[dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode() (#53856)

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* [dygraph]unify _non_static_mode() in_dygraph_mode() and in_dynamic_mode()

* fixed cyclic reference that caused patial import

* fixed bad change

* fix bad import

* fix bad import

* fix bad import

* fix ut failed caused by change in_dynamic_mode

* fix ut failed caused by change in_dynamic_mode

* fixed usage of in_dynamic_mode() or in_dygraph_mode()

* revert python3 to python in .pre-commit-config.yaml

* fix merge conflicts

python/paddle/amp/grad_scaler.py

@@ -19,10 +19,11 @@ from enum import Enum
 import numpy as np
 
 from paddle import _C_ops, _legacy_C_ops
-from paddle.fluid import core, in_dygraph_mode
+from paddle.fluid import core
 from paddle.fluid.data_feeder import check_type
 from paddle.fluid.dygraph import to_variable
 from paddle.fluid.framework import _dygraph_tracer, dygraph_only
+from paddle.framework import in_dynamic_mode
 
 from .auto_cast import amp_global_state
 
@@ -307,7 +308,7 @@ class AmpScaler:
                         else:
                             param_grads_fp32.append(param._grad_ivar())
         else:
-            if in_dygraph_mode():
+            if in_dynamic_mode():
                 # It is very time-consuming to call c++ functions in a loop on the python side.
                 # We put this part of the code on the c++ side to improve the speed in eager mode.
                 (

python/paddle/autograd/autograd.py

@@ -689,7 +689,7 @@ def _grad_for_jacobian(ys, xs, v=None):
             Tensor is the sum of gradients of outputs with respect to the i-th
             inputs.
     """
-    if paddle.fluid._non_static_mode():
+    if paddle.in_dynamic_mode():
         # paddle.grad returns a list though the inputs is a signle Tensor. The
         # follow code snippet fixes the problem by return the first element of
         # xs_grad when the xs is a signle Tensor.

python/paddle/common_ops_import.py

@@ -25,7 +25,6 @@ from paddle.fluid.framework import (  # noqa: F401
     Variable,
     _create_tensor,
     _dygraph_tracer,
-    _non_static_mode,
     convert_np_dtype_to_dtype_,
     default_main_program,
     device_guard,

python/paddle/distributed/auto_parallel/engine.py

@@ -28,7 +28,7 @@ from paddle.distributed import fleet
 from paddle.fluid.executor import _to_name_str
 from paddle.framework import IrGraph
 from paddle.framework import _current_expected_place as _get_device
-from paddle.framework import core, in_dygraph_mode
+from paddle.framework import core, in_dynamic_mode
 from paddle.metric import Metric
 from paddle.static import InputSpec, Operator, Variable, global_scope
 
@@ -312,7 +312,7 @@ class Engine:
         return inputs_spec, labels_spec
 
     def _prepare_data_tensor(self, inputs_spec, labels_spec, inputs, labels):
-        if in_dygraph_mode() or self._dygraph_mode:
+        if in_dynamic_mode() or self._dygraph_mode:
             raise ValueError("Only support static graph mode.")
 
         if inputs_spec:
@@ -561,7 +561,7 @@ class Engine:
         self._has_prepared[mode] = True
 
     def _build(self, mode):
-        if in_dygraph_mode() or self._dygraph_mode:
+        if in_dynamic_mode() or self._dygraph_mode:
             paddle.disable_static()
             self._dygraph_mode = True
             self._logger.info("Building model with 'to_static' method.")
@@ -1789,7 +1789,7 @@ class Engine:
             self._build(mode)
             self._plan(mode)
         else:
-            if in_dygraph_mode() or self._dygraph_mode:
+            if in_dynamic_mode() or self._dygraph_mode:
                 raise ValueError(
                     "Please call `prepare()` or `fit()` or  `evaluate()` or  `predict()` before calling `cost()`."
                 )

python/paddle/distributed/auto_parallel/process_group.py

@@ -16,7 +16,7 @@ from collections import OrderedDict
 
 import paddle
 from paddle import _legacy_C_ops
-from paddle.framework import core, in_dygraph_mode
+from paddle.framework import core, in_dynamic_mode
 from paddle.tensor import fill_constant
 
 from ..collective import _get_global_env, _new_ring_id
@@ -177,7 +177,7 @@ class ProcessGroup:
                 )
             tmp = (
                 paddle.to_tensor([1], dtype="int32")
-                if in_dygraph_mode()
+                if in_dynamic_mode()
                 else fill_constant([0], dtype="int32", value="1")
             )
             # use legacy ops

python/paddle/distributed/collective.py

@@ -19,7 +19,7 @@ import paddle
 
 # (TODO: GhostScreaming) It will be removed later.
 from paddle.fluid import core
-from paddle.framework import in_dygraph_mode
+from paddle.framework import in_dynamic_mode
 
 from .communication.group import Group, _add_new_group, is_initialized
 from .fleet.layers.mpu.mp_ops import _c_concat  # noqa: F401
@@ -128,7 +128,7 @@ def _set_group_map_backend(group, backend):
 
 def _new_ring_id():
     # NOTE(liyurui): For compatible reason, auto parallel and eager mode relay on previous syntax.
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         global _start_ring_id
         _start_ring_id += 1
         return _start_ring_id + max(_get_global_env().nrings, 9)
@@ -198,7 +198,7 @@ def new_group(ranks=None, backend=None, timeout=_default_timeout):
     """
     global _custom_gid
     global _group_map
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         global _default_group_name
         gid = _custom_gid if _custom_gid else _new_ring_id()
         group_name = _default_group_name + str(gid)
@@ -292,7 +292,7 @@ def new_group(ranks=None, backend=None, timeout=_default_timeout):
     # hang caused by cross-creation of new_group
     tmp = (
         paddle.to_tensor([1], dtype="int32")
-        if in_dygraph_mode()
+        if in_dynamic_mode()
         else paddle.full([0], 1, dtype="int32")
     )
     paddle.distributed.all_reduce(tmp, sync_op=True)

python/paddle/distributed/communication/all_gather.py

@@ -102,7 +102,7 @@ def all_gather_object(object_list, obj, group=None):
             # [{'foo': [1, 2, 3]}, {'bar': [4, 5, 6]}] (2 GPUs)
     """
     assert (
-        framework.in_dygraph_mode()
+        framework.in_dynamic_mode()
     ), "all_gather_object doesn't support static graph mode."
 
     tensor, len_of_tensor = convert_object_to_tensor(obj)

python/paddle/distributed/communication/batch_isend_irecv.py

@@ -159,7 +159,7 @@ def batch_isend_irecv(p2p_op_list):
     if _warn_cur_rank_not_in_group(group):
         return
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         backend = group.backend
         tasks = []

python/paddle/distributed/communication/broadcast.py

@@ -102,7 +102,7 @@ def broadcast_object_list(object_list, src, group=None):
             # [{"bar": [4, 5, 6]}] (2 GPUs)
     """
     assert (
-        framework.in_dygraph_mode()
+        framework.in_dynamic_mode()
     ), "broadcast_object_list doesn't support static graph mode."
 
     rank = dist.get_rank()

python/paddle/distributed/communication/gather.py

@@ -55,6 +55,6 @@ def gather(tensor, gather_list=None, dst=0, group=None, sync_op=True):
             # [] (2 GPUs, out for rank 1)
     """
     assert (
-        framework.in_dygraph_mode()
+        framework.in_dynamic_mode()
     ), "gather doesn't support static graph mode yet."
     return stream.gather(tensor, gather_list, dst, group, sync_op)

python/paddle/distributed/communication/group.py

@@ -231,7 +231,7 @@ def get_group(id=0):
 
 
 def _sync_calc_stream(tensor):
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         return paddle._legacy_C_ops.c_sync_calc_stream(tensor, tensor)
     else:
         op_type = 'c_sync_calc_stream'
@@ -244,7 +244,7 @@ def _sync_calc_stream(tensor):
 
 
 def _sync_comm_stream(tensor, ring_id=0):
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         return paddle._legacy_C_ops.c_sync_comm_stream(
             [tensor], [tensor], 'ring_id', ring_id
         )
@@ -318,7 +318,7 @@ def barrier(group=None):
     if group is not None and not group.is_member():
         return
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         place = framework._current_expected_place()
         if isinstance(place, framework.CPUPlace):
@@ -332,7 +332,7 @@ def barrier(group=None):
     ring_id = 0 if group is None else group.id
 
     barrier_tensor = paddle.full([1], 1, dtype="int32")
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         return paddle._legacy_C_ops.barrier(
             barrier_tensor, barrier_tensor, 'ring_id', ring_id
         )

python/paddle/distributed/communication/reduce.py

@@ -56,7 +56,7 @@ class ReduceOp:
 
 
 def _get_reduce_op(reduce_op, func_name):
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         if reduce_op == ReduceOp.SUM:
             return framework.core.ReduceOp.SUM
         elif reduce_op == ReduceOp.MAX:

python/paddle/distributed/communication/scatter.py

@@ -108,7 +108,7 @@ def scatter_object_list(
             # [{'bar': [4, 5, 6]}] (2 GPUs, out for rank 1)
     """
     assert (
-        framework.in_dygraph_mode()
+        framework.in_dynamic_mode()
     ), "scatter_object_list doesn't support static graph mode."
 
     rank = dist.get_rank()

python/paddle/distributed/communication/stream/all_gather.py

@@ -171,7 +171,7 @@ def all_gather(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         if paddle.is_tensor(tensor_or_tensor_list):
             return _all_gather_into_tensor_in_dygraph(
                 tensor_or_tensor_list, tensor, group, sync_op, use_calc_stream

python/paddle/distributed/communication/stream/all_reduce.py

@@ -116,7 +116,7 @@ def all_reduce(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         return _all_reduce_in_dygraph(
             tensor, op, group, sync_op, use_calc_stream

python/paddle/distributed/communication/stream/all_to_all.py

@@ -185,7 +185,7 @@ def alltoall(
     if in_tensor_or_tensor_list is None:
         raise RuntimeError("The input should be specified.")
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         out_is_tensor = paddle.is_tensor(out_tensor_or_tensor_list)
         in_is_tensor = paddle.is_tensor(in_tensor_or_tensor_list)
@@ -335,7 +335,7 @@ def alltoall_single(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         return _alltoall_single_in_dygraph(
             out_tensor,

python/paddle/distributed/communication/stream/broadcast.py

@@ -117,7 +117,7 @@ def broadcast(tensor, src, group=None, sync_op=True, use_calc_stream=False):
             "use_calc_stream can only be True in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         src_rank_in_group = _get_or_throw_group_rank(src, group)
 

python/paddle/distributed/communication/stream/gather.py

@@ -99,7 +99,7 @@ def gather(
     """
 
     assert (
-        framework.in_dygraph_mode()
+        framework.in_dynamic_mode()
     ), "gather doesn't support static graph mode yet."
 
