[Executor] CUDAGraph support RL training (#3265)

* add clear graph opt backend

* cuda graph support rl

* add branch

* 1.fix dynamic_weight_manager bug 2.add clear api for CasualLM

* open test case

* fix typo

* update mkdocs.yaml

* [Docs]Update mkdocs.yml

* update test case

* use unittest in graph test case

fastdeploy/model_executor/models/deepseek_v3.py

@@ -737,6 +737,10 @@ class DeepseekV3ForCausalLM(ModelForCasualLM):
         )
         return hidden_states
 
+    def clear_grpah_opt_backend(self):
+        """Clear graph optimization bakcend, the captured cuda graph will be cleaned"""
+        self.model.clear_grpah_opt_backend(fd_config=self.fd_config)
+
 
 class DeepSeekV3PretrainedModel(PretrainedModel):
     """