diff --git a/test/auto_parallel/random_control_unittest.py b/test/auto_parallel/random_control_unittest.py
index 6fcee26d21201..addc384699885 100644
--- a/test/auto_parallel/random_control_unittest.py
+++ b/test/auto_parallel/random_control_unittest.py
@@ -86,7 +86,7 @@ def compare_mask_between_ranks(
                 mask_tensor_remote = paddle.ones_like(mask_tensor_local)
                 dy_broadcast_helper(mask_tensor_remote)
                 if equal:
-                    assert np.array_equal(
+                    np.testing.assert_array_equal(
                         mask_tensor_remote.numpy(), mask_tensor_local.numpy()
                     )
                 else:
@@ -205,7 +205,7 @@ def test_random_ctrl_with_recompute(self):
         for i in range(7):
             mask_fw = mask_np_list[i].astype("float32")
             mask_rc = mask_np_list[i + 7].astype("float32")
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 mask_fw,
                 mask_rc,
             )
diff --git a/test/collective/fleet/dygraph_dist_save_load.py b/test/collective/fleet/dygraph_dist_save_load.py
index 0d4b50a9eebc4..b2e6216f54612 100644
--- a/test/collective/fleet/dygraph_dist_save_load.py
+++ b/test/collective/fleet/dygraph_dist_save_load.py
@@ -208,7 +208,7 @@ def step_check(path1, path2):
     m1 = paddle.load(path1)
     m2 = paddle.load(path2)
     for v1, v2 in zip(m1, m2):
-        assert np.allclose(v1.numpy(), v2.numpy())
+        np.testing.assert_allclose(v1.numpy(), v2.numpy())
         print(f"value same: {v1.name}")
 
 
diff --git a/test/collective/fleet/dygraph_save_for_auto_infer.py b/test/collective/fleet/dygraph_save_for_auto_infer.py
index 724267fbaf736..f184659197d52 100644
--- a/test/collective/fleet/dygraph_save_for_auto_infer.py
+++ b/test/collective/fleet/dygraph_save_for_auto_infer.py
@@ -267,7 +267,7 @@ def step_check(output_dir):
     m1 = np.load(p1).reshape(-1)
     m2 = np.load(p2).reshape(-1)
     try:
-        assert np.allclose(m1, m2, rtol=1e-5, atol=1e-6)
+        np.testing.assert_allclose(m1, m2, rtol=1e-5, atol=1e-6)
     except:
         diff = m1 - m2
         logger.error(f"max diff{diff.max()}, min diff: {diff.min()}")
diff --git a/test/collective/fleet/fused_attention_pass_with_mp.py b/test/collective/fleet/fused_attention_pass_with_mp.py
index c8a1673e6630d..2f1e657cfc8be 100644
--- a/test/collective/fleet/fused_attention_pass_with_mp.py
+++ b/test/collective/fleet/fused_attention_pass_with_mp.py
@@ -234,7 +234,7 @@ def get_rst(self, use_pass=False):
     def test_pass(self):
         fused_rst = self.get_rst(use_pass=True)
         non_fused_rst = self.get_rst()
-        assert np.allclose(fused_rst, non_fused_rst, atol=1e-5)
+        np.testing.assert_allclose(fused_rst, non_fused_rst, atol=1e-5)
 
 
 if __name__ == "__main__":
diff --git a/test/collective/fleet/hybrid_parallel_communicate_group.py b/test/collective/fleet/hybrid_parallel_communicate_group.py
index e89e807ae4309..1b9febf04f874 100644
--- a/test/collective/fleet/hybrid_parallel_communicate_group.py
+++ b/test/collective/fleet/hybrid_parallel_communicate_group.py
@@ -58,28 +58,28 @@ def test_all(self):
             sync_op=True,
         )
         if dp_rank == 0:
-            assert np.array_equal(result, self.tensor2)
+            np.testing.assert_array_equal(result, self.tensor2)
         elif dp_rank == 1:
-            assert np.array_equal(result, self.tensor1)
+            np.testing.assert_array_equal(result, self.tensor1)
         print("test scatter api ok")
 
         paddle.distributed.broadcast(result, src=1, group=dp_gp, sync_op=True)
-        assert np.array_equal(result, self.tensor1)
+        np.testing.assert_array_equal(result, self.tensor1)
         print("test broadcast api ok")
 
         paddle.distributed.reduce(
             result, dst=dp_src_rank, group=dp_gp, sync_op=True
         )
         if dp_rank == 0:
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 result, paddle.add(self.tensor1, self.tensor1)
             )
         elif dp_rank == 1:
-            assert np.array_equal(result, self.tensor1)
+            np.testing.assert_array_equal(result, self.tensor1)
         print("test reduce api ok")
 
         paddle.distributed.all_reduce(result, sync_op=True)
-        assert np.array_equal(
+        np.testing.assert_array_equal(
             result,
             paddle.add(paddle.add(self.tensor1, self.tensor1), self.tensor1),
         )
@@ -93,8 +93,8 @@ def test_all(self):
         paddle.distributed.all_gather(
             result, self.tensor1, group=dp_gp, sync_op=True
         )
-        assert np.array_equal(result[0], self.tensor1)
-        assert np.array_equal(result[1], self.tensor1)
+        np.testing.assert_array_equal(result[0], self.tensor1)
+        np.testing.assert_array_equal(result[1], self.tensor1)
         print("test all_gather api ok")
 
         paddle.distributed.barrier(group=dp_gp)
diff --git a/test/collective/fleet/new_group.py b/test/collective/fleet/new_group.py
index 1945bae835571..deeea06665019 100644
--- a/test/collective/fleet/new_group.py
+++ b/test/collective/fleet/new_group.py
@@ -36,26 +36,26 @@ def test_all(self):
             result, [self.tensor2, self.tensor1], src=0, group=gp, sync_op=True
         )
         if gp.rank == 0:
-            assert np.array_equal(result, self.tensor2)
+            np.testing.assert_array_equal(result, self.tensor2)
         elif gp.rank == 1:
-            assert np.array_equal(result, self.tensor1)
+            np.testing.assert_array_equal(result, self.tensor1)
         print("test scatter api ok")
 
         paddle.distributed.broadcast(result, src=1, group=gp, sync_op=True)
-        assert np.array_equal(result, self.tensor1)
+        np.testing.assert_array_equal(result, self.tensor1)
         print("test broadcast api ok")
 
         paddle.distributed.reduce(result, dst=0, group=gp, sync_op=True)
         if gp.rank == 0:
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 result, paddle.add(self.tensor1, self.tensor1)
             )
         elif gp.rank == 1:
-            assert np.array_equal(result, self.tensor1)
+            np.testing.assert_array_equal(result, self.tensor1)
         print("test reduce api ok")
 
         paddle.distributed.all_reduce(result, sync_op=True)
-        assert np.array_equal(
+        np.testing.assert_array_equal(
             result,
             paddle.add(paddle.add(self.tensor1, self.tensor1), self.tensor1),
         )
@@ -69,8 +69,8 @@ def test_all(self):
         paddle.distributed.all_gather(
             result, self.tensor1, group=gp, sync_op=True
         )
-        assert np.array_equal(result[0], self.tensor1)
-        assert np.array_equal(result[1], self.tensor1)
+        np.testing.assert_array_equal(result[0], self.tensor1)
+        np.testing.assert_array_equal(result[1], self.tensor1)
         print("test all_gather api ok")
 
         paddle.distributed.barrier(group=gp)
diff --git a/test/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py b/test/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
index 5de19dfb4113b..6f71afb296efb 100644
--- a/test/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
+++ b/test/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
@@ -201,8 +201,8 @@ def scale(self):
                 data = paddle.rand([10, 1024])
             scaler = paddle.amp.AmpScaler(init_loss_scaling=1024)
             scaled_data = scaler.scale(data)
-            self.assertEqual(
-                np.array_equal(scaled_data.numpy(), data.numpy() * 1024), True
+            np.testing.assert_array_equal(
+                scaled_data.numpy(), data.numpy() * 1024
             )
 
     def test_scale(self):
diff --git a/test/collective/multinode/dygraph_hybrid_dp.py b/test/collective/multinode/dygraph_hybrid_dp.py
index 3d2f8f9e5d3b3..d31177d9813c0 100644
--- a/test/collective/multinode/dygraph_hybrid_dp.py
+++ b/test/collective/multinode/dygraph_hybrid_dp.py
@@ -39,7 +39,7 @@ def check_pass(self, *args, **kwargs):
         paddle.distributed.collective.all_reduce(data_part)
         data_reduced = data_part
         data_sumed = np.sum(data, axis=0)
-        assert np.allclose(
+        np.testing.assert_allclose(
             data_sumed, data_reduced.numpy(), rtol=1e-8, atol=1e-8
         )
 
diff --git a/test/collective/multinode/dygraph_hybrid_dpppmp.py b/test/collective/multinode/dygraph_hybrid_dpppmp.py
index 55a130e3f6ad7..4a52d305090b8 100644
--- a/test/collective/multinode/dygraph_hybrid_dpppmp.py
+++ b/test/collective/multinode/dygraph_hybrid_dpppmp.py
@@ -197,7 +197,9 @@ def check_pass(self, *args, **kwargs):
 
             loss_base_arr.append(loss_base.numpy())
             loss_hybrid_arr.append(loss.numpy())
-        assert np.allclose(loss_base_arr, loss_hybrid_arr, rtol=1e-5, atol=1e-5)
+        np.testing.assert_allclose(
+            loss_base_arr, loss_hybrid_arr, rtol=1e-5, atol=1e-5
+        )
 
 
 if __name__ == "__main__":
diff --git a/test/collective/multinode/dygraph_hybrid_fp16.py b/test/collective/multinode/dygraph_hybrid_fp16.py
index 34319a6c018c6..6bd3e4390a190 100644
--- a/test/collective/multinode/dygraph_hybrid_fp16.py
+++ b/test/collective/multinode/dygraph_hybrid_fp16.py
@@ -210,7 +210,9 @@ def check_pass(self, *args, **kwargs):
 
             loss_base_arr.append(loss_base.numpy())
             loss_hybrid_arr.append(loss)
-        assert np.allclose(loss_base_arr, loss_hybrid_arr, rtol=1e-3, atol=1e-3)
+        np.testing.assert_allclose(
+            loss_base_arr, loss_hybrid_arr, rtol=1e-3, atol=1e-3
+        )
 
 
 if __name__ == "__main__":
diff --git a/test/collective/multinode/dygraph_hybrid_recompute.py b/test/collective/multinode/dygraph_hybrid_recompute.py
index 1902b716296d8..6de5e336acf3f 100644
--- a/test/collective/multinode/dygraph_hybrid_recompute.py
+++ b/test/collective/multinode/dygraph_hybrid_recompute.py
@@ -186,7 +186,9 @@ def check_pass(self, *args, **kwargs):
 
             loss_base_arr.append(loss_base.numpy())
             loss_hybrid_arr.append(loss)
-        assert np.allclose(loss_base_arr, loss_hybrid_arr, rtol=1e-5, atol=1e-5)
+        np.testing.assert_allclose(
+            loss_base_arr, loss_hybrid_arr, rtol=1e-5, atol=1e-5
+        )
 
 
 if __name__ == "__main__":
diff --git a/test/collective/process_group_gloo.py b/test/collective/process_group_gloo.py
index 20dcae5928ad9..b6ae187cc5ffa 100644
--- a/test/collective/process_group_gloo.py
+++ b/test/collective/process_group_gloo.py
@@ -76,11 +76,11 @@ def test_create_process_group_gloo(self):
         if rank == 0:
             task = pg.allreduce(tensor_x, core.ReduceOp.MAX)
             task.wait()
-            assert np.array_equal(tensor_x, max_result)
+            np.testing.assert_array_equal(tensor_x, max_result)
         else:
             task = pg.allreduce(tensor_y, core.ReduceOp.MAX)
             task.wait()
-            assert np.array_equal(tensor_y, max_result)
+            np.testing.assert_array_equal(tensor_y, max_result)
 
         print("test allreduce max api ok")
 
@@ -95,10 +95,10 @@ def test_create_process_group_gloo(self):
         broadcast_result = paddle.assign(tensor_x)
         if rank == 0:
             task = pg.broadcast(tensor_x, 0)
-            assert np.array_equal(broadcast_result, tensor_x)
+            np.testing.assert_array_equal(broadcast_result, tensor_x)
         else:
             task = pg.broadcast(tensor_y, 0)
-            assert np.array_equal(broadcast_result, tensor_y)
+            np.testing.assert_array_equal(broadcast_result, tensor_y)
         print("test broadcast api ok")
 
         # test send_recv
@@ -116,11 +116,11 @@ def test_create_process_group_gloo(self):
             task = pg.send(tensor_x, pg.size() - 1, True)
         elif pg.rank() == pg.size() - 1:
             task = pg.recv(tensor_y_1, 0, True)
-            assert np.array_equal(send_recv_result_1, tensor_y_1)
+            np.testing.assert_array_equal(send_recv_result_1, tensor_y_1)
 
         if pg.rank() == 0:
             task = pg.recv(tensor_x, pg.size() - 1, True)
-            assert np.array_equal(send_recv_result_2, tensor_x)
+            np.testing.assert_array_equal(send_recv_result_2, tensor_x)
         elif pg.rank() == pg.size() - 1:
             task = pg.send(tensor_y_2, 0, True)
         print("test send_recv api ok")
@@ -159,8 +159,8 @@ def test_create_process_group_gloo(self):
         out_2 = paddle.slice(
             tensor_out, [0], [out_shape[0] // 2], [out_shape[0]]
         )
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         print("test allgather api ok\n")
 
         # test Reduce
@@ -178,7 +178,7 @@ def test_create_process_group_gloo(self):
             task = pg.reduce(tensor_y, 0)
             task.wait()
         if pg.rank() == 0:
-            assert np.array_equal(tensor_x, sum_result)
+            np.testing.assert_array_equal(tensor_x, sum_result)
         print("test reduce sum api ok\n")
 
         # test Scatter
@@ -199,9 +199,9 @@ def test_create_process_group_gloo(self):
         out1 = paddle.slice(tensor_x, [0], [0], [self.shape[0]])
         out2 = paddle.slice(tensor_x, [0], [self.shape[0]], [self.shape[0] * 2])
         if pg.rank() == 0:
-            assert np.array_equal(tensor_y, out1)
+            np.testing.assert_array_equal(tensor_y, out1)
         else:
-            assert np.array_equal(tensor_y, out2)
+            np.testing.assert_array_equal(tensor_y, out2)
         print("test scatter api ok\n")
 
         # test Gather
@@ -219,7 +219,7 @@ def test_gather(root):
             if pg.rank() == root:
                 task = pg.gather(tensor_y[root], tensor_x, root, True)
                 task.wait()
-                assert np.array_equal(tensor_x, tensor_y)
+                np.testing.assert_array_equal(tensor_x, tensor_y)
             else:
                 task = pg.gather(tensor_y[pg.rank()], tensor_x, root, True)
                 task.wait()
diff --git a/test/collective/process_group_mpi.py b/test/collective/process_group_mpi.py
index f2fc9c498b4e8..b5d9fd4b74377 100644
--- a/test/collective/process_group_mpi.py
+++ b/test/collective/process_group_mpi.py
@@ -69,10 +69,10 @@ def test_allreduce_sum(pg, shape, dtype):
     sum_result = tensor_x + tensor_y
     if pg.rank() == 0:
         task = dist.all_reduce(tensor_x)
-        assert np.array_equal(tensor_x, sum_result)
+        np.testing.assert_array_equal(tensor_x, sum_result)
     else:
         task = dist.all_reduce(tensor_y)
-        assert np.array_equal(tensor_y, sum_result)
+        np.testing.assert_array_equal(tensor_y, sum_result)
     print("test allreduce sum api ok")
 
 
@@ -91,13 +91,13 @@ def test_allreduce_max(pg, shape, dtype):
             tensor_x, dist.ReduceOp.MAX, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, max_result)
+        np.testing.assert_array_equal(tensor_x, max_result)
     else:
         task = dist.all_reduce(
             tensor_y, dist.ReduceOp.MAX, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_y, max_result)
+        np.testing.assert_array_equal(tensor_y, max_result)
     print("test allreduce max api ok")
 
 
@@ -116,13 +116,13 @@ def test_allreduce_min(pg, shape, dtype):
             tensor_x, dist.ReduceOp.MIN, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, min_result)
+        np.testing.assert_array_equal(tensor_x, min_result)
     else:
         task = dist.all_reduce(
             tensor_y, dist.ReduceOp.MIN, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_y, min_result)
+        np.testing.assert_array_equal(tensor_y, min_result)
     print("test allreduce min api ok")
 
 
@@ -141,13 +141,13 @@ def test_allreduce_prod(pg, shape, dtype):
             tensor_x, dist.ReduceOp.PROD, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, prod_result)
+        np.testing.assert_array_equal(tensor_x, prod_result)
     else:
         task = dist.all_reduce(
             tensor_y, dist.ReduceOp.PROD, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_y, prod_result)
+        np.testing.assert_array_equal(tensor_y, prod_result)
     print("test allreduce prod api ok")
 
