[V1] fix torch profiling for V1 offline scenarios (#18445)

Signed-off-by: Divakar Verma <divakar.verma@amd.com>

vllm/v1/worker/gpu_worker.py

@@ -292,6 +292,8 @@ class Worker(WorkerBase):
             self.profiler.start()
         else:
             self.profiler.stop()
+            print(self.profiler.key_averages().table(
+                sort_by="self_cuda_time_total"))
 
     def execute_dummy_batch(self) -> None:
         self.model_runner._dummy_run(1)