Upgrade to vllm 0.17.0 corex v4.1 overlay

vllm/compilation/cuda_graph.py

@@ -21,6 +21,7 @@ from vllm.model_executor.offloader.base import get_offloader
 from vllm.platforms import current_platform
 from vllm.utils.torch_utils import current_stream, weak_ref_tensors
 from vllm.sequence import IntermediateTensors
+
 logger = init_logger(__name__)
 
 
@@ -204,14 +205,14 @@ class CUDAGraphWrapper:
     def unwrap(self) -> Callable[..., Any]:
         # in case we need to access the original runnable.
         return self.runnable
-
+    
     def weak_ref_tensors_with_intermediate(self, output):
         if isinstance(output, IntermediateTensors):
             intermediate_states = IntermediateTensors(
                     tensors={key: weak_ref_tensors(value) for key, value in output.tensors.items()})
             return intermediate_states
         return weak_ref_tensors(output)
-        
+
     def __call__(self, *args: Any, **kwargs: Any) -> Any | None:
         forward_context = get_forward_context()
         batch_descriptor = forward_context.batch_descriptor
@@ -298,12 +299,10 @@ class CUDAGraphWrapper:
                         # the last graph in piecewise cuadgraph mode, because
                         # the output of the last graph will not be used by
                         # any other cuda graph.
-                        # output = weak_ref_tensors(output)
                         output = self.weak_ref_tensors_with_intermediate(output)
 
             # here we always use weak ref for the output
             # to save memory
-            # entry.output = weak_ref_tensors(output)
             entry.output = self.weak_ref_tensors_with_intermediate(output)
             entry.cudagraph = cudagraph