add ablation configs
Browse files- slat_abl_bidirectional_20260428_031441/config.json +114 -0
- slat_abl_diff_causal_cdrop00/config.json +112 -0
- slat_abl_diff_causal_cdrop005/config.json +112 -0
- slat_abl_diff_causal_cdrop01/config.json +112 -0
- slat_abl_diff_causal_cdrop02/config.json +112 -0
- slat_abl_same_noise_20260428_031335/config.json +113 -0
- ss_abl_bidirectional_20260427_175112/config.json +85 -0
- ss_abl_diff_causal_20260427_175045/config.json +84 -0
- ss_abl_same_noise_20260427_175113/config.json +84 -0
slat_abl_bidirectional_20260428_031441/config.json
ADDED
|
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16",
|
| 27 |
+
"causal_attn": false
|
| 28 |
+
}
|
| 29 |
+
}
|
| 30 |
+
},
|
| 31 |
+
"dataset": {
|
| 32 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 33 |
+
"args": {
|
| 34 |
+
"prev_window": 2,
|
| 35 |
+
"val_num_scenes": 20,
|
| 36 |
+
"val_num_frames": 6,
|
| 37 |
+
"snapshot_num_scenes": 4,
|
| 38 |
+
"snapshot_num_frames": 4,
|
| 39 |
+
"image_size": 518,
|
| 40 |
+
"num_cond_views": 6,
|
| 41 |
+
"front_view_prob": 0.4,
|
| 42 |
+
"normalization": {
|
| 43 |
+
"mean": [
|
| 44 |
+
-2.2392261552952513,
|
| 45 |
+
0.7410145409701431,
|
| 46 |
+
0.5079064556582509,
|
| 47 |
+
-0.05298659832508365,
|
| 48 |
+
-0.025759045060578523,
|
| 49 |
+
0.8829246196309438,
|
| 50 |
+
-0.7059389169604169,
|
| 51 |
+
0.9353478047296675
|
| 52 |
+
],
|
| 53 |
+
"std": [
|
| 54 |
+
2.5146836083929074,
|
| 55 |
+
2.396254417443913,
|
| 56 |
+
2.430939371044743,
|
| 57 |
+
2.18786890103016,
|
| 58 |
+
2.9446809932599853,
|
| 59 |
+
2.3331955459252756,
|
| 60 |
+
2.7919380050354676,
|
| 61 |
+
2.6444312498997413
|
| 62 |
+
]
|
| 63 |
+
}
|
| 64 |
+
}
|
| 65 |
+
},
|
| 66 |
+
"trainer": {
|
| 67 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 68 |
+
"args": {
|
| 69 |
+
"max_steps": 100000,
|
| 70 |
+
"batch_size_per_gpu": 4,
|
| 71 |
+
"batch_split": 4,
|
| 72 |
+
"prev_window": 2,
|
| 73 |
+
"coord_drop_prob": 0.05,
|
| 74 |
+
"p_clean": 1.0,
|
| 75 |
+
"max_clean_t": 1.0,
|
| 76 |
+
"cfg_noise_t": 0.9,
|
| 77 |
+
"history_cfg": 0.0,
|
| 78 |
+
"noise_scheme": "independent",
|
| 79 |
+
"optimizer": {
|
| 80 |
+
"name": "AdamW",
|
| 81 |
+
"args": {
|
| 82 |
+
"lr": 0.0001,
|
| 83 |
+
"weight_decay": 0.0
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"ema_rate": [
|
| 87 |
+
0.9999
|
| 88 |
+
],
|
| 89 |
+
"fp16_mode": "inflat_all",
|
| 90 |
+
"fp16_scale_growth": 0.001,
|
| 91 |
+
"grad_clip": {
|
| 92 |
+
"name": "AdaptiveGradClipper",
|
| 93 |
+
"args": {
|
| 94 |
+
"max_norm": 1.0,
|
| 95 |
+
"clip_percentile": 95
|
| 96 |
+
}
|
| 97 |
+
},
|
| 98 |
+
"i_log": 10,
|
| 99 |
+
"i_print": 100,
|
| 100 |
+
"i_sample": 5000,
|
| 101 |
+
"i_save": 2000,
|
| 102 |
+
"p_uncond": 0.1,
|
| 103 |
+
"t_schedule": {
|
| 104 |
+
"name": "logitNormal",
|
| 105 |
+
"args": {
|
| 106 |
+
"mean": 1.0,
|
| 107 |
+
"std": 1.0
|
