{
  "_name_or_path": "dandelin/vilt-b32-mlm",
  "architectures": [
    "ViltForQuestionAnswering"
  ],
  "attention_probs_dropout_prob": 0.0,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 768,
  "id2label": {
    "0": "window",
    "1": "park",
    "2": "giraffes",
    "3": "in car",
    "4": "human",
    "5": "cup",
    "6": "church",
    "7": "they aren't",
    "8": "wine tasting",
    "9": "snowboarding",
    "10": "2",
    "11": "woods",
    "12": "wine",
    "13": "sun",
    "14": "sidewalk",
    "15": "white and blue",
    "16": "white and black",
    "17": "blue",
    "18": "snow",
    "19": "5",
    "20": "purple",
    "21": "air",
    "22": "birthday",
    "23": "lanyard",
    "24": "ice cream",
    "25": "tower",
    "26": "4",
    "27": "desert",
    "28": "cross",
    "29": "camera",
    "30": "security",
    "31": "jeep",
    "32": "bike rack",
    "33": "talking",
    "34": "woman",
    "35": "wedding",
    "36": "forest",
    "37": "bus",
    "38": "smiling",
    "39": "hat",
    "40": "photographer",
    "41": "green",
    "42": "bedroom",
    "43": "full",
    "44": "crossing",
    "45": "skateboarding",
    "46": "soccer",
    "47": "blue and white",
    "48": "hair",
    "49": "hawaii",
    "50": "curtain",
    "51": "calico",
    "52": "big ben",
    "53": "laying down",
    "54": "giraffe",
    "55": "exit",
    "56": "clock",
    "57": "1",
    "58": "skier",
    "59": "small",
    "60": "can't tell",
    "61": "brick",
    "62": "stand",
    "63": "queen",
    "64": "7",
    "65": "resting",
    "66": "ball",
    "67": "many",
    "68": "picnic table",
    "69": "unknown",
    "70": "name tag",
    "71": "red and blue",
    "72": "stripes",
    "73": "zoo",
    "74": "clock tower",
    "75": "table",
    "76": "women",
    "77": "not sure",
    "78": "monitor",
    "79": "girl",
    "80": "black and white",
    "81": "double",
    "82": "ground",
    "83": "sky",
    "84": "on street",
    "85": "door",
    "86": "tired",
    "87": "natural",
    "88": "0",
    "89": "bricks",
    "90": "windows",
    "91": "africa",
    "92": "train",
    "93": "3",
    "94": "crown",
    "95": "red and yellow",
    "96": "8",
    "97": "backpack",
    "98": "french",
    "99": "happy",
    "100": "dirt",
    "101": "lg",
    "102": "tan",
    "103": "screen",
    "104": "fashion",
    "105": "necklace",
    "106": "cloudy",
    "107": "lady",
    "108": "tv",
    "109": "outside",
    "110": "10",
    "111": "station",
    "112": "talking on phone",
    "113": "yes",
    "114": "2010",
    "115": "don't know",
    "116": "dog",
    "117": "shadow",
    "118": "canopy",
    "119": "net",
    "120": "wall",
    "121": "white",
    "122": "nothing",
    "123": "cage",
    "124": "beagle",
    "125": "shelter",
    "126": "on road",
    "127": "snowboarder",
    "128": "7:45",
    "129": "little girl",
    "130": "soccer ball",
    "131": "8:35",
    "132": "street",
    "133": "tent",
    "134": "car",
    "135": "smile",
    "136": "gray",
    "137": "beige",
    "138": "suv",
    "139": "pink",
    "140": "tabby",
    "141": "2000",
    "142": "2013",
    "143": "protection",
    "144": "blonde",
    "145": "plate",
    "146": "cat",
    "147": "chopsticks",
    "148": "skiing",
    "149": "large",
    "150": "bicycles",
    "151": "leather",
    "152": "black",
    "153": "right",
    "154": "clear",
    "155": "6",
    "156": "arrow",
    "157": "at table",
    "158": "boy",
    "159": "plastic",
    "160": "7:35",
    "161": "no",
    "162": "skateboard",
    "163": "man",
    "164": "9:35",
    "165": "trees",
    "166": "platform",
    "167": "shade",
    "168": "yellow",
    "169": "chair",
    "170": "shadows",
    "171": "donut",
    "172": "neon",
    "173": "down",
    "174": "watching",
    "175": "shrimp",
    "176": "plain",
    "177": "king",
    "178": "walking",
    "179": "not there",
    "180": "snowboard",
    "181": "brown",
    "182": "out",
    "183": "person",
    "184": "bikes",
    "185": "style",
    "186": "curtains",
    "187": "doughnut",
    "188": "orange",
    "189": "red",
    "190": "low",
    "191": "roof",
    "192": "lying down",
    "193": "solid",
    "194": "rack",
    "195": "fence",
    "196": "sleeping",
