diff --git a/paddle/contrib/float16/float16_transpiler.py b/paddle/contrib/float16/float16_transpiler.py
index 66e0345c299730c113ffbdc8dd3c1fa32f872f3d..8d95dc0591e1d6bd815cc697528191c2ee8c1cfe 100644
--- a/paddle/contrib/float16/float16_transpiler.py
+++ b/paddle/contrib/float16/float16_transpiler.py
@@ -102,8 +102,8 @@ class Float16Transpiler:
                 continue
             for input_arg in current_op.input_arg_names:
                 if input_arg in self.input_map:
-                    current_op.rename_input(input_arg,
-                                            self.input_map[input_arg])
+                    current_op._rename_input(input_arg,
+                                             self.input_map[input_arg])
 
     def _remove_unused_var(self):
         '''
@@ -187,7 +187,7 @@ class Float16Transpiler:
                     shape=var.shape,
                     persistable=var.persistable)
                 find_op(var)
-                var.op.rename_output(var_name, tmp_var_name)
+                var.op._rename_output(var_name, tmp_var_name)
                 self.block._insert_op(
                     i,
                     type="cast",
diff --git a/paddle/fluid/API.spec b/paddle/fluid/API.spec
index 88b86f50b82680558778f7cd3068b2306ec4c9e8..ad06a494364011a0a51f9adddeae5630eb0d6bb4 100644
--- a/paddle/fluid/API.spec
+++ b/paddle/fluid/API.spec
@@ -6,26 +6,9 @@ paddle.fluid.Program.global_block ArgSpec(args=['self'], varargs=None, keywords=
 paddle.fluid.Program.list_vars ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Program.parse_from_string ArgSpec(args=['binary_str'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Program.to_string ArgSpec(args=['self', 'throw_on_error', 'with_details'], varargs=None, keywords=None, defaults=(False,))
-paddle.fluid.Operator.__init__ ArgSpec(args=['self', 'block', 'desc', 'type', 'inputs', 'outputs', 'attrs'], varargs=None, keywords=None, defaults=(None, None, None, None))
-paddle.fluid.Operator.all_attrs ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.attr ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.attr_type ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.block_attr ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.block_attr_id ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.blocks_attr ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.blocks_attr_ids ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.has_attr ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.has_kernel ArgSpec(args=['self', 'op_type'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.input ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.output ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.rename_input ArgSpec(args=['self', 'old_name', 'new_name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.rename_output ArgSpec(args=['self', 'old_name', 'new_name'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.set_attr ArgSpec(args=['self', 'name', 'val'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.Operator.to_string ArgSpec(args=['self', 'throw_on_error'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.default_startup_program ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.default_main_program ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.program_guard ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
-paddle.fluid.get_var ArgSpec(args=['name', 'program'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.name_scope ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
 paddle.fluid.Executor.__init__ ArgSpec(args=['self', 'place'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Executor.close ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
@@ -170,6 +153,13 @@ paddle.fluid.layers.elementwise_mul ArgSpec(args=['x', 'y', 'out', 'axis', 'use_
 paddle.fluid.layers.elementwise_max ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
 paddle.fluid.layers.elementwise_min ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
 paddle.fluid.layers.elementwise_pow ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.uniform_random_batch_size_like ArgSpec(args=['input', 'shape', 'dtype', 'input_dim_idx', 'output_dim_idx', 'min', 'max', 'seed'], varargs=None, keywords=None, defaults=('float32', 0, 0, -1.0, 1.0, 0))
+paddle.fluid.layers.gaussian_random ArgSpec(args=['shape', 'mean', 'std', 'seed', 'dtype', 'use_mkldnn'], varargs=None, keywords=None, defaults=(0.0, 1.0, 0, 'float32', False))
+paddle.fluid.layers.sampling_id ArgSpec(args=['x', 'min', 'max', 'seed', 'dtype'], varargs=None, keywords=None, defaults=(0.0, 1.0, 0, 'float32'))
+paddle.fluid.layers.gaussian_random_batch_size_like ArgSpec(args=['input', 'shape', 'input_dim_idx', 'output_dim_idx', 'mean', 'std', 'seed', 'dtype'], varargs=None, keywords=None, defaults=(0, 0, 0.0, 1.0, 0, 'float32'))
+paddle.fluid.layers.sum ArgSpec(args=['x', 'use_mkldnn'], varargs=None, keywords=None, defaults=(False,))
+paddle.fluid.layers.slice ArgSpec(args=['input', 'axes', 'starts', 'ends'], varargs=None, keywords=None, defaults=None)
+paddle.fluid.layers.shape ArgSpec(args=['input'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.data ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True))
 paddle.fluid.layers.open_files ArgSpec(args=['filenames', 'shapes', 'lod_levels', 'dtypes', 'thread_num', 'buffer_size', 'pass_num', 'is_test'], varargs=None, keywords=None, defaults=(None, None, 1, None))
 paddle.fluid.layers.read_file ArgSpec(args=['reader'], varargs=None, keywords=None, defaults=None)
@@ -241,13 +231,6 @@ paddle.fluid.layers.logical_and ArgSpec(args=[], varargs='args', keywords='kwarg
 paddle.fluid.layers.logical_or ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.logical_xor ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.logical_not ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.uniform_random_batch_size_like ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.gaussian_random ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.sampling_id ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.gaussian_random_batch_size_like ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.sum ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.slice ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
-paddle.fluid.layers.shape ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.maxout ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.sigmoid ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.logsigmoid ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,))
diff --git a/paddle/fluid/framework/ir/graph_traits.cc b/paddle/fluid/framework/ir/graph_traits.cc
index 8f548913e4e1d9d5bc5bdace8b92db9065cf3b5e..084a4ba2def87eaa8badb3ca2c39865c6e5cb981 100644
--- a/paddle/fluid/framework/ir/graph_traits.cc
+++ b/paddle/fluid/framework/ir/graph_traits.cc
@@ -14,6 +14,8 @@
 
 #include "paddle/fluid/framework/ir/graph_traits.h"
 
