move some fluid apis (#55986)

* move fluid apis

* fix type error

* remove static exponential_decay

* fix some import error

* remove nn.py

* fix some error

* fix type error

python/paddle/__init__.py

@@ -42,7 +42,6 @@ from .framework import disable_static  # noqa: F401
 from .framework import enable_static  # noqa: F401
 from .framework import in_dynamic_mode  # noqa: F401
 from .fluid.dataset import *  # noqa: F401, F403
-from .fluid.lazy_init import LazyGuard  # noqa: F401
 
 from .framework.dtype import iinfo  # noqa: F401
 from .framework.dtype import finfo  # noqa: F401
@@ -437,6 +436,7 @@ import paddle.text  # noqa: F401
 import paddle.vision  # noqa: F401
 
 from .tensor.random import check_shape  # noqa: F401
+from .nn.initializer.lazy_init import LazyGuard  # noqa: F401
 
 # CINN has to set a flag to include a lib
 if is_compiled_with_cinn():

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

@@ -113,7 +113,7 @@ class LocalSGDOptimizer(MetaOptimizerBase):
 
         p2s = self.create_snapshot_vars(main_block.program)
         with program_guard(main_block.program, startup_program):
-            step = paddle.fluid.layers.autoincreased_step_counter(begin=1)
+            step = paddle.optimizer.lr.autoincreased_step_counter(begin=1)
             k_steps = paddle.static.create_global_var(
                 name="k_steps",
                 shape=[1],
@@ -330,7 +330,7 @@ class AdaptiveLocalSGDOptimizer(MetaOptimizerBase):
 
         p2s = self.create_snapshot_vars(main_block.program)
         with program_guard(main_block.program, startup_program):
-            step = paddle.fluid.layers.autoincreased_step_counter(begin=1)
+            step = paddle.optimizer.lr.autoincreased_step_counter(begin=1)
 
             k_steps = paddle.static.create_global_var(
                 name="k_steps",

python/paddle/distributed/passes/ps_server_pass.py

@@ -15,17 +15,15 @@
 import logging
 
 import paddle
-from paddle.fluid.layers.learning_rate_scheduler import (
-    exponential_decay,
-    inverse_time_decay,
-    noam_decay,
-)
 from paddle.optimizer.lr import (
     ExponentialDecay,
     InverseTimeDecay,
     LRScheduler,
     NaturalExpDecay,
     NoamDecay,
+    exponential_decay,
+    inverse_time_decay,
+    noam_decay,
 )
 
 from ..ps.utils.public import (

python/paddle/fluid/initializer.py

@@ -21,7 +21,6 @@ from .framework import (
     default_main_program,
     _current_expected_place,
 )
-from .lazy_init import lazy_init_helper
 from .framework import program_guard
 import numpy as np
 from .core import VarDesc

python/paddle/fluid/layers/__init__.py

@@ -12,17 +12,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from . import nn
-from .nn import *
 from . import io
 from .io import *
 from . import math_op_patch
 from .math_op_patch import *
-from .learning_rate_scheduler import *
 from ..layer_helper import LayerHelper
 
