diff --git a/mindspore/akg/__init__.py b/mindspore/_akg/__init__.py
similarity index 88%
rename from mindspore/akg/__init__.py
rename to mindspore/_akg/__init__.py
index a0c0364bd674b21e6a4ec4f5b2e39b1a4923342d..e3dceaf35e7f95e268a89ac8a853d0df927af31e 100644
--- a/mindspore/akg/__init__.py
+++ b/mindspore/_akg/__init__.py
@@ -18,7 +18,7 @@ import sys
 import os
 
 def AKGAddPath():
-    """akg add path."""
+    """_akg add path."""
     pwd = os.path.dirname(os.path.realpath(__file__))
     tvm_path = os.path.realpath(pwd)
     if tvm_path not in sys.path:
@@ -32,12 +32,12 @@ class AKGMetaPathFinder:
     """class AKGMetaPath finder."""
 
     def find_module(self, fullname, path=None):
-        """method akg find module."""
-        if fullname.startswith("akg.tvm"):
-            rname = fullname[4:]
+        """method _akg find module."""
+        if fullname.startswith("_akg.tvm"):
+            rname = fullname[5:]
             return AKGMetaPathLoader(rname)
-        if fullname.startswith("akg.topi"):
-            rname = fullname[4:]
+        if fullname.startswith("_akg.topi"):
+            rname = fullname[5:]
             return AKGMetaPathLoader(rname)
         return None
 
diff --git a/mindspore/akg/gpu/__init__.py b/mindspore/_akg/gpu/__init__.py
similarity index 100%
rename from mindspore/akg/gpu/__init__.py
rename to mindspore/_akg/gpu/__init__.py
diff --git a/mindspore/akg/gpu/cast.py b/mindspore/_akg/gpu/cast.py
similarity index 86%
rename from mindspore/akg/gpu/cast.py
rename to mindspore/_akg/gpu/cast.py
index 458772a8032cff186c2eb5fe2dfba3674dab48a4..d6b38b6e9ba1a7d1fa4c622fd94365854b7d39cd 100644
--- a/mindspore/akg/gpu/cast.py
+++ b/mindspore/_akg/gpu/cast.py
@@ -14,9 +14,9 @@
 
 """cast"""
 import logging
-import akg.tvm
-from akg.ops.math import cast
-from akg.topi.generic import schedule_elemwise
+import _akg.tvm
+from _akg.ops.math import cast
+from _akg.topi.generic import schedule_elemwise
 
 def Cast(x, dst_type):
     """cast."""
@@ -34,10 +34,10 @@ def gpu_schedule_Cast(outs):
         sch (schedule.Schedule): The created schedule.
     """
     device = 'cuda'
-    ctx = akg.tvm.context(device, 0)
+    ctx = _akg.tvm.context(device, 0)
     if not ctx.exist:
         logging.info("Skip because %s is not enabled", device)
         return None
-    with akg.tvm.target.create(device):
+    with _akg.tvm.target.create(device):
         sch = schedule_elemwise(outs)
     return sch
diff --git a/mindspore/akg/gpu/default_schedule.py b/mindspore/_akg/gpu/default_schedule.py
similarity index 94%
rename from mindspore/akg/gpu/default_schedule.py
rename to mindspore/_akg/gpu/default_schedule.py
index 2e2892c0556b30795e3be6c1f03e57d43dcc8c87..811cc2d7108347a332122a38a1cc970bac73a085 100644
--- a/mindspore/akg/gpu/default_schedule.py
+++ b/mindspore/_akg/gpu/default_schedule.py
@@ -15,7 +15,7 @@
 """default schedule function for GPU"""
 from queue import Queue
 
-import akg.tvm as tvm
+import _akg.tvm as tvm
 
 DEFAULT_GPU_THREAD = 1024
 
@@ -31,7 +31,7 @@ def default_schedule(outs):
         sch (schedule.Schedule): The created schedule.
     """
     if not isinstance(outs, tvm.tensor.Tensor) and not isinstance(outs, list):
-        raise ValueError("outs should be list of akg.tvm.tensor.Tensor or akg.tvm.tensor.Tensor")
+        raise ValueError("outs should be list of _akg.tvm.tensor.Tensor or _akg.tvm.tensor.Tensor")
     device = 'cuda'
     ctx = tvm.context(device, 0)
     if not ctx.exist:
diff --git a/mindspore/akg/gpu/equal.py b/mindspore/_akg/gpu/equal.py
similarity index 85%
rename from mindspore/akg/gpu/equal.py
rename to mindspore/_akg/gpu/equal.py
index 05dce89622d43ca86c9279649a4ee013faa67ea9..3321c10b2c7d06fd26d89d576587fe01f06fd30f 100644
--- a/mindspore/akg/gpu/equal.py
+++ b/mindspore/_akg/gpu/equal.py
@@ -12,9 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """equal"""
-import akg.tvm
-from akg.ops.math import equal
-from akg.topi.generic import schedule_elemwise
+import _akg.tvm
+from _akg.ops.math import equal
+from _akg.topi.generic import schedule_elemwise
 