+#include <vector>
+
 namespace paddle {
 namespace framework {
 namespace ir {
diff --git a/paddle/fluid/operators/sampling_id_op.cc b/paddle/fluid/operators/sampling_id_op.cc
index 724463c95c4a29fb5c00fe791b389d3908771640..a4f41a170426a4650fd3bf8f7fec4758ff34e1b9 100644
--- a/paddle/fluid/operators/sampling_id_op.cc
+++ b/paddle/fluid/operators/sampling_id_op.cc
@@ -53,15 +53,16 @@ class SamplingIdOpMaker : public framework::OpProtoAndCheckerMaker {
 SamplingId Operator.
 A layer for sampling id from multinomial distribution from the
  input. Sampling one id for one sample.)DOC");
-    AddAttr<float>("min", "Minimum value of random. [default 0.0].")
+    AddAttr<float>("min", "Minimum value of random. (float, default 0.0).")
         .SetDefault(0.0f);
-    AddAttr<float>("max", "Maximun value of random. [default 1.0].")
+    AddAttr<float>("max", "Maximun value of random. (float, default 1.0).")
         .SetDefault(1.0f);
-    AddAttr<int>("seed",
-                 "Random seed used for the random number engine. "
-                 "0 means use a seed generated by the system."
-                 "Note that if seed is not 0, this operator will always "
-                 "generate the same random numbers every time. [default 0].")
+    AddAttr<int>(
+        "seed",
+        "Random seed used for the random number engine. "
+        "0 means use a seed generated by the system."
+        "Note that if seed is not 0, this operator will always "
+        "generate the same random numbers every time. (int, default 0).")
         .SetDefault(0);
   }
 };
diff --git a/paddle/fluid/operators/sequence_slice_op.h b/paddle/fluid/operators/sequence_slice_op.h
index b5ea6ff49bbb29571f9a6ef6358ef881acd9be9e..03b59d71cc0ca2eddd1d9912e7ca25348507ba03 100644
--- a/paddle/fluid/operators/sequence_slice_op.h
+++ b/paddle/fluid/operators/sequence_slice_op.h
@@ -75,11 +75,11 @@ class SequenceSliceOpKernel : public framework::OpKernel<T> {
     }
 
     for (size_t i = 0; i < n; ++i) {
-      PADDLE_ENFORCE_LT(0, offset_data[i],
+      PADDLE_ENFORCE_LE(0, offset_data[i],
                         "The offset[%d] must greater than zero.", i);
       PADDLE_ENFORCE_LT(0, length_data[i],
                         "The length[%d] must greater than zero.", i);
-      PADDLE_ENFORCE_LT(lod[0][i] + offset_data[i] + length_data[i],
+      PADDLE_ENFORCE_LE(lod[0][i] + offset_data[i] + length_data[i],
                         lod[0][i + 1], "The target tensor's length overflow.");
     }
 
diff --git a/paddle/fluid/pybind/protobuf.cc b/paddle/fluid/pybind/protobuf.cc
index 67501186d150171728194f23bc02d2c014848dd7..a5bc44122028c1191f511157bdde2e7c2d30c6aa 100644
--- a/paddle/fluid/pybind/protobuf.cc
+++ b/paddle/fluid/pybind/protobuf.cc
@@ -285,12 +285,12 @@ void BindOpDesc(pybind11::module *m) {
       .def("set_output", &pd::OpDesc::SetOutput)
       .def("input_arg_names", &pd::OpDesc::InputArgumentNames)
       .def("output_arg_names", &pd::OpDesc::OutputArgumentNames)
-      .def("rename_input", &pd::OpDesc::RenameInput)
-      .def("rename_output", &pd::OpDesc::RenameOutput)
+      .def("_rename_input", &pd::OpDesc::RenameInput)
+      .def("_rename_output", &pd::OpDesc::RenameOutput)
       .def("has_attr", &pd::OpDesc::HasAttr)
       .def("attr_type", &pd::OpDesc::GetAttrType)
       .def("attr_names", &pd::OpDesc::AttrNames)
-      .def("set_attr", &pd::OpDesc::SetAttr)
+      .def("_set_attr", &pd::OpDesc::SetAttr)
       .def("attr", &pd::OpDesc::GetAttr)
       .def("set_block_attr", &pd::OpDesc::SetBlockAttr)
       .def("set_blocks_attr", &pd::OpDesc::SetBlocksAttr)
@@ -300,8 +300,8 @@ void BindOpDesc(pybind11::module *m) {
              std::string ser(seriralized);
              self.SetAttr(name, ser);
            })
-      .def("block_attr_id", &pd::OpDesc::GetBlockAttrId)
-      .def("blocks_attr_ids", &pd::OpDesc::GetBlocksAttrIds)
+      .def("_block_attr_id", &pd::OpDesc::GetBlockAttrId)
+      .def("_blocks_attr_ids", &pd::OpDesc::GetBlocksAttrIds)
       .def("check_attrs", &pd::OpDesc::CheckAttrs)
       .def("infer_shape", &pd::OpDesc::InferShape)
       .def("infer_var_type", &pd::OpDesc::InferVarType)
diff --git a/paddle/scripts/paddle_build.sh b/paddle/scripts/paddle_build.sh
index e8d2e8e6861b95c94077a72f8e4f7d11b0ab11be..c397f070e947ba787c13397dfc07e4b1e4e37e73 100755
--- a/paddle/scripts/paddle_build.sh
+++ b/paddle/scripts/paddle_build.sh
@@ -629,10 +629,10 @@ EOF
 
 function gen_capi_package() {
     if [[ ${WITH_C_API} == "ON" ]]; then
-        install_prefix="${PADDLE_ROOT}/build/capi_output"
-        rm -rf $install_prefix
-        make DESTDIR="$install_prefix" install
-        cd $install_prefix/usr/local
+        capi_install_prefix=${INSTALL_PREFIX:-/paddle/build}/capi_output
+        rm -rf $capi_install_prefix
+        make DESTDIR="$capi_install_prefix" install
+        cd $capi_install_prefix/
         ls | egrep -v "^Found.*item$" | xargs tar -czf ${PADDLE_ROOT}/build/paddle.tgz
     fi
 }
diff --git a/python/paddle/fluid/backward.py b/python/paddle/fluid/backward.py
index 88eaae10dd55edcc7e811163acf17579eb32cbf1..17fe8dc3c8a28ad129e2d377820da95e8e7a02d9 100644
--- a/python/paddle/fluid/backward.py
+++ b/python/paddle/fluid/backward.py
@@ -38,8 +38,8 @@ def _rename_arg_(op_descs, old_name, new_name, begin_idx=None, end_idx=None):
         op_desc = op_descs[i]
         if isinstance(op_desc, tuple):
             op_desc = op_desc[0]
-        op_desc.rename_input(old_name, new_name)
-        op_desc.rename_output(old_name, new_name)
+        op_desc._rename_input(old_name, new_name)
+        op_desc._rename_output(old_name, new_name)
 
 
 def _create_op_desc_(op_type, inputs, outputs, attrs):
@@ -70,7 +70,7 @@ def _create_op_desc_(op_type, inputs, outputs, attrs):
         if isinstance(val, framework.Block):
             op_desc.set_block_attr(name, val.desc)
         else:
-            op_desc.set_attr(name, val)
+            op_desc._set_attr(name, val)
     return op_desc
 
