Hugging Face's logo

terminusresearch
/
sana-1.6b-1024px

Text-to-Image
Diffusers
Safetensors
SanaPipeline
Model card Files
xet
Community
1
bghira commited on
Commit
804f607
·
verified ·
1 Parent(s): 4e33b93

Update vae/config.json

Browse files
Files changed (1) hide show
  1. vae/config.json +53 -27
vae/config.json CHANGED
@@ -1,17 +1,25 @@
1
  {
2
- "_class_name": "DCAE",
3
  "_diffusers_version": "0.32.0.dev0",
4
- "_name_or_path": "Efficient-Large-Model/dc_ae_f32c32_sana_1.0_diffusers",
5
- "decoder_act": "silu",
6
- "decoder_block_type": [
 
 
 
 
 
 
 
 
7
  "ResBlock",
8
  "ResBlock",
9
  "ResBlock",
10
- "EViTS5GLU",
11
- "EViTS5GLU",
12
- "EViTS5GLU"
13
  ],
14
- "decoder_depth_list": [
15
  3,
16
  3,
17
  3,
@@ -19,8 +27,23 @@
19
  3,
20
  3
21
  ],
22
- "decoder_norm": "rms2d",
23
- "decoder_width_list": [
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24
  128,
25
  256,
26
  512,
@@ -28,17 +51,15 @@
28
  1024,
29
  1024
30
  ],
31
- "downsample_block_type": "Conv",
32
- "encoder_act": "silu",
33
- "encoder_block_type": [
34
  "ResBlock",
35
  "ResBlock",
36
  "ResBlock",
37
- "EViTS5GLU",
38
- "EViTS5GLU",
39
- "EViTS5GLU"
40
  ],
41
- "encoder_depth_list": [
42
  2,
43
  2,
44
  2,
@@ -46,17 +67,22 @@
46
  3,
47
  3
48
  ],
49
- "encoder_norm": "rms2d",
50
- "encoder_width_list": [
51
- 128,
52
- 256,
53
- 512,
54
- 512,
55
- 1024,
56
- 1024
 
 
 
 
 
57
  ],
58
  "in_channels": 3,
59
  "latent_channels": 32,
60
  "scaling_factor": 0.41407,
61
- "upsample_block_type": "InterpolateConv"
62
- }
 
1
  {
2
+ "_class_name": "AutoencoderDC",
3
  "_diffusers_version": "0.32.0.dev0",
4
+ "attention_head_dim": 32,
5
+ "decoder_act_fns": "silu",
6
+ "decoder_block_out_channels": [
7
+ 128,
8
+ 256,
9
+ 512,
10
+ 512,
11
+ 1024,
12
+ 1024
13
+ ],
14
+ "decoder_block_types": [
15
  "ResBlock",
16
  "ResBlock",
17
  "ResBlock",
18
+ "EfficientViTBlock",
19
+ "EfficientViTBlock",
20
+ "EfficientViTBlock"
21
  ],
22
+ "decoder_layers_per_block": [
23
  3,
24
  3,
25
  3,
 
27
  3,
28
  3
29
  ],
30
+ "decoder_norm_types": "rms_norm",
31
+ "decoder_qkv_multiscales": [
32
+ [],
33
+ [],
34
+ [],
35
+ [
36
+ 5
37
+ ],
38
+ [
39
+ 5
40
+ ],
41
+ [
42
+ 5
43
+ ]
44
+ ],
45
+ "downsample_block_type": "Conv",
46
+ "encoder_block_out_channels": [
47
  128,
48
  256,
49
  512,
 
51
  1024,
52
  1024
53
  ],
54
+ "encoder_block_types": [
 
 
55
  "ResBlock",
56
  "ResBlock",
57
  "ResBlock",
58
+ "EfficientViTBlock",
59
+ "EfficientViTBlock",
60
+ "EfficientViTBlock"
61
  ],
62
+ "encoder_layers_per_block": [
63
  2,
64
  2,
65
  2,
 
67
  3,
68
  3
69
  ],
70
+ "encoder_qkv_multiscales": [
71
+ [],
72
+ [],
73
+ [],
74
+ [
75
+ 5
76
+ ],
77
+ [
78
+ 5
79
+ ],
80
+ [
81
+ 5
82
+ ]
83
  ],
84
  "in_channels": 3,
85
  "latent_channels": 32,
86
  "scaling_factor": 0.41407,
87
+ "upsample_block_type": "interpolate"
88
+ }