diff --git a/python/paddle/dataset/wmt16.py b/python/paddle/dataset/wmt16.py
index cc3857aca982a868ddaadd5be7b02e50274e7d67..64ddc2214716e6b95bb6b87473b3acabe178357d 100644
--- a/python/paddle/dataset/wmt16.py
+++ b/python/paddle/dataset/wmt16.py
@@ -62,15 +62,14 @@ def __build_dict(tar_file, dict_size, save_path, lang):
                 word_dict[w] += 1
 
     with open(save_path, "wb") as fout:
-        fout.write(
-            cpt.to_bytes("%s\n%s\n%s\n" % (START_MARK, END_MARK, UNK_MARK)))
+        fout.write(("%s\n%s\n%s\n" % (START_MARK, END_MARK, UNK_MARK)).encode())
         for idx, word in enumerate(
                 sorted(six.iteritems(word_dict),
                        key=lambda x: x[1],
                        reverse=True)):
             if idx + 3 == dict_size: break
-            fout.write(cpt.to_bytes(word[0]))
-            fout.write(cpt.to_bytes('\n'))
+            fout.write(word[0].encode())
+            fout.write(b'\n')
 
 
 def __load_dict(tar_file, dict_size, lang, reverse=False):
diff --git a/python/paddle/fluid/backward.py b/python/paddle/fluid/backward.py
index ac2fd8caf8e85d6f95263165ccd709596ee278fa..e9babc79c3ac88b0e30b96f61d7d534748cb03a3 100755
--- a/python/paddle/fluid/backward.py
+++ b/python/paddle/fluid/backward.py
@@ -391,10 +391,10 @@ def _infer_var_data_type_shape_(grad_var_name, block):
     """
     Infer the data type and shape of given grad variable
     """
-    grad_var = block.desc.find_var(cpt.to_bytes(grad_var_name))
+    grad_var = block.desc.find_var(grad_var_name.encode())
     fwd_name = _strip_grad_suffix_(grad_var_name)
-    if block.desc.has_var_recursive(cpt.to_bytes(fwd_name)):
-        fwd_var = block.desc.find_var_recursive(cpt.to_bytes(fwd_name))
+    if block.desc.has_var_recursive(fwd_name.encode()):
+        fwd_var = block.desc.find_var_recursive(fwd_name.encode())
         grad_var.set_dtype(fwd_var.dtype())
         grad_var.set_shape(fwd_var.shape())
     else:
@@ -1457,7 +1457,7 @@ def _append_backward_vars_(block, start_op_idx, grad_to_var, grad_info_map):
             if grad_var_ins:
                 existing_grad_var_ins = [
                     var for var in grad_var_ins
-                    if block.desc.has_var_recursive(cpt.to_bytes(var))
+                    if block.desc.has_var_recursive(var.encode())
                     or var in parent_op_vars
                 ]
                 if not existing_grad_var_ins:
@@ -1476,10 +1476,10 @@ def _append_backward_vars_(block, start_op_idx, grad_to_var, grad_info_map):
         new_vars = set()
         # create new gradient variables
         for grad_var_name in op_desc.output_arg_names():
-            if block.desc.has_var_recursive(cpt.to_bytes(
-                    grad_var_name)) or grad_var_name == core.empty_var_name():
+            if block.desc.has_var_recursive(grad_var_name.encode(
+            )) or grad_var_name == core.empty_var_name():
                 continue
-            block.desc.var(cpt.to_bytes(grad_var_name))
+            block.desc.var(grad_var_name.encode())
             new_vars.add(grad_var_name)
             if grad_var_name not in grad_to_var:
                 continue
@@ -1930,8 +1930,8 @@ def _get_output_names(cur_block, targets):
             if _some_in_set_(op.desc.output_arg_names(), current_output_names):
                 for name in op.desc.input_arg_names():
                     current_output_names.add(name)
-                    if not block.desc.find_var(cpt.to_bytes(name)) \
-                            and parent_block.desc.find_var(cpt.to_bytes(name)):
+                    if not block.desc.find_var(name.encode()) \
+                            and parent_block.desc.find_var(name.encode()):
                         parent_block_output_names.add(name)
 
