diff --git a/configs/checkpoint/model_cfg.json b/configs/checkpoint/model_cfg.json
index 294c25df8bbfd8ce1de4217a31c94e6672a13b02..85f787fa890c2e6ea45b5dcf56994843d1984c2e 100644
--- a/configs/checkpoint/model_cfg.json
+++ b/configs/checkpoint/model_cfg.json
@@ -170,7 +170,7 @@
         "mlp_experts_flag": true,
         "first_k_dense_replace": 1,
         "moe_layer_freq": 1,
-        "qk_layernorm": true,
+        "qk_layernorm": false,
         "add_qkv_bias": true,
         "router_bias": false
       },
@@ -178,7 +178,7 @@
         "num_layers": "num_hidden_layers",
         "norm_epsilon": "rms_norm_eps",
         "rotary_base": "rope_theta",
-        "num_experts": "num_routed_experts",
+        "num_experts": "n_routed_experts",
         "moe_intermediate_size": "moe_intermediate_size",
         "kv_channels": "head_dim",
         "n_shared_experts": "n_shared_experts"
@@ -190,7 +190,7 @@
         "layers_self_attention_linear_q_proj": "model.layers[layer_idx].self_attn.q_proj",
         "layers_self_attention_linear_k_proj": "model.layers[layer_idx].self_attn.k_proj",
         "layers_self_attention_linear_v_proj": "model.layers[layer_idx].self_attn.v_proj",
-        "layers_self_attention_pre_mlp_layernorm": "transformer.encoder.layers[layer_idx].post_attention_layernorm",
+        "layers_self_attention_pre_mlp_layernorm": "model.layers[layer_idx].post_attention_layernorm",
         "layers_mlp_router": "model.layers[layer_idx].mlp.gate",
         "layers_mlp_experts_gate_proj": "model.layers[layer_idx].mlp.experts[expert_idx].gate_proj",
         "layers_mlp_experts_up_proj": "model.layers[layer_idx].mlp.experts[expert_idx].up_proj",
@@ -199,7 +199,7 @@
         "layers_mlp_shared_experts_up_proj": "model.layers[layer_idx].mlp.shared_experts.up_proj",
         "layers_mlp_shared_experts_linear_fc2": "model.layers[layer_idx].mlp.shared_experts.down_proj",
         "mtp_layers_enorm": "model.layers[layer_idx].enorm",
-        "mtp_layers_hnorm": "model.layers[layer_idx]}.hnorm",
+        "mtp_layers_hnorm": "model.layers[layer_idx].hnorm",
         "mtp_layers_eh_proj": "model.layers[layer_idx].eh_proj",
         "layers_mlp_router_bias": "model.layers[layer_idx].mlp.gate.e_score_correction_bias",
         "mtp_layers_shared_head_norm": "model.layers[layer_idx].shared_head.norm",