diff --git a/test/legacy_test/test_adadelta_op.py b/test/legacy_test/test_adadelta_op.py
index 9202d6e2aa80fa..c7ae043fbac6f0 100644
--- a/test/legacy_test/test_adadelta_op.py
+++ b/test/legacy_test/test_adadelta_op.py
@@ -236,7 +236,7 @@ def test_adadelta_dygraph(self):
         adam.clear_gradients()
 
 
-class TestAdadeltaOpMultiPrecison(unittest.TestCase):
+class TestAdadeltaOpMultiPrecision(unittest.TestCase):
     def _test_adadelta_op_dygraph_place_amp(self, place, use_amp=False):
         import paddle
 
diff --git a/test/legacy_test/test_adagrad_op.py b/test/legacy_test/test_adagrad_op.py
index b1aab3c903248b..4d356e89d4784c 100644
--- a/test/legacy_test/test_adagrad_op.py
+++ b/test/legacy_test/test_adagrad_op.py
@@ -130,7 +130,7 @@ def check_with_place(self, place):
         param_array = np.full((height, row_numel), 5.0).astype("float32")
         param.set(param_array, place)
 
-        # create and initialize LeraningRate Variable
+        # create and initialize LearningRate Variable
         lr = scope.var('LearningRate').get_tensor()
         lr_array = np.full((1), 2.0).astype("float32")
         lr.set(lr_array, place)
@@ -211,7 +211,7 @@ def test_sparse_adagrad(self):
             self.check_with_place(place)
 
 
-class TestAdagradOpMultiPrecison(unittest.TestCase):
+class TestAdagradOpMultiPrecision(unittest.TestCase):
     def _test_adagrad_op_dygraph_place_amp(self, place, use_amp=False):
         import paddle
 
diff --git a/test/legacy_test/test_adam_op.py b/test/legacy_test/test_adam_op.py
index 9a4e3c15553b54..0fc8c80190814b 100644
--- a/test/legacy_test/test_adam_op.py
+++ b/test/legacy_test/test_adam_op.py
@@ -758,7 +758,7 @@ def test_adam_op_with_state_dict(self):
         state_dict = adam.state_dict()
         adam.set_state_dict(state_dict)
 
-        # leanrning_rate is Tensor
+        # learning_rate is Tensor
         with self.assertRaises(TypeError):
             learning_rate = np.array([0.01]).astype("float32")
             learning_rate = paddle.to_tensor(learning_rate)
diff --git a/test/legacy_test/test_adamax_op.py b/test/legacy_test/test_adamax_op.py
index 4087e753982669..48549d6275a9f4 100644
--- a/test/legacy_test/test_adamax_op.py
+++ b/test/legacy_test/test_adamax_op.py
@@ -239,7 +239,7 @@ def test_adamax_op_invalid_input(self):
             )
 
 
-class TestAdamaxOpMultiPrecison(unittest.TestCase):
+class TestAdamaxOpMultiPrecision(unittest.TestCase):
     def _test_adamax_op_dygraph_place_amp(self, place, use_amp=False):
         import paddle
 
diff --git a/test/legacy_test/test_adamw_op.py b/test/legacy_test/test_adamw_op.py
index 752e8076d3b144..1c901e8d4baf57 100644
--- a/test/legacy_test/test_adamw_op.py
+++ b/test/legacy_test/test_adamw_op.py
@@ -404,7 +404,7 @@ def test_adamw_op_dygraph_bypassing_step(self):
             adam.clear_gradients()
 
 
-class TestAdamWOpMultiPrecisonWithMainGrad(unittest.TestCase):
+class TestAdamWOpMultiPrecisionWithMainGrad(unittest.TestCase):
     def _test_adamw_op_dygraph_place_amp_with_maingrad(
         self, place, shape, use_main_grad
     ):
@@ -543,7 +543,7 @@ def test_main(self):
                     )
 
 
-class TestAdamWOpMultiPrecison(unittest.TestCase):
+class TestAdamWOpMultiPrecision(unittest.TestCase):
     def _test_adamw_op_dygraph_place_amp(self, place, use_amp=False):
         paddle.disable_static()
         paddle.seed(10)
diff --git a/test/legacy_test/test_add_n_op.py b/test/legacy_test/test_add_n_op.py
index e543b4f05c74bc..60981b9cd02f0a 100644
--- a/test/legacy_test/test_add_n_op.py
+++ b/test/legacy_test/test_add_n_op.py
@@ -70,15 +70,15 @@ def test_add_n_api(self):
         if not paddle.is_compiled_with_cuda():
             return
         dtypes = ['float32', 'complex64', 'complex128']
-        for dtyte in dtypes:
-            if dtyte == 'complex64' or dtyte == 'complex128':
+        for dtype in dtypes:
+            if dtype == 'complex64' or dtype == 'complex128':
                 self.x_np = (
                     np.random.random([self.l, 16, 256])
                     + 1j * np.random.random([self.l, 16, 256])
-                ).astype(dtyte)
+                ).astype(dtype)
 
-            y_np_32, x_g_np_32 = self.check_main(self.x_np, dtyte)
-            y_np_gt = np.sum(self.x_np, axis=0).astype(dtyte)
+            y_np_32, x_g_np_32 = self.check_main(self.x_np, dtype)
+            y_np_gt = np.sum(self.x_np, axis=0).astype(dtype)
             np.testing.assert_allclose(y_np_32, y_np_gt, rtol=1e-06)
 
 
diff --git a/test/legacy_test/test_arange.py b/test/legacy_test/test_arange.py
index bb4bb0f430b68f..fbcc6eb421da50 100644
--- a/test/legacy_test/test_arange.py
+++ b/test/legacy_test/test_arange.py
@@ -59,7 +59,7 @@ def init_config(self):
         self.case = (0, 5, 1)
 
 
-class TestFloa16ArangeOp(TestArangeOp):
+class TestFloat16ArangeOp(TestArangeOp):
     def init_config(self):
         self.dtype = np.float16
         self.python_api = paddle.arange
diff --git a/test/legacy_test/test_backward.py b/test/legacy_test/test_backward.py
index 2ae9ede04987a5..04aeadc0382133 100644
--- a/test/legacy_test/test_backward.py
+++ b/test/legacy_test/test_backward.py
@@ -378,12 +378,12 @@ def callback(block, context):
 
 
 class TestGradientsWithOptimizer(unittest.TestCase):
-    def _check_grad_op_name(self, forward_list, optimiezed_list):
+    def _check_grad_op_name(self, forward_list, optimized_list):
         backward_list = [op + "_grad" for op in reversed(forward_list)]
-        idx = optimiezed_list.index(backward_list[0], len(backward_list))
+        idx = optimized_list.index(backward_list[0], len(backward_list))
 
         self.assertListEqual(
-            backward_list, optimiezed_list[idx : idx + len(backward_list)]
+            backward_list, optimized_list[idx : idx + len(backward_list)]
         )
 
     def test_gradient_with_optimizer(self):
diff --git a/test/legacy_test/test_bicubic_interp_v2_op.py b/test/legacy_test/test_bicubic_interp_v2_op.py
index 86b998c635648b..d2d0092f1e9f63 100644
--- a/test/legacy_test/test_bicubic_interp_v2_op.py
+++ b/test/legacy_test/test_bicubic_interp_v2_op.py
@@ -674,7 +674,7 @@ def test_case(self):
 
 class TestBicubicOpError(unittest.TestCase):
     def test_imperative_errors(self):
-        # the input of interpoalte must be Variable.
+        # the input of interpolate must be Variable.
         x1 = base.create_lod_tensor(
             np.array([-1, 3, 5, 5]), [[1, 1, 1, 1]], base.CPUPlace()
         )
@@ -687,7 +687,7 @@ def test_mode_type():
             )
 
             out = interpolate(
-                x, size=[12, 12], mode='UNKONWN', align_corners=False
+                x, size=[12, 12], mode='UNKNOWN', align_corners=False
             )
 
         def test_input_shape():
@@ -696,7 +696,7 @@ def test_input_shape():
                 x, size=[12, 12], mode='BICUBIC', align_corners=False
             )
 