 
 __all__ = []
-__all__ += nn.__all__
 __all__ += io.__all__
-__all__ += learning_rate_scheduler.__all__

python/paddle/fluid/layers/nn.py

-# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-All layers just related to the neural network.
-"""
-import os
-import inspect
-import warnings
-
-import numpy as np
-
-import paddle
-from ..layer_helper import LayerHelper
-from ..framework import (
-    Variable,
-    OpProtoHolder,
-    dygraph_only,
-    _dygraph_tracer,
-    default_main_program,
-    _create_tensor,
-    static_only,
-    _global_flags,
-    in_dygraph_mode,
-)
-from ..framework import _current_expected_place
-from .. import dygraph_utils
-from ..param_attr import ParamAttr
-from .layer_function_generator import (
-    autodoc,
-    templatedoc,
-    _generate_doc_string_,
-)
-
-from .. import unique_name
-from .. import core
-from ...utils import deprecated
-from ..data_feeder import (
-    convert_dtype,
-    check_variable_and_dtype,
-    check_type,
-    check_dtype,
-)
-from paddle.utils import deprecated
-from paddle import _C_ops, _legacy_C_ops
-from collections.abc import Iterable
-
-
-__all__ = [
-    'autoincreased_step_counter',
-]
-
-
-def autoincreased_step_counter(counter_name=None, begin=1, step=1):
-    """
-    :api_attr: Static Graph
-
-    Create an auto-increase variable. which will be automatically increased
-    by 1 in every iteration. By default, the first return of this counter is 1,
-    and the step size is 1.
-
-    Args:
-        counter_name(str, optional): The counter name. Default '@STEP_COUNTER@'.
-        begin(int, optional): The first return value of this counter. Default 1.
-        step(int, optional): The step size. Default 1.
-
-    Returns:
-        Variable: The auto-increased Variable with data type int64.
-
-    Examples:
-        .. code-block:: python
-
-           import paddle.fluid as fluid
-           import paddle
-           paddle.enable_static()
-           global_step = fluid.layers.autoincreased_step_counter(
-               counter_name='@LR_DECAY_COUNTER@', begin=0, step=1)
-    """
-    helper = LayerHelper('global_step_counter')
-    if counter_name is None:
-        counter_name = '@STEP_COUNTER@'
-    counter, is_new_var = helper.create_or_get_global_variable(
-        name=counter_name,
-        dtype='int64',
-        shape=[1],
-        persistable=True,
-        belong_to_optimizer=True,
-    )
-    if is_new_var:
-        helper.set_variable_initializer(
-            counter,
-            initializer=paddle.nn.initializer.ConstantInitializer(
-                value=begin - 1, force_cpu=True
-            ),
-        )
-        helper.main_program.global_block()._prepend_op(
-            type='increment',
-            inputs={'X': [counter]},
-            outputs={'Out': [counter]},
-            attrs={'step': float(step)},
-        )
-        counter.stop_gradient = True
-
-    return counter

python/paddle/incubate/distributed/fleet/parameter_server/ir/public.py

@@ -1409,8 +1409,6 @@ def _get_lr_scheduler_program(lr_scheduler, lr_param_dict, lr_decay_steps):
         InverseTimeDecay,
         NaturalExpDecay,
         NoamDecay,
-    )
-    from paddle.static.learning_rate_scheduler import (
         exponential_decay,
         inverse_time_decay,
         natural_exp_decay,

python/paddle/nn/initializer/initializer.py

@@ -18,7 +18,7 @@ import math
 import numpy as np
 
 from ...fluid.framework import default_main_program, in_dygraph_mode
-from ...fluid.lazy_init import lazy_init_helper
+from .lazy_init import lazy_init_helper
 
 __all__ = []
 
@@ -42,7 +42,7 @@ class Initializer:
         return self._lazy_init(param, block)
 
     def forward(self, param, block=None):
-        """Add corresponding initialization operations to the network"""
+        """Add corresponding initialization operations to the network."""
         raise NotImplementedError()
 
     def _lazy_init(self, param, block=None):

python/paddle/fluid/lazy_init.py → python/paddle/nn/initializer/lazy_init.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from . import framework
+from ...fluid import framework
 
 __all__ = ["LazyGuard"]
 

python/paddle/static/__init__.py

@@ -72,8 +72,6 @@ from .nn.control_flow import Print  # noqa: F401
 from ..fluid.param_attr import WeightNormParamAttr  # noqa: F401
 from ..fluid.optimizer import Optimizer  # noqa: F401
 
-from ..fluid.layers import exponential_decay  # noqa: F401
-from ..fluid.layers import learning_rate_scheduler  # noqa: F401
 
 from .nn.metric import auc  # noqa: F401
 from .nn.metric import accuracy  # noqa: F401
@@ -135,5 +133,4 @@ __all__ = [  # noqa
     'create_parameter',
     'set_ipu_shard',
     'ctr_metric_bundle',
-    'exponential_decay',
 ]

python/paddle/static/nn/common.py

@@ -24,7 +24,7 @@ from paddle.common_ops_import import (
     check_type,
     check_variable_and_dtype,
 )
-from paddle.fluid import core, layers, unique_name
+from paddle.fluid import core, unique_name
 from paddle.fluid.data_feeder import check_dtype
 from paddle.fluid.framework import (
     Program,
@@ -4210,7 +4210,7 @@ class ExponentialMovingAverage:
         Update Exponential Moving Average. Should only call this method in
         train program.
         """
-        global_step = layers.autoincreased_step_counter(
+        global_step = paddle.optimizer.lr.autoincreased_step_counter(
             counter_name=self._step_counter_name
         )
         param_master_emas = []

test/collective/fleet/parallel_dygraph_se_resnext.py

@@ -67,7 +67,7 @@ def optimizer_setting(params, parameter_list=None):
         )
     else:
         optimizer = paddle.optimizer.Momentum(
-            learning_rate=fluid.layers.cosine_decay(
+            learning_rate=paddle.optimizer.lr.cosine_decay(
                 learning_rate=lr, step_each_epoch=step, epochs=num_epochs
             ),
             momentum=momentum_rate,

