diff --git a/paddle/fluid/operators/mlu/activation_op_mlu_test.cc b/paddle/fluid/operators/mlu/activation_op_mlu_test.cc
index f88286288317bd8e7c09cbd23ecccfce5df98e7d..884521301750ce92c3f0a2e0b9468c5cc4a57790 100644
--- a/paddle/fluid/operators/mlu/activation_op_mlu_test.cc
+++ b/paddle/fluid/operators/mlu/activation_op_mlu_test.cc
@@ -21,7 +21,6 @@ limitations under the License. */
 
 namespace fw = paddle::framework;
 namespace plat = paddle::platform;
-namespace math = paddle::operators::math;
 
 USE_OP(relu);
 USE_OP_DEVICE_KERNEL(relu, MLU);
diff --git a/paddle/fluid/operators/uniform_random_op_mlu.cc b/paddle/fluid/operators/uniform_random_op_mlu.cc
index 1600bedc6b2fae9ba65a32e831eae4f43abeddf8..2c5f13f5a930788651c2e287febab7ad06aefd20 100644
--- a/paddle/fluid/operators/uniform_random_op_mlu.cc
+++ b/paddle/fluid/operators/uniform_random_op_mlu.cc
@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/uniform_random_op.h"
+#include "paddle/fluid/framework/generator.h"
 #include "paddle/fluid/operators/mlu/mlu_baseop.h"
 
 namespace paddle {
@@ -57,14 +58,45 @@ class MLUUniformRandomKernel : public framework::OpKernel<T> {
 
     tensor->mutable_data<T>(ctx.GetPlace());
     int64_t size = tensor->numel();
-    const float min = static_cast<T>(ctx.Attr<float>("min"));
-    const float max = static_cast<T>(ctx.Attr<float>("max"));
+
+    Tensor cpu_tensor(tensor->dtype());
+    cpu_tensor.Resize(tensor->dims());
+    T *data_cpu = cpu_tensor.mutable_data<T>(platform::CPUPlace());
+
+    std::uniform_real_distribution<T> dist(
+        static_cast<T>(ctx.Attr<float>("min")),
+        static_cast<T>(ctx.Attr<float>("max")));
     unsigned int seed = static_cast<unsigned int>(ctx.Attr<int>("seed"));
-    // make mlu seed
-    MLUCnnlRandomGeneratorDesc random_desc(/*is_mlu200=*/false, seed);
-    cnnlDataType_t data_type = ToCnnlDataType(tensor->type());
-    MLUCnnl::RandomUniform(ctx, size, /*data type=*/data_type,
-                           random_desc.get(), min, max, GetBasePtr(tensor));
+    auto engine = framework::GetCPURandomEngine(seed);
+
+    for (int64_t i = 0; i < size; ++i) {
+      data_cpu[i] = dist(*engine);
+    }
+
+    unsigned int diag_num =
+        static_cast<unsigned int>(ctx.Attr<int>("diag_num"));
+    unsigned int diag_step =
+        static_cast<unsigned int>(ctx.Attr<int>("diag_step"));
+    auto diag_val = static_cast<T>(ctx.Attr<float>("diag_val"));
+    if (diag_num > 0) {
+      PADDLE_ENFORCE_GT(
+          size, (diag_num - 1) * (diag_step + 1),
+          platform::errors::InvalidArgument(
+              "ShapeInvalid: the diagonal's elements is equal (num-1) "
+              "* (step-1) with num %d, step %d,"
+              "It should be smaller than %d, but received %d",
+              diag_num, diag_step, (diag_num - 1) * (diag_step + 1), size));
+      for (int64_t i = 0; i < diag_num; ++i) {
+        int64_t pos = i * diag_step + i;
+        data_cpu[pos] = diag_val;
+      }
+    }
+
+    // copy to MLU
+    framework::TensorCopy(
+        cpu_tensor, ctx.GetPlace(),
+        ctx.template device_context<platform::DeviceContext>(), tensor);
+    ctx.template device_context<paddle::platform::MLUDeviceContext>().Wait();
   }
 };
 