-        def test_align_corcers():
+        def test_align_corners():
             x = paddle.static.data(
                 name="x", shape=[2, 3, 6, 6], dtype="float32"
             )
@@ -887,7 +887,7 @@ def test_input_shape_1():
 
         self.assertRaises(ValueError, test_mode_type)
         self.assertRaises(ValueError, test_input_shape)
-        self.assertRaises(TypeError, test_align_corcers)
+        self.assertRaises(TypeError, test_align_corners)
         self.assertRaises(ValueError, test_attr_data_format)
         self.assertRaises(TypeError, test_actual_shape)
         self.assertRaises(ValueError, test_scale_value)
diff --git a/test/legacy_test/test_checkpoint_saver.py b/test/legacy_test/test_checkpoint_saver.py
index 643ea78816579d..0390d4f8c60f03 100644
--- a/test/legacy_test/test_checkpoint_saver.py
+++ b/test/legacy_test/test_checkpoint_saver.py
@@ -18,7 +18,7 @@
 from paddle.distributed.fleet.utils.fs import HDFSClient
 
 
-class CheckpointerSaverTest(unittest.TestCase):
+class CheckpointSaverTest(unittest.TestCase):
     def test(self):
         fs = HDFSClient("/usr/local/hadoop-2.7.7", None)
         dir_path = "./checkpointsaver_test"
diff --git a/test/legacy_test/test_collective_api_base.py b/test/legacy_test/test_collective_api_base.py
index 8f6a382297a1f1..f71b524344aec7 100644
--- a/test/legacy_test/test_collective_api_base.py
+++ b/test/legacy_test/test_collective_api_base.py
@@ -611,19 +611,19 @@ def convertbf16(origin):
             result1 = []
             result2 = []
 
-            def is_empyt_list(x):
+            def is_empty_list(x):
                 if isinstance(x, list) and len(x) == 0:
                     return True
                 return False
 
             for i in range(tot_expert):
                 for arr in output1[i]:
-                    if is_empyt_list(arr):
+                    if is_empty_list(arr):
                         continue
                     result1.append(arr)
             for i in range(tot_expert):
                 for arr in output2[i]:
-                    if is_empyt_list(arr):
+                    if is_empty_list(arr):
                         continue
                     result2.append(arr)
 
diff --git a/test/legacy_test/test_conv2d_api.py b/test/legacy_test/test_conv2d_api.py
index 9d2398a5782caf..433dafbcd7fed2 100644
--- a/test/legacy_test/test_conv2d_api.py
+++ b/test/legacy_test/test_conv2d_api.py
@@ -201,7 +201,7 @@ def run_5():
 
         self.assertRaises(ValueError, run_5)
 
-        # ValueError: channel dimmention
+        # ValueError: channel dimension
         x = paddle.static.data(
             name="x",
             shape=[2, 5, 5, -1],
diff --git a/test/legacy_test/test_conv3d_layer.py b/test/legacy_test/test_conv3d_layer.py
index d514f56c2631aa..55c4c569568aa0 100644
--- a/test/legacy_test/test_conv3d_layer.py
+++ b/test/legacy_test/test_conv3d_layer.py
@@ -27,7 +27,7 @@ def __init__(
         self,
         methodName='runTest',
         batch_size=4,
-        spartial_shape=(8, 8, 8),
+        spatial_shape=(8, 8, 8),
         num_channels=6,
         num_filters=8,
         filter_size=3,
@@ -43,7 +43,7 @@ def __init__(
         self.batch_size = batch_size
         self.num_channels = num_channels
         self.num_filters = num_filters
-        self.spartial_shape = spartial_shape
+        self.spatial_shape = spatial_shape
         self.filter_size = filter_size
 
         self.padding = padding
@@ -58,13 +58,13 @@ def setUp(self):
         self.channel_last = self.data_format == "NDHWC"
         if self.channel_last:
             input_shape = (
-                (self.batch_size,) + self.spartial_shape + (self.num_channels,)
+                (self.batch_size,) + self.spatial_shape + (self.num_channels,)
             )
         else:
             input_shape = (
                 self.batch_size,
                 self.num_channels,
-            ) + self.spartial_shape
+            ) + self.spatial_shape
         self.input = np.random.randn(*input_shape).astype(self.dtype)
 
         if isinstance(self.filter_size, int):
diff --git a/test/legacy_test/test_conv3d_op.py b/test/legacy_test/test_conv3d_op.py
index cfa39de9220757..cd0d6449020ca0 100644
--- a/test/legacy_test/test_conv3d_op.py
+++ b/test/legacy_test/test_conv3d_op.py
@@ -1212,7 +1212,7 @@ def run_5():
 
             self.assertRaises(ValueError, run_5)
 
-            # ValueError: channel dimmention
+            # ValueError: channel dimension
             x = paddle.static.data(
                 name="x",
                 shape=[2, 5, 5, 5, -1],
diff --git a/test/legacy_test/test_conv3d_transpose_layer.py b/test/legacy_test/test_conv3d_transpose_layer.py
index 7624253ba6031b..910513f6b4176b 100644
--- a/test/legacy_test/test_conv3d_transpose_layer.py
+++ b/test/legacy_test/test_conv3d_transpose_layer.py
@@ -27,7 +27,7 @@ def __init__(
         self,
         methodName='runTest',
         batch_size=2,
-        spartial_shape=(8, 8, 8),
+        spatial_shape=(8, 8, 8),
         num_channels=6,
         num_filters=8,
         filter_size=3,
@@ -44,7 +44,7 @@ def __init__(
         self.batch_size = batch_size
         self.num_channels = num_channels
         self.num_filters = num_filters
-        self.spartial_shape = spartial_shape
+        self.spatial_shape = spatial_shape
         self.filter_size = filter_size
         self.output_size = output_size
 
