config.json

{
  "add_cross_attention": false,
  "architectures": [
    "UdopForConditionalGeneration"
  ],
  "bos_token_id": 0,
  "d_ff": 37,
  "d_kv": 8,
  "d_model": 32,
  "decoder_start_token_id": 0,
  "dense_act_fn": "relu",
  "dropout_rate": 0.1,
  "dtype": "float32",
  "eos_token_id": 1,
  "feed_forward_proj": "relu",
  "image_size": 224,
  "initializer_factor": 0.002,
  "is_decoder": false,
  "is_encoder_decoder": true,
  "is_gated_act": false,
  "layer_norm_epsilon": 1e-06,
  "max_2d_position_embeddings": 1024,
  "model_type": "udop",
  "num_channels": 3,
  "num_decoder_layers": 2,
  "num_heads": 4,
  "num_layers": 2,
  "pad_token_id": 0,
  "patch_size": 16,
  "relative_attention_max_distance": 128,
  "relative_attention_num_buckets": 32,
  "relative_bias_args": [
    {
      "type": "1d"
    },
    {
      "type": "horizontal"
    },
    {
      "type": "vertical"
    }
  ],
  "tie_word_embeddings": true,
  "transformers_version": "5.16.0.dev0",
  "use_cache": true,
  "vocab_size": 33201
}