Hugging Face's logo

aipracticecafe
/
toy-diffusion

Text-to-Image
Safetensors
English
Model card Files
xet
 Community
aipracticecafe commited on
Commit
93503bd
·
verified ·
1 Parent(s): 3bacfca

Upload folder using huggingface_hub

Browse files
Files changed (20) hide show
  1. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/config.json +94 -0
  2. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/model.safetensors +3 -0
  3. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/optimizer.pt +3 -0
  4. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/scheduler.pt +3 -0
  5. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/config.json +94 -0
  6. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/model.safetensors +3 -0
  7. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/optimizer.pt +3 -0
  8. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/scheduler.pt +3 -0
  9. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/config.json +94 -0
  10. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/model.safetensors +3 -0
  11. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/optimizer.pt +3 -0
  12. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/scheduler.pt +3 -0
  13. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/config.json +94 -0
  14. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/model.safetensors +3 -0
  15. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/optimizer.pt +3 -0
  16. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/scheduler.pt +3 -0
  17. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/config.json +94 -0
  18. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/model.safetensors +3 -0
  19. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/optimizer.pt +3 -0
  20. dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/scheduler.pt +3 -0
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/config.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 3,
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": false,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "x",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 78,
19
+ 135
20
+ ],
21
+ "max_seq_len": 135,
22
+ "train_output_only": true,
23
+ "name": "sprint_dual_stream_768_pixel_stage1_a40",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": [
28
+ "data/danbooru_faces",
29
+ "data/synthetic"
30
+ ],
31
+ "image_path": "asuka_ref_clean.jpg",
32
+ "n_samples": 100000,
33
+ "batch_size": 128,
34
+ "num_workers": 8,
35
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
36
+ "compute_normalization": false,
37
+ "is_finetune": false,
38
+ "finetune_orig_ratio": 0.3,
39
+ "use_short_prompts": false,
40
+ "load_into_ram": false,
41
+ "epochs": 25,
42
+ "lr": 0.0002,
43
+ "use_gradient_checkpointing": true,
44
+ "use_cpu_accumulator": false,
45
+ "device": "cuda",
46
+ "save_dir": "results",
47
+ "log_interval": 1,
48
+ "sample_interval": 2,
49
+ "perturb_t": 0.5,
50
+ "perturb_scale": 0.4,
51
+ "start_lr": 1e-06,
52
+ "end_lr": 0.0001,
53
+ "input_perturbation": 0.0,
54
+ "use_ema": false,
55
+ "ema_decay": 0.999,
56
+ "save_interval": 5,
57
+ "use_bitsandbytes": false,
58
+ "cfg_dropout_prob": 0.1,
59
+ "shuffle_tags": true,
60
+ "tag_dropout_prob": 0.1,
61
+ "wd": 0.02,
62
+ "gradient_accumulation_steps": 2,
63
+ "use_cos_scheduler": false,
64
+ "ignore_checkpoint_scheduler": true,
65
+ "ignore_checkpoint_optimizer": true,
66
+ "skip_warmup": false,
67
+ "final_lr_frac": 0.05,
68
+ "is_pixel_training": true,
69
+ "noise_scale": 2.0,
70
+ "sampler_type": "dpm-solver",
71
+ "timestep_fn": "logit-normal",
72
+ "use_ot": false,
73
+ "train_shift": 1.0,
74
+ "sample_shift": 6.93,
75
+ "clip_prediction": false,
76
+ "use_cfm": false,
77
+ "depth": 12,
78
+ "num_heads": 12,
79
+ "dropout": 0.1,
80
+ "compile_model": true,
81
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
82
+ "use_pos": true,
83
+ "cfg_mask_prob": 0.0,
84
+ "drop_ratio": 0.0,
85
+ "use_i1": true,
86
+ "use_rope_text_adapter": false,
87
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
88
