config.json

{
  "add_tail_layers": false,
  "architectures": [
    "Ernie4_5_ForCausalLM"
  ],
  "attention_probs_dropout_prob": 0.0,
  "bos_token_id": 1,
  "compression_ratio": 1.0,
  "dtype": "bfloat16",
  "enable_mtp_magic_send": false,
  "eos_token_id": 2,
  "fuse_gate_detach_matmul": true,
  "fuse_linear": false,
  "fuse_ln": false,
  "fuse_rms_norm": true,
  "fuse_rope": true,
  "fuse_swiglu": true,
  "global_aux_loss": false,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 1024,
  "ignored_index": -100,
  "initializer_range": 0.018041293779826325,
  "intermediate_size": 3072,
  "max_position_embeddings": 131072,
  "model_type": "ernie4_5",
  "moe_aux_loss_lambda": 1e-05,
  "moe_capacity": [],
  "moe_dense_experts_token_type_id": 3,
  "moe_gate": "topk",
  "moe_gate_act": "softmax",
  "moe_group_experts": false,
  "moe_group_orthogonal_loss": true,
  "moe_intermediate_size": 0,
  "moe_k": 2,
  "moe_layer_end_index": 17,
  "moe_layer_interval": 2,
  "moe_layer_start_index": 0,
  "moe_norm_gate_logits": true,
  "moe_num_experts": null,
  "moe_num_shared_experts": 0,
  "moe_orthogonal_loss_lambda": 0.0,
  "moe_reverse_token_drop": false,
  "moe_use_aux_free": false,
  "moe_use_hard_gate": false,
  "moe_z_loss_lambda": 0.0,
  "multi_token_pred_lambda": 0.3,
  "num_attention_heads": 16,
  "num_hidden_layers": 18,
  "num_key_value_heads": 2,
  "num_nextn_predict_layers": 0,
  "pad_token_id": 0,
  "paddleformers_version": "0.1",
  "rms_norm_eps": 1e-05,
  "rope_theta": 500000,
  "sinkhorn_2gate": true,
  "sinkhorn_temp": 0.03,
  "tensor_parallel_degree": -1,
  "tie_word_embeddings": true,
  "token_balance_loss": false,
  "token_balance_seqlen": false,
  "use_bias": false,
  "use_recompute_mtp": false,
  "use_rmsnorm": true,
  "vocab_size": 103424,
  "weight_share_add_bias": true
}