@@ -60,13 +60,13 @@ def setUp(self):
         self.channel_last = self.data_format == "NDHWC"
         if self.channel_last:
             input_shape = (
-                (self.batch_size,) + self.spartial_shape + (self.num_channels,)
+                (self.batch_size,) + self.spatial_shape + (self.num_channels,)
             )
         else:
             input_shape = (
                 self.batch_size,
                 self.num_channels,
-            ) + self.spartial_shape
+            ) + self.spatial_shape
         self.input = np.random.randn(*input_shape).astype(self.dtype)
 
         if isinstance(self.filter_size, int):
diff --git a/test/legacy_test/test_fused_attention_op_api.py b/test/legacy_test/test_fused_attention_op_api.py
index 1570c0b0dd7335..e3ea9491e37820 100644
--- a/test/legacy_test/test_fused_attention_op_api.py
+++ b/test/legacy_test/test_fused_attention_op_api.py
@@ -53,9 +53,9 @@ def layer_norm(x, has_scale, has_bias, weight, bias, epsilon=1e-05):
     batch_size, src_len, d_model = x.shape
     x = x.reshape((batch_size * src_len, d_model))
     mu = np.mean(x, axis=1, keepdims=True)
-    sigma_squar = np.sum(np.square(x - mu), axis=1) / d_model
+    sigma_square = np.sum(np.square(x - mu), axis=1) / d_model
     x1_up = x - mu
-    x1_down_1 = sigma_squar + epsilon
+    x1_down_1 = sigma_square + epsilon
     x1_down = np.sqrt(x1_down_1)
     x1_down = x1_down.reshape((x1_down.shape[0], 1))
     x1 = x1_up / x1_down
diff --git a/test/legacy_test/test_fused_bias_dropout_residual_layer_norm_op_api.py b/test/legacy_test/test_fused_bias_dropout_residual_layer_norm_op_api.py
index dae2f2ba61c886..9efa1cd354cb3b 100644
--- a/test/legacy_test/test_fused_bias_dropout_residual_layer_norm_op_api.py
+++ b/test/legacy_test/test_fused_bias_dropout_residual_layer_norm_op_api.py
@@ -27,9 +27,9 @@ def layer_norm(x, has_scale, has_bias, weight, bias, epsilon=1e-05):
     batch_size, src_len, d_model = x.shape
     x = x.reshape((batch_size * src_len, d_model))
     mu = np.mean(x, axis=1, keepdims=True)
-    sigma_squar = np.sum(np.square(x - mu), axis=1) / d_model
+    sigma_square = np.sum(np.square(x - mu), axis=1) / d_model
     x1_up = x - mu
-    x1_down_1 = sigma_squar + epsilon
+    x1_down_1 = sigma_square + epsilon
     x1_down = np.sqrt(x1_down_1)
     x1_down = x1_down.reshape((x1_down.shape[0], 1))
     x1 = x1_up / x1_down
diff --git a/test/legacy_test/test_log_softmax.py b/test/legacy_test/test_log_softmax.py
index 64be97bb1aba15..494ba127032e42 100644
--- a/test/legacy_test/test_log_softmax.py
+++ b/test/legacy_test/test_log_softmax.py
@@ -194,7 +194,7 @@ def check_api(self, axis=-1):
             out = exe.run(feed={'x': self.x}, fetch_list=[y])
         np.testing.assert_allclose(out[0], ref_out, rtol=1e-05)
 
-        # test dygrapg api
+        # test dygraph api
         paddle.disable_static()
         x = paddle.to_tensor(self.x)
         y = logsoftmax(x)
diff --git a/test/legacy_test/test_paddle_save_load_binary.py b/test/legacy_test/test_paddle_save_load_binary.py
index df7304cf1d19e5..22b62e082cc94a 100644
--- a/test/legacy_test/test_paddle_save_load_binary.py
+++ b/test/legacy_test/test_paddle_save_load_binary.py
@@ -94,7 +94,7 @@ def test_replace_save_load_vars(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
             # test for replace_save_vars/io.load_vars
diff --git a/test/legacy_test/test_scatter_nd_op.py b/test/legacy_test/test_scatter_nd_op.py
index 4936ccdb8989f2..531b865a7c50bd 100644
--- a/test/legacy_test/test_scatter_nd_op.py
+++ b/test/legacy_test/test_scatter_nd_op.py
@@ -29,16 +29,16 @@ def numpy_scatter_nd(ref, index, updates, fun):
     index_shape = index.shape
 
     end_size = index_shape[-1]
-    remain_numl = 1
+    remain_numel = 1
     for i in range(len(index_shape) - 1):
-        remain_numl *= index_shape[i]
+        remain_numel *= index_shape[i]
 
     slice_size = 1
     for i in range(end_size, len(ref_shape)):
         slice_size *= ref_shape[i]
 
-    flat_index = index.reshape([remain_numl] + list(index_shape[-1:]))
-    flat_updates = updates.reshape((remain_numl, slice_size))
+    flat_index = index.reshape([remain_numel] + list(index_shape[-1:]))
+    flat_updates = updates.reshape((remain_numel, slice_size))
     flat_output = ref.reshape(list(ref_shape[:end_size]) + [slice_size])
 
     for i_up, i_out in enumerate(flat_index):
diff --git a/test/legacy_test/test_set_value_op.py b/test/legacy_test/test_set_value_op.py
index c289185e58d212..4113805c663b49 100644
--- a/test/legacy_test/test_set_value_op.py
+++ b/test/legacy_test/test_set_value_op.py
@@ -265,7 +265,7 @@ def _get_answer(self):
 
 
 # 1.2.3 step < 0
-class TestSetValueItemSliceNegetiveStep(TestSetValueApi):
+class TestSetValueItemSliceNegativeStep(TestSetValueApi):
     def set_shape(self):
         self.shape = [5, 2]
 
@@ -283,7 +283,7 @@ def _get_answer(self):
         self.data[5:2:-1] = self.value
 
 
-class TestSetValueItemSliceNegetiveStep2(TestSetValueApi):
+class TestSetValueItemSliceNegativeStep2(TestSetValueApi):
     def set_shape(self):
         self.shape = [5]
 
@@ -301,7 +301,7 @@ def _get_answer(self):
         self.data[1::-1] = self.value
 
 
-class TestSetValueItemSliceNegetiveStep3(TestSetValueApi):
+class TestSetValueItemSliceNegativeStep3(TestSetValueApi):
     def set_shape(self):
         self.shape = [3]
 
@@ -319,7 +319,7 @@ def _get_answer(self):
         self.data[::-1] = self.value
 
 
-class TestSetValueItemSliceNegetiveStep4(TestSetValueApi):
+class TestSetValueItemSliceNegativeStep4(TestSetValueApi):
     def set_shape(self):
         self.shape = [3, 4, 5]
 
