{
  "depth": 32,
  "embed_dim": 1280,
  "mlp_ratio": 4,
  "num_heads": 16,
  "in_chans": 3,
  "hidden_size": 1536,
  "patch_size": 14,
  "spatial_merge_size": 2,
  "spatial_patch_size": 14,
  "temporal_patch_size": 2
}