更改默认的full _cuda_graph启动方式为false

vllm/config.py

@@ -4106,7 +4106,7 @@ class CompilationConfig:
     are always used, it can set this to False. Otherwise, it should
     set this to True, and the compiler will copy the input to an
     internally managed buffer. Default is False."""
-    full_cuda_graph: bool = True
+    full_cuda_graph: bool = False
     """whether to use a full cuda graph for the entire forward pass rather than
     splitting certain operations such as attention into subgraphs. Thus this
     flag cannot be used together with splitting_ops. This may provide
@@ -4948,4 +4948,4 @@ def get_layers_from_vllm_config(vllm_config: VllmConfig,
         for layer_name, layer in
         vllm_config.compilation_config.static_forward_context.items()
         if isinstance(layer, layer_type)
-    }
+    }
\ No newline at end of file