 def Equal(x, y):
     """equal."""
@@ -32,9 +32,9 @@ def gpu_schedule_Equal(outs):
         sch (schedule.Schedule): The created schedule.
     """
     device = 'cuda'
-    ctx = akg.tvm.context(device, 0)
+    ctx = _akg.tvm.context(device, 0)
     if not ctx.exist:
         raise SystemError("Skip because %s is not enabled" % device)
-    with akg.tvm.target.create(device):
+    with _akg.tvm.target.create(device):
         sch = schedule_elemwise(outs)
     return sch
diff --git a/mindspore/akg/gpu/mean.py b/mindspore/_akg/gpu/mean.py
similarity index 97%
rename from mindspore/akg/gpu/mean.py
rename to mindspore/_akg/gpu/mean.py
index a68e929409171c3798404831d157a60455440a7e..e9cdb6d55177cb7678a2cb8d3992f8c7dae11980 100644
--- a/mindspore/akg/gpu/mean.py
+++ b/mindspore/_akg/gpu/mean.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 """mean op compute and schedule"""
-import akg.tvm as tvm
-from akg.ops.math.mean import mean
+import _akg.tvm as tvm
+from _akg.ops.math.mean import mean
 from .default_schedule import DEFAULT_GPU_THREAD
 
 def Mean(x, axis=None, keepdims=True):
diff --git a/mindspore/akg/gpu/mean_grad.py b/mindspore/_akg/gpu/mean_grad.py
similarity index 95%
rename from mindspore/akg/gpu/mean_grad.py
rename to mindspore/_akg/gpu/mean_grad.py
index ef77690a5dd81379317e5674542d9c4adbc7a35d..9d91ee3f403fef3a39e9821fb7a0658bc67a3acc 100644
--- a/mindspore/akg/gpu/mean_grad.py
+++ b/mindspore/_akg/gpu/mean_grad.py
@@ -13,9 +13,9 @@
 # limitations under the License.
 
 """mean_grad"""
-import akg.tvm as tvm
-import akg
-from akg.ops.math import mean
+import _akg.tvm as tvm
+import _akg
+from _akg.ops.math import mean
 from .default_schedule import DEFAULT_GPU_THREAD
 
 
@@ -30,7 +30,7 @@ def mean_ad(head, input_shape, axis, keepdims):
     if tensor_b.op.name == "mean_output":
         tensor_b = tensor_b.op.input_tensors[0]
 
-    jacs = list(akg.differentiate(tensor_b, [tensor_a], head))
+    jacs = list(_akg.differentiate(tensor_b, [tensor_a], head))
     return jacs[0]
 
 
diff --git a/mindspore/akg/gpu/mul.py b/mindspore/_akg/gpu/mul.py
similarity index 93%
rename from mindspore/akg/gpu/mul.py
rename to mindspore/_akg/gpu/mul.py
index 975a237837a81146f746b596d9f5c7f06ae8255f..5c289a62a6f85e02d02d8f60d95284fe6e849b90 100644
--- a/mindspore/akg/gpu/mul.py
+++ b/mindspore/_akg/gpu/mul.py
@@ -13,9 +13,9 @@
 # limitations under the License.
 
 """mul"""
-import akg.topi as topi
-import akg.tvm as tvm
-from akg.ops.math import mul
+import _akg.topi as topi
+import _akg.tvm as tvm
+from _akg.ops.math import mul
 
 def Mul(x, y):
     """mul."""
diff --git a/mindspore/akg/gpu/relu6.py b/mindspore/_akg/gpu/relu6.py
similarity index 95%
rename from mindspore/akg/gpu/relu6.py
rename to mindspore/_akg/gpu/relu6.py
index bdcf23f05a09b919e0685267127cdecc0c210b85..9a0a3d7a451a505abb6b3d6ee8724582c4ac017b 100644
--- a/mindspore/akg/gpu/relu6.py
+++ b/mindspore/_akg/gpu/relu6.py
@@ -13,9 +13,9 @@
 # limitations under the License.
 
