transformer/config.json

{
  "_class_name": "MaskGiTUViT",
  "_version": "0.0.1",
  "add_cross_attention": true,
  "attention_dropout": 0.0,
  "block_out_channels": [
    512,
    1024
  ],
  "codebook_size": 8192,
  "encoder_hidden_size": 768,
  "hidden_dropout": 0.0,
  "hidden_size": 1024,
  "in_channels": 384,
  "initializer_range": 0.02,
  "intermediate_size": 4096,
  "layer_norm_eps": 0.000001,
  "mask_token_id": 8255,
  "max_position_embeddings": 256,
  "norm_type": "rmsnorm",
  "num_attention_heads": 16,
  "num_classes": null,
  "num_hidden_layers": 22,
  "num_res_blocks": 3,
  "num_vq_tokens": 1024,
  "patch_size": 1,
  "project_encoder_hidden_states": false,
  "use_bias": false,
  "use_codebook_size_for_output": true,
  "use_encoder_layernorm": true,
  "use_normformer": false,
  "use_position_embeddings": false,
  "vocab_size": 8256
}