@@ -1504,14 +1504,14 @@ def set_value(t, value):
         np.testing.assert_array_equal(
             inps.grad.numpy(),
             input_grad,
-            err_msg='The gradient of value should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of value should be \n{},\n but received {}'.format(
                 input_grad, inps.grad.numpy()
             ),
         )
         np.testing.assert_array_equal(
             value.grad.numpy(),
             value_grad,
-            err_msg='The gradient of input should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of input should be \n{},\n but received {}'.format(
                 value_grad, value.grad.numpy()
             ),
         )
@@ -1538,14 +1538,14 @@ def set_value(t, value):
         np.testing.assert_array_equal(
             inps2.grad.numpy(),
             input_grad2,
-            err_msg='The gradient of value should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of value should be \n{},\n but received {}'.format(
                 input_grad, inps2.grad.numpy()
             ),
         )
         np.testing.assert_array_equal(
             value2.grad.numpy(),
             value_grad2,
-            err_msg='The gradient of input should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of input should be \n{},\n but received {}'.format(
                 value_grad, value2.grad.numpy()
             ),
         )
@@ -1592,14 +1592,14 @@ def set_value3(t, value):
         np.testing.assert_array_equal(
             inps.grad.numpy(),
             input_grad,
-            err_msg='The gradient of value should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of value should be \n{},\n but received {}'.format(
                 input_grad, inps.grad.numpy()
             ),
         )
         np.testing.assert_array_equal(
             value.grad.numpy(),
             value_grad,
-            err_msg='The gradient of input should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of input should be \n{},\n but received {}'.format(
                 value_grad, value.grad.numpy()
             ),
         )
@@ -1640,14 +1640,14 @@ def set_value4(t, value):
         np.testing.assert_array_equal(
             inps.grad.numpy(),
             input_grad,
-            err_msg='The gradient of value should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of value should be \n{},\n but received {}'.format(
                 input_grad, inps.grad.numpy()
             ),
         )
         np.testing.assert_array_equal(
             value.grad.numpy(),
             value_grad,
-            err_msg='The gradient of input should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of input should be \n{},\n but received {}'.format(
                 value_grad, value.grad.numpy()
             ),
         )
@@ -1692,14 +1692,14 @@ def set_value5(t, value):
         np.testing.assert_array_equal(
             inps.grad.numpy(),
             input_grad,
-            err_msg='The gradient of value should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of value should be \n{},\n but received {}'.format(
                 input_grad, inps.grad.numpy()
             ),
         )
         np.testing.assert_array_equal(
             value.grad.numpy(),
             value_grad,
-            err_msg='The gradient of input should be \n{},\n but reveived {}'.format(
+            err_msg='The gradient of input should be \n{},\n but received {}'.format(
                 value_grad, value.grad.numpy()
             ),
         )
diff --git a/test/legacy_test/test_sgd_op.py b/test/legacy_test/test_sgd_op.py
index d71b2971858922..ba7dbb99d1b878 100644
--- a/test/legacy_test/test_sgd_op.py
+++ b/test/legacy_test/test_sgd_op.py
@@ -85,7 +85,7 @@ def check_with_place(self, place):
         param_array = np.full((height, self.row_numel), 5.0).astype("float32")
         param.set(param_array, place)
 
-        # create and initialize LeraningRate Variable
+        # create and initialize LearningRate Variable
         lr = scope.var('LearningRate').get_tensor()
         lr_array = np.full((1), 2.0).astype("float32")
         lr.set(lr_array, place)
@@ -170,7 +170,7 @@ def check_with_place(self, place):
 
         w_before_optimize = np.array(w_tensor)
 
-        # create and initialize LeraningRate Variable
+        # create and initialize LearningRate Variable
         lr_value = 0.1
         lr = scope.var('LearningRate').get_tensor()
         lr_array = np.full((1), lr_value).astype("float32")
diff --git a/test/legacy_test/test_sgd_op_bf16.py b/test/legacy_test/test_sgd_op_bf16.py
index 3baf0a490cbf53..3af7d9c6bc93af 100644
--- a/test/legacy_test/test_sgd_op_bf16.py
+++ b/test/legacy_test/test_sgd_op_bf16.py
@@ -286,7 +286,7 @@ def _reference(self, data, emb_weight, bf16=False):
         out_dtype = np.uint16 if bf16 else np.float32
         lookup_table_grad = np.zeros(self.w_shape, dtype=out_dtype)
 
-        # indexes may dupplicate
+        # indexes may duplicate
         if bf16:
             for i, idx in enumerate(data):
                 idxv = idx[0]
diff --git a/test/legacy_test/test_signal.py b/test/legacy_test/test_signal.py
index 1d86f15f510951..8d70702a26ad64 100644
--- a/test/legacy_test/test_signal.py
+++ b/test/legacy_test/test_signal.py
@@ -574,7 +574,7 @@ def decorate(cls):
 def setUpModule():
     global rtol
     global atol
-    # All test case will use float64 for compare percision, refs:
+    # All test case will use float64 for compare precision, refs:
     # https://github.com/PaddlePaddle/Paddle/wiki/Upgrade-OP-Precision-to-Float64
     rtol = {
         'float32': 1e-06,
diff --git a/test/legacy_test/test_slice_scatter.py b/test/legacy_test/test_slice_scatter.py
index 075b5a57418865..bb46ddffb9b94f 100644
--- a/test/legacy_test/test_slice_scatter.py
+++ b/test/legacy_test/test_slice_scatter.py
@@ -264,7 +264,7 @@ def init_dtype(self):
         self.dtype = 'float32'
 
 
-class TestSliceScatterApiBroadcase2D(TestSliceScatterApi):
+class TestSliceScatterApiBroadcast2D(TestSliceScatterApi):
     def init_shape(self):
         self.x_shape = [8, 9]
         self.value_shape = [8, 1]
@@ -274,12 +274,12 @@ def init_shape(self):
         self.strides = [2]
 
 
-class TestSliceScatterApiBroadcase2DFloat32(TestSliceScatterApiBroadcase2D):
+class TestSliceScatterApiBroadcast2DFloat32(TestSliceScatterApiBroadcast2D):
     def init_dtype(self):
         self.dtype = 'float32'
 
 
-class TestSliceScatterApiBroadcase3D(TestSliceScatterApi):
+class TestSliceScatterApiBroadcast3D(TestSliceScatterApi):
     def init_shape(self):
         self.x_shape = [8, 9, 6]
         self.value_shape = [1, 9, 1]
@@ -289,7 +289,7 @@ def init_shape(self):
         self.strides = [3, 2]
 
 
-class TestSliceScatterApiBroadcase3DFloat32(TestSliceScatterApiBroadcase3D):
+class TestSliceScatterApiBroadcast3DFloat32(TestSliceScatterApiBroadcast3D):
     def init_dtype(self):
         self.dtype = 'float32'
 
diff --git a/test/legacy_test/test_softmax_op.py b/test/legacy_test/test_softmax_op.py
index 6b9952e569ae5d..1876424cf4d4ba 100644
--- a/test/legacy_test/test_softmax_op.py
+++ b/test/legacy_test/test_softmax_op.py
@@ -59,7 +59,7 @@ def setUp(self):
         self.public_python_api = F.softmax
         self.use_cudnn = False
         self.use_mkldnn = False
-        # explicilty use float32 for ROCm, as MIOpen does not yet support float64
+        # explicitly use float32 for ROCm, as MIOpen does not yet support float64
         self.dtype = np.float32 if core.is_compiled_with_rocm() else np.float64
         self.init_kernel_type()
         self.shape = self.get_x_shape()
@@ -142,7 +142,7 @@ def setUp(self):
         self.public_python_api = F.softmax
         self.use_cudnn = False
         self.use_mkldnn = False
-        # explicilty use float32 for ROCm, as MIOpen does not yet support float64
+        # explicitly use float32 for ROCm, as MIOpen does not yet support float64
         self.dtype = np.float32 if core.is_compiled_with_rocm() else np.float64
         self.init_kernel_type()
 
