File size: 2,156 Bytes
1e7a7fa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

=== scheduled policy 1/2 ===
slot_key=fixed:0
model_id=wandb://self-improving/real-dagger-mining-01b/real01b-routing-d1-baseline-uniform-r0-dp-r18-sepenc-imagen-dbc10b90-92-init0-seed-1-umirel-schemep-proprioanchor-knf2zjdr-final:v0
wandb: [wandb.Api()] Loaded credentials for https://api.wandb.ai from /home/lars/.netrc.
wandb: Downloading large artifact 'real01b-routing-d1-baseline-uniform-r0-dp-r18-sepenc-imagen-dbc10b90-92-init0-seed-1-umirel-schemep-proprioanchor-knf2zjdr-final:v0', 362.56MB. 6 files...
wandb: \ 1 of 6 files downloaded...
wandb: | 4 of 6 files downloaded...
wandb:   6 of 6 files downloaded.  
Done. 00:00:37.5 (9.7MB/s)
Loading weights from local directory
/home/lars/self-improving-project/self-improving-robots/robot-env/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py:1061: UserWarning: Can't initialize NVML
  raw_cnt = _raw_device_count_nvml()
  Diffusion action horizon: predict 12 (horizon), execute 6 (n_action_steps).
  Scheduler: DDIMScheduler (8 inference steps)
  Image resolution: 224x224 (auto-detected from checkpoint)
  Per-camera replacement crop: {'wrist_left': (180, 0, 639, 413), 'side_1': (140, 120, 560, 470), 'side_2': (130, 120, 440, 445)} (from policy config)

=== scheduled policy 2/2 ===
slot_key=fixed:1
model_id=wandb://self-improving/real-dagger-mining-01b/real01b-routing-d1-ours-sobol-r0-dp-r18-sepenc-imagenet-pre-56c61020-92-init0-seed-1-umirel-schemep-proprioanchor-1wbp1jd8-final:v0
wandb: Downloading large artifact 'real01b-routing-d1-ours-sobol-r0-dp-r18-sepenc-imagenet-pre-56c61020-92-init0-seed-1-umirel-schemep-proprioanchor-1wbp1jd8-final:v0', 362.56MB. 6 files...
wandb: \ 1 of 6 files downloaded...
wandb: | 4 of 6 files downloaded...
wandb:   6 of 6 files downloaded.  
Done. 00:00:33.9 (10.7MB/s)
Loading weights from local directory
  Diffusion action horizon: predict 12 (horizon), execute 6 (n_action_steps).
  Scheduler: DDIMScheduler (8 inference steps)
  Image resolution: 224x224 (auto-detected from checkpoint)
  Per-camera replacement crop: {'wrist_left': (180, 0, 639, 413), 'side_1': (140, 120, 560, 470), 'side_2': (130, 120, 440, 445)} (from policy config)