+ "skip_checkpointing_layers": 4,
89
+ "patch_size": 16,
90
+ "use_pixel_decoder": true,
91
+ "sample_file": "sample.toml",
92
+ "cfg_scale": 4.5,
93
+ "projection_dim": 2
94
+ }
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e44458d1aa5b5f137e1fac67bc9f2eb71447938019dad068b941b9d48c8de79
3
+ size 24169478
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63e326e91fa9b98bd34914d61214b2f1492afb2725b9a8bfcc934398e8cf6609
3
+ size 96694027
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_10/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:189780bda0d5c779d75bb1f1e31ba130ea20dffce93cf8aa38f2d6e6356801ee
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/config.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 3,
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": false,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "x",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 78,
19
+ 135
20
+ ],
21
+ "max_seq_len": 135,
22
+ "train_output_only": true,
23
+ "name": "sprint_dual_stream_768_pixel_stage1_a40",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": [
28
+ "data/danbooru_faces",
29
+ "data/synthetic"
30
+ ],
31
+ "image_path": "asuka_ref_clean.jpg",
32
+ "n_samples": 100000,
33
+ "batch_size": 128,
34
+ "num_workers": 8,
35
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
36
+ "compute_normalization": false,
37
+ "is_finetune": false,
38
+ "finetune_orig_ratio": 0.3,
39
+ "use_short_prompts": false,
40
+ "load_into_ram": false,
41
+ "epochs": 25,
42
+ "lr": 0.0002,
43
+ "use_gradient_checkpointing": true,
44
+ "use_cpu_accumulator": false,
45
+ "device": "cuda",
46
+ "save_dir": "results",
47
+ "log_interval": 1,
48
+ "sample_interval": 2,
49
+ "perturb_t": 0.5,
50
+ "perturb_scale": 0.4,
51
+ "start_lr": 1e-06,
52
+ "end_lr": 0.0001,
53
+ "input_perturbation": 0.0,
54
+ "use_ema": false,
55
+ "ema_decay": 0.999,
56
+ "save_interval": 5,
57
+ "use_bitsandbytes": false,
58
+ "cfg_dropout_prob": 0.1,
59
+ "shuffle_tags": true,
60
+ "tag_dropout_prob": 0.1,
61
+ "wd": 0.02,
62
+ "gradient_accumulation_steps": 2,
63
+ "use_cos_scheduler": false,
64
+ "ignore_checkpoint_scheduler": true,
65
+ "ignore_checkpoint_optimizer": true,
66
+ "skip_warmup": false,
67
+ "final_lr_frac": 0.05,
68
+ "is_pixel_training": true,
69
+ "noise_scale": 2.0,
70
+ "sampler_type": "dpm-solver",
71
+ "timestep_fn": "logit-normal",
72
+ "use_ot": false,
73
+ "train_shift": 1.0,
74
+ "sample_shift": 6.93,
75
+ "clip_prediction": false,
76
+ "use_cfm": false,
77
+ "depth": 12,
78
+ "num_heads": 12,
79
+ "dropout": 0.1,
80
+ "compile_model": true,
81
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
82
+ "use_pos": true,
83
+ "cfg_mask_prob": 0.0,
84
+ "drop_ratio": 0.0,
85
+ "use_i1": true,
86
+ "use_rope_text_adapter": false,
87
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
88
+ "skip_checkpointing_layers": 4,
89
+ "patch_size": 16,
90
+ "use_pixel_decoder": true,
91
+ "sample_file": "sample.toml",
92
+ "cfg_scale": 4.5,
93
+ "projection_dim": 2
94
+ }
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e65b24be0cdaaffe3c739e937b4ed3e8eba06da38e0fab634a4903e7286e78ba
3
+ size 24169478
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4b8b47dc1517e17adedf1bf45871a6dea239944953ef94be28a27d2fc4249a84
3
+ size 96694027
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_15/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c7161657e37e5f5c8b892f18625f0a4ef4f6221d934d4615f51f8132cd7427d
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/config.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 3,
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": false,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "x",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 78,
19
+ 135
20
+ ],
21
+ "max_seq_len": 135,
22
+ "train_output_only": true,
23
+ "name": "sprint_dual_stream_768_pixel_stage1_a40",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": [
28
+ "data/danbooru_faces",
29
+ "data/synthetic"
30
+ ],
31
+ "image_path": "asuka_ref_clean.jpg",
32
+ "n_samples": 100000,
33