@@ -190,7 +190,7 @@ def setUp(self):
         self.prim_op_type = "comp"
         self.use_cudnn = True
         self.use_mkldnn = False
-        # explicilty use float32 for ROCm, as MIOpen does not yet support float64
+        # explicitly use float32 for ROCm, as MIOpen does not yet support float64
         self.dtype = np.float32 if core.is_compiled_with_rocm() else np.float64
 
         np.random.seed(0)
diff --git a/test/legacy_test/test_softmax_with_cross_entropy_op.py b/test/legacy_test/test_softmax_with_cross_entropy_op.py
index 62f475cd922a7c..8bafae13efc703 100644
--- a/test/legacy_test/test_softmax_with_cross_entropy_op.py
+++ b/test/legacy_test/test_softmax_with_cross_entropy_op.py
@@ -99,7 +99,7 @@ def initParams(self):
         self.python_out_sig = ["Loss", "Softmax"]
         self.numeric_stable_mode = False
         self.soft_label = False
-        # explicilty use float32 for ROCm, as MIOpen does not yet support float64
+        # explicitly use float32 for ROCm, as MIOpen does not yet support float64
         self.dtype = np.float32 if core.is_compiled_with_rocm() else np.float64
         self.axis = -1
         self.ignore_index = -1
diff --git a/test/legacy_test/test_sparse_attention_op.py b/test/legacy_test/test_sparse_attention_op.py
index 48946522864aa7..24272c5e416b0c 100644
--- a/test/legacy_test/test_sparse_attention_op.py
+++ b/test/legacy_test/test_sparse_attention_op.py
@@ -438,7 +438,7 @@ def test_dygraph(self):
         paddle_key = paddle.to_tensor(key, place=self.place)
         paddle_value = paddle.to_tensor(value, place=self.place)
         paddle_offset = paddle.to_tensor(offset, place=self.place)
-        paddle_colunmns = paddle.to_tensor(columns, place=self.place)
+        paddle_columns = paddle.to_tensor(columns, place=self.place)
         paddle_kp_mask = paddle.to_tensor(key_padding_mask, place=self.place)
         paddle_attn_mask = paddle.to_tensor(attn_mask, place=self.place)
 
@@ -448,7 +448,7 @@ def test_dygraph(self):
                 paddle_key,
                 paddle_value,
                 paddle_offset,
-                paddle_colunmns,
+                paddle_columns,
                 key_padding_mask=paddle_kp_mask,
                 attn_mask=paddle_attn_mask,
             )
@@ -469,7 +469,7 @@ def test_dygraph(self):
                 paddle_key,
                 paddle_value,
                 paddle_offset,
-                paddle_colunmns,
+                paddle_columns,
             )
 
             numpy_result, __, __ = ref_batch_sparse_attention(
diff --git a/test/legacy_test/test_split_op.py b/test/legacy_test/test_split_op.py
index 9311a5f2d99579..0f780b2ddfff63 100644
--- a/test/legacy_test/test_split_op.py
+++ b/test/legacy_test/test_split_op.py
@@ -567,9 +567,11 @@ def test_out1(self):
             eager_x2_out = x2.numpy()
             loss = x0.sum()
             loss.backward()
-            manul_grad = np.zeros_like(input_1)
-            manul_grad[:, :2, :] = 1
-            np.testing.assert_allclose(input.gradient(), manul_grad, rtol=1e-05)
+            manual_grad = np.zeros_like(input_1)
+            manual_grad[:, :2, :] = 1
+            np.testing.assert_allclose(
+                input.gradient(), manual_grad, rtol=1e-05
+            )
             np.testing.assert_allclose(ex_x0, eager_x0_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x1, eager_x1_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x2, eager_x2_out, rtol=1e-05)
@@ -597,9 +599,11 @@ def test_out2(self):
             eager_x2_out = x2.numpy()
             loss = x0.sum()
             loss.backward()
-            manul_grad = np.zeros_like(input_1)
-            manul_grad[:, :2, :] = 1
-            np.testing.assert_allclose(input.gradient(), manul_grad, rtol=1e-05)
+            manual_grad = np.zeros_like(input_1)
+            manual_grad[:, :2, :] = 1
+            np.testing.assert_allclose(
+                input.gradient(), manual_grad, rtol=1e-05
+            )
             np.testing.assert_allclose(ex_x0, eager_x0_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x1, eager_x1_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x2, eager_x2_out, rtol=1e-05)
@@ -630,9 +634,11 @@ def test_out1(self):
             eager_x2_out = x2.numpy()
             loss = x0.sum()
             loss.backward()
-            manul_grad = np.zeros_like(input_1)
-            manul_grad[:, :2, :] = 1
-            np.testing.assert_allclose(input.gradient(), manul_grad, rtol=1e-05)
+            manual_grad = np.zeros_like(input_1)
+            manual_grad[:, :2, :] = 1
+            np.testing.assert_allclose(
+                input.gradient(), manual_grad, rtol=1e-05
+            )
             np.testing.assert_allclose(ex_x0, eager_x0_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x1, eager_x1_out, rtol=1e-05)
             np.testing.assert_allclose(ex_x2, eager_x2_out, rtol=1e-05)
diff --git a/test/legacy_test/test_static_pylayer.py b/test/legacy_test/test_static_pylayer.py
index fd5075a4904aa3..34a5afe577a67a 100644
--- a/test/legacy_test/test_static_pylayer.py
+++ b/test/legacy_test/test_static_pylayer.py
@@ -133,7 +133,7 @@ def backward_fn(dy):
         self.assertEqual(x_grad.shape, ())
 
     @test_with_pir_api
-    def test_return_var_typle(self):
+    def test_return_var_type(self):
         def forward_fn(a, b):
             return 3 * a, -2 * b
 
diff --git a/test/legacy_test/test_static_save_load.py b/test/legacy_test/test_static_save_load.py
index ca1adaed4ef0ab..f662ee3f95e699 100644
--- a/test/legacy_test/test_static_save_load.py
+++ b/test/legacy_test/test_static_save_load.py
@@ -340,7 +340,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -357,7 +357,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
@@ -480,7 +480,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -497,7 +497,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
@@ -613,7 +613,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -630,7 +630,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
@@ -751,7 +751,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -768,7 +768,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             # base.load(test_program, "./test_1", None )
@@ -807,7 +807,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.set_program_state(test_program, program_state_1)
@@ -829,7 +829,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.set_program_state(test_program, program_state_2)
@@ -851,7 +851,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.set_program_state(test_program, program_state_3)
@@ -954,7 +954,7 @@ def set_var(var, ndarray):
                 t = np.array(
                     base.global_scope().find_var(var.name).get_tensor()
                 )
-                # make sure all the paramerter or optimizer var have been update
+                # make sure all the parameter or optimizer var have been update
                 base_map[var.name] = t
 
         for var in program.list_vars():
@@ -1073,7 +1073,7 @@ def test_load_from_old_interface(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -1091,7 +1091,7 @@ def test_load_from_old_interface(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
@@ -1214,7 +1214,7 @@ def test_load_from_old_interface_var_list(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -1237,7 +1237,7 @@ def test_load_from_old_interface_var_list(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
@@ -1355,7 +1355,7 @@ def test_load_from_old_interface(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
             save_dir = os.path.join(temp_dir.name, "test_path")
@@ -1373,7 +1373,7 @@ def test_load_from_old_interface(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             file_model_path = os.path.join(save_dir, "model_single")
@@ -1556,7 +1556,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
             save_dir = os.path.join(self.temp_dir.name, "test_program_1")
@@ -1571,7 +1571,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             # case 1: load basic
@@ -1731,7 +1731,7 @@ def test_ptb_rnn_cpu_float32(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
@@ -1749,7 +1749,7 @@ def test_ptb_rnn_cpu_float32(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             # base.load(test_program, "./test_1", None )
@@ -1816,7 +1816,7 @@ def test_pickle_protocol(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
 
