config.json

{
  "approx_mode": "full",
  "architectures": [
    "MraForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.1,
  "block_per_row": 8,
  "bos_token_id": 0,
  "dim": 768,
  "dropout_prob": 0.1,
  "embedding_dim": 768,
  "eos_token_id": 2,
  "head_dim": 64,
  "hidden_act": "gelu",
  "hidden_dim": 3072,
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initial_prior_diagonal_n_blocks": 1,
  "initial_prior_first_n_blocks": 3,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-05,
  "max_position_embeddings": 4096,
  "mixed_precision": true,
  "model_type": "mra",
  "num_attention_heads": 12,
  "num_head": 12,
  "num_hidden_layers": 12,
  "num_layers": 12,
  "num_sen_type": 1,
  "pad_token_id": 1,
  "position_embedding_type": "absolute",
  "shared_weight": false,
  "torch_dtype": "float32",
  "transformers_version": "4.31.0.dev0",
  "type_vocab_size": 1,
  "vocab_size": 50265
}