 """relu6"""
-import akg.topi as topi
-import akg.tvm as tvm
-from akg.topi import tag
+import _akg.topi as topi
+import _akg.tvm as tvm
+from _akg.topi import tag
 
 @tvm.tag_scope(tag=tag.ELEMWISE)
 def topi_nn_relu6(x):
diff --git a/mindspore/akg/gpu/relu6_grad.py b/mindspore/_akg/gpu/relu6_grad.py
similarity index 97%
rename from mindspore/akg/gpu/relu6_grad.py
rename to mindspore/_akg/gpu/relu6_grad.py
index e0590cf6efd4724a03ddb145135b7ca250cfdd75..62aeabb4c0771cf6b958407b3e0ee82c4ba40f2f 100644
--- a/mindspore/akg/gpu/relu6_grad.py
+++ b/mindspore/_akg/gpu/relu6_grad.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 """relu6 grad"""
-import akg.topi as topi
-import akg.tvm as tvm
+import _akg.topi as topi
+import _akg.tvm as tvm
 
 def ReLU6Grad(y_grad, x):
     """
diff --git a/mindspore/akg/gpu/squeeze.py b/mindspore/_akg/gpu/squeeze.py
similarity index 96%
rename from mindspore/akg/gpu/squeeze.py
rename to mindspore/_akg/gpu/squeeze.py
index 34fa423b8c7065aafe461d46a94bc08b1cccc36e..b5f55facaac70343ec6a40d2a9b58f99092f9bb5 100644
--- a/mindspore/akg/gpu/squeeze.py
+++ b/mindspore/_akg/gpu/squeeze.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 """squeeze"""
-import akg.topi as topi
-import akg.tvm as tvm
+import _akg.topi as topi
+import _akg.tvm as tvm
 
 def Squeeze(x, axis=None):
     """
diff --git a/mindspore/akg/gpu/squeeze_grad.py b/mindspore/_akg/gpu/squeeze_grad.py
similarity index 98%
rename from mindspore/akg/gpu/squeeze_grad.py
rename to mindspore/_akg/gpu/squeeze_grad.py
index ef6a4242baba6e29488d0aa5432e2d8b6bc77b12..8180ff9638d1604e5757e1dbf45a756f6261f645 100644
--- a/mindspore/akg/gpu/squeeze_grad.py
+++ b/mindspore/_akg/gpu/squeeze_grad.py
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 """squeeze grad"""
-import akg.topi as topi
+import _akg.topi as topi
 
 def SqueezeGrad(y_grad, x_shape, axis=None):
     """
diff --git a/mindspore/akg/gpu/tile.py b/mindspore/_akg/gpu/tile.py
similarity index 85%
rename from mindspore/akg/gpu/tile.py
rename to mindspore/_akg/gpu/tile.py
index cd3c663f978dfbb0930199149e6ab26f2da70c17..1eb6979b0999b464f5420129993b33674e8f632c 100644
--- a/mindspore/akg/gpu/tile.py
+++ b/mindspore/_akg/gpu/tile.py
@@ -12,9 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """tile"""
-import akg.tvm
-from akg.ops.array import tile
-from akg.topi.generic import schedule_elemwise
+import _akg.tvm
+from _akg.ops.array import tile
+from _akg.topi.generic import schedule_elemwise
 
 def Tile(x, multiples):
     """tile."""
@@ -31,9 +31,9 @@ def gpu_schedule_Tile(outs):
         sch (schedule.Schedule): The created schedule.
     """
     device = 'cuda'
-    ctx = akg.tvm.context(device, 0)
+    ctx = _akg.tvm.context(device, 0)
     if not ctx.exist:
         raise SystemError("Skip because %s is not enabled" % device)
-    with akg.tvm.target.create(device):
+    with _akg.tvm.target.create(device):
         s = schedule_elemwise(outs)
     return s
diff --git a/mindspore/akg/message.py b/mindspore/_akg/message.py
similarity index 94%
rename from mindspore/akg/message.py
rename to mindspore/_akg/message.py
index 86bdf2899cf4ef764bc6e65ab4212a9d4ee0e5f8..4528771848df46dbbbb7ded87620c6ef4447ba71 100644
--- a/mindspore/akg/message.py
+++ b/mindspore/_akg/message.py
@@ -20,9 +20,9 @@ import logging
 import traceback
 import os.path
 from pathlib import Path
-import akg.tvm
-from akg.utils import validation_check as vc_util
-from akg.utils.dsl_create import TensorUtils
+import _akg.tvm
+from _akg.utils import validation_check as vc_util
+from _akg.utils.dsl_create import TensorUtils
 from . import gpu
 from . import op_build
 