 
@@ -164,10 +164,10 @@ def test_broadcast(pg, shape, dtype):
         task = dist.broadcast(tensor_x, 0, use_calc_stream=False)
         task.synchronize()
         assert task.is_completed()
-        assert np.array_equal(broadcast_result, tensor_x)
+        np.testing.assert_array_equal(broadcast_result, tensor_x)
     else:
         task = dist.broadcast(tensor_y, 0)
-        assert np.array_equal(broadcast_result, tensor_y)
+        np.testing.assert_array_equal(broadcast_result, tensor_y)
     print("test broadcast api ok")
 
 
@@ -205,8 +205,8 @@ def test_allgather(pg, shape, dtype):
         tensor_out = paddle.concat(tensor_out_list)
     out_1 = paddle.slice(tensor_out, [0], [0], [out_shape[0] // 2])
     out_2 = paddle.slice(tensor_out, [0], [out_shape[0] // 2], [out_shape[0]])
-    assert np.array_equal(tensor_x, out_1)
-    assert np.array_equal(tensor_y, out_2)
+    np.testing.assert_array_equal(tensor_x, out_1)
+    np.testing.assert_array_equal(tensor_y, out_2)
     print("test allgather api ok\n")
 
     if pg.rank() == 0:
@@ -219,8 +219,8 @@ def test_allgather(pg, shape, dtype):
         tensor_out = paddle.concat(tensor_out_list)
     out_1 = paddle.slice(tensor_out, [0], [0], [out_shape[0] // 2])
     out_2 = paddle.slice(tensor_out, [0], [out_shape[0] // 2], [out_shape[0]])
-    assert np.array_equal(tensor_x, out_1)
-    assert np.array_equal(tensor_y, out_2)
+    np.testing.assert_array_equal(tensor_x, out_1)
+    np.testing.assert_array_equal(tensor_y, out_2)
     print("test allgather api2 ok\n")
 
 
@@ -249,9 +249,9 @@ def test_all2all(pg, shape, dtype):
     out1_2 = paddle.slice(tensor_out1, [0], [shape[0] // 2], [shape[0]])
     out2_1 = paddle.slice(tensor_out2, [0], [0], [shape[0] // 2])
     if pg.rank() == 0:
-        assert np.array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
+        np.testing.assert_array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
     else:
-        assert np.array_equal(out2_1, raw_tensor_x_2)
+        np.testing.assert_array_equal(out2_1, raw_tensor_x_2)
     print("test alltoall api ok\n")
 
     x = np.random.random(shape).astype(dtype)
@@ -277,9 +277,9 @@ def test_all2all(pg, shape, dtype):
     out1_2 = paddle.slice(tensor_out1, [0], [shape[0] // 2], [shape[0]])
     out2_1 = paddle.slice(tensor_out2, [0], [0], [shape[0] // 2])
     if pg.rank() == 0:
-        assert np.array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
+        np.testing.assert_array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
     else:
-        assert np.array_equal(out2_1, raw_tensor_x_2)
+        np.testing.assert_array_equal(out2_1, raw_tensor_x_2)
     print("test alltoall api2 ok\n")
 
 
@@ -297,7 +297,7 @@ def test_reduce_sum(pg, shape, dtype):
         task = dist.reduce(tensor_y, 0, use_calc_stream=False)
         task.wait()
     if pg.rank() == 0:
-        assert np.array_equal(tensor_x, sum_result)
+        np.testing.assert_array_equal(tensor_x, sum_result)
     print("test reduce sum api ok\n")
 
 
@@ -316,7 +316,7 @@ def test_reduce_max(pg, shape, dtype):
             tensor_x, 0, dist.ReduceOp.MAX, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, max_result)
+        np.testing.assert_array_equal(tensor_x, max_result)
     else:
         task = dist.reduce(
             tensor_y, 0, dist.ReduceOp.MAX, use_calc_stream=False
@@ -340,7 +340,7 @@ def test_reduce_min(pg, shape, dtype):
             tensor_x, 0, dist.ReduceOp.MIN, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, min_result)
+        np.testing.assert_array_equal(tensor_x, min_result)
     else:
         task = dist.reduce(
             tensor_y, 0, dist.ReduceOp.MIN, use_calc_stream=False
@@ -364,7 +364,7 @@ def test_reduce_prod(pg, shape, dtype):
             tensor_x, 0, dist.ReduceOp.PROD, use_calc_stream=False
         )
         task.wait()
-        assert np.array_equal(tensor_x, prod_result)
+        np.testing.assert_array_equal(tensor_x, prod_result)
     else:
         task = dist.reduce(
             tensor_y, 0, dist.ReduceOp.PROD, use_calc_stream=False
@@ -391,9 +391,9 @@ def test_scatter(pg, shape, dtype):
     out1 = paddle.slice(tensor_x, [0], [0], [shape[0]])
     out2 = paddle.slice(tensor_x, [0], [shape[0]], [shape[0] * 2])
     if pg.rank() == 0:
-        assert np.array_equal(tensor_y, out1)
+        np.testing.assert_array_equal(tensor_y, out1)
     else:
-        assert np.array_equal(tensor_y, out2)
+        np.testing.assert_array_equal(tensor_y, out2)
     print("test scatter api ok\n")
 
 
@@ -411,7 +411,7 @@ def test_send_recv(pg, sub_group, shape, dtype):
     elif pg.rank() == 1:
         task = dist.recv(tensor_y, 0, group=sub_group, use_calc_stream=False)
         task.wait()
-        assert np.array_equal(tensor_y, tensor_x)
+        np.testing.assert_array_equal(tensor_y, tensor_x)
 
     print("test send api ok")
 
@@ -427,7 +427,7 @@ def test_send_recv(pg, sub_group, shape, dtype):
         task = dist.send(tensor_x, 1, group=sub_group, use_calc_stream=True)
     elif pg.rank() == 1:
         task = dist.recv(tensor_y, 0, group=sub_group, use_calc_stream=True)
-        assert np.array_equal(tensor_y, tensor_x)
+        np.testing.assert_array_equal(tensor_y, tensor_x)
 
     print("test send api ok")
 
diff --git a/test/collective/process_group_nccl.py b/test/collective/process_group_nccl.py
index 713e0a01b4abb..67815eab2bc83 100644
--- a/test/collective/process_group_nccl.py
+++ b/test/collective/process_group_nccl.py
@@ -64,10 +64,10 @@ def test_create_process_group_nccl(self):
         sum_result = tensor_x + tensor_y
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x)
-            assert np.array_equal(tensor_x, sum_result)
+            np.testing.assert_array_equal(tensor_x, sum_result)
         else:
             task = dist.all_reduce(tensor_y)
-            assert np.array_equal(tensor_y, sum_result)
+            np.testing.assert_array_equal(tensor_y, sum_result)
 
         print("test allreduce sum api ok")
 
@@ -82,10 +82,10 @@ def test_create_process_group_nccl(self):
         sum_result = tensor_x + tensor_y
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x)
-            assert np.array_equal(tensor_x, sum_result)
+            np.testing.assert_array_equal(tensor_x, sum_result)
         else:
             task = dist.all_reduce(tensor_y)
-            assert np.array_equal(tensor_y, sum_result)
+            np.testing.assert_array_equal(tensor_y, sum_result)
 
         print("test allreduce sum api with = [] ok")
 
@@ -102,11 +102,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, max_result)
+            np.testing.assert_array_equal(tensor_x, max_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, max_result)
+            np.testing.assert_array_equal(tensor_y, max_result)
 
         print("test allreduce max api ok")
 
@@ -123,11 +123,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, max_result)
+            np.testing.assert_array_equal(tensor_x, max_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, max_result)
+            np.testing.assert_array_equal(tensor_y, max_result)
 
         print("test allreduce max api with shape = [] ok")
 
@@ -144,11 +144,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, min_result)
+            np.testing.assert_array_equal(tensor_x, min_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, min_result)
+            np.testing.assert_array_equal(tensor_y, min_result)
 
         print("test allreduce min api ok")
 
@@ -165,11 +165,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, min_result)
+            np.testing.assert_array_equal(tensor_x, min_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, min_result)
+            np.testing.assert_array_equal(tensor_y, min_result)
 
         print("test allreduce min api with shape [] ok")
 
@@ -186,11 +186,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, prod_result)
+            np.testing.assert_array_equal(tensor_x, prod_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, prod_result)
+            np.testing.assert_array_equal(tensor_y, prod_result)
 
         print("test allreduce prod api ok")
 
@@ -207,11 +207,11 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.all_reduce(tensor_x, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, prod_result)
+            np.testing.assert_array_equal(tensor_x, prod_result)
         else:
             task = dist.all_reduce(tensor_y, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, prod_result)
+            np.testing.assert_array_equal(tensor_y, prod_result)
 
         print("test allreduce prod api with shape = [] ok")
 
@@ -229,11 +229,11 @@ def test_create_process_group_nccl(self):
             task.synchronize()
             paddle.device.cuda.synchronize()
             assert task.is_completed()
-            assert np.array_equal(broadcast_result, tensor_x)
+            np.testing.assert_array_equal(broadcast_result, tensor_x)
         else:
             task = dist.broadcast(tensor_y, 0)
             paddle.device.cuda.synchronize()
-            assert np.array_equal(broadcast_result, tensor_y)
+            np.testing.assert_array_equal(broadcast_result, tensor_y)
 
         print("test broadcast api ok")
 
@@ -251,11 +251,11 @@ def test_create_process_group_nccl(self):
             task.synchronize()
             paddle.device.cuda.synchronize()
             assert task.is_completed()
-            assert np.array_equal(broadcast_result, tensor_x)
+            np.testing.assert_array_equal(broadcast_result, tensor_x)
         else:
             task = dist.broadcast(tensor_y, 0)
             paddle.device.cuda.synchronize()
-            assert np.array_equal(broadcast_result, tensor_y)
+            np.testing.assert_array_equal(broadcast_result, tensor_y)
         assert tensor_y.shape == []
 
         print("test broadcast api with shape=[] ok")
@@ -298,8 +298,8 @@ def test_create_process_group_nccl(self):
         out_2 = paddle.slice(
             tensor_out, [0], [out_shape[0] // 2], [out_shape[0]]
         )
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         print("test allgather api ok\n")
 
         if pg.rank() == 0:
@@ -316,8 +316,8 @@ def test_create_process_group_nccl(self):
         out_2 = paddle.slice(
             tensor_out, [0], [out_shape[0] // 2], [out_shape[0]]
         )
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         print("test allgather api2 ok\n")
 
         # test allgather with shape = []
@@ -337,8 +337,8 @@ def test_create_process_group_nccl(self):
             paddle.device.cuda.synchronize()
         out_1 = tensor_out_list[0]
         out_2 = tensor_out_list[1]
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         print("test allgather api with shape [] ok\n")
 
         # test alltoall
@@ -371,9 +371,11 @@ def test_create_process_group_nccl(self):
         )
         out2_1 = paddle.slice(tensor_out2, [0], [0], [self.shape[0] // 2])
         if pg.rank() == 0:
-            assert np.array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
+            np.testing.assert_array_equal(
+                out1_2.numpy(), raw_tensor_y_1.numpy()
+            )
         else:
-            assert np.array_equal(out2_1, raw_tensor_x_2)
+            np.testing.assert_array_equal(out2_1, raw_tensor_x_2)
         print("test alltoall api ok\n")
 
         x = np.random.random(self.shape).astype(self.dtype)
@@ -404,9 +406,11 @@ def test_create_process_group_nccl(self):
         )
         out2_1 = paddle.slice(tensor_out2, [0], [0], [self.shape[0] // 2])
         if pg.rank() == 0:
-            assert np.array_equal(out1_2.numpy(), raw_tensor_y_1.numpy())
+            np.testing.assert_array_equal(
+                out1_2.numpy(), raw_tensor_y_1.numpy()
+            )
         else:
-            assert np.array_equal(out2_1, raw_tensor_x_2)
+            np.testing.assert_array_equal(out2_1, raw_tensor_x_2)
         print("test alltoall api2 ok\n")
 
         # test Reduce
@@ -425,7 +429,7 @@ def test_create_process_group_nccl(self):
             task.wait()
             paddle.device.cuda.synchronize()
         if pg.rank() == 0:
-            assert np.array_equal(tensor_x, sum_result)
+            np.testing.assert_array_equal(tensor_x, sum_result)
         print("test reduce sum api ok\n")
 
         # test reduce max
@@ -441,7 +445,7 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.reduce(tensor_x, 0, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, max_result)
+            np.testing.assert_array_equal(tensor_x, max_result)
         else:
             task = dist.reduce(tensor_y, 0, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
@@ -461,7 +465,7 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.reduce(tensor_x, 0, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, min_result)
+            np.testing.assert_array_equal(tensor_x, min_result)
         else:
             task = dist.reduce(tensor_y, 0, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
@@ -481,7 +485,7 @@ def test_create_process_group_nccl(self):
         if pg.rank() == 0:
             task = dist.reduce(tensor_x, 0, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_x, prod_result)
+            np.testing.assert_array_equal(tensor_x, prod_result)
         else:
             task = dist.reduce(tensor_y, 0, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
@@ -511,9 +515,9 @@ def test_create_process_group_nccl(self):
         out1 = paddle.slice(tensor_x, [0], [0], [self.shape[0]])
         out2 = paddle.slice(tensor_x, [0], [self.shape[0]], [self.shape[0] * 2])
         if pg.rank() == 0:
-            assert np.array_equal(tensor_y, out1)
+            np.testing.assert_array_equal(tensor_y, out1)
         else:
-            assert np.array_equal(tensor_y, out2)
+            np.testing.assert_array_equal(tensor_y, out2)
         print("test scatter api ok\n")
 
         # test Scatter with shape=[]
@@ -534,9 +538,9 @@ def test_create_process_group_nccl(self):
         out1 = paddle.assign(tensor_x)
         out2 = paddle.assign(tensor_x + 1)
         if pg.rank() == 0:
-            assert np.array_equal(tensor_y, out1)
+            np.testing.assert_array_equal(tensor_y, out1)
         else:
-            assert np.array_equal(tensor_y, out2), f"{tensor_y}, {out2}"
+            np.testing.assert_array_equal(tensor_y, out2)
         assert tensor_y.shape == []
         print("test scatter api with shape=[] ok\n")
 
@@ -554,7 +558,7 @@ def test_create_process_group_nccl(self):
         else:
             task = dist.recv(tensor_y, 0, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, tensor_x)
+            np.testing.assert_array_equal(tensor_y, tensor_x)
 
         print("test send api ok")
 
@@ -570,7 +574,7 @@ def test_create_process_group_nccl(self):
             task = dist.send(tensor_x, 1, sync_op=True)
         else:
             task = dist.recv(tensor_y, 0, sync_op=True)
-            assert np.array_equal(tensor_y, tensor_x)
+            np.testing.assert_array_equal(tensor_y, tensor_x)
 
         print("test send api ok")
 
diff --git a/test/collective/strategy_group.py b/test/collective/strategy_group.py
index 5197cf632b402..a70fbdf151822 100644
--- a/test/collective/strategy_group.py
+++ b/test/collective/strategy_group.py
@@ -31,7 +31,7 @@ def _check_using_all_reduce(group):
     data = paddle.to_tensor([1, 2, 3])
     result = paddle.to_tensor([2, 4, 6])
     dist.all_reduce(data, group=group)
-    assert np.array_equal(data, result)
+    np.testing.assert_array_equal(data, result)
 
 
 def _check_using_send(group, dst):
@@ -43,7 +43,7 @@ def _check_using_recv(group, src):
     result = paddle.to_tensor([1, 2, 3])
     data = paddle.to_tensor([0, 0, 0])
     dist.recv(data, src=src, group=group)
-    assert np.array_equal(data, result)
+    np.testing.assert_array_equal(data, result)
 
 
 class TestStrategyGroupAPI(unittest.TestCase):
diff --git a/test/dygraph_to_static/test_cycle_gan.py b/test/dygraph_to_static/test_cycle_gan.py
index 7974371fc0f13..19c14115afc0f 100644
--- a/test/dygraph_to_static/test_cycle_gan.py
+++ b/test/dygraph_to_static/test_cycle_gan.py
@@ -698,17 +698,13 @@ def test_train(self):
         st_out = self.train(to_static=True)
         dy_out = self.train(to_static=False)
 
-        assert_func = np.allclose
         # Note(Aurelius84): Because we disable BN on GPU,
         # but here we enhance the check on CPU by `np.array_equal`
         # which means the dy_out and st_out shall be exactly same.
         if not fluid.is_compiled_with_cuda():
-            assert_func = np.array_equal
-
-        self.assertTrue(
-            assert_func(dy_out, st_out),
-            msg=f"dy_out:\n {dy_out}\n st_out:\n{st_out}",
-        )
+            np.testing.assert_array_equal(dy_out, st_out)
+        else:
+            np.testing.assert_allclose(dy_out, st_out, rtol=1e-5, atol=1e-8)
 
 
 if __name__ == "__main__":
diff --git a/test/ir/inference/test_inference_predictor_run.py b/test/ir/inference/test_inference_predictor_run.py
index 99ba29956c5da..c6a8c5db9f3c1 100644
--- a/test/ir/inference/test_inference_predictor_run.py
+++ b/test/ir/inference/test_inference_predictor_run.py
@@ -119,7 +119,7 @@ def test_output(self):
         inorder_output = self.get_inorder_output()
         disorder_output = self.get_disorder_output()
 