| 108 |
+
}
|
| 109 |
+
},
|
| 110 |
+
"sigma_min": 1e-05,
|
| 111 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 112 |
+
}
|
| 113 |
+
}
|
| 114 |
+
}
|
slat_abl_diff_causal_cdrop00/config.json
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16"
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dataset": {
|
| 31 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 32 |
+
"args": {
|
| 33 |
+
"prev_window": 2,
|
| 34 |
+
"val_num_scenes": 20,
|
| 35 |
+
"val_num_frames": 6,
|
| 36 |
+
"snapshot_num_scenes": 4,
|
| 37 |
+
"snapshot_num_frames": 4,
|
| 38 |
+
"image_size": 518,
|
| 39 |
+
"num_cond_views": 6,
|
| 40 |
+
"front_view_prob": 0.4,
|
| 41 |
+
"normalization": {
|
| 42 |
+
"mean": [
|
| 43 |
+
-2.2392261552952513,
|
| 44 |
+
0.7410145409701431,
|
| 45 |
+
0.5079064556582509,
|
| 46 |
+
-0.05298659832508365,
|
| 47 |
+
-0.025759045060578523,
|
| 48 |
+
0.8829246196309438,
|
| 49 |
+
-0.7059389169604169,
|
| 50 |
+
0.9353478047296675
|
| 51 |
+
],
|
| 52 |
+
"std": [
|
| 53 |
+
2.5146836083929074,
|
| 54 |
+
2.396254417443913,
|
| 55 |
+
2.430939371044743,
|
| 56 |
+
2.18786890103016,
|
| 57 |
+
2.9446809932599853,
|
| 58 |
+
2.3331955459252756,
|
| 59 |
+
2.7919380050354676,
|
| 60 |
+
2.6444312498997413
|
| 61 |
+
]
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"trainer": {
|
| 66 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_steps": 100000,
|
| 69 |
+
"batch_size_per_gpu": 4,
|
| 70 |
+
"batch_split": 4,
|
| 71 |
+
"prev_window": 2,
|
| 72 |
+
"coord_drop_prob": 0.0,
|
| 73 |
+
"p_clean": 1.0,
|
| 74 |
+
"max_clean_t": 1.0,
|
| 75 |
+
"cfg_noise_t": 0.9,
|
| 76 |
+
"history_cfg": 0.0,
|
| 77 |
+
"optimizer": {
|
| 78 |
+
"name": "AdamW",
|
| 79 |
+
"args": {
|
| 80 |
+
"lr": 0.0001,
|
| 81 |
+
"weight_decay": 0.0
|
| 82 |
+
}
|
| 83 |
+
},
|
| 84 |
+
"ema_rate": [
|
| 85 |
+
0.9999
|
| 86 |
+
],
|
| 87 |
+
"fp16_mode": "inflat_all",
|
| 88 |
+
"fp16_scale_growth": 0.001,
|
| 89 |
+
"grad_clip": {
|
| 90 |
+
"name": "AdaptiveGradClipper",
|
| 91 |
+
"args": {
|
| 92 |
+
"max_norm": 1.0,
|
| 93 |
+
"clip_percentile": 95
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"i_log": 10,
|
| 97 |
+
"i_print": 100,
|
| 98 |
+
"i_sample": 5000,
|
| 99 |
+
"i_save": 2000,
|
| 100 |
+
"p_uncond": 0.1,
|
| 101 |
+
"t_schedule": {
|
| 102 |
+
"name": "logitNormal",
|
| 103 |
+
"args": {
|
| 104 |
+
"mean": 1.0,
|
| 105 |
+
"std": 1.0
|
| 106 |
+
}
|
| 107 |
+
},
|
| 108 |
+
"sigma_min": 1e-05,
|
| 109 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 110 |
+
}
|
| 111 |
+
}
|
| 112 |
+
}
|
slat_abl_diff_causal_cdrop005/config.json
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16"
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dataset": {
|
| 31 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 32 |
+
"args": {
|
| 33 |
+
"prev_window": 2,
|
| 34 |
+
"val_num_scenes": 20,
|
| 35 |
+
"val_num_frames": 6,
