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": 11008,
    "model_type": "",
    "num_attention_heads": 16,
    "num_key_value_heads": 16,
    "residual_dropout": 0.0,
    "text_hidden_size": 2048,
    "vit_layers": [
      -3,
      -9
    ]
  },
  "architectures": [
    "MolmoActForActionReasoning"
  ],
  "auto_map": {
    "AutoConfig": "configuration_molmoact.MolmoActConfig",
    "AutoModelForImageTextToText": "modeling_molmoact.MolmoActForActionReasoning"
  },
  "image_patch_id": 151938,
  "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": 2048,
    "initializer_range": 0.02,
    "intermediate_size": 11008,
    "layer_norm_eps": 1e-06,
    "max_position_embeddings": 4096,
    "model_type": "molmoact_llm",
    "norm_after": false,
    "num_attention_heads": 16,
    "num_hidden_layers": 36,
    "num_key_value_heads": 2,
    "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": 151936
  },
  "model_type": "molmoact",
  "n_action_bins": 256,
  "norm_stats": {
    "norm_stats": {
      "1023": {
        "action": {
          "max": [
            0.04367176070809364,
            0.022778866812586784,
            0.01708386093378067,
            0.028339775279164314,
            0.04235227778553963,
            0.06355218589305878,
            1.0
          ],
          "mean": [
            0.00020232585666235536,
            -0.0004970032023265958,
            -0.000430602376582101,
            -0.00015123763296287507,
            4.535434709396213e-05,
            -0.00013272084470372647,
            0.44365325570106506
          ],
          "min": [
            -0.023753266781568527,
            -0.027570929378271103,
            -0.01796768791973591,
            -0.04000410810112953,
            -0.0816362202167511,
            -0.040603041648864746,
            0.0
          ],
          "q01": [
            -0.006955103948712349,
            -0.014739521779119968,
            -0.007455943152308464,
            -0.00743744894862175,
            -0.007610292173922062,
            -0.008179591968655586,
            0.0
          ],
          "q99": [
            0.007702148985117674,
            0.010903404094278812,
            0.009033547714352608,
            0.0069551002234220505,
            0.006484407931566238,
            0.00784524530172348,
            1.0
          ],
          "std": [
            0.0027527061756700277,
            0.004528957884758711,
            0.0034716646187007427,
            0.00254280143417418,
            0.002804982475936413,
            0.0028773832600563765,
            0.4968259632587433
          ]
        },
        "num_entries": 114035
      }
    }
  },
  "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
  }
}