config.json

{
  "architectures": [
    "VJEPA2ForVideoClassification"
  ],
  "attention_dropout": 0.0,
  "attention_probs_dropout_prob": 0.0,
  "crop_size": 256,
  "drop_path_rate": 0.0,
  "frames_per_clip": 16,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 1024,
  "id2label": {
    "0": "ApplyLipstick",
    "1": "BaseballPitch",
    "2": "BasketballDunk",
    "3": "BenchPress",
    "4": "BalanceBeam",
    "5": "BabyCrawling",
    "6": "Archery",
    "7": "BandMarching",
    "8": "ApplyEyeMakeup",
    "9": "Basketball"
  },
  "image_size": 256,
  "in_chans": 3,
  "initializer_range": 0.02,
  "label2id": {
    "ApplyEyeMakeup": 8,
    "ApplyLipstick": 0,
    "Archery": 6,
    "BabyCrawling": 5,
    "BalanceBeam": 4,
    "BandMarching": 7,
    "BaseballPitch": 1,
    "Basketball": 9,
    "BasketballDunk": 2,
    "BenchPress": 3
  },
  "layer_norm_eps": 1e-06,
  "mlp_ratio": 4,
  "model_type": "vjepa2",
  "num_attention_heads": 16,
  "num_hidden_layers": 24,
  "num_pooler_layers": 3,
  "patch_size": 16,
  "pred_hidden_size": 384,
  "pred_mlp_ratio": 4.0,
  "pred_num_attention_heads": 12,
  "pred_num_hidden_layers": 12,
  "pred_num_mask_tokens": 10,
  "pred_zero_init_mask_tokens": true,
  "problem_type": "single_label_classification",
  "qkv_bias": true,
  "torch_dtype": "float32",
  "transformers_version": "4.53.0.dev0",
  "tubelet_size": 2,
  "use_SiLU": false,
  "wide_SiLU": true
}