-        assert np.allclose(
+        np.testing.assert_allclose(
             inorder_output.numpy().flatten(), disorder_output.numpy().flatten()
         )
 
diff --git a/test/legacy_test/test_activation_op.py b/test/legacy_test/test_activation_op.py
index c8984da8514d2..3657621614189 100644
--- a/test/legacy_test/test_activation_op.py
+++ b/test/legacy_test/test_activation_op.py
@@ -3294,9 +3294,15 @@ def test_api(self):
                 fetch_list=[out_1, out_2, res, out_6],
             )
 
-            assert np.allclose(res_1, np.power(input, 2))
-            assert np.allclose(res_2, np.power(input, 3))
-            assert np.allclose(res_6, np.power(input, 3))
+            np.testing.assert_allclose(
+                res_1, np.power(input, 2), rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_2, np.power(input, 3), rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_6, np.power(input, 3), rtol=1e-5, atol=1e-8
+            )
 
 
 def ref_stanh(x, scale_a=0.67, scale_b=1.7159):
diff --git a/test/legacy_test/test_adaptive_avg_pool2d.py b/test/legacy_test/test_adaptive_avg_pool2d.py
index a2b0066235f87..663ac74781597 100644
--- a/test/legacy_test/test_adaptive_avg_pool2d.py
+++ b/test/legacy_test/test_adaptive_avg_pool2d.py
@@ -148,15 +148,21 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_4, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
-
-            assert np.allclose(res_2, self.res_2_np)
-
-            assert np.allclose(res_3, self.res_3_np)
-
-            assert np.allclose(res_4, self.res_4_np)
-
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(
+                res_1, self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_2, self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_3, self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_4, self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_5, self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -169,36 +175,38 @@ def test_dynamic_graph(self):
             out_1 = paddle.nn.functional.adaptive_avg_pool2d(
                 x=x, output_size=[3, 3]
             )
-
             out_2 = paddle.nn.functional.adaptive_avg_pool2d(x=x, output_size=5)
-
             out_3 = paddle.nn.functional.adaptive_avg_pool2d(
                 x=x, output_size=[2, 5]
             )
-
             out_4 = paddle.nn.functional.adaptive_avg_pool2d(
                 x=x, output_size=[3, 3], data_format="NHWC"
             )
-
             out_5 = paddle.nn.functional.adaptive_avg_pool2d(
                 x=x, output_size=[None, 3]
             )
-
             out_6 = paddle.nn.functional.interpolate(
                 x=x, mode="area", size=[2, 5]
             )
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
-
-            assert np.allclose(out_2.numpy(), self.res_2_np)
-
-            assert np.allclose(out_3.numpy(), self.res_3_np)
-
-            assert np.allclose(out_4.numpy(), self.res_4_np)
-
-            assert np.allclose(out_5.numpy(), self.res_5_np)
-
-            assert np.allclose(out_6.numpy(), self.res_3_np)
+            np.testing.assert_allclose(
+                out_1.numpy(), self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_2.numpy(), self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_3.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_4.numpy(), self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_5.numpy(), self.res_5_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_6.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
 
 
 class TestAdaptiveAvgPool2DClassAPI(unittest.TestCase):
@@ -260,15 +268,21 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_4, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
-
-            assert np.allclose(res_2, self.res_2_np)
-
-            assert np.allclose(res_3, self.res_3_np)
-
-            assert np.allclose(res_4, self.res_4_np)
-
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(
+                res_1, self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_2, self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_3, self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_4, self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_5, self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -297,15 +311,21 @@ def test_dynamic_graph(self):
             )
             out_5 = adaptive_avg_pool(x=x)
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
-
-            assert np.allclose(out_2.numpy(), self.res_2_np)
-
-            assert np.allclose(out_3.numpy(), self.res_3_np)
-
-            assert np.allclose(out_4.numpy(), self.res_4_np)
-
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(
+                out_1.numpy(), self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_2.numpy(), self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_3.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_4.numpy(), self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_5.numpy(), self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
 
 class TestOutputSizeTensor(UnittestBase):
diff --git a/test/legacy_test/test_adaptive_avg_pool3d.py b/test/legacy_test/test_adaptive_avg_pool3d.py
index 99afe85996ce6..d5054ba2107af 100755
--- a/test/legacy_test/test_adaptive_avg_pool3d.py
+++ b/test/legacy_test/test_adaptive_avg_pool3d.py
@@ -169,15 +169,21 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_4, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
-
-            assert np.allclose(res_2, self.res_2_np)
-
-            assert np.allclose(res_3, self.res_3_np)
-
-            assert np.allclose(res_4, self.res_4_np)
-
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(
+                res_1, self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_2, self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_3, self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_4, self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_5, self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -209,17 +215,24 @@ def test_dynamic_graph(self):
                 x=x, mode="area", size=[2, 3, 5]
             )
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
-
-            assert np.allclose(out_2.numpy(), self.res_2_np)
-
-            assert np.allclose(out_3.numpy(), self.res_3_np)
-
-            assert np.allclose(out_4.numpy(), self.res_4_np)
-
-            assert np.allclose(out_5.numpy(), self.res_5_np)
-
-            assert np.allclose(out_6.numpy(), self.res_3_np)
+            np.testing.assert_allclose(
+                out_1.numpy(), self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_2.numpy(), self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_3.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_4.numpy(), self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_5.numpy(), self.res_5_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_6.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
 
 
 class TestAdaptiveAvgPool3DClassAPI(unittest.TestCase):
@@ -288,15 +301,21 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_4, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
-
-            assert np.allclose(res_2, self.res_2_np)
-
-            assert np.allclose(res_3, self.res_3_np)
-
-            assert np.allclose(res_4, self.res_4_np)
-
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(
+                res_1, self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_2, self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_3, self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_4, self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                res_5, self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -329,15 +348,21 @@ def test_dynamic_graph(self):
             )
             out_5 = adaptive_avg_pool(x=x)
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
-
-            assert np.allclose(out_2.numpy(), self.res_2_np)
-
-            assert np.allclose(out_3.numpy(), self.res_3_np)
-
-            assert np.allclose(out_4.numpy(), self.res_4_np)
-
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(
+                out_1.numpy(), self.res_1_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_2.numpy(), self.res_2_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_3.numpy(), self.res_3_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_4.numpy(), self.res_4_np, rtol=1e-5, atol=1e-8
+            )
+            np.testing.assert_allclose(
+                out_5.numpy(), self.res_5_np, rtol=1e-5, atol=1e-8
+            )
 
 
 if __name__ == '__main__':
diff --git a/test/legacy_test/test_adaptive_max_pool2d.py b/test/legacy_test/test_adaptive_max_pool2d.py
index 62a4dee7e58a0..104271b955257 100644
--- a/test/legacy_test/test_adaptive_max_pool2d.py
+++ b/test/legacy_test/test_adaptive_max_pool2d.py
@@ -149,15 +149,15 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
+            np.testing.assert_allclose(res_1, self.res_1_np)
 
-            assert np.allclose(res_2, self.res_2_np)
+            np.testing.assert_allclose(res_2, self.res_2_np)
 
-            assert np.allclose(res_3, self.res_3_np)
+            np.testing.assert_allclose(res_3, self.res_3_np)
 
-            # assert np.allclose(res_4, self.res_4_np)
+            # np.testing.assert_allclose(res_4, self.res_4_np)
 
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(res_5, self.res_5_np)
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -184,15 +184,15 @@ def test_dynamic_graph(self):
                 x=x, output_size=[None, 3]
             )
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
+            np.testing.assert_allclose(out_1.numpy(), self.res_1_np)
 
-            assert np.allclose(out_2.numpy(), self.res_2_np)
+            np.testing.assert_allclose(out_2.numpy(), self.res_2_np)
 
-            assert np.allclose(out_3.numpy(), self.res_3_np)
+            np.testing.assert_allclose(out_3.numpy(), self.res_3_np)
 
-            # assert np.allclose(out_4.numpy(), self.res_4_np)
+            # np.testing.assert_allclose(out_4.numpy(), self.res_4_np)
 
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(out_5.numpy(), self.res_5_np)
 
 
 class TestAdaptiveMaxPool2DClassAPI(unittest.TestCase):
@@ -255,15 +255,15 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
+            np.testing.assert_allclose(res_1, self.res_1_np)
 
-            assert np.allclose(res_2, self.res_2_np)
+            np.testing.assert_allclose(res_2, self.res_2_np)
 
-            assert np.allclose(res_3, self.res_3_np)
+            np.testing.assert_allclose(res_3, self.res_3_np)
 
-            # assert np.allclose(res_4, self.res_4_np)
+            # np.testing.assert_allclose(res_4, self.res_4_np)
 
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(res_5, self.res_5_np)
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -291,15 +291,15 @@ def test_dynamic_graph(self):
             )
             out_5 = adaptive_max_pool(x=x)
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
+            np.testing.assert_allclose(out_1.numpy(), self.res_1_np)
 
-            assert np.allclose(out_2.numpy(), self.res_2_np)
+            np.testing.assert_allclose(out_2.numpy(), self.res_2_np)
 
-            assert np.allclose(out_3.numpy(), self.res_3_np)
+            np.testing.assert_allclose(out_3.numpy(), self.res_3_np)
 
-            # assert np.allclose(out_4.numpy(), self.res_4_np)
+            # np.testing.assert_allclose(out_4.numpy(), self.res_4_np)
 
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(out_5.numpy(), self.res_5_np)
 
 
 class TestOutDtype(unittest.TestCase):
diff --git a/test/legacy_test/test_adaptive_max_pool3d.py b/test/legacy_test/test_adaptive_max_pool3d.py
index f221964097001..13eed4823d88f 100755
--- a/test/legacy_test/test_adaptive_max_pool3d.py
+++ b/test/legacy_test/test_adaptive_max_pool3d.py
@@ -170,15 +170,15 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
+            np.testing.assert_allclose(res_1, self.res_1_np)
 
-            assert np.allclose(res_2, self.res_2_np)
+            np.testing.assert_allclose(res_2, self.res_2_np)
 
-            assert np.allclose(res_3, self.res_3_np)
+            np.testing.assert_allclose(res_3, self.res_3_np)
 
-            # assert np.allclose(res_4, self.res_4_np)
+            # np.testing.assert_allclose(res_4, self.res_4_np)
 
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(res_5, self.res_5_np)
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -205,15 +205,15 @@ def test_dynamic_graph(self):
                 x=x, output_size=[None, 3, None]
             )
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
+            np.testing.assert_allclose(out_1.numpy(), self.res_1_np)
 
-            assert np.allclose(out_2.numpy(), self.res_2_np)
+            np.testing.assert_allclose(out_2.numpy(), self.res_2_np)
 
-            assert np.allclose(out_3.numpy(), self.res_3_np)
+            np.testing.assert_allclose(out_3.numpy(), self.res_3_np)
 
-            # assert np.allclose(out_4.numpy(), self.res_4_np)
+            # np.testing.assert_allclose(out_4.numpy(), self.res_4_np)
 
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(out_5.numpy(), self.res_5_np)
 
 
 class TestAdaptiveMaxPool3DClassAPI(unittest.TestCase):
@@ -280,15 +280,15 @@ def test_static_graph(self):
                 fetch_list=[out_1, out_2, out_3, out_5],
             )
 
-            assert np.allclose(res_1, self.res_1_np)
+            np.testing.assert_allclose(res_1, self.res_1_np)
 
-            assert np.allclose(res_2, self.res_2_np)
+            np.testing.assert_allclose(res_2, self.res_2_np)
 
-            assert np.allclose(res_3, self.res_3_np)
+            np.testing.assert_allclose(res_3, self.res_3_np)
 
             #     assert np.allclose(res_4, self.res_4_np)
 
-            assert np.allclose(res_5, self.res_5_np)
+            np.testing.assert_allclose(res_5, self.res_5_np)
 
     def test_dynamic_graph(self):
         for use_cuda in (
@@ -320,15 +320,15 @@ def test_dynamic_graph(self):
             )
             out_5 = adaptive_max_pool(x=x)
 
-            assert np.allclose(out_1.numpy(), self.res_1_np)
+            np.testing.assert_allclose(out_1.numpy(), self.res_1_np)
 
-            assert np.allclose(out_2.numpy(), self.res_2_np)
+            np.testing.assert_allclose(out_2.numpy(), self.res_2_np)
 
-            assert np.allclose(out_3.numpy(), self.res_3_np)
+            np.testing.assert_allclose(out_3.numpy(), self.res_3_np)
 
             #     assert np.allclose(out_4.numpy(), self.res_4_np)
 
-            assert np.allclose(out_5.numpy(), self.res_5_np)
+            np.testing.assert_allclose(out_5.numpy(), self.res_5_np)
 
 
 class TestOutDtype(unittest.TestCase):
diff --git a/test/legacy_test/test_addmm_op.py b/test/legacy_test/test_addmm_op.py
index 66a86961e885d..1f92270cbeeac 100644
--- a/test/legacy_test/test_addmm_op.py
+++ b/test/legacy_test/test_addmm_op.py
@@ -328,7 +328,9 @@ def test_api_with_dygraph(self):
             x = fluid.dygraph.to_variable(np_x)
             y = fluid.dygraph.to_variable(np_y)
             out = paddle.tensor.addmm(input, x, y)
-            assert np.allclose(np_input + np.dot(np_x, np_y), out.numpy())
+            np.testing.assert_allclose(
+                np_input + np.dot(np_x, np_y), out.numpy(), rtol=1e-5, atol=1e-8
+            )
 
 
 class TestAddMMAPI(unittest.TestCase):
diff --git a/test/legacy_test/test_batch_norm_op_v2.py b/test/legacy_test/test_batch_norm_op_v2.py
index a55c478996678..618513a0d044b 100644
--- a/test/legacy_test/test_batch_norm_op_v2.py
+++ b/test/legacy_test/test_batch_norm_op_v2.py
@@ -173,7 +173,7 @@ def compute_v2(x):
 
                     bn = paddle.nn.BatchNorm2D(shape[1])
                     eag_y = bn(paddle.to_tensor(x))
-                    assert np.allclose(eag_y.numpy(), y.numpy())
+                    np.testing.assert_allclose(eag_y.numpy(), y.numpy())
                 return y.numpy()
 
             def compute_v3(x, is_test, trainable_statistics):
@@ -351,10 +351,10 @@ def test_1d_opt(self):
             y.backward()
             y2.backward()
 
-            assert np.allclose(
+            np.testing.assert_allclose(
                 y.numpy().flatten(), y2.numpy().flatten(), atol=1e-5, rtol=1e-5
             )
-            assert np.allclose(
+            np.testing.assert_allclose(
                 bn1d.weight.grad.numpy().flatten(),
                 bn2d.weight.grad.numpy().flatten(),
                 atol=1e-5,
diff --git a/test/legacy_test/test_broadcast_to_op.py b/test/legacy_test/test_broadcast_to_op.py
index 63d3b91417512..e2da6a1117296 100644
--- a/test/legacy_test/test_broadcast_to_op.py
+++ b/test/legacy_test/test_broadcast_to_op.py
@@ -66,9 +66,9 @@ def test_api(self):
             },
             fetch_list=[out_1, out_2, out_3],
         )
-        assert np.array_equal(res_1, np.tile(input, (1, 1)))
-        assert np.array_equal(res_2, np.tile(input, (1, 1)))
-        assert np.array_equal(res_3, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_1, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_2, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_3, np.tile(input, (1, 1)))
 
     def test_api_fp16_gpu(self):
         if paddle.fluid.core.is_compiled_with_cuda():
@@ -101,9 +101,9 @@ def test_api_fp16_gpu(self):
                     },
                     fetch_list=[out_1, out_2, out_3],
                 )
-                assert np.array_equal(res_1, np.tile(input, (1, 1)))
-                assert np.array_equal(res_2, np.tile(input, (1, 1)))
-                assert np.array_equal(res_3, np.tile(input, (1, 1)))
+                np.testing.assert_array_equal(res_1, np.tile(input, (1, 1)))
+                np.testing.assert_array_equal(res_2, np.tile(input, (1, 1)))
+                np.testing.assert_array_equal(res_3, np.tile(input, (1, 1)))
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_channel_shuffle.py b/test/legacy_test/test_channel_shuffle.py
index efecb886a757f..f8b6ef1df9514 100644
--- a/test/legacy_test/test_channel_shuffle.py
+++ b/test/legacy_test/test_channel_shuffle.py
@@ -120,8 +120,8 @@ def test_static_graph_functional(self):
                 use_prune=True,
             )
 
-            assert np.allclose(res_1, self.out_1_np)
-            assert np.allclose(res_2, self.out_2_np)
+            np.testing.assert_allclose(res_1[0], self.out_1_np)
+            np.testing.assert_allclose(res_2[0], self.out_2_np)
 
     # same test between layer and functional in this op.
     def test_static_graph_layer(self):
@@ -160,8 +160,8 @@ def test_static_graph_layer(self):
                 use_prune=True,
             )
 