diff --git a/test/legacy_test/test_static_save_load_bf16.py b/test/legacy_test/test_static_save_load_bf16.py
index 1ca1bec7d15e76..bc91f34b3f60c1 100644
--- a/test/legacy_test/test_static_save_load_bf16.py
+++ b/test/legacy_test/test_static_save_load_bf16.py
@@ -127,7 +127,7 @@ def test_ptb_rnn_cpu_bfloat16(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
             save_dir = os.path.join(self.temp_dir.name, "test_1")
@@ -142,7 +142,7 @@ def test_ptb_rnn_cpu_bfloat16(self):
                     new_t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been set to zero
+                    # make sure all the parameter or optimizer var have been set to zero
                     self.assertTrue(np.sum(np.abs(new_t)) == 0)
 
             paddle.static.load(
diff --git a/test/legacy_test/test_static_save_load_large.py b/test/legacy_test/test_static_save_load_large.py
index 2011dd45cfaf82..d0ef10ae3a9abf 100644
--- a/test/legacy_test/test_static_save_load_large.py
+++ b/test/legacy_test/test_static_save_load_large.py
@@ -49,7 +49,7 @@ def test_large_parameters_static_save(self):
                     t = np.array(
                         base.global_scope().find_var(var.name).get_tensor()
                     )
-                    # make sure all the paramerter or optimizer var have been update
+                    # make sure all the parameter or optimizer var have been update
                     self.assertTrue(np.sum(np.abs(t)) != 0)
                     base_map[var.name] = t
             temp_dir = tempfile.TemporaryDirectory()
diff --git a/test/legacy_test/test_static_shape_inferrence_for_shape_tensor.py b/test/legacy_test/test_static_shape_inferrence_for_shape_tensor.py
index a38fd888819370..eaa910637bbd2a 100644
--- a/test/legacy_test/test_static_shape_inferrence_for_shape_tensor.py
+++ b/test/legacy_test/test_static_shape_inferrence_for_shape_tensor.py
@@ -17,7 +17,7 @@
 import paddle
 
 
-class StaticShapeInferrenceTest(unittest.TestCase):
+class StaticShapeInferenceTest(unittest.TestCase):
     def test_static_graph(self):
         paddle.enable_static()
         data = paddle.static.data(name="x", shape=[-1, 2], dtype='float32')
diff --git a/test/legacy_test/test_sum_op.py b/test/legacy_test/test_sum_op.py
index cd2f9a4f6cdec1..a2570a566c3488 100644
--- a/test/legacy_test/test_sum_op.py
+++ b/test/legacy_test/test_sum_op.py
@@ -86,16 +86,16 @@ def setUp(self):
         self.init_kernel_type()
 
     def check_with_place(self, place, inplace):
-        self.check_input_and_optput(
+        self.check_input_and_output(
             core.Scope(), place, inplace, True, True, True
         )
-        self.check_input_and_optput(
+        self.check_input_and_output(
             core.Scope(), place, inplace, False, True, True
         )
-        self.check_input_and_optput(
+        self.check_input_and_output(
             core.Scope(), place, inplace, False, False, True
         )
-        self.check_input_and_optput(
+        self.check_input_and_output(
             core.Scope(), place, inplace, False, False, False
         )
 
@@ -108,7 +108,7 @@ def _get_array(self, rows, row_numel):
             array[i] *= rows[i]
         return array
 
-    def check_input_and_optput(
+    def check_input_and_output(
         self,
         scope,
         place,
@@ -198,7 +198,7 @@ def _get_array(self, rows, row_numel):
         else:
             return np.ndarray((0, row_numel), dtype=self.dtype)
 
-    def check_input_and_optput(
+    def check_input_and_output(
         self,
         scope,
         place,
diff --git a/test/legacy_test/test_svd_op.py b/test/legacy_test/test_svd_op.py
index a74404e4085244..6b41fa7cead8d5 100644
--- a/test/legacy_test/test_svd_op.py
+++ b/test/legacy_test/test_svd_op.py
@@ -90,7 +90,7 @@ def test_check_grad(self):
 
 
 class TestSvdCheckGrad2(TestSvdOp):
-    # NOTE(xiongkun03): because we want to construct some full rank matrics,
+    # NOTE(xiongkun03): because we want to construct some full rank matrices,
     #                   so we can't specifize matrices which numel() > 100
 
     no_need_check_grad = True
diff --git a/test/legacy_test/test_sync_batch_norm_op.py b/test/legacy_test/test_sync_batch_norm_op.py
index 17daa24996b4ff..3b7ea63cb5963e 100644
--- a/test/legacy_test/test_sync_batch_norm_op.py
+++ b/test/legacy_test/test_sync_batch_norm_op.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-test for sync bachnorm op.
+test for sync batchnorm op.
 for both FP64 and FP16 input.
 """
 
diff --git a/test/legacy_test/test_tensor.py b/test/legacy_test/test_tensor.py
index 8c007d46753906..9207d3a181789f 100644
--- a/test/legacy_test/test_tensor.py
+++ b/test/legacy_test/test_tensor.py
@@ -339,7 +339,7 @@ def test_print_tensor(self):
             print(tensor)
             self.assertTrue(isinstance(str(tensor), str))
 
-    def test_tensor_poiter(self):
+    def test_tensor_pointer(self):
         place = core.CPUPlace()
         scope = core.Scope()
         var = scope.var("test_tensor")
diff --git a/test/legacy_test/test_tensor_register_hook.py b/test/legacy_test/test_tensor_register_hook.py
index 29c614713dff30..c7826c983adcd7 100644
--- a/test/legacy_test/test_tensor_register_hook.py
+++ b/test/legacy_test/test_tensor_register_hook.py
@@ -589,7 +589,7 @@ def test_register_backward_hook_for_var_without_gradient(self):
             x._register_backward_hook(global_void_hook)
 
 
-class TestRegsiterBackwardFinalHook(unittest.TestCase):
+class TestRegisterBackwardFinalHook(unittest.TestCase):
     def setUp(self):
         self.devices = ["cpu"]
         if paddle.is_compiled_with_cuda():
diff --git a/test/legacy_test/test_tensor_uva.py b/test/legacy_test/test_tensor_uva.py
index 34d7e59609e0bc..e7b6d03fe8bd93 100644
--- a/test/legacy_test/test_tensor_uva.py
+++ b/test/legacy_test/test_tensor_uva.py
@@ -53,7 +53,7 @@ def test_uva_tensor_creation(self):
                 np.testing.assert_allclose(tensor.numpy(), data, rtol=1e-05)
                 np.testing.assert_allclose(tensor2.numpy(), data, rtol=1e-05)
 
