{
  "d_model": 96,
  "n_layers": 4,
  "n_heads": 4,
  "ff": 256,
  "prefix_len": 8,
  "seq_vis": 256,
  "caps_len": 32,
  "pos_total": 264,
  "text_vocab": 64,
  "vis_vocab": 32,
  "img": 16,
  "sampling": {
    "temperature": 0.85,
    "top_k": 10
  }
}