Upload config.json with huggingface_hub
Browse files- config.json +116 -0
config.json
ADDED
|
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"transformer": {
|
| 3 |
+
"resolution": 512,
|
| 4 |
+
"patch_size": 2,
|
| 5 |
+
"mlp_ratio": 4,
|
| 6 |
+
"use_qknorm": true,
|
| 7 |
+
"use_swiglu": true,
|
| 8 |
+
"use_rope": true,
|
| 9 |
+
"use_rmsnorm": true,
|
| 10 |
+
"wo_shift": false,
|
| 11 |
+
"dim": 1536,
|
| 12 |
+
"depth": 36,
|
| 13 |
+
"num_heads": 24,
|
| 14 |
+
"ae_type": "qwen",
|
| 15 |
+
"ae_path": "preset/ckpts/Qwen-Image-vae-2d",
|
| 16 |
+
"time_dist": [
|
| 17 |
+
"lognorm",
|
| 18 |
+
0,
|
| 19 |
+
1.0
|
| 20 |
+
],
|
| 21 |
+
"cfg_ratio": 0.1,
|
| 22 |
+
"cfg_scale": 1.5,
|
| 23 |
+
"dinov2_size": 448,
|
| 24 |
+
"enc_type": "dinov2l",
|
| 25 |
+
"enc_dim": 1024,
|
| 26 |
+
"layer_dinov2b_list": [
|
| 27 |
+
17
|
| 28 |
+
],
|
| 29 |
+
"interp_type": "lin",
|
| 30 |
+
"encdim_ratio": 3,
|
| 31 |
+
"weak_cond_strength_aelq_list": [
|
| 32 |
+
0,
|
| 33 |
+
0
|
| 34 |
+
],
|
| 35 |
+
"cond_strength_aelq_list": [
|
| 36 |
+
5,
|
| 37 |
+
1
|
| 38 |
+
],
|
| 39 |
+
"t_start": 0,
|
| 40 |
+
"t_end": 1,
|
| 41 |
+
"auxiliary_time_cond": false,
|
| 42 |
+
"distill_type": "onestep"
|
| 43 |
+
},
|
| 44 |
+
"vae": {
|
| 45 |
+
"_class_name": "AutoencoderKLQwenImage2D",
|
| 46 |
+
"_diffusers_version": "0.35.0",
|
| 47 |
+
"attn_scales": [],
|
| 48 |
+
"base_dim": 96,
|
| 49 |
+
"dim_mult": [
|
| 50 |
+
1,
|
| 51 |
+
2,
|
| 52 |
+
4,
|
| 53 |
+
4
|
| 54 |
+
],
|
| 55 |
+
"dropout": 0.0,
|
| 56 |
+
"latents_mean": [
|
| 57 |
+
-0.7571,
|
| 58 |
+
-0.7089,
|
| 59 |
+
-0.9113,
|
| 60 |
+
0.1075,
|
| 61 |
+
-0.1745,
|
| 62 |
+
0.9653,
|
| 63 |
+
-0.1517,
|
| 64 |
+
1.5508,
|
| 65 |
+
0.4134,
|
| 66 |
+
-0.0715,
|
| 67 |
+
0.5517,
|
| 68 |
+
-0.3632,
|
| 69 |
+
-0.1922,
|
| 70 |
+
-0.9497,
|
| 71 |
+
0.2503,
|
| 72 |
+
-0.2921
|
| 73 |
+
],
|
| 74 |
+
"latents_std": [
|
| 75 |
+
2.8184,
|
| 76 |
+
1.4541,
|
| 77 |
+
2.3275,
|
| 78 |
+
2.6558,
|
| 79 |
+
1.2196,
|
| 80 |
+
1.7708,
|
| 81 |
+
2.6052,
|
| 82 |
+
2.0743,
|
| 83 |
+
3.2687,
|
| 84 |
+
2.1526,
|
| 85 |
+
2.8652,
|
| 86 |
+
1.5579,
|
| 87 |
+
1.6382,
|
| 88 |
+
1.1253,
|
| 89 |
+
2.8251,
|
| 90 |
+
1.916
|
| 91 |
+
],
|
| 92 |
+
"non_linearity": "silu",
|
| 93 |
+
"num_res_blocks": 2,
|
| 94 |
+
"temperal_downsample": [
|
| 95 |
+
false,
|
| 96 |
+
true,
|
| 97 |
+
true
|
| 98 |
+
],
|
| 99 |
+
"z_dim": 16
|
| 100 |
+
},
|
| 101 |
+
"dinov2": {
|
| 102 |
+
"arch": "vit_large",
|
| 103 |
+
"patch_size": 14,
|
| 104 |
+
"embed_dim": 1024,
|
| 105 |
+
"depth_kept": 18,
|
| 106 |
+
"num_heads": 16,
|
| 107 |
+
"mlp_ratio": 4,
|
| 108 |
+
"layerscale": true,
|
| 109 |
+
"qkv_bias": true,
|
| 110 |
+
"ln_eps": 1e-06,
|
| 111 |
+
"input_size": 448,
|
| 112 |
+
"feature_layer": 17,
|
| 113 |
+
"pos_embed": "interpolated to 448\u00b2 (offset 0.1, bicubic)"
|
| 114 |
+
},
|
| 115 |
+
"key_contract": "conv (O,kh,kw,I); Sequential children \u2192 layers.N; VAE gamma (C); DINOv2 blocks 0..17 only"
|
| 116 |
+
}
|