-    def test_uva_tensor_corectness(self):
+    def test_uva_tensor_correctness(self):
         if paddle.base.core.is_compiled_with_cuda():
             a = np.arange(0, 100, dtype="int32")
             a = a.reshape([10, 10])
diff --git a/test/legacy_test/test_traced_layer_err_msg.py b/test/legacy_test/test_traced_layer_err_msg.py
index 16880dafdcbf75..4927fdea82a549 100644
--- a/test/legacy_test/test_traced_layer_err_msg.py
+++ b/test/legacy_test/test_traced_layer_err_msg.py
@@ -241,7 +241,7 @@ def test_linear_net_with_none(self):
 
 
 class TestTracedLayerSaveInferenceModel(unittest.TestCase):
-    """test save_inference_model will automaticlly create non-exist dir"""
+    """test save_inference_model will automatically create non-exist dir"""
 
     def setUp(self):
         self.temp_dir = tempfile.TemporaryDirectory()
diff --git a/test/legacy_test/test_trans_layout_op.py b/test/legacy_test/test_trans_layout_op.py
index da59301aacfc3c..b936abc95df954 100644
--- a/test/legacy_test/test_trans_layout_op.py
+++ b/test/legacy_test/test_trans_layout_op.py
@@ -45,17 +45,17 @@ def test_check_output(self):
 class LayoutAutoTune(unittest.TestCase):
     def test_config(self):
         paddle.base.core.enable_layout_autotune()
-        if self.use_autoune():
+        if self.use_autotune():
             self.assertEqual(paddle.base.core.use_layout_autotune(), True)
             paddle.base.core.disable_layout_autotune()
         self.assertEqual(paddle.base.core.use_layout_autotune(), False)
-        self.use_autoune()
+        self.use_autotune()
 
     def setUp(self):
         paddle.disable_static()
-        self.use_autoune()
+        self.use_autotune()
 
-    def use_autoune(self):
+    def use_autotune(self):
         if paddle.is_compiled_with_cuda():
             paddle.incubate.autotune.set_config(
                 config={"layout": {"enable": True}}
diff --git a/test/legacy_test/test_transformer_api.py b/test/legacy_test/test_transformer_api.py
index 5945ac4a7d5d24..fd55abde22093f 100644
--- a/test/legacy_test/test_transformer_api.py
+++ b/test/legacy_test/test_transformer_api.py
@@ -252,9 +252,9 @@ def layer_norm(x, normalized_shape, norm, epsilon=1e-05, act=None):
         batch_size, src_len, d_model = x.shape
         x = x.reshape((batch_size * src_len, d_model))
         mu = np.mean(x, axis=1, keepdims=True)
-        sigma_squar = np.sum(np.square(x - mu), axis=1) / d_model
+        sigma_square = np.sum(np.square(x - mu), axis=1) / d_model
         x1_up = x - mu
-        x1_down_1 = sigma_squar + epsilon
+        x1_down_1 = sigma_square + epsilon
         x1_down = np.sqrt(x1_down_1)
         x1_down = x1_down.reshape((x1_down.shape[0], 1))
         x1 = x1_up / x1_down
diff --git a/test/legacy_test/test_tril_triu_op.py b/test/legacy_test/test_tril_triu_op.py
index 36a2ddb0383a77..391b7f83f9dc39 100644
--- a/test/legacy_test/test_tril_triu_op.py
+++ b/test/legacy_test/test_tril_triu_op.py
@@ -116,7 +116,7 @@ def test_check_grad_normal(self):
 def case_generator(op_type, Xshape, diagonal, expected, dtype):
     """
     Generate testcases with the params shape of X, diagonal and op_type.
-    If arg`expercted` is 'success', it will register an Optest case and expect to pass.
+    If arg `expected` is 'success', it will register an OpTest case and expect to pass.
     Otherwise, it will register an API case and check the expect failure.
     """
     cls_name = (
diff --git a/test/legacy_test/test_truncated_gaussian_random_op.py b/test/legacy_test/test_truncated_gaussian_random_op.py
index eb8b502b082d4c..0572d0da6face2 100644
--- a/test/legacy_test/test_truncated_gaussian_random_op.py
+++ b/test/legacy_test/test_truncated_gaussian_random_op.py
@@ -22,7 +22,7 @@
 from paddle.base.executor import Executor
 
 
-class TestTrunctedGaussianRandomOp(unittest.TestCase):
+class TestTruncatedGaussianRandomOp(unittest.TestCase):
     def setUp(self):
         self.op_type = "truncated_gaussian_random"
         self.inputs = {}
diff --git a/test/legacy_test/test_vision_models.py b/test/legacy_test/test_vision_models.py
index b53f1100305496..150ae03c58fe27 100644
--- a/test/legacy_test/test_vision_models.py
+++ b/test/legacy_test/test_vision_models.py
@@ -20,7 +20,7 @@
 from paddle.vision import models
 
 
-class TestVisonModels(unittest.TestCase):
+class TestVisionModels(unittest.TestCase):
     def models_infer(self, arch, pretrained=False, batch_norm=False):
         x = np.array(np.random.random((2, 3, 224, 224)), dtype=np.float32)
         if batch_norm:
diff --git a/test/legacy_test/test_viterbi_decode_op.py b/test/legacy_test/test_viterbi_decode_op.py
index 91f79565a2caa7..fd5ff4b61f789e 100644
--- a/test/legacy_test/test_viterbi_decode_op.py
+++ b/test/legacy_test/test_viterbi_decode_op.py
@@ -31,7 +31,7 @@ def __call__(self, inputs, length):
         bs, seq_len, n_label = inputs.shape
         inputs_t = np.transpose(inputs, (1, 0, 2))
         trans_exp = np.expand_dims(self.transitions, axis=0)
-        historys = []
+        histories = []
         left_length = np.array(length)
         max_seq_len = np.amax(left_length)
         left_length = np.expand_dims(left_length, 1)
@@ -49,7 +49,7 @@ def __call__(self, inputs, length):
             alpha_exp = np.expand_dims(alpha, 2)
             alpha_trn_sum = alpha_exp + trans_exp
             max_res = np.amax(alpha_trn_sum, 1), np.argmax(alpha_trn_sum, 1)
-            historys = historys + [max_res[1]] if i >= 1 else []
+            histories = histories + [max_res[1]] if i >= 1 else []
             alpha_nxt = max_res[0] + logit
             mask = left_length > 0
             alpha = mask * alpha_nxt + (1 - mask) * alpha
@@ -61,7 +61,7 @@ def __call__(self, inputs, length):
         last_ids_update = last_ids * (left_length >= 0)
         batch_path = [last_ids_update]
         batch_offset = np.arange(bs) * n_label
-        for hist in reversed(historys):
+        for hist in reversed(histories):
             left_length = left_length + 1
             gather_idx = batch_offset + last_ids
             last_ids_update = np.take(hist, gather_idx) * (left_length > 0)
diff --git a/test/legacy_test/test_warpctc_op.py b/test/legacy_test/test_warpctc_op.py
index e747f381af608e..9355eeec21ad59 100644
--- a/test/legacy_test/test_warpctc_op.py
+++ b/test/legacy_test/test_warpctc_op.py
@@ -110,7 +110,7 @@ def forward_a_sequence(self, softmax_a_sequence, labels_a_sequence):
         required_times = labels_a_sequence.shape[0]
         old_label = -1
         for i in range(labels_a_sequence.shape[0]):
-            # two contingous labels with the same value
+            # two contiguous labels with the same value
             if labels_a_sequence[i, 0] == old_label:
                 required_times = required_times + 1
             old_label = labels_a_sequence[i, 0]
diff --git a/test/legacy_test/test_where_op.py b/test/legacy_test/test_where_op.py
index 16c6020d546506..d88b1b3b3a5a77 100644
--- a/test/legacy_test/test_where_op.py
+++ b/test/legacy_test/test_where_op.py
@@ -319,7 +319,7 @@ def __test_where_with_broadcast_static(self, cond_shape, x_shape, y_shape):
                 np.testing.assert_array_equal(out[0], expect)
 