         block = parent_block
diff --git a/python/paddle/fluid/dygraph/dygraph_to_static/partial_program.py b/python/paddle/fluid/dygraph/dygraph_to_static/partial_program.py
index 8592e9e8919520549a21df94c17547f2e69788e3..2e545a91a990916800638a37d070e4ad012b2155 100644
--- a/python/paddle/fluid/dygraph/dygraph_to_static/partial_program.py
+++ b/python/paddle/fluid/dygraph/dygraph_to_static/partial_program.py
@@ -30,8 +30,7 @@ from paddle.fluid.framework import _apply_pass
 from paddle.fluid.contrib.mixed_precision.decorator import AutoMixedPrecisionLists
 from paddle.fluid.contrib.mixed_precision.fp16_utils import rewrite_program, cast_model_to_fp16
 from paddle.fluid.dygraph.amp.auto_cast import _in_amp_guard, _in_pure_fp16_guard
-import paddle.compat as cpt
-from paddle import _C_ops, _legacy_C_ops
+from paddle import _legacy_C_ops
 
 
 class NestSequence(object):
@@ -833,8 +832,7 @@ class PartialProgramLayer:
         # be user wanted result.
         for param in params:
             grad_name = param.name + core.grad_var_suffix()
-            grad_var = train_program.desc.block(0).find_var(
-                cpt.to_bytes(grad_name))
+            grad_var = train_program.desc.block(0).find_var(grad_name.encode())
             # NOTE: cannot find var desc maybe no problem, such as in batch_norm
             if grad_var is None:
                 continue
diff --git a/python/paddle/fluid/dygraph/io.py b/python/paddle/fluid/dygraph/io.py
index cf5a85b44f0b432906e2b57ac813adc021e5e626..6671499e1c4ef9078203d11571522291cc2e0b74 100644
--- a/python/paddle/fluid/dygraph/io.py
+++ b/python/paddle/fluid/dygraph/io.py
@@ -145,7 +145,7 @@ def _append_loaded_suffix_to_var(program_desc):
         var_desc.set_name(new_name)
         for block_idx in six.moves.range(program_desc.num_blocks()):
             block = program_desc.block(block_idx)
-            block._rename_var(cpt.to_bytes(old_name), cpt.to_bytes(new_name))
+            block._rename_var(old_name.encode(), new_name.encode())
             for op_idx in six.moves.range(block.op_size()):
                 op = block.op(op_idx)
                 op._rename_input(old_name, new_name)
@@ -224,8 +224,7 @@ def _rename_var_program_desc(program_desc, include=None, exclude=None):
             else:
                 name_new = name_old
             if name_old != name_new:
-                cur_block._rename_var(cpt.to_bytes(name_old),
-                                      cpt.to_bytes(name_new))
+                cur_block._rename_var(name_old.encode(), name_new.encode())
             if not is_double_grad_var:
                 dict_rename_var_old_new[name_old] = name_new
                 dict_rename_var_new_old[name_new] = name_old
@@ -259,11 +258,11 @@ def _rename_var_program_desc(program_desc, include=None, exclude=None):
                         op._rename_input(
                             input_arg_name,
                             dict_rename_var_old_new[input_arg_name])
-                        if cur_block.has_var(cpt.to_bytes(input_arg_name)):
+                        if cur_block.has_var(input_arg_name.encode()):
                             cur_block._rename_var(
-                                cpt.to_bytes(input_arg_name),
-                                cpt.to_bytes(
-                                    dict_rename_var_old_new[input_arg_name]))
+                                input_arg_name.encode(),
+                                dict_rename_var_old_new[input_arg_name].encode(
+                                ))
             for output_arg_name in op.output_arg_names():
                 if output_arg_name in dict_rename_var_old_new:
                     if output_arg_name != dict_rename_var_old_new[
@@ -271,11 +270,11 @@ def _rename_var_program_desc(program_desc, include=None, exclude=None):
                         op._rename_output(
                             output_arg_name,
                             dict_rename_var_old_new[output_arg_name])
-                        if cur_block.has_var(cpt.to_bytes(output_arg_name)):
+                        if cur_block.has_var(output_arg_name.encode()):
                             cur_block._rename_var(
-                                cpt.to_bytes(output_arg_name),
-                                cpt.to_bytes(
-                                    dict_rename_var_old_new[output_arg_name]))
+                                output_arg_name.encode(),
+                                dict_rename_var_old_new[output_arg_name].encode(
+                                ))
     program_desc.flush()
     return dict_rename_var_new_old, dict_rename_var_old_new
 
