diff --git a/paddle/fluid/API.spec b/paddle/fluid/API.spec
index 964be52ee9a9a30a427eab6951c8395d247153df..075634a306344f9f391ce7c5b4e2a0d2a8fdd037 100644
--- a/paddle/fluid/API.spec
+++ b/paddle/fluid/API.spec
@@ -77,8 +77,8 @@ paddle.fluid.io.save_persistables ArgSpec(args=['executor', 'dirname', 'main_pro
 paddle.fluid.io.load_vars ArgSpec(args=['executor', 'dirname', 'main_program', 'vars', 'predicate', 'filename'], varargs=None, keywords=None, defaults=(None, None, None, None))
 paddle.fluid.io.load_params ArgSpec(args=['executor', 'dirname', 'main_program', 'filename'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.io.load_persistables ArgSpec(args=['executor', 'dirname', 'main_program', 'filename'], varargs=None, keywords=None, defaults=(None, None))
-paddle.fluid.io.load_inference_model ArgSpec(args=['dirname', 'executor', 'model_filename', 'params_filename', 'pserver_endpoints'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.io.save_inference_model ArgSpec(args=['dirname', 'feeded_var_names', 'target_vars', 'executor', 'main_program', 'model_filename', 'params_filename', 'export_for_deployment'], varargs=None, keywords=None, defaults=(None, None, None, True))
+paddle.fluid.io.load_inference_model ArgSpec(args=['dirname', 'executor', 'model_filename', 'params_filename', 'pserver_endpoints'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.io.get_inference_program ArgSpec(args=['target_vars', 'main_program'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.initializer.ConstantInitializer.__init__ ArgSpec(args=['self', 'value', 'force_cpu'], varargs=None, keywords=None, defaults=(0.0, False))
 paddle.fluid.initializer.UniformInitializer.__init__ ArgSpec(args=['self', 'low', 'high', 'seed'], varargs=None, keywords=None, defaults=(-1.0, 1.0, 0))
diff --git a/python/paddle/fluid/io.py b/python/paddle/fluid/io.py
index f4a1fa78b8956b2d555b87e64feb75fd692fa433..049c0e40e6b5538fea6f45e64fffa5fd920322d6 100644
--- a/python/paddle/fluid/io.py
+++ b/python/paddle/fluid/io.py
@@ -672,12 +672,11 @@ def save_inference_model(dirname,
     save_persistables(executor, dirname, inference_program, params_filename)
 
     # if there is lookup table, the trainer 0 will notify all pserver to save.
-    if main_program._is_distributed and main_program._is_chief:
-        if main_program._distributed_lookup_table:
-            lookup_table_filename = os.path.join(dirname, "__lookup_table__")
-            _save_lookup_tables_by_notify(
-                executor, lookup_table_filename,
-                main_program._distributed_lookup_table, main_program._endpoints)
+    if main_program._is_distributed and main_program._is_chief and main_program._distributed_lookup_table:
+        lookup_table_filename = os.path.join(dirname, "__lookup_table__")
+        _save_lookup_tables_by_notify(executor, lookup_table_filename,
+                                      main_program._distributed_lookup_table,
+                                      main_program._endpoints)
 
 
 def load_inference_model(dirname,
diff --git a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
index 0543e6238109bf5bab68a9f08fc34678936e756f..b88fd2707d58ffefad6ad85c9880b64725d7e421 100644
--- a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
+++ b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
@@ -536,6 +536,19 @@ class TestAsyncDistLookupTable(TestDistLookupTableBase):
         self.assertEqual([op.type for op in trainer.blocks[0].ops], ops)
 
 
+class TestDistArgsInProgram(TestDistLookupTableBase):
+    def net_conf(self):
+        self.network_with_table(is_sparse=True, is_distributed=True)
+
+    def transpiler_test_impl(self):
+        config = fluid.DistributeTranspilerConfig()
+        pserver1, _ = self.get_pserver(self.pserver1_ep, config, False)
+
+        self.assertTrue(pserver1._is_chief)
+        self.assertTrue(pserver1._is_distributed)
+        self.assertEqual(pserver1._distributed_lookup_table)
+
+
 class TestRMSPropOptimizer(TranspilerTest):
     def net_conf(self):
         x = fluid.layers.data(name='x', shape=[1000], dtype='float32')
@@ -566,5 +579,41 @@ class TestRMSPropOptimizer(TranspilerTest):
         self.assertEqual(moment_var.shape, (500, 1000))
 
 
+class TestLoadSliceVar(TranspilerTest):
+    def net_conf(self):
+        x = fluid.layers.data(name='x', shape=[1000], dtype='float32')
+        y_predict = fluid.layers.fc(input=x,
+                                    size=1000,
+                                    act=None,
+                                    param_attr=fluid.ParamAttr(name='fc_w'),
+                                    bias_attr=fluid.ParamAttr(name='fc_b'))
+        y = fluid.layers.data(name='y', shape=[1], dtype='float32')
+        cost = fluid.layers.square_error_cost(input=y_predict, label=y)
+        avg_cost = fluid.layers.mean(cost)
+        optimizer = fluid.optimizer.RMSProp(learning_rate=0.1)
+        optimizer.minimize(avg_cost)
+        return
+
+    def transpiler_test_impl(self):
+        pserver, _ = self.get_pserver(self.pserver1_ep)
+        pserver2, _ = self.get_pserver(self.pserver2_ep)
+
+        self.assertTrue(pserver._slice_vars_and_atts)
+        self.assertTrue(pserver2._slice_vars_and_atts)
+
+        for idx in xrange(len(pserver._slice_vars_and_atts)):
+            self.assertEqual(pserver._slice_vars_and_atts[idx][0],
+                             pserver2._slice_vars_and_atts[idx][0])
+
+            total_numel = reduce(lambda x, y: x * y,
+                                 pserver._slice_vars_and_atts[idx][0].shape)
+            self.assertEqual(
+                total_numel,
+                reduce(lambda x, y: x * y,
+                       pserver._slice_vars_and_atts[idx][2].shape) + reduce(
+                           lambda x, y: x * y,
+                           pserver2._slice_vars_and_atts[idx][2].shape))
+
+
 if __name__ == "__main__":
     unittest.main()