config.json

{
  "architectures": [
    "SubSegDeBERTaForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.1,
  "bos_token_id": 1,
  "char_decoder_hidden_size": 256,
  "char_decoder_num_layers": 1,
  "char_embed_dim": 128,
  "cls_token_id": 1,
  "dropout": 0.1,
  "dtype": "float32",
  "eos_token_id": 2,
  "eow_token_id": 5,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-07,
  "lex_vocab_size": 10001,
  "loss_normalization": "word",
  "mask_token_id": 3,
  "max_position_embeddings": 1024,
  "max_relative_positions": -1,
  "max_seg_len": 5,
  "mlm_probability": 0.4,
  "model_type": "subsegdeberta",
  "n_alpha": 263,
  "norm_rel_ebd": "layer_norm",
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 0,
  "pos_att_type": "p2c|c2p",
  "position_buckets": 256,
  "relative_attention": true,
  "seg_end_token_id": 6,
  "sep_token_id": 2,
  "transformers_version": "4.57.1",
  "unk_token_id": 4,
  "vocab_size": 418,
  "word_context_hidden_size": 768,
  "word_context_num_layers": 1
}