base_config.json

{
  "emb_size": 768,
  "feedforward_size": 3072,
  "hidden_size": 768,
  "hidden_act": "gelu",
  "heads_num": 12,
  "layers_num": 12,
  "dropout": 0.1,
  "data_processor": "dalle",
  "embedding": ["word", "pos", "seg"],
  "encoder": "transformer",
  "mask": "prefix",
  "target": ["lm"],
  "image_height": 256,
  "image_width": 256,
  "patch_size": 16,
  "tokenizer": "multimodal",
  "image_tokenizer": {
    "is_gumbel": false,
    "is_transformer": true,
    "image_vocab_size": 1024,
    "frame_size": 16
  }
}