Hugging Face's logo

ruwwww
/
batik-pixelgen-exp

Model card Files
xet
 Community
ruwwww commited on
Commit
05ed813
·
verified ·
1 Parent(s): 7185b05

Add checkpoint(s) [via HfApi]

Browse files
Files changed (5) hide show
  1. jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602082333.yaml +219 -0
  2. jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602120157.yaml +219 -0
  3. jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602120215.yaml +219 -0
  4. jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/epoch=2555-step=800000.ckpt +3 -0
  5. jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/uploaded_by.txt +1 -0
jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602082333.yaml ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # lightning.pytorch==2.6.1
2
+ seed_everything: 1234
3
+ torch_hub_dir: null
4
+ huggingface_cache_dir: null
5
+ trainer:
6
+ accelerator: auto
7
+ strategy: auto
8
+ devices: auto
9
+ num_nodes: 1
10
+ precision: bf16-mixed
11
+ logger:
12
+ class_path: lightning.pytorch.loggers.WandbLogger
13
+ init_args:
14
+ name: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
15
+ save_dir: .
16
+ version: null
17
+ offline: false
18
+ dir: null
19
+ id: null
20
+ anonymous: null
21
+ project: jit_b_pixel_ablation
22
+ log_model: false
23
+ experiment: null
24
+ prefix: ''
25
+ checkpoint_name: null
26
+ add_file_policy: mutable
27
+ entity: null
28
+ notes: null
29
+ tags: null
30
+ config: null
31
+ config_exclude_keys: null
32
+ config_include_keys: null
33
+ allow_val_change: null
34
+ group: null
35
+ job_type: null
36
+ mode: null
37
+ force: null
38
+ reinit: null
39
+ resume: null
40
+ resume_from: null
41
+ fork_from: null
42
+ save_code: null
43
+ tensorboard: null
44
+ sync_tensorboard: null
45
+ monitor_gym: null
46
+ settings: null
47
+ callbacks:
48
+ - class_path: src.callbacks.model_checkpoint.CheckpointHook
49
+ init_args:
50
+ dirpath: null
51
+ filename: null
52
+ monitor: null
53
+ verbose: false
54
+ save_last: true
55
+ save_top_k: -1
56
+ save_on_exception: false
57
+ save_weights_only: false
58
+ mode: min
59
+ auto_insert_metric_name: true
60
+ every_n_train_steps: 50000
61
+ train_time_interval: null
62
+ every_n_epochs: null
63
+ save_on_train_epoch_end: null
64
+ enable_version_counter: true
65
+ - class_path: src.callbacks.save_images.SaveImagesHook
66
+ init_args:
67
+ save_dir: val_ode50_cfg2.0
68
+ save_compressed: true
69
+ fast_dev_run: false
70
+ max_epochs: null
71
+ min_epochs: null
72
+ max_steps: 1000000
73
+ min_steps: null
74
+ max_time: null
75
+ limit_train_batches: null
76
+ limit_val_batches: null
77
+ limit_test_batches: null
78
+ limit_predict_batches: null
79
+ overfit_batches: 0.0
80
+ val_check_interval: null
81
+ check_val_every_n_epoch: 1000000
82
+ num_sanity_val_steps: 0
83
+ log_every_n_steps: 50
84
+ enable_checkpointing: null
85
+ enable_progress_bar: null
86
+ enable_model_summary: null
87
+ accumulate_grad_batches: 1
88
+ gradient_clip_val: null
89
+ gradient_clip_algorithm: null
90
+ deterministic: null
91
+ benchmark: null
92
+ inference_mode: true
93
+ use_distributed_sampler: false
94
+ profiler: null
95
+ detect_anomaly: false
96
+ barebones: false
97
+ plugins:
98
+ - class_path: src.plugins.bd_env.BDEnvironment
99
+ sync_batchnorm: false
100
+ reload_dataloaders_every_n_epochs: 0
101
+ default_root_dir: ./universal_pix_workdirs
102
+ enable_autolog_hparams: true
103
+ model_registry: null
104
+ model:
105
+ vae:
106
+ class_path: src.models.autoencoder.pixel.PixelAE
107
+ init_args:
108
+ scale: 1.0
109
+ shift: 0.0
110
+ conditioner:
