdf8c31086c0b78f70ecad3676183026725395f915605c2d
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b469da12553f89d783f8b9e2c183c76f781bb104584c0d54f627ced87a182078
3
+ size 436537805
dual_stream_768_12_trans_2_llm_te_t5/epoch_1200/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdc7ba97a4506f630fa8786b37fb93fa3f7ab61efd6cb6861c60edc15d869591
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_t5/epoch_800/config.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 32,
5
+ "out_channels": 3,
6
+ "hidden_dim": 768,
7
+ "num_layers": 3,
8
+ "ch_mult": 2,
9
+ "cross_attention_dim": 768,
10
+ "is_conditional": true,
11
+ "is_latents": true,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "v",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 72,
19
+ 123
20
+ ],
21
+ "max_seq_len": 123,
22
+ "train_output_only": false,
23
+ "name": "dual_stream_768_13_llm_te_t5_random_drop",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": "data/danbooru_eva_faces_v2_flux",
28
+ "image_path": "asuka_ref_clean.jpg",
29
+ "n_samples": 100000,
30
+ "batch_size": 128,
31
+ "num_workers": 4,
32
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
33
+ "compute_normalization": true,
34
+ "is_finetune": false,
35
+ "finetune_orig_ratio": 0.3,
36
+ "use_short_prompts": true,
37
+ "load_into_ram": true,
38
+ "epochs": 1200,
39
+ "lr": 0.0004,
40
+ "use_gradient_checkpointing": true,
41
+ "use_cpu_accumulator": false,
42
+ "device": "cuda",
43
+ "save_dir": "results",
44
+ "log_interval": 1,
45
+ "sample_interval": 10,
46
+ "perturb_t": 0.5,
47
+ "perturb_scale": 0.4,
48
+ "start_lr": 1e-06,
49
+ "end_lr": 0.0001,
50
+ "input_perturbation": 0.0,
51
+ "use_ema": true,
52
+ "ema_decay": 0.999,
53
+ "ema_start_epoch": 300,
54
+ "save_interval": 100,
55
+ "resume_from_checkpoint": "results/checkpoints/20261101-075507/epoch_750",
56
+ "use_bitsandbytes": true,
57
+ "cfg_dropout_prob": 0.1,
58
+ "shuffle_tags": true,
59
+ "tag_dropout_prob": 0.1,
60
+ "wd": 0.02,
61
+ "gradient_accumulation_steps": 2,
62
+ "use_cos_scheduler": false,
63
+ "ignore_checkpoint_scheduler": false,
64
+ "ignore_checkpoint_optimizer": false,
65
+ "skip_warmup": false,
66
+ "final_lr_frac": 0.1,
67
+ "is_pixel_training": false,
68
+ "sampler_type": "dpm-solver",
69
+ "timestep_fn": "logit-normal",
70
+ "use_ot": true,
71
+ "train_shift": 1.0,
72
+ "sample_shift": 3.0,
73
+ "clip_prediction": false,
74
+ "use_cfm": false,
75
+ "depth": 12,
76
+ "num_heads": 12,
77
+ "dropout": 0.1,
78
+ "compile_model": true,
79
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
80
+ "use_pos": true,
81
+ "cfg_mask_prob": 0.0,
82
+ "drop_ratio": 0.0,
83
+ "use_i1": true,
84
+ "use_rope_text_adapter": false,
85
+ "hf_text_encoder": "google/t5gemma-2-270m-270m",
86
+ "skip_checkpointing_layers": 6,
87
+ "patch_size": 2,
88
+ "use_pixel_decoder": false,
89
+ "sample_file": "sample.toml",
90
+ "cfg_scale": 6.0,
91
+ "dynamic_thresholding": false,
92
+ "projection_dim": 2
93
+ }
dual_stream_768_12_trans_2_llm_te_t5/epoch_800/ema_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:654426a4fdf360997e35e4712b38659c9bc080d8824c20c6ca02a34b13d83e51
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_800/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6feead7826321e3dd3f9b07ceb864527e785fd55dff5559a527a87791322308
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_800/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96cd02bddef5d7490060623c3ac89658d22ef34e0e83d1ea73497cd20a1e741e
3
+ size 436537805