@@ -413,25 +412,25 @@ class _ProgramHolder(object):
             op = root_block.op(i)
             if op.type() == 'feed':
                 ops_to_remove.append(i)
-                feed_var_name = cpt.to_bytes(op.input('X')[0])
+                feed_var_name = op.input('X')[0].encode()
                 root_block._remove_var(feed_var_name)
                 self._input_descs.append(
-                    root_block.find_var(cpt.to_bytes(op.output('Out')[0])))
+                    root_block.find_var(op.output('Out')[0].encode()))
             elif op.type() == 'scale' and op.output('Out')[0].startswith(
                     'save_infer_model/scale_'):
                 ops_to_remove.append(i)
-                out_var_name = cpt.to_bytes(op.output('Out')[0])
+                out_var_name = op.output('Out')[0].encode()
                 root_block._remove_var(out_var_name)
                 self._output_descs.append(
-                    root_block.find_var(cpt.to_bytes(op.input('X')[0])))
+                    root_block.find_var(op.input('X')[0].encode()))
             elif op.type() == 'fetch':
                 ops_to_remove.append(i)
-                fetch_var_name = cpt.to_bytes(op.output('Out')[0])
+                fetch_var_name = op.output('Out')[0].encode()
                 root_block._remove_var(fetch_var_name)
                 # NOTE: some old pre-train models have no extra scale_op
                 if not op.input('X')[0].startswith('save_infer_model/scale_'):
                     self._output_descs.append(
-                        root_block.find_var(cpt.to_bytes(op.input('X')[0])))
+                        root_block.find_var(op.input('X')[0].encode()))
             else:
                 if op.has_attr("op_callstack"):
                     op.remove_attr("op_callstack")
@@ -937,7 +936,7 @@ def _run_dygraph(instance, input, program_holder):
     # be user wanted result.
     for persistable_var in persistable_vars:
         grad_var_name = persistable_var.name + core.grad_var_suffix()
-        grad_var = trace_program.block(0).find_var(cpt.to_bytes(grad_var_name))
+        grad_var = trace_program.block(0).find_var(grad_var_name.encode())
         # NOTE: cannot find var desc maybe not problem,
         # such as in batch_norm
         if grad_var is None:
diff --git a/python/paddle/fluid/executor.py b/python/paddle/fluid/executor.py
index 3eb8c23a07b69e1e6204dcb6aad994bece924d22..09fe42e705742b67dd65825fca4be0072dbfbed1 100755
--- a/python/paddle/fluid/executor.py
+++ b/python/paddle/fluid/executor.py
@@ -27,7 +27,6 @@ from .framework import convert_np_dtype_to_dtype_, _apply_pass
 from . import core
 from . import unique_name
 from . import compiler
-from .. import compat as cpt
 from .trainer_factory import TrainerFactory
 from .trainer_factory import FetchHandlerMonitor
 import copy
@@ -1674,7 +1673,7 @@ class Executor(object):
             else:
                 global_block = program.global_block()
             for varname in global_block.vars:
-                vardesc = global_block.desc.find_var(cpt.to_bytes(varname))
+                vardesc = global_block.desc.find_var(varname.encode())
                 varobj = global_block.vars[varname]
 
