Hugging Face's logo

aipracticecafe
/
dual-stream-dit-spatial-rope

sprint_dual
Model card Files
xet
 Community
aipracticecafe commited on
Commit
ed8babe
·
verified ·
1 Parent(s): 41c2fa7

Upload folder using huggingface_hub

Browse files
Files changed (15) hide show
  1. epoch_191_step_260000/config.json +110 -0
  2. epoch_191_step_260000/optimizer.pt +3 -0
  3. epoch_191_step_260000/scheduler.pt +3 -0
  4. epoch_191_step_260000/training_state.pt +3 -0
  5. epoch_191_step_260000/unet.safetensors +3 -0
  6. epoch_196_step_266500/config.json +110 -0
  7. epoch_196_step_266500/optimizer.pt +3 -0
  8. epoch_196_step_266500/scheduler.pt +3 -0
  9. epoch_196_step_266500/training_state.pt +3 -0
  10. epoch_196_step_266500/unet.safetensors +3 -0
  11. epoch_199_step_271680/config.json +110 -0
  12. epoch_199_step_271680/optimizer.pt +3 -0
  13. epoch_199_step_271680/scheduler.pt +3 -0
  14. epoch_199_step_271680/training_state.pt +3 -0
  15. epoch_199_step_271680/unet.safetensors +3 -0
epoch_191_step_260000/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.75,
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": "results/spatial_rope/epoch_174_step_233640",
18
+ "cfg_mask_prob": 0.0,
19
+ "skip_checkpointing_layers": 2,
20
+ "use_random_drop": false,
21
+ "vae_batch_size": 8,
22
+ "use_calibrated_spatial": true,
23
+ "train": {
24
+ "batch_size": 64,
25
+ "lr": 0.00025,
26
+ "epochs": 200,
27
+ "gradient_accumulation_steps": 4,
28
+ "dtype": "fp32",
29
+ "use_checkpointing": true,
30
+ "train_te": false,
31
+ "wd": 0.01,
32
+ "warmup": 0.04,
33
+ "min_lr_ratio": 0.01,
34
+ "reset_scheduler": true,
35
+ "reset_optimizer": false,
36
+ "use_cos_scheduler": false,
37
+ "objective": "flow_matching",
38
+ "timestep_fn": "logit-normal",
39
+ "shift": 2.0,
40
+ "snr_gamma": 5.0,
41
+ "input_perturb": 0.1,
42
+ "cfg_dropout_prob": 0.1,
43
+ "seed": 232,
44
+ "use_bitsandbytes": true,
45
+ "train_only_output": false,
46
+ "skip_save_n_times": 4,
47
+ "run_stability_check": false,
48
+ "world_size": 1,
49
+ "use_ot": false,
50
+ "use_ema": false,
51
+ "ema_decay": 0.999,
52
+ "use_kahan_sum": false,
53
+ "use_cpu_accumulator": false,
54
+ "compile_model": true,
55
+ "gpu_peak_tflops": 165.2
56
+ },
57
+ "data": {
58
+ "dataset_type": "streaming",
59
+ "h5_path": "data",
60
+ "streaming_dataset_name": "aipracticecafe/curated-danbooru-2026-512px-flux2-vae",
61
+ "num_workers": 6,
62
+ "prefetch_factor": 6,
63
+ "load_into_ram": true,
64
+ "tag_dropout": 0.1,
65
+ "use_shift_crop": true,
66
+ "resolution": 512,
67
+ "buffer_size": 2500,
68
+ "cache_latents_to_ram": false,
69
+ "is_latent": true,
70
+ "precompute_num_workers": 2,
71
+ "precompute_prefetch_factor": 2
72
+ },
73
+ "models": {
74
+ "model_type": "sprint_dual",
75
+ "resume_from_checkpoint": "results/spatial_rope/epoch_174_step_233640",
76
+ "output_head_path": null,
77
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
78
+ "hidden_size": 768,
79
+ "depth": 13,
80
+ "num_heads": 12,
81
+ "encoder_depth": 2,
82
+ "decoder_depth": 2,
83
+ "drop_ratio": 0.75,
84
+ "drop_target": "image",
85
+ "residual_type": "concat_linear",
86
+ "cfg_mask_prob": 0.0,
87
+ "skip_checkpointing_layers": 2,
88
+ "use_random_drop": false,
89
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
90
+ "vae_mean": 0.05106564902828182,
91
+ "vae_std": 1.69915383699,
92
+ "vae_batch_size": 8,
93
+ "in_channels": 32,
94
+ "use_calibrated_spatial": true
95
+ },
96
+ "paths": {
97
+ "models": "models"
98
+ },
99
+ "logging": {
100
+ "project": "dual-dit",
101
+ "wandb_entity": "aipracticecafe-aipracticecafe",
102
+ "save_interval": 6500,
103
+ "hf_repo": "dual-stream-dit-spatial-rope",
104
+ "wandb_run_name_prefix": "4090-sprint-512px-latents-spatial-rope"
105
+ },
106
+ "sampling": {
107
+ "interval": 1300,
108