-            assert np.allclose(res_1, out_1_np)
-            assert np.allclose(res_2, out_2_np)
+            np.testing.assert_allclose(res_1[0], out_1_np)
+            np.testing.assert_allclose(res_2[0], out_2_np)
 
     def run_dygraph(self, groups, data_format):
         n, c, h, w = 2, 9, 4, 4
diff --git a/test/legacy_test/test_concat_op.py b/test/legacy_test/test_concat_op.py
index 1176ba32b20d4..db848d7b5cff4 100644
--- a/test/legacy_test/test_concat_op.py
+++ b/test/legacy_test/test_concat_op.py
@@ -472,9 +472,15 @@ def test_fluid_api(self):
             feed={"x_1": input_2, "x_2": input_2, "x_3": input_3},
             fetch_list=[out_1, out_2, out_3],
         )
-        assert np.array_equal(res_1, np.concatenate((input_2, input_3), axis=1))
-        assert np.array_equal(res_2, np.concatenate((input_2, input_3), axis=1))
-        assert np.array_equal(res_3, np.concatenate((input_2, input_3), axis=1))
+        np.testing.assert_array_equal(
+            res_1, np.concatenate((input_2, input_3), axis=1)
+        )
+        np.testing.assert_array_equal(
+            res_2, np.concatenate((input_2, input_3), axis=1)
+        )
+        np.testing.assert_array_equal(
+            res_3, np.concatenate((input_2, input_3), axis=1)
+        )
 
     def test_api(self):
         paddle.enable_static()
@@ -501,10 +507,18 @@ def test_api(self):
             feed={"x_1": input_2, "x_2": input_2, "x_3": input_3},
             fetch_list=[out_1, out_2, out_3, out_4],
         )
-        assert np.array_equal(res_1, np.concatenate((input_2, input_3), axis=1))
-        assert np.array_equal(res_2, np.concatenate((input_2, input_3), axis=1))
-        assert np.array_equal(res_3, np.concatenate((input_2, input_3), axis=1))
-        assert np.array_equal(res_4, np.concatenate((input_2, input_3), axis=1))
+        np.testing.assert_array_equal(
+            res_1, np.concatenate((input_2, input_3), axis=1)
+        )
+        np.testing.assert_array_equal(
+            res_2, np.concatenate((input_2, input_3), axis=1)
+        )
+        np.testing.assert_array_equal(
+            res_3, np.concatenate((input_2, input_3), axis=1)
+        )
+        np.testing.assert_array_equal(
+            res_4, np.concatenate((input_2, input_3), axis=1)
+        )
 
     def test_imperative(self):
         in1 = np.array([[1, 2, 3], [4, 5, 6]])
diff --git a/test/legacy_test/test_cuda_graph_partial_graph_static_run.py b/test/legacy_test/test_cuda_graph_partial_graph_static_run.py
index 822bef54b5a63..2e301bdbd94da 100644
--- a/test/legacy_test/test_cuda_graph_partial_graph_static_run.py
+++ b/test/legacy_test/test_cuda_graph_partial_graph_static_run.py
@@ -124,7 +124,7 @@ def test_static_mode_cuda_graph(self):
         x_data = np.random.random((3, 10)).astype('float32')
         cuda_graph_rst = self.run_with_cuda_graph(x_data)
         normal_run_rst = self.normal_run(x_data)
-        assert np.array_equal(cuda_graph_rst, normal_run_rst)
+        np.testing.assert_array_equal(cuda_graph_rst, normal_run_rst)
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_detach.py b/test/legacy_test/test_detach.py
index a17178ac28716..4b3d25a1cde73 100644
--- a/test/legacy_test/test_detach.py
+++ b/test/legacy_test/test_detach.py
@@ -172,7 +172,9 @@ def test_NoDetachMulti_DetachMulti(self):
     def test_NoDetachSingle_DetachMulti(self):
         array_no_detach_single = self.no_detach_single()
         array_detach_multi = self.detach_multi()
-        assert np.array_equal(array_no_detach_single, array_detach_multi)
+        np.testing.assert_array_equal(
+            array_no_detach_single, array_detach_multi
+        )
 
 
 class TestInplace(unittest.TestCase):
diff --git a/test/legacy_test/test_egr_python_api.py b/test/legacy_test/test_egr_python_api.py
index 8e7d07f8d2f41..10f86571af101 100644
--- a/test/legacy_test/test_egr_python_api.py
+++ b/test/legacy_test/test_egr_python_api.py
@@ -844,6 +844,7 @@ def test_set_value(self):
         ori_place = egr_tensor.place
 
         new_arr = np.random.rand(4, 16, 16, 32).astype('float32')
+
         self.assertFalse(np.array_equal(egr_tensor.numpy(), new_arr))
 
         egr_tensor.set_value(new_arr)
@@ -964,6 +965,7 @@ def test_set_value(self):
         linear = paddle.nn.Linear(1, 3)
         ori_place = linear.weight.place
         new_weight = np.ones([1, 3]).astype('float32')
+
         self.assertFalse(np.array_equal(linear.weight.numpy(), new_weight))
 
         linear.weight.set_value(new_weight)
diff --git a/test/legacy_test/test_einsum_v2.py b/test/legacy_test/test_einsum_v2.py
index 434c59b5b804e..6b4bf6958b946 100644
--- a/test/legacy_test/test_einsum_v2.py
+++ b/test/legacy_test/test_einsum_v2.py
@@ -581,7 +581,7 @@ def test_shape(self):
         A = paddle.to_tensor(np.array([1.0, 2.0]))
         A_expect = paddle.to_tensor([[1.0, 0.0], [0.0, 2.0]])
         A_actual = paddle.einsum('i->ii', A)
-        assert np.array_equal(A_expect.numpy(), A_actual.numpy())
+        np.testing.assert_array_equal(A_expect.numpy(), A_actual.numpy())
 
 
 class TestSimpleUndiagonal2(unittest.TestCase):
@@ -595,7 +595,7 @@ def test_shape(self):
         B = paddle.to_tensor(np.array([1.0, 1.0]))
         A_expect = paddle.to_tensor([[2.0, 0.0], [0.0, 4.0]])
         A_actual = paddle.einsum('i,j->ii', A, B)
-        assert np.array_equal(A_expect.numpy(), A_actual.numpy())
+        np.testing.assert_array_equal(A_expect.numpy(), A_actual.numpy())
 
 
 class TestSimpleComplexGrad(unittest.TestCase):
diff --git a/test/legacy_test/test_expand_as_v2_op.py b/test/legacy_test/test_expand_as_v2_op.py
index db866144eaf96..68c9801acb4d8 100755
--- a/test/legacy_test/test_expand_as_v2_op.py
+++ b/test/legacy_test/test_expand_as_v2_op.py
@@ -280,7 +280,7 @@ def test_api(self):
             feed={"x": input1, "target_tensor": input2},
             fetch_list=[out_1],
         )
-        assert np.array_equal(res_1[0], np.tile(input1, (2, 1, 1)))
+        np.testing.assert_array_equal(res_1[0], np.tile(input1, (2, 1, 1)))
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_expand_v2_op.py b/test/legacy_test/test_expand_v2_op.py
index 128bdda6da019..9b9f6b631a2bf 100644
--- a/test/legacy_test/test_expand_v2_op.py
+++ b/test/legacy_test/test_expand_v2_op.py
@@ -318,9 +318,9 @@ def test_api(self):
             },
             fetch_list=[out_1, out_2, out_3],
         )
-        assert np.array_equal(res_1, np.tile(input, (1, 1)))
-        assert np.array_equal(res_2, np.tile(input, (1, 1)))
-        assert np.array_equal(res_3, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_1, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_2, np.tile(input, (1, 1)))
+        np.testing.assert_array_equal(res_3, np.tile(input, (1, 1)))
 
 
 class TestExpandInferShape(unittest.TestCase):
diff --git a/test/legacy_test/test_fill_constant_op.py b/test/legacy_test/test_fill_constant_op.py
index e74b24c6e5c19..614cd29668d88 100644
--- a/test/legacy_test/test_fill_constant_op.py
+++ b/test/legacy_test/test_fill_constant_op.py
@@ -338,14 +338,30 @@ def test_api(self):
             fetch_list=[out_1, out_2, out_3, out_4, out_5, out_6, out_7, out_8],
         )
 
-        assert np.array_equal(res_1, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_2, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_3, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_4, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_5, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_6, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_7, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_8, np.full([1, 2], 1.1, dtype="float32"))
+        np.testing.assert_array_equal(
+            res_1, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_2, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_3, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_4, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_5, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_6, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_7, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_8, np.full([1, 2], 1.1, dtype="float32")
+        )
 
 
 class TestFillConstantImperative(unittest.TestCase):
@@ -369,16 +385,16 @@ def test_api(self):
             res4 = paddle.tensor.fill_constant(
                 shape=shape, dtype='int32', value=value
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res1.numpy(), np.full([1, 2], 1.1, dtype="float32")
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res2.numpy(), np.full([1, 2], 1.1, dtype="float32")
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res3.numpy(), np.full([1, 2], 1.1, dtype="float32")
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res4.numpy(), np.full([1, 2], 88, dtype="int32")
             )
 
diff --git a/test/legacy_test/test_flatten_op.py b/test/legacy_test/test_flatten_op.py
index 76f8ef4a1a462..0803db60c73c0 100644
--- a/test/legacy_test/test_flatten_op.py
+++ b/test/legacy_test/test_flatten_op.py
@@ -90,7 +90,7 @@ def test_fp16_with_gpu(self):
                         fetch_list=[y],
                     )
 
-                    assert np.array_equal(res[0].shape, [12 * 14])
+                    np.testing.assert_array_equal(res[0].shape, [12 * 14])
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_full_op.py b/test/legacy_test/test_full_op.py
index 473289447410a..9a5c95044927f 100644
--- a/test/legacy_test/test_full_op.py
+++ b/test/legacy_test/test_full_op.py
@@ -74,13 +74,27 @@ def test_api(self):
             fetch_list=[out_1, out_2, out_3, out_4, out_5, out_6, out_7],
         )
 
-        assert np.array_equal(res_1, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_2, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_3, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_4, np.full([1, 2], 1.2, dtype="float32"))
-        assert np.array_equal(res_5, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_6, np.full([1, 2], 1.1, dtype="float32"))
-        assert np.array_equal(res_7, np.full([1, 2], 1.1, dtype="float32"))
+        np.testing.assert_array_equal(
+            res_1, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_2, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_3, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_4, np.full([1, 2], 1.2, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_5, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_6, np.full([1, 2], 1.1, dtype="float32")
+        )
+        np.testing.assert_array_equal(
+            res_7, np.full([1, 2], 1.1, dtype="float32")
+        )
 
     def test_api_eager(self):
         with fluid.dygraph.base.guard():
@@ -134,18 +148,36 @@ def test_api_eager(self):
                 out_7, dtype=np.float32, fill_value=np.abs(1.1)
             )
 
-            assert np.array_equal(out_1, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_2, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_3, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_4, np.full([1, 2], 1.2, dtype="float32"))
-            assert np.array_equal(out_5, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_6, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_7, np.full([1, 2], 1.1, dtype="float32"))
-            assert np.array_equal(out_8, np.full([2], 1.1, dtype="float32"))
-            assert np.array_equal(
+            np.testing.assert_array_equal(
+                out_1, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_2, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_3, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_4, np.full([1, 2], 1.2, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_5, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_6, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_7, np.full([1, 2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
+                out_8, np.full([2], 1.1, dtype="float32")
+            )
+            np.testing.assert_array_equal(
                 out_9, np.full([2, 2, 4], 1.1, dtype="float32")
             )
-            assert np.array_equal(out_10, np.full([1, 2], 1.1, dtype="float32"))
+            np.testing.assert_array_equal(
+                out_10, np.full([1, 2], 1.1, dtype="float32")
+            )
 
 
 class TestFullOpError(unittest.TestCase):
diff --git a/test/legacy_test/test_fused_attention_pass.py b/test/legacy_test/test_fused_attention_pass.py
index 0f64bf3458a12..263ff746c710f 100644
--- a/test/legacy_test/test_fused_attention_pass.py
+++ b/test/legacy_test/test_fused_attention_pass.py
@@ -185,7 +185,9 @@ def get_rst(self, use_pass=False):
     def test_pass(self):
         fused_rst = self.get_rst(use_pass=True)
         non_fused_rst = self.get_rst()
-        assert np.allclose(fused_rst, non_fused_rst)
+        np.testing.assert_allclose(
+            fused_rst, non_fused_rst, rtol=1e-5, atol=1e-8
+        )
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_fused_feedforward_pass.py b/test/legacy_test/test_fused_feedforward_pass.py
index e72de143b37bf..1abbfec2201bd 100644
--- a/test/legacy_test/test_fused_feedforward_pass.py
+++ b/test/legacy_test/test_fused_feedforward_pass.py
@@ -165,7 +165,9 @@ def test_pass(self):
                         self.use_dropout_2 = use_dropout_2
                         ret_loss = self.get_value()
                         ret_loss_fused = self.get_value(use_pass=True)
-                        assert np.allclose(ret_loss, ret_loss_fused)
+                        np.testing.assert_allclose(
+                            ret_loss, ret_loss_fused, rtol=1e-5, atol=1e-8
+                        )
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_imperative_layer_children.py b/test/legacy_test/test_imperative_layer_children.py
index b0fb822f48c0e..bf440d511c566 100644
--- a/test/legacy_test/test_imperative_layer_children.py
+++ b/test/legacy_test/test_imperative_layer_children.py
@@ -60,8 +60,8 @@ def test_func_apply_init_weight(self):
         self.ori_y1, self.ori_y2 = self.func_apply_init_weight()
 
         # compare ori dygraph and new egr
-        assert np.array_equal(self.ori_y1.numpy(), self.new_y1.numpy())
-        assert np.array_equal(self.ori_y2.numpy(), self.new_y2.numpy())
+        np.testing.assert_array_equal(self.ori_y1.numpy(), self.new_y1.numpy())
+        np.testing.assert_array_equal(self.ori_y2.numpy(), self.new_y2.numpy())
 
 
 if __name__ == '__main__':
diff --git a/test/legacy_test/test_imperative_numpy_bridge.py b/test/legacy_test/test_imperative_numpy_bridge.py
index 58059a295539d..0adb69adcf878 100644
--- a/test/legacy_test/test_imperative_numpy_bridge.py
+++ b/test/legacy_test/test_imperative_numpy_bridge.py
@@ -44,6 +44,7 @@ def test_tensor_from_numpy(self):
             data_np[0][0] = -1
             self.assertEqual(data_np[0][0], -1)
             self.assertNotEqual(var2[0][0].numpy(), -1)
+
             self.assertFalse(np.array_equal(var2.numpy(), data_np))
 
 
diff --git a/test/legacy_test/test_layers.py b/test/legacy_test/test_layers.py
index ded9e08da74cf..8cea6edd1c501 100644
--- a/test/legacy_test/test_layers.py
+++ b/test/legacy_test/test_layers.py
@@ -637,8 +637,8 @@ def test_embeding(self):
             dy_rlt = emb2(base.to_variable(inp_word))
             dy_rlt_value = dy_rlt.numpy()
 
-        self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(dy_rlt_value, static_rlt))
+        np.testing.assert_allclose(static_rlt2[0], static_rlt)
+        np.testing.assert_allclose(dy_rlt_value[0], static_rlt)
 
         with self.dynamic_graph():
             custom_weight = np.random.randn(dict_size, 32).astype("float32")
diff --git a/test/legacy_test/test_linspace.py b/test/legacy_test/test_linspace.py
index f36a5e7c8cb1b..6468ad08c8fb5 100644
--- a/test/legacy_test/test_linspace.py
+++ b/test/legacy_test/test_linspace.py
@@ -169,7 +169,7 @@ def test_dtype(self):
             res_1, res_2, res_3 = exe.run(
                 fluid.default_main_program(), fetch_list=[out_1, out_2, out_3]
             )
-            assert np.array_equal(res_1, res_2)
+            np.testing.assert_array_equal(res_1, res_2)
 
     def test_name(self):
         with paddle_static_guard():
diff --git a/test/legacy_test/test_logspace.py b/test/legacy_test/test_logspace.py
index e68dba46fefc6..0587846bc4841 100644
--- a/test/legacy_test/test_logspace.py
+++ b/test/legacy_test/test_logspace.py
@@ -179,7 +179,7 @@ def test_dtype(self):
 
         exe = paddle.static.Executor()
         res_1, res_2 = exe.run(prog, fetch_list=[out_1, out_2])
-        assert np.array_equal(res_1, res_2)
+        np.testing.assert_array_equal(res_1, res_2)
         paddle.disable_static()
 
     def test_name(self):
diff --git a/test/legacy_test/test_lrn_op.py b/test/legacy_test/test_lrn_op.py
index ff087fa44823e..df9b1ebccf481 100644
--- a/test/legacy_test/test_lrn_op.py
+++ b/test/legacy_test/test_lrn_op.py
@@ -371,7 +371,7 @@ def test_static_fp16_gpu(self):
                         fetch_list=[y],
                     )
 
-                    assert np.array_equal(res[0].shape, input.shape)
+                    np.testing.assert_array_equal(res[0].shape, input.shape)
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_meshgrid_op.py b/test/legacy_test/test_meshgrid_op.py
index 377699e3855ec..d2f7b0c2eca89 100644
--- a/test/legacy_test/test_meshgrid_op.py
+++ b/test/legacy_test/test_meshgrid_op.py
@@ -162,8 +162,8 @@ def test_api(self):
             feed={'x': input_1, 'y': input_2},
             fetch_list=[grid_x, grid_y],
         )
-        assert np.array_equal(res_1, out_1)
-        assert np.array_equal(res_2, out_2)
+        np.testing.assert_array_equal(res_1, out_1)
+        np.testing.assert_array_equal(res_2, out_2)
 
 
 class TestMeshgridOp4(unittest.TestCase):
@@ -199,8 +199,8 @@ def test_list_input(self):
             fetch_list=[grid_x, grid_y],
         )
 