diff --git a/paddle/scripts/paddle_build.sh b/paddle/scripts/paddle_build.sh
index ed70a8638bf73561dab52d8420b734ff24987c15..41e5e0469dcb4099a89a0517207b8f6d8d39632b 100755
--- a/paddle/scripts/paddle_build.sh
+++ b/paddle/scripts/paddle_build.sh
@@ -1269,6 +1269,8 @@ function card_test() {
         CUDA_DEVICE_COUNT=1
     elif [ "${WITH_ROCM}" == "ON" ];then
         CUDA_DEVICE_COUNT=$(rocm-smi -i | grep GPU | wc -l)
+    elif [ "${WITH_MLU}" == "ON" ];then
+        CUDA_DEVICE_COUNT=1
     else
         CUDA_DEVICE_COUNT=$(nvidia-smi -L | wc -l)
     fi
@@ -2102,6 +2104,130 @@ set -ex
     fi   
 }
 
+function parallel_test_base_mlu() {
+    mkdir -p ${PADDLE_ROOT}/build
+    cd ${PADDLE_ROOT}/build/python/paddle/fluid/tests/unittests/mlu
+    if [ ${WITH_TESTING:-ON} == "ON" ] ; then
+    cat <<EOF
+    ========================================
+    Running unit mlu tests ...
+    ========================================
+EOF
+
+set +x
+        test_cases=$(ctest -N -V) # get all test cases
+        get_quickly_disable_ut||disable_ut_quickly='disable_ut'   # indicate whether the case was in quickly disable list
+        while read -r line; do
+            if [[ "$line" == "" ]]; then
+                continue
+            fi
+            read testcase <<< $(echo "$line"|grep -oEi "\w+$")
+            if [[ "$single_card_tests" == "" ]]; then
+                single_card_tests="^$testcase$"
+            else
+                single_card_tests="$single_card_tests|^$testcase$"
+            fi
+        done <<< "$test_cases";
+
+        ut_actual_total_startTime_s=`date +%s`
+
+        card_test "$single_card_tests" 1 # run cases 1 job each time with single MLU
+        collect_failed_tests
+
+        # add unit test retry for MLU
+        rm -f $tmp_dir/*
+        exec_times=0
+        retry_unittests_record=''
+        retry_time=4
+        exec_time_array=('first' 'second' 'third' 'fourth')
+        parallel_failed_tests_exec_retry_threshold=120
+        exec_retry_threshold=30
+        is_retry_execuate=0
+        rerun_ut_startTime_s=`date +%s`
+        if [ -n "$failed_test_lists" ];then
+            if [ ${TIMEOUT_DEBUG_HELP:-OFF} == "ON" ];then
+                bash $PADDLE_ROOT/tools/timeout_debug_help.sh "$failed_test_lists"    # cat logs for tiemout uts which killed by ctest
+            fi
+            need_retry_ut_str=$(echo "$failed_test_lists" | grep -oEi "\-.+\(.+\)" | sed 's/(.\+)//' | sed 's/- //' )
+            need_retry_ut_arr=(${need_retry_ut_str})
+            need_retry_ut_count=${#need_retry_ut_arr[@]}
+            retry_unittests=$(echo "$failed_test_lists" | grep -oEi "\-.+\(.+\)" | sed 's/(.\+)//' | sed 's/- //' )
+            while ( [ $exec_times -lt $retry_time ] )
+                do
+                    if [[ "${exec_times}" == "0" ]] ;then
+                        if [ $need_retry_ut_count -lt $parallel_failed_tests_exec_retry_threshold ];then
+                            is_retry_execuate=0
+                        else
+                            is_retry_execuate=1
+                        fi
+                    elif [[ "${exec_times}" == "1" ]] ;then
+                        need_retry_ut_str=$(echo "$failed_test_lists" | grep -oEi "\-.+\(.+\)" | sed 's/(.\+)//' | sed 's/- //' )
+                        need_retry_ut_arr=(${need_retry_ut_str})
+                        need_retry_ut_count=${#need_retry_ut_arr[@]} 
+                        if [ $need_retry_ut_count -lt $exec_retry_threshold ];then
+                            is_retry_execuate=0
+                        else
+                            is_retry_execuate=1
+                        fi
+                    fi
+                    if [[ "$is_retry_execuate" == "0" ]];then
+                        set +e
+                        retry_unittests_record="$retry_unittests_record$failed_test_lists"
