diff --git a/vllm/config/model.py b/vllm/config/model.py
index 7f6313005737b..84e59dfc83d2e 100644
--- a/vllm/config/model.py
+++ b/vllm/config/model.py
@@ -508,12 +508,7 @@ class ModelConfig:
         self.hf_image_processor_config = get_hf_image_processor_config(
             self.model, hf_token=self.hf_token, revision=self.revision
         )
-        self.model_arch_config = None
-        convertor_cls = MODEL_ARCH_CONFIG_CONVERTORS.get(
-            hf_config.model_type, ModelArchConfigConvertorBase
-        )
-        convertor = convertor_cls(hf_config)
-        self.model_arch_config = convertor.convert(self.model, self.revision)
+        self.model_arch_config = self.get_model_arch_config()
 
         architectures = self.architectures
         registry = self.registry
@@ -717,6 +712,13 @@ class ModelConfig:
         self._verify_cuda_graph()
         self._verify_bnb_config()
 
+    def get_model_arch_config(self) -> ModelArchitectureConfig:
+        convertor_cls = MODEL_ARCH_CONFIG_CONVERTORS.get(
+            self.hf_config.model_type, ModelArchConfigConvertorBase
+        )
+        convertor = convertor_cls(self.hf_config)
+        return convertor.convert(self.model, self.revision)
+
     @field_validator("tokenizer_mode", mode="after")
     def _lowercase_tokenizer_mode(cls, tokenizer_mode: str) -> str:
         return tokenizer_mode.lower()
diff --git a/vllm/config/vllm.py b/vllm/config/vllm.py
index 614a3226cb711..f03d9d768c740 100644
--- a/vllm/config/vllm.py
+++ b/vllm/config/vllm.py
@@ -421,6 +421,7 @@ class VllmConfig:
 
         model_config = copy.deepcopy(self.model_config)
         model_config.hf_config = hf_config
+        model_config.model_arch_config = model_config.get_model_arch_config()
 
         return replace(self, model_config=model_config)