@@ -67,7 +67,7 @@ def compilewithjson(json_str):
             tensor_shape = input_desc[0]['shape']
             tensor_shape = (1,) if not tensor_shape else tensor_shape
             vc_util.shape_dtype_max_size_check(tensor_shape)
-            args[input_desc[0]['name']] = akg.tvm.placeholder(
+            args[input_desc[0]['name']] = _akg.tvm.placeholder(
                 shape=tensor_shape, name=input_desc[0]['tensor_name'], dtype=input_desc[0]['data_type'])
             tsr.append(args[input_desc[0]['name']])
         else:
@@ -76,7 +76,7 @@ def compilewithjson(json_str):
                 tensor_shape = tmp_desc['shape']
                 tensor_shape = (1,) if not tensor_shape else tensor_shape
                 vc_util.shape_dtype_max_size_check(tensor_shape)
-                tmp_input.append(akg.tvm.placeholder(
+                tmp_input.append(_akg.tvm.placeholder(
                     shape=tensor_shape, name=tmp_desc['tensor_name'], dtype=tmp_desc['data_type']))
             args[input_desc[0]['name']] = tmp_input
             tsr = tsr + tmp_input
diff --git a/mindspore/akg/op_build.py b/mindspore/_akg/op_build.py
similarity index 88%
rename from mindspore/akg/op_build.py
rename to mindspore/_akg/op_build.py
index e3d3ec2b78b08b722e0d1ce5fdee17ffb14647fd..44a250bd9e3794083d8f89a16c78cd7184252c9c 100644
--- a/mindspore/akg/op_build.py
+++ b/mindspore/_akg/op_build.py
@@ -19,10 +19,10 @@ import types
 import typing
 import logging
 import traceback
-import akg.tvm
-import akg
-from akg import save_gpu_param as gpu_utils
-from akg.utils import validation_check as vc_util
+import _akg.tvm
+import _akg
+from _akg import save_gpu_param as gpu_utils
+from _akg.utils import validation_check as vc_util
 
 MS_CUDA_KERNEL_PATH = "/tmp/cuda_meta/"
 
@@ -38,21 +38,21 @@ def op_build(opnames, computes, args, custom_schedule, device, kernel_name, attr
             return None
 
         schedule_name = 'gpu_schedule_' + opnames[0]
-        schedule_func = getattr(akg.gpu, schedule_name)
+        schedule_func = getattr(_akg.gpu, schedule_name)
         if not isinstance(schedule_func, (types.FunctionType, typing.Callable)):
             logging.error("no schedule func found %s", str(schedule_name))
             return None
 
         ptx_file = os.path.realpath(MS_CUDA_KERNEL_PATH + kernel_name + ".ptx")
         if os.path.exists(ptx_file):
-            os.remove(ptx_file)
+            os.chmod(ptx_file, 0o600)
         try:
             with open(ptx_file, 'at') as file:
                 fcntl.flock(file.fileno(), fcntl.LOCK_EX)
                 file.seek(0, 2)
                 if file.tell() == 0:
                     s = schedule_func(computes)
-                    foo = akg.tvm.build(s, args, device, name=kernel_name)
+                    foo = _akg.tvm.build(s, args, device, name=kernel_name)
                     ptx_code = foo.imported_modules[0].get_source("ptx")
                     file.write(ptx_code)
                     json_file = os.path.realpath(MS_CUDA_KERNEL_PATH + kernel_name + ".json")
diff --git a/mindspore/akg/ops/__init__.py b/mindspore/_akg/ops/__init__.py
similarity index 100%
rename from mindspore/akg/ops/__init__.py
rename to mindspore/_akg/ops/__init__.py
diff --git a/mindspore/akg/ops/array/__init__.py b/mindspore/_akg/ops/array/__init__.py
similarity index 100%
rename from mindspore/akg/ops/array/__init__.py
rename to mindspore/_akg/ops/array/__init__.py
diff --git a/mindspore/akg/ops/array/tile.py b/mindspore/_akg/ops/array/tile.py
similarity index 84%
rename from mindspore/akg/ops/array/tile.py
rename to mindspore/_akg/ops/array/tile.py
index e60fcc4ffb094595584b42253a3ca32f699fb0a5..2fa485ea36c67fa153350edbe2a42f89bf053076 100644
--- a/mindspore/akg/ops/array/tile.py
+++ b/mindspore/_akg/ops/array/tile.py
@@ -13,12 +13,12 @@
 # limitations under the License.
 
 """operator dsl function: tile"""
-import akg.tvm
-import akg.topi
-from akg.utils import validation_check as vc_util
+import _akg.tvm
+import _akg.topi
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, (list, tuple))
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, (list, tuple))
 def tile(data, multiples):
     """
     Repeats the data in the specified dimensions according to the multiples.
@@ -32,5 +32,5 @@ def tile(data, multiples):
     """
     vc_util.check_shape(data.shape)
     vc_util.check_int_list(multiples, "multiples")
-    output = akg.topi.tile(data, multiples)
+    output = _akg.topi.tile(data, multiples)
     return output
diff --git a/mindspore/akg/ops/math/__init__.py b/mindspore/_akg/ops/math/__init__.py
similarity index 100%
rename from mindspore/akg/ops/math/__init__.py
rename to mindspore/_akg/ops/math/__init__.py
diff --git a/mindspore/akg/ops/math/cast.py b/mindspore/_akg/ops/math/cast.py
similarity index 83%
rename from mindspore/akg/ops/math/cast.py
rename to mindspore/_akg/ops/math/cast.py
index 7266fd60c1daad5009a7b9a57d7d0f883af29c26..78140bfe271eea225c55916d44b386fa4cebd9dc 100644
--- a/mindspore/akg/ops/math/cast.py
+++ b/mindspore/_akg/ops/math/cast.py
@@ -13,12 +13,12 @@
 # limitations under the License.
 
