diff --git a/python/paddle/distributed/fleet/utils/hybrid_parallel_util.py b/python/paddle/distributed/fleet/utils/hybrid_parallel_util.py
index 7e95bfe7f331aadd3b5aeb6adc99a1a5706d7c6b..7e527eced3f0419f5601efa2138b0f51917194c4 100644
--- a/python/paddle/distributed/fleet/utils/hybrid_parallel_util.py
+++ b/python/paddle/distributed/fleet/utils/hybrid_parallel_util.py
@@ -197,7 +197,7 @@ def fused_allreduce_gradients_with_group(
         else _apply_collective_grads
     )
     with framework.no_grad():
-        apply_func(parameter_list, group, bucket_size)
+        apply_func(parameter_list, group, bucket_size, scale)
 
 
 def fused_allreduce_gradients(parameter_list, hcg):