diff --git a/python/paddle/fluid/contrib/slim/tests/test_moving_average_abs_max_scale_op.py b/python/paddle/fluid/contrib/slim/tests/test_moving_average_abs_max_scale_op.py
index 1ec463192d9198c527fd84c9d8ee805ce001ca7a..201aa9c1b4ba24276cb38607f0c49f07c40e73a7 100644
--- a/python/paddle/fluid/contrib/slim/tests/test_moving_average_abs_max_scale_op.py
+++ b/python/paddle/fluid/contrib/slim/tests/test_moving_average_abs_max_scale_op.py
@@ -60,7 +60,7 @@ class TestMovingAverageAbsMaxScaleOp(unittest.TestCase):
         moving_average_abs_max_scale_ops = [
             op
             for op in main_program.blocks[0].ops
-            if op.type == u'moving_average_abs_max_scale'
+            if op.type == 'moving_average_abs_max_scale'
         ]
         assert (
             len(moving_average_abs_max_scale_ops) == 1
diff --git a/python/paddle/fluid/data_feeder.py b/python/paddle/fluid/data_feeder.py
index 1f900eff0df87f7636bc5667b547ad61fa9525e7..2728cbfb8d28f77c3e2266347bea983f86eb350f 100644
--- a/python/paddle/fluid/data_feeder.py
+++ b/python/paddle/fluid/data_feeder.py
@@ -50,6 +50,7 @@ def convert_dtype(dtype):
         if dtype in _PADDLE_DTYPE_2_NUMPY_DTYPE:
             return _PADDLE_DTYPE_2_NUMPY_DTYPE[dtype]
     elif isinstance(dtype, type):
+        # This branch is for NumPy scalar types
         if dtype in [
             bool,
             np.float16,
@@ -66,6 +67,7 @@ def convert_dtype(dtype):
         ]:
             return dtype.__name__
     else:
+        # This branch is for np.dtype and str
         if dtype in [
             'bool',
             'float16',
@@ -79,24 +81,10 @@ def convert_dtype(dtype):
             'uint8',
             'complex64',
             'complex128',
-            u'bool',
-            u'float16',
-            u'uint16',
-            u'float32',
-            u'float64',
-            u'int8',
-            u'int16',
-            u'int32',
-            u'int64',
-            u'uint8',
-            u'complex64',
-            u'complex128',
         ]:
-            # this code is a little bit dangerous, since error could happen
-            # when casting no-ascii code to str in python2.
-            # but since the set itself is limited, so currently, it is good.
-            # however, jointly supporting python2 and python3, (as well as python4 maybe)
-            # may still be a long-lasting problem.
+            # NOTE(SigureMo): Since the np.dtype object is not an instance of
+            # type, so it will not be handled by the previous branch. We need
+            # to convert it to str here.
             return str(dtype)
         # NOTE(zhangbo): Now numpy does not support bfloat, and paddle use uint16 to represent bfloat16, and there binaries are consistent.
         if dtype in ['bfloat16']:
@@ -104,7 +92,8 @@ def convert_dtype(dtype):
 
     raise TypeError(
         "dtype must be any of [bool, float16, uint16, float32, float64, int8, int16, "
-        "int32, int64, uint8, complex64, complex128], but received %s" % dtype
+        "int32, int64, uint8, complex64, complex128, bfloat16], but received %s"
+        % dtype
     )
 
 
diff --git a/python/paddle/fluid/tests/unittests/test_backward.py b/python/paddle/fluid/tests/unittests/test_backward.py
index 03b2c95105c83e2bf22565c414e2c09556f6fac0..98c2c05e60c58c4bf71de27b526149684a4fcd4c 100644
--- a/python/paddle/fluid/tests/unittests/test_backward.py
+++ b/python/paddle/fluid/tests/unittests/test_backward.py
@@ -188,24 +188,24 @@ class SimpleNet(BackwardNet):
         super().__init__()
         self.stop_gradient_grad_vars = set(
             [
-                u'x_no_grad@GRAD',
-                u'x2_no_grad@GRAD',
-                u'x3_no_grad@GRAD',
-                u'label_no_grad@GRAD',
+                'x_no_grad@GRAD',
+                'x2_no_grad@GRAD',
+                'x3_no_grad@GRAD',
+                'label_no_grad@GRAD',
             ]
         )
         self.no_grad_vars = set()
-        self.params_names = set([u'w2v', u'fc_predict.b_0', u'fc_w'])
+        self.params_names = set(['w2v', 'fc_predict.b_0', 'fc_w'])
         self.op_path = [
-            u'lookup_table_v2',
-            u'lookup_table_v2',  # embedding
-            u'elementwise_add',  # merge
-            u'mul',
-            u'elementwise_add',
-            u'softmax',  # fc
-            u'elementwise_sub',
-            u'square',
-            u'reduce_mean',
+            'lookup_table_v2',
+            'lookup_table_v2',  # embedding
+            'elementwise_add',  # merge
+            'mul',
+            'elementwise_add',
+            'softmax',  # fc
+            'elementwise_sub',
+            'square',
+            'reduce_mean',
         ]  # loss
         self.shape = [16, 50]
 