 
@@ -346,7 +346,7 @@ def _append_backward_ops_(block,
         grad_sub_block_list = []
         # If the op has its own sub-block, deal with the sub-block first
         if op.has_attr("sub_block"):
-            sub_block = program.block(op.block_attr_id("sub_block"))
+            sub_block = program.block(op._block_attr_id("sub_block"))
             grad_sub_block = program._create_block()
             grad_sub_block._set_forward_block_idx(sub_block.idx)
             cb = _callback_lookup_(op)
@@ -382,7 +382,7 @@ def _append_backward_ops_(block,
     for op_desc in grad_op_descs:
         new_op_desc = target_block.desc.append_op()
         new_op_desc.copy_from(op_desc)
-        new_op_desc.set_attr(op_role_attr_name, backward)
+        new_op_desc._set_attr(op_role_attr_name, backward)
         grad_to_var["__current_op_desc__"] = new_op_desc
         if callbacks is not None:
             assert (isinstance(callbacks, list))
@@ -408,7 +408,7 @@ def _append_backward_vars_(block, start_op_idx, grad_to_var, grad_info_map):
     for op_idx in range(start_op_idx, block.desc.op_size()):
         op_desc = block.desc.op(op_idx)
         if op_desc.has_attr("sub_block"):
-            sub_block = block.program.block(op_desc.block_attr_id("sub_block"))
+            sub_block = block.program.block(op_desc._block_attr_id("sub_block"))
             _append_backward_vars_(sub_block, 0, grad_to_var, grad_info_map)
         new_vars = set()
         # create new gradient variables
@@ -438,12 +438,12 @@ def _rename_grad_(block, start_op_idx, grad_to_var, target_grad_map):
         op_desc = block.desc.op(op_idx)
         for name in op_desc.input_arg_names():
             if name in var_map:
-                op_desc.rename_input(name, var_map[name])
+                op_desc._rename_input(name, var_map[name])
 
         for name in op_desc.output_arg_names():
             if block.desc.find_var(name.encode("ascii")):
                 new_name = unique_name.generate(name)
-                op_desc.rename_output(name, new_name)
+                op_desc._rename_output(name, new_name)
                 var_map[name] = new_name
 
     for g, ng in six.iteritems(var_map):
@@ -542,9 +542,9 @@ def append_backward(loss, parameter_list=None, no_grad_set=None,
         if loss.op is None:
             raise ValueError("loss.op is None. Should not happend")
 
-    loss.op.set_attr(core.op_proto_and_checker_maker.kOpRoleAttrName(),
-                     int(core.op_proto_and_checker_maker.OpRole.Forward) |
-                     int(core.op_proto_and_checker_maker.OpRole.Loss))
+    loss.op._set_attr(core.op_proto_and_checker_maker.kOpRoleAttrName(),
+                      int(core.op_proto_and_checker_maker.OpRole.Forward) |
+                      int(core.op_proto_and_checker_maker.OpRole.Loss))
 
     if callbacks is not None:
         isinstance(callbacks, list)
@@ -631,7 +631,7 @@ def append_backward(loss, parameter_list=None, no_grad_set=None,
         attr_val = [p.name, g.name]
         if g.op.has_attr(op_role_var_attr_name):
             attr_val.extend(g.op.attr(op_role_var_attr_name))
-        g.op.set_attr(op_role_var_attr_name, attr_val)
+        g.op._set_attr(op_role_var_attr_name, attr_val)
 
     return params_and_grads
 
diff --git a/python/paddle/fluid/clip.py b/python/paddle/fluid/clip.py
index 32b8f1189fd65ba1e8da5aeaf316fc0ae05af552..e884185528282021fd16289ccc6a3533e22b9967 100644
--- a/python/paddle/fluid/clip.py
+++ b/python/paddle/fluid/clip.py
@@ -75,8 +75,8 @@ class ErrorClipByValue(BaseErrorClipAttr):
         clip_op_desc.set_type("clip")
         clip_op_desc.set_input("X", [grad_name])
         clip_op_desc.set_output("Out", [grad_name])
-        clip_op_desc.set_attr("min", self.min)
-        clip_op_desc.set_attr("max", self.max)
+        clip_op_desc._set_attr("min", self.min)
+        clip_op_desc._set_attr("max", self.max)
 
 
 def error_clip_callback(block, context):
diff --git a/python/paddle/fluid/framework.py b/python/paddle/fluid/framework.py
index fc61bcbea66de07350ee778abb16e81f8f8bc8db..7ece6a7fafa91b24b03feeabb2bbefa0d3a1b24b 100644
--- a/python/paddle/fluid/framework.py
+++ b/python/paddle/fluid/framework.py
@@ -37,11 +37,9 @@ from . import unique_name
 
 __all__ = [
     'Program',
-    'Operator',
     'default_startup_program',
     'default_main_program',
     'program_guard',
-    'get_var',
     'name_scope',
 ]
 
@@ -654,11 +652,11 @@ class Operator(object):
                 self._update_desc_attr(attr_name, attr_val)
 
         self.desc.check_attrs()
-        if self.has_kernel(type):
+        if self._has_kernel(type):
             self.desc.infer_var_type(self.block.desc)
             self.desc.infer_shape(self.block.desc)
 
-    def has_kernel(self, op_type):
+    def _has_kernel(self, op_type):
         return op_type not in self.OP_WITHOUT_KERNEL_SET
 
     def to_string(self, throw_on_error):
@@ -699,7 +697,7 @@ class Operator(object):
         """
         return self.desc.input(name)
 
-    def rename_input(self, old_name, new_name):
+    def _rename_input(self, old_name, new_name):
         """
         Rename the `old_name` to `new_name`.
 
@@ -710,9 +708,9 @@ class Operator(object):
         Returns:
             None
         """
-        self.desc.rename_input(old_name, new_name)
+        self.desc._rename_input(old_name, new_name)
 
-    def rename_output(self, old_name, new_name):
+    def _rename_output(self, old_name, new_name):
         """
         Rename the `old_name` to `new_name`.
 
