Hugging Face's logo

dzungpham
/
font-architect

Image-to-Image
Diffusers
Safetensors
English
font-diffusion
contrastive-learning
font-generation
character-synthesis
style-transfer
dpm-solver
Model card Files
xet
 Community
1
dzungpham commited on
Commit
a33c4fe
·
verified ·
1 Parent(s): d4cf9be

DRO training phase 2 for 8000 steps

Browse files
Files changed (37) hide show
  1. DRO-20260226-8P2/FontDiffuserDRO_training_phase_2_config.yaml +159 -0
  2. DRO-20260226-8P2/checkpoint_step_1000/content_encoder.safetensors +3 -0
  3. DRO-20260226-8P2/checkpoint_step_1000/fst_module.safetensors +3 -0
  4. DRO-20260226-8P2/checkpoint_step_1000/fst_projection.safetensors +3 -0
  5. DRO-20260226-8P2/checkpoint_step_1000/mss_encoder.safetensors +3 -0
  6. DRO-20260226-8P2/checkpoint_step_1000/original_style_projection.safetensors +3 -0
  7. DRO-20260226-8P2/checkpoint_step_1000/scr.safetensors +3 -0
  8. DRO-20260226-8P2/checkpoint_step_1000/style_encoder.safetensors +3 -0
  9. DRO-20260226-8P2/checkpoint_step_1000/training_state.pt +3 -0
  10. DRO-20260226-8P2/checkpoint_step_1000/unet.safetensors +3 -0
  11. DRO-20260226-8P2/checkpoint_step_504/content_encoder.safetensors +3 -0
  12. DRO-20260226-8P2/checkpoint_step_504/fst_module.safetensors +3 -0
  13. DRO-20260226-8P2/checkpoint_step_504/fst_projection.safetensors +3 -0
  14. DRO-20260226-8P2/checkpoint_step_504/mss_encoder.safetensors +3 -0
  15. DRO-20260226-8P2/checkpoint_step_504/original_style_projection.safetensors +3 -0
  16. DRO-20260226-8P2/checkpoint_step_504/scr.safetensors +3 -0
  17. DRO-20260226-8P2/checkpoint_step_504/style_encoder.safetensors +3 -0
  18. DRO-20260226-8P2/checkpoint_step_504/training_state.pt +3 -0
  19. DRO-20260226-8P2/checkpoint_step_504/unet.safetensors +3 -0
  20. DRO-20260226-8P2/checkpoint_step_750/content_encoder.safetensors +3 -0
  21. DRO-20260226-8P2/checkpoint_step_750/fst_module.safetensors +3 -0
  22. DRO-20260226-8P2/checkpoint_step_750/fst_projection.safetensors +3 -0
  23. DRO-20260226-8P2/checkpoint_step_750/mss_encoder.safetensors +3 -0
  24. DRO-20260226-8P2/checkpoint_step_750/original_style_projection.safetensors +3 -0
  25. DRO-20260226-8P2/checkpoint_step_750/scr.safetensors +3 -0
  26. DRO-20260226-8P2/checkpoint_step_750/style_encoder.safetensors +3 -0
  27. DRO-20260226-8P2/checkpoint_step_750/training_state.pt +3 -0
  28. DRO-20260226-8P2/checkpoint_step_750/unet.safetensors +3 -0
  29. DRO-20260226-8P2/final/content_encoder.safetensors +3 -0
  30. DRO-20260226-8P2/final/fst_module.safetensors +3 -0
  31. DRO-20260226-8P2/final/fst_projection.safetensors +3 -0
  32. DRO-20260226-8P2/final/mss_encoder.safetensors +3 -0
  33. DRO-20260226-8P2/final/original_style_projection.safetensors +3 -0
  34. DRO-20260226-8P2/final/scr.safetensors +3 -0
  35. DRO-20260226-8P2/final/style_encoder.safetensors +3 -0
  36. DRO-20260226-8P2/final/training_state.pt +3 -0
  37. DRO-20260226-8P2/final/unet.safetensors +3 -0
DRO-20260226-8P2/FontDiffuserDRO_training_phase_2_config.yaml ADDED
@@ -0,0 +1,159 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ adam_beta1: 0.9
2
+ adam_beta2: 0.999
3
+ adam_epsilon: 1.0e-08
4
+ adam_weight_decay: 0.01
5
+ algorithm_type: dpmsolver++
6
+ batch_size: 1
7
+ beta_scheduler: scaled_linear
8
+ channel_attn: true
9
+ channels_last: false
10
+ character_input: false
11
+ characters: null
12
+ characters_file: null
13
+ ckpt_dir: null
14
+ ckpt_interval: 250
15
+ compile: true
16
+ compute_fid: false
17
+ consistency_loss_weight: 0.1
18
+ content_character: null
19
+ content_encoder_downsample_size: 3
20
+ content_image_path: null
21
+ content_image_size: !!python/tuple
22
+ - 96
23
+ - 96
24
+ content_start_channel: 64
25
+ controlnet: false
26
+ correcting_x0_fn: null
27
+ data_root: .