111
+ class_path: src.models.conditioner.class_label.LabelConditioner
112
+ init_args:
113
+ num_classes: 20
114
+ denoiser:
115
+ class_path: src.models.transformer.JiT.JiT
116
+ init_args:
117
+ input_size: 256
118
+ patch_size: 16
119
+ in_channels: 3
120
+ hidden_size: 768
121
+ depth: 12
122
+ num_heads: 12
123
+ mlp_ratio: 4.0
124
+ attn_drop: 0.0
125
+ proj_drop: 0.0
126
+ num_classes: 20
127
+ bottleneck_dim: 128
128
+ use_bottleneck: true
129
+ in_context_len: 32
130
+ in_context_start: 4
131
+ diffusion_trainer:
132
+ class_path: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.REPATrainer
133
+ init_args:
134
+ scheduler:
135
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
136
+ loss_weight_fn: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.constant
137
+ feat_loss_weight: 0.5
138
+ lognorm_t: true
139
+ timeshift: 1.0
140
+ encoder:
141
+ class_path: src.models.encoder.DINOv2
142
+ init_args:
143
+ weight_path: ~/.cache/torch/hub/checkpoints/dinov2_vitb14_pretrain.pth
144
+ base_patch_size: 16
145
+ align_layer: 4
146
+ proj_denoiser_dim: 768
147
+ proj_hidden_dim: 768
148
+ proj_encoder_dim: 768
149
+ P_mean: -0.8
150
+ P_std: 0.8
151
+ t_eps: 0.05
152
+ num_classes: 1000
153
+ lpips_weight: 0.1
154
+ dino_weight: 0.01
155
+ percept_t_threshold: 0.3
156
+ noise_scale: 1.0
157
+ patch_size: 16
158
+ percept_ratio: 1.0
159
+ null_condition_p: 0.1
160
+ diffusion_sampler:
161
+ class_path: src.diffusion.flow_matching.sampling.HeunSamplerJiT
162
+ init_args:
163
+ scheduler:
164
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
165
+ w_scheduler:
166
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
167
+ exact_henu: true
168
+ guidance_interval_min: 0.1
169
+ guidance_interval_max: 0.9
170
+ timeshift: 1.0
171
+ step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
172
+ last_step: null
173
+ last_step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
174
+ guidance_fn: src.diffusion.base.guidance.simple_guidance_fn
175
+ num_steps: 50
176
+ guidance: 2.0
177
+ ema_tracker:
178
+ class_path: src.callbacks.simple_ema.SimpleEMA
179
+ init_args:
180
+ decay: 0.9999
181
+ every_n_steps: 1
182
+ optimizer:
183
+ class_path: torch.optim.AdamW
184
+ init_args:
185
+ lr: 0.0001
186
+ betas:
187
+ - 0.9
188
+ - 0.999
189
+ eps: 1.0e-08
190
+ weight_decay: 0.0
191
+ amsgrad: false
192
+ maximize: false
193
+ foreach: null
194
+ capturable: false
195
+ differentiable: false
196
+ fused: null
197
+ lr_scheduler: null
198
+ eval_original_model: false
199
+ data:
200
+ train_dataset:
201
+ class_path: src.data.dataset.imagenet.PixImageNet
202
+ init_args:
203
+ root: /data/batik-256/images/
204
+ resolution: 256
205
+ random_crop: false
206
+ random_flip: false
207
+ eval_dataset: null
208
+ pred_dataset: null
209
+ train_batch_size: 64
210
+ train_num_workers: 8
211
+ train_prefetch_factor: 8
212
+ eval_batch_size: 32
213
+ eval_num_workers: 4
214
+ pred_batch_size: 32
215
+ pred_num_workers: 1
216
+ tags:
217
+ exp: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
218
+ ckpt_path: null
219
+ weights_only: null
jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602120157.yaml ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # lightning.pytorch==2.6.1
2
+ seed_everything: 1234
3
+ torch_hub_dir: null
4
+ huggingface_cache_dir: null
5
+ trainer:
6
+ accelerator: auto
7
+ strategy: auto
8
+ devices: auto
9
+ num_nodes: 1
10
+ precision: bf16-mixed
11
+ logger:
12
+ class_path: lightning.pytorch.loggers.WandbLogger
13
+ init_args:
14
+ name: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
15
+ save_dir: .
