Push model using huggingface_hub.
Browse files- config.json +109 -1
config.json
CHANGED
@@ -1,8 +1,116 @@
|
|
1 |
{
|
2 |
-
"ckpt_path":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
"cond_stage_key": "text",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
"first_stage_key": "surface",
|
5 |
"ignore_keys": [],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
"scale_by_std": false,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
"z_scale_factor": 1.0
|
8 |
}
|
|
|
1 |
{
|
2 |
+
"ckpt_path": null,
|
3 |
+
"cond_stage_config": {
|
4 |
+
"params": {
|
5 |
+
"max_length": 77,
|
6 |
+
"version": "openai/clip-vit-large-patch14",
|
7 |
+
"zero_embedding_radio": 0.1
|
8 |
+
},
|
9 |
+
"target": "michelangelo.models.conditional_encoders.encoder_factory.FrozenAlignedCLIPTextEmbedder"
|
10 |
+
},
|
11 |
"cond_stage_key": "text",
|
12 |
+
"denoiser_cfg": {
|
13 |
+
"params": {
|
14 |
+
"context_dim": 768,
|
15 |
+
"heads": 12,
|
16 |
+
"init_scale": 1.0,
|
17 |
+
"input_channels": 64,
|
18 |
+
"layers": 8,
|
19 |
+
"n_ctx": 256,
|
20 |
+
"output_channels": 64,
|
21 |
+
"skip_ln": true,
|
22 |
+
"use_checkpoint": true,
|
23 |
+
"width": 768
|
24 |
+
},
|
25 |
+
"target": "michelangelo.models.asl_diffusion.asl_udt.ConditionalASLUDTDenoiser"
|
26 |
+
},
|
27 |
+
"first_stage_config": {
|
28 |
+
"params": {
|
29 |
+
"aligned_module_cfg": {
|
30 |
+
"params": {
|
31 |
+
"clip_model_version": "openai/clip-vit-large-patch14"
|
32 |
+
},
|
33 |
+
"target": "michelangelo.models.tsal.clip_asl_module.CLIPAlignedShapeAsLatentModule"
|
34 |
+
},
|
35 |
+
"loss_cfg": {
|
36 |
+
"target": "torch.nn.Identity"
|
37 |
+
},
|
38 |
+
"shape_module_cfg": {
|
39 |
+
"params": {
|
40 |
+
"embed_dim": 64,
|
41 |
+
"heads": 12,
|
42 |
+
"include_pi": false,
|
43 |
+
"init_scale": 0.25,
|
44 |
+
"num_decoder_layers": 16,
|
45 |
+
"num_encoder_layers": 8,
|
46 |
+
"num_freqs": 8,
|
47 |
+
"num_latents": 256,
|
48 |
+
"point_feats": 3,
|
49 |
+
"qkv_bias": false,
|
50 |
+
"use_checkpoint": true,
|
51 |
+
"use_ln_post": true,
|
52 |
+
"width": 768
|
53 |
+
},
|
54 |
+
"target": "michelangelo.models.tsal.sal_perceiver.AlignedShapeLatentPerceiver"
|
55 |
+
}
|
56 |
+
},
|
57 |
+
"target": "michelangelo.models.tsal.asl_pl_module.AlignedShapeAsLatentPLModule"
|
58 |
+
},
|
59 |
"first_stage_key": "surface",
|
60 |
"ignore_keys": [],
|
61 |
+
"loss_cfg": {
|
62 |
+
"loss_type": "mse"
|
63 |
+
},
|
64 |
+
"optimizer_cfg": {
|
65 |
+
"optimizer": {
|
66 |
+
"params": {
|
67 |
+
"betas": [
|
68 |
+
0.9,
|
69 |
+
0.99
|
70 |
+
],
|
71 |
+
"eps": 1e-06,
|
72 |
+
"weight_decay": 0.01
|
73 |
+
},
|
74 |
+
"target": "torch.optim.AdamW"
|
75 |
+
},
|
76 |
+
"scheduler": {
|
77 |
+
"params": {
|
78 |
+
"f_max": 1.0,
|
79 |
+
"f_min": 0.001,
|
80 |
+
"f_start": 1e-06,
|
81 |
+
"warm_up_steps": 5000
|
82 |
+
},
|
83 |
+
"target": "michelangelo.utils.trainings.lr_scheduler.LambdaWarmUpCosineFactorScheduler"
|
84 |
+
}
|
85 |
+
},
|
86 |
"scale_by_std": false,
|
87 |
+
"scheduler_cfg": {
|
88 |
+
"denoise": {
|
89 |
+
"params": {
|
90 |
+
"beta_end": 0.012,
|
91 |
+
"beta_schedule": "scaled_linear",
|
92 |
+
"beta_start": 0.00085,
|
93 |
+
"clip_sample": false,
|
94 |
+
"num_train_timesteps": 1000,
|
95 |
+
"set_alpha_to_one": false,
|
96 |
+
"steps_offset": 1
|
97 |
+
},
|
98 |
+
"target": "diffusers.schedulers.DDIMScheduler"
|
99 |
+
},
|
100 |
+
"eta": 0.0,
|
101 |
+
"guidance_scale": 7.5,
|
102 |
+
"noise": {
|
103 |
+
"params": {
|
104 |
+
"beta_end": 0.012,
|
105 |
+
"beta_schedule": "scaled_linear",
|
106 |
+
"beta_start": 0.00085,
|
107 |
+
"clip_sample": false,
|
108 |
+
"num_train_timesteps": 1000,
|
109 |
+
"variance_type": "fixed_small"
|
110 |
+
},
|
111 |
+
"target": "diffusers.schedulers.DDPMScheduler"
|
112 |
+
},
|
113 |
+
"num_inference_steps": 50
|
114 |
+
},
|
115 |
"z_scale_factor": 1.0
|
116 |
}
|