diff --git a/vllm_omni/worker/gpu_model_runner.py b/vllm_omni/worker/gpu_model_runner.py
index ca072fea9..ba6f9e367 100644
--- a/vllm_omni/worker/gpu_model_runner.py
+++ b/vllm_omni/worker/gpu_model_runner.py
@@ -927,7 +927,8 @@ class OmniGPUModelRunner(GPUModelRunner):
     def _update_additional_information(self, scheduler_output: "SchedulerOutput") -> None:
         for new_req in scheduler_output.scheduled_new_reqs:
             payload_info = getattr(new_req, "additional_information", None)
-            self._merge_additional_information_update(new_req.req_id, payload_info)
+            if isinstance(payload_info, dict):
+                self._merge_additional_information_update(new_req.req_id, payload_info)
 
         if hasattr(scheduler_output.scheduled_cached_reqs, "additional_information"):
             cached_infos = getattr(scheduler_output.scheduled_cached_reqs, "additional_information", {})