Upload config.json with huggingface_hub
Browse files- config.json +89 -0
config.json
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"type": "hybrid_act_diffusion",
|
| 3 |
+
"n_obs_steps": 1,
|
| 4 |
+
"normalization_mapping": {
|
| 5 |
+
"VISUAL": "MEAN_STD",
|
| 6 |
+
"STATE": "MEAN_STD",
|
| 7 |
+
"ACTION": "MEAN_STD"
|
| 8 |
+
},
|
| 9 |
+
"input_features": {
|
| 10 |
+
"observation.images.top": {
|
| 11 |
+
"type": "VISUAL",
|
| 12 |
+
"shape": [
|
| 13 |
+
3,
|
| 14 |
+
480,
|
| 15 |
+
640
|
| 16 |
+
]
|
| 17 |
+
},
|
| 18 |
+
"observation.state": {
|
| 19 |
+
"type": "STATE",
|
| 20 |
+
"shape": [
|
| 21 |
+
14
|
| 22 |
+
]
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"output_features": {
|
| 26 |
+
"action": {
|
| 27 |
+
"type": "ACTION",
|
| 28 |
+
"shape": [
|
| 29 |
+
14
|
| 30 |
+
]
|
| 31 |
+
}
|
| 32 |
+
},
|
| 33 |
+
"device": "cuda",
|
| 34 |
+
"use_amp": false,
|
| 35 |
+
"chunk_size": 100,
|
| 36 |
+
"n_action_steps": 100,
|
| 37 |
+
"temporal_ensemble_coeff": null,
|
| 38 |
+
"vision_backbone": "resnet18",
|
| 39 |
+
"pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
|
| 40 |
+
"replace_final_stride_with_dilation": false,
|
| 41 |
+
"use_group_norm": true,
|
| 42 |
+
"pre_norm": false,
|
| 43 |
+
"dim_model": 512,
|
| 44 |
+
"n_heads": 8,
|
| 45 |
+
"dim_feedforward": 3200,
|
| 46 |
+
"feedforward_activation": "relu",
|
| 47 |
+
"n_encoder_layers": 4,
|
| 48 |
+
"n_decoder_layers": 1,
|
| 49 |
+
"use_vae": true,
|
| 50 |
+
"latent_dim": 32,
|
| 51 |
+
"n_vae_encoder_layers": 4,
|
| 52 |
+
"noise_scheduler_type": "DDPM",
|
| 53 |
+
"num_train_timesteps": 50,
|
| 54 |
+
"beta_schedule": "squaredcos_cap_v2",
|
| 55 |
+
"beta_start": 0.0001,
|
| 56 |
+
"beta_end": 0.02,
|
| 57 |
+
"prediction_type": "epsilon",
|
| 58 |
+
"clip_sample": true,
|
| 59 |
+
"clip_sample_range": 1.0,
|
| 60 |
+
"diffusion_refine_steps": 3,
|
| 61 |
+
"spatial_softmax_num_keypoints": 32,
|
| 62 |
+
"diffusion_down_dims": [
|
| 63 |
+
256,
|
| 64 |
+
512
|
| 65 |
+
],
|
| 66 |
+
"diffusion_kernel_size": 3,
|
| 67 |
+
"diffusion_n_groups": 8,
|
| 68 |
+
"diffusion_step_embed_dim": 64,
|
| 69 |
+
"diffusion_use_film_scale_modulation": true,
|
| 70 |
+
"hybrid_mode": "sequential",
|
| 71 |
+
"hybrid_weight": 0.5,
|
| 72 |
+
"keyframe_indices": [
|
| 73 |
+
0,
|
| 74 |
+
49,
|
| 75 |
+
99
|
| 76 |
+
],
|
| 77 |
+
"diffusion_loss_weight": 1.0,
|
| 78 |
+
"dropout": 0.1,
|
| 79 |
+
"kl_weight": 1.0,
|
| 80 |
+
"smoothness_weight": 0.0,
|
| 81 |
+
"smoothness_type": "velocity",
|
| 82 |
+
"optimizer_lr": 1e-05,
|
| 83 |
+
"optimizer_weight_decay": 0.0001,
|
| 84 |
+
"optimizer_lr_backbone": 1e-05,
|
| 85 |
+
"optimizer_lr_diffusion": 0.0001,
|
| 86 |
+
"scheduler_name": "cosine",
|
| 87 |
+
"scheduler_num_warmup_steps": 500,
|
| 88 |
+
"scheduler_num_training_steps": 30000
|
| 89 |
+
}
|