     if _warn_cur_rank_not_in_group(group):

python/paddle/distributed/communication/stream/recv.py

@@ -105,7 +105,7 @@ def recv(tensor, src=0, group=None, sync_op=True, use_calc_stream=False):
             "use_calc_stream can only be True in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         src_rank_in_group = _get_or_throw_group_rank(src, group)
 

python/paddle/distributed/communication/stream/reduce.py

@@ -131,7 +131,7 @@ def reduce(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         dst_rank_in_group = _get_or_throw_group_rank(dst, group)
         return _reduce_in_dygraph(

python/paddle/distributed/communication/stream/reduce_scatter.py

@@ -158,7 +158,7 @@ def reduce_scatter(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         if paddle.is_tensor(tensor_or_tensor_list):
             return _reduce_scatter_tensor_in_dygraph(
@@ -243,7 +243,7 @@ def _reduce_scatter_base(
             "use_calc_stream can only be true in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         return _reduce_scatter_tensor_in_dygraph(
             out_tensor,

python/paddle/distributed/communication/stream/scatter.py

@@ -201,7 +201,7 @@ def scatter(
             )
         tensor_or_tensor_list = []
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         src_rank_in_group = _get_or_throw_group_rank(src, group)
         if paddle.is_tensor(tensor_or_tensor_list):

python/paddle/distributed/communication/stream/send.py

@@ -104,7 +104,7 @@ def send(tensor, dst=0, group=None, sync_op=True, use_calc_stream=False):
             "use_calc_stream can only be True in sync op behavior."
         )
 
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = _get_global_group() if group is None else group
         dst_rank_in_group = _get_or_throw_group_rank(dst, group)
 

python/paddle/distributed/fleet/base/role_maker.py

@@ -1176,7 +1176,7 @@ class PaddleCloudRoleMaker(RoleMakerBase):
             else:
                 self._collective_env()
             self._role_is_generated = True
-            if not paddle.framework.in_dynamic_mode():
+            if not paddle.in_dynamic_mode():
                 self._gloo_init()
 
 

python/paddle/distributed/fleet/fleet.py

@@ -17,9 +17,8 @@ import os
 
 import paddle
 from paddle.fluid import compiler
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.wrapped_decorator import wrap_decorator
-from paddle.framework import _global_flags
+from paddle.framework import _global_flags, in_dynamic_mode
 from paddle.framework.ir import apply_build_strategy
 
 from .base import topology as tp
@@ -281,7 +280,7 @@ class Fleet:
                         "CUDA_VISIBLE_DEVICES shoule be set only 1 card if you use `python` to launch fleet program."
                     )
 
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             if self.worker_num() == 1:
                 # if worker_num is 1, should construct default topology & hcg
                 self._topology = tp.CommunicateTopology()
@@ -1270,7 +1269,7 @@ class Fleet:
             )
         else:
             if (
-                in_dygraph_mode()
+                in_dynamic_mode()
                 or self._role_maker._is_non_distributed()
                 or self._is_collective
             ):
@@ -1286,7 +1285,7 @@ class Fleet:
         context["user_defined_strategy"] = copy.deepcopy(
             self._user_defined_strategy
         )
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             # imitate target optimizer retrieval
             target_opt = self.user_defined_optimizer
             self._context = context

python/paddle/distributed/fleet/layers/mpu/mp_ops.py

@@ -16,7 +16,7 @@ import paddle
 from paddle import _legacy_C_ops
 from paddle.distributed import collective
 from paddle.fluid.data_feeder import check_dtype, check_variable_and_dtype
-from paddle.framework import LayerHelper, _create_tensor, in_dygraph_mode
+from paddle.framework import LayerHelper, _create_tensor, in_dynamic_mode
 from paddle.nn import Layer
 from paddle.nn.utils import dygraph_utils
 
@@ -39,7 +39,7 @@ def _c_identity(tensor, group=None):
         return
     ring_id = 0 if group is None else group.id
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         from paddle.autograd import PyLayer
 
         class c_identity_eager(PyLayer):
@@ -108,7 +108,7 @@ def _c_concat(tensor, group=None):
     rank = group.rank
     nranks = group.nranks
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.c_concat(
             tensor,
             'ring_id',
@@ -174,7 +174,7 @@ def _c_split(tensor, group=None):
         else group.nranks
     )
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.c_split(
             tensor,
             'use_calc_stream',
@@ -226,7 +226,7 @@ def _mp_allreduce(
     if group is not None and not group.is_member():
         return
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         group = collective._get_default_group() if group is None else group
         assert op == ReduceOp.SUM, f"Unknown parameter: {op}."
 
@@ -308,7 +308,7 @@ def _c_lookup_table(table, index, start_index=0, name=None):
     Returns:
         Tensor.
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.c_embedding(
             table, index, "start_index", start_index
         )
@@ -401,7 +401,7 @@ def _c_softmax_with_cross_entropy(
     if input_dims - 1 == label_dims:
         label = paddle.unsqueeze(label, axis=-1)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         softmax, loss = _legacy_C_ops.c_softmax_with_cross_entropy(
             logits,
             label,
@@ -445,7 +445,7 @@ def _linear(x, weight, bias=None, name=None):
     """
     Fuction Linear
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         pre_bias = _create_tensor(dtype=x.dtype)
         _legacy_C_ops.matmul(
             x,
@@ -810,7 +810,7 @@ def split(
             supported_operations
         )
     )
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         raise ValueError(
             "paddle.distributed.split cannot be used in dynamic "
             "graph mode, plese use ParallelEmbedding, ParallelRowLinear, "

python/paddle/distributed/fleet/layers/mpu/random.py

@@ -21,8 +21,7 @@ from paddle import _legacy_C_ops
 from paddle.common_ops_import import Variable
 from paddle.fluid import core
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import in_dygraph_mode
-from paddle.framework import LayerHelper
+from paddle.framework import LayerHelper, in_dynamic_mode
 
 __all__ = []
 
@@ -218,7 +217,7 @@ def dropout(
     )  # semantic transfer
 
     # dygraph using tracker, doesn't need determinate seed
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out, mask = _legacy_C_ops.dropout(
             x,
             'dropout_prob',

python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py

@@ -22,7 +22,7 @@ import paddle
 from paddle.common_ops_import import LayerHelper
 from paddle.fluid.dygraph import base as imperative_base
 from paddle.fluid.optimizer import Momentum, Optimizer
-from paddle.framework import core, in_dygraph_mode
+from paddle.framework import core, in_dynamic_mode
 from paddle.nn.clip import ClipGradByNorm, append_gradient_clip_ops
 from paddle.regularizer import L1Decay, L2Decay
 from paddle.static import create_global_var
@@ -46,7 +46,7 @@ class DGCMomentumOptimizer(Optimizer):
         grad_clip=None,
         name=None,
     ):
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             raise Exception("In dygraph, don't support DGCMomentumOptimizer.")
 
         assert (

python/paddle/distributed/fleet/meta_parallel/parallel_layers/pp_layers.py

@@ -533,7 +533,7 @@ class PipelineLayer(nn.Layer):
         for key, comm in self.shared_comm.items():
             param = getattr(self.shared_layers[key], comm['weight_attr'])
             # need use trace_op to allreduce weight
-            if framework.in_dygraph_mode():
+            if framework.in_dynamic_mode():
                 with paddle.framework.no_grad():
                     paddle.distributed.all_reduce(
                         param.grad

python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py

@@ -555,7 +555,7 @@ class PipelineParallelWithInterleave(PipelineParallel):
         super().__init__(layers=layers, hcg=hcg, strategy=strategy)
         assert layers.get_num_virtual_stages() > 1
         assert (
-            framework.in_dygraph_mode()
+            framework.in_dynamic_mode()
         ), "virtual pipeline stage with interleave only support eager dygraph mode"
         # setup for interleave scheduler
         self.num_model_chunks = layers.get_num_virtual_stages()

python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py

@@ -220,7 +220,7 @@ def _partial_send_op(
     tensor, group, use_calc_stream, ring_id, dst, nranks, rank_id
 ):
     dst_rank_in_group = dst if group is None else group.get_group_rank(dst)
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         group = (
             paddle.distributed.collective._get_default_group()
             if group is None
@@ -291,7 +291,7 @@ def recv_partial(
     else:
         if use_calc_stream:
             recv_op = paddle.distributed.recv
-        elif framework.in_dygraph_mode():
+        elif framework.in_dynamic_mode():
             recv_op = paddle.distributed.irecv
         return recv_op(tensor.detach(), src=src_rank, group=group)
 
@@ -656,7 +656,7 @@ def _p2p_helper(
                     tasks.append(task)
         _xpu_comm_group_end()
     if not sync_recv:
-        if framework.in_dygraph_mode():
+        if framework.in_dynamic_mode():
             # wait irecv tasks in eager dygraph mode with new comm library
             for task in tasks:
                 assert task is not None

python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_stage3.py

@@ -38,7 +38,7 @@ def _all_gather(tensor, buffer_size, group):
     """
 
     assert group is not None
-    if framework.in_dygraph_mode():
+    if framework.in_dynamic_mode():
         out = paddle.zeros([buffer_size], dtype=tensor.dtype)
         task = group.process_group.all_gather(tensor, out)
         return out, task

python/paddle/distributed/fleet/optimizer.py

@@ -15,7 +15,7 @@
 import copy
 
 from paddle.distributed import fleet
-from paddle.fluid.framework import in_dygraph_mode
+from paddle.framework import in_dynamic_mode
 
 from .meta_optimizers import HeterParallelOptimizer, HybridParallelOptimizer
 from .utils.log_util import logger
@@ -81,7 +81,7 @@ def _dygraph_distributed_optimizer(optimizer, strategy=None):
 
 
 def distributed_optimizer(*args, **kwargs):
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _dygraph_distributed_optimizer(*args, **kwargs)
     else:
         return fleet.fleet.distributed_optimizer(*args, **kwargs)

python/paddle/distributed/fleet/recompute/recompute.py

@@ -21,7 +21,7 @@ from paddle.autograd import PyLayer
 from paddle.distributed.fleet.meta_parallel.parallel_layers.random import (
     get_rng_state_tracker,
 )
-from paddle.framework import core, in_dygraph_mode
+from paddle.framework import core, in_dynamic_mode
 
 from ..utils.log_util import logger
 
@@ -198,7 +198,7 @@ class RecomputeFunction(PyLayer):
                     forward_outputs_with_grad, backward_inputs_with_grad
                 )
 
-            if in_dygraph_mode():
+            if in_dynamic_mode():
                 grads = tuple(
                     inp._grad_ivar()
                     for inp in detached_inputs

python/paddle/distributed/fleet/recompute/recompute_hybrid.py

@@ -161,7 +161,7 @@ class _HPRecomputeFunction(PyLayer):
                 #  If not marked non_differentiable, all output tensors' attr `stop gradient`
                 #  will be reset to `False` in c++ backend.
                 #  See https://github.com/PaddlePaddle/Paddle/blob/9d62efb0e6e5373823039d9eda96cd5905426c0a/paddle/fluid/pybind/eager_py_layer.cc#L388
-                if framework.in_dygraph_mode() and state:
+                if framework.in_dynamic_mode() and state:
                     ctx.mark_non_differentiable(arg)
             else:
                 ctx.inputs.append(arg)

python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py

@@ -20,8 +20,7 @@ from paddle.distributed import fleet
 
 # (TODO: GhostScreaming) It will be removed later.
 from paddle.fluid import core
-from paddle.fluid.framework import in_dygraph_mode
-from paddle.framework import Block, Program
+from paddle.framework import Block, Program, in_dynamic_mode
 
 
 class HybridParallelInferenceHelper:
@@ -205,7 +204,7 @@ class HybridParallelInferenceHelper:
             self._device = "gpu"
         assert self._device, "Only gpu are supported."
 
