transformer/config.json

{
  "image_dim": 4,
  "image_size": 512,
  "image_stride": 8,
  "text_token_dim": 2560,
  "text_token_len": 256,
  "video_base_size": [
    1,
    16,
    16
  ],
  "image_base_size": [
    32,
    32
  ],
  "arch": [
    "vit_d16w1024",
    "vit_d32w1024",
    "mlp_d3w1280"
  ]
}