 """operator dsl function: cast"""
-import akg.tvm
-import akg.topi
-from akg.utils import validation_check as vc_util
+import _akg.tvm
+import _akg.topi
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, str)
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, str)
 def cast(data, dst_type):
     """
     cast data to target type.
@@ -31,6 +31,6 @@ def cast(data, dst_type):
         tvm.tensor.Tensor, type is dst_type.
     """
     vc_util.check_shape(data.shape)
-    out = akg.topi.cast(data, dst_type)
+    out = _akg.topi.cast(data, dst_type)
 
     return out
diff --git a/mindspore/akg/ops/math/equal.py b/mindspore/_akg/ops/math/equal.py
similarity index 63%
rename from mindspore/akg/ops/math/equal.py
rename to mindspore/_akg/ops/math/equal.py
index eb446ac52bb309fb062a05154a16e8db205a8953..2dbb1ba73330d81add8e29738126376d3e2b6f43 100644
--- a/mindspore/akg/ops/math/equal.py
+++ b/mindspore/_akg/ops/math/equal.py
@@ -13,13 +13,13 @@
 # limitations under the License.
 
 """operator dsl function: equal"""
-import akg.tvm
-import akg.topi
-from akg.utils.dsl_create import produce_shapes
-from akg.utils import validation_check as vc_util
+import _akg.tvm
+import _akg.topi
+from _akg.utils.dsl_create import produce_shapes
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, akg.tvm.tensor.Tensor)
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, _akg.tvm.tensor.Tensor)
 def equal(input1, input2):
     """
     check whether input1 equals to input2.
@@ -42,13 +42,13 @@ def equal(input1, input2):
     dtype = input1.dtype
 
     # get equal compute
-    t_value = akg.tvm.compute(shape, lambda *indice: akg.tvm.const(1, dtype), "T")
-    f_value = akg.tvm.compute(shape, lambda *indice: akg.tvm.const(0, dtype), "F")
-
-    input1_bro = akg.topi.broadcast_to(input1, shape)
-    input2_bro = akg.topi.broadcast_to(input2, shape)
-    c_out = akg.tvm.compute(shape, lambda *indice: akg.tvm.expr.Select(input1_bro[indice] == input2_bro[indice],
-                                                                       t_value[indice], f_value[indice]), name="C")
-    res = akg.tvm.compute(shape, lambda *indice: c_out(*indice).astype("bool"), name="res")
+    t_value = _akg.tvm.compute(shape, lambda *indice: _akg.tvm.const(1, dtype), "T")
+    f_value = _akg.tvm.compute(shape, lambda *indice: _akg.tvm.const(0, dtype), "F")
+
+    input1_bro = _akg.topi.broadcast_to(input1, shape)
+    input2_bro = _akg.topi.broadcast_to(input2, shape)
+    c_out = _akg.tvm.compute(shape, lambda *indice: _akg.tvm.expr.Select(input1_bro[indice] == input2_bro[indice],
+                                                                         t_value[indice], f_value[indice]), name="C")
+    res = _akg.tvm.compute(shape, lambda *indice: c_out(*indice).astype("bool"), name="res")
 
     return res
diff --git a/mindspore/akg/ops/math/mean.py b/mindspore/_akg/ops/math/mean.py
similarity index 82%
rename from mindspore/akg/ops/math/mean.py
rename to mindspore/_akg/ops/math/mean.py
index a26bc29087eee4b46beda1ae8d27bba53d2095b5..8764387d33e076622416959fbeb1d9839336d288 100644
--- a/mindspore/akg/ops/math/mean.py
+++ b/mindspore/_akg/ops/math/mean.py
@@ -13,14 +13,14 @@
 # limitations under the License.
 
 """operator dsl function: mean"""
-import akg.topi
-import akg.tvm
-from akg.utils import format_transform as ft_util
-from akg.utils import validation_check as vc_util
-from akg.ops.math import sum
+import _akg.topi
+import _akg.tvm
+from _akg.utils import format_transform as ft_util
+from _akg.utils import validation_check as vc_util
+from _akg.ops.math import sum
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, (list, tuple, int, type(None)), (bool, type(None)))
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, (list, tuple, int, type(None)), (bool, type(None)))
 def mean(data, axis=None, keepdims=False):
     """
     Computes the mean of the values of a Tensor over the whole dataset.
@@ -42,6 +42,6 @@ def mean(data, axis=None, keepdims=False):
     for i in axis:
         count *= shape[i]
     output, _ = sum.sum_value(data, axis, keepdims)
-    res = akg.topi.divide(output, count)
+    res = _akg.topi.divide(output, count)
 
     return res
diff --git a/mindspore/akg/ops/math/mul.py b/mindspore/_akg/ops/math/mul.py
similarity index 86%
rename from mindspore/akg/ops/math/mul.py
rename to mindspore/_akg/ops/math/mul.py
index 8377a63d6979a0bbaf56c74531d4f26853d4a838..a690089da2d9b38b22cf0899dc2cde64a777b5ca 100644
--- a/mindspore/akg/ops/math/mul.py
+++ b/mindspore/_akg/ops/math/mul.py
@@ -13,11 +13,11 @@
 # limitations under the License.
 