+ "batch_size": 128,
34
+ "num_workers": 8,
35
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
36
+ "compute_normalization": false,
37
+ "is_finetune": false,
38
+ "finetune_orig_ratio": 0.3,
39
+ "use_short_prompts": false,
40
+ "load_into_ram": false,
41
+ "epochs": 25,
42
+ "lr": 0.0002,
43
+ "use_gradient_checkpointing": true,
44
+ "use_cpu_accumulator": false,
45
+ "device": "cuda",
46
+ "save_dir": "results",
47
+ "log_interval": 1,
48
+ "sample_interval": 2,
49
+ "perturb_t": 0.5,
50
+ "perturb_scale": 0.4,
51
+ "start_lr": 1e-06,
52
+ "end_lr": 0.0001,
53
+ "input_perturbation": 0.0,
54
+ "use_ema": false,
55
+ "ema_decay": 0.999,
56
+ "save_interval": 5,
57
+ "use_bitsandbytes": false,
58
+ "cfg_dropout_prob": 0.1,
59
+ "shuffle_tags": true,
60
+ "tag_dropout_prob": 0.1,
61
+ "wd": 0.02,
62
+ "gradient_accumulation_steps": 2,
63
+ "use_cos_scheduler": false,
64
+ "ignore_checkpoint_scheduler": true,
65
+ "ignore_checkpoint_optimizer": true,
66
+ "skip_warmup": false,
67
+ "final_lr_frac": 0.05,
68
+ "is_pixel_training": true,
69
+ "noise_scale": 2.0,
70
+ "sampler_type": "dpm-solver",
71
+ "timestep_fn": "logit-normal",
72
+ "use_ot": false,
73
+ "train_shift": 1.0,
74
+ "sample_shift": 6.93,
75
+ "clip_prediction": false,
76
+ "use_cfm": false,
77
+ "depth": 12,
78
+ "num_heads": 12,
79
+ "dropout": 0.1,
80
+ "compile_model": true,
81
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
82
+ "use_pos": true,
83
+ "cfg_mask_prob": 0.0,
84
+ "drop_ratio": 0.0,
85
+ "use_i1": true,
86
+ "use_rope_text_adapter": false,
87
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
88
+ "skip_checkpointing_layers": 4,
89
+ "patch_size": 16,
90
+ "use_pixel_decoder": true,
91
+ "sample_file": "sample.toml",
92
+ "cfg_scale": 4.5,
93
+ "projection_dim": 2
94
+ }
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c4ac9138f09ea6045e28db961499519ac3039c1c0276168d7b252f424e58d6c2
3
+ size 24169478
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd8a7d08e28221692b1b59180f64dc85ff9952fa670a85c101fd34afec902b69
3
+ size 96694027
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_20/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78d4293b339598b5ce03f9ed2cc510121354d6601af679543b3dfd87b30f3d07
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/config.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 3,
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": false,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "x",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 78,
19
+ 135
20
+ ],
21
+ "max_seq_len": 135,
22
+ "train_output_only": true,
23
+ "name": "sprint_dual_stream_768_pixel_stage1_a40",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": [
28
+ "data/danbooru_faces",
29
+ "data/synthetic"
30
+ ],
31
+ "image_path": "asuka_ref_clean.jpg",
32
+ "n_samples": 100000,
33
+ "batch_size": 128,
34
+ "num_workers": 8,
35
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
36
+ "compute_normalization": false,
37
+ "is_finetune": false,
38
+ "finetune_orig_ratio": 0.3,
39
+ "use_short_prompts": false,
40
+ "load_into_ram": false,
41
+ "epochs": 25,
42
+ "lr": 0.0002,
43
+ "use_gradient_checkpointing": true,
44
+ "use_cpu_accumulator": false,
45
+ "device": "cuda",
46
+ "save_dir": "results",
47
+ "log_interval": 1,
48
+ "sample_interval": 2,
49
+ "perturb_t": 0.5,
50
+ "perturb_scale": 0.4,
51
+ "start_lr": 1e-06,
52
+ "end_lr": 0.0001,
53
+ "input_perturbation": 0.0,
54
+ "use_ema": false,
55
+ "ema_decay": 0.999,
56
+ "save_interval": 5,
57
+ "use_bitsandbytes": false,
58
+ "cfg_dropout_prob": 0.1,
59
+ "shuffle_tags": true,
60
+ "tag_dropout_prob": 0.1,
61
+ "wd": 0.02,
62
+ "gradient_accumulation_steps": 2,
63
+ "use_cos_scheduler": false,
64
+ "ignore_checkpoint_scheduler": true,
65
+ "ignore_checkpoint_optimizer": true,
66
+ "skip_warmup": false,
67
+ "final_lr_frac": 0.05,
68
+ "is_pixel_training": true,
69
+ "noise_scale": 2.0,
70
+ "sampler_type": "dpm-solver",
71
+ "timestep_fn": "logit-normal",
72
+ "use_ot": false,
73
+ "train_shift": 1.0,
74
+ "sample_shift": 6.93,
75
