Conditionally recapture cuda graph after model weight update from disk (#12060)

python/sglang/srt/model_executor/model_runner.py

@@ -862,6 +862,7 @@ class ModelRunner:
         model_path: str,
         load_format: str,
         weight_name_filter: Optional[Callable[[str], bool]] = None,
+        recapture_cuda_graph: bool = False,
     ) -> tuple[bool, str]:
         """Update engine weights in-place from the disk."""
         logger.info(
@@ -917,6 +918,9 @@ class ModelRunner:
         self.server_args.load_format = load_format
         self.load_config = load_config
 
+        if recapture_cuda_graph and self.device == "cuda":
+            self.init_device_graphs()
+
         logger.info("Update weights end.")
         return True, "Succeeded to update model weights."