-        assert np.array_equal(res_1, out_1)
-        assert np.array_equal(res_2, out_2)
+        np.testing.assert_array_equal(res_1, out_1)
+        np.testing.assert_array_equal(res_2, out_2)
 
 
 class TestMeshgridOp5(unittest.TestCase):
@@ -236,8 +236,8 @@ def test_tuple_input(self):
             fetch_list=[grid_x, grid_y],
         )
 
-        assert np.array_equal(res_1, out_1)
-        assert np.array_equal(res_2, out_2)
+        np.testing.assert_array_equal(res_1, out_1)
+        np.testing.assert_array_equal(res_2, out_2)
 
 
 class TestMeshgridOp6(unittest.TestCase):
@@ -262,8 +262,8 @@ def test_api_with_dygraph(self):
             tensor_4 = fluid.dygraph.to_variable(input_4)
             res_3, res_4 = paddle.tensor.meshgrid(tensor_3, tensor_4)
 
-            assert np.array_equal(res_3.shape, [100, 200])
-            assert np.array_equal(res_4.shape, [100, 200])
+            np.testing.assert_array_equal(res_3.shape, [100, 200])
+            np.testing.assert_array_equal(res_4.shape, [100, 200])
 
 
 class TestMeshgridOp7(unittest.TestCase):
@@ -288,8 +288,8 @@ def test_api_with_dygraph_list_input(self):
             tensor_4 = fluid.dygraph.to_variable(input_4)
             res_3, res_4 = paddle.tensor.meshgrid([tensor_3, tensor_4])
 
-            assert np.array_equal(res_3.shape, [100, 200])
-            assert np.array_equal(res_4.shape, [100, 200])
+            np.testing.assert_array_equal(res_3.shape, [100, 200])
+            np.testing.assert_array_equal(res_4.shape, [100, 200])
 
 
 class TestMeshgridOp8(unittest.TestCase):
@@ -314,8 +314,8 @@ def test_api_with_dygraph_tuple_input(self):
             tensor_4 = fluid.dygraph.to_variable(input_4)
             res_3, res_4 = paddle.tensor.meshgrid((tensor_3, tensor_4))
 
-            assert np.array_equal(res_3.shape, [100, 200])
-            assert np.array_equal(res_4.shape, [100, 200])
+            np.testing.assert_array_equal(res_3.shape, [100, 200])
+            np.testing.assert_array_equal(res_4.shape, [100, 200])
 
 
 class TestMeshGrid_ZeroDim(TestMeshgridOp):
diff --git a/test/legacy_test/test_multiprocess_dataloader_dataset.py b/test/legacy_test/test_multiprocess_dataloader_dataset.py
index 9e2b89b12860c..7eb99ef7006e7 100755
--- a/test/legacy_test/test_multiprocess_dataloader_dataset.py
+++ b/test/legacy_test/test_multiprocess_dataloader_dataset.py
@@ -84,8 +84,8 @@ def run_main(self, num_workers, places):
                 assert label.shape == [1, 1]
                 assert isinstance(input, fluid.core.eager.Tensor)
                 assert isinstance(label, fluid.core.eager.Tensor)
-                assert np.allclose(input.numpy(), input_np[i])
-                assert np.allclose(label.numpy(), label_np[i])
+                np.testing.assert_allclose(input.numpy(), input_np[i])
+                np.testing.assert_allclose(label.numpy(), label_np[i])
 
     def test_main(self):
         places = [paddle.CPUPlace()]
@@ -109,10 +109,10 @@ def test_main(self):
             input1, label1, input2, label2 = dataset[i]
             input1_t, label1_t = dataset1[i]
             input2_t, label2_t = dataset2[i]
-            assert np.allclose(input1, input1_t)
-            assert np.allclose(label1, label1_t)
-            assert np.allclose(input2, input2_t)
-            assert np.allclose(label2, label2_t)
+            np.testing.assert_allclose(input1, input1_t)
+            np.testing.assert_allclose(label1, label1_t)
+            np.testing.assert_allclose(input2, input2_t)
+            np.testing.assert_allclose(label2, label2_t)
 
 
 class TestRandomSplitApi(unittest.TestCase):
@@ -226,12 +226,12 @@ def run_main(self, num_workers, places):
 
         idx = 0
         for image, label in iter(dataset1):
-            assert np.allclose(image, samples[idx][0])
-            assert np.allclose(label, samples[idx][1])
+            np.testing.assert_allclose(image, samples[idx][0])
+            np.testing.assert_allclose(label, samples[idx][1])
             idx += 1
         for image, label in iter(dataset2):
-            assert np.allclose(image, samples[idx][0])
-            assert np.allclose(label, samples[idx][1])
+            np.testing.assert_allclose(image, samples[idx][0])
+            np.testing.assert_allclose(label, samples[idx][1])
             idx += 1
 
     def test_main(self):
diff --git a/test/legacy_test/test_nan_to_num_op.py b/test/legacy_test/test_nan_to_num_op.py
index 2aad0ff5bdac3..a5e0bbe62f4e2 100644
--- a/test/legacy_test/test_nan_to_num_op.py
+++ b/test/legacy_test/test_nan_to_num_op.py
@@ -79,10 +79,10 @@ def test_static(self):
             exe = paddle.static.Executor(self.place)
             res = exe.run(feed={'X': x_np}, fetch_list=[out1, out2, out3, out4])
 
-        self.assertTrue(np.allclose(out1_np, res[0]))
-        self.assertTrue(np.allclose(out2_np, res[1]))
-        self.assertTrue(np.allclose(out3_np, res[2]))
-        self.assertTrue(np.allclose(out4_np, res[3]))
+        np.testing.assert_allclose(out1_np, res[0])
+        np.testing.assert_allclose(out2_np, res[1])
+        np.testing.assert_allclose(out3_np, res[2])
+        np.testing.assert_allclose(out4_np, res[3])
 
     def test_dygraph(self):
         paddle.disable_static(place=self.place)
@@ -97,23 +97,23 @@ def test_dygraph(self):
 
             out_tensor = paddle.nan_to_num(x_tensor)
             out_np = np_nan_to_num(x_np)
-            self.assertTrue(np.allclose(out_tensor.numpy(), out_np))
+            np.testing.assert_allclose(out_tensor.numpy(), out_np)
 
             out_tensor = paddle.nan_to_num(x_tensor, 1.0, None, None)
             out_np = np_nan_to_num(x_np, 1, None, None)
-            self.assertTrue(np.allclose(out_tensor.numpy(), out_np))
+            np.testing.assert_allclose(out_tensor.numpy(), out_np)
 
             out_tensor = paddle.nan_to_num(x_tensor, 1.0, 2.0, None)
             out_np = np_nan_to_num(x_np, 1, 2, None)
-            self.assertTrue(np.allclose(out_tensor.numpy(), out_np))
+            np.testing.assert_allclose(out_tensor.numpy(), out_np)
 
             out_tensor = paddle.nan_to_num(x_tensor, 1.0, None, -10.0)
             out_np = np_nan_to_num(x_np, 1, None, -10)
-            self.assertTrue(np.allclose(out_tensor.numpy(), out_np))
+            np.testing.assert_allclose(out_tensor.numpy(), out_np)
 
             out_tensor = paddle.nan_to_num(x_tensor, 1.0, 100.0, -10.0)
             out_np = np_nan_to_num(x_np, 1, 100, -10)
-            self.assertTrue(np.allclose(out_tensor.numpy(), out_np))
+            np.testing.assert_allclose(out_tensor.numpy(), out_np)
 
         paddle.enable_static()
 
@@ -128,7 +128,7 @@ def test_check_grad(self):
         dx = paddle.grad(y, x_tensor)[0].numpy()
 
         np_grad = np_nan_to_num_grad(x_np, np.ones_like(x_np))
-        self.assertTrue(np.allclose(np_grad, dx))
+        np.testing.assert_allclose(np_grad, dx)
 
         paddle.enable_static()
 
diff --git a/test/legacy_test/test_number_count_op.py b/test/legacy_test/test_number_count_op.py
index d2a1dcbf52824..07185a8dfeefc 100644
--- a/test/legacy_test/test_number_count_op.py
+++ b/test/legacy_test/test_number_count_op.py
@@ -71,7 +71,7 @@ def test_api_dygraph(self):
         paddle.disable_static()
         x = paddle.to_tensor(self.x)
         out = utils._number_count(x, self.upper_num)
-        assert np.allclose(out.numpy(), self.out)
+        np.testing.assert_allclose(out.numpy(), self.out)
 
 
 if __name__ == '__main__':
diff --git a/test/legacy_test/test_numel_op.py b/test/legacy_test/test_numel_op.py
index 9d87d242a87f8..b4b18ccbe07f7 100644
--- a/test/legacy_test/test_numel_op.py
+++ b/test/legacy_test/test_numel_op.py
@@ -120,10 +120,10 @@ def test_numel_static(self):
                 },
                 fetch_list=[out_1, out_2],
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res_1, np.array(np.size(input_1)).astype("int64")
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res_2, np.array(np.size(input_2)).astype("int64")
             )
 
@@ -135,8 +135,8 @@ def test_numel_imperative(self):
         x_2 = paddle.to_tensor(input_2)
         out_1 = paddle.numel(x_1)
         out_2 = paddle.numel(x_2)
-        assert np.array_equal(out_1.numpy().item(0), np.size(input_1))
-        assert np.array_equal(out_2.numpy().item(0), np.size(input_2))
+        np.testing.assert_array_equal(out_1.numpy().item(0), np.size(input_1))
+        np.testing.assert_array_equal(out_2.numpy().item(0), np.size(input_2))
         paddle.enable_static()
 
     def test_error(self):
diff --git a/test/legacy_test/test_pixel_shuffle_op.py b/test/legacy_test/test_pixel_shuffle_op.py
index c20ba8678d806..7c6f18479fd11 100644
--- a/test/legacy_test/test_pixel_shuffle_op.py
+++ b/test/legacy_test/test_pixel_shuffle_op.py
@@ -186,17 +186,17 @@ def test_static_graph_functional(self):
                 feed={"x": self.x_1_np},
                 fetch_list=out_1,
                 use_prune=True,
-            )
+            )[0]
 
             res_2 = exe.run(
                 fluid.default_main_program(),
                 feed={"x2": self.x_2_np},
                 fetch_list=out_2,
                 use_prune=True,
-            )
+            )[0]
 
-            assert np.allclose(res_1, self.out_1_np)
-            assert np.allclose(res_2, self.out_2_np)
+            np.testing.assert_allclose(res_1, self.out_1_np)
+            np.testing.assert_allclose(res_2, self.out_2_np)
 
     def test_api_fp16(self):
         paddle.enable_static()
@@ -226,15 +226,15 @@ def test_api_fp16(self):
                     feed={"x": self.x_1_np},
                     fetch_list=out_1,
                     use_prune=True,
-                )
+                )[0]
                 res_2 = exe.run(
                     fluid.default_main_program(),
                     feed={"x2": self.x_2_np},
                     fetch_list=out_2,
                     use_prune=True,
-                )
-                assert np.allclose(res_1, out_1_np)
-                assert np.allclose(res_2, out_2_np)
+                )[0]
+                np.testing.assert_allclose(res_1, out_1_np)
+                np.testing.assert_allclose(res_2, out_2_np)
 
     # same test between layer and functional in this op.
     def test_static_graph_layer(self):
@@ -264,17 +264,17 @@ def test_static_graph_layer(self):
                 feed={"x": self.x_1_np},
                 fetch_list=out_1,
                 use_prune=True,
-            )
+            )[0]
 
             res_2 = exe.run(
                 fluid.default_main_program(),
                 feed={"x2": self.x_2_np},
                 fetch_list=out_2,
                 use_prune=True,
-            )
+            )[0]
 
-            assert np.allclose(res_1, out_1_np)
-            assert np.allclose(res_2, out_2_np)
+            np.testing.assert_allclose(res_1, out_1_np, rtol=1e-5, atol=1e-8)
+            np.testing.assert_allclose(res_2, out_2_np, rtol=1e-5, atol=1e-8)
 
     def run_dygraph(self, up_factor, data_format):
         n, c, h, w = 2, 9, 4, 4
diff --git a/test/legacy_test/test_pixel_unshuffle.py b/test/legacy_test/test_pixel_unshuffle.py
index b2cfd457603c4..2353ca0192c7e 100644
--- a/test/legacy_test/test_pixel_unshuffle.py
+++ b/test/legacy_test/test_pixel_unshuffle.py
@@ -225,17 +225,17 @@ def test_static_graph_functional(self):
                 feed={"x": self.x_1_np},
                 fetch_list=out_1,
                 use_prune=True,
-            )
+            )[0]
 
             res_2 = exe.run(
                 fluid.default_main_program(),
                 feed={"x2": self.x_2_np},
                 fetch_list=out_2,
                 use_prune=True,
-            )
+            )[0]
 
-            assert np.allclose(res_1, self.out_1_np)
-            assert np.allclose(res_2, self.out_2_np)
+            np.testing.assert_allclose(res_1, self.out_1_np)
+            np.testing.assert_allclose(res_2, self.out_2_np)
 
     # same test between layer and functional in this op.
     def test_static_graph_layer(self):
@@ -267,17 +267,17 @@ def test_static_graph_layer(self):
                 feed={"x": self.x_1_np},
                 fetch_list=out_1,
                 use_prune=True,
-            )
+            )[0]
 
             res_2 = exe.run(
                 fluid.default_main_program(),
                 feed={"x2": self.x_2_np},
                 fetch_list=out_2,
                 use_prune=True,
-            )
+            )[0]
 
-            assert np.allclose(res_1, out_1_np)
-            assert np.allclose(res_2, out_2_np)
+            np.testing.assert_allclose(res_1, out_1_np)
+            np.testing.assert_allclose(res_2, out_2_np)
 
     def run_dygraph(self, down_factor, data_format):
         '''run_dygraph'''
diff --git a/test/legacy_test/test_pool3d_api.py b/test/legacy_test/test_pool3d_api.py
index 80f22f1467e27..a9e849fb91d41 100644
--- a/test/legacy_test/test_pool3d_api.py
+++ b/test/legacy_test/test_pool3d_api.py
@@ -391,7 +391,7 @@ def test_static_fp16_gpu(self):
                     fetch_list=[y],
                 )
 
-                assert np.array_equal(res[0].shape, [1, 2, 1, 16, 16])
+                np.testing.assert_array_equal(res[0].shape, [1, 2, 1, 16, 16])
 
     def test_static_bf16_gpu(self):
         paddle.enable_static()
@@ -421,7 +421,7 @@ def test_static_bf16_gpu(self):
                     fetch_list=[y],
                 )
 
-                assert np.array_equal(res[0].shape, [1, 2, 1, 16, 16])
+                np.testing.assert_array_equal(res[0].shape, [1, 2, 1, 16, 16])
 
 
 class TestPool3DError_API(unittest.TestCase):
diff --git a/test/legacy_test/test_prune_gate_by_capacity_op.py b/test/legacy_test/test_prune_gate_by_capacity_op.py
index e52d67185e374..d6103be8d13d5 100644
--- a/test/legacy_test/test_prune_gate_by_capacity_op.py
+++ b/test/legacy_test/test_prune_gate_by_capacity_op.py
@@ -62,7 +62,7 @@ def prune_gate_by_capacity(gate_idx, expert_count, n_expert, n_worker):
 def assert_allclose(output, expected, n_expert):
     c1 = count(output, n_expert)
     c2 = count(expected, n_expert)
-    assert np.allclose(c1, c2)
+    np.testing.assert_allclose(c1, c2)
 
 
 @unittest.skipIf(
diff --git a/test/legacy_test/test_random_routing_op.py b/test/legacy_test/test_random_routing_op.py
index e7f0412eecfef..8977e30011044 100644
--- a/test/legacy_test/test_random_routing_op.py
+++ b/test/legacy_test/test_random_routing_op.py
@@ -59,7 +59,7 @@ def test_api_dygraph(self):
         value = paddle.to_tensor(self.topk_value)
         prob = paddle.to_tensor(self.prob)
         out = utils._random_routing(x, value, prob)
-        assert np.allclose(out.numpy(), self.out)
+        np.testing.assert_allclose(out.numpy(), self.out)
 
 
 @unittest.skipIf(
diff --git a/test/legacy_test/test_reshape_op.py b/test/legacy_test/test_reshape_op.py
index 2feecb5005b14..dc85f407aceab 100755
--- a/test/legacy_test/test_reshape_op.py
+++ b/test/legacy_test/test_reshape_op.py
@@ -400,10 +400,10 @@ def _test_api(self):
             fetch_list=[out_1, out_2, out_3, out_4],
         )
 
-        assert np.array_equal(res_1, input.reshape(shape))
-        assert np.array_equal(res_2, input.reshape(shape))
-        assert np.array_equal(res_3, input.reshape([5, 10]))
-        assert np.array_equal(res_4, input.reshape(shape))
+        np.testing.assert_array_equal(res_1, input.reshape(shape))
+        np.testing.assert_array_equal(res_2, input.reshape(shape))
+        np.testing.assert_array_equal(res_3, input.reshape([5, 10]))
+        np.testing.assert_array_equal(res_4, input.reshape(shape))
 
     def test_paddle_api(self):
         self._set_paddle_api()
@@ -424,9 +424,9 @@ def test_imperative(self):
             shape_tensor = self.to_tensor(np.array([2, 5, 5]).astype("int32"))
             out_3 = self.reshape(x, shape=shape_tensor)
 
