diff --git a/python/paddle/distributed/fleet/meta_optimizers/sharding_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/sharding_optimizer.py
index 52468ab5334967b35532711133a380b54706f8de..c4d42f90615fc1251ff19126bf992b38cc02d11b 100755
--- a/python/paddle/distributed/fleet/meta_optimizers/sharding_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/sharding_optimizer.py
@@ -1621,13 +1621,8 @@ class ShardingOptimizer(MetaOptimizerBase):
             persistable=True,
             force_cpu=True)
 
-        cond_var = layers.create_global_var(
-            name="gradient_merge_cond",
-            shape=[1],
-            value=bool(0),
-            dtype='bool',
-            persistable=False,
-            force_cpu=True)
+        cond_var = main_block.create_var(
+            name="gradient_merge_cond", shape=[1], dtype='bool')
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step
diff --git a/python/paddle/distributed/passes/auto_parallel_gradient_merge.py b/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
index 7668dff36207ed700f5aa6378cb0f5532cfedd3f..accac81133825cd41a826e9efeed847a70ef4a22 100644
--- a/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
+++ b/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
@@ -107,13 +107,8 @@ def _get_gm_cond_var(main_program, k_steps, dist_context):
         force_cpu=True)
     set_var_dist_attr(dist_context, step_var, [-1], world_process_group.ranks)
 
-    cond_var = layers.create_global_var(
-        name="gradient_merge_cond",
-        shape=[1],
-        value=bool(0),
-        dtype='bool',
-        persistable=False,
-        force_cpu=True)
+    cond_var = main_block.create_var(
+        name="gradient_merge_cond", shape=[1], dtype='bool')
     set_var_dist_attr(dist_context, cond_var, [-1], world_process_group.ranks)
 
     with device_guard("cpu"):
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 7bf4608de89c9cb57e7f2309e6165dbb636ce41a..8242d8e3392ec02ebb0f335b099d24eebd9fff06 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7098,13 +7098,8 @@ class GradientMergeOptimizer(object):
             persistable=True,
             force_cpu=True)
 
-        cond_var = layers.create_global_var(
-            name="gradient_merge_cond",
-            shape=[1],
-            value=bool(0),
-            dtype='bool',
-            persistable=False,
-            force_cpu=True)
+        cond_var = main_block.create_var(
+            name="gradient_merge_cond", shape=[1], dtype='bool')
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step