28
+ dataset_split: train_original
29
+ demo: false
30
+ deterministic: false
31
+ device: cuda:0
32
+ dro_div_weight: 0.0
33
+ dro_lpips_weight: 1.0
34
+ dro_max_timestep_frac: 0.3
35
+ dro_normalise_reward: false
36
+ dro_reward_scale: 1.0
37
+ dro_sharp_weight: 0.0
38
+ dro_ssim_weight: 1.0
39
+ dro_warmup_steps: 0
40
+ dro_weight: 0.1
41
+ drop_prob: 0.1
42
+ enable_attention_slicing: false
43
+ enable_style_transform: false
44
+ enable_xformers: false
45
+ end_line: null
46
+ evaluate: false
47
+ experience_name: FontDiffuserDRO_training_phase_2
48
+ export_onnx: false
49
+ fast_sampling: false
50
+ feature_dim: 512
51
+ ffn_dim: 2048
52
+ fp16: false
53
+ freeze_modules: ''
54
+ frequency_filter_type: gaussian
55
+ frequency_low_cutoff: 0.1
56
+ frequency_mid_cutoff: 0.4
57
+ frequency_mid_target: both
58
+ frequency_use_mid_band: true
59
+ fst_ckpt_path: null
60
+ fst_feature_channels: 64,128,256,512,1024
61
+ fst_num_queries: 220
62
+ fst_num_scales: 5
63
+ fst_query_dim: 256
64
+ gradient_accumulation_steps: 2
65
+ ground_truth_dir: null
66
+ guidance_scale: 7.5
67
+ guidance_type: classifier-free
68
+ hidden_dim: 256
69
+ identity_adaptive_max_weight: 1.0
70
+ identity_adaptive_min_weight: 0.1
71
+ identity_log_metrics: true
72
+ identity_loss_type: frobenius
73
+ identity_loss_weight: 0.1
74
+ identity_matrix_size: null
75
+ identity_metric_interval: 100
76
+ identity_pair_mode: random
77
+ identity_pooled_reduction: mean
78
+ identity_reg_weight: 0.01
79
+ identity_regularization: orthogonal
80
+ identity_similarity_threshold: 0.8
81
+ instructpix2pix: false
82
+ learning_rate: 5.0e-05
83
+ local_rank: 0
84
+ log_interval: 50
85
+ logging_dir: logs
86
+ lr_scheduler: cosine
87
+ lr_warmup_steps: 200
88
+ max_grad_norm: 1.0
89
+ max_train_steps: 1000
90
+ method: multistep
91
+ mixed_precision: 'no'
92
+ mode: refinement
93
+ model_type: noise
94
+ mss_base_channels: 64
95
+ mss_num_scales: 5
96
+ nce_layers: 0,1,2,3
97
+ num_consistency_pairs: 3
98
+ num_heads: 8
99
+ num_identity_pairs: 0
100
+ num_inference_steps: 20
101
+ num_neg: 34
102
+ num_workers: 3
103
+ offset_coefficient: 0.3
104
+ onnx_export_dir: null
105
+ onnx_opset_version: 17
106
+ order: 2
107
+ output_dir: outputs/FontArchitect/DRO-20260226-8P2
108
+ perceptual_coefficient: 0.03
109
+ phase_1: false
110
+ phase_1_ckpt_dir: ckpt/DRO-20260225-8P2/checkpoint_step_500
111
+ phase_2: true
112
+ report_to: wandb
113
+ resolution: 96
114
+ resume_from_checkpoint: ckpt/DRO-20260225-8P2/checkpoint_step_500
115
+ save_image: false
116
+ save_image_dir: null
117
+ save_interval: 10
118
+ sc_coefficient: 0.03
119
+ scale_lr: false
120
+ scr_ckpt_path: ckpt/DRO-20260225-8P2/checkpoint_step_500/scr.safetensors
121
+ scr_image_size: 96
122
+ seed: 123
123
+ skeleton_distance_method: hybrid
124
+ skeleton_fusion_method: concat
125
+ skeleton_max_distance: 12.0
126
+ skeleton_method: medial_axis
127
+ skeleton_output_mode: dual_channel
128
+ skeleton_sigma: 1.5
129
+ skip_type: time_uniform
130
+ start_line: 1
131
+ style_image_path: null
132
+ style_image_size: !!python/tuple
133
+ - 96
134
+ - 96
135
+ style_images: null
136
+ style_source_same_prob: 0.5
137
+ style_start_channel: 64
138
+ style_transform_coefficient: 0.1
139
+ summary: false
140
+ t_end: null
141
+ t_start: null
142
+ temperature: 0.07
143
+ train_batch_size: 4
144
+ ttf_path: ttf/KaiXinSongA.ttf