dual_stream_768_12_trans_2_llm_te_t5/epoch_800/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a52ada1e4e76b9ae9c4f0b26e6d92a58e6beb288e27ffa870a91e57676abf24
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_t5/epoch_900/config.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 32,
5
+ "out_channels": 3,
6
+ "hidden_dim": 768,
7
+ "num_layers": 3,
8
+ "ch_mult": 2,
9
+ "cross_attention_dim": 768,
10
+ "is_conditional": true,
11
+ "is_latents": true,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "v",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 72,
19
+ 123
20
+ ],
21
+ "max_seq_len": 123,
22
+ "train_output_only": false,
23
+ "name": "dual_stream_768_13_llm_te_t5_random_drop",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": "data/danbooru_eva_faces_v2_flux",
28
+ "image_path": "asuka_ref_clean.jpg",
29
+ "n_samples": 100000,
30
+ "batch_size": 128,
31
+ "num_workers": 4,
32
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
33
+ "compute_normalization": true,
34
+ "is_finetune": false,
35
+ "finetune_orig_ratio": 0.3,
36
+ "use_short_prompts": true,
37
+ "load_into_ram": true,
38
+ "epochs": 1200,
39
+ "lr": 0.0004,
40
+ "use_gradient_checkpointing": true,
41
+ "use_cpu_accumulator": false,
42
+ "device": "cuda",
43
+ "save_dir": "results",
44
+ "log_interval": 1,
45
+ "sample_interval": 10,
46
+ "perturb_t": 0.5,
47
+ "perturb_scale": 0.4,
48
+ "start_lr": 1e-06,
49
+ "end_lr": 0.0001,
50
+ "input_perturbation": 0.0,
51
+ "use_ema": true,
52
+ "ema_decay": 0.999,
53
+ "ema_start_epoch": 300,
54
+ "save_interval": 100,
55
+ "resume_from_checkpoint": "results/checkpoints/20261101-075507/epoch_750",
56
+ "use_bitsandbytes": true,
57
+ "cfg_dropout_prob": 0.1,
58
+ "shuffle_tags": true,
59
+ "tag_dropout_prob": 0.1,
60
+ "wd": 0.02,
61
+ "gradient_accumulation_steps": 2,
62
+ "use_cos_scheduler": false,
63
+ "ignore_checkpoint_scheduler": false,
64
+ "ignore_checkpoint_optimizer": false,
65
+ "skip_warmup": false,
66
+ "final_lr_frac": 0.1,
67
+ "is_pixel_training": false,
68
+ "sampler_type": "dpm-solver",
69
+ "timestep_fn": "logit-normal",
70
+ "use_ot": true,
71
+ "train_shift": 1.0,
72
+ "sample_shift": 3.0,
73
+ "clip_prediction": false,
74
+ "use_cfm": false,
75
+ "depth": 12,
76
+ "num_heads": 12,
77
+ "dropout": 0.1,
78
+ "compile_model": true,
79
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
80
+ "use_pos": true,
81
+ "cfg_mask_prob": 0.0,
82
+ "drop_ratio": 0.0,
83
+ "use_i1": true,
84
+ "use_rope_text_adapter": false,
85
+ "hf_text_encoder": "google/t5gemma-2-270m-270m",
86
+ "skip_checkpointing_layers": 6,
87
+ "patch_size": 2,
88
+ "use_pixel_decoder": false,
89
+ "sample_file": "sample.toml",
90
+ "cfg_scale": 6.0,
91
+ "dynamic_thresholding": false,
92
+ "projection_dim": 2
93
+ }
dual_stream_768_12_trans_2_llm_te_t5/epoch_900/ema_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:588ec2d2e8d16d71379556434c14e33eaf6984741e12f61a9ba50130e001cd06
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_900/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfa1659e75b6ab44f799e78c44c6c300aa9f6a8ba074154a80f330aea0972414
3
+ size 438882976
dual_stream_768_12_trans_2_llm_te_t5/epoch_900/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a9e8895fd34750289947fccf8a19838edb12bfc2fa3efb05d47b85b4aaa2d8a
3
+ size 436537805
dual_stream_768_12_trans_2_llm_te_t5/epoch_900/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad5e518105411d38fba518cf9e4e880b4848c8980728197ae8d4465b6cb914fc
3
+ size 1657