test/legacy_test/dist_se_resnext.py

@@ -248,7 +248,7 @@ class DistSeResneXt2x2(TestDistRunnerBase):
         else:
             optimizer = (
                 paddle.distributed.fleet.meta_optimizers.DGCMomentumOptimizer(
-                    learning_rate=fluid.layers.piecewise_decay(
+                    learning_rate=paddle.optimizer.lr.piecewise_decay(
                         boundaries=bd, values=lr
                     ),
                     momentum=0.9,

test/legacy_test/test_dist_transpiler.py

@@ -477,7 +477,7 @@ class TestLRDecayConditional(TranspilerTest):
         cost = paddle.nn.functional.square_error_cost(input=y_predict, label=y)
         avg_cost = paddle.mean(cost)
         sgd_optimizer = paddle.optimizer.SGD(
-            learning_rate=fluid.layers.piecewise_decay(
+            learning_rate=paddle.optimizer.lr.piecewise_decay(
                 [10000, 20000], [1.0, 0.5, 1.0]
             )
         )
@@ -581,7 +581,7 @@ class TestL2DecayWithPiecewise(TranspilerTest):
         bd = [1, 10, 20, 30]
         lr = [base_lr * (0.1**i) for i in range(len(bd) + 1)]
         sgd_optimizer = paddle.optimizer.Momentum(
-            learning_rate=fluid.layers.piecewise_decay(
+            learning_rate=paddle.optimizer.lr.piecewise_decay(
                 boundaries=bd, values=lr
             ),
             momentum=0.9,

test/legacy_test/test_imperative_ocr_attention_model.py

@@ -451,7 +451,7 @@ class TestDygraphOCRAttention(unittest.TestCase):
             ocr_attention = OCRAttention()
 
             if Config.learning_rate_decay == "piecewise_decay":
-                learning_rate = fluid.layers.piecewise_decay(
+                learning_rate = paddle.optimizer.lr.piecewise_decay(
                     [50000], [Config.LR, Config.LR * 0.01]
                 )
             else:
@@ -527,7 +527,7 @@ class TestDygraphOCRAttention(unittest.TestCase):
             ocr_attention = OCRAttention()
 
             if Config.learning_rate_decay == "piecewise_decay":
-                learning_rate = fluid.layers.piecewise_decay(
+                learning_rate = paddle.optimizer.lr.piecewise_decay(
                     [50000], [Config.LR, Config.LR * 0.01]
                 )
             else:

test/legacy_test/test_imperative_resnet.py

@@ -67,7 +67,7 @@ def optimizer_setting(params, parameter_list=None):
         # TODO(minqiyang): Add learning rate scheduler support to dygraph mode
         #  optimizer = fluid.optimizer.Momentum(
         #  learning_rate=params["lr"],
-        #  learning_rate=fluid.layers.piecewise_decay(
+        #  learning_rate=paddle.optimizer.lr.piecewise_decay(
         #  boundaries=bd, values=lr),
         #  momentum=0.9,
         #  regularization=paddle.regularizer.L2Decay(1e-4))

test/legacy_test/test_imperative_resnet_sorted_gradient.py

@@ -63,7 +63,7 @@ def optimizer_setting(params, parameter_list=None):
         # TODO(minqiyang): Add learning rate scheduler support to dygraph mode
         #  optimizer = fluid.optimizer.Momentum(
         #  learning_rate=params["lr"],
-        #  learning_rate=fluid.layers.piecewise_decay(
+        #  learning_rate=paddle.optimizer.lr.piecewise_decay(
         #  boundaries=bd, values=lr),
         #  momentum=0.9,
         #  regularization=paddle.regularizer.L2Decay(1e-4))

test/legacy_test/test_imperative_transformer_sorted_gradient.py

@@ -1137,7 +1137,7 @@ class TestDygraphTransformerSortGradient(unittest.TestCase):
                 is_sparse=is_sparse,
             )
             if sync:
-                lr_decay = fluid.layers.learning_rate_scheduler.noam_decay(
+                lr_decay = paddle.optimizer.lr.noam_decay(
                     ModelHyperParams.d_model, TrainTaskConfig.warmup_steps
                 )
                 with fluid.default_main_program()._lr_schedule_guard():

test/legacy_test/test_learning_rate_scheduler.py

@@ -20,7 +20,7 @@ import numpy as np
 
 import paddle
 from paddle import fluid
-from paddle.fluid import core, framework, layers
+from paddle.fluid import core, framework
 