-        assert not in_dygraph_mode(), "Only static graph mode is supported."
+        assert not in_dynamic_mode(), "Only static graph mode is supported."
 
         op_maker = core.op_proto_and_checker_maker
         self._op_role = op_maker.OpRole

python/paddle/distributed/fleet/utils/hybrid_parallel_util.py

@@ -17,7 +17,7 @@ from paddle import framework
 from paddle.distributed.parallel import (
     _split_tensors,
     build_groups,
-    in_dygraph_mode,
+    in_dynamic_mode,
     sync_params_buffers,
 )
 
@@ -131,7 +131,7 @@ def _broadcast_data_help(data, shape, dtype, hcg):
     )
 
     if mp_rank != 0:
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             data._clear_data()
             input_data._share_buffer_to(data)
         else:
@@ -174,7 +174,7 @@ def broadcast_input_data(hcg, *inputs, **kwargs):
     for v in inputs:
         if isinstance(v, core.eager.Tensor):
             with framework.no_grad():
-                if in_dygraph_mode() and not eval(f"v.place.is_{dev}_place")():
+                if in_dynamic_mode() and not eval(f"v.place.is_{dev}_place")():
                     v_gpu = v._copy_to(place, True)
                     v._clear_data()
                     v_gpu._share_buffer_to(v)
@@ -185,7 +185,7 @@ def broadcast_input_data(hcg, *inputs, **kwargs):
     for k, v in kwargs.items():
         if isinstance(v, core.eager.Tensor):
             with framework.no_grad():
-                if in_dygraph_mode() and not eval(f"v.place.is_{dev}_place")():
+                if in_dynamic_mode() and not eval(f"v.place.is_{dev}_place")():
                     v_gpu = v._copy_to(place, True)
                     v._clear_data()
                     v_gpu._share_buffer_to(v)
@@ -217,7 +217,7 @@ def fused_allreduce_gradients_with_group(
 ):
     apply_func = (
         _apply_collective_grads_eager
-        if in_dygraph_mode()
+        if in_dynamic_mode()
         else _apply_collective_grads
     )
     with framework.no_grad():

python/paddle/distributed/fleet/utils/mix_precision_utils.py

@@ -120,7 +120,7 @@ class MixPrecisionOptimizer:
                 if param.stop_gradient:
                     continue
                 grad_var = param.main_grad
-                if framework.in_dygraph_mode():
+                if paddle.in_dynamic_mode():
                     if (
                         hasattr(grad_var, "is_selected_rows")
                         and grad_var.is_selected_rows()
@@ -151,7 +151,7 @@ class MixPrecisionOptimizer:
                     if param.stop_gradient:
                         continue
                     grad_var = param.main_grad
-                    if framework.in_dygraph_mode():
+                    if paddle.in_dynamic_mode():
                         if (
                             hasattr(grad_var, "is_selected_rows")
                             and grad_var.is_selected_rows()

python/paddle/distributed/models/moe/utils.py

@@ -14,7 +14,7 @@
 
 from paddle import _legacy_C_ops
 from paddle.common_ops_import import check_variable_and_dtype
-from paddle.framework import LayerHelper, in_dygraph_mode
+from paddle.framework import LayerHelper, in_dynamic_mode
 
 
 def _number_count(numbers, upper_range):
@@ -39,7 +39,7 @@ def _number_count(numbers, upper_range):
             number_count = paddle.distributed.utils.number_count(numbers, upper_range)
             print(number_count) # the result: [2, 0, 2, 0, 0, 0]
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.number_count(numbers, 'upper_range', upper_range)
     else:
         op_type = 'number_count'
@@ -86,7 +86,7 @@ def _assign_pos(x, cum_count):
             pos = paddle.distributed.utils.assign_pos(x=numbers, cum_count=num_cum)
             print(pos) # the result: (2, 0, 3, 1)
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.assign_pos(x, cum_count, cum_count[-1])
     else:
         op_type = 'assign_pos'
@@ -121,7 +121,7 @@ def _random_routing(topk_idx, topk_value, prob, topk=2):
         prob: random prob, shape=(topk_idx.shape[0],)
     """
     if topk == 2:
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             return _legacy_C_ops.random_routing(prob, topk_value, topk_idx)
         else:
             raise RuntimeError("Not supporting static graph mode now")
@@ -150,7 +150,7 @@ def _limit_by_capacity(expert_count, capacity, n_worker):
             out = paddle.distributed.utils.limit_by_capacity(expert_count, capacity, n_work)
             print(out) # the result: [1, 2, 2, 4, 3, 3]
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.limit_by_capacity(
             expert_count, capacity, 'n_worker', n_worker
         )
@@ -195,7 +195,7 @@ def _prune_gate_by_capacity(gate_idx, expert_count, n_expert, n_worker):
             # Tensor(shape=[8], dtype=int32, place=CUDAPlace(0), stop_gradient=True,
               [1, 3, 3, 3, -1, 2, 1, 1])
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.prune_gate_by_capacity(
             gate_idx, expert_count, "n_expert", n_expert, "n_worker", n_worker
         )

python/paddle/distributed/parallel.py

@@ -47,7 +47,7 @@ from paddle.distributed.fleet.launch_utils import check_backend
 # (TODO: GhostScreaming) It will be removed later.
 from paddle.framework import _set_expected_place
 from paddle.framework import base as imperative_base
-from paddle.framework import core, in_dygraph_mode
+from paddle.framework import core, in_dynamic_mode
 from paddle.nn.layer import layers
 from paddle.utils import deprecated
 
@@ -101,7 +101,7 @@ def _reshape_inplace(x, shape):
 
 @framework.dygraph_only
 def _split_tensors(coalesced_grads_and_grad_vars):
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         for (
             coalesced_grad,
             origin_grad_vars,
@@ -356,7 +356,7 @@ class DataParallel(layers.Layer):
         super().__init__(layers.full_name() + "_data_parallel")
 
         assert (
-            in_dygraph_mode()
+            in_dynamic_mode()
         ), "It's not supported to construct DataParallel in static graph mode."
 
         self._layers = layers
@@ -381,7 +381,7 @@ class DataParallel(layers.Layer):
                 "constructing the DataParallel."
             )
 
-            if in_dygraph_mode():
+            if in_dynamic_mode():
                 self.group = (
                     paddle.distributed.collective._get_default_group()
                     if self.group is None
@@ -456,7 +456,7 @@ class DataParallel(layers.Layer):
             check_layer_sparse(sublayer) for sublayer, _ in layers_param
         ]
 
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             self.group_indices = core.eager_assign_group_by_size(
                 trainable_parameters,
                 is_sparse_gradient,
@@ -1041,7 +1041,7 @@ def init_parallel_env():
 
     group = None
 
-    if backend in _valid_backend_list and in_dygraph_mode():
+    if backend in _valid_backend_list and in_dynamic_mode():
         if _default_group_name in _get_group_map_by_name():
             return _get_group_map_by_name()[_default_group_name]
         _set_default_backend(backend)
@@ -1212,7 +1212,7 @@ def get_rank(group=None):
             print("The rank is %d" % dist.get_rank())
             # The rank is 0
     """
-    if in_dygraph_mode() and group:
+    if in_dynamic_mode() and group:
         return group.rank
 
     assert group is None, "Only support group argument in eager mode."
@@ -1244,7 +1244,7 @@ def get_world_size(group=None):
             print("The world_size is %d" % dist.get_world_size())
             # The world_size is 1
     """
-    if in_dygraph_mode() and group:
+    if in_dynamic_mode() and group:
         return group.world_size
 
     assert group is None, "Only support group argument in eager mode."

python/paddle/distributed/utils/moe_utils.py

@@ -14,7 +14,7 @@
 
 from paddle import _legacy_C_ops
 from paddle.common_ops_import import check_variable_and_dtype
-from paddle.framework import LayerHelper, in_dygraph_mode
+from paddle.framework import LayerHelper, in_dynamic_mode
 
 
 def global_scatter(
@@ -102,7 +102,7 @@ def global_scatter(
         return
 
     ring_id = 0 if group is None else group.id
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.global_scatter(
             x,
             local_count,
@@ -219,7 +219,7 @@ def global_gather(
         return
 
     ring_id = 0 if group is None else group.id
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.global_gather(
             x,
             local_count,

python/paddle/distribution/bernoulli.py

@@ -18,8 +18,8 @@ import numpy as np
 import paddle
 from paddle.distribution import exponential_family
 from paddle.fluid.data_feeder import check_type, convert_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layers import tensor
+from paddle.framework import in_dynamic_mode
 from paddle.nn.functional import (
     binary_cross_entropy_with_logits,
     sigmoid,
@@ -93,7 +93,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
 
     def __init__(self, probs, name=None):
         self.name = name or 'Bernoulli'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 probs,
                 'probs',
@@ -110,7 +110,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
             self.dtype = paddle.get_default_dtype()
 
         # Check probs range [0, 1].
-        if _non_static_mode():
+        if in_dynamic_mode():
             """Not use `paddle.any` in static mode, which always be `True`."""
             if (
                 paddle.any(self.probs < 0)
@@ -176,7 +176,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 # [100, 2, 2]
         """
         name = self.name + '_sample'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 shape,
                 'shape',
@@ -255,7 +255,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 #        288.66418457)
         """
         name = self.name + '_rsample'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 shape,
                 'shape',
@@ -317,7 +317,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 #        [1.])
         """
         name = self.name + '_cdf'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(value, 'value', tensor.Variable, name)
 
         value = self._check_values_dtype_in_probs(self.probs, value)
@@ -355,7 +355,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 #        [-1.20397282])
         """
         name = self.name + '_log_prob'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(value, 'value', tensor.Variable, name)
 
         value = self._check_values_dtype_in_probs(self.probs, value)
@@ -394,7 +394,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 #        [0.29999998])
         """
         name = self.name + '_prob'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(value, 'value', tensor.Variable, name)
 
         return self.log_prob(value).exp(name=name)
@@ -459,7 +459,7 @@ class Bernoulli(exponential_family.ExponentialFamily):
                 #        0.33891910)
         """
         name = self.name + '_kl_divergence'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(other, 'other', Bernoulli, name)
 
         a_logits = self.logits

python/paddle/distribution/categorical.py

@@ -17,8 +17,8 @@ import numpy as np
 import paddle
 from paddle.distribution import distribution
 from paddle.fluid.data_feeder import check_type, convert_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layers import tensor
+from paddle.framework import in_dynamic_mode
 from paddle.tensor import multinomial
 
 
@@ -90,7 +90,7 @@ class Categorical(distribution.Distribution):
             logits(list|tuple|numpy.ndarray|Tensor): The logits input of categorical distribution. The data type is float32 or float64.
             name(str, optional): Name for the operation (optional, default is None). For more information, please refer to :ref:`api_guide_Name`.
         """
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 logits,
                 'logits',
@@ -146,7 +146,7 @@ class Categorical(distribution.Distribution):
 
         """
         name = self.name + '_sample'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(shape, 'shape', (list), 'sample')
 
         num_samples = np.prod(np.array(shape))
@@ -208,7 +208,7 @@ class Categorical(distribution.Distribution):
 