 """operator dsl function: mul"""
-import akg.topi
-from akg.utils import validation_check as vc_util
+import _akg.topi
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, akg.tvm.tensor.Tensor)
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, _akg.tvm.tensor.Tensor)
 def mul(l_input, r_input):
     """
     Calculate x * y element-wise.
@@ -38,6 +38,6 @@ def mul(l_input, r_input):
     vc_util.check_shape(shape2)
     vc_util.auto_broadcast_check(shape1, shape2)
     vc_util.elemwise_dtype_check(l_input.dtype, r_input.dtype)
-    output = akg.topi.multiply(l_input, r_input)
+    output = _akg.topi.multiply(l_input, r_input)
 
     return output
diff --git a/mindspore/akg/ops/math/sub.py b/mindspore/_akg/ops/math/sub.py
similarity index 86%
rename from mindspore/akg/ops/math/sub.py
rename to mindspore/_akg/ops/math/sub.py
index a4a85b0a09feefa37565d90321fe31ddba48626f..6ae2ee51efceac888913835c8f06827f1129079a 100644
--- a/mindspore/akg/ops/math/sub.py
+++ b/mindspore/_akg/ops/math/sub.py
@@ -13,12 +13,12 @@
 # limitations under the License.
 
 """operator dsl function: sub"""
-import akg.topi
-import akg.tvm
-from akg.utils import validation_check as vc_util
+import _akg.topi
+import _akg.tvm
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, akg.tvm.tensor.Tensor)
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, _akg.tvm.tensor.Tensor)
 def sub(data1, data2):
     """
     Computes data1 - data2 elementwise, broadcast is supported.
@@ -35,6 +35,6 @@ def sub(data1, data2):
     vc_util.check_shape(data2.shape)
     vc_util.auto_broadcast_check(data1.shape, data2.shape)
 
-    res = akg.topi.subtract(data1, data2)
+    res = _akg.topi.subtract(data1, data2)
 
     return res
diff --git a/mindspore/akg/ops/math/sum.py b/mindspore/_akg/ops/math/sum.py
similarity index 79%
rename from mindspore/akg/ops/math/sum.py
rename to mindspore/_akg/ops/math/sum.py
index ea71bab9c48b78dcc9bed47b42cc5898c36d91d7..b9720469a6036939599b1502b558d496f0429868 100644
--- a/mindspore/akg/ops/math/sum.py
+++ b/mindspore/_akg/ops/math/sum.py
@@ -14,13 +14,13 @@
 
 """operator dsl function: sum"""
 
-import akg.topi
-import akg.tvm
-from akg.utils import format_transform as ft_util
-from akg.utils import validation_check as vc_util
+import _akg.topi
+import _akg.tvm
+from _akg.utils import format_transform as ft_util
+from _akg.utils import validation_check as vc_util
 
 
-@vc_util.check_input_type(akg.tvm.tensor.Tensor, (list, tuple, int, type(None)), (bool, type(None)))
+@vc_util.check_input_type(_akg.tvm.tensor.Tensor, (list, tuple, int, type(None)), (bool, type(None)))
 def sum_value(inputs, axis=None, keepdims=False):
     """
     Compute the sum of elements across dimensions of a tensor.
@@ -38,8 +38,8 @@ def sum_value(inputs, axis=None, keepdims=False):
     vc_util.check_shape(inputs.shape)
 
     if not axis:
-        output = akg.topi.identity(inputs)
+        output = _akg.topi.identity(inputs)
     else:
-        output = akg.topi.sum(inputs, axis=axis, keepdims=keepdims)
+        output = _akg.topi.sum(inputs, axis=axis, keepdims=keepdims)
 
     return output
diff --git a/mindspore/akg/save_gpu_param.py b/mindspore/_akg/save_gpu_param.py
similarity index 95%
rename from mindspore/akg/save_gpu_param.py
rename to mindspore/_akg/save_gpu_param.py
index 228bdf32ca53f91529f6f2be1c66f0e68ec987aa..ed2c9fe23abd3def8acdd4325a86b56ce21465dc 100644
--- a/mindspore/akg/save_gpu_param.py
+++ b/mindspore/_akg/save_gpu_param.py
@@ -15,9 +15,9 @@
 """save gpu param"""
 import os
 import hashlib
-import akg.tvm
-from akg.tvm import schedule
-from akg.utils import validation_check as vc_util
+import _akg.tvm
+from _akg.tvm import schedule
+from _akg.utils import validation_check as vc_util
 
