diff --git a/python/paddle/distributed/communication/all_gather.py b/python/paddle/distributed/communication/all_gather.py
index 18f4bbab7ce08dd786bc3c59a9e11c1bd2f51c1b..791586104d559ef0495066c98ee2bc9013b7c751 100644
--- a/python/paddle/distributed/communication/all_gather.py
+++ b/python/paddle/distributed/communication/all_gather.py
@@ -16,7 +16,7 @@ import numpy as np
 
 import paddle
 import paddle.distributed.communication.stream as stream
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 
 from .serialization_utils import (
     convert_object_to_tensor,
diff --git a/python/paddle/distributed/communication/batch_isend_irecv.py b/python/paddle/distributed/communication/batch_isend_irecv.py
index a85fdcbacbf94ac89aefa9733178173795345815..1dab6f533bd2ea4a00f6bea3df2ee5394d406c6a 100644
--- a/python/paddle/distributed/communication/batch_isend_irecv.py
+++ b/python/paddle/distributed/communication/batch_isend_irecv.py
@@ -15,8 +15,7 @@
 import contextlib
 
 import paddle.distributed as dist
-import paddle.fluid.core as core
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _warn_cur_rank_not_in_group,
@@ -79,12 +78,12 @@ class P2POp:
 @contextlib.contextmanager
 def _with_batch_p2p_guard(backend):
     if backend == "NCCL":
-        core.ProcessGroupNCCL.group_start()
+        framework.core.ProcessGroupNCCL.group_start()
     try:
         yield
     finally:
         if backend == "NCCL":
-            core.ProcessGroupNCCL.group_end()
+            framework.core.ProcessGroupNCCL.group_end()
 
 
 def _check_p2p_op_list(p2p_op_list):
diff --git a/python/paddle/distributed/communication/broadcast.py b/python/paddle/distributed/communication/broadcast.py
index fd6c2219c8b25a2258032c927276625e4d82c6cf..2e5dde826be0a5e034e29af0be61a3095b21b205 100644
--- a/python/paddle/distributed/communication/broadcast.py
+++ b/python/paddle/distributed/communication/broadcast.py
@@ -15,7 +15,7 @@
 import paddle
 import paddle.distributed as dist
 import paddle.distributed.communication.stream as stream
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 
 from .serialization_utils import (
     convert_object_to_tensor,
diff --git a/python/paddle/distributed/communication/group.py b/python/paddle/distributed/communication/group.py
index f0236a2bdbb39f38c4502fa7e8d8031d7eed6deb..70f6c936545b57960b4288b05f3983fd605273b2 100644
--- a/python/paddle/distributed/communication/group.py
+++ b/python/paddle/distributed/communication/group.py
@@ -16,9 +16,7 @@ import warnings
 
 import paddle
 import paddle.distributed as dist
-import paddle.fluid.core as core
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 
 
 class Group:
@@ -239,7 +237,7 @@ def _sync_calc_stream(tensor):
         return paddle._legacy_C_ops.c_sync_calc_stream(tensor, tensor)
     else:
         op_type = 'c_sync_calc_stream'
-        helper = layer_helper.LayerHelper(op_type, **locals())
+        helper = framework.LayerHelper(op_type, **locals())
         helper.append_op(
             type=op_type,
             inputs={'X': [tensor]},
@@ -254,7 +252,7 @@ def _sync_comm_stream(tensor, ring_id=0):
         )
     else:
         op_type = 'c_sync_comm_stream'
-        helper = layer_helper.LayerHelper(op_type, **locals())
+        helper = framework.LayerHelper(op_type, **locals())
         helper.append_op(
             type=op_type,
             inputs={'X': [tensor]},
@@ -325,7 +323,7 @@ def barrier(group=None):
     if framework.in_dygraph_mode():
         group = _get_global_group() if group is None else group
         place = framework._current_expected_place()
-        if isinstance(place, core.CPUPlace):
+        if isinstance(place, framework.CPUPlace):
             task = group.process_group.barrier()
         else:
             device_id = place.get_device_id()
@@ -344,7 +342,7 @@ def barrier(group=None):
         op_type = 'barrier'
         if not isinstance(ring_id, int):
             raise ValueError("The type of 'group' for barrier must be int.")
-        helper = layer_helper.LayerHelper(op_type, **locals())
+        helper = framework.LayerHelper(op_type, **locals())
         helper.append_op(
             type=op_type,
             inputs={'X': [barrier_tensor]},
diff --git a/python/paddle/distributed/communication/reduce.py b/python/paddle/distributed/communication/reduce.py
index 973e003e0604e6612ce80e8d5860c539d09b6df5..9d4f9548d3ed8e119a064f3a4d8b07dd3d3fb0e0 100644
--- a/python/paddle/distributed/communication/reduce.py
+++ b/python/paddle/distributed/communication/reduce.py
@@ -14,8 +14,7 @@
 