diff --git a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
index 9e2ff12bcad1f160ee939f4a344e14055bf39b47..7b5fe2c11434bf628bd4b4ac475929a0167f0047 100644
--- a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
+++ b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
@@ -1471,9 +1471,9 @@ class TestRemoteHsigmoid(TestDistLookupTableBase):
                 for name in ["epmap", "table_names", "epmap"]:
                     assert op.has_attr(name)
                     if name == "epmap":
-                        assert op.attr(name)[0] == u'127.0.0.1:6174'
+                        assert op.attr(name)[0] == '127.0.0.1:6174'
                     elif name == "table_names":
-                        assert op.attr(name)[0] == u'hierarchical_sigmoid_0.w_0'
+                        assert op.attr(name)[0] == 'hierarchical_sigmoid_0.w_0'
                     else:
                         assert op.attr(name) == 3
             elif op.type == "lookup_table":
@@ -1484,7 +1484,7 @@ class TestRemoteHsigmoid(TestDistLookupTableBase):
         for op in trainer.blocks[0].ops:
             if op.type == "recv":
                 assert len(op.output("Out")) == 1
-                assert op.output("Out")[0] == u'hierarchical_sigmoid_0.b_0'
+                assert op.output("Out")[0] == 'hierarchical_sigmoid_0.b_0'
                 op_count += 1
         assert op_count == 1
 
diff --git a/python/paddle/fluid/tests/unittests/test_feed_data_check_shape_type.py b/python/paddle/fluid/tests/unittests/test_feed_data_check_shape_type.py
index 3b885f26683cd45c80c7a25a7f0600d16784933e..764e5e75067480df86d2c408903b2031ac0667d8 100644
--- a/python/paddle/fluid/tests/unittests/test_feed_data_check_shape_type.py
+++ b/python/paddle/fluid/tests/unittests/test_feed_data_check_shape_type.py
@@ -106,7 +106,7 @@ class TestFeedData(unittest.TestCase):
                     "The fed Variable %r should have dimensions = %r, "
                     "shape = %r, but received fed shape %r on each device"
                     % (
-                        u'data',
+                        'data',
                         len(in_shape_tuple),
                         in_shape_tuple,
                         error_shape_list,
@@ -120,7 +120,7 @@ class TestFeedData(unittest.TestCase):
                 self.assertEqual(
                     str(dtype_mismatch_err.exception),
                     "The data type of fed Variable %r must be 'int64', but "
-                    "received 'float64'" % (u'label'),
+                    "received 'float64'" % ('label'),
                 )
 
     def _test_feed_data_dtype_mismatch(self, use_cuda, use_parallel_executor):
diff --git a/python/paddle/fluid/tests/unittests/test_print_op.py b/python/paddle/fluid/tests/unittests/test_print_op.py
index 0ce1230f52e868b3e133e9b58f1e20802b6f5f44..372515f4f025d1762daee8d77e3c1647cd337984 100755
--- a/python/paddle/fluid/tests/unittests/test_print_op.py
+++ b/python/paddle/fluid/tests/unittests/test_print_op.py
@@ -125,7 +125,7 @@ class TestPrintOpBackward(unittest.TestCase):
             loss = paddle.static.Print(loss)
             paddle.optimizer.Adam().minimize(loss)
 
