config.json

{
  "act_adaptor": "mlp3x_silu",
  "act_pos_emb_config": [
    [
      "state",
      1
    ],
    [
      "action",
      16
    ]
  ],
  "action_dim": 76,
  "ema": {
    "inv_gamma": 1.0,
    "max_value": 0.9999,
    "min_value": 0.0,
    "power": 0.75,
    "update_after_step": 0
  },
  "hrdt": {
    "depth": 16,
    "ffn_dim_multiplier": null,
    "hidden_size": 2176,
    "multiple_of": 256,
    "norm_eps": 1e-05,
    "num_heads": 16,
    "num_kv_heads": 8,
    "output_size": 76,
    "use_flash_attn": true
  },
  "img_adapter": "mlp2x_silu",
  "img_pos_emb_config": [
    [
      "image",
      [
        1,
        1,
        -729
      ]
    ]
  ],
  "lang_adapter": "mlp2x_silu",
  "lang_pos_emb_config": [
    [
      "language",
      -1024
    ]
  ],
  "loss_mode": null,
  "max_img_len": 729,
  "max_lang_len": 1024,
  "mode": "pretrain",
  "noise_scheduler": {
    "num_inference_timesteps": 5,
    "sampler_type": "uniform",
    "timestep_max": 0.999
  },
  "pred_horizon": 16,
  "pretrained_backbone_path": null,
  "st_adaptor": "mlp3x_silu",
  "state_dim": 76,
  "text": {
    "feature_dim": 4096
  },
  "time_noise": {
    "a": 5,
    "beta_m": 100
  },
  "training_mode": "lang",
  "video_adapter": "mlp2x_silu",
  "vision": {
    "feature_dim": 2176
  }
}