{
  "model_type": "nova_vla",
  "architectures": ["NovaVLA"],
  "model": {
    "image_size": 224,
    "patch_size": 16,
    "vis_width": 384,
    "vis_depth": 12,
    "vis_heads": 6,
    "vocab_size": 8192,
    "max_text_len": 32,
    "text_width": 384,
    "text_depth": 6,
    "text_heads": 6,
    "d_model": 512,
    "heads": 8,
    "mlp_ratio": 4,
    "dropout": 0.0,
    "perceive_depth": 12,
    "think_steps": 3,
    "plan_tokens": 8,
    "action_dim": 7,
    "chunk": 16,
    "flow_depth": 6,
    "flow_heads": 8,
    "flow_sample_steps": 10,
    "proprio_dim": 8,
    "aux_future_offsets": [5, 10],
    "aux_weight": 0.05
  },
  "normalization": {
    "state_mean": [0.3103575683105713, 0.03385267682753605, 0.07539214932109646, 0.0019429612792045642, -0.09182343426452422, 0.11483689370041507, 0.7289980988365603, 0.0],
    "state_std": [0.05759261126114262, 0.08664525178991958, 0.050983596210794846, 0.12618106493472714, 0.1652598155695422, 0.45545984538043605, 0.3420460553904895, 1.0],
    "action_mean": [0.00030262056748640604, 4.503544677976835e-05, 0.0008777064659451161, -0.0008190429105794883, -0.002905744384466832, 0.001450979582147165, 0.607390816217698],
    "action_std": [0.010445865322303785, 0.016080933468225835, 0.015232868987278048, 0.028649411223907066, 0.029377470968555214, 0.11161144330245538, 0.48518597376928313]
  }
}