{
  "_moe_implementation": "fused",
  "architectures": [
    "MegrezMoeForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoConfig": "configuration_megrez_moe.MegrezMoeConfig",
    "AutoModel": "modeling_megrez_moe.MegrezMoeModel",
    "AutoModelForCausalLM": "modeling_megrez_moe.MegrezMoeForCausalLM"
  },
  "aux_loss_alpha": 0.001,
  "bos_token_id": null,
  "dtype": "bfloat16",
  "eos_token_id": 120005,
  "ep_size": 1,
  "experts_shared_frequency": 3,
  "first_k_dense_replace": 1,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 10944,
  "max_position_embeddings": 163840,
  "model_type": "megrez_moe",
  "moe_intermediate_size": 1408,
  "moe_layer_freq": 1,
  "n_group": 1,
  "n_routed_experts": 64,
  "n_shared_experts": 4,
  "norm_topk_prob": true,
  "num_attention_heads": 16,
  "num_experts_per_tok": 6,
  "num_hidden_layers": 31,
  "num_key_value_heads": 4,
  "pad_token_id": 120002,
  "pre_gate": true,
  "pretraining_tp": 1,
  "quantize": true,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 5000000,
  "routed_scaling_factor": 1.0,
  "scoring_func": "sigmoid",
  "seq_aux": true,
  "tie_word_embeddings": false,
  "topk_group": 1,
  "topk_method": "noaux_tc",
  "transformers_version": "4.56.0",
  "use_cache": true,
  "vocab_size": 122880
}