-        assert np.array_equal(out_1.numpy(), input.reshape(shape))
-        assert np.array_equal(out_2.numpy(), input.reshape([5, 10]))
-        assert np.array_equal(out_3.numpy(), input.reshape(shape))
+        np.testing.assert_array_equal(out_1.numpy(), input.reshape(shape))
+        np.testing.assert_array_equal(out_2.numpy(), input.reshape([5, 10]))
+        np.testing.assert_array_equal(out_3.numpy(), input.reshape(shape))
 
 
 class TestStaticReshape_(TestReshapeAPI):
@@ -448,9 +448,9 @@ def test_imperative(self):
             shape_tensor = self.to_tensor(np.array([2, 5, 5]).astype("int32"))
             out_3 = self.reshape(x, shape=shape_tensor)
 
-        assert np.array_equal(out_1.numpy(), input.reshape(shape))
-        assert np.array_equal(out_2.numpy(), input.reshape(shape))
-        assert np.array_equal(out_3.numpy(), input.reshape(shape))
+        np.testing.assert_array_equal(out_1.numpy(), input.reshape(shape))
+        np.testing.assert_array_equal(out_2.numpy(), input.reshape(shape))
+        np.testing.assert_array_equal(out_3.numpy(), input.reshape(shape))
 
 
 # Test Input Error
diff --git a/test/legacy_test/test_size_op.py b/test/legacy_test/test_size_op.py
index dfff90b742fca..0bb3ac64bce75 100644
--- a/test/legacy_test/test_size_op.py
+++ b/test/legacy_test/test_size_op.py
@@ -83,10 +83,10 @@ def test_size_static(self):
                 },
                 fetch_list=[out_1, out_2],
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res_1, np.array(np.size(input_1)).astype("int64")
             )
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 res_2, np.array(np.size(input_2)).astype("int64")
             )
 
@@ -98,8 +98,8 @@ def test_size_imperative(self):
         x_2 = paddle.to_tensor(input_2)
         out_1 = paddle.numel(x_1)
         out_2 = paddle.numel(x_2)
-        assert np.array_equal(out_1.numpy().item(0), np.size(input_1))
-        assert np.array_equal(out_2.numpy().item(0), np.size(input_2))
+        np.testing.assert_array_equal(out_1.numpy().item(0), np.size(input_1))
+        np.testing.assert_array_equal(out_2.numpy().item(0), np.size(input_2))
         paddle.enable_static()
 
     def test_error(self):
diff --git a/test/legacy_test/test_slice_op.py b/test/legacy_test/test_slice_op.py
index 9e6ebd6f2a186..ced40c38605a9 100644
--- a/test/legacy_test/test_slice_op.py
+++ b/test/legacy_test/test_slice_op.py
@@ -631,13 +631,13 @@ def test_1(self):
                 fetch_list=[out_1, out_2, out_3, out_4, out_5, out_6, out_7],
             )
 
-            assert np.array_equal(res_1, input[-3:3, 0:100, 2:-1, :])
-            assert np.array_equal(res_2, input[-3:3, 0:100, :, 2:-1])
-            assert np.array_equal(res_3, input[-3:3, 0:100, :, 2:-1])
-            assert np.array_equal(res_4, input[-3:3, 0:100, 2:-1, :])
-            assert np.array_equal(res_5, input[-3:3, 0:100, 2:-1, :])
-            assert np.array_equal(res_6, input[-3:3, 0:100, :, 2:-1])
-            assert np.array_equal(res_7, input[-1, 0:100, :, 2:-1])
+            np.testing.assert_array_equal(res_1, input[-3:3, 0:100, 2:-1, :])
+            np.testing.assert_array_equal(res_2, input[-3:3, 0:100, :, 2:-1])
+            np.testing.assert_array_equal(res_3, input[-3:3, 0:100, :, 2:-1])
+            np.testing.assert_array_equal(res_4, input[-3:3, 0:100, 2:-1, :])
+            np.testing.assert_array_equal(res_5, input[-3:3, 0:100, 2:-1, :])
+            np.testing.assert_array_equal(res_6, input[-3:3, 0:100, :, 2:-1])
+            np.testing.assert_array_equal(res_7, input[-1, 0:100, :, 2:-1])
 
 
 class TestSliceApiWithTensor(unittest.TestCase):
diff --git a/test/legacy_test/test_sparse_conv_op.py b/test/legacy_test/test_sparse_conv_op.py
index 7ca55d58a6a4b..2d2af3c11fcb6 100644
--- a/test/legacy_test/test_sparse_conv_op.py
+++ b/test/legacy_test/test_sparse_conv_op.py
@@ -94,7 +94,7 @@ def test_conv3d(self):
         )
         out.backward(out)
         out = paddle.sparse.coalesce(out)
-        assert np.array_equal(correct_out_values, out.values().numpy())
+        np.testing.assert_array_equal(correct_out_values, out.values().numpy())
 
     def test_subm_conv2d(self):
         indices = [[0, 0, 0, 0], [0, 0, 1, 2], [1, 3, 2, 3]]
@@ -126,7 +126,9 @@ def test_subm_conv3d(self):
         y = paddle.sparse.nn.functional.subm_conv3d(
             sparse_x, weight, key='subm_conv'
         )
-        assert np.array_equal(sparse_x.indices().numpy(), y.indices().numpy())
+        np.testing.assert_array_equal(
+            sparse_x.indices().numpy(), y.indices().numpy()
+        )
 
     def test_Conv2D(self):
         # (3, non_zero_num), 3-D:(N, H, W)
@@ -223,7 +225,7 @@ def test_SubmConv3D(self):
 
         sparse_out = subm_conv3d(sparse_input)
         # the output shape of subm_conv is same as input shape
-        assert np.array_equal(indices, sparse_out.indices().numpy())
+        np.testing.assert_array_equal(indices, sparse_out.indices().numpy())
 
         # test errors
         with self.assertRaises(ValueError):
@@ -294,14 +296,16 @@ def test_Conv3D_bias(self):
         dense_out = sp_out.to_dense()
         sp_loss = dense_out.mean()
         sp_loss.backward()
-        assert np.allclose(out.numpy(), dense_out.numpy(), atol=1e-3, rtol=1e-3)
-        assert np.allclose(
+        np.testing.assert_allclose(
+            out.numpy(), dense_out.numpy(), atol=1e-3, rtol=1e-3
+        )
+        np.testing.assert_allclose(
             conv3d.weight.grad.numpy().transpose(2, 3, 4, 1, 0),
             sp_conv3d.weight.grad.numpy(),
             atol=1e-3,
             rtol=1e-3,
         )
-        assert np.allclose(
+        np.testing.assert_allclose(
             conv3d.bias.grad.numpy(),
             sp_conv3d.bias.grad.numpy(),
             atol=1e-5,
diff --git a/test/legacy_test/test_sparse_copy_op.py b/test/legacy_test/test_sparse_copy_op.py
index 237b980650676..a97148c703fc5 100644
--- a/test/legacy_test/test_sparse_copy_op.py
+++ b/test/legacy_test/test_sparse_copy_op.py
@@ -30,7 +30,7 @@ def test_copy_sparse_coo(self):
         dense_x_2 = paddle.to_tensor(np_x_2, dtype='float32')
         coo_x_2 = dense_x_2.to_sparse_coo(2)
         coo_x_2.copy_(coo_x, True)
-        assert np.array_equal(np_values, coo_x_2.values().numpy())
+        np.testing.assert_array_equal(np_values, coo_x_2.values().numpy())
 
     def test_copy_sparse_csr(self):
         np_x = [[0, 1.0, 0], [2.0, 0, 0], [0, 3.0, 0]]
@@ -42,4 +42,4 @@ def test_copy_sparse_csr(self):
         dense_x_2 = paddle.to_tensor(np_x_2, dtype='float32')
         csr_x_2 = dense_x_2.to_sparse_csr()
         csr_x_2.copy_(csr_x, True)
-        assert np.array_equal(np_values, csr_x_2.values().numpy())
+        np.testing.assert_array_equal(np_values, csr_x_2.values().numpy())
diff --git a/test/legacy_test/test_sparse_model.py b/test/legacy_test/test_sparse_model.py
index 2b7c646b3ab49..9e71757f90342 100644
--- a/test/legacy_test/test_sparse_model.py
+++ b/test/legacy_test/test_sparse_model.py
@@ -54,13 +54,19 @@ def test(self):
         sparse_loss = sparse_out.values().mean()
         sparse_loss.backward(retain_graph=True)
 
-        assert np.allclose(dense_out.numpy(), sparse_out.to_dense().numpy())
-        assert np.allclose(x.grad.numpy(), sparse_x.grad.to_dense().numpy())
+        np.testing.assert_allclose(
+            dense_out.numpy(), sparse_out.to_dense().numpy()
+        )
+        np.testing.assert_allclose(
+            x.grad.numpy(), sparse_x.grad.to_dense().numpy()
+        )
 
         loss.backward()
         sparse_loss.backward()
 
-        assert np.allclose(x.grad.numpy(), sparse_x.grad.to_dense().numpy())
+        np.testing.assert_allclose(
+            x.grad.numpy(), sparse_x.grad.to_dense().numpy()
+        )
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_sparse_norm_op.py b/test/legacy_test/test_sparse_norm_op.py
index c17a252ee75a6..ca513ef38aa1f 100644
--- a/test/legacy_test/test_sparse_norm_op.py
+++ b/test/legacy_test/test_sparse_norm_op.py
@@ -48,7 +48,7 @@ def test(self):
 
         sparse_y = sparse_batch_norm(sparse_x)
         # compare the result with dense batch_norm
-        assert np.allclose(
+        np.testing.assert_allclose(
             dense_y.flatten().numpy(),
             sparse_y.values().flatten().numpy(),
             atol=1e-5,
@@ -57,7 +57,7 @@ def test(self):
 
         # test backward
         sparse_y.backward(sparse_y)
-        assert np.allclose(
+        np.testing.assert_allclose(
             dense_x.grad.flatten().numpy(),
             sparse_x.grad.values().flatten().numpy(),
             atol=1e-5,
@@ -85,7 +85,9 @@ def test2(self):
         dense_bn = paddle.nn.BatchNorm1D(channels)
         dense_x = dense_x.reshape((-1, dense_x.shape[-1]))
         dense_out = dense_bn(dense_x)
-        assert np.allclose(dense_out.numpy(), batch_norm_out.values().numpy())
+        np.testing.assert_allclose(
+            dense_out.numpy(), batch_norm_out.values().numpy()
+        )
         # [1, 6, 6, 6, 3]
 
     def check(self, shape):
@@ -137,7 +139,7 @@ def test_sync_batch_norm(self):
             dense_sync_bn = paddle.nn.SyncBatchNorm(2)
             x = x.reshape((-1, x.shape[-1]))
             dense_hidden = dense_sync_bn(x)
-            assert np.allclose(
+            np.testing.assert_allclose(
                 sparse_hidden.values().numpy(), dense_hidden.numpy()
             )
 
diff --git a/test/legacy_test/test_sparse_pooling_op.py b/test/legacy_test/test_sparse_pooling_op.py
index 1a031329fa584..f8d0cabd304ff 100644
--- a/test/legacy_test/test_sparse_pooling_op.py
+++ b/test/legacy_test/test_sparse_pooling_op.py
@@ -64,8 +64,10 @@ def test(self):
         dense_out.backward(dense_out)
 
         # compare with dense
-        assert np.allclose(dense_out.numpy(), out.numpy())
-        assert np.allclose(dense_x.grad.numpy(), self.dense_x.grad.numpy())
+        np.testing.assert_allclose(dense_out.numpy(), out.numpy())
+        np.testing.assert_allclose(
+            dense_x.grad.numpy(), self.dense_x.grad.numpy()
+        )
 
 
 class TestStride(TestMaxPool3DFunc):
@@ -111,7 +113,7 @@ def test(self):
         dense_out = paddle.nn.functional.max_pool3d(
             dense_x, 3, data_format='NDHWC'
         )
-        assert np.allclose(dense_out.numpy(), out.numpy())
+        np.testing.assert_allclose(dense_out.numpy(), out.numpy())
 
 
 if __name__ == "__main__":
diff --git a/test/legacy_test/test_sparse_utils_op.py b/test/legacy_test/test_sparse_utils_op.py
index 60cf3a7a5208e..2b7583db92175 100644
--- a/test/legacy_test/test_sparse_utils_op.py
+++ b/test/legacy_test/test_sparse_utils_op.py
@@ -33,17 +33,17 @@ def test_create_coo_by_tensor(self):
             dense_indices, dense_elements, dense_shape, stop_gradient=False
         )
         # test the to_string.py
-        assert np.array_equal(indices, coo.indices().numpy())
-        assert np.array_equal(values, coo.values().numpy())
+        np.testing.assert_array_equal(indices, coo.indices().numpy())
+        np.testing.assert_array_equal(values, coo.values().numpy())
 
     def test_create_coo_by_np(self):
         indices = [[0, 1, 2], [1, 2, 0]]
         values = [1.0, 2.0, 3.0]
         dense_shape = [3, 3]
         coo = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape)
-        assert np.array_equal(3, coo.nnz())
-        assert np.array_equal(indices, coo.indices().numpy())
-        assert np.array_equal(values, coo.values().numpy())
+        np.testing.assert_array_equal(3, coo.nnz())
+        np.testing.assert_array_equal(indices, coo.indices().numpy())
+        np.testing.assert_array_equal(values, coo.values().numpy())
 
     def test_create_csr_by_tensor(self):
         crows = [0, 2, 3, 5]
@@ -69,10 +69,10 @@ def test_create_csr_by_np(self):
         dense_shape = [3, 4]
         csr = paddle.sparse.sparse_csr_tensor(crows, cols, values, dense_shape)
         # test the to_string.py
-        assert np.array_equal(5, csr.nnz())
-        assert np.array_equal(crows, csr.crows().numpy())
-        assert np.array_equal(cols, csr.cols().numpy())
-        assert np.array_equal(values, csr.values().numpy())
+        np.testing.assert_array_equal(5, csr.nnz())
+        np.testing.assert_array_equal(crows, csr.crows().numpy())
+        np.testing.assert_array_equal(cols, csr.cols().numpy())
+        np.testing.assert_array_equal(values, csr.values().numpy())
 
     def test_place(self):
         place = core.CPUPlace()
@@ -132,8 +132,8 @@ def test_to_sparse_coo(self):
         values = [1.0, 2.0, 3.0, 4.0, 5.0]
         dense_x = paddle.to_tensor(x, dtype='float32', stop_gradient=False)
         out = dense_x.to_sparse_coo(2)
-        assert np.array_equal(out.indices().numpy(), indices)
-        assert np.array_equal(out.values().numpy(), values)
+        np.testing.assert_array_equal(out.indices().numpy(), indices)
+        np.testing.assert_array_equal(out.values().numpy(), values)
         # test to_sparse_coo_grad backward
         out_grad_indices = [[0, 1], [0, 1]]
         out_grad_values = [2.0, 3.0]
@@ -144,7 +144,9 @@ def test_to_sparse_coo(self):
             stop_gradient=True,
         )
         out.backward(out_grad)
-        assert np.array_equal(dense_x.grad.numpy(), out_grad.to_dense().numpy())
+        np.testing.assert_array_equal(
+            dense_x.grad.numpy(), out_grad.to_dense().numpy()
+        )
 
     def test_coo_to_dense(self):
         indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
@@ -168,7 +170,7 @@ def test_coo_to_dense(self):
             dense_tensor.backward(paddle.to_tensor(out_grad))
             # mask the out_grad by sparse_x.indices()
             correct_x_grad = [2.0, 4.0, 7.0, 9.0, 10.0]
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 correct_x_grad, sparse_x.grad.values().numpy()
             )
 
@@ -182,7 +184,7 @@ def test_coo_to_dense(self):
             sparse_x_cpu.retain_grads()
             dense_tensor_cpu = sparse_x_cpu.to_dense()
             dense_tensor_cpu.backward(paddle.to_tensor(out_grad))
-            assert np.array_equal(
+            np.testing.assert_array_equal(
                 correct_x_grad, sparse_x_cpu.grad.values().numpy()
             )
 
