framework:
  name: QwenGR00T_with_Language
  qwenvl:
    base_vlm: Qwen/Qwen2.5-VL-3B-Instruct
    attn_implementation: flash_attention_2
    vl_hidden_dim: 2048
  dino:
    dino_backbone: dinov2_vits14
  action_model:
    action_model_type: DiT-B
    action_hidden_dim: 2560
    hidden_size: 1024
    add_pos_embed: true
    max_seq_len: 1024
    action_dim: 11
    state_dim: 11
    future_action_window_size: 4
    action_horizon: 5
    past_action_window_size: 0
    num_inference_timesteps: 4
    num_target_vision_tokens: 32
    noise_beta_alpha: 1.5
    noise_beta_beta: 1.0
    noise_s: 0.999
    num_timestep_buckets: 1000
    diffusion_model_cfg:
      cross_attention_dim: 2048
      dropout: 0.2
      final_dropout: true
      interleave_self_attention: true
      norm_type: ada_norm
      num_layers: 16
      output_dim: 1024
      positional_embeddings: null

datasets:
  vla_data:
    history_window_size: 7
    action_chunk_size: 5
  vla_data2:
    default_image_resolution:
    - 3
    - 224
    - 224
    image_size:
    - 224
    - 224
    action_chunk_size: 5