File size: 1,495 Bytes
1b90f1c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
datasets:
  vla_data:
    data_mix: robotwin
    data_root_dir: /inspire/ssd/project/embodied-basic-model/zhangjianing-253108140206/DATASET/robotwin_lerobot
    dataset_py: lerobot_datasets
    image_size:
    - 448
    - 448
    per_device_batch_size: 8
    video_backend: torchvision_av
framework:
  action_model:
    action_dim: 14
    action_hidden_dim: 2560
    action_model_type: DiT-B
    future_action_window_size: 15
    past_action_window_size: 0
  name: QwenOFT
  qwenvl:
    base_vlm: /inspire/qb-ilm/project/embodied-basic-model/zhangjianing-253108140206/model/cubev0-200000-Qwen3-VL
output_dir: /inspire/qb-ilm/project/embodied-basic-model/zhangjianing-253108140206/checkpoints/cubev0-robotwin-finetune-oft/cubev0_robotwin_200000_groot
run_id: cubev0_robotwin_200000_groot
run_root_dir: /inspire/qb-ilm/project/embodied-basic-model/zhangjianing-253108140206/checkpoints/cubev0-robotwin-finetune-oft
seed: 42
trainer:
  eval_interval: 1000
  freeze_modules: true
  gradient_accumulation_steps: 1
  gradient_clipping: 1.0
  is_resume: false
  learning_rate:
    action_model: 0.0001
    base: 1.0e-05
    qwen_vl_interface: 1.0e-05
  logging_frequency: 50
  lr_scheduler_type: cosine_with_min_lr
  max_train_steps: 30000
  num_warmup_steps: 100
  optimizer:
    betas:
    - 0.9
    - 0.95
    eps: 1.0e-08
    weight_decay: 1.0e-08
  save_interval: 5000
  scheduler_specific_kwargs:
    min_lr: 5.0e-07
wandb_entity: zaleni-tongji-university
wandb_project: cubev0-robotwin-finetune