145
+ unet_channels: !!python/tuple
146
+ - 64
147
+ - 128
148
+ - 256
149
+ - 512
150
+ use_adaptive_identity_loss: false
151
+ use_dro: true
152
+ use_frequency_decomp: true
153
+ use_fst: true
154
+ use_pooled_identity_loss: false
155
+ use_skeleton_content: false
156
+ use_wandb: true
157
+ val_interval: 100
158
+ wandb_project: fontdiffuser-eval
159
+ wandb_run_name: null
DRO-20260226-8P2/checkpoint_step_1000/content_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ee03233703578e23371127fb0cd29a6f4ee36e107972ebd23902f5dd1e0faf6
3
+ size 4756580
DRO-20260226-8P2/checkpoint_step_1000/fst_module.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ce801b59701fc4086a1f5c1a0957381e6d32b7d1aa67b7e1b8abf6e68604341
3
+ size 238575652
DRO-20260226-8P2/checkpoint_step_1000/fst_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a4e8d198cc62958f215910b4e7ddc3995364d83cb511debc6af94867bd02b89
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_1000/mss_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49e75571c80f882f856092c1b68c227966a51147d4248c8bde91e4cedc1f536a
3
+ size 25261992
DRO-20260226-8P2/checkpoint_step_1000/original_style_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3e15d6d81bcad13f6ce723ba7b2b7bc75f008720f905b995857831b5796e556
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_1000/scr.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fbfecbf3e4485517a5d4004f99a394f21ebd318f06a2484374722cfb3d238c4a
3
+ size 284227968
DRO-20260226-8P2/checkpoint_step_1000/style_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:718afe6904ec0c143e3c848e7216fa5f8cac6d62fa3f48201d50debbcb68783c
3
+ size 82394556
DRO-20260226-8P2/checkpoint_step_1000/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57eea6b1a2e751f01725101ee90d463efd31b954f54962547ff328f991ae5513
3
+ size 1349134749
DRO-20260226-8P2/checkpoint_step_1000/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8276d3ddb5660fb56b8e1b53ccb155b306fafe7f459432ed8c369f539470d50f
3
+ size 314927748
DRO-20260226-8P2/checkpoint_step_504/content_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04461be19b5eaaf9e1d83fb8ef84b7a7a51ea87c8095f0d2320bb1659dfea9a9
3
+ size 4756580
DRO-20260226-8P2/checkpoint_step_504/fst_module.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a8f683f69b672d8b8568f81fdf6bd8fb97d07b8c5930549e533c504836129fd
3
+ size 238575652
DRO-20260226-8P2/checkpoint_step_504/fst_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:051fec3c5ac27ee1501afb81755353bb108774158726290c1ed8d6ecf2430f3e
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_504/mss_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e32a73972ed28d92e8a1c5a12b8b6fa5cf4180d2aba3f11b55c69382b040d1b9
3
+ size 25261992
DRO-20260226-8P2/checkpoint_step_504/original_style_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d5f53162ed3b9dccddfe324684b4f3596f51d5376687b5c3edb8579de6f95d3
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_504/scr.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:720c3f5aabc331be7e584da0f3d1372be324f514f6d3eb84e5f53b3f894ec8c1
3
+ size 284227968
DRO-20260226-8P2/checkpoint_step_504/style_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12eafee3b242d25e6e7288d05fdb5e18cdd3880b8a77d778dee8efdb0a32868b
3
+ size 82394556
DRO-20260226-8P2/checkpoint_step_504/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:acdf2384de8d9d49587cdccaee0a0d5202c01873ae245323d99f27a7df149f8a