@@ -723,7 +721,7 @@ class Operator(object):
         Returns:
             None
         """
-        self.desc.rename_output(old_name, new_name)
+        self.desc._rename_output(old_name, new_name)
 
     @property
     def input_names(self):
@@ -787,7 +785,7 @@ class Operator(object):
         """
         return self.desc.attr_type(name)
 
-    def set_attr(self, name, val):
+    def _set_attr(self, name, val):
         """
         Set the value of attribute by attribute's name.
 
@@ -820,7 +818,7 @@ class Operator(object):
                 isinstance(val, core.ProgramDesc):
             self.desc.set_serialized_attr(name, val.serialize_to_string())
         else:
-            self.desc.set_attr(name, val)
+            self.desc._set_attr(name, val)
 
     @property
     def attr_names(self):
@@ -839,7 +837,7 @@ class Operator(object):
         """
         return self.desc.attr(name)
 
-    def block_attr_id(self, name):
+    def _block_attr_id(self, name):
         """
         Get the block attribute's id by name.
 
@@ -849,9 +847,9 @@ class Operator(object):
         Returns:
             int: the block index.
         """
-        return self.desc.block_attr_id(name)
+        return self.desc._block_attr_id(name)
 
-    def block_attr(self, name):
+    def _block_attr(self, name):
         """
         Get the block attribute  by name.
 
@@ -862,11 +860,11 @@ class Operator(object):
             block: the block attribute.
         """
 
-        id = self.block_attr_id(name)
+        id = self._block_attr_id(name)
         assert (id >= 0 and id < len(self.block.program.blocks))
         return self.block.program.blocks[id]
 
-    def blocks_attr(self, name):
+    def _blocks_attr(self, name):
         """
         Get the blocks attribute  by name.
 
@@ -877,13 +875,13 @@ class Operator(object):
             list: list of the blocks attribute.
         """
         attrs = []
-        for i in self.blocks_attr_ids(name):
+        for i in self._blocks_attr_ids(name):
             assert (i >= 0 and i < len(self.block.program.blocks))
             attrs.append(self.block.program.blocks[i])
 
         return attrs
 
-    def blocks_attr_ids(self, name):
+    def _blocks_attr_ids(self, name):
         """
         Get the blocks attribute's ids by name.
 
@@ -894,7 +892,7 @@ class Operator(object):
             list: list of the blocks ids.
         """
 
-        return self.desc.blocks_attr_ids(name)
+        return self.desc._blocks_attr_ids(name)
 
     def all_attrs(self):
         """
@@ -908,11 +906,11 @@ class Operator(object):
         for n in attr_names:
             attr_type = self.desc.attr_type(n)
             if attr_type == core.AttrType.BLOCK:
-                attr_map[n] = self.block_attr(n)
+                attr_map[n] = self._block_attr(n)
                 continue
 
             if attr_type == core.AttrType.BLOCKS:
-                attr_map[n] = self.blocks_attr(n)
+                attr_map[n] = self._blocks_attr(n)
                 continue
 
             attr_map[n] = self.attr(n)
@@ -1786,7 +1784,7 @@ class Program(object):
             for j in six.moves.range(block.op_size()):
                 op = block.op(j)
                 if op.has_attr('is_test'):
-                    op.set_attr('is_test', True)
+                    op._set_attr('is_test', True)
         res.blocks = [
             Block(res, i) for i in six.moves.range(res.desc.num_blocks())
         ]
@@ -2160,7 +2158,7 @@ def program_guard(main_program, startup_program=None):
         switch_startup_program(startup_program)
 
 
-def get_var(name, program=None):
+def _get_var(name, program=None):
     """
     Get a variable by name from the global block of a program.
 
diff --git a/python/paddle/fluid/layers/detection.py b/python/paddle/fluid/layers/detection.py
index 574d0d727cba9fa9de0cffbe116f71b9e65a7092..9772c65738a2c5373f657164e3bc379404ba642e 100644
--- a/python/paddle/fluid/layers/detection.py
+++ b/python/paddle/fluid/layers/detection.py
@@ -284,7 +284,7 @@ def detection_output(loc,
         target_box=loc,
         code_type='decode_center_size')
     compile_shape = scores.shape
-    run_shape = ops.shape(scores)
+    run_shape = nn.shape(scores)
     scores = nn.flatten(x=scores, axis=2)
     scores = nn.softmax(input=scores)
     scores = nn.reshape(x=scores, shape=compile_shape, actual_shape=run_shape)
@@ -697,7 +697,7 @@ def ssd_loss(location,
         raise ValueError("Only support mining_type == max_negative now.")
 
     num, num_prior, num_class = confidence.shape
-    conf_shape = ops.shape(confidence)
+    conf_shape = nn.shape(confidence)
 
     def __reshape_to_2d(var):
         return nn.flatten(x=var, axis=2)
@@ -724,7 +724,7 @@ def ssd_loss(location,
     target_label.stop_gradient = True
     conf_loss = nn.softmax_with_cross_entropy(confidence, target_label)
     # 3. Mining hard examples
-    actual_shape = ops.slice(conf_shape, axes=[0], starts=[0], ends=[2])
+    actual_shape = nn.slice(conf_shape, axes=[0], starts=[0], ends=[2])
     actual_shape.stop_gradient = True
     conf_loss = nn.reshape(
         x=conf_loss, shape=(num, num_prior), actual_shape=actual_shape)
diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py
index 2cb61a9cd25c744710ab7ac9ea591902740f78da..a9696ac20060d1069a99a02a79a755a740e760f0 100644
--- a/python/paddle/fluid/layers/nn.py
+++ b/python/paddle/fluid/layers/nn.py
@@ -29,110 +29,29 @@ from .. import unique_name
 from functools import reduce
 
