diff --git a/vllm/model_executor/offloader/base.py b/vllm/model_executor/offloader/base.py
index 2780ecdd08..3206df34e8 100644
--- a/vllm/model_executor/offloader/base.py
+++ b/vllm/model_executor/offloader/base.py
@@ -129,14 +129,14 @@ def get_offloader() -> BaseOffloader:
     return _instance
 
 
-def set_offloader(instance: BaseOffloader) -> None:
-    """Set the global offloader instance."""
+def set_offloader(instance: BaseOffloader | None) -> None:
+    """Set or reset the global offloader instance."""
     global _instance
-    _instance = instance
-    if isinstance(instance, NoopOffloader):
+    _instance = NoopOffloader() if instance is None else instance
+    if isinstance(_instance, NoopOffloader):
         logger.debug_once("Offloader set to NoopOffloader (no offloading).")
     else:
-        logger.info_once("Offloader set to %s", type(instance).__name__)
+        logger.info_once("Offloader set to %s", type(_instance).__name__)
 
 
 def create_offloader(offload_config: "OffloadConfig") -> BaseOffloader:
diff --git a/vllm/v1/worker/gpu/model_runner.py b/vllm/v1/worker/gpu/model_runner.py
index a5e5ac2497..57f0180edc 100644
--- a/vllm/v1/worker/gpu/model_runner.py
+++ b/vllm/v1/worker/gpu/model_runner.py
@@ -2248,6 +2248,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         torch.accelerator.synchronize()
         if self.aux_output_connector is not None:
             self.aux_output_connector.close()
+        set_offloader(None)
         self.cudagraph_manager = None
         self.fast_prefill = None
         self.pooling_runner = None