diff --git a/example_trainer/vllm_api_server.py b/example_trainer/vllm_api_server.py
index 3b6b7d33..4bab25b5 100644
--- a/example_trainer/vllm_api_server.py
+++ b/example_trainer/vllm_api_server.py
@@ -747,7 +747,6 @@ async def lora_unload() -> JSONResponse:
 
 def build_app(args: Namespace) -> FastAPI:
     """Build the FastAPI application."""
-    global app
     app.root_path = args.root_path
     return app
 
diff --git a/example_trainer/vllm_patching/patched_gpu_runner.py b/example_trainer/vllm_patching/patched_gpu_runner.py
index 8fd56708..e4b0ae32 100644
--- a/example_trainer/vllm_patching/patched_gpu_runner.py
+++ b/example_trainer/vllm_patching/patched_gpu_runner.py
@@ -19,12 +19,6 @@ from __future__ import annotations
 
 import os
 import sys
-from typing import TYPE_CHECKING
-
-# Lazy imports to avoid circular dependencies
-if TYPE_CHECKING:
-    from vllm.v1.worker.gpu_model_runner import GPUModelRunner
-
 
 # Flag to track if patches have been applied
 _PATCHES_APPLIED = False