                 # Can not check var build by fluid.layers.data(), bucause fluid.layers.data() had not set need_check_feed
diff --git a/python/paddle/fluid/framework.py b/python/paddle/fluid/framework.py
index 49c0bb24d6d203ef62e8fd877041ae229de6caa7..e709a02b388af020b1942c371b20b484fd02975f 100644
--- a/python/paddle/fluid/framework.py
+++ b/python/paddle/fluid/framework.py
@@ -1391,10 +1391,10 @@ class Variable(object):
 
         is_new_var = False
         name = cpt.to_text(name)
-        self.desc = self.block.desc.find_var(cpt.to_bytes(name))
+        self.desc = self.block.desc.find_var(name.encode())
 
         if self.desc is None:
-            self.desc = self.block.desc.var(cpt.to_bytes(name))
+            self.desc = self.block.desc.var(name.encode())
             is_new_var = True
 
         if is_new_var:
@@ -3691,9 +3691,9 @@ class Block(object):
         else:
             raise ValueError("unsupported var type: %s", type(v))
         orig_var_type = v.type
-        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
+        self.desc._rename_var(name.encode(), new_name.encode())
         # NOTE: v is destroyed by C++ after calling _rename_var.
-        d = self.desc.find_var(cpt.to_bytes(new_name))
+        d = self.desc.find_var(new_name.encode())
         if var_type == "Parameter":
             if in_dygraph_mode():
                 var = EagerParamBase(d.shape(),
@@ -3744,7 +3744,7 @@ class Block(object):
     def _remove_var(self, name, sync=True):
         if sync == True:
             self._sync_with_cpp()
-        self.desc._remove_var(cpt.to_bytes(name))
+        self.desc._remove_var(name.encode())
         del self.vars[name]
 
     def create_parameter(self, *args, **kwargs):
@@ -3953,7 +3953,7 @@ class Block(object):
 
         # sync variables removed from c++ end
         for var in list(self.vars.keys()):
-            if not self.desc.find_var(cpt.to_bytes(var)):
+            if not self.desc.find_var(var.encode()):
                 self.vars.pop(var)
 
         # sync operators from cpp
@@ -5839,7 +5839,7 @@ class Program(object):
                 root_block._remove_op(0, read_op_idx + 1)
             for var in root_block.all_vars():
                 if var.type() == core.VarDesc.VarType.READER:
-                    root_block._remove_var(cpt.to_bytes(var.name()))
+                    root_block._remove_var(var.name().encode())
 
         # change all `is_test` attributes to True
         for i in six.moves.range(res.desc.num_blocks()):
diff --git a/python/paddle/fluid/ir.py b/python/paddle/fluid/ir.py
index aca134a1df55a9982a3779e03313481cce712e5f..3e58af416d506fe738431c67ef0745f18c19cdda 100644
--- a/python/paddle/fluid/ir.py
+++ b/python/paddle/fluid/ir.py
@@ -158,7 +158,7 @@ class RegisterPassHelper(object):
     def _prune_program_desc(self, ops):
         for op_desc in ops:
             default_attrs = core.get_op_attrs_default_value(
-                paddle.compat.to_bytes(op_desc.type))
+                op_desc.type.encode())
             remove_attrs = list()
             for attr in op_desc.attrs:
                 # attr must not in
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index a6044e926d03cdcb3f87a5611efad6fec4a3d9fc..289cb9b9044abdfeccb26cb14486fba6473dd522 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -2450,10 +2450,10 @@ class ConditionalBlock(object):
 
         new_vars = set()
         for grad_var_name in new_op_desc.output_arg_names():
-            if grad_sub_block.desc.has_var_recursive(cpt.to_bytes(
-                    grad_var_name)) or grad_var_name == core.empty_var_name():
+            if grad_sub_block.desc.has_var_recursive(grad_var_name.encode(
+            )) or grad_var_name == core.empty_var_name():
                 continue
-            grad_sub_block.desc.var(cpt.to_bytes(grad_var_name))
+            grad_sub_block.desc.var(grad_var_name.encode())
             new_vars.add(grad_var_name)
             if grad_var_name not in op_grad_to_var:
                 continue
diff --git a/python/paddle/fluid/tests/unittests/ir/inference/program_config.py b/python/paddle/fluid/tests/unittests/ir/inference/program_config.py
index abb353977b4e9d01fe0e95b3278ec931f0a7df86..5ad6f1c885cf95a28faa8f2c7a4cdf161fc9ed27 100644
--- a/python/paddle/fluid/tests/unittests/ir/inference/program_config.py
+++ b/python/paddle/fluid/tests/unittests/ir/inference/program_config.py
@@ -18,7 +18,6 @@ import enum
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
-from paddle import compat as cpt
 from paddle.fluid.initializer import NumpyArrayInitializer
 from paddle.fluid.framework import convert_np_dtype_to_dtype_
 
