[torch.compile] rework compile control with piecewise cudagraph (#9715)

Signed-off-by: youkaichao <youkaichao@gmail.com>

vllm/plugins/__init__.py

@@ -1,7 +1,8 @@
 import logging
-from typing import Callable, Dict, Optional, Union
+from typing import Callable, Optional, Union
 
 import vllm.envs as envs
+from vllm.compilation.config import CompilationConfig
 
 logger = logging.getLogger(__name__)
 
@@ -44,13 +45,13 @@ def get_torch_compile_backend() -> Optional[Union[Callable, str]]:
     return _torch_compile_backend
 
 
-_inductor_additional_configs: Dict = {}
+_compilation_config: Optional[CompilationConfig] = None
 
 
-def set_inductor_additional_configs(configs: Dict):
-    global _inductor_additional_configs
-    _inductor_additional_configs = configs
+def set_compilation_config(config: Optional[CompilationConfig]):
+    global _compilation_config
+    _compilation_config = config
 
 
-def get_inductor_additional_configs() -> Dict:
-    return _inductor_additional_configs
+def get_compilation_config() -> Optional[CompilationConfig]:
+    return _compilation_config