 import paddle
 import paddle.distributed.communication.stream as stream
-import paddle.fluid.core as core
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 
 
 class ReduceOp:
@@ -59,13 +58,13 @@ class ReduceOp:
 def _get_reduce_op(reduce_op, func_name):
     if framework.in_dygraph_mode():
         if reduce_op == ReduceOp.SUM:
-            return core.ReduceOp.SUM
+            return framework.core.ReduceOp.SUM
         elif reduce_op == ReduceOp.MAX:
-            return core.ReduceOp.MAX
+            return framework.core.ReduceOp.MAX
         elif reduce_op == ReduceOp.MIN:
-            return core.ReduceOp.MIN
+            return framework.core.ReduceOp.MIN
         elif reduce_op == ReduceOp.PROD:
-            return core.ReduceOp.PRODUCT
+            return framework.core.ReduceOp.PRODUCT
     else:
         if reduce_op == ReduceOp.SUM:
             return 'c_{}_sum'.format(func_name)
diff --git a/python/paddle/distributed/communication/scatter.py b/python/paddle/distributed/communication/scatter.py
index ee5886c414d61d2340d9724916c87cb013a664f8..455bb5d1cf78a652f8b5cc38526e16d562864008 100644
--- a/python/paddle/distributed/communication/scatter.py
+++ b/python/paddle/distributed/communication/scatter.py
@@ -17,7 +17,7 @@ import numpy as np
 import paddle
 import paddle.distributed as dist
 import paddle.distributed.communication.stream as stream
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 
 from .serialization_utils import (
     convert_object_to_tensor,
diff --git a/python/paddle/distributed/communication/stream/all_gather.py b/python/paddle/distributed/communication/stream/all_gather.py
index 779a3c8f64cf7138b61cf50ad59abf920892d2a3..f5d21a35daac1a10adad9071e5ac7a504ff857ef 100644
--- a/python/paddle/distributed/communication/stream/all_gather.py
+++ b/python/paddle/distributed/communication/stream/all_gather.py
@@ -15,8 +15,7 @@
 import paddle
 import paddle.distributed as dist
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import _get_global_group
 
 
@@ -62,7 +61,7 @@ def _all_gather_in_dygraph(
 
 def _all_gather_in_static_mode(tensor_list, tensor, group, sync_op):
     op_type = 'c_allgather'
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     out = helper.create_variable_for_type_inference(dtype=tensor.dtype)
     for elem in tensor_list:
         data_feeder.check_variable_and_dtype(
diff --git a/python/paddle/distributed/communication/stream/all_reduce.py b/python/paddle/distributed/communication/stream/all_reduce.py
index 412085b1b1720ab84f2a7378b5ae49441aed4b55..786e4284c755645fdedf0e4b722fbe0e2734a07e 100644
--- a/python/paddle/distributed/communication/stream/all_reduce.py
+++ b/python/paddle/distributed/communication/stream/all_reduce.py
@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _warn_cur_rank_not_in_group,
@@ -60,7 +59,7 @@ def _all_reduce_in_static_mode(tensor, op, group, sync_op, use_calc_stream):
 
     # TODO: Support task and use task.wait in static graph mode
     #       Use use_calc_stream rather than sync_op
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     helper.append_op(
         type=op_type,
         inputs={'X': [tensor]},
diff --git a/python/paddle/distributed/communication/stream/all_to_all.py b/python/paddle/distributed/communication/stream/all_to_all.py
index d64ccb742ef08cac09217bffd2b2f59b8d185ceb..4a804cf4ababf15559d97300d9b6f778f282bf83 100644
--- a/python/paddle/distributed/communication/stream/all_to_all.py
+++ b/python/paddle/distributed/communication/stream/all_to_all.py
@@ -15,8 +15,7 @@
 import paddle
 import paddle.distributed as dist
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _warn_cur_rank_not_in_group,
@@ -73,7 +72,7 @@ def _all_to_all_in_static_mode(
     op_type = 'alltoall'
     ring_id = 0 if group is None else group.id
     nranks = dist.get_world_size()
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
 