16
+ version: null
17
+ offline: false
18
+ dir: null
19
+ id: null
20
+ anonymous: null
21
+ project: jit_b_pixel_ablation
22
+ log_model: false
23
+ experiment: null
24
+ prefix: ''
25
+ checkpoint_name: null
26
+ add_file_policy: mutable
27
+ entity: null
28
+ notes: null
29
+ tags: null
30
+ config: null
31
+ config_exclude_keys: null
32
+ config_include_keys: null
33
+ allow_val_change: null
34
+ group: null
35
+ job_type: null
36
+ mode: null
37
+ force: null
38
+ reinit: null
39
+ resume: null
40
+ resume_from: null
41
+ fork_from: null
42
+ save_code: null
43
+ tensorboard: null
44
+ sync_tensorboard: null
45
+ monitor_gym: null
46
+ settings: null
47
+ callbacks:
48
+ - class_path: src.callbacks.model_checkpoint.CheckpointHook
49
+ init_args:
50
+ dirpath: null
51
+ filename: null
52
+ monitor: null
53
+ verbose: false
54
+ save_last: true
55
+ save_top_k: -1
56
+ save_on_exception: false
57
+ save_weights_only: false
58
+ mode: min
59
+ auto_insert_metric_name: true
60
+ every_n_train_steps: 50000
61
+ train_time_interval: null
62
+ every_n_epochs: null
63
+ save_on_train_epoch_end: null
64
+ enable_version_counter: true
65
+ - class_path: src.callbacks.save_images.SaveImagesHook
66
+ init_args:
67
+ save_dir: val_ode50_cfg2.0
68
+ save_compressed: true
69
+ fast_dev_run: false
70
+ max_epochs: null
71
+ min_epochs: null
72
+ max_steps: 1000000
73
+ min_steps: null
74
+ max_time: null
75
+ limit_train_batches: null
76
+ limit_val_batches: null
77
+ limit_test_batches: null
78
+ limit_predict_batches: null
79
+ overfit_batches: 0.0
80
+ val_check_interval: null
81
+ check_val_every_n_epoch: 1000000
82
+ num_sanity_val_steps: 0
83
+ log_every_n_steps: 50
84
+ enable_checkpointing: null
85
+ enable_progress_bar: null
86
+ enable_model_summary: null
87
+ accumulate_grad_batches: 1
88
+ gradient_clip_val: null
89
+ gradient_clip_algorithm: null
90
+ deterministic: null
91
+ benchmark: null
92
+ inference_mode: true
93
+ use_distributed_sampler: false
94
+ profiler: null
95
+ detect_anomaly: false
96
+ barebones: false
97
+ plugins:
98
+ - class_path: src.plugins.bd_env.BDEnvironment
99
+ sync_batchnorm: false
100
+ reload_dataloaders_every_n_epochs: 0
101
+ default_root_dir: ./universal_pix_workdirs
102
+ enable_autolog_hparams: true
103
+ model_registry: null
104
+ model:
105
+ vae:
106
+ class_path: src.models.autoencoder.pixel.PixelAE
107
+ init_args:
108
+ scale: 1.0
109
+ shift: 0.0
110
+ conditioner:
111
+ class_path: src.models.conditioner.class_label.LabelConditioner
112
+ init_args:
113
+ num_classes: 20
114
+ denoiser:
115
+ class_path: src.models.transformer.JiT.JiT
116
+ init_args:
117
+ input_size: 256
118
+ patch_size: 16
119
+ in_channels: 3
120
+ hidden_size: 768
121
+ depth: 12
122
+ num_heads: 12
123
+ mlp_ratio: 4.0
124
+ attn_drop: 0.0
125
+ proj_drop: 0.0
126
+ num_classes: 20
127
+ bottleneck_dim: 128
128
+ use_bottleneck: true
129
+ in_context_len: 32
130
+ in_context_start: 4
131
+ diffusion_trainer:
132
+ class_path: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.REPATrainer
