training complete :>
Browse files- config.json +102 -0
- model.safetensors +3 -0
config.json
ADDED
|
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"type": "diffusion",
|
| 3 |
+
"n_obs_steps": 2,
|
| 4 |
+
"normalization_mapping": {
|
| 5 |
+
"STATE": "MIN_MAX",
|
| 6 |
+
"ENV": "MIN_MAX",
|
| 7 |
+
"ACTION": "MIN_MAX",
|
| 8 |
+
"VISUAL": "IDENTITY"
|
| 9 |
+
},
|
| 10 |
+
"input_features": {
|
| 11 |
+
"observation.image": {
|
| 12 |
+
"type": "VISUAL",
|
| 13 |
+
"shape": [
|
| 14 |
+
3,
|
| 15 |
+
96,
|
| 16 |
+
96
|
| 17 |
+
]
|
| 18 |
+
},
|
| 19 |
+
"observation.state": {
|
| 20 |
+
"type": "STATE",
|
| 21 |
+
"shape": [
|
| 22 |
+
2
|
| 23 |
+
]
|
| 24 |
+
},
|
| 25 |
+
"observation.environment_state": {
|
| 26 |
+
"type": "ENV",
|
| 27 |
+
"shape": [
|
| 28 |
+
16
|
| 29 |
+
]
|
| 30 |
+
}
|
| 31 |
+
},
|
| 32 |
+
"output_features": {
|
| 33 |
+
"action": {
|
| 34 |
+
"type": "ACTION",
|
| 35 |
+
"shape": [
|
| 36 |
+
2
|
| 37 |
+
]
|
| 38 |
+
}
|
| 39 |
+
},
|
| 40 |
+
"device": "cuda",
|
| 41 |
+
"use_amp": false,
|
| 42 |
+
"horizon": 16,
|
| 43 |
+
"n_action_steps": 8,
|
| 44 |
+
"drop_n_last_frames": 7,
|
| 45 |
+
"vision_backbone": "resnet18",
|
| 46 |
+
"crop_shape": [
|
| 47 |
+
84,
|
| 48 |
+
84
|
| 49 |
+
],
|
| 50 |
+
"crop_is_random": true,
|
| 51 |
+
"pretrained_backbone_weights": null,
|
| 52 |
+
"use_group_norm": true,
|
| 53 |
+
"spatial_softmax_num_keypoints": 32,
|
| 54 |
+
"use_separate_rgb_encoder_per_camera": false,
|
| 55 |
+
"state_backbone": "MLP",
|
| 56 |
+
"state_encoder_block_channels": [
|
| 57 |
+
32
|
| 58 |
+
],
|
| 59 |
+
"state_encoder_feature_dim": 64,
|
| 60 |
+
"state_encoder_use_layernorm": true,
|
| 61 |
+
"multimodal": true,
|
| 62 |
+
"multimodal_fusion_block_channels": [
|
| 63 |
+
128
|
| 64 |
+
],
|
| 65 |
+
"multimodal_encoder_feature_dim": 256,
|
| 66 |
+
"use_transformer_fusion": true,
|
| 67 |
+
"multimodal_hidden_dim": 256,
|
| 68 |
+
"multimodal_num_layers": 2,
|
| 69 |
+
"multimodal_num_heads": 4,
|
| 70 |
+
"multimodal_dropout": 0.1,
|
| 71 |
+
"down_dims": [
|
| 72 |
+
512,
|
| 73 |
+
1024,
|
| 74 |
+
2048
|
| 75 |
+
],
|
| 76 |
+
"kernel_size": 5,
|
| 77 |
+
"n_groups": 8,
|
| 78 |
+
"diffusion_step_embed_dim": 128,
|
| 79 |
+
"use_film_scale_modulation": true,
|
| 80 |
+
"noise_scheduler_type": "DDPM",
|
| 81 |
+
"num_train_timesteps": 100,
|
| 82 |
+
"beta_schedule": "squaredcos_cap_v2",
|
| 83 |
+
"beta_start": 0.0001,
|
| 84 |
+
"beta_end": 0.02,
|
| 85 |
+
"prediction_type": "epsilon",
|
| 86 |
+
"clip_sample": true,
|
| 87 |
+
"clip_sample_range": 1.0,
|
| 88 |
+
"variance_type": "fixed_small",
|
| 89 |
+
"num_inference_steps": null,
|
| 90 |
+
"keypoint_drop_prob": 0.37,
|
| 91 |
+
"do_mask_loss_for_padding": false,
|
| 92 |
+
"optimizer_type": "adamw",
|
| 93 |
+
"optimizer_lr": 0.0001,
|
| 94 |
+
"optimizer_betas": [
|
| 95 |
+
0.95,
|
| 96 |
+
0.999
|
| 97 |
+
],
|
| 98 |
+
"optimizer_eps": 1e-08,
|
| 99 |
+
"optimizer_weight_decay": 1e-06,
|
| 100 |
+
"scheduler_name": "cosine",
|
| 101 |
+
"scheduler_warmup_steps": 500
|
| 102 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9057db4fde63753fafa18632a3e70fcbe4649b3681790404e9d6b1e0842cd65a
|
| 3 |
+
size 1101178264
|