{
  "axes_dims": [
    32,
    48,
    48
  ],
  "axes_lens": [
    2048,
    512,
    512
  ],
  "cap_feat_dim": 7680,
  "dim": 3584,
  "gist_max_text_len": 2048,
  "gist_min_text_len": 256,
  "gist_step": 64,
  "in_channels": 32,
  "injection_layer_ids": [
    6,
    14,
    22,
    30,
    38
  ],
  "max_num_gist_tokens": 512,
  "min_num_gist_tokens": 256,
  "n_heads": 28,
  "n_layers": 44,
  "n_refiner_layers": 2,
  "norm_eps": 1e-05,
  "patch_size": 2,
  "perceiver_num_layers": 2,
  "rope_theta": 256.0,
  "t_scale": 1000.0
}