         """
         name = self.name + '_kl_divergence'
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(other, 'other', Categorical, 'kl_divergence')
 
         logits = self.logits - paddle.max(self.logits, axis=-1, keepdim=True)

python/paddle/distribution/dirichlet.py

@@ -15,8 +15,8 @@
 import paddle
 from paddle.distribution import exponential_family
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 
 class Dirichlet(exponential_family.ExponentialFamily):
@@ -159,7 +159,7 @@ class Dirichlet(exponential_family.ExponentialFamily):
 
 def _dirichlet(concentration, name=None):
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return paddle._C_ops.dirichlet(concentration)
     else:
         op_type = 'dirichlet'

python/paddle/distribution/distribution.py

@@ -26,8 +26,8 @@ import numpy as np
 import paddle
 from paddle import _C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype, convert_dtype
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.layers import tensor
+from paddle.framework import in_dynamic_mode
 
 
 class Distribution:
@@ -221,7 +221,7 @@ class Distribution:
         Returns:
             value (Tensor): Change value's dtype if value's dtype is different from param.
         """
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             if value.dtype != param.dtype and convert_dtype(value.dtype) in [
                 'float32',
                 'float64',

python/paddle/distribution/exponential_family.py

@@ -14,7 +14,7 @@
 
 import paddle
 from paddle.distribution import distribution
-from paddle.fluid.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 
 
 class ExponentialFamily(distribution.Distribution):
@@ -61,7 +61,7 @@ class ExponentialFamily(distribution.Distribution):
 
         log_norm = self._log_normalizer(*natural_parameters)
 
-        if _non_static_mode():
+        if in_dynamic_mode():
             grads = paddle.grad(
                 log_norm.sum(), natural_parameters, create_graph=True
             )

python/paddle/distribution/kl.py

@@ -27,7 +27,7 @@ from paddle.distribution.laplace import Laplace
 from paddle.distribution.lognormal import LogNormal
 from paddle.distribution.normal import Normal
 from paddle.distribution.uniform import Uniform
-from paddle.fluid.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 
 __all__ = ["register_kl", "kl_divergence"]
 
@@ -229,7 +229,7 @@ def _kl_expfamily_expfamily(p, q):
     p_log_norm = p._log_normalizer(*p_natural_params)
 
     try:
-        if _non_static_mode():
+        if in_dynamic_mode():
             p_grads = paddle.grad(
                 p_log_norm, p_natural_params, create_graph=True
             )

python/paddle/distribution/normal.py

@@ -20,8 +20,8 @@ import numpy as np
 import paddle
 from paddle.distribution import distribution
 from paddle.fluid.data_feeder import check_type, convert_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layers import tensor
+from paddle.framework import in_dynamic_mode
 from paddle.tensor import random
 
 
@@ -87,7 +87,7 @@ class Normal(distribution.Distribution):
     """
 
     def __init__(self, loc, scale, name=None):
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 loc,
                 'loc',
@@ -166,7 +166,7 @@ class Normal(distribution.Distribution):
         if not isinstance(shape, Iterable):
             raise TypeError('sample shape must be Iterable object.')
 
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(seed, 'seed', (int), 'sample')
 
         shape = list(shape)
@@ -321,7 +321,7 @@ class Normal(distribution.Distribution):
             Tensor, kl-divergence between two normal distributions.The data type is float32.
 
         """
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(other, 'other', Normal, 'kl_divergence')
 
         name = self.name + '_kl_divergence'

python/paddle/distribution/uniform.py

@@ -18,8 +18,8 @@ import paddle
 from paddle import _C_ops
 from paddle.distribution import distribution
 from paddle.fluid.data_feeder import check_type, convert_dtype
-from paddle.fluid.framework import _non_static_mode, in_dygraph_mode
 from paddle.fluid.layers import tensor
+from paddle.framework import in_dynamic_mode
 from paddle.tensor import random
 
 
@@ -92,7 +92,7 @@ class Uniform(distribution.Distribution):
     """
 
     def __init__(self, low, high, name=None):
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(
                 low,
                 'low',
@@ -152,7 +152,7 @@ class Uniform(distribution.Distribution):
             Tensor, A tensor with prepended dimensions shape. The data type is float32.
 
         """
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             check_type(shape, 'shape', (list), 'sample')
             check_type(seed, 'seed', (int), 'sample')
 
@@ -205,7 +205,7 @@ class Uniform(distribution.Distribution):
 
         """
         value = self._check_values_dtype_in_probs(self.low, value)
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             # ensure value in [low, high]
             lb_bool = self.low < value
             ub_bool = value < self.high
@@ -234,7 +234,7 @@ class Uniform(distribution.Distribution):
 
         """
         value = self._check_values_dtype_in_probs(self.low, value)
-        if in_dygraph_mode():
+        if in_dynamic_mode():
             lb_bool = self.low < value
             ub_bool = value < self.high
             lb = _C_ops.cast(lb_bool, value.dtype)

python/paddle/fft.py

@@ -20,8 +20,8 @@ import paddle
 
 from . import _C_ops
 from .fluid.data_feeder import check_variable_and_dtype
-from .fluid.framework import in_dygraph_mode
 from .fluid.layer_helper import LayerHelper
+from .framework import in_dynamic_mode
 from .tensor.attribute import is_floating_point, is_integer
 from .tensor.creation import _complex_to_real_dtype, _real_to_complex_dtype
 
@@ -1437,7 +1437,7 @@ def fft_c2c(x, n, axis, norm, forward, name):
         s = [n]
         x = _resize_fft_input(x, s, axes)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out = _C_ops.fft_c2c(x, axes, norm, forward)
     else:
         op_type = 'fft_c2c'
@@ -1468,7 +1468,7 @@ def fft_r2c(x, n, axis, norm, forward, onesided, name):
         _check_fft_n(n)
         s = [n]
         x = _resize_fft_input(x, s, axes)
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out = _C_ops.fft_r2c(x, axes, norm, forward, onesided)
     else:
         op_type = 'fft_r2c'
@@ -1511,7 +1511,7 @@ def fft_c2r(x, n, axis, norm, forward, name):
         s = [n // 2 + 1]
         x = _resize_fft_input(x, s, axes)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         if n is not None:
             out = _C_ops.fft_c2r(x, axes, norm, forward, n)
         else:
@@ -1570,7 +1570,7 @@ def fftn_c2c(x, s, axes, norm, forward, name):
     if s is not None:
         x = _resize_fft_input(x, s, axes)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out = _C_ops.fft_c2c(x, axes, norm, forward)
     else:
         op_type = 'fft_c2c'
@@ -1620,7 +1620,7 @@ def fftn_r2c(x, s, axes, norm, forward, onesided, name):
     if s is not None:
         x = _resize_fft_input(x, s, axes)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out = _C_ops.fft_r2c(x, axes, norm, forward, onesided)
     else:
         op_type = 'fft_r2c'
@@ -1684,7 +1684,7 @@ def fftn_c2r(x, s, axes, norm, forward, name):
         fft_input_shape[-1] = fft_input_shape[-1] // 2 + 1
         x = _resize_fft_input(x, fft_input_shape, axes)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         if s is not None:
             out = _C_ops.fft_c2r(x, axes, norm, forward, s[-1])
         else:

python/paddle/fluid/contrib/optimizer.py

@@ -229,7 +229,7 @@ class Momentum(Optimizer):
             else None
         )
 
-        if framework._non_static_mode():
+        if framework.in_dygraph_mode():
             _, _, _ = _legacy_C_ops.momentum(
                 param_and_grad[0],
                 param_and_grad[1],

python/paddle/fluid/data_feeder.py

@@ -22,8 +22,8 @@ import struct
 from .framework import (
     Variable,
     default_main_program,
+    in_dygraph_mode,
     _current_expected_place,
-    _non_static_mode,
 )
 from .framework import _cpu_num, _cuda_ids
 
@@ -140,7 +140,7 @@ def check_type(input, input_name, expected_type, op_name, extra_message=''):
     # in dynamic graph mode.
     # 2. Performance considerations. Because these checks are executed at
     # each step in dynamic graph mode, it will bring a heavy performance burden.
-    if _non_static_mode():
+    if in_dygraph_mode():
         return
 
     # NOTE: `in_declarative_mode` is used to determined whether this op is called under
@@ -171,7 +171,7 @@ def check_dtype(
     input_dtype, input_name, expected_dtype, op_name, extra_message=''
 ):
     # See NOTE [ Why skip dynamic graph check ]
-    if _non_static_mode():
+    if in_dygraph_mode():
         return
     if convert_dtype(input_dtype) in ['float16']:
         warnings.warn(
@@ -208,7 +208,7 @@ def check_shape(
     expected_tensor_dtype=('int32', 'int64'),
 ):
     # See NOTE [ Why skip dynamic graph check ]
-    if _non_static_mode():
+    if in_dygraph_mode():
         return
     check_type(shape, 'shape', expected_shape_type, op_name)
     if expected_element_type is not None and not isinstance(shape, Variable):

python/paddle/fluid/dygraph/base.py

@@ -106,7 +106,7 @@ def program_desc_tracing_guard(enable):
 @signature_safe_contextmanager
 def param_guard(parameters):
     # Note: parameters is a reference of self._parameters or self._buffers
-    if in_declarative_mode() and not framework.in_dygraph_mode() and parameters:
+    if in_declarative_mode() and not paddle.in_dynamic_mode() and parameters:
         origin_parameters = parameters.copy()
         for name, var_base in parameters.items():
             if isinstance(var_base, list):

python/paddle/fluid/dygraph/tensor_patch_methods.py

@@ -270,7 +270,7 @@ def monkey_patch_tensor():
                 # 4: [5000.]
 