3
+ size 1349134749
DRO-20260226-8P2/checkpoint_step_504/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:918533fca1607a6728205df672fb33bc1fbcb32e84446dbe301ee5c2b29f3160
3
+ size 314927748
DRO-20260226-8P2/checkpoint_step_750/content_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7990432ee514a92373ca9ed9f463a46ef06e45dffac0699a3f8ccbdf180fcc5
3
+ size 4756580
DRO-20260226-8P2/checkpoint_step_750/fst_module.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d3daeafa96c57ff05a09d764d46da8390f8b3989ad9cad31e55b1ee6749079b
3
+ size 238575652
DRO-20260226-8P2/checkpoint_step_750/fst_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b666a5d267e3671599bf0ce3fc423e70a367a14bdbbf9339be8633e85d78313
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_750/mss_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:835bd5bae385537eb22bbf79a5ffc8bf7c509cf0470743faf24fa547af24af14
3
+ size 25261992
DRO-20260226-8P2/checkpoint_step_750/original_style_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e5da38afa61824ed08cfeeb7d583dfe78027bb5825e18b59b32fd90e0dae9d94
3
+ size 4198552
DRO-20260226-8P2/checkpoint_step_750/scr.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73242967dc4debbe2539b581c62cca03231c3e52736642f2d3fbd1b1a2886b6f
3
+ size 284227968
DRO-20260226-8P2/checkpoint_step_750/style_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aaf3fb629ee98a43ccf0c0bb064850c9a7eb9e62efa3ac02d048321ba9f7d92b
3
+ size 82394556
DRO-20260226-8P2/checkpoint_step_750/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92475e240099a8cdb71a7d87f3d4cbf4ca58b7d22b73d265abc79d98c9a825ea
3
+ size 1349134749
DRO-20260226-8P2/checkpoint_step_750/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bab830f7afe3a4cb85ab510097158b6533a0e29c14baae762cd10661940d888
3
+ size 314927748
DRO-20260226-8P2/final/content_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ee03233703578e23371127fb0cd29a6f4ee36e107972ebd23902f5dd1e0faf6
3
+ size 4756580
DRO-20260226-8P2/final/fst_module.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ce801b59701fc4086a1f5c1a0957381e6d32b7d1aa67b7e1b8abf6e68604341
3
+ size 238575652
DRO-20260226-8P2/final/fst_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a4e8d198cc62958f215910b4e7ddc3995364d83cb511debc6af94867bd02b89
3
+ size 4198552
DRO-20260226-8P2/final/mss_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49e75571c80f882f856092c1b68c227966a51147d4248c8bde91e4cedc1f536a
3
+ size 25261992
DRO-20260226-8P2/final/original_style_projection.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3e15d6d81bcad13f6ce723ba7b2b7bc75f008720f905b995857831b5796e556
3
+ size 4198552
DRO-20260226-8P2/final/scr.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fbfecbf3e4485517a5d4004f99a394f21ebd318f06a2484374722cfb3d238c4a
3
+ size 284227968
DRO-20260226-8P2/final/style_encoder.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:718afe6904ec0c143e3c848e7216fa5f8cac6d62fa3f48201d50debbcb68783c
3
+ size 82394556
DRO-20260226-8P2/final/training_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57eea6b1a2e751f01725101ee90d463efd31b954f54962547ff328f991ae5513
3
+ size 1349134749
DRO-20260226-8P2/final/unet.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8276d3ddb5660fb56b8e1b53ccb155b306fafe7f459432ed8c369f539470d50f
3
+ size 314927748