config.json

{
  "action_chunk_size": 5,
  "bet_softmax_temperature": 0.1,
  "crop_is_random": true,
  "crop_shape": [
    84,
    84
  ],
  "dropout": 0.1,
  "gpt_block_size": 500,
  "gpt_hidden_dim": 512,
  "gpt_input_dim": 512,
  "gpt_n_head": 8,
  "gpt_n_layer": 8,
  "gpt_output_dim": 512,
  "input_normalization_modes": {
    "observation.image.cam_left": "mean_std",
    "observation.image.cam_right": "mean_std",
    "observation.image.cam_wrist": "mean_std",
    "observation.state": "mean_std"
  },
  "input_shapes": {
    "observation.image.cam_left": [
      3,
      96,
      96
    ],
    "observation.image.cam_right": [
      3,
      96,
      96
    ],
    "observation.image.cam_wrist": [
      3,
      96,
      96
    ],
    "observation.state": [
      7
    ]
  },
  "mlp_hidden_dim": 1024,
  "n_action_pred_token": 7,
  "n_obs_steps": 5,
  "n_vqvae_training_steps": 20000,
  "offset_loss_weight": 10000.0,
  "output_normalization_modes": {
    "action": "mean_std"
  },
  "output_shapes": {
    "action": [
      7
    ]
  },
  "pretrained_backbone_weights": null,
  "primary_code_loss_weight": 5.0,
  "secondary_code_loss_weight": 0.5,
  "sequentially_select": false,
  "spatial_softmax_num_keypoints": 32,
  "use_group_norm": true,
  "vision_backbone": "resnet18",
  "vqvae_embedding_dim": 256,
  "vqvae_enc_hidden_dim": 128,
  "vqvae_n_embed": 16
}