dek924 commited on
Commit
86b2d29
·
verified ·
1 Parent(s): ca4805a

Upload config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. config.json +110 -0
config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "ehrxdiff",
3
+ "linear_start": 0.0015,
4
+ "linear_end": 0.0295,
5
+ "num_timesteps_cond": 1,
6
+ "log_every_t": 200,
7
+ "timesteps": 1000,
8
+ "first_stage_key": "target_img",
9
+ "cond_stage_key": "table, prev_img",
10
+ "image_size": 64,
11
+ "channels": 3,
12
+ "cond_stage_trainable": true,
13
+ "conditioning_key": "crossattn",
14
+ "monitor": "val/loss_simple_ema",
15
+ "scale_factor": 0.18215,
16
+ "use_ema": true,
17
+ "load_ema": false,
18
+ "unet_config": {
19
+ "target": "cheff.ldm.modules.diffusionmodules.openaimodel.UNetModel",
20
+ "params": {
21
+ "image_size": 64,
22
+ "in_channels": 3,
23
+ "out_channels": 3,
24
+ "model_channels": 224,
25
+ "attention_resolutions": [
26
+ 8,
27
+ 4,
28
+ 2
29
+ ],
30
+ "num_res_blocks": 2,
31
+ "channel_mult": [
32
+ 1,
33
+ 2,
34
+ 4,
35
+ 4
36
+ ],
37
+ "num_heads": 8,
38
+ "use_spatial_transformer": true,
39
+ "transformer_depth": 1,
40
+ "context_dim": 768,
41
+ "use_checkpoint": true,
42
+ "legacy": false
43
+ }
44
+ },
45
+ "first_stage_config": {
46
+ "target": "cheff.ldm.models.autoencoder.AutoencoderKL",
47
+ "params": {
48
+ "embed_dim": 3,
49
+ "ckpt_path": null,
50
+ "ddconfig": {
51
+ "double_z": true,
52
+ "z_channels": 3,
53
+ "resolution": 256,
54
+ "in_channels": 3,
55
+ "out_ch": 3,
56
+ "ch": 128,
57
+ "ch_mult": [
58
+ 1,
59
+ 2,
60
+ 4
61
+ ],
62
+ "num_res_blocks": 2,
63
+ "attn_resolutions": [],
64
+ "dropout": 0.0
65
+ },
66
+ "lossconfig": {
67
+ "target": "torch.nn.Identity"
68
+ }
69
+ }
70
+ },
71
+ "cond_stage_config": {
72
+ "target": "cheff.ldm.modules.encoders.modules.MultiModalTransformerAdaptor",
73
+ "params": {
74
+ "autoencoder_config": {
75
+ "embed_dim": 3,
76
+ "ckpt_path": null,
77
+ "ddconfig": {
78
+ "double_z": true,
79
+ "z_channels": 3,
80
+ "resolution": 256,
81
+ "in_channels": 3,
82
+ "out_ch": 3,
83
+ "ch": 128,
84
+ "ch_mult": [
85
+ 1,
86
+ 2,
87
+ 4
88
+ ],
89
+ "num_res_blocks": 2,
90
+ "attn_resolutions": [],
91
+ "dropout": 0.0
92
+ },
93
+ "lossconfig": {
94
+ "target": "torch.nn.Identity"
95
+ }
96
+ },
97
+ "clip_visual_enc_config": {
98
+ "input_resolution": 256,
99
+ "layers": 12,
100
+ "width": 768,
101
+ "patch_size": 32,
102
+ "heads": 12
103
+ },
104
+ "clip_enc_checkpoint": "checkpoints/clip_vit32_256_1024.ckpt",
105
+ "context_dim": 768,
106
+ "condition_feat_dim": 1024,
107
+ "clip_trainable": true
108
+ }
109
+ }
110
+ }