@@ -120,7 +119,7 @@ class BlockConfig:
 
     def fill_block_desc(self, block_desc):
         for name in self.vars:
-            var_desc = block_desc.var(cpt.to_bytes(name))
+            var_desc = block_desc.var(name.encode())
             var_desc.set_type(core.VarDesc.VarType.LOD_TENSOR)
             if self.vars_lod_level is not None and name in self.vars_lod_level.keys(
             ):
@@ -147,9 +146,9 @@ class BlockConfig:
             for name, values in op_config.outputs.items():
                 op_desc.set_output(name, values)
                 for v in values:
-                    if block_desc.has_var_recursive(cpt.to_bytes(v)):
+                    if block_desc.has_var_recursive(v.encode()):
                         continue
-                    var_desc = block_desc.var(cpt.to_bytes(v))
+                    var_desc = block_desc.var(v.encode())
                     var_desc.set_type(core.VarDesc.VarType.LOD_TENSOR)
                     if op_config.outputs_var_type is not None and v in op_config.outputs_var_type.keys(
                     ):
@@ -216,13 +215,13 @@ def create_fake_model(program_config):
     util_program = fluid.Program()
     main_block_desc = main_program_desc.block(0)
 
-    var_desc = main_block_desc.var(cpt.to_bytes("feed"))
+    var_desc = main_block_desc.var(b"feed")
     var_desc.set_type(core.VarDesc.VarType.FEED_MINIBATCH)
     var_desc.set_persistable(True)
 
     index = 0
     for name, tensor_config in program_config.inputs.items():
-        var_desc = main_block_desc.var(cpt.to_bytes(name))
+        var_desc = main_block_desc.var(name.encode())
         var_desc.set_type(core.VarDesc.VarType.LOD_TENSOR)
         var_desc.set_dtype(convert_np_dtype_to_dtype_(tensor_config.dtype))
         var_desc.set_shape(tensor_config.shape)
@@ -239,7 +238,7 @@ def create_fake_model(program_config):
 
     save_var_map = {}
     for name, tensor_config in program_config.weights.items():
-        var_desc = main_block_desc.var(cpt.to_bytes(name))
+        var_desc = main_block_desc.var(name.encode())
         var_desc.set_type(core.VarDesc.VarType.LOD_TENSOR)
         var_desc.set_dtype(convert_np_dtype_to_dtype_(tensor_config.dtype))
         var_desc.set_shape(tensor_config.shape)
@@ -280,9 +279,9 @@ def create_fake_model(program_config):
         for name, values in op_config.outputs.items():
             op_desc.set_output(name, values)
             for v in values:
-                if main_block_desc.has_var_recursive(cpt.to_bytes(v)):
+                if main_block_desc.has_var_recursive(v.encode()):
                     continue
-                var_desc = main_block_desc.var(cpt.to_bytes(v))
+                var_desc = main_block_desc.var(v.encode())
                 var_desc.set_type(core.VarDesc.VarType.LOD_TENSOR)
                 if op_config.outputs_var_type is not None and v in op_config.outputs_var_type.keys(
                 ):
@@ -303,7 +302,7 @@ def create_fake_model(program_config):
         op_desc.check_attrs()
 
