config.json

{
  "architectures": [
    "LxmertForQuestionAnswering"
  ],
  "attention_probs_dropout_prob": 0.1,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "l_layers": 9,
  "layer_norm_eps": 1e-12,
  "max_position_embeddings": 512,
  "model_type": "lxmert",
  "num_attention_heads": 12,
  "num_attr_labels": 400,
  "num_hidden_layers": {
    "cross_encoder": 5,
    "language": 9,
    "vision": 5
  },
  "num_object_labels": 1600,
  "num_qa_labels": 1842,
  "r_layers": 5,
  "task_mask_lm": true,
  "task_matched": true,
  "task_obj_predict": true,
  "task_qa": true,
  "type_vocab_size": 2,
  "visual_attr_loss": true,
  "visual_feat_dim": 2048,
  "visual_feat_loss": true,
  "visual_loss_normalizer": 6.67,
  "visual_obj_loss": true,
  "visual_pos_dim": 4,
  "vocab_size": 30522,
  "x_layers": 5
}