+ "config_file": "sample.toml"
109
+ }
110
+ }
epoch_191_step_260000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ff593f25edbccff2aff0dbda99ebe7125b54dd5ae513a9701ececd986ecaa9d
3
+ size 419905931
epoch_191_step_260000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bb9447ddd1be57760bc8b54e9f23900bb3568ddfe4c70122b841e9f5aba6ca4
3
+ size 1721
epoch_191_step_260000/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47245af7d5c2fda9b0ade14d1c7365f4e682a469607efcabc51ca8fa2a65f5f6
3
+ size 1303
epoch_191_step_260000/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:378ec745c603d9571f9dc0764c59cdfbe5bb5936197facd6d3642612c11b681a
3
+ size 425266512
epoch_196_step_266500/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.75,
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": "results/spatial_rope/epoch_174_step_233640",
18
+ "cfg_mask_prob": 0.0,
19
+ "skip_checkpointing_layers": 2,
20
+ "use_random_drop": false,
21
+ "vae_batch_size": 8,
22
+ "use_calibrated_spatial": true,
23
+ "train": {
24
+ "batch_size": 64,
25
+ "lr": 0.00025,
26
+ "epochs": 200,
27
+ "gradient_accumulation_steps": 4,
28
+ "dtype": "fp32",
29
+ "use_checkpointing": true,
30
+ "train_te": false,
31
+ "wd": 0.01,
32
+ "warmup": 0.04,
33
+ "min_lr_ratio": 0.01,
34
+ "reset_scheduler": true,
35
+ "reset_optimizer": false,
36
+ "use_cos_scheduler": false,
37
+ "objective": "flow_matching",
38
+ "timestep_fn": "logit-normal",
39
+ "shift": 2.0,
40
+ "snr_gamma": 5.0,
41
+ "input_perturb": 0.1,
42
+ "cfg_dropout_prob": 0.1,
43
+ "seed": 232,
44
+ "use_bitsandbytes": true,
45
+ "train_only_output": false,
46
+ "skip_save_n_times": 4,
47
+ "run_stability_check": false,
48
+ "world_size": 1,
49
+ "use_ot": false,
50
+ "use_ema": false,
51
+ "ema_decay": 0.999,
52
+ "use_kahan_sum": false,
53
+ "use_cpu_accumulator": false,
54
+ "compile_model": true,
55
+ "gpu_peak_tflops": 165.2
56
+ },
57
+ "data": {
58
+ "dataset_type": "streaming",
59
+ "h5_path": "data",
60
+ "streaming_dataset_name": "aipracticecafe/curated-danbooru-2026-512px-flux2-vae",
61
+ "num_workers": 6,
62
+ "prefetch_factor": 6,
63
+ "load_into_ram": true,
64
+ "tag_dropout": 0.1,
65
+ "use_shift_crop": true,
66
+ "resolution": 512,
67
+ "buffer_size": 2500,
68
+ "cache_latents_to_ram": false,
69
+ "is_latent": true,
70
+ "precompute_num_workers": 2,
71
+ "precompute_prefetch_factor": 2
72
+ },
73
+ "models": {
74
+ "model_type": "sprint_dual",
75
+ "resume_from_checkpoint": "results/spatial_rope/epoch_174_step_233640",
76
+ "output_head_path": null,
77
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
78
+ "hidden_size": 768,
79
+ "depth": 13,
80
+ "num_heads": 12,
81
+ "encoder_depth": 2,
82
+ "decoder_depth": 2,
83
+ "drop_ratio": 0.75,
84
+ "drop_target": "image",
85
+ "residual_type": "concat_linear",
86
+ "cfg_mask_prob": 0.0,
87
+ "skip_checkpointing_layers": 2,
88
+ "use_random_drop": false,
89
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
90
+ "vae_mean": 0.05106564902828182,
91
+ "vae_std": 1.69915383699,
92
+ "vae_batch_size": 8,
93
+ "in_channels": 32,
94
+ "use_calibrated_spatial": true
95
+ },
96
+ "paths": {
97
+ "models": "models"
98
+ },
99
+ "logging": {
100
+ "project": "dual-dit",
101
+ "wandb_entity": "aipracticecafe-aipracticecafe",
102
+ "save_interval": 6500,
103
+ "hf_repo": "dual-stream-dit-spatial-rope",
104
+ "wandb_run_name_prefix": "4090-sprint-512px-latents-spatial-rope"
105
+ },
106
+ "sampling": {
107
+ "interval": 1300,
108
+ "config_file": "sample.toml"
109
+ }
110
+ }
epoch_196_step_266500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:362506acd38ff4127a8c6919d484d15e8dc4288cfb73820289e6fe5365c88385
3
+ size 419905931
epoch_196_step_266500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b249f0995b66c91df15846f1c743a53457c46b7502f07f706faf9f7ebc5b50f