     in_tensor = in_tensor_or_tensor_list
     if isinstance(in_tensor_or_tensor_list, list):
diff --git a/python/paddle/distributed/communication/stream/broadcast.py b/python/paddle/distributed/communication/stream/broadcast.py
index cb6fbc75d152807c6586fff93eb4321fb1be1f07..c7f6fc3203de34f8981db89e277ec6fad1f70041 100644
--- a/python/paddle/distributed/communication/stream/broadcast.py
+++ b/python/paddle/distributed/communication/stream/broadcast.py
@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _get_or_throw_group_rank,
@@ -57,7 +56,7 @@ def _broadcast_in_static_mode(
     )
 
     op_type = 'c_broadcast'
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     ring_id = 0 if group is None else group.id
 
     helper.append_op(
diff --git a/python/paddle/distributed/communication/stream/recv.py b/python/paddle/distributed/communication/stream/recv.py
index fcd007e6d333dba5d3bdaa7d9574c6a60b53a850..7b623d67203c2838eb362baf53c18cc63d45ca8e 100644
--- a/python/paddle/distributed/communication/stream/recv.py
+++ b/python/paddle/distributed/communication/stream/recv.py
@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _get_or_throw_group_rank,
@@ -48,7 +47,7 @@ def _recv_in_static_mode(
         'recv',
     )
     ring_id = 0 if group is None else group.id
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     helper.append_op(
         type=op_type,
         outputs={'Out': [tensor]},
diff --git a/python/paddle/distributed/communication/stream/reduce.py b/python/paddle/distributed/communication/stream/reduce.py
index 8bd81bd586a98b1d2bece7375acb10e9976278d3..e81ff4de9c047d72db28ef934603f9c828fe950c 100644
--- a/python/paddle/distributed/communication/stream/reduce.py
+++ b/python/paddle/distributed/communication/stream/reduce.py
@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _get_or_throw_group_rank,
@@ -63,7 +62,7 @@ def _reduce_in_static_mode(
     op_type = _get_reduce_op(op, "reduce")
     ring_id = 0 if group is None else group.id
 
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     helper.append_op(
         type=op_type,
         inputs={'X': [tensor]},
diff --git a/python/paddle/distributed/communication/stream/reduce_scatter.py b/python/paddle/distributed/communication/stream/reduce_scatter.py
index 3442365863002efae8008317dfadad3ae7f3ce66..b0776246d0af98735e4e90e2713899e0632afc43 100644
--- a/python/paddle/distributed/communication/stream/reduce_scatter.py
+++ b/python/paddle/distributed/communication/stream/reduce_scatter.py
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import paddle
-import paddle.fluid.framework as framework
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _warn_cur_rank_not_in_group,
diff --git a/python/paddle/distributed/communication/stream/scatter.py b/python/paddle/distributed/communication/stream/scatter.py
index 6f332fbbd6fb3301b5c010fc0e899896239bcb44..db2010e4116dc88d9344eb4bd61c753e34e08ab3 100644
--- a/python/paddle/distributed/communication/stream/scatter.py
+++ b/python/paddle/distributed/communication/stream/scatter.py
@@ -17,8 +17,7 @@ import warnings
 import paddle
 import paddle.distributed as dist
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _get_or_throw_group_rank,
@@ -113,7 +112,7 @@ def _scatter_in_static_mode(
     )
 
     op_type = 'c_scatter'
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     helper.append_op(
         type=op_type,
         inputs={'X': [input_tensor]},
diff --git a/python/paddle/distributed/communication/stream/send.py b/python/paddle/distributed/communication/stream/send.py
index e18a9a5738482f10ac424cdce7138cf621b32aed..30a965e82154ce72900b511e97371d836aca14fc 100644
--- a/python/paddle/distributed/communication/stream/send.py
+++ b/python/paddle/distributed/communication/stream/send.py
@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid.data_feeder as data_feeder
-import paddle.fluid.framework as framework
-import paddle.fluid.layer_helper as layer_helper
+import paddle.framework as framework
 from paddle.distributed.communication.group import (
     _get_global_group,
     _get_or_throw_group_rank,
@@ -49,7 +48,7 @@ def _send_in_static_mode(
     )
 
     ring_id = 0 if group is None else group.id
-    helper = layer_helper.LayerHelper(op_type, **locals())
+    helper = framework.LayerHelper(op_type, **locals())
     helper.append_op(
         type=op_type,
         inputs={'X': [tensor]},