diff --git a/imperative/python/megengine/distributed/helper.py b/imperative/python/megengine/distributed/helper.py
index 79be16e1000bbdf3e2c50de98204b079994093ea..0755ea2ce3f275612d60312bda6b1087cee7065c 100644
--- a/imperative/python/megengine/distributed/helper.py
+++ b/imperative/python/megengine/distributed/helper.py
@@ -211,6 +211,8 @@ class AllreduceCallback:
         self._grad_origin_device = dict()
 
     def _pack(self, dtype):
+        if len(self._packing_list[dtype]) == 0:
+            return
         grad_list = [self._gradients_dict[p] for p in self._packing_list[dtype]]
         shapes = [p.shape for p in self._packing_list[dtype]]
         reduced_grads = pack_allreduce_split(