{
  "_name_or_path": "sonoisa/t5-base-japanese",
  "activation_dropout": 0.1,
  "architectures": [
    "VLT5ModelWrapper"
  ],
  "attention_dropout": 0.1,
  "bos_token_id": 0,
  "classifier": false,
  "d_ff": 3072,
  "d_kv": 64,
  "d_model": 768,
  "decoder_start_token_id": 0,
  "dropout": 0.1,
  "dropout_rate": 0.1,
  "eos_token_id": 1,
  "eos_token_ids": [
    1
  ],
  "feat_dim": 2048,
  "feed_forward_proj": "relu",
  "gradient_checkpointing": false,
  "individual_vis_layer_norm": true,
  "initializer_factor": 1.0,
  "is_encoder_decoder": true,
  "layer_norm_epsilon": 1e-06,
  "losses": "lm,qa,ground_caption,refer,itm",
  "max_length": 512,
  "model_type": "t5",
  "n_images": 2,
  "n_positions": 512,
  "num_beams": 4,
  "num_decoder_layers": 12,
  "num_heads": 12,
  "num_layers": 12,
  "pad_token_id": 0,
  "pos_dim": 4,
  "relative_attention_num_buckets": 32,
  "share_vis_lang_layer_norm": false,
  "transformers_version": "4.7.0",
  "use_cache": true,
  "use_vis_layer_norm": true,
  "use_vis_order_embedding": true,
  "vocab_size": 32200
}