     for index, name in enumerate(program_config.outputs):
-        var_desc = main_block_desc.var(cpt.to_bytes("fetch"))
+        var_desc = main_block_desc.var(b"fetch")
         var_desc.set_type(core.VarDesc.VarType.FETCH_LIST)
         var_desc.set_need_check_feed(True)
         op_desc = main_block_desc.append_op()
diff --git a/python/paddle/fluid/tests/unittests/npu/test_run_program_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_run_program_op_npu.py
index 61df52cfd08348f559ccd36c22bc69fe1681a098..5401228d2835c515c4436c589c46fb58ed8247a2 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_run_program_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_run_program_op_npu.py
@@ -24,7 +24,6 @@ from op_test import OpTest
 import paddle
 from paddle import _C_ops, _legacy_C_ops
 import paddle.fluid as fluid
-from paddle import compat as cpt
 from paddle.fluid import core, framework, executor
 from paddle.fluid.layers.utils import _hash_with_id
 from paddle.fluid.framework import _in_eager_mode_
@@ -241,7 +240,7 @@ class RunProgramNPUOpTest(unittest.TestCase):
         grad_name = name + core.grad_var_suffix()
         for i in six.moves.range(self.program_desc.num_blocks()):
             block = self.program_desc.block(i)
-            var_desc = block.find_var_recursive(cpt.to_bytes(grad_name))
+            var_desc = block.find_var_recursive(grad_name.encode())
             return var_desc.type() if var_desc is not None else None
 
 
diff --git a/python/paddle/fluid/tests/unittests/test_eager_run_program.py b/python/paddle/fluid/tests/unittests/test_eager_run_program.py
index ad4ada7c3a5fd03f324bfc4c8f6023bb56fff53a..df151f27ec21a5ffd9026ac7e9a52a4cfb76a7d6 100644
--- a/python/paddle/fluid/tests/unittests/test_eager_run_program.py
+++ b/python/paddle/fluid/tests/unittests/test_eager_run_program.py
@@ -48,7 +48,7 @@ def _append_backward_desc(main_program, outs):
 #     for param in params:
 #         grad_name = param.name + core.grad_var_suffix()
 #         grad_var = train_program.desc.block(0).find_var(
-#             cpt.to_bytes(grad_name))
+#             grad_name.encode())
 #         # NOTE: cannot find var desc maybe no problem, such as in batch_norm
 #         if grad_var is None:
 #             continue
diff --git a/python/paddle/fluid/tests/unittests/test_protobuf_descs.py b/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
index ee522bc77f753ee9a86367ae46aab43b020869c4..6022bf02602cbd01e35a4f4e04b6811d92821d56 100644
--- a/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
+++ b/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
@@ -14,7 +14,6 @@
 
 import unittest
 import paddle.fluid.core as core
-import paddle.compat as cpt
 from paddle.fluid.framework import Program
 
 
@@ -112,7 +111,7 @@ class TestVarDesc(unittest.TestCase):
     def test_shape(self):
         program_desc = core.ProgramDesc()
         block = program_desc.block(0)
-        var = block.var(cpt.to_bytes('my_var'))
+        var = block.var(b'my_var')
         var.set_type(core.VarDesc.VarType.SELECTED_ROWS)
         src_shape = [3, 2, 10, 8]
         var.set_shape(src_shape)
@@ -123,7 +122,7 @@ class TestVarDesc(unittest.TestCase):
     def test_multiple_shape(self):
         program_desc = core.ProgramDesc()
         block = program_desc.block(0)
-        var = block.var(cpt.to_bytes('my_reader'))
+        var = block.var(b'my_reader')
         var.set_type(core.VarDesc.VarType.READER)
         src_shapes = [[2, 3, 3], [4, 5], [6, 7, 8, 9]]
         var.set_shapes(src_shapes)
@@ -134,7 +133,7 @@ class TestVarDesc(unittest.TestCase):
     def test_dtype(self):
         program_desc = core.ProgramDesc()
         block = program_desc.block(0)
-        var = block.var(cpt.to_bytes('my_var'))
+        var = block.var(b'my_var')
         var.set_type(core.VarDesc.VarType.LOD_TENSOR)
         var.set_dtype(core.VarDesc.VarType.INT32)
         self.assertEqual(core.VarDesc.VarType.INT32, var.dtype())
@@ -143,7 +142,7 @@ class TestVarDesc(unittest.TestCase):
     def test_multiple_dtype(self):
         program_desc = core.ProgramDesc()
         block = program_desc.block(0)
-        var = block.var(cpt.to_bytes('my_reader'))
+        var = block.var(b'my_reader')
         var.set_type(core.VarDesc.VarType.READER)
         src_types = [
             core.VarDesc.VarType.INT32, core.VarDesc.VarType.FP64,
@@ -156,7 +155,7 @@ class TestVarDesc(unittest.TestCase):
     def test_multiple_lod_level(self):
         program_desc = core.ProgramDesc()
         block = program_desc.block(0)
-        var = block.var(cpt.to_bytes('my_reader'))
+        var = block.var(b'my_reader')
         var.set_type(core.VarDesc.VarType.READER)
         src_types = [3, 1, 2]
         var.set_lod_levels(src_types)
@@ -171,12 +170,12 @@ class TestBlockDesc(unittest.TestCase):
         self.assertIsNotNone(program_desc)
         block = program_desc.block(0)
         self.assertIsNotNone(block)
-        var1 = block.var(cpt.to_bytes("var1"))
-        var2 = block.var(cpt.to_bytes("var2"))
-        var3 = block.var(cpt.to_bytes("var3"))
+        var1 = block.var(b"var1")
+        var2 = block.var(b"var2")
+        var3 = block.var(b"var3")
         all_vars = block.all_vars()
         self.assertEqual(set(all_vars), {var1, var2, var3})
-        var2_re = block.find_var(cpt.to_bytes("var2"))
+        var2_re = block.find_var(b"var2")
         self.assertEqual(var2_re, var2)
 