|
| 36 |
+
"snapshot_num_scenes": 4,
|
| 37 |
+
"snapshot_num_frames": 4,
|
| 38 |
+
"image_size": 518,
|
| 39 |
+
"num_cond_views": 6,
|
| 40 |
+
"front_view_prob": 0.4,
|
| 41 |
+
"normalization": {
|
| 42 |
+
"mean": [
|
| 43 |
+
-2.2392261552952513,
|
| 44 |
+
0.7410145409701431,
|
| 45 |
+
0.5079064556582509,
|
| 46 |
+
-0.05298659832508365,
|
| 47 |
+
-0.025759045060578523,
|
| 48 |
+
0.8829246196309438,
|
| 49 |
+
-0.7059389169604169,
|
| 50 |
+
0.9353478047296675
|
| 51 |
+
],
|
| 52 |
+
"std": [
|
| 53 |
+
2.5146836083929074,
|
| 54 |
+
2.396254417443913,
|
| 55 |
+
2.430939371044743,
|
| 56 |
+
2.18786890103016,
|
| 57 |
+
2.9446809932599853,
|
| 58 |
+
2.3331955459252756,
|
| 59 |
+
2.7919380050354676,
|
| 60 |
+
2.6444312498997413
|
| 61 |
+
]
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"trainer": {
|
| 66 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_steps": 100000,
|
| 69 |
+
"batch_size_per_gpu": 4,
|
| 70 |
+
"batch_split": 4,
|
| 71 |
+
"prev_window": 2,
|
| 72 |
+
"coord_drop_prob": 0.05,
|
| 73 |
+
"p_clean": 1.0,
|
| 74 |
+
"max_clean_t": 1.0,
|
| 75 |
+
"cfg_noise_t": 0.9,
|
| 76 |
+
"history_cfg": 0.0,
|
| 77 |
+
"optimizer": {
|
| 78 |
+
"name": "AdamW",
|
| 79 |
+
"args": {
|
| 80 |
+
"lr": 0.0001,
|
| 81 |
+
"weight_decay": 0.0
|
| 82 |
+
}
|
| 83 |
+
},
|
| 84 |
+
"ema_rate": [
|
| 85 |
+
0.9999
|
| 86 |
+
],
|
| 87 |
+
"fp16_mode": "inflat_all",
|
| 88 |
+
"fp16_scale_growth": 0.001,
|
| 89 |
+
"grad_clip": {
|
| 90 |
+
"name": "AdaptiveGradClipper",
|
| 91 |
+
"args": {
|
| 92 |
+
"max_norm": 1.0,
|
| 93 |
+
"clip_percentile": 95
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"i_log": 10,
|
| 97 |
+
"i_print": 100,
|
| 98 |
+
"i_sample": 5000,
|
| 99 |
+
"i_save": 2000,
|
| 100 |
+
"p_uncond": 0.1,
|
| 101 |
+
"t_schedule": {
|
| 102 |
+
"name": "logitNormal",
|
| 103 |
+
"args": {
|
| 104 |
+
"mean": 1.0,
|
| 105 |
+
"std": 1.0
|
| 106 |
+
}
|
| 107 |
+
},
|
| 108 |
+
"sigma_min": 1e-05,
|
| 109 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 110 |
+
}
|
| 111 |
+
}
|
| 112 |
+
}
|
slat_abl_diff_causal_cdrop01/config.json
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16"
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dataset": {
|
| 31 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 32 |
+
"args": {
|
| 33 |
+
"prev_window": 2,
|
| 34 |
+
"val_num_scenes": 20,
|
| 35 |
+
"val_num_frames": 6,
|
| 36 |
+
"snapshot_num_scenes": 4,
|
| 37 |
+
"snapshot_num_frames": 4,
|
| 38 |
+
"image_size": 518,
|
| 39 |
+
"num_cond_views": 6,
|
| 40 |
+
"front_view_prob": 0.4,
|
| 41 |
+
"normalization": {
|
| 42 |
+
"mean": [
|
| 43 |
+
-2.2392261552952513,
|
| 44 |
+
0.7410145409701431,
|
| 45 |
+
0.5079064556582509,
|
| 46 |
+
-0.05298659832508365,
|
| 47 |
+
-0.025759045060578523,
|
| 48 |
+
0.8829246196309438,
|
| 49 |
+
-0.7059389169604169,
|
| 50 |
+
0.9353478047296675
|
| 51 |