 __all__ = [
-    'fc',
-    'embedding',
-    'dynamic_lstm',
-    'dynamic_lstmp',
-    'dynamic_gru',
-    'gru_unit',
-    'linear_chain_crf',
-    'crf_decoding',
-    'cos_sim',
-    'cross_entropy',
-    'square_error_cost',
-    'chunk_eval',
-    'sequence_conv',
-    'conv2d',
-    'conv3d',
-    'sequence_pool',
-    'sequence_softmax',
-    'softmax',
-    'pool2d',
-    'pool3d',
-    'batch_norm',
-    'beam_search_decode',
-    'conv2d_transpose',
-    'conv3d_transpose',
-    'sequence_expand',
-    'sequence_expand_as',
-    'sequence_pad',
-    'lstm_unit',
-    'reduce_sum',
-    'reduce_mean',
-    'reduce_max',
-    'reduce_min',
-    'reduce_prod',
-    'sequence_first_step',
-    'sequence_last_step',
-    'dropout',
-    'split',
-    'ctc_greedy_decoder',
-    'edit_distance',
-    'l2_normalize',
-    'matmul',
-    'topk',
-    'warpctc',
-    'sequence_reshape',
-    'transpose',
-    'im2sequence',
-    'nce',
-    'hsigmoid',
-    'beam_search',
-    'row_conv',
-    'multiplex',
-    'layer_norm',
-    'softmax_with_cross_entropy',
-    'smooth_l1',
-    'one_hot',
-    'autoincreased_step_counter',
-    'reshape',
-    'squeeze',
-    'unsqueeze',
-    'lod_reset',
-    'lrn',
-    'pad',
-    'pad_constant_like',
-    'label_smooth',
-    'roi_pool',
-    'dice_loss',
-    'image_resize',
-    'image_resize_short',
-    'resize_bilinear',
-    'gather',
-    'scatter',
-    'sequence_scatter',
-    'random_crop',
-    'mean_iou',
-    'relu',
-    'log',
-    'crop',
-    'rank_loss',
-    'elu',
-    'relu6',
-    'pow',
-    'stanh',
-    'hard_sigmoid',
-    'swish',
-    'prelu',
-    'brelu',
-    'leaky_relu',
-    'soft_relu',
-    'flatten',
-    'sequence_mask',
-    'stack',
-    'pad2d',
-    'unstack',
-    'sequence_enumerate',
-    'expand',
-    'sequence_concat',
-    'scale',
-    'elementwise_add',
-    'elementwise_div',
-    'elementwise_sub',
-    'elementwise_mul',
-    'elementwise_max',
-    'elementwise_min',
-    'elementwise_pow',
+    'fc', 'embedding', 'dynamic_lstm', 'dynamic_lstmp', 'dynamic_gru',
+    'gru_unit', 'linear_chain_crf', 'crf_decoding', 'cos_sim', 'cross_entropy',
+    'square_error_cost', 'chunk_eval', 'sequence_conv', 'conv2d', 'conv3d',
+    'sequence_pool', 'sequence_softmax', 'softmax', 'pool2d', 'pool3d',
+    'batch_norm', 'beam_search_decode', 'conv2d_transpose', 'conv3d_transpose',
+    'sequence_expand', 'sequence_expand_as', 'sequence_pad', 'lstm_unit',
+    'reduce_sum', 'reduce_mean', 'reduce_max', 'reduce_min', 'reduce_prod',
+    'sequence_first_step', 'sequence_last_step', 'dropout', 'split',
+    'ctc_greedy_decoder', 'edit_distance', 'l2_normalize', 'matmul', 'topk',
+    'warpctc', 'sequence_reshape', 'transpose', 'im2sequence', 'nce',
+    'hsigmoid', 'beam_search', 'row_conv', 'multiplex', 'layer_norm',
+    'softmax_with_cross_entropy', 'smooth_l1', 'one_hot',
+    'autoincreased_step_counter', 'reshape', 'squeeze', 'unsqueeze',
+    'lod_reset', 'lrn', 'pad', 'pad_constant_like', 'label_smooth', 'roi_pool',
+    'dice_loss', 'image_resize', 'image_resize_short', 'resize_bilinear',
+    'gather', 'scatter', 'sequence_scatter', 'random_crop', 'mean_iou', 'relu',
+    'log', 'crop', 'rank_loss', 'elu', 'relu6', 'pow', 'stanh', 'hard_sigmoid',
+    'swish', 'prelu', 'brelu', 'leaky_relu', 'soft_relu', 'flatten',
+    'sequence_mask', 'stack', 'pad2d', 'unstack', 'sequence_enumerate',
+    'expand', 'sequence_concat', 'scale', 'elementwise_add', 'elementwise_div',
+    'elementwise_sub', 'elementwise_mul', 'elementwise_max', 'elementwise_min',
+    'elementwise_pow', 'uniform_random_batch_size_like', 'gaussian_random',
+    'sampling_id', 'gaussian_random_batch_size_like', 'sum', 'slice', 'shape'
 ]
 
