Recapture cuda graph after model weight update to resolve IMA error (#11780)

python/sglang/srt/model_executor/model_runner.py

@@ -981,6 +981,10 @@ class ModelRunner:
         self.server_args.load_format = load_format
         self.load_config = load_config
 
+        # Recapture device graph after model weight update.
+        if not self.server_args.disable_cuda_graph and self.device == "cuda":
+            self.init_device_graphs()
+
         logger.info("Update weights end.")
         return True, "Succeeded to update model weights."