{
  "latent_channels": 16,
  "cond_dim": 128,
  "num_blocks": 2,
  "num_heads": 2,
  "head_dim": 32,
  "T_inner": 2,
  "T_outer": 1,
  "ffn_mult": 2.0,
  "dropout": 0.0,
  "use_ift": false,
  "encoder_base_ch": 32,
  "decoder_base_ch": 64,
  "vocab_size": 32000,
  "max_text_length": 77,
  "text_layers": 2,
  "text_heads": 2
}