 
@@ -6463,6 +6382,246 @@ def expand(x, expand_times, name=None):
     return out
 
 
+from paddle.fluid.framework import convert_np_dtype_to_dtype_
+
+
+@templatedoc()
+def uniform_random_batch_size_like(input,
+                                   shape,
+                                   dtype='float32',
+                                   input_dim_idx=0,
+                                   output_dim_idx=0,
+                                   min=-1.0,
+                                   max=1.0,
+                                   seed=0):
+    """
+    ${comment}
+
+    Args:
+        input (Variable): ${input_comment}
+        shape (tuple|list): ${shape_comment}
+        input_dim_idx (Int): ${input_dim_idx_comment}
+        output_dim_idx (Int): ${output_dim_idx_comment}
+        min (Float): ${min_comment}
+        max (Float): ${max_comment}
+        seed (Int): ${seed_comment}
+        dtype(np.dtype|core.VarDesc.VarType|str): The type of data : float32, float_16, int etc
+    Returns:
+        out (Variable): ${out_comment}
+
+    """
+
+    helper = LayerHelper('uniform_random_batch_size_like', **locals())
+    out = helper.create_tmp_variable(dtype)
+    c_dtype = convert_np_dtype_to_dtype_(dtype)
+    helper.append_op(
+        type='uniform_random_batch_size_like',
+        inputs={'Input': input},
+        outputs={'Out': out},
+        attrs={
+            'shape': shape,
+            'input_dim_idx': input_dim_idx,
+            'output_dim_idx': output_dim_idx,
+            'min': min,
+            'max': max,
+            'seed': seed,
+            'dtype': c_dtype
+        })
+
+    return out
+
+
+@templatedoc()
+def gaussian_random(shape,
+                    mean=0.0,
+                    std=1.0,
+                    seed=0,
+                    dtype='float32',
+                    use_mkldnn=False):
+    """
+    ${comment}
+
+    Args:
+        shape (tuple|list): ${shape_comment}
+        mean (Float): ${mean_comment}
+        std (Float): ${std_comment}
+        seed (Int): ${seed_comment}
+        dtype(np.dtype|core.VarDesc.VarType|str): Output data type.
+        use_mkldnn (Bool): Only used in mkldnn kernel.
+
+    Returns:
+        out (Variable): ${out_comment}
+
+    """
+
+    helper = LayerHelper('gaussian_random', **locals())
+    out = helper.create_tmp_variable(dtype)
+    c_dtype = convert_np_dtype_to_dtype_(dtype)
+    helper.append_op(
+        type='gaussian_random',
+        outputs={'Out': out},
+        attrs={
+            'shape': shape,
+            'mean': mean,
+            'std': std,
+            'seed': seed,
+            'dtype': c_dtype,
+            'use_mkldnn': use_mkldnn
+        })
+
+    return out
+
+
+@templatedoc()
+def sampling_id(x, min=0.0, max=1.0, seed=0, dtype='float32'):
+    """
+    ${comment}
+
+    Args:
+        x (Variable): ${x_comment}
+        min (Float): ${min_comment}
+        max (Float): ${max_comment}
+        seed (Float): ${seed_comment}
+        dtype(np.dtype|core.VarDesc.VarType|str): The type of output data : float32, float_16, int etc
+
+    Returns:
+        out (Variable): ${out_comment}
+
+    """
+
+    helper = LayerHelper('sampling_id', **locals())
+    out = helper.create_tmp_variable(dtype)
+    helper.append_op(
+        type='sampling_id',
+        inputs={'X': x},
+        outputs={'Out': out},
+        attrs={'min': min,
+               'max': max,
+               'seed': seed})
+
+    return out
+
+
+@templatedoc()
+def gaussian_random_batch_size_like(input,
+                                    shape,
+                                    input_dim_idx=0,
+                                    output_dim_idx=0,
+                                    mean=0.0,
+                                    std=1.0,
+                                    seed=0,
+                                    dtype='float32'):
+    """
+    ${comment}
+
+    Args:
+        input (Variable): ${input_comment}
+        shape (tuple|list): ${shape_comment}
+        input_dim_idx (Int): ${input_dim_idx_comment}
+        output_dim_idx (Int): ${output_dim_idx_comment}
+        mean (Float): ${mean_comment}
+        std (Float): ${std_comment}
+        seed (Int): ${seed_comment}
+        dtype(np.dtype|core.VarDesc.VarType|str): The type of output data : float32, float_16, int etc
+
+    Returns:
+        out (Variable): ${out_comment}
+    """
+
+    helper = LayerHelper('gaussian_random_batch_size_like', **locals())
+    out = helper.create_tmp_variable(dtype)
+    c_dtype = convert_np_dtype_to_dtype_(dtype)
+    helper.append_op(
+        type='gaussian_random_batch_size_like',
+        inputs={'Input': input},
+        outputs={'Out': out},
+        attrs={
+            'shape': shape,
+            'input_dim_idx': input_dim_idx,
+            'output_dim_idx': output_dim_idx,
+            'mean': mean,
+            'std': std,
+            'seed': seed,
+            'dtype': c_dtype
+        })
+
+    return out
+
+
+@templatedoc()
+def sum(x, use_mkldnn=False):
+    """
+    ${comment}
+
+    Args:
+        x (Variable): ${x_comment}
+        use_mkldnn (Bool): ${use_mkldnn_comment}
+
+    Returns:
+        out (Variable): ${out_comment}
+    """
+
+    helper = LayerHelper('sum', **locals())
+    out = helper.create_tmp_variable(dtype=helper.input_dtype('x'))
+    helper.append_op(
+        type='sum',
+        inputs={'X': x},
+        outputs={'Out': out},
+        attrs={'use_mkldnn': use_mkldnn})
+
+    return out
+
+
+@templatedoc()
+def slice(input, axes, starts, ends):
+    """
+    ${comment}
+
+    Args:
+        input (Variable): ${input_comment}.
+        axes (List): ${axes_comment}
+        starts (List): ${starts_comment}
+        ends (List): ${ends_comment}
+
+    Returns:
+        out (Variable): ${out_comment}
+
+    """
+
+    helper = LayerHelper('slice', **locals())
+    out = helper.create_tmp_variable(dtype=helper.input_dtype('input'))
+    helper.append_op(
+        type='slice',
+        inputs={'Input': input},
+        outputs={'Out': out},
+        attrs={'axes': axes,
+               'starts': starts,
+               'ends': ends})
+
+    return out
+
+
+@templatedoc()
+def shape(input):
+    """
+    ${comment}
+
+    Args:
+        input (Variable): ${input_comment}
+
+    Returns:
+        out (Variable): ${out_comment}
+
+    """
+
+    helper = LayerHelper('shape', **locals())
+    out = helper.create_tmp_variable(dtype=helper.input_dtype('input'))
+    helper.append_op(
+        type='shape', inputs={'Input': input}, outputs={'Out': out})
+
+    return out
+
+
 def _elementwise_op(helper):
     op_type = helper.layer_type
     x = helper.kwargs.get('x', None)
diff --git a/python/paddle/fluid/layers/ops.py b/python/paddle/fluid/layers/ops.py
index 7867bfe00e25711643eab1ab8d0141dbbad3da52..8a533035b0a3d175073cb0b0884aa507bcff782c 100644
--- a/python/paddle/fluid/layers/ops.py
+++ b/python/paddle/fluid/layers/ops.py
@@ -45,13 +45,6 @@ __all__ = [
     'logical_or',
     'logical_xor',
     'logical_not',
-    'uniform_random_batch_size_like',
-    'gaussian_random',
-    'sampling_id',
-    'gaussian_random_batch_size_like',
-    'sum',
-    'slice',
-    'shape',
     'maxout',
 ]
 
diff --git a/python/paddle/fluid/tests/unittests/dist_transformer.py b/python/paddle/fluid/tests/unittests/dist_transformer.py
index 3e536b7da1bdac73f13e6c633b508dab26f1580a..a2cc57425841100a2b61279d1b447b88ed4b9a54 100644
--- a/python/paddle/fluid/tests/unittests/dist_transformer.py
+++ b/python/paddle/fluid/tests/unittests/dist_transformer.py
@@ -1488,7 +1488,7 @@ def wrap_decoder(trg_vocab_size,
     if weight_sharing:
         predict = layers.matmul(
             x=dec_output,
-            y=fluid.get_var(word_emb_param_names[0]),
+            y=fluid.framework._get_var(word_emb_param_names[0]),
             transpose_y=True)
     else:
         predict = layers.fc(input=dec_output,
diff --git a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
index ecde407e6d85ea1bfc0181b4b60e095ea496fb1a..54a1c68a37f6929890aab697b48d621e6effb7d8 100644
--- a/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
+++ b/python/paddle/fluid/tests/unittests/test_dist_transpiler.py
@@ -264,6 +264,25 @@ class TestLRDecay(TranspilerTest):
         ])
 
