diff --git a/python/paddle/fluid/tests/unittests/test_parallel_executor_crf.py b/python/paddle/fluid/tests/unittests/test_parallel_executor_crf.py
index d6dbedcf875b06f5af4597e6dad71a39f286a471..84b0aad8acb096a32f625e32fb640599f2882d97 100644
--- a/python/paddle/fluid/tests/unittests/test_parallel_executor_crf.py
+++ b/python/paddle/fluid/tests/unittests/test_parallel_executor_crf.py
@@ -16,6 +16,7 @@ from __future__ import print_function
 
 import paddle.dataset.conll05 as conll05
 import paddle.fluid as fluid
+import paddle.fluid.core as core
 import unittest
 import paddle
 import numpy as np
@@ -177,32 +178,36 @@ class TestCRFModel(unittest.TestCase):
     def test_update_sparse_parameter_all_reduce(self):
         build_strategy = fluid.BuildStrategy()
         build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.AllReduce
-        self.check_network_convergence(
-            is_sparse=True, build_strategy=build_strategy, use_cuda=True)
+        if core.is_compiled_with_cuda():
+            self.check_network_convergence(
+                is_sparse=True, build_strategy=build_strategy, use_cuda=True)
         self.check_network_convergence(
             is_sparse=True, build_strategy=build_strategy, use_cuda=False)
 
     def test_update_dense_parameter_all_reduce(self):
         build_strategy = fluid.BuildStrategy()
         build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.AllReduce
-        self.check_network_convergence(
-            is_sparse=False, build_strategy=build_strategy, use_cuda=True)
+        if core.is_compiled_with_cuda():
+            self.check_network_convergence(
+                is_sparse=False, build_strategy=build_strategy, use_cuda=True)
         self.check_network_convergence(
             is_sparse=False, build_strategy=build_strategy, use_cuda=False)
 
     def test_update_sparse_parameter_reduce(self):
         build_strategy = fluid.BuildStrategy()
         build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.Reduce
-        self.check_network_convergence(
-            is_sparse=True, build_strategy=build_strategy, use_cuda=True)
+        if core.is_compiled_with_cuda():
+            self.check_network_convergence(
+                is_sparse=True, build_strategy=build_strategy, use_cuda=True)
         self.check_network_convergence(
             is_sparse=True, build_strategy=build_strategy, use_cuda=False)
 
     def test_update_dense_parameter_reduce(self):
         build_strategy = fluid.BuildStrategy()
         build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.Reduce
-        self.check_network_convergence(
-            is_sparse=False, build_strategy=build_strategy, use_cuda=True)
+        if core.is_compiled_with_cuda():
+            self.check_network_convergence(
+                is_sparse=False, build_strategy=build_strategy, use_cuda=True)
         self.check_network_convergence(
             is_sparse=False, build_strategy=build_strategy, use_cuda=False)