+
],
|
| 52 |
+
"std": [
|
| 53 |
+
2.5146836083929074,
|
| 54 |
+
2.396254417443913,
|
| 55 |
+
2.430939371044743,
|
| 56 |
+
2.18786890103016,
|
| 57 |
+
2.9446809932599853,
|
| 58 |
+
2.3331955459252756,
|
| 59 |
+
2.7919380050354676,
|
| 60 |
+
2.6444312498997413
|
| 61 |
+
]
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"trainer": {
|
| 66 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_steps": 100000,
|
| 69 |
+
"batch_size_per_gpu": 4,
|
| 70 |
+
"batch_split": 4,
|
| 71 |
+
"prev_window": 2,
|
| 72 |
+
"coord_drop_prob": 0.1,
|
| 73 |
+
"p_clean": 1.0,
|
| 74 |
+
"max_clean_t": 1.0,
|
| 75 |
+
"cfg_noise_t": 0.9,
|
| 76 |
+
"history_cfg": 0.0,
|
| 77 |
+
"optimizer": {
|
| 78 |
+
"name": "AdamW",
|
| 79 |
+
"args": {
|
| 80 |
+
"lr": 0.0001,
|
| 81 |
+
"weight_decay": 0.0
|
| 82 |
+
}
|
| 83 |
+
},
|
| 84 |
+
"ema_rate": [
|
| 85 |
+
0.9999
|
| 86 |
+
],
|
| 87 |
+
"fp16_mode": "inflat_all",
|
| 88 |
+
"fp16_scale_growth": 0.001,
|
| 89 |
+
"grad_clip": {
|
| 90 |
+
"name": "AdaptiveGradClipper",
|
| 91 |
+
"args": {
|
| 92 |
+
"max_norm": 1.0,
|
| 93 |
+
"clip_percentile": 95
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"i_log": 10,
|
| 97 |
+
"i_print": 100,
|
| 98 |
+
"i_sample": 5000,
|
| 99 |
+
"i_save": 2000,
|
| 100 |
+
"p_uncond": 0.1,
|
| 101 |
+
"t_schedule": {
|
| 102 |
+
"name": "logitNormal",
|
| 103 |
+
"args": {
|
| 104 |
+
"mean": 1.0,
|
| 105 |
+
"std": 1.0
|
| 106 |
+
}
|
| 107 |
+
},
|
| 108 |
+
"sigma_min": 1e-05,
|
| 109 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 110 |
+
}
|
| 111 |
+
}
|
| 112 |
+
}
|
slat_abl_diff_causal_cdrop02/config.json
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16"
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dataset": {
|
| 31 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 32 |
+
"args": {
|
| 33 |
+
"prev_window": 2,
|
| 34 |
+
"val_num_scenes": 20,
|
| 35 |
+
"val_num_frames": 6,
|
| 36 |
+
"snapshot_num_scenes": 4,
|
| 37 |
+
"snapshot_num_frames": 4,
|
| 38 |
+
"image_size": 518,
|
| 39 |
+
"num_cond_views": 6,
|
| 40 |
+
"front_view_prob": 0.4,
|
| 41 |
+
"normalization": {
|
| 42 |
+
"mean": [
|
| 43 |
+
-2.2392261552952513,
|
| 44 |
+
0.7410145409701431,
|
| 45 |
+
0.5079064556582509,
|
| 46 |
+
-0.05298659832508365,
|
| 47 |
+
-0.025759045060578523,
|
| 48 |
+
0.8829246196309438,
|
| 49 |
+
-0.7059389169604169,
|
| 50 |
+
0.9353478047296675
|
| 51 |
+
],
|
| 52 |
+
"std": [
|
| 53 |
+
2.5146836083929074,
|
| 54 |
+
2.396254417443913,
|
| 55 |
+
2.430939371044743,
|
| 56 |
+
2.18786890103016,
|
| 57 |
+
2.9446809932599853,
|
| 58 |
+
2.3331955459252756,
|
| 59 |
+
2.7919380050354676,
|
| 60 |
+
2.6444312498997413
|
| 61 |
+
]
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"trainer": {
|
| 66 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_steps": 100000,
|
| 69 |
+
"batch_size_per_gpu": 4,
|
| 70 |
+
"batch_split": 4,
