remove the AdamOptimizer、SGDOptimizer、MomentumOptimizer、ModelAverage、LookaheadOptimizer、FtrlOptimizer、DecayedAdagradOptimizer、DpsgdOptimizer in fluid and relocate the ExponentialMovingAverage、PipelineOptimizer、GradientMergeOptimizer and change optimizer base for LarsMomentumOptimizer and RecomputeOptimizer

python/paddle/amp/grad_scaler.py

@@ -75,8 +75,8 @@ class AmpScaler:
 
         data = np.random.uniform(-1, 1, [10, 3, 32, 32]).astype('float32')
         model = paddle.nn.Conv2D(3, 2, 3)
-        optimizer = paddle.optimizer.SGDOptimizer(
-                learning_rate=0.01, parameter_list=model.parameters())
+        optimizer = paddle.optimizer.SGD(
+                learning_rate=0.01, parameters=model.parameters())
         scaler = paddle.amp.AmpScaler(init_loss_scaling=1024)
         data = paddle.to_tensor(data)
         with paddle.amp.amp_guard():
@@ -168,8 +168,8 @@ def scale(self, var):
 
                 data = np.random.uniform(-1, 1, [10, 3, 32, 32]).astype('float32')
                 model = paddle.nn.Conv2D(3, 2, 3)
-                optimizer = paddle.optimizer.SGDOptimizer(
-                        learning_rate=0.01, parameter_list=model.parameters())
+                optimizer = paddle.optimizer.SGD(
+                        learning_rate=0.01, parameters=model.parameters())
                 scaler = paddle.amp.AmpScaler(init_loss_scaling=1024)
                 data = paddle.to_tensor(data)
                 with paddle.amp.amp_guard():
@@ -221,8 +221,8 @@ def minimize(self, optimizer, *args, **kwargs):
 
                 data = np.random.uniform(-1, 1, [10, 3, 32, 32]).astype('float32')
                 model = paddle.nn.Conv2D(3, 2, 3)
-                optimizer = paddle.optimizer.SGDOptimizer(
-                        learning_rate=0.01, parameter_list=model.parameters())
+                optimizer = paddle.optimizer.SGD(
+                        learning_rate=0.01, parameters=model.parameters())
                 scaler = paddle.amp.AmpScaler(init_loss_scaling=1024)
                 data = paddle.to_tensor(data)
                 with paddle.amp.amp_guard():

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

@@ -21,9 +21,10 @@
 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.fluid.optimizer import Optimizer
 from paddle.framework import core, in_dynamic_mode
 from paddle.nn.clip import ClipGradByNorm, append_gradient_clip_ops
+from paddle.optimizer import Momentum
 from paddle.regularizer import L1Decay, L2Decay
 from paddle.static import create_global_var
 

python/paddle/distributed/fleet/meta_optimizers/fp16_allreduce_optimizer.py

@@ -152,6 +152,6 @@ def fp16_compression(param_and_grads):
 
     def apply_optimize(self, loss, startup_program, params_grads):
         new_params_grads = self.fp16_compression(params_grads)
-        return self.inner_opt.apply_optimize(
+        return self.inner_opt._apply_optimize(
             loss, startup_program=startup_program, params_grads=new_params_grads
         )

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

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 
-from paddle.fluid.optimizer import GradientMergeOptimizer as GM
+from paddle.incubate.optimizer import GradientMergeOptimizer as GM
 
 from .meta_optimizer_base import MetaOptimizerBase
 

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

@@ -14,7 +14,7 @@
 import logging
 
 import paddle
-from paddle.fluid.optimizer import AdamOptimizer
+from paddle.optimizer import Adam
 
 from .meta_optimizer_base import MetaOptimizerBase
 
@@ -38,7 +38,7 @@ def _set_basic_info(
         )
 
         opt = self.inner_opt
-        if not isinstance(opt, AdamOptimizer):
+        if not isinstance(opt, Adam):
             return
 
         configs = self.user_defined_strategy.lamb_configs
@@ -72,7 +72,7 @@ def _can_apply(self):
             return False
 
         if self.user_defined_strategy.lamb:
-            if not isinstance(self.inner_opt, AdamOptimizer):
+            if not isinstance(self.inner_opt, Adam):
                 logging.warn(
                     "lamb need the inner optimizer to be AdamOptimizer optimizer but got {}.".format(
                         self.inner_opt.type