3
+ size 1721
epoch_196_step_266500/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ade524160bf1a56475d321330f9c964dfebb24ca412727e4d97ccd46d33ee69f
3
+ size 1303
epoch_196_step_266500/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d699e5a9feb7b98e6dad86613e389a1868b045877f13ffcae2bd96e7738feec
3
+ size 425266512
epoch_199_step_271680/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.75,
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": "results/spatial_rope/epoch_174_step_233640",
18
+ "cfg_mask_prob": 0.0,
19
+ "skip_checkpointing_layers": 2,
20
+ "use_random_drop": false,
21
+ "vae_batch_size": 8,
22
+ "use_calibrated_spatial": true,
23
+ "train": {
24
+ "batch_size": 64,
25
+ "lr": 0.00025,
26
+ "epochs": 200,
27
+ "gradient_accumulation_steps": 4,
28
+ "dtype": "fp32",
29
+ "use_checkpointing": true,
30
+ "train_te": false,
31
+ "wd": 0.01,
32
+ "warmup": 0.04,
33
+ "min_lr_ratio": 0.01,
34
+ "reset_scheduler": true,
35
+ "reset_optimizer": false,
36
+ "use_cos_scheduler": false,
37
+ "objective": "flow_matching",
38
+ "timestep_fn": "logit-normal",
39
+ "shift": 2.0,
40
+ "snr_gamma": 5.0,
41
+ "input_perturb": 0.1,
42
+ "cfg_dropout_prob": 0.1,
43
+ "seed": 232,
44
+ "use_bitsandbytes": true,
45
+ "train_only_output": false,
46
+ "skip_save_n_times": 4,
47
+ "run_stability_check": false,
48
+ "world_size": 1,
49
+ "use_ot": false,
50
+ "use_ema": false,
51
+ "ema_decay": 0.999,
52
+ "use_kahan_sum": false,
53
+ "use_cpu_accumulator": false,
54
+ "compile_model": true,
55
+ "gpu_peak_tflops": 165.2
56
+ },
57
+ "data": {
58
+ "dataset_type": "streaming",
59
+ "h5_path": "data",
60
+ "streaming_dataset_name": "aipracticecafe/curated-danbooru-2026-512px-flux2-vae",
61
+ "num_workers": 6,
62
+ "prefetch_factor": 6,
63
+ "load_into_ram": true,
64
+ "tag_dropout": 0.1,
65
+ "use_shift_crop": true,
66
+ "resolution": 512,
67
+ "buffer_size": 2500,
68
+ "cache_latents_to_ram": false,
69
+ "is_latent": true,
70
+ "precompute_num_workers": 2,
71
+ "precompute_prefetch_factor": 2
72
+ },
73
+ "models": {
74
+ "model_type": "sprint_dual",
75
+ "resume_from_checkpoint": "results/spatial_rope/epoch_174_step_233640",
76
+ "output_head_path": null,
77
+ "hf_text_encoder": "LiquidAI/LFM2.5-230M",
78
+ "hidden_size": 768,
79
+ "depth": 13,
80
+ "num_heads": 12,
81
+ "encoder_depth": 2,
82
+ "decoder_depth": 2,
83
+ "drop_ratio": 0.75,
84
+ "drop_target": "image",
85
+ "residual_type": "concat_linear",
86
+ "cfg_mask_prob": 0.0,
87
+ "skip_checkpointing_layers": 2,
88
+ "use_random_drop": false,
89
+ "hf_vae": "kaiyuyue/FLUX.2-dev-vae",
90
+ "vae_mean": 0.05106564902828182,
91
+ "vae_std": 1.69915383699,
92
+ "vae_batch_size": 8,
93
+ "in_channels": 32,
94
+ "use_calibrated_spatial": true
95
+ },
96
+ "paths": {
97
+ "models": "models"
98
+ },
99
+ "logging": {
100
+ "project": "dual-dit",
101
+ "wandb_entity": "aipracticecafe-aipracticecafe",
102
+ "save_interval": 6500,
103
+ "hf_repo": "dual-stream-dit-spatial-rope",
104
+ "wandb_run_name_prefix": "4090-sprint-512px-latents-spatial-rope"
105
+ },
106
+ "sampling": {
107
+ "interval": 1300,
108
+ "config_file": "sample.toml"
109
+ }
110
+ }
epoch_199_step_271680/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e684ff3b93d633eafb91eaadc6de4508bd393209df82714a8f56b7fce2eb30fa
3
+ size 419905931
epoch_199_step_271680/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:14945d79a8ff200a2bdaf81e5205eb63d998e2e545921ca9d8cf07637fe9f43b
3
+ size 1721
epoch_199_step_271680/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c03b1c6a23eaa16764863cf4230443c17961544327351cf0cc70fcb25f0376bc
3
+ size 1303
epoch_199_step_271680/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:033e423f0f158002f69c2627ebde8132365acd7369a7a5bb96f285595c948b6c
3
+ size 425266512