{
  "architectures": [
    "HawkForCausalLM"
  ],
  "auto_map": {
    "AutoConfig": "modeling_hawk_param.HawkConfig",
    "AutoModel": "modeling_hawk_param.HawkModel",
    "AutoModelForCausalLM": "modeling_hawk_param.HawkForCausalLM",
    "AutoModelForSequenceClassification": "modeling_hawk_param.HawkForSequenceClassification",
    "AutoModelForTokenClassification": "modeling_hawk_param.HawkForTokenClassification"
  },
  "bos_token_id": 1,
  "boundary_token_ids": [
    47,
    70,
    60,
    81,
    88,
    23436
  ],
  "classifier_dropout": 0.1,
  "conv_kernel": 4,
  "dtype": "float32",
  "eos_token_id": 2,
  "gumbel_temperature": 0.5,
  "k_params": 3,
  "lambda_param_sup": 0.1,
  "logit_cap": 6.0,
  "max_position_embeddings": 1024,
  "mlp_expansion": 3,
  "model_type": "hawk_rglru",
  "n_embd": 704,
  "n_layer": 12,
  "pad_token_id": 3,
  "param_inject": "embed",
  "param_label_smooth": 0.05,
  "param_reg": 0.01,
  "parametric_mode": "switch",
  "rglru_c": 8.0,
  "rmsnorm_eps": 1e-06,
  "rnn_width": 768,
  "slot_hold_eps": 0.02,
  "tie_word_embeddings": true,
  "transformers_version": "5.3.0",
  "use_parallel_scan": true,
  "vocab_size": 39953,
  "wh_decay_gamma": 1.0,
  "wh_slot_channels": 0,
  "wh_token_ids": []
}