     def test_add_op(self):
diff --git a/python/paddle/fluid/tests/unittests/test_pybind_interface.py b/python/paddle/fluid/tests/unittests/test_pybind_interface.py
index efb2c1d7da5943c3587f9c4bb05b642d51556c5a..854aa7f3e9e76a99425e3e4b8704fe3ccad66020 100644
--- a/python/paddle/fluid/tests/unittests/test_pybind_interface.py
+++ b/python/paddle/fluid/tests/unittests/test_pybind_interface.py
@@ -14,14 +14,13 @@
 
 import unittest
 from paddle.fluid import core
-from paddle import compat as cpt
 
 
 class TestPybindInference(unittest.TestCase):
 
     # call get_op_attrs_default_value for c++ coverage rate
     def test_get_op_attrs_default_value(self):
-        core.get_op_attrs_default_value(cpt.to_bytes("fill_constant"))
+        core.get_op_attrs_default_value(b"fill_constant")
 
         # the default values of Op 'fill_constant'
         #
diff --git a/python/paddle/fluid/tests/unittests/test_run_program_op.py b/python/paddle/fluid/tests/unittests/test_run_program_op.py
index 56fda34c209387e54ffb04fcfdb64c667d4a882a..89642bb66168b411fae88de445a40c95799e5b99 100644
--- a/python/paddle/fluid/tests/unittests/test_run_program_op.py
+++ b/python/paddle/fluid/tests/unittests/test_run_program_op.py
@@ -20,7 +20,6 @@ import six
 import paddle
 from paddle import _legacy_C_ops
 import paddle.fluid as fluid
-from paddle import compat as cpt
 from paddle.fluid import core, framework
 from paddle.fluid.layers.utils import _hash_with_id
 from paddle.fluid.framework import _in_eager_mode_
@@ -301,7 +300,7 @@ class RunProgramOpTest(unittest.TestCase):
         grad_name = name + core.grad_var_suffix()
         for i in six.moves.range(self.program_desc.num_blocks()):
             block = self.program_desc.block(i)
-            var_desc = block.find_var_recursive(cpt.to_bytes(grad_name))
+            var_desc = block.find_var_recursive(grad_name.encode())
             return var_desc.type() if var_desc is not None else None
 
 
diff --git a/python/paddle/incubate/autograd/primx.py b/python/paddle/incubate/autograd/primx.py
index e61a77791fd19a77fd0e479400ee29178d0cd98c..c57f651d344c34b156e2d37fab39d1049db33adc 100644
--- a/python/paddle/incubate/autograd/primx.py
+++ b/python/paddle/incubate/autograd/primx.py
@@ -15,7 +15,6 @@
 from collections import OrderedDict
 
 import paddle
-from paddle import compat as cpt
 from paddle.fluid import framework as framework
 from paddle.fluid.framework import Operator, default_main_program
 from paddle.incubate.autograd.utils import as_tensors
@@ -220,7 +219,7 @@ class Transform(object):
         block = self.block
         for var in vars_to_erase:
             name = var.name
-            block.desc._remove_var(cpt.to_bytes(name))
+            block.desc._remove_var(name.encode())
             del block.vars[name]
         block._sync_with_cpp()
 