133
+ init_args:
134
+ scheduler:
135
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
136
+ loss_weight_fn: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.constant
137
+ feat_loss_weight: 0.5
138
+ lognorm_t: true
139
+ timeshift: 1.0
140
+ encoder:
141
+ class_path: src.models.encoder.DINOv2
142
+ init_args:
143
+ weight_path: ~/.cache/torch/hub/checkpoints/dinov2_vitb14_pretrain.pth
144
+ base_patch_size: 16
145
+ align_layer: 4
146
+ proj_denoiser_dim: 768
147
+ proj_hidden_dim: 768
148
+ proj_encoder_dim: 768
149
+ P_mean: -0.8
150
+ P_std: 0.8
151
+ t_eps: 0.05
152
+ num_classes: 1000
153
+ lpips_weight: 0.1
154
+ dino_weight: 0.01
155
+ percept_t_threshold: 0.3
156
+ noise_scale: 1.0
157
+ patch_size: 16
158
+ percept_ratio: 1.0
159
+ null_condition_p: 0.1
160
+ diffusion_sampler:
161
+ class_path: src.diffusion.flow_matching.sampling.HeunSamplerJiT
162
+ init_args:
163
+ scheduler:
164
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
165
+ w_scheduler:
166
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
167
+ exact_henu: true
168
+ guidance_interval_min: 0.1
169
+ guidance_interval_max: 0.9
170
+ timeshift: 1.0
171
+ step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
172
+ last_step: null
173
+ last_step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
174
+ guidance_fn: src.diffusion.base.guidance.simple_guidance_fn
175
+ num_steps: 50
176
+ guidance: 2.0
177
+ ema_tracker:
178
+ class_path: src.callbacks.simple_ema.SimpleEMA
179
+ init_args:
180
+ decay: 0.9999
181
+ every_n_steps: 1
182
+ optimizer:
183
+ class_path: torch.optim.AdamW
184
+ init_args:
185
+ lr: 0.0001
186
+ betas:
187
+ - 0.9
188
+ - 0.999
189
+ eps: 1.0e-08
190
+ weight_decay: 0.0
191
+ amsgrad: false
192
+ maximize: false
193
+ foreach: null
194
+ capturable: false
195
+ differentiable: false
196
+ fused: null
197
+ lr_scheduler: null
198
+ eval_original_model: false
199
+ data:
200
+ train_dataset:
201
+ class_path: src.data.dataset.imagenet.PixImageNet
202
+ init_args:
203
+ root: /data/batik-256/images/
204
+ resolution: 256
205
+ random_crop: false
206
+ random_flip: false
207
+ eval_dataset: null
208
+ pred_dataset: null
209
+ train_batch_size: 64
210
+ train_num_workers: 8
211
+ train_prefetch_factor: 8
212
+ eval_batch_size: 32
213
+ eval_num_workers: 4
214
+ pred_batch_size: 32
215
+ pred_num_workers: 1
216
+ tags:
217
+ exp: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
218
+ ckpt_path: null
219
+ weights_only: null
jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/config-fit-2602120215.yaml ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # lightning.pytorch==2.6.1
2
+ seed_everything: 1234
3
+ torch_hub_dir: null
4
+ huggingface_cache_dir: null
5
+ trainer:
6
+ accelerator: auto
7
+ strategy: auto
8
+ devices: auto
9
+ num_nodes: 1
10
+ precision: bf16-mixed
11
+ logger:
12
+ class_path: lightning.pytorch.loggers.WandbLogger
13
+ init_args:
14
+ name: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
15
+ save_dir: .
