config.json

{
  "d_model": 2048,
  "decoder_attention_heads": 32,
  "decoder_ffn_dim": 4096,
  "decoder_layers": 24,
  "decoder_start_token_id": 16384,
  "encoder_attention_heads": 32,
  "encoder_ffn_dim": 4096,
  "encoder_layers": 24,
  "encoder_vocab_size": 50272,
  "image_length": 256,
  "image_vocab_size": 16415,
  "max_text_length": 64
}