         """
-        if framework._non_static_mode():
+        if framework.in_dygraph_mode():
             if in_profiler_mode():
                 record_event = profiler.RecordEvent(
                     "Gradient Backward", profiler.TracerEventType.Backward
@@ -978,21 +978,20 @@ def monkey_patch_tensor():
         ("values", values),
         ("to_dense", to_dense),
         ("to_sparse_coo", to_sparse_coo),
+        ("_set_grad_ivar", _set_grad_ivar),
+        ("value", value),
+        ("cpu", cpu),
+        ("cuda", cuda),
+        ("pin_memory", pin_memory),
+        ("_slice", _slice),
+        ("_numel", _numel),
+        ("_uva", _uva),
+        ("_clear_data", _clear_data),
+        ("__hash__", __hash__),
+        ("_use_gpudnn", _use_gpudnn),
     ):
         setattr(core.eager.Tensor, method_name, method)
 
-    setattr(core.eager.Tensor, "_set_grad_ivar", _set_grad_ivar)
-    setattr(core.eager.Tensor, "value", value)
-    setattr(core.eager.Tensor, "cpu", cpu)
-    setattr(core.eager.Tensor, "cuda", cuda)
-    setattr(core.eager.Tensor, "pin_memory", pin_memory)
-    setattr(core.eager.Tensor, "_slice", _slice)
-    setattr(core.eager.Tensor, "_numel", _numel)
-    setattr(core.eager.Tensor, "_uva", _uva)
-    setattr(core.eager.Tensor, "_clear_data", _clear_data)
-    setattr(core.eager.Tensor, "__hash__", __hash__)
-    setattr(core.eager.Tensor, "_use_gpudnn", _use_gpudnn)
-
     global _already_patch_repr
     if not _already_patch_repr:
         # NOTE(zhiqiu): pybind11 will set a default __str__ method of enum class.

python/paddle/fluid/framework.py

@@ -52,7 +52,6 @@ __all__ = [
     'cpu_places',
     'xpu_places',
     'cuda_pinned_places',
-    '_non_static_mode',
     'in_dygraph_mode',
     'is_compiled_with_cinn',
     'is_compiled_with_cuda',
@@ -156,28 +155,6 @@ extra_op_attrs = {
     "unique": ["is_sorted"],
 }
 
-# Some explanation of our execution system 2022.03
-# For now we have 3 kinds of execution system, since we refactored dygraph mode to
-# build a fast execution system for dynamic mode. But we can't just remove all legacy
-# code once we present the new system for some historical reason. That's why we have
-# these flags.
-#
-# 1. _non_static_mode():
-# _non_static_mode means  we are now running in legacy dygraph mode or dygraph mode.
-# 2. dygraph_mode():
-# This flags inidicates we are now running in dygraph mode which called eager mode before.
-# 3. _in_legacy_dygraph():
-# This flags has been deprecated
-#
-# They have a relation ship as below:
-# Since _in_legacy_graph is deprecated, so dygraph_mode is _non_static_mode
-#
-# Why we have to make different of _in_legacy_dygraph and dygraph_mode?
-# In some performance issue, we find that python if statement cause server performance problem
-# and we need our new dygraph mode becomes as fast as it could be. That's why we make these flags
-# to make sure in most case, we find new dygraph mode first with only one if statement.
-
-
 # FIXME(dev): We haven't fully verified eager mode on XPU et.al but
 # only GPU/CPU. Remove this after we improve this feature.
 _is_first_import_ = True
@@ -213,10 +190,6 @@ def in_dygraph_mode():
     return global_var._dygraph_tracer_ is not None
 
 
-def _non_static_mode():
-    return global_var._dygraph_tracer_ is not None
-
-
 global_ipu_index = -1
 global_ipu_stage = -1
 ipu_index_attr_name = 'ipu_index'
@@ -459,7 +432,7 @@ def require_version(min_version, max_version=None):
 
 def _dygraph_not_support_(func):
     def __impl__(*args, **kwargs):
-        assert not _non_static_mode(), (
+        assert not in_dygraph_mode(), (
             "We don't support %s in dynamic graph mode" % func.__name__
         )
         return func(*args, **kwargs)
@@ -469,7 +442,7 @@ def _dygraph_not_support_(func):
 
 def _dygraph_only_(func):
     def __impl__(*args, **kwargs):
-        assert _non_static_mode(), (
+        assert in_dygraph_mode(), (
             "We only support '%s()' in dynamic graph mode, please call 'paddle.disable_static()' to enter dynamic graph mode."
             % func.__name__
         )
@@ -482,7 +455,7 @@ def _non_static_only_(func):
     def __impl__(*args, **kwargs):
         from .dygraph.base import in_declarative_mode
 
-        assert _non_static_mode() or in_declarative_mode(), (
+        assert in_dygraph_mode() or in_declarative_mode(), (
             "We only support '%s()' in dynamic graph mode, please call 'paddle.disable_static()' to enter dynamic graph mode."
             % func.__name__
         )
@@ -493,7 +466,7 @@ def _non_static_only_(func):
 
 def _static_only_(func):
     def __impl__(*args, **kwargs):
-        assert not _non_static_mode(), (
+        assert not in_dygraph_mode(), (
             "In PaddlePaddle 2.x, we turn on dynamic graph mode by default, and '%s()' is only supported in static graph mode. So if you want to use this api, please call 'paddle.enable_static()' before this api to enter static graph mode."
             % func.__name__
         )
@@ -971,7 +944,7 @@ def name_scope(prefix=None):
     """
     # TODO(panyx0718): Only [0-9a-z].
     # in dygraph we don't need namescope since it will cause mem leak
-    if _non_static_mode():
+    if in_dygraph_mode():
         yield
     else:
         assert prefix, "namescope prefix can not be empty."
@@ -2738,7 +2711,7 @@ class Operator:
         except ValueError:
             pass
 
-        if _non_static_mode():
+        if in_dygraph_mode():
             if type is None:
                 raise ValueError(
                     "`type` to initialized an Operator can not be None."
@@ -2924,7 +2897,7 @@ class Operator:
                         else:
                             out_arg_names.append(arg.name)
                         # TODO(minqiyang): could we remove variable's op in static graph mode?
-                        if not _non_static_mode():
+                        if not in_dygraph_mode():
                             if isinstance(arg, str):
                                 block.var(arg).op = self
                             else:
@@ -3799,7 +3772,7 @@ class Block:
         )
 
     def create_var(self, *args, **kwargs):
-        if _non_static_mode():
+        if in_dygraph_mode():
             var = _create_tensor(*args, **kwargs)
         else:
             var = Variable(block=self, *args, **kwargs)
@@ -3956,7 +3929,7 @@ class Block:
             Operator: the append Operator.
         """
         op_type = kwargs.get("type", None)
-        if _non_static_mode():
+        if in_dygraph_mode():
             attrs = kwargs.get("attrs", {})
             inplace_map = kwargs.get("inplace_map", None)
             warnings.warn(
@@ -4093,7 +4066,7 @@ class Block:
         return self.ops[start:end]
 
     def _prepend_op(self, *args, **kwargs):
-        if _non_static_mode():
+        if in_dygraph_mode():
             type = kwargs.get("type", None)
             attrs = kwargs.get("attrs", {})
             op = Operator(
@@ -7469,7 +7442,7 @@ def _cuda_graph_guard(cuda_graph_attr=None):
                                    cuda_graph_capture_mode;memory_pool_id;cuda_graph_id
     """
     assert (
-        not _non_static_mode()
+        not in_dygraph_mode()
     ), "cuda_graph_guard only works under static graph mode"
     assert (
         core.is_compiled_with_cuda()

python/paddle/fluid/incubate/checkpoint/auto_checkpoint.py

@@ -24,7 +24,7 @@ from contextlib import contextmanager
 
 from paddle.fluid import unique_name, compiler
 from .checkpoint_saver import SerializableBase, CheckpointSaver, PaddleModel
-from paddle.fluid.framework import _non_static_mode, Program
+from paddle.fluid.framework import in_dygraph_mode, Program
 
 g_train_epoch_range = None
 g_checker = None
@@ -138,7 +138,7 @@ class AutoCheckpointChecker:
         return self._save_checkpoint_inter
 
     def valid(self):
-        if _non_static_mode():
+        if in_dygraph_mode():
             return False
 
         return (

python/paddle/fluid/layer_helper.py

@@ -17,7 +17,7 @@ import paddle
 from .framework import (
     Parameter,
     dtype_is_floating,
-    _non_static_mode,
+    in_dygraph_mode,
     OpProtoHolder,
     _global_flags,
 )
@@ -159,7 +159,7 @@ class LayerHelper(LayerHelperBase):
         if use_mkldnn:
             act['use_mkldnn'] = use_mkldnn
         act_type = act.pop('type')
-        if _non_static_mode():
+        if in_dygraph_mode():
             res = _append_activation_in_dygraph(
                 input_var, act_type, use_cudnn, use_mkldnn
             )

python/paddle/fluid/layer_helper_base.py

@@ -20,7 +20,7 @@ from .framework import (
     Variable,
     default_main_program,
     default_startup_program,
-    _non_static_mode,
+    in_dygraph_mode,
     _current_expected_place,
 )
 from . import unique_name
@@ -409,7 +409,7 @@ class LayerHelperBase:
             param = self._create_weight_normalize(attr, shape, dtype)
             WeightNormParamAttr.params_with_weight_norm.append(param)
             return param
-        if _non_static_mode():
+        if in_dygraph_mode():
             # In dygraph mode, we want the returned parameter to be
             # initialized so that it can be used imperatively.
             # check parameter name
@@ -527,7 +527,7 @@ class LayerHelperBase:
             initializer: initializer to use
         """
         assert isinstance(var, Variable)
-        if _non_static_mode():
+        if in_dygraph_mode():
             initializer(var, self.main_program.global_block())
         else:
             self.startup_program.global_block().create_var(

python/paddle/fluid/lazy_init.py

@@ -37,7 +37,7 @@ class LazyInitHelper:
         if self._state:
             return
         assert (
-            framework._non_static_mode()
+            framework.in_dygraph_mode()
         ), "LazyInit.enable() is only available in dygraph mode."
         self._state = True
 

python/paddle/fluid/reader.py

@@ -26,7 +26,7 @@ from .framework import (
     program_guard,
     default_main_program,
     default_startup_program,
-    _non_static_mode,
+    in_dygraph_mode,
     cpu_places,
     _current_expected_place,
 )
@@ -417,7 +417,7 @@ class DataLoader:
                             epoch_id, batch_id, np.mean(loss.numpy())))
 
         """
-        if _non_static_mode():
+        if in_dygraph_mode():
             return DygraphGeneratorLoader(
                 feed_list,
                 capacity,
@@ -1605,7 +1605,7 @@ class DatasetLoader(DataLoaderBase):
             dataset, paddle.distributed.fleet.dataset.DatasetBase
         ), "dataset must be type of DatasetBase"
         assert (
-            not _non_static_mode()
+            not in_dygraph_mode()
         ), "DatasetLoader is not supported in dygraph mode yet"
         if isinstance(places, (list, tuple)):
             places = _get_paddle_place_list(places)

python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_pp_recompute.py

@@ -87,7 +87,7 @@ class TransformerNet(Layer):
 
 class EmbeddingPipe(EmbeddingNet):
     def forward(self, tensors):
-        if framework.in_dygraph_mode():
+        if framework.in_dynamic_mode():
             stable, x = tensors
             return stable, super().forward(x)
         else:
@@ -96,7 +96,7 @@ class EmbeddingPipe(EmbeddingNet):
 
 class TransformerNetPipe(TransformerNet):
     def forward(self, tensors):
-        if framework.in_dygraph_mode():
+        if framework.in_dynamic_mode():
             stable, x = tensors
             output = super().forward(x)
             return stable, output
@@ -109,7 +109,7 @@ class CriterionPipe(Layer):
         super().__init__()
 
     def forward(self, out, label):
-        if framework.in_dygraph_mode():
+        if framework.in_dynamic_mode():
             out = out[-1]
         loss = out.mean()
         return loss
@@ -179,7 +179,7 @@ class TestDistPPTraning(unittest.TestCase):
             x_data = np.random.randint(0, vocab_size, size=[batch_size, length])
             x = paddle.to_tensor(x_data)
             x.stop_gradient = True
-            input_ = (x, x) if framework.in_dygraph_mode() else x
+            input_ = (x, x) if framework.in_dynamic_mode() else x
             loss = model.train_batch([input_, x], optimizer, scheduler)
             # TODO(shenliang03) add utest for loss
             print("loss: ", loss)

python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py

@@ -55,7 +55,7 @@ def optimizer_setting(params, parameter_list=None):
     bd = [step * e for e in ls["epochs"]]
     lr = params["lr"]
     num_epochs = params["num_epochs"]
-    if fluid._non_static_mode():
+    if fluid.in_dygraph_mode():
         optimizer = fluid.optimizer.Momentum(
             learning_rate=fluid.layers.cosine_decay(
                 learning_rate=lr, step_each_epoch=step, epochs=num_epochs

python/paddle/fluid/tests/unittests/test_context_manager.py

@@ -29,8 +29,8 @@ class TestContextManagerRaiseException(unittest.TestCase):
 
     def test_func2(self):
         # After test_func1 executed, if fluid.dygraph.guard() in test_func1 safely exited,
-        # fluid._non_static_mode() should be false.
-        self.assertEqual(fluid._non_static_mode(), False)
+        # fluid.in_dygraph_mode() should be false.
+        self.assertEqual(fluid.in_dygraph_mode(), False)
 
 
 if __name__ == '__main__':

python/paddle/fluid/tests/unittests/test_dropout_nd_op.py

@@ -21,7 +21,7 @@ import paddle
 from paddle import _legacy_C_ops, fluid
 from paddle.fluid import core
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode
+from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.layer_helper import LayerHelper
 from paddle.static import default_main_program
 