 
 def exponential_decay(
@@ -239,7 +239,9 @@ class TestLearningRateDecayDygraph(unittest.TestCase):
             d_model = 0.01
             warmup_steps = 200
             learning_rate = 2.0
-            lr = fluid.layers.noam_decay(d_model, warmup_steps, learning_rate)
+            lr = paddle.optimizer.lr.noam_decay(
+                d_model, warmup_steps, learning_rate
+            )
             for step in range(5):
                 step += 1
                 right_result = noam_decay(
@@ -278,7 +280,7 @@ class TestLearningRateDecayDygraph(unittest.TestCase):
                 np.testing.assert_allclose(t, right_result[i], rtol=1e-05)
 
             with self.assertRaises(TypeError):
-                lr = fluid.layers.linear_lr_warmup(
+                lr = paddle.optimizer.lr.linear_lr_warmup(
                     learning_rate="fake_lr",
                     warmup_steps=2,
                     start_lr=0.0,
@@ -443,39 +445,59 @@ class TestLearningRateDecay(unittest.TestCase):
         common_kwargs_false["staircase"] = False
 
         decay_fns = [
-            (exponential_decay, layers.exponential_decay, common_kwargs_true),
-            (exponential_decay, layers.exponential_decay, common_kwargs_false),
-            (natural_exp_decay, layers.natural_exp_decay, common_kwargs_true),
-            (natural_exp_decay, layers.natural_exp_decay, common_kwargs_false),
-            (inverse_time_decay, layers.inverse_time_decay, common_kwargs_true),
+            (
+                exponential_decay,
+                paddle.optimizer.lr.exponential_decay,
+                common_kwargs_true,
+            ),
+            (
+                exponential_decay,
+                paddle.optimizer.lr.exponential_decay,
+                common_kwargs_false,
+            ),
+            (
+                natural_exp_decay,
+                paddle.optimizer.lr.natural_exp_decay,
+                common_kwargs_true,
+            ),
+            (
+                natural_exp_decay,
+                paddle.optimizer.lr.natural_exp_decay,
+                common_kwargs_false,
+            ),
+            (
+                inverse_time_decay,
+                paddle.optimizer.lr.inverse_time_decay,
+                common_kwargs_true,
+            ),
             (
                 inverse_time_decay,
-                layers.inverse_time_decay,
+                paddle.optimizer.lr.inverse_time_decay,
                 common_kwargs_false,
             ),
             (
                 polynomial_decay,
-                layers.polynomial_decay,
+                paddle.optimizer.lr.polynomial_decay,
                 {"learning_rate": 1.0, "decay_steps": 5, "cycle": True},
             ),
             (
                 polynomial_decay,
-                layers.polynomial_decay,
+                paddle.optimizer.lr.polynomial_decay,
                 {"learning_rate": 1.0, "decay_steps": 5, "cycle": False},
             ),
             (
                 piecewise_decay,
-                layers.piecewise_decay,
+                paddle.optimizer.lr.piecewise_decay,
                 {"boundaries": [3, 6, 9], "values": [0.1, 0.2, 0.3, 0.4]},
             ),
             (
                 cosine_decay,
-                layers.cosine_decay,
+                paddle.optimizer.lr.cosine_decay,
                 {"learning_rate": 0.1, "step_each_epoch": 100, "epochs": 120},
             ),
             (
                 noam_decay,
-                layers.noam_decay,
+                paddle.optimizer.lr.noam_decay,
                 {"d_model": 0.01, "warmup_steps": 200, "learning_rate": 2.0},
             ),
         ]
@@ -507,7 +529,7 @@ class TestLinearWamrupLearningRateDecay(unittest.TestCase):
         end_lr = 0.1
 
         with fluid.program_guard(main_prog, startup_prog):
-            decayed_lr = layers.linear_lr_warmup(
+            decayed_lr = paddle.optimizer.lr.linear_lr_warmup(
                 fluid_decay_fn(**kwargs), warmup_steps, start_lr, end_lr
             )
 
@@ -548,7 +570,7 @@ class TestLinearWamrupLearningRateDecayWithScalarInput(unittest.TestCase):
         warmup_steps = 10
 
         with fluid.program_guard(main_prog, startup_prog):
-            decayed_lr = layers.linear_lr_warmup(
+            decayed_lr = paddle.optimizer.lr.linear_lr_warmup(
                 lr, warmup_steps, start_lr, end_lr
             )