 
 def get_dim(dim, axis=True):
@@ -66,7 +66,7 @@ def save_gpu_params(s, args, kernel_info):
     ptx_code = kernel_info[0]
     file_name = kernel_info[1]
     kernel_name = kernel_info[2]
-    ir = str(akg.tvm.lower(s, args, simple_mode=True))
+    ir = str(_akg.tvm.lower(s, args, simple_mode=True))
     file_path = os.path.realpath(file_name)
     if os.path.exists(file_path):
         os.remove(file_path)
diff --git a/mindspore/akg/utils/__init__.py b/mindspore/_akg/utils/__init__.py
similarity index 100%
rename from mindspore/akg/utils/__init__.py
rename to mindspore/_akg/utils/__init__.py
diff --git a/mindspore/akg/utils/dsl_create.py b/mindspore/_akg/utils/dsl_create.py
similarity index 91%
rename from mindspore/akg/utils/dsl_create.py
rename to mindspore/_akg/utils/dsl_create.py
index aaea913143ee393b4e23efd37cb9295febd3d042..9d27039b28a4b3e5ef753907871846960d040b2f 100644
--- a/mindspore/akg/utils/dsl_create.py
+++ b/mindspore/_akg/utils/dsl_create.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 """dsl create helping function"""
-import akg
-from akg.utils import format_transform as ft_util
+import _akg
+from _akg.utils import format_transform as ft_util
 
 class TensorUtils:
     """Class for creating tensor."""
@@ -33,11 +33,11 @@ class TensorUtils:
         """update tensor attrs."""
         tensor_attrs = cls.get_tensor_attrs(tensor)
         tensor_attrs.update(attrs)
-        tensor = akg.tvm.compute(tensor.shape,
-                                 lambda *indice: tensor[indice],
-                                 name=tensor.op.name,
-                                 tag=tensor.op.tag,
-                                 attrs=tensor_attrs)
+        tensor = _akg.tvm.compute(tensor.shape,
+                                  lambda *indice: tensor[indice],
+                                  name=tensor.op.name,
+                                  tag=tensor.op.tag,
+                                  attrs=tensor_attrs)
         return tensor
 
     @classmethod
@@ -61,7 +61,7 @@ class TensorUtils:
             raise RuntimeError("Shape of the input_tensor and the output_tensor should be equal, "
                                "but got %s and %s"%(input_tensor_shape, output_tensor_shape))
         output_tensor = cls.update_tensor_attrs(output_tensor, {cls.CREATE_SCH_ONLY: 1})
-        data_buf = akg.tvm.decl_buffer(input_tensor.shape, input_tensor.dtype, name=buffer_name)
+        data_buf = _akg.tvm.decl_buffer(input_tensor.shape, input_tensor.dtype, name=buffer_name)
         binds_info = {input_tensor: data_buf, output_tensor: data_buf}
         return output_tensor, binds_info
 
diff --git a/mindspore/akg/utils/format_transform.py b/mindspore/_akg/utils/format_transform.py
similarity index 86%
rename from mindspore/akg/utils/format_transform.py
rename to mindspore/_akg/utils/format_transform.py
index 816cbcaadb89af36f708c6f33765f2a79e2adb23..f83130a32a24343c509ff07fd191ca0e87234bee 100644
--- a/mindspore/akg/utils/format_transform.py
+++ b/mindspore/_akg/utils/format_transform.py
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 """format transform function"""
-import akg
+import _akg
 
 def refine_reduce_axis(input, axis):
     """make reduce axis legal."""
@@ -43,11 +43,11 @@ def refine_reduce_axis(input, axis):
 
 
 def get_shape_from_tensor(data):
-    """translate akg.tvm.shape to list type in python."""
+    """translate _akg.tvm.shape to list type in python."""
     tvm_shape = data.shape
     py_shape = []
     for i in tvm_shape:
-        if isinstance(i, akg.tvm.expr.Var):
+        if isinstance(i, _akg.tvm.expr.Var):
             py_shape.append(i)
         else:
             py_shape.append(i.value)
@@ -55,10 +55,10 @@ def get_shape_from_tensor(data):
 
 
 def tvm_shape_to_list(tvm_shape):
-    """translate akg.tvm.shape to list type in python."""
+    """translate _akg.tvm.shape to list type in python."""
     py_shape = []
     for i in tvm_shape:
-        if isinstance(i, akg.tvm.expr.Var):
+        if isinstance(i, _akg.tvm.expr.Var):
             py_shape.append(i)
         else:
             py_shape.append(i.value)
@@ -67,9 +67,9 @@ def tvm_shape_to_list(tvm_shape):
 
 def get_shape(data):
     """get shape and save it as list."""
-    if isinstance(data, akg.tvm.tensor.Tensor):
+    if isinstance(data, _akg.tvm.tensor.Tensor):
         shape = get_shape_from_tensor(data)
-    elif isinstance(data, akg.tvm.container.Array):
+    elif isinstance(data, _akg.tvm.container.Array):
         shape = tvm_shape_to_list(data)
     elif isinstance(data, int):
         shape = [data]
diff --git a/mindspore/akg/utils/validation_check.py b/mindspore/_akg/utils/validation_check.py
similarity index 99%
rename from mindspore/akg/utils/validation_check.py
rename to mindspore/_akg/utils/validation_check.py
index 72494c528160894250745d1460146a45f614b8f5..1231b3110ed8f6ad876bceb364a89a8bd095c4a6 100644
--- a/mindspore/akg/utils/validation_check.py
+++ b/mindspore/_akg/utils/validation_check.py
@@ -14,7 +14,7 @@
 