16
+ version: null
17
+ offline: false
18
+ dir: null
19
+ id: null
20
+ anonymous: null
21
+ project: jit_b_pixel_ablation
22
+ log_model: false
23
+ experiment: null
24
+ prefix: ''
25
+ checkpoint_name: null
26
+ add_file_policy: mutable
27
+ entity: null
28
+ notes: null
29
+ tags: null
30
+ config: null
31
+ config_exclude_keys: null
32
+ config_include_keys: null
33
+ allow_val_change: null
34
+ group: null
35
+ job_type: null
36
+ mode: null
37
+ force: null
38
+ reinit: null
39
+ resume: null
40
+ resume_from: null
41
+ fork_from: null
42
+ save_code: null
43
+ tensorboard: null
44
+ sync_tensorboard: null
45
+ monitor_gym: null
46
+ settings: null
47
+ callbacks:
48
+ - class_path: src.callbacks.model_checkpoint.CheckpointHook
49
+ init_args:
50
+ dirpath: null
51
+ filename: null
52
+ monitor: null
53
+ verbose: false
54
+ save_last: true
55
+ save_top_k: -1
56
+ save_on_exception: false
57
+ save_weights_only: false
58
+ mode: min
59
+ auto_insert_metric_name: true
60
+ every_n_train_steps: 50000
61
+ train_time_interval: null
62
+ every_n_epochs: null
63
+ save_on_train_epoch_end: null
64
+ enable_version_counter: true
65
+ - class_path: src.callbacks.save_images.SaveImagesHook
66
+ init_args:
67
+ save_dir: val_ode50_cfg2.0
68
+ save_compressed: true
69
+ fast_dev_run: false
70
+ max_epochs: null
71
+ min_epochs: null
72
+ max_steps: 1000000
73
+ min_steps: null
74
+ max_time: null
75
+ limit_train_batches: null
76
+ limit_val_batches: null
77
+ limit_test_batches: null
78
+ limit_predict_batches: null
79
+ overfit_batches: 0.0
80
+ val_check_interval: null
81
+ check_val_every_n_epoch: 1000000
82
+ num_sanity_val_steps: 0
83
+ log_every_n_steps: 50
84
+ enable_checkpointing: null
85
+ enable_progress_bar: null
86
+ enable_model_summary: null
87
+ accumulate_grad_batches: 1
88
+ gradient_clip_val: null
89
+ gradient_clip_algorithm: null
90
+ deterministic: null
91
+ benchmark: null
92
+ inference_mode: true
93
+ use_distributed_sampler: false
94
+ profiler: null
95
+ detect_anomaly: false
96
+ barebones: false
97
+ plugins:
98
+ - class_path: src.plugins.bd_env.BDEnvironment
99
+ sync_batchnorm: false
100
+ reload_dataloaders_every_n_epochs: 0
101
+ default_root_dir: ./universal_pix_workdirs
102
+ enable_autolog_hparams: true
103
+ model_registry: null
104
+ model:
105
+ vae:
106
+ class_path: src.models.autoencoder.pixel.PixelAE
107
+ init_args:
108
+ scale: 1.0
109
+ shift: 0.0
110
+ conditioner:
111
+ class_path: src.models.conditioner.class_label.LabelConditioner
112
+ init_args:
113
+ num_classes: 20
114
+ denoiser:
115
+ class_path: src.models.transformer.JiT.JiT
116
+ init_args:
117
+ input_size: 256
118
+ patch_size: 16
119
+ in_channels: 3
120
+ hidden_size: 768
121
+ depth: 12
122
+ num_heads: 12
123
+ mlp_ratio: 4.0
124
+ attn_drop: 0.0
125
+ proj_drop: 0.0
126
+ num_classes: 20
127
+ bottleneck_dim: 128
128
+ use_bottleneck: true
129
+ in_context_len: 32
130
+ in_context_start: 4
131