 
+class TestDecayedAdagrad(TranspilerTest):
+    def net_conf(self):
+        x = fluid.layers.data(name='x', shape=[1000], dtype='float32')
+        y_predict = fluid.layers.fc(input=x,
+                                    size=1000,
+                                    act=None,
+                                    param_attr=fluid.ParamAttr(name='fc_w'),
+                                    bias_attr=fluid.ParamAttr(name='fc_b'))
+        y = fluid.layers.data(name='y', shape=[1], dtype='float32')
+        cost = fluid.layers.square_error_cost(input=y_predict, label=y)
+        avg_cost = fluid.layers.mean(cost)
+        opt = fluid.optimizer.DecayedAdagrad(learning_rate=0.1)
+        opt.minimize(avg_cost)
+
+    def transpiler_test_impl(self):
+        pserver, startup = self.get_pserver(self.pserver1_ep)
+        trainer, _ = self.get_trainer()
+
+
 class TestLRDecayConditional(TranspilerTest):
     def net_conf(self):
         x = fluid.layers.data(name='x', shape=[1000], dtype='float32')
diff --git a/python/paddle/fluid/tests/unittests/test_infer_shape.py b/python/paddle/fluid/tests/unittests/test_infer_shape.py
index a3d700aad8236fea7bb0e6d043323ad3bd0851f2..fdff22cacc28731a91ff4fd17407bd9edbdd9d8b 100644
--- a/python/paddle/fluid/tests/unittests/test_infer_shape.py
+++ b/python/paddle/fluid/tests/unittests/test_infer_shape.py
@@ -76,8 +76,8 @@ class TestInferShape(unittest.TestCase):
         mul_op_desc.set_input("X", ["x"])
         mul_op_desc.set_input("Y", ["y"])
         mul_op_desc.set_output("Out", ["out"])
-        mul_op_desc.set_attr("x_num_col_dims", 1)
-        mul_op_desc.set_attr("y_num_col_dims", 1)
+        mul_op_desc._set_attr("x_num_col_dims", 1)
+        mul_op_desc._set_attr("y_num_col_dims", 1)
 
         mul_op_desc.check_attrs()
         mul_op_desc.infer_shape(block)
diff --git a/python/paddle/fluid/tests/unittests/test_layers.py b/python/paddle/fluid/tests/unittests/test_layers.py
index f474cdae2054531d44724e0e3e0e58a35fb8ddcd..b8dc9e8ad7cd7cd100d5c3cb99319e6f5a37da91 100644
--- a/python/paddle/fluid/tests/unittests/test_layers.py
+++ b/python/paddle/fluid/tests/unittests/test_layers.py
@@ -541,7 +541,7 @@ class TestBook(unittest.TestCase):
         with program_guard(program):
             input = layers.data(
                 name="input", shape=[3, 100, 100], dtype="float32")
-            out = layers.shape(input, name="shape")
+            out = layers.shape(input)
             self.assertIsNotNone(out)
         print(str(program))
 
@@ -758,6 +758,65 @@ class TestBook(unittest.TestCase):
             out = layers.expand(x, [1, 2])
         print(str(program))
 
+    def test_uniform_random_batch_size_like(self):
+        program = Program()
+        with program_guard(program):
+            input = layers.data(name="input", shape=[13, 11], dtype='float32')
+            out = layers.uniform_random_batch_size_like(input, [-1, 11])
+            self.assertIsNotNone(out)
+        print(str(program))
+
+    def test_gaussian_random(self):
+        program = Program()
+        with program_guard(program):
+            out = layers.gaussian_random(shape=[20, 30])
+            self.assertIsNotNone(out)
+        print(str(program))
+
+    def test_sampling_id(self):
+        program = Program()
+        with program_guard(program):
+            x = layers.data(
+                name="X",
+                shape=[13, 11],
+                dtype='float32',
+                append_batch_size=False)
+
+            out = layers.sampling_id(x)
+            self.assertIsNotNone(out)
+        print(str(program))
+
+    def test_gaussian_random_batch_size_like(self):
+        program = Program()
+        with program_guard(program):
+            input = layers.data(name="input", shape=[13, 11], dtype='float32')
+
+            out = layers.gaussian_random_batch_size_like(
+                input, shape=[-1, 11], mean=1.0, std=2.0)
+            self.assertIsNotNone(out)
+        print(str(program))
+
+    def test_sum(self):
+        program = Program()
+        with program_guard(program):
+            input = layers.data(name="input", shape=[13, 11], dtype='float32')
+
+            out = layers.sum(input)
+            self.assertIsNotNone(out)
+        print(str(program))
+
+    def test_slice(self):
+        starts = [1, 0, 2]
+        ends = [3, 3, 4]
+        axes = [0, 1, 2]
+
+        program = Program()
+        with program_guard(program):
+            input = layers.data(
+                name="input", shape=[3, 4, 5, 6], dtype='float32')
+
+            out = layers.slice(input, axes=axes, starts=starts, ends=ends)
+
     def test_softshrink(self):
         program = Program()
         with program_guard(program):
diff --git a/python/paddle/fluid/tests/unittests/test_protobuf_descs.py b/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
index d24b5cbd06ddf9f332c1369ebd513bef27b77e14..7fb2171f611adea434d6f2710465810fb69d6979 100644
--- a/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
+++ b/python/paddle/fluid/tests/unittests/test_protobuf_descs.py
@@ -38,40 +38,40 @@ class TestOpDesc(unittest.TestCase):
         self.assertEqual(['z'], op.output("Out"))
         self.assertEqual(["Out"], op.output_names())
 
-        op.set_attr("int_attr", 1)
+        op._set_attr("int_attr", 1)
         self.assertEqual(1, op.attr("int_attr"))
         self.assertTrue(op.has_attr("int_attr"))
         self.assertEqual(core.AttrType.INT, op.attr_type("int_attr"))
 
-        op.set_attr("float_attr", -1.32)
+        op._set_attr("float_attr", -1.32)
         self.assertAlmostEqual(-1.32, op.attr("float_attr"), delta=1e-4)
         self.assertTrue(op.has_attr("float_attr"))
 
-        op.set_attr("bool_attr", False)
+        op._set_attr("bool_attr", False)
         self.assertFalse(op.attr("bool_attr"))
 