    "197": "bicycle",
    "198": "gray and black"
  },
  "image_size": 384,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "label2id": {
    "0": 88,
    "1": 57,
    "10": 110,
    "2": 10,
    "2000": 141,
    "2010": 114,
    "2013": 142,
    "3": 93,
    "4": 26,
    "5": 19,
    "6": 155,
    "7": 64,
    "7:35": 160,
    "7:45": 128,
    "8": 96,
    "8:35": 131,
    "9:35": 164,
    "africa": 91,
    "air": 21,
    "arrow": 156,
    "at table": 157,
    "backpack": 97,
    "ball": 66,
    "beagle": 124,
    "bedroom": 42,
    "beige": 137,
    "bicycle": 197,
    "bicycles": 150,
    "big ben": 52,
    "bike rack": 32,
    "bikes": 184,
    "birthday": 22,
    "black": 152,
    "black and white": 80,
    "blonde": 144,
    "blue": 17,
    "blue and white": 47,
    "boy": 158,
    "brick": 61,
    "bricks": 89,
    "brown": 181,
    "bus": 37,
    "cage": 123,
    "calico": 51,
    "camera": 29,
    "can't tell": 60,
    "canopy": 118,
    "car": 134,
    "cat": 146,
    "chair": 169,
    "chopsticks": 147,
    "church": 6,
    "clear": 154,
    "clock": 56,
    "clock tower": 74,
    "cloudy": 106,
    "cross": 28,
    "crossing": 44,
    "crown": 94,
    "cup": 5,
    "curtain": 50,
    "curtains": 186,
    "desert": 27,
    "dirt": 100,
    "dog": 116,
    "don't know": 115,
    "donut": 171,
    "door": 85,
    "double": 81,
    "doughnut": 187,
    "down": 173,
    "exit": 55,
    "fashion": 104,
    "fence": 195,
    "forest": 36,
    "french": 98,
    "full": 43,
    "giraffe": 54,
    "giraffes": 2,
    "girl": 79,
    "gray": 136,
    "gray and black": 198,
    "green": 41,
    "ground": 82,
    "hair": 48,
    "happy": 99,
    "hat": 39,
    "hawaii": 49,
    "human": 4,
    "ice cream": 24,
    "in car": 3,
    "jeep": 31,
    "king": 177,
    "lady": 107,
    "lanyard": 23,
    "large": 149,
    "laying down": 53,
    "leather": 151,
    "lg": 101,
    "little girl": 129,
    "low": 190,
    "lying down": 192,
    "man": 163,
    "many": 67,
    "monitor": 78,
    "name tag": 70,
    "natural": 87,
    "necklace": 105,
    "neon": 172,
    "net": 119,
    "no": 161,
    "not sure": 77,
    "not there": 179,
    "nothing": 122,
    "on road": 126,
    "on street": 84,
    "orange": 188,
    "out": 182,
    "outside": 109,
    "park": 1,
    "person": 183,
    "photographer": 40,
    "picnic table": 68,
    "pink": 139,
    "plain": 176,
    "plastic": 159,
    "plate": 145,
    "platform": 166,
    "protection": 143,
    "purple": 20,
    "queen": 63,
    "rack": 194,
    "red": 189,
    "red and blue": 71,
    "red and yellow": 95,
    "resting": 65,
    "right": 153,
    "roof": 191,
    "screen": 103,
    "security": 30,
    "shade": 167,
    "shadow": 117,
    "shadows": 170,
    "shelter": 125,
    "shrimp": 175,
    "sidewalk": 14,
    "skateboard": 162,
    "skateboarding": 45,
    "skier": 58,
    "skiing": 148,
    "sky": 83,
    "sleeping": 196,
    "small": 59,
    "smile": 135,
    "smiling": 38,
    "snow": 18,
    "snowboard": 180,
    "snowboarder": 127,
    "snowboarding": 9,
    "soccer": 46,
    "soccer ball": 130,
    "solid": 193,
    "stand": 62,
    "station": 111,
    "street": 132,
    "stripes": 72,
    "style": 185,
    "sun": 13,
    "suv": 138,
    "tabby": 140,
    "table": 75,
    "talking": 33,
    "talking on phone": 112,
    "tan": 102,
    "tent": 133,
    "they aren't": 7,
    "tired": 86,
    "tower": 25,
    "train": 92,
    "trees": 165,
    "tv": 108,
    "unknown": 69,
    "walking": 178,
    "wall": 120,
    "watching": 174,
    "wedding": 35,
    "white": 121,
    "white and black": 16,
    "white and blue": 15,
    "window": 0,
    "windows": 90,
    "wine": 12,
    "wine tasting": 8,
    "woman": 34,
    "women": 76,
    "woods": 11,
    "yellow": 168,
    "yes": 113,
    "zoo": 73
  },
  "layer_norm_eps": 1e-12,
  "max_image_length": -1,
  "max_position_embeddings": 40,
  "modality_type_vocab_size": 2,
  "model_type": "vilt",
  "num_attention_heads": 12,
  "num_channels": 3,
  "num_hidden_layers": 12,
  "num_images": -1,
  "patch_size": 32,
  "qkv_bias": true,
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "transformers_version": "4.39.3",
  "type_vocab_size": 2,
  "vocab_size": 30522
}