Text-to-Image
Diffusers
PyTorch
flow-matching
rectified-flow
dit
dual-stream-dit
sprint
anime
aipracticecafe commited on
Commit
e82c626
·
verified ·
1 Parent(s): cae307e

Upload folder using huggingface_hub

Browse files
epoch_215_step_311350/config.json ADDED
@@ -0,0 +1,113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "sprint_dual",
3
+ "model_type": "sprint_dual",
4
+ "in_channels": 32,
5
+ "hidden_size": 768,
6
+ "depth": 13,
7
+ "num_heads": 12,
8
+ "encoder_depth": 2,
9
+ "decoder_depth": 2,
10
+ "drop_ratio": 0.0,
11
+ "drop_target": "image",
12
+ "residual_type": "concat_linear",
13
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
14
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
15
+ "vae_mean": 0.05106564902828182,
16
+ "vae_std": 1.69915383699,
17
+ "resume_from_checkpoint": "models/epoch_213_step_304200/epoch_213_step_304200",
18
+ "cfg_mask_prob": 0.0,
19
+ "skip_checkpointing_layers": 0,
20
+ "use_random_drop": false,
21
+ "vae_batch_size": 8,
22
+ "use_calibrated_spatial": true,
23
+ "train": {
24
+ "batch_size": 8,
25
+ "lr": 8e-05,
26
+ "epochs": 217,
27
+ "gradient_accumulation_steps": 16,
28
+ "dtype": "fp32",
29
+ "use_checkpointing": false,
30
+ "train_te": false,
31
+ "wd": 0.01,
32
+ "warmup": 0.04,
33
+ "min_lr_ratio": 0.05,
34
+ "reset_scheduler": true,
35
+ "reset_optimizer": false,
36
+ "use_cos_scheduler": false,
37
+ "sched_type": "wsd",
38
+ "decay_ratio": 0.15,
39
+ "objective": "flow_matching",
40
+ "timestep_fn": "logit-normal",
41
+ "shift": 3.0,
42
+ "snr_gamma": 5.0,
43
+ "input_perturb": 0.1,
44
+ "cfg_dropout_prob": 0.1,
45
+ "seed": 232,
46
+ "use_bitsandbytes": true,
47
+ "train_only_output": false,
48
+ "skip_save_n_times": 4,
49
+ "run_stability_check": false,
50
+ "world_size": 1,
51
+ "use_ot": false,
52
+ "use_ema": false,
53
+ "ema_decay": 0.999,
54
+ "use_kahan_sum": false,
55
+ "use_cpu_accumulator": false,
56
+ "compile_model": true,
57
+ "gpu_peak_tflops": 149.7
58
+ },
59
+ "data": {
60
+ "dataset_type": "streaming",
61
+ "h5_path": "data",
62
+ "dataset_path": "data",
63
+ "streaming_dataset_name": "aipracticecafe/curated-danbooru-2026-512px-flux2-vae",
64
+ "num_workers": 6,
65
+ "prefetch_factor": 6,
66
+ "load_into_ram": true,
67
+ "tag_dropout": 0.1,
68
+ "use_shift_crop": true,
69
+ "resolution": 1024,
70
+ "buffer_size": 2500,
71
+ "cache_latents_to_ram": false,
72
+ "is_latent": false,
73
+ "precompute_num_workers": 2,
74
+ "precompute_prefetch_factor": 2
75
+ },
76
+ "models": {
77
+ "model_type": "sprint_dual",
78
+ "resume_from_checkpoint": "models/epoch_213_step_304200/epoch_213_step_304200",
79
+ "output_head_path": null,
80
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
81
+ "hidden_size": 768,
82
+ "depth": 13,
83
+ "num_heads": 12,
84
+ "encoder_depth": 2,
85
+ "decoder_depth": 2,
86
+ "drop_ratio": 0.0,
87
+ "drop_target": "image",
88
+ "residual_type": "concat_linear",
89
+ "cfg_mask_prob": 0.0,
90
+ "skip_checkpointing_layers": 0,
91
+ "use_random_drop": false,
92
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
93
+ "vae_mean": 0.05106564902828182,
94
+ "vae_std": 1.69915383699,
95
+ "vae_batch_size": 8,
96
+ "in_channels": 32,
97
+ "use_calibrated_spatial": true
98
+ },
99
+ "paths": {
100
+ "models": "models"
101
+ },
102
+ "logging": {
103
+ "project": "dual-dit",
104
+ "wandb_entity": "aipracticecafe-aipracticecafe",
105
+ "save_interval": 650,
106
+ "hf_repo": "dual-stream-dit-spatial-rope",
107
+ "wandb_run_name_prefix": "4090-sprint-1MP-latents-spatial-rope"
108
+ },
109
+ "sampling": {
110
+ "interval": 325,
111
+ "config_file": "sample.toml"
112
+ }
113
+ }
epoch_215_step_311350/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa3762c442db0706125b232813adbcb76c54a330d90afc05d0425539facd2a68
3
+ size 419909323
epoch_215_step_311350/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f7191b559a1e00e8b385f62eb19e4f24e8a59f51b89e86bca331bd2f2c38dc1
3
+ size 1913
epoch_215_step_311350/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:655a0d647b13af10ac6040b2d1c50ab91f0bbe446bc278311c71c4a23af60025
3
+ size 1303