+ "clip_prediction": false,
76
+ "use_cfm": false,
77
+ "depth": 12,
78
+ "num_heads": 12,
79
+ "dropout": 0.1,
80
+ "compile_model": true,
81
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
82
+ "use_pos": true,
83
+ "cfg_mask_prob": 0.0,
84
+ "drop_ratio": 0.0,
85
+ "use_i1": true,
86
+ "use_rope_text_adapter": false,
87
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
88
+ "skip_checkpointing_layers": 4,
89
+ "patch_size": 16,
90
+ "use_pixel_decoder": true,
91
+ "sample_file": "sample.toml",
92
+ "cfg_scale": 4.5,
93
+ "projection_dim": 2
94
+ }
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d579dbdbaa923f331d729a271ba7532996bba506844c269601aef6a2d35c19d5
3
+ size 24169478
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82389377b3649adacf0eae911d9ba79323e872aea5ce4f7fea1fba049321656a
3
+ size 96694027
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_25/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:818d57ee61fa55849cdf0bd2b45a97f9b4209b27a65f4d4dcf9096f0db4f05b5
3
+ size 1657
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/config.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "dual_stream",
3
+ "model_type": "dual_stream",
4
+ "in_channels": 3,
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": false,
12
+ "vae_scale": 0.5885282298938924,
13
+ "vae_shift": 0.05106564902828182,
14
+ "schedule_type": "linear",
15
+ "prediction_target": "x",
16
+ "loss_target": "v",
17
+ "tiers_len": [
18
+ 78,
19
+ 135
20
+ ],
21
+ "max_seq_len": 135,
22
+ "train_output_only": true,
23
+ "name": "sprint_dual_stream_768_pixel_stage1_a40",
24
+ "mode": "train",
25
+ "seed": 1995,
26
+ "dataset_type": "kanji",
27
+ "data_path": [
28
+ "data/danbooru_faces",
29
+ "data/synthetic"
30
+ ],
31
+ "image_path": "asuka_ref_clean.jpg",
32
+ "n_samples": 100000,
33
+ "batch_size": 128,
34
+ "num_workers": 8,
35
+ "font_path": "/usr/share/fonts/opentype/ipafont-gothic/ipag.ttf",
36
+ "compute_normalization": false,
37
+ "is_finetune": false,
38
+ "finetune_orig_ratio": 0.3,
39
+ "use_short_prompts": false,
40
+ "load_into_ram": false,
41
+ "epochs": 25,
42
+ "lr": 0.0002,
43
+ "use_gradient_checkpointing": true,
44
+ "use_cpu_accumulator": false,
45
+ "device": "cuda",
46
+ "save_dir": "results",
47
+ "log_interval": 1,
48
+ "sample_interval": 2,
49
+ "perturb_t": 0.5,
50
+ "perturb_scale": 0.4,
51
+ "start_lr": 1e-06,
52
+ "end_lr": 0.0001,
53
+ "input_perturbation": 0.0,
54
+ "use_ema": false,
55
+ "ema_decay": 0.999,
56
+ "save_interval": 5,
57
+ "use_bitsandbytes": false,
58
+ "cfg_dropout_prob": 0.1,
59
+ "shuffle_tags": true,
60
+ "tag_dropout_prob": 0.1,
61
+ "wd": 0.02,
62
+ "gradient_accumulation_steps": 2,
63
+ "use_cos_scheduler": false,
64
+ "ignore_checkpoint_scheduler": true,
65
+ "ignore_checkpoint_optimizer": true,
66
+ "skip_warmup": false,
67
+ "final_lr_frac": 0.05,
68
+ "is_pixel_training": true,
69
+ "noise_scale": 2.0,
70
+ "sampler_type": "dpm-solver",
71
+ "timestep_fn": "logit-normal",
72
+ "use_ot": false,
73
+ "train_shift": 1.0,
74
+ "sample_shift": 6.93,
75
+ "clip_prediction": false,
76
+ "use_cfm": false,
77
+ "depth": 12,
78
+ "num_heads": 12,
79
+ "dropout": 0.1,
80
+ "compile_model": true,
81
+ "vae_pretrained": "kaiyuyue/FLUX.2-dev-vae",
82
+ "use_pos": true,
83
+ "cfg_mask_prob": 0.0,
84
+ "drop_ratio": 0.0,
85
+ "use_i1": true,
86
+ "use_rope_text_adapter": false,
87
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
88
+ "skip_checkpointing_layers": 4,
89
+ "patch_size": 16,
90
+ "use_pixel_decoder": true,
91
+ "sample_file": "sample.toml",
92
+ "cfg_scale": 4.5,
93
+ "projection_dim": 2
94
+ }
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb4924161712e32d2e5f9165bd03b79311ae688e80e90ffcceec94641a1220ea
3
+ size 24169478
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17d21481e8b71f252afe33096b78e2907fe17542c75a6ba6625f6d09a02108fe
3
+ size 96694027
dual_stream_768_12_trans_2_llm_te_log_normal_pixel_stage_1/epoch_5/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6949e1b390911ef5f653c8b0a5d18d0d2817004db5afb7d295231aa27f625f2b
3
+ size 1657