@@ -34,7 +34,7 @@ def dropout_nd(
     mode = (
         'downgrade_in_infer' if mode == 'downscale_in_infer' else mode
     )  # semantic transfer
-    if _non_static_mode():
+    if in_dygraph_mode():
         if default_main_program().random_seed != 0:
             seed = default_main_program().random_seed
 

python/paddle/fluid/tests/unittests/test_imperative_decorator.py

@@ -27,7 +27,7 @@ class TestTracerMode(unittest.TestCase):
         self.init_mode = True
 
     def get_tracer_mode(self):
-        assert fluid._non_static_mode(), "Dygraph mode must be enabled"
+        assert framework.in_dygraph_mode(), "Dygraph mode must be enabled"
 
     @fluid.dygraph.no_grad
     def no_grad_func(self, a):

python/paddle/fluid/tests/unittests/test_imperative_double_grad.py

@@ -26,7 +26,7 @@ from paddle.vision.models import resnet50, resnet101
 
 def _dygraph_guard_(func):
     def __impl__(*args, **kwargs):
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             return func(*args, **kwargs)
         else:
             with fluid.dygraph.guard():

python/paddle/fluid/tests/unittests/test_imperative_ocr_attention_model.py

@@ -206,7 +206,7 @@ class EncoderNet(paddle.nn.Layer):
             initializer=paddle.nn.initializer.Normal(0.0, 0.02),
             learning_rate=2.0,
         )
-        if fluid.framework._non_static_mode():
+        if fluid.framework.in_dygraph_mode():
             h_0 = np.zeros(
                 (Config.batch_size, rnn_hidden_size), dtype="float32"
             )

python/paddle/fluid/tests/unittests/test_imperative_resnet.py

@@ -58,7 +58,7 @@ def optimizer_setting(params, parameter_list=None):
         base_lr = params["lr"]
         lr = []
         lr = [base_lr * (0.1**i) for i in range(len(bd) + 1)]
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             optimizer = fluid.optimizer.SGD(
                 learning_rate=0.01, parameter_list=parameter_list
             )

python/paddle/fluid/tests/unittests/test_imperative_resnet_sorted_gradient.py

@@ -54,7 +54,7 @@ def optimizer_setting(params, parameter_list=None):
         base_lr = params["lr"]
         lr = []
         lr = [base_lr * (0.1**i) for i in range(len(bd) + 1)]
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             optimizer = fluid.optimizer.SGD(
                 learning_rate=0.01, parameter_list=parameter_list
             )

python/paddle/fluid/tests/unittests/test_imperative_se_resnext.py

@@ -54,7 +54,7 @@ def optimizer_setting(params, parameter_list=None):
         # bd = [step * e for e in ls["epochs"]]
         # base_lr = params["lr"]
         # lr = [base_lr * (0.1**i) for i in range(len(bd) + 1)]
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             optimizer = fluid.optimizer.SGD(
                 learning_rate=0.01, parameter_list=parameter_list
             )

python/paddle/fluid/tests/unittests/test_imperative_star_gan_with_gradient_penalty.py

@@ -114,7 +114,7 @@ class InstanceNorm(paddle.nn.Layer):
         self.bias = self.create_parameter(shape=[num_channels], is_bias=True)
 
     def forward(self, input):
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             out, _, _ = _legacy_C_ops.instance_norm(
                 input, self.scale, self.bias, 'epsilon', self.epsilon
             )
@@ -387,7 +387,7 @@ def loss_cls(cls, label, cfg):
 
 
 def calc_gradients(outputs, inputs, no_grad_set):
-    if fluid._non_static_mode():
+    if fluid.in_dygraph_mode():
         return fluid.dygraph.grad(
             outputs=outputs,
             inputs=inputs,
@@ -481,7 +481,7 @@ def build_optimizer(layer, cfg, loss=None):
     learning_rate = 1e-3
     beta1 = 0.5
     beta2 = 0.999
-    if fluid._non_static_mode():
+    if fluid.in_dygraph_mode():
         return fluid.optimizer.Adam(
             learning_rate=learning_rate,
             beta1=beta1,

python/paddle/fluid/tests/unittests/test_imperative_tensor_clear_gradient.py

@@ -24,7 +24,7 @@ from paddle.fluid.wrapped_decorator import wrap_decorator
 
 def _dygraph_guard_(func):
     def __impl__(*args, **kwargs):
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             return func(*args, **kwargs)
         else:
             with fluid.dygraph.guard():

python/paddle/fluid/tests/unittests/test_imperative_triple_grad.py

@@ -24,7 +24,7 @@ from paddle.fluid.wrapped_decorator import wrap_decorator
 
 def _dygraph_guard_(func):
     def __impl__(*args, **kwargs):
-        if fluid._non_static_mode():
+        if fluid.in_dygraph_mode():
             return func(*args, **kwargs)
         else:
             with fluid.dygraph.guard():

python/paddle/fluid/tests/unittests/test_logical_op.py

@@ -18,7 +18,7 @@ import numpy as np
 from eager_op_test import convert_float_to_uint16
 
 import paddle
-from paddle.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 from paddle.static import Executor, Program, program_guard
 
 SUPPORTED_DTYPES = [
@@ -182,11 +182,11 @@ def test_type_error(unit_test, use_gpu, type_str_map):
         if binary_op:
             if type_str_map['x'] != type_str_map['y']:
                 unit_test.assertRaises(error_type, op, x=x, y=y)
-            if not _non_static_mode():
+            if not in_dynamic_mode():
                 error_type = TypeError
                 unit_test.assertRaises(error_type, op, x=x, y=y, out=1)
         else:
-            if not _non_static_mode():
+            if not in_dynamic_mode():
                 error_type = TypeError
                 unit_test.assertRaises(error_type, op, x=x, out=1)
 

python/paddle/fluid/tests/unittests/test_multiclass_nms_op.py

@@ -19,8 +19,8 @@ import numpy as np
 from eager_op_test import OpTest
 
 import paddle
-from paddle import _C_ops, _legacy_C_ops
-from paddle.fluid import _non_static_mode, core, in_dygraph_mode
+from paddle import _C_ops
+from paddle.fluid import core
 from paddle.fluid.layer_helper import LayerHelper
 
 
@@ -42,7 +42,7 @@ def multiclass_nms3(
 
     helper = LayerHelper('multiclass_nms3', **locals())
 
-    if in_dygraph_mode():
+    if paddle.in_dynamic_mode():
         attrs = (
             score_threshold,
             nms_top_k,
@@ -58,30 +58,6 @@ def multiclass_nms3(
         if not return_index:
             index = None
         return output, index, nms_rois_num
-    elif _non_static_mode():
-        attrs = (
-            'background_label',
-            background_label,
-            'score_threshold',
-            score_threshold,
-            'nms_top_k',
-            nms_top_k,
-            'nms_threshold',
-            nms_threshold,
-            'keep_top_k',
-            keep_top_k,
-            'nms_eta',
-            nms_eta,
-            'normalized',
-            normalized,
-        )
-        output, index, nms_rois_num = _legacy_C_ops.multiclass_nms3(
-            bboxes, scores, rois_num, *attrs
-        )
-        if not return_index:
-            index = None
-        return output, index, nms_rois_num
-
     else:
         output = helper.create_variable_for_type_inference(dtype=bboxes.dtype)
         index = helper.create_variable_for_type_inference(dtype='int32')

python/paddle/fluid/tests/unittests/test_squared_l2_norm_op.py

@@ -20,11 +20,11 @@ from numpy import linalg as LA
 
 import paddle
 from paddle import _C_ops, _legacy_C_ops
-from paddle.framework import in_dygraph_mode
+from paddle.framework import in_dynamic_mode
 
 
 def test_squared_l2_norm(x):
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.squared_l2_norm(x)
     else:
         return _legacy_C_ops.squared_l2_norm(x)

python/paddle/fluid/unique_name.py

@@ -120,9 +120,9 @@ def generate(key):
 # NOTE(zhiqiu): use c++ unique_name_generator in dygraph mode,
 # in order to keep name consistency.
 def generate_with_ignorable_key(key):
-    from .framework import _non_static_mode, _dygraph_tracer
+    from .framework import in_dygraph_mode, _dygraph_tracer
 
-    if _non_static_mode():
+    if in_dygraph_mode():
         return _dygraph_tracer()._generate_unique_name()
 
     return generator(key)

python/paddle/fluid/variable_index.py

@@ -436,8 +436,7 @@ def _getitem_impl_(var, item):
                 start = 0 if step > 0 else MAX_INTEGER
             if end is None:
                 if (
-                    paddle.fluid.framework._non_static_mode()
-                    or not is_tensor_array
+                    paddle.in_dynamic_mode() or not is_tensor_array
                 ) and var.shape[dim] != -1:
                     end = var.shape[dim] if step > 0 else -1
                 else:
@@ -550,7 +549,7 @@ def _getitem_impl_(var, item):
     out = var
     if len(axes) > 0:
         op_type = "strided_slice" if use_strided_slice else "slice"
-        if paddle.fluid.framework.in_dygraph_mode() and op_type == "slice":
+        if paddle.in_dynamic_mode() and op_type == "slice":
             if "StartsTensorList" in inputs.keys():
                 st = inputs['StartsTensorList']
             else:
@@ -620,11 +619,11 @@ def _setitem_for_tensor_array(var, item, value):
     If item is case (1), we perform paddle.tensor.array_write,
     in other cases, we raise a NotImplementedError.
     """
-    from ..framework import LayerHelper, core, _non_static_mode
+    from ..framework import LayerHelper, core
     from .framework import Variable
 
     assert (
-        not _non_static_mode()
+        not paddle.in_dynamic_mode()
     ), "setitem for tensor_array must be called in static graph mode."
     if isinstance(item, (Variable, int)):
         from paddle.jit.dy2static.variable_trans_func import (
@@ -808,7 +807,7 @@ def _setitem_impl_(var, item, value):
             )
         )
 