 """validation check functions"""
 from functools import wraps, reduce
-from akg.utils.format_transform import get_shape
+from _akg.utils.format_transform import get_shape
 
 MAX_DATA_SIZE = 2 ** 31
 
diff --git a/mindspore/_extends/parallel_compile/multi_compiler.py b/mindspore/_extends/parallel_compile/multi_compiler.py
index 542167888b1c06cbc18d17f0236a7054072e3f7c..86e1b684d2a8215fd4d5a9dbd0e5ce6dcd2ad882 100644
--- a/mindspore/_extends/parallel_compile/multi_compiler.py
+++ b/mindspore/_extends/parallel_compile/multi_compiler.py
@@ -32,7 +32,7 @@ def _compiletask(platform, *jsons):
 
         """
     if platform == "AKG":
-        p = __import__("akg", globals(), locals(), ['ms'], 0)
+        p = __import__("_akg", globals(), locals(), ['ms'], 0)
         func = getattr(p.ms, "compilewithjson")
         for json_item in jsons:
             res = func(json_item)
diff --git a/mindspore/ccsrc/kernel/common_utils.h b/mindspore/ccsrc/kernel/common_utils.h
index 6e3635d9045c62cf5b592bc56641aba2c04e8814..07f191cc7b7eced0b83fedc7161a0c6d6bad273a 100644
--- a/mindspore/ccsrc/kernel/common_utils.h
+++ b/mindspore/ccsrc/kernel/common_utils.h
@@ -37,7 +37,7 @@ constexpr auto kProcessorCuda = "cuda";
 constexpr auto kJsonSuffix = ".json";
 constexpr auto kInfoSuffix = ".info";
 constexpr unsigned int AUTODIFF_COMPILE_OVERTIME = 600;
-constexpr auto kAkgModule = "akg";
+constexpr auto kAkgModule = "_akg";
 constexpr auto kArgDataformat = "data_format";
 
 const std::vector<std::string> support_devices = {"aicore", "aicpu", "cuda"};
diff --git a/package.sh b/package.sh
index 0c75a1bbfd0d7c9b4f966ffd61100381600146cf..685fa1205a18eb0ccbf16a675bbe539bf118d4f8 100755
--- a/package.sh
+++ b/package.sh
@@ -77,11 +77,11 @@ cp -rf "${BUILD_PATH}/../mindspore/ops" "${PACKAGE_PATH}/mindspore"
 cp -rf "${BUILD_PATH}/../mindspore/communication" "${PACKAGE_PATH}/mindspore"
 
 if [[ "X$2" = "Xgpu" ]]; then
-    echo "package akg when gpu enable."
-    cp -rf "${BASEPATH}/mindspore/akg" "${PACKAGE_PATH}"
+    echo "package _akg when gpu enable."
+    cp -rf "${BASEPATH}/mindspore/_akg" "${PACKAGE_PATH}"
     if [[ -d "${BUILD_PATH}/mindspore/incubator-tvm" ]]; then
-        cp -rf "${BUILD_PATH}/mindspore/incubator-tvm/topi/python/topi" "${PACKAGE_PATH}/akg"
-        cp -rf "${BUILD_PATH}/mindspore/incubator-tvm/python/tvm" "${PACKAGE_PATH}/akg"
+        cp -rf "${BUILD_PATH}/mindspore/incubator-tvm/topi/python/topi" "${PACKAGE_PATH}/_akg"
+        cp -rf "${BUILD_PATH}/mindspore/incubator-tvm/python/tvm" "${PACKAGE_PATH}/_akg"
     fi
 fi
 
diff --git a/setup.py b/setup.py
index e009a9b31284660df004772604d5ccf1e5dc8b9f..221c6dc4f2b24205091294601193682c0844f78b 100644
--- a/setup.py
+++ b/setup.py
@@ -137,7 +137,7 @@ class BuildPy(build_py):
         super().run()
         mindspore_dir = os.path.join(pkg_dir, 'build', 'lib', 'mindspore')
         update_permissions(mindspore_dir)
-        mindspore_dir = os.path.join(pkg_dir, 'build', 'lib', 'akg')
+        mindspore_dir = os.path.join(pkg_dir, 'build', 'lib', '_akg')
         update_permissions(mindspore_dir)