config.json

{
  "architectures": [
    "VJEPA21Model"
  ],
  "attention_probs_dropout_prob": 0.0,
  "auto_map": {
    "AutoConfig": "configuration_vjepa21.VJEPA21Config",
    "AutoModel": "modeling_vjepa21.VJEPA21Model",
    "AutoModelForVideoClassification": "modeling_vjepa21.VJEPA21ForVideoClassification",
    "AutoVideoProcessor": "video_processing_vjepa21.VJEPA21VideoProcessor"
  },
  "crop_size": 384,
  "drop_path_rate": 0.0,
  "dtype": "float32",
  "frames_per_clip": 64,
  "has_cls_first": false,
  "hidden_act": "gelu",
  "hidden_size": 768,
  "img_temporal_dim_size": 1,
  "in_chans": 3,
  "initializer_range": 0.02,
  "interpolate_rope": true,
  "layer_norm_eps": 1e-06,
  "mlp_ratio": 4.0,
  "modality_embedding": true,
  "model_type": "vjepa21",
  "n_output_distillation": 1,
  "n_registers": 0,
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "num_pooler_heads": 16,
  "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": 8,
  "pred_return_all_tokens": true,
  "pred_teacher_embed_dim": 1664,
  "pred_zero_init_mask_tokens": true,
  "qkv_bias": true,
  "transformers_version": "5.14.1",
  "tubelet_size": 2,
  "wide_silu": true
}