datasets:
  vla_data:
    action_mode: abs
    data_mix: robodojo_v21_all_h50_q99
    data_root_dir: ./playground/Datasets
    dataset_py: lerobot_datasets
    obs_image_size:
    - 224
    - 224
    per_device_batch_size: 16
    video_backend: torchvision_av
framework:
  action_model:
    action_dim: 14
    action_hidden_dim: 2560
    action_horizon: 50
    action_model_type: MLP
  name: QwenOFT
  qwenvl:
    attn_implementation: flash_attention_2
    base_vlm: Qwen/Qwen3-VL-4B-Instruct
output_dir: ./playground/Checkpoints/robodojo_v21_all_h50_q99_qwenoft
run_id: robodojo_v21_all_h50_q99_qwenoft
run_root_dir: ./playground/Checkpoints
seed: 42
trainer:
  eval_interval: 100
  freeze_modules: ''
  gradient_clipping: 1.0
  is_resume: true
  learning_rate:
    action_model: 0.0001
    base: 1.0e-05
    qwen_vl_interface: 1.0e-05
  logging_frequency: 10
  lr_scheduler_type: cosine_with_min_lr
  max_train_steps: 150000
  num_warmup_steps: 5000
  optimizer:
    betas:
    - 0.9
    - 0.95
    eps: 1.0e-08
    weight_decay: 1.0e-08
  save_interval: 10000
  scheduler_specific_kwargs:
    min_lr: 5.0e-07
wandb_entity: your_wandb_entity
wandb_project: llavavla