epoch_215_step_311350/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b5c4abcc9664d18b1a2f240a2f2c2da25256e2781a9b7830f2ed9675eed633c
3
+ size 425266512
epoch_215_step_312000/config.json ADDED
@@ -0,0 +1,113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "sprint_dual",
3
+ "model_type": "sprint_dual",
4
+ "in_channels": 32,
5
+ "hidden_size": 768,
6
+ "depth": 13,
7
+ "num_heads": 12,
8
+ "encoder_depth": 2,
9
+ "decoder_depth": 2,
10
+ "drop_ratio": 0.0,
11
+ "drop_target": "image",
12
+ "residual_type": "concat_linear",
13
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
14
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
15
+ "vae_mean": 0.05106564902828182,
16
+ "vae_std": 1.69915383699,
17
+ "resume_from_checkpoint": "models/epoch_213_step_304200/epoch_213_step_304200",
18
+ "cfg_mask_prob": 0.0,
19
+ "skip_checkpointing_layers": 0,
20
+ "use_random_drop": false,
21
+ "vae_batch_size": 8,
22
+ "use_calibrated_spatial": true,
23
+ "train": {
24
+ "batch_size": 8,
25
+ "lr": 8e-05,
26
+ "epochs": 217,
27
+ "gradient_accumulation_steps": 16,
28
+ "dtype": "fp32",
29
+ "use_checkpointing": false,
30
+ "train_te": false,
31
+ "wd": 0.01,
32
+ "warmup": 0.04,
33
+ "min_lr_ratio": 0.05,
34
+ "reset_scheduler": true,
35
+ "reset_optimizer": false,
36
+ "use_cos_scheduler": false,
37
+ "sched_type": "wsd",
38
+ "decay_ratio": 0.15,
39
+ "objective": "flow_matching",
40
+ "timestep_fn": "logit-normal",
41
+ "shift": 3.0,
42
+ "snr_gamma": 5.0,
43
+ "input_perturb": 0.1,
44
+ "cfg_dropout_prob": 0.1,
45
+ "seed": 232,
46
+ "use_bitsandbytes": true,
47
+ "train_only_output": false,
48
+ "skip_save_n_times": 4,
49
+ "run_stability_check": false,
50
+ "world_size": 1,
51
+ "use_ot": false,
52
+ "use_ema": false,
53
+ "ema_decay": 0.999,
54
+ "use_kahan_sum": false,
55
+ "use_cpu_accumulator": false,
56
+ "compile_model": true,
57
+ "gpu_peak_tflops": 149.7
58
+ },
59
+ "data": {
60
+ "dataset_type": "streaming",
61
+ "h5_path": "data",
62
+ "dataset_path": "data",
63
+ "streaming_dataset_name": "aipracticecafe/curated-danbooru-2026-512px-flux2-vae",
64
+ "num_workers": 6,
65
+ "prefetch_factor": 6,
66
+ "load_into_ram": true,
67
+ "tag_dropout": 0.1,
68
+ "use_shift_crop": true,
69
+ "resolution": 1024,
70
+ "buffer_size": 2500,
71
+ "cache_latents_to_ram": false,
72
+ "is_latent": false,
73
+ "precompute_num_workers": 2,
74
+ "precompute_prefetch_factor": 2
75
+ },
76
+ "models": {
77
+ "model_type": "sprint_dual",
78
+ "resume_from_checkpoint": "models/epoch_213_step_304200/epoch_213_step_304200",
79
+ "output_head_path": null,
80
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
81
+ "hidden_size": 768,
82
+ "depth": 13,
83
+ "num_heads": 12,
84
+ "encoder_depth": 2,
85
+ "decoder_depth": 2,
86
+ "drop_ratio": 0.0,
87
+ "drop_target": "image",
88
+ "residual_type": "concat_linear",
89
+ "cfg_mask_prob": 0.0,
90
+ "skip_checkpointing_layers": 0,
91
+ "use_random_drop": false,
92
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
93
+ "vae_mean": 0.05106564902828182,
94
+ "vae_std": 1.69915383699,
95
+ "vae_batch_size": 8,
96
+ "in_channels": 32,
97
+ "use_calibrated_spatial": true
98
+ },
99
+ "paths": {
100
+ "models": "models"
101
+ },
102
+ "logging": {
103
+ "project": "dual-dit",
104
+ "wandb_entity": "aipracticecafe-aipracticecafe",
105
+ "save_interval": 650,
106
+ "hf_repo": "dual-stream-dit-spatial-rope",
107
+ "wandb_run_name_prefix": "4090-sprint-1MP-latents-spatial-rope"
108
+ },
109
+ "sampling": {
110
+ "interval": 325,
111
+ "config_file": "sample.toml"
112
+ }
113
+ }
epoch_215_step_312000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a85b5f026e16c588be40a0deaa6919414f86e5dcf704493881fdef103ed15c7
3
+ size 419909323
epoch_215_step_312000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a32150b8149ddd4d16131d25a211fe8bf5453a7c23d2262df366b86c5b7c4662
3
+ size 1913
epoch_215_step_312000/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1625f51095f55986873d5ded3b04a1322bf64d4bf3a58a2de9397f624294a913
3
+ size 1303
epoch_215_step_312000/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2c80d8de7405b62b9172317538452606e815e8e20baf2044816bb88c90b582d
3
+ size 425266512