-        op.set_attr("string_attr", "abc")
+        op._set_attr("string_attr", "abc")
         self.assertEqual("abc", op.attr("string_attr"))
         self.assertTrue(op.has_attr("string_attr"))
 
-        op.set_attr("ints_attr", [1, 2, 3])
+        op._set_attr("ints_attr", [1, 2, 3])
         self.assertEqual([1, 2, 3], op.attr("ints_attr"))
 
         expected = [1.2, 2.3, 3.4]
-        op.set_attr("floats_attr", expected)
+        op._set_attr("floats_attr", expected)
         for e, a in zip(expected, op.attr("floats_attr")):
             self.assertAlmostEqual(e, a, delta=1e-4)
 
-        op.set_attr("strings_attr", ["a", "b", "c"])
+        op._set_attr("strings_attr", ["a", "b", "c"])
         self.assertEqual(["a", "b", "c"], op.attr("strings_attr"))
 
-        op.set_attr("bools_attr", [True, False, True])
+        op._set_attr("bools_attr", [True, False, True])
         self.assertEqual([True, False, True], op.attr("bools_attr"))
 
         self.assertEqual(8, len(op.attr_names()))
 
-        op.set_block_attr("block_attr", program_desc.block(0))
-        self.assertEqual(0, op.block_attr_id("block_attr"))
+        op.set_block_attr("_block_attr", program_desc.block(0))
+        self.assertEqual(0, op._block_attr_id("_block_attr"))
 
         mul_op = block.append_op()
         mul_op.set_type("mul")
diff --git a/python/paddle/fluid/transpiler/details/program_utils.py b/python/paddle/fluid/transpiler/details/program_utils.py
index 59899e7e9ab98f661699d5ac0645c92bd23a1512..391d6aa12bdd70b9ef988898bee8e86cd0a0d765 100644
--- a/python/paddle/fluid/transpiler/details/program_utils.py
+++ b/python/paddle/fluid/transpiler/details/program_utils.py
@@ -128,7 +128,7 @@ def op_to_code(op):
         attr_type = op.desc.attr_type(name)
         if attr_type == core.AttrType.BLOCK:
             a = "{name} = block[{value}]".format(
-                name=name, type=attr_type, value=op.block_attr_id(name))
+                name=name, type=attr_type, value=op._block_attr_id(name))
             attrs_str += a
             if i != len(attr_names) - 1:
                 attrs_str += ", "
@@ -136,7 +136,7 @@ def op_to_code(op):
 
         if attr_type == core.AttrType.BLOCKS:
             a = "{name} = blocks{value}".format(
-                name=name, type=attr_type, value=op.blocks_attr_ids(name))
+                name=name, type=attr_type, value=op._blocks_attr_ids(name))
             attrs_str += a
             if i != len(attr_names) - 1:
                 attrs_str += ", "
diff --git a/python/paddle/fluid/transpiler/distribute_transpiler.py b/python/paddle/fluid/transpiler/distribute_transpiler.py
index 57dceffc634b1a191a066d6f4a56dfcdd927b6ab..ecdbe27f4d90268d755a712e25289cfaf4715f29 100644
--- a/python/paddle/fluid/transpiler/distribute_transpiler.py
+++ b/python/paddle/fluid/transpiler/distribute_transpiler.py
@@ -470,7 +470,10 @@ class DistributeTranspiler(object):
         """
         # remove optimize ops and add a send op to main_program
         # FIXME(typhoonzero): Also ops like clip_gradient, lrn_decay?
+        lr_ops = self._get_lr_ops()
         delete_ops(self.origin_program.global_block(), self.optimize_ops)
+        delete_ops(self.origin_program.global_block(), lr_ops)
+
         self.origin_program.__str__()
 
         if wait_port:
@@ -668,7 +671,7 @@ in a single call.")
                 __clone_lr_op_sub_block__(cloned_op, program, new_sub_block)
 
             # reset the block of op
-            op.set_attr('sub_block', new_sub_block)
+            op._set_attr('sub_block', new_sub_block)
 
         # append lr decay ops to the child block if exists
         lr_ops = self._get_lr_ops()
@@ -862,7 +865,7 @@ to transpile() call.")
                 if op.type in [
                         "gaussian_random", "fill_constant", "uniform_random"
                 ]:
-                    op.set_attr("shape", list(new_outputs["Out"].shape))
+                    op._set_attr("shape", list(new_outputs["Out"].shape))
                 s_prog.global_block().append_op(
                     type=op.type,
                     inputs=new_inputs,
@@ -1428,6 +1431,9 @@ to transpile() call.")
         elif op_type == "rmsprop":
             if varkey in ["Moment", "MeanSquare"]:
                 return param_shape
+        elif op_type == "decayed_adagrad":
+            if varkey == "Moment":
+                return param_shape
         elif op_type == "sgd":
             pass
         return orig_shape
diff --git a/python/paddle/fluid/transpiler/inference_transpiler.py b/python/paddle/fluid/transpiler/inference_transpiler.py
index 49ba2cfd55bc881ed753fcefbd41f5b8fd4ebaf7..43d51b03e81895d7322d9e28a9c40b6d7cc69206 100644
--- a/python/paddle/fluid/transpiler/inference_transpiler.py
+++ b/python/paddle/fluid/transpiler/inference_transpiler.py
@@ -163,7 +163,7 @@ class InferenceTranspiler(object):
                 next_op = self.block.ops[i + 1]
                 if next_op.type == 'relu':
                     # modify bnorm OP to include relu
-                    current_op.set_attr("fuse_with_relu", True)
+                    current_op._set_attr("fuse_with_relu", True)
                     # remove relu OP
                     self.block._remove_op(i + 1)
             i = i + 1
@@ -377,7 +377,7 @@ class InferenceTranspiler(object):
                 type=old_var.type,
                 dtype=old_var.dtype,
                 shape=old_var.shape)
-            op.rename_input(old_param_name, new_param_name)
+            op._rename_input(old_param_name, new_param_name)
             self.scope.var(new_param_name)
 
             tensor = self.scope.find_var(new_param_name).get_tensor()
@@ -463,8 +463,8 @@ class InferenceTranspiler(object):
             current_op = self.block.ops[i]
             for input_arg in current_op.input_arg_names:
                 if input_arg in self.input_map:
-                    current_op.rename_input(input_arg,
-                                            self.input_map[input_arg])
+                    current_op._rename_input(input_arg,
+                                             self.input_map[input_arg])
 
     def _remove_unused_var(self):
         '''