|
| 71 |
+
"prev_window": 2,
|
| 72 |
+
"coord_drop_prob": 0.2,
|
| 73 |
+
"p_clean": 1.0,
|
| 74 |
+
"max_clean_t": 1.0,
|
| 75 |
+
"cfg_noise_t": 0.9,
|
| 76 |
+
"history_cfg": 0.0,
|
| 77 |
+
"optimizer": {
|
| 78 |
+
"name": "AdamW",
|
| 79 |
+
"args": {
|
| 80 |
+
"lr": 0.0001,
|
| 81 |
+
"weight_decay": 0.0
|
| 82 |
+
}
|
| 83 |
+
},
|
| 84 |
+
"ema_rate": [
|
| 85 |
+
0.9999
|
| 86 |
+
],
|
| 87 |
+
"fp16_mode": "inflat_all",
|
| 88 |
+
"fp16_scale_growth": 0.001,
|
| 89 |
+
"grad_clip": {
|
| 90 |
+
"name": "AdaptiveGradClipper",
|
| 91 |
+
"args": {
|
| 92 |
+
"max_norm": 1.0,
|
| 93 |
+
"clip_percentile": 95
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"i_log": 10,
|
| 97 |
+
"i_print": 100,
|
| 98 |
+
"i_sample": 5000,
|
| 99 |
+
"i_save": 2000,
|
| 100 |
+
"p_uncond": 0.1,
|
| 101 |
+
"t_schedule": {
|
| 102 |
+
"name": "logitNormal",
|
| 103 |
+
"args": {
|
| 104 |
+
"mean": 1.0,
|
| 105 |
+
"std": 1.0
|
| 106 |
+
}
|
| 107 |
+
},
|
| 108 |
+
"sigma_min": 1e-05,
|
| 109 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 110 |
+
}
|
| 111 |
+
}
|
| 112 |
+
}
|
slat_abl_same_noise_20260428_031335/config.json
ADDED
|
@@ -0,0 +1,113 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingElasticSLatFlowModelV3",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 64,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 2,
|
| 15 |
+
"num_io_res_blocks": 2,
|
| 16 |
+
"io_block_channels": [
|
| 17 |
+
128
|
| 18 |
+
],
|
| 19 |
+
"pe_mode": "ape",
|
| 20 |
+
"qk_rms_norm": true,
|
| 21 |
+
"use_fp16": true,
|
| 22 |
+
"use_checkpoint": true,
|
| 23 |
+
"window_size": 3,
|
| 24 |
+
"trainable_scope": "self_attn",
|
| 25 |
+
"temporal_pe_mode": "rope",
|
| 26 |
+
"pretrained_base": "microsoft/TRELLIS-image-large/ckpts/slat_flow_img_dit_L_64l8p2_fp16"
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dataset": {
|
| 31 |
+
"name": "DynamicSequentialImageConditionedSLat",
|
| 32 |
+
"args": {
|
| 33 |
+
"prev_window": 2,
|
| 34 |
+
"val_num_scenes": 20,
|
| 35 |
+
"val_num_frames": 6,
|
| 36 |
+
"snapshot_num_scenes": 4,
|
| 37 |
+
"snapshot_num_frames": 4,
|
| 38 |
+
"image_size": 518,
|
| 39 |
+
"num_cond_views": 6,
|
| 40 |
+
"front_view_prob": 0.4,
|
| 41 |
+
"normalization": {
|
| 42 |
+
"mean": [
|
| 43 |
+
-2.2392261552952513,
|
| 44 |
+
0.7410145409701431,
|
| 45 |
+
0.5079064556582509,
|
| 46 |
+
-0.05298659832508365,
|
| 47 |
+
-0.025759045060578523,
|
| 48 |
+
0.8829246196309438,
|
| 49 |
+
-0.7059389169604169,
|
| 50 |
+
0.9353478047296675
|
| 51 |
+
],
|
| 52 |
+
"std": [
|
| 53 |
+
2.5146836083929074,
|
| 54 |
+
2.396254417443913,
|
| 55 |
+
2.430939371044743,
|
| 56 |
+
2.18786890103016,
|
| 57 |
+
2.9446809932599853,
|
| 58 |
+
2.3331955459252756,
|
| 59 |
+
2.7919380050354676,
|
| 60 |
+
2.6444312498997413
|
| 61 |
+
]
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"trainer": {
|
| 66 |
+
