diff --git a/python/sglang/srt/model_executor/model_runner.py b/python/sglang/srt/model_executor/model_runner.py
index 4ae426b3c..02330c0e9 100644
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -922,14 +922,6 @@ class ModelRunner(ModelRunnerKVCacheMixin):
                 server_args=self.server_args,
                 model_config=self.model_config,
             )
-            if self.uses_cp_shared_kv:
-                from sglang.srt.mem_cache.cp_shared_kv_layout import CpSharedKVLayout
-
-                self.cp_shared_kv_layout = CpSharedKVLayout(
-                    page_size=self.page_size,
-                    cp_size=self.server_args.attn_cp_size,
-                    cp_rank=get_attention_cp_rank(),
-                )
             if is_npu():
                 register_sgl_tp_rank(self.gpu_id)
 
@@ -952,6 +944,15 @@ class ModelRunner(ModelRunnerKVCacheMixin):
                     f"(tp_size={self.tp_size}, pp_size={self.pp_size}, ep_size={self.moe_ep_size})"
                 )
 
+        if self.uses_cp_shared_kv:
+            from sglang.srt.mem_cache.cp_shared_kv_layout import CpSharedKVLayout
+
+            self.cp_shared_kv_layout = CpSharedKVLayout(
+                page_size=self.page_size,
+                cp_size=self.server_args.attn_cp_size,
+                cp_rank=get_attention_cp_rank(),
+            )
+
         pre_model_load_memory = get_available_gpu_memory(
             self.device,
             self.gpu_id,