+                        failed_test_lists_ult=`echo "${failed_test_lists}" |grep -Po '[^ ].*$'`
+                        set -e
+                        if [[ "${exec_times}" == "1" ]] || [[ "${exec_times}" == "3" ]];then
+                            if [[ "${failed_test_lists}" == "" ]];then
+                                break
+                            else
+                                retry_unittests=$(echo "$failed_test_lists" | grep -oEi "\-.+\(.+\)" | sed 's/(.\+)//' | sed 's/- //' )
+                            fi
+                        fi
+                        echo "========================================="
+                        echo "This is the ${exec_time_array[$exec_times]} time to re-run"
+                        echo "========================================="
+                        echo "The following unittest will be re-run:"
+                        echo "${retry_unittests}"                    
+                        for line in ${retry_unittests[@]} ;
+                            do
+                                tmp_one_tmp="$( echo $single_card_tests | grep -oEi $line )"
+
+                                if [[ "$tmp_one_tmp" != ""  ]]; then
+                                    if [[ "$one_card_retry" == "" ]]; then
+                                        one_card_retry="^$line$"
+                                    else
+                                        one_card_retry="$one_card_retry|^$line$"
+                                    fi
+                                fi
+
+                            done
+
+                        if [[ "$one_card_retry" != "" ]]; then
+                            card_test "$one_card_retry" 1 # run cases 1 job each time with single GPU
+                        fi
+                        exec_times=$[$exec_times+1]
+                        failed_test_lists=''
+                        collect_failed_tests
+                        rm -f $tmp_dir/*
+                        one_card_retry=''
+                    else 
+                        break
+                    fi
+
+                done
+        fi
+
+        rerun_ut_endTime_s=`date +%s`
+        
+        echo "ipipe_log_param_Rerun_TestCases_Total_Time: $[ $rerun_ut_endTime_s - $rerun_ut_startTime_s ]s" >> ${PADDLE_ROOT}/build/build_summary.txt
+        ut_actual_total_endTime_s=`date +%s`
+        echo "ipipe_log_param_actual_TestCases_Total_Time: $[ $ut_actual_total_endTime_s - $ut_actual_total_startTime_s ]s" >> ${PADDLE_ROOT}/build/build_summary.txt
+        if [[ "$EXIT_CODE" != "0" ]]; then
+            show_ut_retry_result
+        fi
+set -ex
+    fi   
+}
+
 function parallel_test() {
     mkdir -p ${PADDLE_ROOT}/build
     cd ${PADDLE_ROOT}/build
@@ -2117,6 +2243,8 @@ function parallel_test() {
         parallel_test_base_xpu
     elif [ "$WITH_ASCEND_CL" == "ON" ];then
         parallel_test_base_npu
+    elif [ "$WITH_MLU" == "ON" ];then
+        parallel_test_base_mlu
     else
         parallel_test_base_cpu ${PROC_RUN:-1}
     fi
@@ -2873,6 +3001,11 @@ function main() {
         parallel_test
         check_coverage
         ;;
+      check_mlu_coverage)
+        cmake_gen_and_build ${PYTHON_ABI:-""} ${parallel_number}
+        parallel_test
+        check_coverage
+        ;;
       reuse_so_cicheck_py35)
         reuse_so_cache
         parallel_test
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_accuracy_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_accuracy_op_mlu.py
index e229966c12d245921b1b059efa011d07405ada94..5e5c4c9a301e9353497289082f322f63ae9a981e 100755
--- a/python/paddle/fluid/tests/unittests/mlu/test_accuracy_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_accuracy_op_mlu.py
@@ -23,6 +23,8 @@ import paddle
 import paddle.fluid as fluid
 from paddle.fluid import compiler, Program, program_guard
 