+ diffusion_trainer:
132
+ class_path: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.REPATrainer
133
+ init_args:
134
+ scheduler:
135
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
136
+ loss_weight_fn: src.diffusion.flow_matching.training_repa_JiT_LPIPS_DINO_NoiseGating.constant
137
+ feat_loss_weight: 0.5
138
+ lognorm_t: true
139
+ timeshift: 1.0
140
+ encoder:
141
+ class_path: src.models.encoder.DINOv2
142
+ init_args:
143
+ weight_path: ~/.cache/torch/hub/checkpoints/dinov2_vitb14_pretrain.pth
144
+ base_patch_size: 16
145
+ align_layer: 4
146
+ proj_denoiser_dim: 768
147
+ proj_hidden_dim: 768
148
+ proj_encoder_dim: 768
149
+ P_mean: -0.8
150
+ P_std: 0.8
151
+ t_eps: 0.05
152
+ num_classes: 1000
153
+ lpips_weight: 0.1
154
+ dino_weight: 0.01
155
+ percept_t_threshold: 0.3
156
+ noise_scale: 1.0
157
+ patch_size: 16
158
+ percept_ratio: 1.0
159
+ null_condition_p: 0.1
160
+ diffusion_sampler:
161
+ class_path: src.diffusion.flow_matching.sampling.HeunSamplerJiT
162
+ init_args:
163
+ scheduler:
164
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
165
+ w_scheduler:
166
+ class_path: src.diffusion.flow_matching.scheduling.LinearScheduler
167
+ exact_henu: true
168
+ guidance_interval_min: 0.1
169
+ guidance_interval_max: 0.9
170
+ timeshift: 1.0
171
+ step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
172
+ last_step: null
173
+ last_step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
174
+ guidance_fn: src.diffusion.base.guidance.simple_guidance_fn
175
+ num_steps: 50
176
+ guidance: 2.0
177
+ ema_tracker:
178
+ class_path: src.callbacks.simple_ema.SimpleEMA
179
+ init_args:
180
+ decay: 0.9999
181
+ every_n_steps: 1
182
+ optimizer:
183
+ class_path: torch.optim.AdamW
184
+ init_args:
185
+ lr: 0.0001
186
+ betas:
187
+ - 0.9
188
+ - 0.999
189
+ eps: 1.0e-08
190
+ weight_decay: 0.0
191
+ amsgrad: false
192
+ maximize: false
193
+ foreach: null
194
+ capturable: false
195
+ differentiable: false
196
+ fused: null
197
+ lr_scheduler: null
198
+ eval_original_model: false
199
+ data:
200
+ train_dataset:
201
+ class_path: src.data.dataset.imagenet.PixImageNet
202
+ init_args:
203
+ root: /data/batik-256/images/
204
+ resolution: 256
205
+ random_crop: false
206
+ random_flip: false
207
+ eval_dataset: null
208
+ pred_dataset: null
209
+ train_batch_size: 64
210
+ train_num_workers: 8
211
+ train_prefetch_factor: 8
212
+ eval_batch_size: 32
213
+ eval_num_workers: 4
214
+ pred_batch_size: 32
215
+ pred_num_workers: 1
216
+ tags:
217
+ exp: jit_b_pixel_bs64_repa_4_lpips_pdino_adamw
218
+ ckpt_path: null
219
+ weights_only: null
jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/epoch=2555-step=800000.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aaa2eee7cf53377d830fd86b31d06b6d93bccdb28cc6bafefa601a9cc280637c
3
+ size 2109052911
jit_b_pixel_bs64_repa_4_lpips_pdino_adamw/step_800000/uploaded_by.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ source_dir=/workspace