@@ -193,12 +195,12 @@ def test_to_sparse_csr(self):
         values = [1, 2, 3, 4, 5]
         dense_x = paddle.to_tensor(x)
         out = dense_x.to_sparse_csr()
-        assert np.array_equal(out.crows().numpy(), crows)
-        assert np.array_equal(out.cols().numpy(), cols)
-        assert np.array_equal(out.values().numpy(), values)
+        np.testing.assert_array_equal(out.crows().numpy(), crows)
+        np.testing.assert_array_equal(out.cols().numpy(), cols)
+        np.testing.assert_array_equal(out.values().numpy(), values)
 
         dense_tensor = out.to_dense()
-        assert np.array_equal(dense_tensor.numpy(), x)
+        np.testing.assert_array_equal(dense_tensor.numpy(), x)
 
     def test_coo_values_grad(self):
         indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
@@ -214,7 +216,7 @@ def test_coo_values_grad(self):
         out_grad = [2.0, 3.0, 5.0, 8.0, 9.0]
         # test coo_values_grad
         values_tensor.backward(paddle.to_tensor(out_grad))
-        assert np.array_equal(out_grad, sparse_x.grad.values().numpy())
+        np.testing.assert_array_equal(out_grad, sparse_x.grad.values().numpy())
         indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
         values = [
             [1.0, 1.0],
@@ -240,7 +242,7 @@ def test_coo_values_grad(self):
         ]
         # test coo_values_grad
         values_tensor.backward(paddle.to_tensor(out_grad))
-        assert np.array_equal(out_grad, sparse_x.grad.values().numpy())
+        np.testing.assert_array_equal(out_grad, sparse_x.grad.values().numpy())
 
     def test_sparse_coo_tensor_grad(self):
         for device in devices:
@@ -266,7 +268,9 @@ def test_sparse_coo_tensor_grad(self):
                 )
                 sparse_x.backward(sparse_out_grad)
                 correct_values_grad = [0, 3]
-                assert np.array_equal(correct_values_grad, values.grad.numpy())
+                np.testing.assert_array_equal(
+                    correct_values_grad, values.grad.numpy()
+                )
 
                 # test the non-zero values is a vector
                 values = [[1, 1], [2, 2]]
@@ -283,7 +287,9 @@ def test_sparse_coo_tensor_grad(self):
                 )
                 sparse_x.backward(sparse_out_grad)
                 correct_values_grad = [[0, 0], [3, 3]]
-                assert np.array_equal(correct_values_grad, values.grad.numpy())
+                np.testing.assert_array_equal(
+                    correct_values_grad, values.grad.numpy()
+                )
 
     def test_sparse_coo_tensor_sorted(self):
         for device in devices:
@@ -300,10 +306,12 @@ def test_sparse_coo_tensor_sorted(self):
                 sparse_x = paddle.sparse.coalesce(sparse_x)
                 indices_sorted = [[0, 1], [1, 0]]
                 values_sorted = [5.0, 1.0]
-                assert np.array_equal(
+                np.testing.assert_array_equal(
                     indices_sorted, sparse_x.indices().numpy()
                 )
-                assert np.array_equal(values_sorted, sparse_x.values().numpy())
+                np.testing.assert_array_equal(
+                    values_sorted, sparse_x.values().numpy()
+                )
 
                 # test the non-zero values is a vector
                 values = [[1.0, 1.0], [2.0, 2.0], [3.0, 3.0]]
@@ -311,16 +319,18 @@ def test_sparse_coo_tensor_sorted(self):
                 sparse_x = paddle.sparse.sparse_coo_tensor(indices, values)
                 sparse_x = paddle.sparse.coalesce(sparse_x)
                 values_sorted = [[5.0, 5.0], [1.0, 1.0]]
-                assert np.array_equal(
+                np.testing.assert_array_equal(
                     indices_sorted, sparse_x.indices().numpy()
                 )
-                assert np.array_equal(values_sorted, sparse_x.values().numpy())
+                np.testing.assert_array_equal(
+                    values_sorted, sparse_x.values().numpy()
+                )
 
     def test_batch_csr(self):
         def verify(dense_x):
             sparse_x = dense_x.to_sparse_csr()
             out = sparse_x.to_dense()
-            assert np.allclose(out.numpy(), dense_x.numpy())
+            np.testing.assert_allclose(out.numpy(), dense_x.numpy())
 
         shape = np.random.randint(low=1, high=10, size=3)
         shape = list(shape)
diff --git a/test/legacy_test/test_split_op.py b/test/legacy_test/test_split_op.py
index b6d45b4e455d8..8f7781f925cea 100644
--- a/test/legacy_test/test_split_op.py
+++ b/test/legacy_test/test_split_op.py
@@ -323,12 +323,12 @@ def test_api(self):
         )
 
         out = np.split(input_1, [2, 3], 1)
-        assert np.array_equal(res_0, out[0])
-        assert np.array_equal(res_1, out[1])
-        assert np.array_equal(res_2, out[2])
-        assert np.array_equal(res_3, out[0])
-        assert np.array_equal(res_4, out[1])
-        assert np.array_equal(res_5, out[2])
+        np.testing.assert_array_equal(res_0, out[0])
+        np.testing.assert_array_equal(res_1, out[1])
+        np.testing.assert_array_equal(res_2, out[2])
+        np.testing.assert_array_equal(res_3, out[0])
+        np.testing.assert_array_equal(res_4, out[1])
+        np.testing.assert_array_equal(res_5, out[2])
 
 
 class TestSplitOpError(unittest.TestCase):
diff --git a/test/legacy_test/test_strided_slice_op.py b/test/legacy_test/test_strided_slice_op.py
index 3181083e6968f..dde0bbf4e5c25 100644
--- a/test/legacy_test/test_strided_slice_op.py
+++ b/test/legacy_test/test_strided_slice_op.py
@@ -605,13 +605,13 @@ def test_1(self):
             },
             fetch_list=[out_1, out_2, out_3, out_4, out_5, out_6, out_7],
         )
-        assert np.array_equal(res_1, input[-3:3, 0:100, 2:-1, :])
-        assert np.array_equal(res_2, input[-3:3, 0:100, :, 2:-1])
-        assert np.array_equal(res_3, input[-3:3, 0:100, :, 2:-1])
-        assert np.array_equal(res_4, input[-3:3, 0:100, 2:-1, :])
-        assert np.array_equal(res_5, input[-3:3, 0:100:2, -1:2:-1, :])
-        assert np.array_equal(res_6, input[-3:3, 0:100:2, :, -1:2:-1])
-        assert np.array_equal(res_7, input[-1, 0:100:2, :, -1:2:-1])
+        np.testing.assert_array_equal(res_1, input[-3:3, 0:100, 2:-1, :])
+        np.testing.assert_array_equal(res_2, input[-3:3, 0:100, :, 2:-1])
+        np.testing.assert_array_equal(res_3, input[-3:3, 0:100, :, 2:-1])
+        np.testing.assert_array_equal(res_4, input[-3:3, 0:100, 2:-1, :])
+        np.testing.assert_array_equal(res_5, input[-3:3, 0:100:2, -1:2:-1, :])
+        np.testing.assert_array_equal(res_6, input[-3:3, 0:100:2, :, -1:2:-1])
+        np.testing.assert_array_equal(res_7, input[-1, 0:100:2, :, -1:2:-1])
 
     def test_dygraph_op(self):
         x = paddle.zeros(shape=[3, 4, 5, 6], dtype="float32")
diff --git a/test/legacy_test/test_tile_op.py b/test/legacy_test/test_tile_op.py
index ab8d289aeae03..282a0a52e86e8 100644
--- a/test/legacy_test/test_tile_op.py
+++ b/test/legacy_test/test_tile_op.py
@@ -387,9 +387,9 @@ def test_api(self):
             out_2 = paddle.tile(x, repeat_times=[positive_2, 3])
             out_3 = paddle.tile(x, repeat_times=repeat_times)
 
-            assert np.array_equal(out_1.numpy(), np.tile(np_x, (2, 3)))
-            assert np.array_equal(out_2.numpy(), np.tile(np_x, (2, 3)))
-            assert np.array_equal(out_3.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_1.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_2.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_3.numpy(), np.tile(np_x, (2, 3)))
 
 
 class TestTileDoubleGradCheck(unittest.TestCase):
diff --git a/test/legacy_test/test_unbind_op.py b/test/legacy_test/test_unbind_op.py
index 80c3db774a7bd..670433a84c8b5 100644
--- a/test/legacy_test/test_unbind_op.py
+++ b/test/legacy_test/test_unbind_op.py
@@ -38,8 +38,8 @@ def test_unbind(self):
             fetch_list=[out_0, out_1],
         )
 
-        assert np.array_equal(res_1, input_1[0, 0:100])
-        assert np.array_equal(res_2, input_1[1, 0:100])
+        np.testing.assert_array_equal(res_1, input_1[0, 0:100])
+        np.testing.assert_array_equal(res_2, input_1[1, 0:100])
 
     def test_unbind_static_fp16_gpu(self):
         if paddle.fluid.core.is_compiled_with_cuda():
@@ -61,8 +61,8 @@ def test_unbind_static_fp16_gpu(self):
                     fetch_list=[y],
                 )
 
-                assert np.array_equal(res[0], input[0, :])
-                assert np.array_equal(res[1], input[1, :])
+                np.testing.assert_array_equal(res[0], input[0, :])
+                np.testing.assert_array_equal(res[1], input[1, :])
 
     def test_unbind_dygraph(self):
         with fluid.dygraph.guard():
@@ -96,8 +96,8 @@ def test_layers_unbind(self):
             fetch_list=[out_0, out_1],
         )
 
-        assert np.array_equal(res_1, input_1[0, 0:100])
-        assert np.array_equal(res_2, input_1[1, 0:100])
+        np.testing.assert_array_equal(res_1, input_1[0, 0:100])
+        np.testing.assert_array_equal(res_2, input_1[1, 0:100])
 
 
 class TestUnbindOp(OpTest):
diff --git a/test/legacy_test/test_unsqueeze2_op.py b/test/legacy_test/test_unsqueeze2_op.py
index 2ba8d1204b90b..df4115eb0c57a 100755
--- a/test/legacy_test/test_unsqueeze2_op.py
+++ b/test/legacy_test/test_unsqueeze2_op.py
@@ -275,11 +275,11 @@ def test_api(self):
             fetch_list=[out_1, out_2, out_3, out_4, out_5],
         )
 
-        assert np.array_equal(res_1, input.reshape([3, 1, 1, 2, 5, 1]))
-        assert np.array_equal(res_2, input.reshape([3, 1, 1, 2, 5, 1]))
-        assert np.array_equal(res_3, input.reshape([3, 1, 1, 2, 5, 1]))
-        assert np.array_equal(res_4, input.reshape([3, 2, 5, 1]))
-        assert np.array_equal(res_5, input.reshape([3, 1, 1, 2, 5, 1]))
+        np.testing.assert_array_equal(res_1, input.reshape([3, 1, 1, 2, 5, 1]))
+        np.testing.assert_array_equal(res_2, input.reshape([3, 1, 1, 2, 5, 1]))
+        np.testing.assert_array_equal(res_3, input.reshape([3, 1, 1, 2, 5, 1]))
+        np.testing.assert_array_equal(res_4, input.reshape([3, 2, 5, 1]))
+        np.testing.assert_array_equal(res_5, input.reshape([3, 1, 1, 2, 5, 1]))
 
     def test_error(self):
         def test_axes_type():
diff --git a/test/legacy_test/test_update_loss_scaling_op.py b/test/legacy_test/test_update_loss_scaling_op.py
index 6060236c5c5c9..56ffc0499699a 100644
--- a/test/legacy_test/test_update_loss_scaling_op.py
+++ b/test/legacy_test/test_update_loss_scaling_op.py
@@ -274,14 +274,20 @@ def loss_scaling_check(self, use_cuda=True, scope=fluid.Scope()):
                     ],
                 )
 
-            assert np.array_equal(result_v[0], a_v)
-            assert np.array_equal(result_v[1], b_v)
-            assert np.array_equal(result_v[0], result_v[2])
-            assert np.array_equal(result_v[1], result_v[3])
-            assert np.array_equal(result_v[4], found_inf_v)
-            assert np.array_equal(result_v[5], prev_loss_scaling_v * incr_ratio)
-            assert np.array_equal(result_v[6], np.zeros_like(num_good_steps_v))
-            assert np.array_equal(result_v[7], np.zeros_like(num_bad_steps_v))
+            np.testing.assert_array_equal(result_v[0], a_v)
+            np.testing.assert_array_equal(result_v[1], b_v)
+            np.testing.assert_array_equal(result_v[0], result_v[2])
+            np.testing.assert_array_equal(result_v[1], result_v[3])
+            np.testing.assert_array_equal(result_v[4], found_inf_v)
+            np.testing.assert_array_equal(
+                result_v[5], prev_loss_scaling_v * incr_ratio
+            )
+            np.testing.assert_array_equal(
+                result_v[6], np.zeros_like(num_good_steps_v)
+            )
+            np.testing.assert_array_equal(
+                result_v[7], np.zeros_like(num_bad_steps_v)
+            )
 
     def loss_scaling_check_inf(self, use_cuda=True, scope=fluid.Scope()):
         with paddle_static_guard():
@@ -353,14 +359,20 @@ def loss_scaling_check_inf(self, use_cuda=True, scope=fluid.Scope()):
                         num_bad_steps,
                     ],
                 )
-            assert np.array_equal(result_v[0], np.zeros_like(a_v))
-            assert np.array_equal(result_v[1], np.zeros_like(b_v))
-            assert np.array_equal(result_v[2], np.zeros_like(a_v))
-            assert np.array_equal(result_v[3], np.zeros_like(b_v))
-            assert np.array_equal(result_v[4], found_inf_v)
-            assert np.array_equal(result_v[5], prev_loss_scaling_v * decr_ratio)
-            assert np.array_equal(result_v[6], np.zeros_like(num_good_steps_v))
-            assert np.array_equal(result_v[7], np.zeros_like(num_bad_steps_v))
+            np.testing.assert_array_equal(result_v[0], np.zeros_like(a_v))
+            np.testing.assert_array_equal(result_v[1], np.zeros_like(b_v))
+            np.testing.assert_array_equal(result_v[2], np.zeros_like(a_v))
+            np.testing.assert_array_equal(result_v[3], np.zeros_like(b_v))
+            np.testing.assert_array_equal(result_v[4], found_inf_v)
+            np.testing.assert_array_equal(
+                result_v[5], prev_loss_scaling_v * decr_ratio
+            )
+            np.testing.assert_array_equal(
+                result_v[6], np.zeros_like(num_good_steps_v)
+            )
+            np.testing.assert_array_equal(
+                result_v[7], np.zeros_like(num_bad_steps_v)
+            )
 
     def test_loss_scaling_cpu(self):
         with paddle_static_guard():
diff --git a/test/legacy_test/test_where_op.py b/test/legacy_test/test_where_op.py
index b68afe65c3a08..aa03f7276c1b3 100644
--- a/test/legacy_test/test_where_op.py
+++ b/test/legacy_test/test_where_op.py
@@ -166,17 +166,17 @@ def test_api(self, use_cuda=False):
                             feed={'cond': self.cond, 'x': self.x, 'y': self.y},
                             fetch_list=fetch_list,
                         )
-                        assert np.array_equal(out[0], self.out)
+                        np.testing.assert_array_equal(out[0], self.out)
                         if x_stop_gradient is False:
-                            assert np.array_equal(
+                            np.testing.assert_array_equal(
                                 out[2], self.ref_x_backward(out[1])
                             )
                             if y.stop_gradient is False:
-                                assert np.array_equal(
+                                np.testing.assert_array_equal(
                                     out[3], self.ref_y_backward(out[1])
                                 )
                         elif y.stop_gradient is False:
-                            assert np.array_equal(
+                            np.testing.assert_array_equal(
                                 out[2], self.ref_y_backward(out[1])
                             )
 
@@ -202,7 +202,9 @@ def test_api_broadcast(self, use_cuda=False):
                     feed={'x': x_i, 'y': y_i},
                     fetch_list=[result],
                 )
-                assert np.array_equal(out[0], np.where((x_i > 1), x_i, y_i))
+                np.testing.assert_array_equal(
+                    out[0], np.where((x_i > 1), x_i, y_i)
+                )
 
     def test_scalar(self):
         paddle.enable_static()
@@ -228,7 +230,7 @@ def test_scalar(self):
                     fetch_list=[result],
                 )
                 expect = np.where(cond_data, x_data, y_data)
-                assert np.array_equal(out[0], expect)
+                np.testing.assert_array_equal(out[0], expect)
 
     def __test_where_with_broadcast_static(self, cond_shape, x_shape, y_shape):
         paddle.enable_static()
@@ -262,7 +264,7 @@ def __test_where_with_broadcast_static(self, cond_shape, x_shape, y_shape):
                     fetch_list=[result],
                 )
                 expect = np.where(cond_data, x_data, y_data)
-                assert np.array_equal(out[0], expect)
+                np.testing.assert_array_equal(out[0], expect)
 
     def test_static_api_broadcast_1(self):
         cond_shape = [2, 4]
@@ -323,7 +325,9 @@ def test_api(self):
             y = fluid.dygraph.to_variable(y_i)
             cond = fluid.dygraph.to_variable(cond_i)
             out = paddle.where(cond, x, y)
-            assert np.array_equal(out.numpy(), np.where(cond_i, x_i, y_i))
+            np.testing.assert_array_equal(
+                out.numpy(), np.where(cond_i, x_i, y_i)
+            )
 
     def test_scalar(self):
         with fluid.dygraph.guard():
@@ -332,7 +336,7 @@ def test_scalar(self):
             y = 2.0
             cond = fluid.dygraph.to_variable(cond_i)
             out = paddle.where(cond, x, y)
-            assert np.array_equal(out.numpy(), np.where(cond_i, x, y))
+            np.testing.assert_array_equal(out.numpy(), np.where(cond_i, x, y))
 
     def __test_where_with_broadcast_dygraph(self, cond_shape, a_shape, b_shape):
         with fluid.dygraph.guard():
diff --git a/test/mkldnn/check_flags_mkldnn_ops_on_off.py b/test/mkldnn/check_flags_mkldnn_ops_on_off.py
index 188f70bca1215..f00f3967225ca 100644
--- a/test/mkldnn/check_flags_mkldnn_ops_on_off.py
+++ b/test/mkldnn/check_flags_mkldnn_ops_on_off.py
@@ -54,7 +54,7 @@ def check():
         np_res = np.add(a_np, b_np)
         np_res = np.matmul(np_res, np.transpose(b_np, (0, 2, 1)))
         np_res = np.maximum(np_res, 0)
-    assert np.allclose(res1.numpy(), np_res, atol=1e-3)
+    np.testing.assert_allclose(res1.numpy(), np_res, atol=1e-3)
 
 
 if __name__ == '__main__':
diff --git a/test/mkldnn/check_flags_use_mkldnn.py b/test/mkldnn/check_flags_use_mkldnn.py
index 4c9dd32e54443..07b4829743cd6 100644
--- a/test/mkldnn/check_flags_use_mkldnn.py
+++ b/test/mkldnn/check_flags_use_mkldnn.py
@@ -38,7 +38,7 @@ def check():
         a = fluid.dygraph.to_variable(a_np)
         res1 = func(a)
         res2 = np.maximum(a_np, 0)
-    assert np.array_equal(res1.numpy(), res2)
+    np.testing.assert_array_equal(res1.numpy(), res2)
 
 
 if __name__ == '__main__':
diff --git a/test/quantization/test_quant2_int8_mkldnn_pass.py b/test/quantization/test_quant2_int8_mkldnn_pass.py
index 61c700d23b7f4..e51da1db81ba8 100644
--- a/test/quantization/test_quant2_int8_mkldnn_pass.py
+++ b/test/quantization/test_quant2_int8_mkldnn_pass.py
@@ -92,7 +92,7 @@ def test_dequantize_op_weights(self):
             param.set(self.variables_mul["mul_weights"], self.place)
             qpass._dequantize_op_weights(graph, op_node, "Y", "Out")
 