+paddle.enable_static()
+
 
 class TestAccuracyOp(OpTest):
     def setUp(self):
@@ -132,5 +134,4 @@ class TestAccuracyAPI(unittest.TestCase):
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu.py
index 2150e06381fac37e527d9d593f8752eb38ba1596..4cbff21dfc4965d8aa47955437ac90a7b62dd13e 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu.py
@@ -29,6 +29,7 @@ import paddle.fluid as fluid
 from paddle.fluid import Program, program_guard
 
 _set_use_system_allocator(True)
+paddle.enable_static()
 
 
 def _reference_testing(x, scale, offset, mean, var, epsilon, data_format):
@@ -698,5 +699,4 @@ class TestDygraphBatchNormOpenReserveSpace(unittest.TestCase):
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu_v2.py b/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu_v2.py
index f608344f6e0363864a76a23f8d8c10dace130149..7dd9dcdee57f99e71c4bb889e9a03ef248dbcfe7 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu_v2.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu_v2.py
@@ -26,6 +26,8 @@ import paddle.fluid as fluid
 from paddle.fluid import Program, program_guard
 import paddle
 
+paddle.enable_static()
+
 
 class TestBatchNorm(unittest.TestCase):
     def test_name(self):
@@ -291,5 +293,4 @@ class TestBatchNormUseGlobalStatsCase3(TestBatchNormUseGlobalStats):
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_cast_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_cast_op_mlu.py
index 71f79c34d2312eda3952447a5748e66b44d1ab82..10356b124b2ea3a020bbfbf95b9cc0778f193a99 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_cast_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_cast_op_mlu.py
@@ -25,6 +25,8 @@ import paddle.fluid.core as core
 import paddle.fluid as fluid
 from paddle.fluid import compiler, Program, program_guard
 
+paddle.enable_static()
+
 
 class TestCastOpFp32ToFp16(OpTest):
     def setUp(self):
@@ -119,17 +121,7 @@ class TestCastOpError(unittest.TestCase):
             x1 = fluid.create_lod_tensor(
                 np.array([[-1]]), [[1]], fluid.MLUPlace(0))
             self.assertRaises(TypeError, fluid.layers.cast, x1, 'int32')
-            # The input dtype of cast_op must be bool, float16, float32, float64, int32, int64, uint8.
-            x2 = fluid.layers.data(name='x2', shape=[4], dtype='int16')
-            self.assertRaises(TypeError, fluid.layers.cast, x2, 'int32')
-
-            def test_dtype_type():
-                x4 = fluid.layers.data(name='x4', shape=[4], dtype='int32')
-                output = fluid.layers.cast(x=x4, dtype='int16')
-
-            self.assertRaises(TypeError, test_dtype_type)
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_concat_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_concat_op_mlu.py
index 3bfa96b70011238b48f55d628ad17794b84ff5de..ba37fcee15472a46cd2c64ddd5418f525e759569 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_concat_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_concat_op_mlu.py
@@ -176,7 +176,7 @@ def create_test_AxisTensor(parent):
     class TestConcatAxisTensor(parent):
         def setUp(self):
             self.op_type = "concat"
-            self.dtype = self.init_dtype()
+            self.init_dtype()
             self.init_test_data()
 
             self.inputs = {
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_elementwise_add_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_elementwise_add_op_mlu.py
index 5b6db6903fba0dc1d77b6026623a3bc3c101013c..3dc711c7d75e1e93bfc727e3979af8a9e0d55bd1 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_elementwise_add_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_elementwise_add_op_mlu.py
@@ -23,6 +23,8 @@ from op_test import OpTest, skip_check_grad_ci
 import paddle.fluid as fluid
 from paddle.fluid import compiler, Program, program_guard
 