@@ -512,7 +511,7 @@ def _lower(block, reverse, blacklist):
         assert var_name in to_bind_rev, 'var_name "{}" is not in to_bind_rev.'.format(
             var_name)
         if var_name != to_bind_rev[var_name]:
-            block.desc._remove_var(cpt.to_bytes(var_name))
+            block.desc._remove_var(var_name.encode())
             del block.vars[var_name]
     block._sync_with_cpp()
 
diff --git a/python/paddle/text/datasets/wmt16.py b/python/paddle/text/datasets/wmt16.py
index 4cac2f1ed387a142423624d0705ffd90aab0fd41..40e464cb0ebae45f275fb7af883fc4bca03cc131 100644
--- a/python/paddle/text/datasets/wmt16.py
+++ b/python/paddle/text/datasets/wmt16.py
@@ -170,14 +170,14 @@ class WMT16(Dataset):
 
         with open(dict_path, "wb") as fout:
             fout.write(
-                cpt.to_bytes("%s\n%s\n%s\n" % (START_MARK, END_MARK, UNK_MARK)))
+                ("%s\n%s\n%s\n" % (START_MARK, END_MARK, UNK_MARK)).encode())
             for idx, word in enumerate(
                     sorted(six.iteritems(word_dict),
                            key=lambda x: x[1],
                            reverse=True)):
                 if idx + 3 == dict_size: break
-                fout.write(cpt.to_bytes(word[0]))
-                fout.write(cpt.to_bytes('\n'))
+                fout.write(word[0].encode())
+                fout.write(b'\n')
 
     def _load_data(self):
         # the index for start mark, end mark, and unk are the same in source
diff --git a/tools/infrt/generate_pd_op_dialect_from_paddle_op_maker.py b/tools/infrt/generate_pd_op_dialect_from_paddle_op_maker.py
index f2c04f3cba8f857eb151718a8fe7ac0c3c653880..087c88d937be632c0d92a77926c56921c6c91067 100644
--- a/tools/infrt/generate_pd_op_dialect_from_paddle_op_maker.py
+++ b/tools/infrt/generate_pd_op_dialect_from_paddle_op_maker.py
@@ -14,7 +14,6 @@
 
 import paddle.fluid.framework as framework
 from paddle.fluid import core
-from paddle import compat as cpt
 
 
 # collect original ops: op which has both inference and grid defination
@@ -59,7 +58,7 @@ QUANT = "quant"
 
 
 def get_attr_default_value(op_name):
-    return core.get_op_attrs_default_value(cpt.to_bytes(op_name))
+    return core.get_op_attrs_default_value(op_name.encode())
 
 
 def get_vars_info(op_vars_proto):
diff --git a/tools/print_op_desc.py b/tools/print_op_desc.py
index b85103a7a25e164c83880b518caa803d53923892..cf46849add4b1751b579ffdb25945f4b2d44e32f 100644
--- a/tools/print_op_desc.py
+++ b/tools/print_op_desc.py
@@ -45,7 +45,6 @@ Usage:
 import paddle.fluid.framework as framework
 from paddle.fluid import core
 import json
-from paddle import compat as cpt
 
 INPUTS = "Inputs"
 OUTPUTS = "Outputs"
@@ -64,7 +63,7 @@ QUANT = "quant"
 
 
 def get_attr_default_value(op_name):
-    return core.get_op_attrs_default_value(cpt.to_bytes(op_name))
+    return core.get_op_attrs_default_value(op_name.encode())
 
 
 def get_vars_info(op_vars_proto):