"name": "DiffusionForcingImageConditionedSparseFlowMatchingCFGTrainer",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_steps": 100000,
|
| 69 |
+
"batch_size_per_gpu": 4,
|
| 70 |
+
"batch_split": 4,
|
| 71 |
+
"prev_window": 2,
|
| 72 |
+
"coord_drop_prob": 0.05,
|
| 73 |
+
"p_clean": 1.0,
|
| 74 |
+
"max_clean_t": 1.0,
|
| 75 |
+
"cfg_noise_t": 0.9,
|
| 76 |
+
"history_cfg": 0.0,
|
| 77 |
+
"noise_scheme": "shared",
|
| 78 |
+
"optimizer": {
|
| 79 |
+
"name": "AdamW",
|
| 80 |
+
"args": {
|
| 81 |
+
"lr": 0.0001,
|
| 82 |
+
"weight_decay": 0.0
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"ema_rate": [
|
| 86 |
+
0.9999
|
| 87 |
+
],
|
| 88 |
+
"fp16_mode": "inflat_all",
|
| 89 |
+
"fp16_scale_growth": 0.001,
|
| 90 |
+
"grad_clip": {
|
| 91 |
+
"name": "AdaptiveGradClipper",
|
| 92 |
+
"args": {
|
| 93 |
+
"max_norm": 1.0,
|
| 94 |
+
"clip_percentile": 95
|
| 95 |
+
}
|
| 96 |
+
},
|
| 97 |
+
"i_log": 10,
|
| 98 |
+
"i_print": 100,
|
| 99 |
+
"i_sample": 5000,
|
| 100 |
+
"i_save": 2000,
|
| 101 |
+
"p_uncond": 0.1,
|
| 102 |
+
"t_schedule": {
|
| 103 |
+
"name": "logitNormal",
|
| 104 |
+
"args": {
|
| 105 |
+
"mean": 1.0,
|
| 106 |
+
"std": 1.0
|
| 107 |
+
}
|
| 108 |
+
},
|
| 109 |
+
"sigma_min": 1e-05,
|
| 110 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 111 |
+
}
|
| 112 |
+
}
|
| 113 |
+
}
|
ss_abl_bidirectional_20260427_175112/config.json
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingTemporalSparseStructureFlowModel",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 16,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 1,
|
| 15 |
+
"pe_mode": "ape",
|
| 16 |
+
"qk_rms_norm": true,
|
| 17 |
+
"use_fp16": true,
|
| 18 |
+
"use_checkpoint": false,
|
| 19 |
+
"window_size": 3,
|
| 20 |
+
"trainable_scope": "self_attn_only",
|
| 21 |
+
"causal_attn": false
|
| 22 |
+
}
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"dataset": {
|
| 26 |
+
"name": "DynamicSequentialImageConditionedSS",
|
| 27 |
+
"args": {
|
| 28 |
+
"image_size": 518,
|
| 29 |
+
"val_num_scenes": 20,
|
| 30 |
+
"val_num_frames": null,
|
| 31 |
+
"snapshot_num_scenes": 1,
|
| 32 |
+
"snapshot_num_frames": 4,
|
| 33 |
+
"num_cond_views": 6,
|
| 34 |
+
"front_view_prob": 0.4,
|
| 35 |
+
"min_frames": 3,
|
| 36 |
+
"pretrained_ss_dec": "microsoft/TRELLIS-image-large/ckpts/ss_dec_conv3d_16l8_fp16"
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"trainer": {
|
| 40 |
+
"name": "DiffusionForcingTemporalSSImageConditionedCFGTrainer",
|
| 41 |
+
"args": {
|
| 42 |
+
"max_steps": 200000,
|
| 43 |
+
"batch_size_per_gpu": 4,
|
| 44 |
+
"batch_split": 4,
|
| 45 |
+
"window_size": 3,
|
| 46 |
+
"snapshot_steps": 50,
|
| 47 |
+
"snapshot_cfg_strength": 3.0,
|
| 48 |
+
"snapshot_rollout_frames": 4,
|
| 49 |
+
"noise_scheme": "independent",
|
| 50 |
+
"history_guidance_mode": "none",
|
| 51 |
+
"history_guidance_strength": 1.0,
|
| 52 |
+
"history_guidance_sigma": 0.5,