+paddle.enable_static()
+
 
 class TestElementwiseAddOp(OpTest):
     def set_mlu(self):
@@ -523,5 +525,4 @@ class TestBoolAddFloatElementwiseAddop(unittest.TestCase):
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_fill_constant_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_fill_constant_op_mlu.py
index 6610127d382bd3a715b64ad359c500fefc595936..a43b7d0164d7bb3154999db1dcf61e79a537e021 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_fill_constant_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_fill_constant_op_mlu.py
@@ -27,6 +27,8 @@ import paddle.fluid as fluid
 import numpy as np
 from paddle.fluid import compiler, Program, program_guard
 
+paddle.enable_static()
+
 
 # Situation 1: Attr(shape) is a list(without tensor)
 class TestFillConstantOp1(OpTest):
@@ -449,5 +451,4 @@ class TestFillConstantOpError(unittest.TestCase):
 
 
 if __name__ == "__main__":
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_gaussian_random_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_gaussian_random_op_mlu.py
index 97a945dc905715206a8042a7fa2ea9bacb5f015c..6f64196a586dd0069f9340286a97077a3676809b 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_gaussian_random_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_gaussian_random_op_mlu.py
@@ -26,6 +26,8 @@ sys.path.append('..')
 from op_test import OpTest
 import paddle
 
+paddle.enable_static()
+
 
 class TestGaussianRandomOp(OpTest):
     def setUp(self):
@@ -74,5 +76,4 @@ class TestMeanStdAreInt(TestGaussianRandomOp):
 
 
 if __name__ == "__main__":
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_momentum_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_momentum_op_mlu.py
index af09eabe787dc3d0ca4aa84852847f06cea39927..a2cd69fee325a6074ac356885b5f851b5140649c 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_momentum_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_momentum_op_mlu.py
@@ -26,6 +26,8 @@ import paddle.fluid as fluid
 import numpy
 from test_momentum_op import calculate_momentum_by_numpy
 
+paddle.enable_static()
+
 
 class TestMomentumOp1(OpTest):
     def setUp(self):
@@ -608,5 +610,4 @@ class TestMultiTensorMomentumStatic(unittest.TestCase):
 
 
 if __name__ == "__main__":
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_pool2d_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_pool2d_op_mlu.py
index fd442c6205e98d26b4797ff2ef4499b376bc8bdd..1be3d2d85a4220406c90d8b164d7d9b3731b9f87 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_pool2d_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_pool2d_op_mlu.py
@@ -27,6 +27,8 @@ sys.path.append('..')
 from op_test import OpTest
 from test_pool2d_op import pool2D_forward_naive, avg_pool2D_forward_naive, max_pool2D_forward_naive, adaptive_start_index, adaptive_end_index
 
+paddle.enable_static()
+
 
 def pool2d_backward_navie(x,
                           ksize,
@@ -1016,5 +1018,4 @@ class TestDygraphPool2DAPI(unittest.TestCase):
 
 
 if __name__ == '__main__':
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_scale_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_scale_op_mlu.py
index bb7f438c4ab2b5eaae7ac2df2f7c1978186b1a05..53254c738d985db028c7fdfe80b715a14ceaec03 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_scale_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_scale_op_mlu.py
@@ -25,6 +25,8 @@ import paddle.fluid.core as core
 from paddle.fluid.op import Operator
 from paddle.static import Program, program_guard
 
+paddle.enable_static()
+
 
 class TestScaleOp(OpTest):
     def setUp(self):
@@ -201,5 +203,4 @@ class TestScaleInplaceApiDygraph(TestScaleApiDygraph):
 
 
 if __name__ == "__main__":
-    paddle.enable_static()
     unittest.main()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_top_k_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_top_k_op_mlu.py
index 8ad0e787ab0cc277a4086f6c98f21f7d15fc09d7..366f783ce0d2a1bf482e5bbba0377cc445d26d1e 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_top_k_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_top_k_op_mlu.py
@@ -22,6 +22,8 @@ from op_test import OpTest
 import paddle
 import paddle.fluid.core as core
 
+paddle.enable_static()
+
 
 class TestTopkOp(OpTest):
     def setUp(self):
@@ -69,5 +71,4 @@ class TestTopkFP16Op(TestTopkOp):
 
 
 if __name__ == "__main__":
-    paddle.enable_static()
     unittest.main()