-        print_ops = [op for op in main.blocks[0].ops if op.type == u'print']
+        print_ops = [op for op in main.blocks[0].ops if op.type == 'print']
         assert len(print_ops) == 2, "The number of print op should be 2"
 
         place = paddle.CUDAPlace(0) if use_cuda else paddle.CPUPlace()
diff --git a/python/paddle/framework/framework.py b/python/paddle/framework/framework.py
index e366b13077d2a35050ed2c1350e8fd8d1b8d2632..cafabb0d73ba85a0ccea29f236a7ec6138e51528 100644
--- a/python/paddle/framework/framework.py
+++ b/python/paddle/framework/framework.py
@@ -40,6 +40,7 @@ def set_default_dtype(d):
 
     """
     if isinstance(d, type):
+        # This branch is for NumPy scalar types
         if d in [np.float16, np.float32, np.float64]:
             d = d.__name__
         else:
@@ -48,19 +49,11 @@ def set_default_dtype(d):
                 ", but received %s" % d.__name__
             )
     else:
-        if d in [
-            'float16',
-            'float32',
-            'float64',
-            u'float16',
-            u'float32',
-            u'float64',
-        ]:
-            # this code is a little bit dangerous, since error could happen
-            # when casting no-ascii code to str in python2.
-            # but since the set itself is limited, so currently, it is good.
-            # however, jointly supporting python2 and python3, (as well as python4 maybe)
-            # may still be a long-lasting problem.
+        # This branch is for np.dtype and str
+        if d in ['float16', 'float32', 'float64']:
+            # NOTE(SigureMo): Since the np.dtype object is not an instance of
+            # type, so it will not be handled by the previous branch. We need
+            # to convert it to str here.
             d = str(d)
         else:
             raise TypeError(
diff --git a/python/paddle/tensor/math.py b/python/paddle/tensor/math.py
index 22aef853314c72fdabe57f95cda3533856a1f41f..a47ca0ed06b9cba00c72d6e2ad272f3dbe0023f1 100644
--- a/python/paddle/tensor/math.py
+++ b/python/paddle/tensor/math.py
@@ -1357,14 +1357,6 @@ def sum(x, axis=None, dtype=None, keepdim=False, name=None):
             'int64',
             'complex64',
             'complex128',
-            u'bool',
-            u'float16',
-            u'float32',
-            u'float64',
-            u'int32',
-            u'int64',
-            u'complex64',
-            u'complex128',
         ],
         'sum',
     )
diff --git a/tools/prune_for_jetson.py b/tools/prune_for_jetson.py
index 8bfe9ebf9f8dd7b7a4c24036ea4a9b8090d37504..14d1fb3ef2e2878c3b7d3a43ade0a6bce6e2021d 100644
--- a/tools/prune_for_jetson.py
+++ b/tools/prune_for_jetson.py
@@ -90,7 +90,7 @@ def prune_phi_kernels():
             content = content.replace(p, '')
 
         with open(op_file, 'w', encoding='utf-8') as f:
-            f.write(u'{}'.format(content))
+            f.write(content)
 
     print('We erase all grad op and kernel for Paddle-Inference lib.')
     print('%50s%10s' % ('type', 'count'))
@@ -137,7 +137,7 @@ def append_fluid_kernels():
         return False
 
     with open(file_name, 'w', encoding='utf-8') as f:
-        f.write(u'{}'.format(new_content))
+        f.write(new_content)
 
     # 2. add op and kernel register
     op_white_list.append("tensorrt_engine")
@@ -170,7 +170,7 @@ def append_fluid_kernels():
                         matches[0], matches[0].replace(k, k + "__")
                     )
                     with open(op_file, 'w', encoding='utf-8') as f:
-                        f.write(u'{}'.format(content))
+                        f.write(content)
 
     return True
 
diff --git a/tools/remove_grad_op_and_kernel.py b/tools/remove_grad_op_and_kernel.py
index 8aa456bdb671f6e75d34b5cf58779055a08ad21a..85bfed9640da8406d5522cb1f720016d3eaaeb16 100644
--- a/tools/remove_grad_op_and_kernel.py
+++ b/tools/remove_grad_op_and_kernel.py
@@ -170,7 +170,7 @@ if __name__ == '__main__':
             content = content.replace(i, '')
 
         with open(op_file, 'w', encoding='utf-8') as f:
-            f.write(u'{}'.format(content))
+            f.write(content)
 
     # 2. update operators/CMakeLists.txt
     cmake_file = os.path.join(