|
| 53 |
+
"history_guidance_num_noised": 1,
|
| 54 |
+
"optimizer": {
|
| 55 |
+
"name": "AdamW",
|
| 56 |
+
"args": {
|
| 57 |
+
"lr": 0.0001,
|
| 58 |
+
"weight_decay": 0.0
|
| 59 |
+
}
|
| 60 |
+
},
|
| 61 |
+
"ema_rate": [
|
| 62 |
+
0.9999
|
| 63 |
+
],
|
| 64 |
+
"fp16_mode": "inflat_all",
|
| 65 |
+
"fp16_scale_growth": 0.001,
|
| 66 |
+
"grad_clip": {
|
| 67 |
+
"name": "AdaptiveGradClipper",
|
| 68 |
+
"args": {
|
| 69 |
+
"max_norm": 1.0,
|
| 70 |
+
"clip_percentile": 95
|
| 71 |
+
}
|
| 72 |
+
},
|
| 73 |
+
"i_log": 10,
|
| 74 |
+
"i_print": 100,
|
| 75 |
+
"i_sample": 5000,
|
| 76 |
+
"i_save": 2000,
|
| 77 |
+
"p_uncond": 0.1,
|
| 78 |
+
"t_schedule": {
|
| 79 |
+
"name": "uniform"
|
| 80 |
+
},
|
| 81 |
+
"sigma_min": 1e-05,
|
| 82 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
}
|
ss_abl_diff_causal_20260427_175045/config.json
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingTemporalSparseStructureFlowModel",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 16,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 1,
|
| 15 |
+
"pe_mode": "ape",
|
| 16 |
+
"qk_rms_norm": true,
|
| 17 |
+
"use_fp16": true,
|
| 18 |
+
"use_checkpoint": false,
|
| 19 |
+
"window_size": 3,
|
| 20 |
+
"trainable_scope": "self_attn_only"
|
| 21 |
+
}
|
| 22 |
+
}
|
| 23 |
+
},
|
| 24 |
+
"dataset": {
|
| 25 |
+
"name": "DynamicSequentialImageConditionedSS",
|
| 26 |
+
"args": {
|
| 27 |
+
"image_size": 518,
|
| 28 |
+
"val_num_scenes": 20,
|
| 29 |
+
"val_num_frames": null,
|
| 30 |
+
"snapshot_num_scenes": 1,
|
| 31 |
+
"snapshot_num_frames": 4,
|
| 32 |
+
"num_cond_views": 6,
|
| 33 |
+
"front_view_prob": 0.4,
|
| 34 |
+
"min_frames": 3,
|
| 35 |
+
"pretrained_ss_dec": "microsoft/TRELLIS-image-large/ckpts/ss_dec_conv3d_16l8_fp16"
|
| 36 |
+
}
|
| 37 |
+
},
|
| 38 |
+
"trainer": {
|
| 39 |
+
"name": "DiffusionForcingTemporalSSImageConditionedCFGTrainer",
|
| 40 |
+
"args": {
|
| 41 |
+
"max_steps": 200000,
|
| 42 |
+
"batch_size_per_gpu": 4,
|
| 43 |
+
"batch_split": 4,
|
| 44 |
+
"window_size": 3,
|
| 45 |
+
"snapshot_steps": 50,
|
| 46 |
+
"snapshot_cfg_strength": 3.0,
|
| 47 |
+
"snapshot_rollout_frames": 4,
|
| 48 |
+
"noise_scheme": "independent",
|
| 49 |
+
"history_guidance_mode": "none",
|
| 50 |
+
"history_guidance_strength": 1.0,
|
| 51 |
+
"history_guidance_sigma": 0.5,
|
| 52 |
+
"history_guidance_num_noised": 1,
|
| 53 |
+
"optimizer": {
|
| 54 |
+
"name": "AdamW",
|
| 55 |
+
"args": {
|
| 56 |
+
"lr": 0.0001,
|
| 57 |
+
"weight_decay": 0.0
|
| 58 |
+
}
|
| 59 |
+
},
|
| 60 |
+
"ema_rate": [
|
| 61 |
+
0.9999
|
| 62 |
+
],
|
| 63 |
+
"fp16_mode": "inflat_all",
|
| 64 |
+
"fp16_scale_growth": 0.001,
|
| 65 |
+
"grad_clip": {
|
| 66 |
+
"name": "AdaptiveGradClipper",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_norm": 1.0,
|
| 69 |
+
"clip_percentile": 95
|
| 70 |
+
}
|
| 71 |
+
},
|
| 72 |
+
"i_log": 10,
|