-            assert np.allclose(
+            np.testing.assert_allclose(
                 self.scope.find_var("mul_weights").get_tensor(),
                 [
                     [
diff --git a/test/xpu/process_group_bkcl.py b/test/xpu/process_group_bkcl.py
index 9c9b88862feab..e78b0bd3d98da 100644
--- a/test/xpu/process_group_bkcl.py
+++ b/test/xpu/process_group_bkcl.py
@@ -86,11 +86,11 @@ def test_create_process_group_bkcl(self):
             # XPU don't support event query by now, so just use sync op here
             task = dist.broadcast(tensor_x, 0)
             paddle.device.xpu.synchronize()
-            assert np.array_equal(broadcast_result, tensor_x)
+            np.testing.assert_array_equal(broadcast_result, tensor_x)
         else:
             task = dist.broadcast(tensor_y, 0)
             paddle.device.xpu.synchronize()
-            assert np.array_equal(broadcast_result, tensor_y)
+            np.testing.assert_array_equal(broadcast_result, tensor_y)
 
         sys.stdout.write(f"rank {pg.rank()}: test broadcast api ok\n")
 
@@ -132,8 +132,8 @@ def test_create_process_group_bkcl(self):
         out_2 = paddle.slice(
             tensor_out, [0], [out_shape[0] // 2], [out_shape[0]]
         )
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         sys.stdout.write(f"rank {pg.rank()}: test allgather api ok\n")
 
         if pg.rank() == 0:
@@ -150,8 +150,8 @@ def test_create_process_group_bkcl(self):
         out_2 = paddle.slice(
             tensor_out, [0], [out_shape[0] // 2], [out_shape[0]]
         )
-        assert np.array_equal(tensor_x, out_1)
-        assert np.array_equal(tensor_y, out_2)
+        np.testing.assert_array_equal(tensor_x, out_1)
+        np.testing.assert_array_equal(tensor_y, out_2)
         sys.stdout.write(f"rank {pg.rank()}: test allgather api2 ok\n")
 
         # test Reduce
@@ -171,8 +171,8 @@ def test_create_process_group_bkcl(self):
             task.wait()
             paddle.device.xpu.synchronize()
         if pg.rank() == 0:
-            assert np.array_equal(tensor_x, sum_result)
-        assert np.array_equal(tensor_y, old_tensor_y)
+            np.testing.assert_array_equal(tensor_x, sum_result)
+        np.testing.assert_array_equal(tensor_y, old_tensor_y)
         sys.stdout.write(f"rank {pg.rank()}: test reduce sum api ok\n")
 
         # test reduce_scatter
@@ -196,9 +196,9 @@ def test_create_process_group_bkcl(self):
             task.wait()
         paddle.device.xpu.synchronize()
         if pg.rank() == 0:
-            assert np.array_equal(need_result0, tensor_out)
+            np.testing.assert_array_equal(need_result0, tensor_out)
         else:
-            assert np.array_equal(need_result1, tensor_out)
+            np.testing.assert_array_equal(need_result1, tensor_out)
         sys.stdout.write(f"rank {pg.rank()}: test reduce_scatter sum api ok\n")
 
         # test send async api
@@ -215,7 +215,7 @@ def test_create_process_group_bkcl(self):
         else:
             task = dist.recv(tensor_y, 0, sync_op=False)
             task.wait()
-            assert np.array_equal(tensor_y, tensor_x)
+            np.testing.assert_array_equal(tensor_y, tensor_x)
 
         # test send sync api
         # rank 0
@@ -229,7 +229,7 @@ def test_create_process_group_bkcl(self):
             task = dist.send(tensor_x, 1, sync_op=True)
         else:
             task = dist.recv(tensor_y, 0, sync_op=True)
-            assert np.array_equal(tensor_y, tensor_x)
+            np.testing.assert_array_equal(tensor_y, tensor_x)
 
         # test send 0-d tensor
         # rank 0
diff --git a/test/xpu/test_expand_as_v2_op_xpu.py b/test/xpu/test_expand_as_v2_op_xpu.py
index 1843748e8ae0c..41f345091054c 100644
--- a/test/xpu/test_expand_as_v2_op_xpu.py
+++ b/test/xpu/test_expand_as_v2_op_xpu.py
@@ -147,7 +147,7 @@ def test_api(self):
             feed={"x": x_np, "target_tensor": y_np},
             fetch_list=[out_1],
         )
-        assert np.array_equal(res_1[0], np.tile(x_np, (2, 1, 1)))
+        np.testing.assert_array_equal(res_1[0], np.tile(x_np, (2, 1, 1)))
 
 
 support_types = get_xpu_op_support_types('expand_as_v2')
diff --git a/test/xpu/test_expand_v2_op_xpu.py b/test/xpu/test_expand_v2_op_xpu.py
index 9d869d14b32e2..ad5397ff3bcb2 100644
--- a/test/xpu/test_expand_v2_op_xpu.py
+++ b/test/xpu/test_expand_v2_op_xpu.py
@@ -226,9 +226,9 @@ def test_static(self):
                 fetch_list=[out_1, out_2, out_3],
             )
 
-            assert np.array_equal(res_1, np.tile(input, (1, 1)))
-            assert np.array_equal(res_2, np.tile(input, (1, 1)))
-            assert np.array_equal(res_3, np.tile(input, (1, 1)))
+            np.testing.assert_array_equal(res_1, np.tile(input, (1, 1)))
+            np.testing.assert_array_equal(res_2, np.tile(input, (1, 1)))
+            np.testing.assert_array_equal(res_3, np.tile(input, (1, 1)))
 
 
 support_types = get_xpu_op_support_types('expand_v2')
diff --git a/test/xpu/test_masked_select_op_xpu.py b/test/xpu/test_masked_select_op_xpu.py
index 3c29714a3d3b8..f2ed82cd1e8d7 100644
--- a/test/xpu/test_masked_select_op_xpu.py
+++ b/test/xpu/test_masked_select_op_xpu.py
@@ -86,7 +86,7 @@ def test_imperative_mode(self):
         mask = paddle.to_tensor(np_mask)
         out = paddle.masked_select(x, mask)
         np_out = np_masked_select(np_x, np_mask)
-        self.assertEqual(np.allclose(out.numpy(), np_out), True)
+        np.testing.assert_allclose(out.numpy(), np_out)
         paddle.enable_static()
 
     def test_static_mode(self):
diff --git a/test/xpu/test_sparse_utils_op_xpu.py b/test/xpu/test_sparse_utils_op_xpu.py
index 37e0d39130ffa..5a282563447a6 100644
--- a/test/xpu/test_sparse_utils_op_xpu.py
+++ b/test/xpu/test_sparse_utils_op_xpu.py
@@ -29,17 +29,17 @@ def test_create_coo_by_tensor(self):
         coo = paddle.sparse.sparse_coo_tensor(
             dense_indices, dense_elements, dense_shape, stop_gradient=False
         )
-        assert np.array_equal(indices, coo.indices().numpy())
-        assert np.array_equal(values, coo.values().numpy())
+        np.testing.assert_array_equal(indices, coo.indices().numpy())
+        np.testing.assert_array_equal(values, coo.values().numpy())
 
     def test_create_coo_by_np(self):
         indices = [[0, 1, 2], [1, 2, 0]]
         values = [1.0, 2.0, 3.0]
         dense_shape = [3, 3]
         coo = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape)
-        assert np.array_equal(3, coo.nnz())
-        assert np.array_equal(indices, coo.indices().numpy())
-        assert np.array_equal(values, coo.values().numpy())
+        np.testing.assert_array_equal(3, coo.nnz())
+        np.testing.assert_array_equal(indices, coo.indices().numpy())
+        np.testing.assert_array_equal(values, coo.values().numpy())
 
     def test_place(self):
         indices = [[0, 1], [0, 1]]
diff --git a/test/xpu/test_tile_op_xpu.py b/test/xpu/test_tile_op_xpu.py
index 2e661199a0928..ae2e1b2f0d7dd 100644
--- a/test/xpu/test_tile_op_xpu.py
+++ b/test/xpu/test_tile_op_xpu.py
@@ -219,9 +219,9 @@ def test_api(self):
             out_2 = paddle.tile(x, repeat_times=[positive_2, 3])
             out_3 = paddle.tile(x, repeat_times=repeat_times)
 
-            assert np.array_equal(out_1.numpy(), np.tile(np_x, (2, 3)))
-            assert np.array_equal(out_2.numpy(), np.tile(np_x, (2, 3)))
-            assert np.array_equal(out_3.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_1.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_2.numpy(), np.tile(np_x, (2, 3)))
+            np.testing.assert_array_equal(out_3.numpy(), np.tile(np_x, (2, 3)))
 
 
 class TestTileAPI_ZeroDim(unittest.TestCase):
diff --git a/test/xpu/test_unbind_op_xpu.py b/test/xpu/test_unbind_op_xpu.py
index fa77e80fb6806..3ec10511a7e90 100644
--- a/test/xpu/test_unbind_op_xpu.py
+++ b/test/xpu/test_unbind_op_xpu.py
@@ -50,8 +50,8 @@ def test_unbind(self):
                 fetch_list=[out_0, out_1],
             )
 
-            assert np.array_equal(res_1, input_1[0, 0:100])
-            assert np.array_equal(res_2, input_1[1, 0:100])
+            np.testing.assert_array_equal(res_1, input_1[0, 0:100])
+            np.testing.assert_array_equal(res_2, input_1[1, 0:100])
 
         def test_unbind_dygraph(self):
             with fluid.dygraph.guard():
@@ -89,8 +89,8 @@ def test_layers_unbind(self):
                 fetch_list=[out_0, out_1],
             )
 
-            assert np.array_equal(res_1, input_1[0, 0:100])
-            assert np.array_equal(res_2, input_1[1, 0:100])
+            np.testing.assert_array_equal(res_1, input_1[0, 0:100])
+            np.testing.assert_array_equal(res_2, input_1[1, 0:100])
 
     class TestUnbindOp(XPUOpTest):
         def initParameters(self):
diff --git a/test/xpu/test_update_loss_scaling_op_xpu.py b/test/xpu/test_update_loss_scaling_op_xpu.py
index 86e6aac6badb5..c8e398a3d7782 100644
--- a/test/xpu/test_update_loss_scaling_op_xpu.py
+++ b/test/xpu/test_update_loss_scaling_op_xpu.py
@@ -174,14 +174,20 @@ def loss_scaling_check(self, scope=fluid.Scope()):
                         num_bad_steps,
                     ],
                 )
-            assert np.array_equal(result_v[0], a_v)
-            assert np.array_equal(result_v[1], b_v)
-            assert np.array_equal(result_v[0], result_v[2])
-            assert np.array_equal(result_v[1], result_v[3])
-            assert np.array_equal(result_v[4], found_inf_v)
-            assert np.array_equal(result_v[5], prev_loss_scaling_v * incr_ratio)
-            assert np.array_equal(result_v[6], np.zeros_like(num_good_steps_v))
-            assert np.array_equal(result_v[7], np.zeros_like(num_bad_steps_v))
+            np.testing.assert_array_equal(result_v[0], a_v)
+            np.testing.assert_array_equal(result_v[1], b_v)
+            np.testing.assert_array_equal(result_v[0], result_v[2])
+            np.testing.assert_array_equal(result_v[1], result_v[3])
+            np.testing.assert_array_equal(result_v[4], found_inf_v)
+            np.testing.assert_array_equal(
+                result_v[5], prev_loss_scaling_v * incr_ratio
+            )
+            np.testing.assert_array_equal(
+                result_v[6], np.zeros_like(num_good_steps_v)
+            )
+            np.testing.assert_array_equal(
+                result_v[7], np.zeros_like(num_bad_steps_v)
+            )
 
         def loss_scaling_check_inf(self, use_cuda=True, scope=fluid.Scope()):
             a = paddle.static.data(
@@ -252,14 +258,20 @@ def loss_scaling_check_inf(self, use_cuda=True, scope=fluid.Scope()):
                         num_bad_steps,
                     ],
                 )
-            assert np.array_equal(result_v[0], np.zeros_like(a_v))
-            assert np.array_equal(result_v[1], np.zeros_like(b_v))
-            assert np.array_equal(result_v[2], np.zeros_like(a_v))
-            assert np.array_equal(result_v[3], np.zeros_like(b_v))
-            assert np.array_equal(result_v[4], found_inf_v)
-            assert np.array_equal(result_v[5], prev_loss_scaling_v * decr_ratio)
-            assert np.array_equal(result_v[6], np.zeros_like(num_good_steps_v))
-            assert np.array_equal(result_v[7], np.zeros_like(num_bad_steps_v))
+            np.testing.assert_array_equal(result_v[0], np.zeros_like(a_v))
+            np.testing.assert_array_equal(result_v[1], np.zeros_like(b_v))
+            np.testing.assert_array_equal(result_v[2], np.zeros_like(a_v))
+            np.testing.assert_array_equal(result_v[3], np.zeros_like(b_v))
+            np.testing.assert_array_equal(result_v[4], found_inf_v)
+            np.testing.assert_array_equal(
+                result_v[5], prev_loss_scaling_v * decr_ratio
+            )
+            np.testing.assert_array_equal(
+                result_v[6], np.zeros_like(num_good_steps_v)
+            )
+            np.testing.assert_array_equal(
+                result_v[7], np.zeros_like(num_bad_steps_v)
+            )
 
         def test_loss_scaling(self):
             main = fluid.Program()
diff --git a/test/xpu/test_where_op_xpu.py b/test/xpu/test_where_op_xpu.py
index 8dd7500517aed..13ec8c8c446a7 100644
--- a/test/xpu/test_where_op_xpu.py
+++ b/test/xpu/test_where_op_xpu.py
@@ -132,18 +132,18 @@ def test_api(self):
                         feed={'cond': self.cond, 'x': self.x, 'y': self.y},
                         fetch_list=fetch_list,
                     )
-                    assert np.array_equal(out[0], self.out)
+                    np.testing.assert_array_equal(out[0], self.out)
 
                     if x_stop_gradient is False:
-                        assert np.array_equal(
+                        np.testing.assert_array_equal(
                             out[2], self.ref_x_backward(out[1])
                         )
                         if y.stop_gradient is False:
-                            assert np.array_equal(
+                            np.testing.assert_array_equal(
                                 out[3], self.ref_y_backward(out[1])
                             )
                     elif y.stop_gradient is False:
-                        assert np.array_equal(
+                        np.testing.assert_array_equal(
                             out[2], self.ref_y_backward(out[1])
                         )
 
@@ -165,7 +165,7 @@ def test_api_broadcast(self, use_cuda=False):
             out = exe.run(
                 train_prog, feed={'x': x_i, 'y': y_i}, fetch_list=[result]
             )
-            assert np.array_equal(out[0], np.where(x_i > 1, x_i, y_i))
+            np.testing.assert_array_equal(out[0], np.where(x_i > 1, x_i, y_i))
 
 
 class TestWhereDygraphAPI(unittest.TestCase):
@@ -178,7 +178,9 @@ def test_api(self):
             y = fluid.dygraph.to_variable(y_i)
             cond = fluid.dygraph.to_variable(cond_i)
             out = paddle.where(cond, x, y)
-            assert np.array_equal(out.numpy(), np.where(cond_i, x_i, y_i))
+            np.testing.assert_array_equal(
+                out.numpy(), np.where(cond_i, x_i, y_i)
+            )
 
 
 if __name__ == '__main__':