use device param in load_model method (#13037)

vllm/worker/model_runner.py

@@ -1107,7 +1107,7 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
 
     def load_model(self) -> None:
         logger.info("Starting to load model %s...", self.model_config.model)
-        with DeviceMemoryProfiler() as m:
+        with DeviceMemoryProfiler(self.device) as m:
             self.model = get_model(vllm_config=self.vllm_config)
 
         self.model_memory_usage = m.consumed_memory