| 73 |
+
"i_print": 100,
|
| 74 |
+
"i_sample": 5000,
|
| 75 |
+
"i_save": 2000,
|
| 76 |
+
"p_uncond": 0.1,
|
| 77 |
+
"t_schedule": {
|
| 78 |
+
"name": "uniform"
|
| 79 |
+
},
|
| 80 |
+
"sigma_min": 1e-05,
|
| 81 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 82 |
+
}
|
| 83 |
+
}
|
| 84 |
+
}
|
ss_abl_same_noise_20260427_175113/config.json
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"models": {
|
| 3 |
+
"denoiser": {
|
| 4 |
+
"name": "DiffusionForcingTemporalSparseStructureFlowModel",
|
| 5 |
+
"args": {
|
| 6 |
+
"resolution": 16,
|
| 7 |
+
"in_channels": 8,
|
| 8 |
+
"out_channels": 8,
|
| 9 |
+
"model_channels": 1024,
|
| 10 |
+
"cond_channels": 1024,
|
| 11 |
+
"num_blocks": 24,
|
| 12 |
+
"num_heads": 16,
|
| 13 |
+
"mlp_ratio": 4,
|
| 14 |
+
"patch_size": 1,
|
| 15 |
+
"pe_mode": "ape",
|
| 16 |
+
"qk_rms_norm": true,
|
| 17 |
+
"use_fp16": true,
|
| 18 |
+
"use_checkpoint": false,
|
| 19 |
+
"window_size": 3,
|
| 20 |
+
"trainable_scope": "self_attn_only"
|
| 21 |
+
}
|
| 22 |
+
}
|
| 23 |
+
},
|
| 24 |
+
"dataset": {
|
| 25 |
+
"name": "DynamicSequentialImageConditionedSS",
|
| 26 |
+
"args": {
|
| 27 |
+
"image_size": 518,
|
| 28 |
+
"val_num_scenes": 20,
|
| 29 |
+
"val_num_frames": null,
|
| 30 |
+
"snapshot_num_scenes": 1,
|
| 31 |
+
"snapshot_num_frames": 4,
|
| 32 |
+
"num_cond_views": 6,
|
| 33 |
+
"front_view_prob": 0.4,
|
| 34 |
+
"min_frames": 3,
|
| 35 |
+
"pretrained_ss_dec": "microsoft/TRELLIS-image-large/ckpts/ss_dec_conv3d_16l8_fp16"
|
| 36 |
+
}
|
| 37 |
+
},
|
| 38 |
+
"trainer": {
|
| 39 |
+
"name": "DiffusionForcingTemporalSSImageConditionedCFGTrainer",
|
| 40 |
+
"args": {
|
| 41 |
+
"max_steps": 200000,
|
| 42 |
+
"batch_size_per_gpu": 4,
|
| 43 |
+
"batch_split": 4,
|
| 44 |
+
"window_size": 3,
|
| 45 |
+
"snapshot_steps": 50,
|
| 46 |
+
"snapshot_cfg_strength": 3.0,
|
| 47 |
+
"snapshot_rollout_frames": 4,
|
| 48 |
+
"noise_scheme": "shared",
|
| 49 |
+
"history_guidance_mode": "none",
|
| 50 |
+
"history_guidance_strength": 1.0,
|
| 51 |
+
"history_guidance_sigma": 0.5,
|
| 52 |
+
"history_guidance_num_noised": 1,
|
| 53 |
+
"optimizer": {
|
| 54 |
+
"name": "AdamW",
|
| 55 |
+
"args": {
|
| 56 |
+
"lr": 0.0001,
|
| 57 |
+
"weight_decay": 0.0
|
| 58 |
+
}
|
| 59 |
+
},
|
| 60 |
+
"ema_rate": [
|
| 61 |
+
0.9999
|
| 62 |
+
],
|
| 63 |
+
"fp16_mode": "inflat_all",
|
| 64 |
+
"fp16_scale_growth": 0.001,
|
| 65 |
+
"grad_clip": {
|
| 66 |
+
"name": "AdaptiveGradClipper",
|
| 67 |
+
"args": {
|
| 68 |
+
"max_norm": 1.0,
|
| 69 |
+
"clip_percentile": 95
|
| 70 |
+
}
|
| 71 |
+
},
|
| 72 |
+
"i_log": 10,
|
| 73 |
+
"i_print": 100,
|
| 74 |
+
"i_sample": 5000,
|
| 75 |
+
"i_save": 2000,
|
| 76 |
+
"p_uncond": 0.1,
|
| 77 |
+
"t_schedule": {
|
| 78 |
+
"name": "uniform"
|
| 79 |
+
},
|
| 80 |
+
"sigma_min": 1e-05,
|
| 81 |
+
"image_cond_model": "dinov2_vitl14_reg"
|
| 82 |
+
}
|
| 83 |
+
}
|
| 84 |
+
}
|