apply some suggestions

python/paddle/fluid/dygraph/base.py

@@ -44,7 +44,7 @@
     'to_variable',
 ]
 
-# Flag that indicates whether running code under `@to_static`
+NON_PERSISTABLE_VAR_NAME_SUFFIX = "__non_persistable"
 
 
 def in_declarative_mode():
@@ -147,17 +147,14 @@ def _convert_into_variable(tensor):
             # But if its shape is empty while created from `create_variable()`, we consider this buffer
             # non-persistable. See case of `dropout_state` in lstm api.
             is_persistable = True
-            if "dropout_state" in tensor.name:
+            if tensor.name.endswith(NON_PERSISTABLE_VAR_NAME_SUFFIX):
                 is_persistable = False
 
             new_var = tensor._to_static_var(
                 to_parameter=False, persistable=is_persistable
             )
         # add param into parameter recorder to collect all the params used in this program.
         if new_var.persistable is True:
-            # TODO(@xiongkun): 0d-tensor may be affected at present,
-            # but there is no particularly good method to identify whether 0d-tensor
-            # is used as buffer or "dropout_state" in LSTM buffer variable.
             from paddle.jit.dy2static.program_translator import (
                 ProgramTranslator,
             )

python/paddle/fluid/framework.py

@@ -27,11 +27,9 @@
 import multiprocessing
 import sys
 import logging
-import paddle
 
 from .proto import framework_pb2, data_feed_pb2
 
-
 from . import core
 from . import unique_name
 import paddle.version as fluid_version
@@ -1719,12 +1717,14 @@ def backward(self, retain_graph=False):
                 loss.backward()
 
         """
+        from .backward import append_backward
+
         if retain_graph is True:
             raise AssertionError(
                 "`retain_graph` == True is not supported in @to_static function."
                 "please set retain_graph = False."
             )
-        param_grad_list = paddle.fluid.backward.append_backward(self)
+        param_grad_list = append_backward(self)
         for param, param_grad in param_grad_list:
             # set grad to simulate dygraph loss.backward() in static mode.
             setattr(param, "grad", param_grad)
@@ -7625,12 +7625,11 @@ def _get_var(name, program=None):
 @signature_safe_contextmanager
 def dygraph_guard_if_declarative():
     from .dygraph.base import in_declarative_mode
+    from .dygraph import Tracer
 
     if in_declarative_mode():
         # Under @paddle.jit.to_static decorator, we switch back dygraph mode temporarily.
-        with paddle.fluid.framework._dygraph_guard(
-            tracer=paddle.fluid.dygraph.Tracer()
-        ):
+        with _dygraph_guard(tracer=Tracer()):
             yield
     else:
         yield

python/paddle/fluid/tests/unittests/dygraph_to_static/CMakeLists.txt

@@ -39,9 +39,13 @@ if(WITH_PYTHON)
   set_tests_properties(test_lac PROPERTIES TIMEOUT 120)
 endif()
 
+# Disable on Windows CPU CI for timeout
 if(WIN32 AND NOT WITH_GPU)
-  list(REMOVE_ITEM TEST_OPS test_resnet_amp
-  )# disable on Windows CPU CI for timeout
+  list(REMOVE_ITEM TEST_OPS test_resnet_amp)
+  # TODO(SigureMo): Temporarily disable train step on Windows CPU CI.
+  # We should remove this after fix the performance issue.
+  list(REMOVE_ITEM TEST_OPS test_train_step_resnet18_adam)
+  list(REMOVE_ITEM TEST_OPS test_train_step_resnet18_sgd)
 endif()
 
 foreach(TEST_OP ${TEST_OPS})
@@ -66,8 +70,6 @@ set_tests_properties(test_transformer PROPERTIES TIMEOUT 200)
 set_tests_properties(test_bmn PROPERTIES TIMEOUT 120)
 #set_tests_properties(test_mnist PROPERTIES TIMEOUT 120)
 set_tests_properties(test_build_strategy PROPERTIES TIMEOUT 120)
-set_tests_properties(test_train_step_resnet18_sgd PROPERTIES TIMEOUT 1200)
-set_tests_properties(test_train_step_resnet18_adam PROPERTIES TIMEOUT 1200)
 
 if(NOT WIN32)
   set_tests_properties(test_resnet_v2 PROPERTIES TIMEOUT 120)

python/paddle/nn/layer/rnn.py

@@ -22,6 +22,7 @@
 from paddle import _C_ops, _legacy_C_ops, framework, in_dynamic_mode
 from paddle.common_ops_import import Variable
 from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
+from paddle.fluid.dygraph.base import NON_PERSISTABLE_VAR_NAME_SUFFIX
 from paddle.fluid.framework import (
     _non_static_mode,
     default_startup_program,
@@ -47,7 +48,7 @@ def rnn(
     sequence_length=None,
     time_major=False,
     is_reverse=False,
-    **kwargs
+    **kwargs,
 ):
     r"""
     rnn creates a recurrent neural network specified by RNNCell `cell`,
@@ -109,7 +110,7 @@ def rnn(
             sequence_length,
             time_major,
             is_reverse,
-            **kwargs
+            **kwargs,
         )
     else:
         return _rnn_static_graph(
@@ -119,7 +120,7 @@ def rnn(
             sequence_length,
             time_major,
             is_reverse,
-            **kwargs
+            **kwargs,
         )
 
 
@@ -155,7 +156,7 @@ def _rnn_dynamic_graph(
     sequence_length=None,
     time_major=False,
     is_reverse=False,
-    **kwargs
+    **kwargs,
 ):
     time_step_index = 0 if time_major else 1
     flat_inputs = paddle.utils.flatten(inputs)
@@ -223,7 +224,7 @@ def _rnn_static_graph(
     sequence_length=None,
     time_major=False,
     is_reverse=False,
-    **kwargs
+    **kwargs,
 ):
     check_type(inputs, 'inputs', (Variable, list, tuple), 'rnn')
     if isinstance(inputs, (list, tuple)):
@@ -359,7 +360,7 @@ def birnn(
     initial_states=None,
     sequence_length=None,
     time_major=False,
-    **kwargs
+    **kwargs,
 ):
     r"""
     birnn creates a bidirectional recurrent neural network specified by
@@ -432,7 +433,7 @@ def birnn(
         states_fw,
         sequence_length,
         time_major=time_major,
-        **kwargs
+        **kwargs,
     )
 
     outputs_bw, states_bw = rnn(
@@ -442,7 +443,7 @@ def birnn(
         sequence_length,
         time_major=time_major,
         is_reverse=True,
-        **kwargs
+        **kwargs,
     )
 
     outputs = paddle.utils.map_structure(
@@ -1209,7 +1210,7 @@ def forward(
             sequence_length=sequence_length,
             time_major=self.time_major,
             is_reverse=self.is_reverse,
-            **kwargs
+            **kwargs,
         )
         return final_outputs, final_states
 
@@ -1296,7 +1297,7 @@ def forward(
             initial_states,
             sequence_length,
             self.time_major,
-            **kwargs
+            **kwargs,
         )
         return outputs, final_states
 
@@ -1429,7 +1430,7 @@ def flatten_parameters(self):
             # should dropout state be persistable for static-graph
             self._dropout_state = self.create_variable(
                 dtype=core.VarDesc.VarType.UINT8,
-                name="dropout_state",
+                name=f"dropout_state{NON_PERSISTABLE_VAR_NAME_SUFFIX}",
             )
             if in_dynamic_mode():
                 with paddle.no_grad():