-    if paddle.fluid.framework._non_static_mode():
+    if paddle.in_dynamic_mode():
         var._bump_inplace_version()
 
     cur_block = default_main_program().current_block()

python/paddle/framework/__init__.py

@@ -55,10 +55,7 @@ from ..fluid.framework import set_flags  # noqa: F401
 from ..fluid.framework import Parameter
 from ..fluid.dygraph.base import enable_dygraph as disable_static  # noqa: F401
 from ..fluid.dygraph.base import disable_dygraph as enable_static  # noqa: F401
-from ..fluid.framework import _non_static_mode as in_dynamic_mode  # noqa: F401
-from ..fluid.framework import (  # noqa: F401
-    _non_static_mode,  # temporary used for hackson
-)
+from ..fluid.framework import in_dygraph_mode as in_dynamic_mode  # noqa: F401
 from ..fluid.framework import (
     _current_expected_place,
     _get_paddle_place,
@@ -74,7 +71,6 @@ from ..fluid.framework import _dygraph_tracer  # noqa: F401
 from ..fluid.framework import generate_control_dev_var_name  # noqa: F401
 
 from ..fluid.layer_helper import LayerHelper  # noqa: F401
-from ..fluid.framework import in_dygraph_mode  # noqa: F401
 from ..fluid.framework import _global_flags  # noqa: F401
 from ..fluid.framework import _apply_pass  # noqa: F401
 from ..fluid.framework import switch_main_program

python/paddle/framework/io.py

@@ -34,7 +34,7 @@ from paddle.fluid.framework import (
     _create_tensor,
     _current_expected_place,
     _dygraph_tracer,
-    _non_static_mode,
+    in_dygraph_mode,
 )
 
 from .io_utils import (
@@ -438,7 +438,7 @@ def _to_LodTensor(ndarray):
 def _tuple_to_tensor(obj, return_numpy):
     if return_numpy:
         return obj[1]
-    if _non_static_mode():
+    if in_dygraph_mode():
         t = paddle.to_tensor(obj[1])
         # This function does modify the name of return value.
         # Loading the same variable multiple times may cause the same name.
@@ -451,7 +451,7 @@ def _tuple_to_tensor(obj, return_numpy):
 def _ndarray_to_tensor(obj, return_numpy):
     if return_numpy:
         return obj
-    if _non_static_mode():
+    if in_dygraph_mode():
         return paddle.to_tensor(obj)
     else:
         return _to_LodTensor(obj)
@@ -508,7 +508,7 @@ def _parse_load_result(obj, return_numpy):
         return obj
 
     if _contain_x(obj, is_layer):
-        if not _non_static_mode():
+        if not in_dygraph_mode():
             raise ValueError(
                 "Layer can only be loaded in dynamic graph mode, but now in static graph mode."
             )
@@ -819,7 +819,7 @@ def save(obj, path, protocol=4, **configs):
                 f.write(obj.desc.serialize_to_string())
 
         elif _is_state_dict(obj):
-            if _non_static_mode():
+            if in_dygraph_mode():
                 _legacy_save(obj, path, protocol)
             else:
                 _legacy_static_save(obj, path, protocol)
@@ -1110,7 +1110,7 @@ def load(path, **configs):
                     if config.return_numpy:
                         return np.array(tensor)
                     else:
-                        if _non_static_mode():
+                        if in_dygraph_mode():
                             return _lod_tensor2varbase(tensor)
                         return tensor
                 except:

python/paddle/geometric/math.py

@@ -14,8 +14,8 @@
 
 from paddle import _C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 __all__ = []
 
@@ -50,7 +50,7 @@ def segment_sum(data, segment_ids, name=None):
             #Outputs: [[4., 4., 4.], [4., 5., 6.]]
 
     """
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.segment_pool(data, segment_ids, "SUM")
     else:
         check_variable_and_dtype(
@@ -107,7 +107,7 @@ def segment_mean(data, segment_ids, name=None):
 
     """
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.segment_pool(data, segment_ids, "MEAN")
     else:
 
@@ -164,7 +164,7 @@ def segment_min(data, segment_ids, name=None):
 
     """
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.segment_pool(data, segment_ids, "MIN")
     else:
         check_variable_and_dtype(
@@ -220,7 +220,7 @@ def segment_max(data, segment_ids, name=None):
 
     """
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.segment_pool(data, segment_ids, "MAX")
     else:
         check_variable_and_dtype(

python/paddle/geometric/message_passing/send_recv.py

@@ -20,8 +20,9 @@ from paddle.fluid.data_feeder import (
     check_type,
     check_variable_and_dtype,
 )
-from paddle.fluid.framework import Variable, in_dygraph_mode
+from paddle.fluid.framework import Variable
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 from .utils import (
     convert_out_size_to_list,
@@ -118,7 +119,7 @@ def send_u_recv(
 
     # TODO(daisiming): Should we add judgement for out_size: max(dst_index) + 1.
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out_size = convert_out_size_to_list(out_size)
         return _C_ops.send_u_recv(
             x, src_index, dst_index, reduce_op.upper(), out_size
@@ -295,7 +296,7 @@ def send_ue_recv(
 
     # TODO(daisiming): Should we add judgement for out_size: max(dst_index) + 1.
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         out_size = convert_out_size_to_list(out_size)
         return _C_ops.send_ue_recv(
             x,
@@ -451,7 +452,7 @@ def send_uv(x, y, src_index, dst_index, message_op="add", name=None):
         message_op = 'mul'
         y = 1.0 / (y + 1e-12)
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         return _C_ops.send_uv(x, y, src_index, dst_index, message_op.upper())
     else:
 

python/paddle/geometric/reindex.py

@@ -15,8 +15,9 @@
 import paddle
 from paddle import _C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import Variable, _non_static_mode
+from paddle.fluid.framework import Variable
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 __all__ = []
 
@@ -86,7 +87,7 @@ def reindex_graph(
         True if value_buffer is not None and index_buffer is not None else False
     )
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         reindex_src, reindex_dst, out_nodes = _C_ops.reindex_graph(
             x,
             neighbors,
@@ -205,7 +206,7 @@ def reindex_heter_graph(
         True if value_buffer is not None and index_buffer is not None else False
     )
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         neighbors = paddle.concat(neighbors, axis=0)
         count = paddle.concat(count, axis=0)
         reindex_src, reindex_dst, out_nodes = _C_ops.reindex_graph(

python/paddle/geometric/sampling/neighbors.py

@@ -14,8 +14,8 @@
 
 from paddle import _C_ops, _legacy_C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode, in_dygraph_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 __all__ = []
 
@@ -100,7 +100,7 @@ def sample_neighbors(
 
     use_perm_buffer = True if perm_buffer is not None else False
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         (
             out_neighbors,
             out_count,
@@ -251,7 +251,7 @@ def weighted_sample_neighbors(
                 "`eids` should not be None if `return_eids` is True."
             )
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         (
             out_neighbors,
             out_count,

python/paddle/hapi/model.py

@@ -33,7 +33,8 @@ from paddle.fluid.dygraph.base import to_variable
 from paddle.fluid.executor import global_scope
 from paddle.fluid.framework import Variable
 from paddle.fluid.framework import _current_expected_place as _get_device
-from paddle.fluid.framework import _get_paddle_place, _non_static_mode
+from paddle.fluid.framework import _get_paddle_place
+from paddle.framework import in_dynamic_mode
 from paddle.framework.io_utils import is_belong_to_optimizer
 from paddle.io import DataLoader, Dataset, DistributedBatchSampler
 from paddle.jit.translated_layer import INFER_MODEL_SUFFIX, INFER_PARAMS_SUFFIX
@@ -256,7 +257,7 @@ def prepare_distributed_context(place=None):
             exe = fluid.Executor(place)
             exe.run(communicator_prog)
 
-        if fluid._non_static_mode():
+        if in_dynamic_mode():
             fluid.disable_dygraph()
             _init_context()
             fluid.enable_dygraph(place)
@@ -1170,7 +1171,7 @@ class Model:
         self._test_dataloader = None
         self.stop_training = False
 
-        if not _non_static_mode():
+        if not in_dynamic_mode():
             if not isinstance(inputs, (list, tuple, dict, Input)):
                 raise TypeError(
                     "'inputs' must be list or tuple or dict, and couldn't be None."
@@ -1182,7 +1183,7 @@ class Model:
         self._labels = self._verify_spec(labels)
 
         # init backend
-        if fluid._non_static_mode():
+        if in_dynamic_mode():
             self._adapter = DynamicGraphAdapter(self)
         else:
             self._adapter = StaticGraphAdapter(self)
@@ -1238,7 +1239,7 @@ class Model:
 
         """
         loss = self._adapter.train_batch(inputs, labels, update)
-        if fluid._non_static_mode() and self._input_info is None:
+        if in_dynamic_mode() and self._input_info is None:
             self._update_inputs()
         return loss
 
@@ -1292,7 +1293,7 @@ class Model:
 
         """
         loss = self._adapter.eval_batch(inputs, labels)
-        if fluid._non_static_mode() and self._input_info is None:
+        if in_dynamic_mode() and self._input_info is None:
             self._update_inputs()
         return loss
 
@@ -1341,7 +1342,7 @@ class Model:
 
         """
         loss = self._adapter.predict_batch(inputs)
-        if fluid._non_static_mode() and self._input_info is None:
+        if in_dynamic_mode() and self._input_info is None:
             self._update_inputs()
         return loss
 
@@ -1527,7 +1528,7 @@ class Model:
         )
 
         # TODO: support save/load scaler state in static graph
-        if _non_static_mode():
+        if in_dynamic_mode():
             scaler_state = None
             if hasattr(self, '_scaler') and self._scaler is not None:
                 if os.path.exists(path + '.pdscaler'):
@@ -1644,7 +1645,7 @@ class Model:
                 )
 
             if 'use_fp16_guard' in amp_config_key_set:
-                if _non_static_mode():
+                if in_dynamic_mode():
                     raise ValueError(
                         "'use_fp16_guard' is supported in static graph mode only."
                     )
@@ -1702,7 +1703,7 @@ class Model:
                 paddle.distributed.ParallelEnv().nranks > 1
                 and not _parallel_context_initialized
             ):
-                if fluid._non_static_mode():
+                if in_dynamic_mode():
                     main_prog_seed = fluid.default_main_program().random_seed
                     startup_prog_seed = (
                         fluid.default_startup_program().random_seed
@@ -2228,7 +2229,7 @@ class Model:
             None
         """
 
-        if fluid._non_static_mode():
+        if in_dynamic_mode():
             with fluid.framework._dygraph_guard(None):
                 layer = self.network
                 if self._input_info is None:  # No provided or inferred
@@ -2428,7 +2429,7 @@ class Model:
                 if (
                     shapes is not None
                     and dtypes is not None
-                    and fluid._non_static_mode()
+                    and in_dynamic_mode()
                 ):
                     out_specs = [
                         Input(name=n, dtype=dtypes[i], shape=shapes[i])

python/paddle/incubate/autograd/functional.py

@@ -69,7 +69,7 @@ def vjp(func, xs, v=None):
 
     # ``_seprate`` breaks the dependencies between ``xs`` and other
     # variables. See more ``_seprate`` .
-    if paddle.fluid._non_static_mode() or not utils.prim_enabled():
+    if framework.in_dygraph_mode() or not utils.prim_enabled():
         xs, v = _separate(xs), _separate(v)
     ys = func(*xs) if isinstance(xs, typing.Sequence) else func(xs)
     _check_v_shape(v, ys)
@@ -130,12 +130,12 @@ def jvp(func, xs, v=None):
     _check_inputs(func, xs, v)
     # ``_seprate`` breaks the dependencies between ``xs`` and other
     # variables. See more ``_seprate`` .
-    if paddle.fluid._non_static_mode() or not utils.prim_enabled():
+    if framework.in_dygraph_mode() or not utils.prim_enabled():
         xs, v = _separate(xs), _separate(v)
     ys = func(*xs) if isinstance(xs, typing.Sequence) else func(xs)
     _check_v_shape(v, xs)
 