     def __test_where_with_type_promotion(
-        self, x_dtype, y_dtype, expeced_dtype=None
+        self, x_dtype, y_dtype, expected_dtype=None
     ):
         paddle.enable_static()
         main_program = paddle.static.Program()
@@ -367,7 +367,7 @@ def __test_where_with_type_promotion(
                     )
                     expect = np.where(cond_data, x_data_convert, y_data_convert)
                     np.testing.assert_array_equal(out[0], expect)
-                    self.assertEqual(out[0].dtype.__str__(), expeced_dtype)
+                    self.assertEqual(out[0].dtype.__str__(), expected_dtype)
                 else:
                     expect = np.where(cond_data, x_data, y_data)
                     np.testing.assert_array_equal(out[0], expect)
diff --git a/test/legacy_test/test_while_loop_op.py b/test/legacy_test/test_while_loop_op.py
index ec63375043e05d..3b8123d48df4ce 100644
--- a/test/legacy_test/test_while_loop_op.py
+++ b/test/legacy_test/test_while_loop_op.py
@@ -684,7 +684,7 @@ def type_error_cond_returns_not_variable():
 
             self.assertRaises(TypeError, type_error_cond_returns_not_variable)
 
-            # The type of `cond` returns in Op(while_loop) must be a bollean variable
+            # The type of `cond` returns in Op(while_loop) must be a boolean variable
             def type_error_cond_returns_not_boolean():
                 out = paddle.static.nn.while_loop(
                     cond_returns_not_bool_tensor, body, [data_1d]
diff --git a/test/legacy_test/test_while_op.py b/test/legacy_test/test_while_op.py
index d763576f9ff3a0..69dc4e1b8c070c 100644
--- a/test/legacy_test/test_while_op.py
+++ b/test/legacy_test/test_while_op.py
@@ -235,7 +235,7 @@ def body(i, s, x):
             x = paddle.static.data(shape=[-1], name='x', dtype='float32')
             func(x)
 
-        # NOTE(winter-wang): The while_op in pir mode  doesn't need following constrait, so hre only check when in non-pir mode.
+        # NOTE(winter-wang): The while_op in pir mode doesn't need following constraint, so here only check when in non-pir mode.
         if not in_pir_mode():
             for op in main_program.block(0).ops:
                 if op.type == "while":
diff --git a/test/legacy_test/test_zeros_like_op.py b/test/legacy_test/test_zeros_like_op.py
index 538556cd4f1fcd..4768a2506f2493 100644
--- a/test/legacy_test/test_zeros_like_op.py
+++ b/test/legacy_test/test_zeros_like_op.py
@@ -59,7 +59,7 @@ def test_api(self):
             self.assertEqual((outs[i] == np.zeros(shape, dtype)).all(), True)
 
 
-class TestZerosLikeImpeartive(unittest.TestCase):
+class TestZerosLikeImperative(unittest.TestCase):
     def test_out(self):
         shape = [3, 4]
         place = (
diff --git a/test/mkldnn/test_elementwise_add_mkldnn_op.py b/test/mkldnn/test_elementwise_add_mkldnn_op.py
index 9d62f1cf55e977..8b9dded0129bdd 100644
--- a/test/mkldnn/test_elementwise_add_mkldnn_op.py
+++ b/test/mkldnn/test_elementwise_add_mkldnn_op.py
@@ -102,21 +102,21 @@ def test_check_grad_ignore_x(self):
         pass
 
 
-class TestOneDNNlementwiseAddOpZeroDim(TestOneDNNElementwiseAddOp):
+class TestOneDNNElementwiseAddOpZeroDim(TestOneDNNElementwiseAddOp):
     def init_input_output(self):
         self.x = np.random.random((100,)).astype(self.dtype)
         self.y = np.array(3.0).astype(self.dtype)
         self.out = np.add(self.x, self.y)
 
 
-class TestOneDNNlementwiseAddOpZeroDim2(TestOneDNNElementwiseAddOp):
+class TestOneDNNElementwiseAddOpZeroDim2(TestOneDNNElementwiseAddOp):
     def init_input_output(self):
         self.x = np.array(3.0).astype(self.dtype)
         self.y = np.random.random((100,)).astype(self.dtype)
         self.out = np.add(self.x, self.y)
 
 
-class TestOneDNNlementwiseAddOpZeroDim3(TestOneDNNElementwiseAddOp):
+class TestOneDNNElementwiseAddOpZeroDim3(TestOneDNNElementwiseAddOp):
     def init_input_output(self):
         self.x = np.array(3.0).astype(self.dtype)
         self.y = np.array(3.0).astype(self.dtype)
@@ -127,7 +127,7 @@ def init_input_output(self):
 
 
 @skip_check_grad_ci(
-    reason="oneDNN's int8 elementwise_ops don't implemend grad kernel."
+    reason="oneDNN's int8 elementwise_ops don't implement grad kernel."
 )
 class TestInt8(TestElementwiseAddOp):
     def init_kernel_type(self):
diff --git a/test/mkldnn/test_fused_vit_attention.py b/test/mkldnn/test_fused_vit_attention.py
index 8c4876e4281b25..b980f8bff912a8 100644
--- a/test/mkldnn/test_fused_vit_attention.py
+++ b/test/mkldnn/test_fused_vit_attention.py
@@ -23,7 +23,7 @@
 np.random.seed(0)
 
 
-def test_fuse_resenet_unit():
+def test_fuse_resnet_unit():
     tests = [[1, 4096, 768, 12], [10, 4097, 756, 12], [10, 4097, 756, 12]]
     for test in tests:
         batch_size = test[0]
@@ -74,4 +74,4 @@ def test_fuse_resenet_unit():
 
 
 if __name__ == '__main__':
-    test_fuse_resenet_unit()
+    test_fuse_resnet_unit()
diff --git a/test/mkldnn/test_fusion_gru_bf16_mkldnn_op.py b/test/mkldnn/test_fusion_gru_bf16_mkldnn_op.py
index ae44798dce4eb3..fd9925df082e07 100644
--- a/test/mkldnn/test_fusion_gru_bf16_mkldnn_op.py
+++ b/test/mkldnn/test_fusion_gru_bf16_mkldnn_op.py
@@ -59,7 +59,7 @@ def setUp(self):
         N = len(self.lod[0])
 
         # fp32 X input for reference implementation and
-        # corressponding bf16 data as input to GRU oneDNN bf16 kernel
+        # corresponding bf16 data as input to GRU oneDNN bf16 kernel
         x_fp32 = np.random.rand(T, self.M).astype('float32')
         x_bf16 = convert_float_to_uint16(x_fp32)