config.json

{
  "adapter_config": {
    "attention_dropout": 0.0,
    "float32_attention": true,
    "head_dim": 72,
    "hidden_act": "silu",
    "hidden_size": 1152,
    "image_feature_dropout": 0.0,
    "image_padding_embed": null,
    "initializer_range": 0.02,
    "intermediate_size": 18944,
    "model_type": "",
    "num_attention_heads": 16,
    "num_key_value_heads": 16,
    "residual_dropout": 0.0,
    "text_hidden_size": 3584,
    "vit_layers": [
      -3,
      -9
    ]
  },
  "architectures": [
    "MolmoActForActionReasoning"
  ],
  "auto_map": {
    "AutoConfig": "configuration_molmoact.MolmoActConfig",
    "AutoModelForImageTextToText": "modeling_molmoact.MolmoActForActionReasoning"
  },
  "image_patch_id": 152066,
  "initializer_range": 0.02,
  "llm_config": {
    "additional_vocab_size": 128,
    "attention_dropout": 0.0,
    "embedding_dropout": 0.0,
    "head_dim": 128,
    "hidden_act": "silu",
    "hidden_size": 3584,
    "initializer_range": 0.02,
    "intermediate_size": 18944,
    "layer_norm_eps": 1e-06,
    "max_position_embeddings": 4096,
    "model_type": "molmoact_llm",
    "norm_after": false,
    "num_attention_heads": 28,
    "num_hidden_layers": 28,
    "num_key_value_heads": 4,
    "qk_norm_type": "olmo",
    "qkv_bias": true,
    "residual_dropout": 0.0,
    "rope_scaling": null,
    "rope_theta": 1000000.0,
    "use_cache": true,
    "use_qk_norm": false,
    "vocab_size": 152064
  },
  "model_type": "molmoact",
  "n_action_bins": 256,
  "norm_stats": {
    "libero_10_no_noops_modified": {
      "action": {
        "max": [
          0.9375,
          0.9375,
          0.9375,
          0.30000001192092896,
          0.29357144236564636,
          0.375,
          1.0
        ],
        "mean": [
          0.01820324920117855,
          0.05858374014496803,
          -0.05592384561896324,
          0.004626928828656673,
          0.00289608770981431,
          -0.007673131301999092,
          0.5457824468612671
        ],
        "min": [
          -0.9375,
          -0.9375,
          -0.9375,
          -0.23642857372760773,
          -0.3053571283817291,
          -0.3675000071525574,
          0.0
        ],
        "q01": [
          -0.6348214149475098,
          -0.7741071581840515,
          -0.7633928656578064,
          -0.09749999642372131,
          -0.14819999992847435,
          -0.2742857038974762,
          0.0
        ],
        "q99": [
          0.7714285850524902,
          0.8464285731315613,
          0.9375,
          0.13928571343421936,
          0.15964286029338837,
          0.3246428668498993,
          1.0
        ],
        "std": [
          0.2825464606285095,
          0.35904666781425476,
          0.3673802614212036,
          0.03770702704787254,
          0.05429719388484955,
          0.08725254982709885,
          0.49815231561660767
        ]
      },
      "num_trajectories": 379,
      "num_transitions": 101469,
      "proprio": {
        "max": [
          0.21031762659549713,
          0.39128610491752625,
          1.3332009315490723,
          3.6714255809783936,
          3.560650587081909,
          1.386339545249939,
          0.0,
          0.04160946607589722,
          0.0013633022317662835
        ],
        "mean": [
          -0.04190658777952194,
          0.03539430722594261,
          0.8257141709327698,
          2.908308267593384,
          -0.5562185049057007,
          -0.16649018228054047,
          0.0,
          0.028316624462604523,
          -0.028561657294631004
        ],
        "min": [
          -0.4828203022480011,
          -0.3255046010017395,
          0.445506751537323,
          1.1321442127227783,
          -3.641430377960205,
          -1.842738389968872,
          0.0,
          -0.0010040868073701859,
          -0.04111652821302414
        ],
        "q01": [
          -0.3899900782108307,
          -0.2838300323486328,
          0.44795057058334353,
          1.8810229921340942,
          -2.886677579879761,
          -1.1599004411697387,
          0.0,
          0.002066459748893976,
          -0.04001387819647789
        ],
        "q99": [
          0.1530261474847791,
          0.32915401458740223,
          1.2546923208236693,
          3.303542451858519,
          2.7496529006957933,
          0.6893712210655194,
          0.0,
          0.040048558115959164,
          -0.0017598449345678235
        ],
        "std": [
          0.10743364691734314,
          0.14424669742584229,
          0.2572328448295593,
          0.3441362977027893,
          1.234421730041504,
          0.3579835891723633,
          0.0,
          0.013308707624673843,
          0.013174631632864475
        ]
      }
    }
  },
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "transformers_version": "4.52.3",
  "use_cache": true,
  "vit_config": {
    "attention_dropout": 0.0,
    "float32_attention": true,
    "head_dim": 72,
    "hidden_act": "gelu_pytorch_tanh",
    "hidden_size": 1152,
    "image_default_input_size": [
      378,
      378
    ],
    "image_num_pos": 729,
    "image_patch_size": 14,
    "initializer_range": 0.02,
    "intermediate_size": 4304,
    "layer_norm_eps": 1e-06,
    "model_type": "molmoact_vit",
    "num_attention_heads": 16,
    "num_hidden_layers": 27,
    "num_key_value_heads": 16,
    "patch_bias": true,
    "pre_layernorm": false,
    "residual_dropout": 0.0,
    "use_cls_token": false
  }
}