-    if not paddle.fluid._non_static_mode() and utils.prim_enabled():
+    if not framework.in_dygraph_mode() and utils.prim_enabled():
         return ys, primapi.forward_grad(ys, xs, v)
     else:
         return ys, _double_backward_trick(ys, xs, v)
@@ -352,7 +352,7 @@ class _Jacobian:
     def __init__(self, func, xs):
         # Skip separating in prim mode temporarily, as detach and clone are not
         # primitive operators.
-        if not paddle.fluid._non_static_mode() and utils.prim_enabled():
+        if not framework.in_dygraph_mode() and utils.prim_enabled():
             self._xs = xs
         else:
             self._xs = _separate(xs)
@@ -580,7 +580,7 @@ def _grad(ys, xs, v=None):
             Tensor is the sum of gradients of outputs with respect to the i-th
             inputs.
     """
-    if paddle.fluid._non_static_mode():
+    if framework.in_dygraph_mode():
         # paddle.grad returns a list though the inputs is a signle Tensor. The
         # follow code snippet fixes the problem by return the first element of
         # xs_grad when the xs is a signle Tensor.

python/paddle/incubate/distributed/models/moe/moe_layer.py

@@ -26,7 +26,7 @@ from paddle import nn
 from paddle.autograd import PyLayer
 from paddle.distributed.utils.moe_utils import global_gather, global_scatter
 from paddle.distributed.utils.nccl_utils import check_nccl_version_for_p2p
-from paddle.framework import in_dygraph_mode
+from paddle.framework import in_dynamic_mode
 from paddle.incubate.distributed.fleet import recompute_hybrid
 
 from .gate import BaseGate, GShardGate, NaiveGate, SwitchGate
@@ -63,7 +63,7 @@ def _all_gather(tensor, group=None, use_calc_stream=True):
     if group is not None and not group.is_member():
         return
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         group = (
             paddle.distributed.collective._get_default_group()
             if group is None

python/paddle/incubate/distributed/models/moe/utils.py

@@ -26,14 +26,14 @@ from paddle.distributed.models.moe.utils import (
     _number_count,
     _prune_gate_by_capacity,
 )
-from paddle.framework import in_dygraph_mode
+from paddle.framework import in_dynamic_mode
 
 
 def _alltoall(in_tensor_list, group=None, use_calc_stream=True):
     if group is not None and not group.is_member():
         return
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         group = (
             paddle.distributed.collective._get_default_group()
             if group is None

python/paddle/incubate/layers/nn.py

@@ -1174,7 +1174,7 @@ def bilateral_slice(x, guide, grid, has_offset, name=None):
             output = paddle.incubate.layers.bilateral_slice(x, guide, grid, has_offset=True)
 
     """
-    if paddle.fluid._non_static_mode():
+    if paddle.in_dynamic_mode():
         attrs = ('has_offset', has_offset)
         return _legacy_C_ops.bilateral_slice(x, grid, guide, *attrs)
 
@@ -1252,7 +1252,7 @@ def correlation(
 
     """
 
-    if paddle.fluid._non_static_mode():
+    if paddle.in_dynamic_mode():
         attrs = (
             "pad_size",
             pad_size,
@@ -1501,7 +1501,7 @@ def fused_bn_add_act(
 def pow2_decay_with_linear_warmup(
     warmup_steps, total_steps, base_lr, end_lr, dtype='float32', name=None
 ):
-    if paddle.fluid._non_static_mode():
+    if paddle.in_dynamic_mode():
         raise NotImplementedError(
             "pow2_decay_with_linear_warmup does not support dygraph mode yet."
         )

python/paddle/incubate/nn/functional/fused_dropout_add.py

@@ -16,8 +16,7 @@
 from paddle import _C_ops
 from paddle.common_ops_import import default_main_program
 from paddle.fluid import core
-from paddle.fluid.framework import in_dygraph_mode
-from paddle.framework import LayerHelper
+from paddle.framework import LayerHelper, in_dynamic_mode
 
 
 def fused_dropout_add(
@@ -73,7 +72,7 @@ def fused_dropout_add(
             "mode argument should be 'downscale_in_infer' or 'upscale_in_train'"
         )
     seed = None
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         if default_main_program().random_seed != 0:
             seed = default_main_program().random_seed
         out, seed_offset = _C_ops.fused_dropout_add(

python/paddle/incubate/nn/functional/fused_gate_attention.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 from paddle import _legacy_C_ops
-from paddle.fluid.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 
 
 def fused_gate_attention(
@@ -142,7 +142,7 @@ def fused_gate_attention(
             # [2, 4, 2, 4]
 
     """
-    if _non_static_mode():
+    if in_dynamic_mode():
         _, _, _, _, _, _, _, _, out = _legacy_C_ops.fused_gate_attention(
             query,
             key,

python/paddle/incubate/nn/functional/fused_matmul_bias.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 
 from paddle import _legacy_C_ops
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 from paddle.tensor.linalg import matmul
 
 
@@ -53,7 +53,7 @@ def fused_matmul_bias(
     """
     if bias is None:
         return matmul(x, y, transpose_x, transpose_y, name)
-    if _non_static_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.fused_gemm_epilogue(
             x, y, bias, 'trans_x', transpose_x, 'trans_y', transpose_y
         )

python/paddle/incubate/nn/functional/fused_transformer.py

@@ -15,8 +15,9 @@
 from paddle import _legacy_C_ops
 from paddle.fluid import core
 from paddle.fluid.data_feeder import check_dtype, check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode, default_main_program
+from paddle.fluid.framework import default_main_program
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 __all__ = []
 
@@ -132,7 +133,7 @@ def fused_feedforward(
         'downgrade_in_infer' if mode == 'downscale_in_infer' else mode
     )  # semantic transfer
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         if default_main_program().random_seed != 0:
             seed = default_main_program().random_seed
         out, _, _, _, _, _, _, _, _, _, _ = _legacy_C_ops.fused_feedforward(
@@ -363,7 +364,7 @@ def fused_bias_dropout_residual_layer_norm(
             x.shape[len(x.shape) - 1] == ln_bias.shape[0]
         ), "The dim of ln_bias must equal to the last dim of x."
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         if default_main_program().random_seed != 0:
             seed = default_main_program().random_seed
         (
@@ -620,7 +621,7 @@ def fused_multi_head_attention(
             f"The rank of the x should be 3, but received {x.ndim}."
         )
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         if default_main_program().random_seed != 0:
             seed = default_main_program().random_seed
         # pre_ln_mean, pre_ln_variance, pre_ln_out, qkv_out, qkv_bias_out, transpose_out, qk_out,
@@ -1046,7 +1047,7 @@ def fused_multi_transformer(
         'downgrade_in_infer' if mode == 'downscale_in_infer' else mode
     )  # semantic transfer
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         cache_kv_out, final_out = _legacy_C_ops.fused_multi_transformer(
             x,
             ln_scales,

python/paddle/incubate/nn/layer/fused_dropout_nd.py

@@ -14,7 +14,7 @@
 
 import paddle
 from paddle import _legacy_C_ops
-from paddle.fluid.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 
 
 class FusedDropout(paddle.nn.Layer):
@@ -104,7 +104,7 @@ class FusedDropout(paddle.nn.Layer):
         if self.p == 0:
             return input
 
-        if self.axis is not None and _non_static_mode():
+        if self.axis is not None and in_dynamic_mode():
             seed = None
             if paddle.static.default_main_program().random_seed != 0:
                 seed = paddle.static.default_main_program().random_seed

python/paddle/incubate/nn/layer/fused_transformer.py

@@ -17,7 +17,8 @@ import paddle
 from paddle.fluid import core
 from paddle.fluid.core import VarDesc
 from paddle.fluid.dygraph import no_grad
-from paddle.fluid.framework import _non_static_mode, convert_np_dtype_to_dtype_
+from paddle.fluid.framework import convert_np_dtype_to_dtype_
+from paddle.framework import in_dynamic_mode
 from paddle.incubate.nn import functional as incubate_f
 from paddle.nn import Layer
 from paddle.nn.initializer import Constant
@@ -34,7 +35,7 @@ def _set_var_distributed(var):
 
     var.is_distributed = True
 
-    if not _non_static_mode():
+    if not in_dynamic_mode():
         # NOTE: use current_block and find_var_recursive to support while_loop
         startup_block = paddle.static.default_startup_program().current_block()
         main_block = paddle.static.default_main_program().current_block()

python/paddle/incubate/nn/loss.py

@@ -15,7 +15,7 @@
 from paddle import _legacy_C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
 from paddle.fluid.layer_helper import LayerHelper
-from paddle.framework import _non_static_mode
+from paddle.framework import in_dynamic_mode
 
 
 def identity_loss(x, reduction="none"):
@@ -59,7 +59,7 @@ def identity_loss(x, reduction="none"):
         if reduction is None:
             raise Exception("Unsupported reduction type.")
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         return _legacy_C_ops.identity_loss(x, "reduction", reduction)
 
     check_variable_and_dtype(x, 'x', ['float32', 'float64'], "identity_loss")

python/paddle/incubate/nn/memory_efficient_attention.py

@@ -21,8 +21,8 @@
 
 import paddle
 from paddle import _C_ops
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 from .attn_bias import (
     BlockDiagonalCausalMask,
@@ -99,7 +99,7 @@ def memory_efficient_attention(
     bias = _get_tensor_bias(attn_bias)
     is_test = not training
 
-    if in_dygraph_mode():
+    if in_dynamic_mode():
         output, logsumexp, seed_and_offset = _C_ops.memory_efficient_attention(
             query,
             key,

python/paddle/incubate/operators/graph_khop_sampler.py

@@ -14,8 +14,8 @@
 
 from paddle import _legacy_C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 
 
 def graph_khop_sampler(
@@ -84,7 +84,7 @@ def graph_khop_sampler(
 
     """
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         if return_eids:
             if sorted_eids is None:
                 raise ValueError(

python/paddle/incubate/operators/graph_reindex.py

@@ -14,8 +14,8 @@
 
 from paddle import _C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 from paddle.utils import deprecated
 
 
@@ -116,7 +116,7 @@ def graph_reindex(
                 "be None if `flag_buffer_hashtable` is True."
             )
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         reindex_src, reindex_dst, out_nodes = _C_ops.reindex_graph(
             x,
             neighbors,

python/paddle/incubate/operators/graph_sample_neighbors.py

@@ -14,8 +14,8 @@
 
 from paddle import _legacy_C_ops
 from paddle.fluid.data_feeder import check_variable_and_dtype
-from paddle.fluid.framework import _non_static_mode
 from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dynamic_mode
 from paddle.utils import deprecated
 
 
@@ -109,7 +109,7 @@ def graph_sample_neighbors(
                 "is True."
             )
 
-    if _non_static_mode():
+    if in_dynamic_mode():
         (
             out_neighbors,
             out_count,