diff --git a/vllm/config/model.py b/vllm/config/model.py
index 8e28e34bf..3c89658f0 100644
--- a/vllm/config/model.py
+++ b/vllm/config/model.py
@@ -106,6 +106,10 @@ class ModelConfig:
     """Name or path of the Hugging Face model to use. It is also used as the
     content for `model_name` tag in metrics output when `served_model_name` is
     not specified."""
+    model_weights: str = ""
+    """Original model weights path. Used when the model is pulled from object 
+    storage (e.g., RunAI) to preserve the original URI while `model` points to 
+    the local directory."""
     runner: RunnerOption = "auto"
     """The type of model runner to use. Each vLLM instance only supports one
     model runner, even if the same model can be used for multiple types."""
@@ -705,6 +709,10 @@ class ModelConfig:
             tokenizer: Tokenizer name or path
         """
 
+        # Skip if model_weights is already set (model already pulled)
+        if self.model_weights:
+            return
+
         if not (is_runai_obj_uri(model) or is_runai_obj_uri(tokenizer)):
             return
 
diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index 77201f668..a6a4f780a 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -354,6 +354,7 @@ class EngineArgs:
     """Arguments for vLLM engine."""
 
     model: str = ModelConfig.model
+    model_weights: str = ModelConfig.model_weights
     served_model_name: str | list[str] | None = ModelConfig.served_model_name
     tokenizer: str | None = ModelConfig.tokenizer
     hf_config_path: str | None = ModelConfig.hf_config_path
@@ -1206,6 +1207,7 @@ class EngineArgs:
 
         return ModelConfig(
             model=self.model,
+            model_weights=self.model_weights,
             hf_config_path=self.hf_config_path,
             runner=self.runner,
             convert=self.convert,
@@ -1349,6 +1351,7 @@ class EngineArgs:
 
         model_config = self.create_model_config()
         self.model = model_config.model
+        self.model_weights = model_config.model_weights
         self.tokenizer = model_config.tokenizer
 
         self._check_feature_supported(model_config)
diff --git a/vllm/model_executor/model_loader/runai_streamer_loader.py b/vllm/model_executor/model_loader/runai_streamer_loader.py
index fb33d3c64..9d3ade4cd 100644
--- a/vllm/model_executor/model_loader/runai_streamer_loader.py
+++ b/vllm/model_executor/model_loader/runai_streamer_loader.py
@@ -108,8 +108,8 @@ class RunaiModelStreamerLoader(BaseModelLoader):
     def load_weights(self, model: nn.Module, model_config: ModelConfig) -> None:
         """Load weights into a model."""
         model_weights = model_config.model
-        if hasattr(model_config, "model_weights"):
-            model_weights = model_config.model_weights
+        if model_weights_override := model_config.model_weights:
+            model_weights = model_weights_override
         model.load_weights(
             self._get_weights_iterator(model_weights, model_config.revision)
         )
diff --git a/vllm/model_executor/model_loader/sharded_state_loader.py b/vllm/model_executor/model_loader/sharded_state_loader.py
index 1538f0c2a..e27cedd99 100644
--- a/vllm/model_executor/model_loader/sharded_state_loader.py
+++ b/vllm/model_executor/model_loader/sharded_state_loader.py
@@ -110,8 +110,8 @@ class ShardedStateLoader(BaseModelLoader):
         from vllm.distributed import get_tensor_model_parallel_rank
 
         model_weights = model_config.model
-        if hasattr(model_config, "model_weights"):
-            model_weights = model_config.model_weights
+        if model_weights_override := model_config.model_weights:
+            model_weights = model_weights_override
         local_model_path = model_weights
 
         rank = get_tensor_model_parallel_rank()