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

@@ -13,7 +13,8 @@
 
 import logging
 
-from paddle.fluid.optimizer import LarsMomentumOptimizer, Momentum
+from paddle.incubate.optimizer import LarsMomentumOptimizer
+from paddle.optimizer import Momentum
 
 from .meta_optimizer_base import MetaOptimizerBase
 
@@ -98,7 +99,7 @@ def apply_gradients(self, params_grads):
         return self.lars_opt.apply_gradients(params_grads=params_grads)
 
     def apply_optimize(self, loss, startup_program, params_grads):
-        return self.lars_opt.apply_optimize(
+        return self.lars_opt._apply_optimize(
             loss, startup_program=startup_program, params_grads=params_grads
         )
 

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

@@ -49,9 +49,7 @@ def _can_apply(self):
             self.inner_opt,
             (
                 paddle.optimizer.momentum.Momentum,
-                paddle.fluid.optimizer.Momentum,
                 paddle.optimizer.sgd.SGD,
-                paddle.fluid.optimizer.SGD,
             ),
         )
 
@@ -235,9 +233,7 @@ def _can_apply(self):
             self.inner_opt,
             (
                 paddle.optimizer.Momentum,
-                paddle.fluid.optimizer.Momentum,
                 paddle.optimizer.sgd.SGD,
-                paddle.fluid.optimizer.SGD,
             ),
         )
 

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

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle.fluid.optimizer import Optimizer
+from paddle.optimizer import Optimizer
 
 __all__ = []
 
@@ -81,7 +81,7 @@ def backward(
         )
 
     def apply_optimize(self, loss, startup_program, params_grads):
-        return self.inner_opt.apply_optimize(
+        return self.inner_opt._apply_optimize(
             loss, startup_program=startup_program, params_grads=params_grads
         )
 

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

@@ -17,7 +17,6 @@
 import subprocess
 
 import paddle
-from paddle import fluid
 from paddle.framework import core
 
 from ..base.private_helper_function import wait_server_ready
@@ -293,7 +292,7 @@ def get_sys_free_mem():
                     % (platform.system())
                 )
 
-        if not isinstance(self.inner_opt, fluid.optimizer.SGDOptimizer):
+        if not isinstance(self.inner_opt, paddle.optimizer.SGD):
             return False
 
         free = get_sys_free_mem()

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

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 
 import paddle
-from paddle.fluid.optimizer import PipelineOptimizer as PO
+from paddle.incubate.optimizer import PipelineOptimizer as PO
 
 from .common import (
     OP_ROLE_KEY,

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

@@ -204,7 +204,7 @@ def get_sys_free_mem():
                     % (platform.system())
                 )
 
-        if not isinstance(self.inner_opt, paddle.fluid.optimizer.SGDOptimizer):
+        if not isinstance(self.inner_opt, paddle.optimizer.SGD):
             return False
 
         free = get_sys_free_mem()

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

@@ -96,9 +96,9 @@ def minimize_impl(
     ):
         optimize_ops, params_grads = self.inner_opt.minimize(
             loss,
-            startup_program=startup_program,
-            parameter_list=parameter_list,
-            no_grad_set=no_grad_set,
+            startup_program,
+            parameter_list,
+            no_grad_set,
         )
         device = paddle.device.get_device()
         place = paddle.set_device(device)

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

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 
-from paddle.fluid.optimizer import RecomputeOptimizer as RO
+from paddle.incubate.optimizer import RecomputeOptimizer as RO
 
 from .meta_optimizer_base import MetaOptimizerBase
 

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

@@ -15,7 +15,7 @@
 import os
 
 from paddle.fluid import core
-from paddle.fluid.optimizer import PipelineOptimizer
+from paddle.incubate.optimizer import PipelineOptimizer
 from paddle.static import (
     create_global_var,
     default_startup_program,

python/paddle/distributed/passes/auto_parallel_fp16.py

@@ -917,7 +917,7 @@ def _apply_single_impl(self, main_program, startup_program, context):
 
             if self.target_dtype == "fp16":
                 if isinstance(
-                    base_opt, (paddle.static.Adam, paddle.optimizer.AdamW)
+                    base_opt, (paddle.optimizer.Adam, paddle.optimizer.AdamW)
                 ):
                     with main_program._optimized_guard([]):
                         # found_inf = paddle.tensor.creation._memcpy(

python/paddle/distributed/transpiler/distribute_transpiler.py

@@ -295,7 +295,7 @@ class DistributeTranspiler:
             cost =paddle.nn.functional.square_error_cost(input=y_predict, label=y)
             avg_loss = paddle.mean(cost)
 
-            sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
+            sgd_optimizer = paddle.optimizer.SGD(learning_rate=0.001)
             sgd_optimizer.minimize(avg_loss)
 
             # for pserver mode