datasets:
  vla_data:
    data_mix: robochallenge_table30v2_shred_paper
    data_root_dir: /home/Travor/starvla-assets/Datasets/RoboChallenge_table30v2
    dataset_py: lerobot_datasets
    delete_pause_frame: false
    num_workers: 2
    per_device_batch_size: 1
    persistent_workers: true
    pin_memory: true
    prefetch_factor: 4
    sequential_step_sampling: false
    video_backend: decord
framework:
  action_model:
    action_dim: 8
    action_hidden_dim: 1024
    action_horizon: 8
    action_model_type: MLP
  name: QwenOFT
  qwenvl:
    attn_implementation: sdpa
    base_vlm: /home/Travor/starvla-assets/Pretrained_models/Qwen3.5-0.8B
output_dir: /tmp/starvla-full-validation-results/dataloader-throughput-fulltrain-100step-workers
run_id: dataloader-throughput-fulltrain-100step-workers
run_root_dir: /tmp/starvla-full-validation-results
seed: 42
trainer:
  eval_interval: 100000
  freeze_modules: qwen_vl_interface.model
  gradient_clipping: 1.0
  learning_rate:
    action_model: 0.0001
    base: 2.5e-05
    qwen_vl_interface: 1.0e-05
  logging_frequency: 5
  lr_scheduler_type: cosine_with_min_lr
  max_train_steps: 100
  num_warmup_steps: 10
  optimizer:
    betas:
    - 0.9
    - 0.95
    eps: 1.0e-08
    weight_decay: 1.0e-08
  save_interval: 100
  scheduler_specific_kwargs:
    min_lr: 1.0e-06
wandb_entity: your_wandb_entity
wandb_project: starVLA_local_full_validation