[rm fluid] for the non distribution (#50313)

python/paddle/fluid/incubate/fleet/parameter_server/distribute_transpiler/__init__.py

@@ -12,20 +12,22 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-Convert the fluid program to distributed data-parallelism programs.
+Convert the static program to distributed data-parallelism programs.
 """
 
 import os
 import sys
 import warnings
 
-from paddle import fluid
-from paddle.fluid import core
-from paddle.fluid.framework import default_main_program
-from paddle.fluid.framework import default_startup_program
-from paddle.fluid.framework import Program
+import paddle
+from paddle.framework import core
+from paddle.static import (
+    default_main_program,
+    default_startup_program,
+    Program,
+    Executor,
+)
 from paddle.fluid.compiler import CompiledProgram
-from paddle.fluid.executor import Executor
 from paddle.fluid.parallel_executor import ParallelExecutor
 from paddle.fluid.optimizer import Optimizer
 
@@ -274,7 +276,7 @@ class FleetTranspiler(Fleet):
                 )
             )
 
-            fluid.io.load_vars(
+            paddle.static.load_vars(
                 self._executor,
                 main_program=self.main_program,
                 dirname=model_dir,
@@ -430,8 +432,8 @@ class FleetTranspiler(Fleet):
             )
 
         # Todo(MrChengmo): support recv&save GPU-Kernel for ps-gpu model save
-        if not isinstance(executor.place, fluid.CPUPlace):
-            save_executor = Executor(fluid.CPUPlace())
+        if not isinstance(executor.place, paddle.CPUPlace):
+            save_executor = Executor(paddle.CPUPlace())
         else:
             save_executor = executor
 
@@ -440,7 +442,7 @@ class FleetTranspiler(Fleet):
                 raise TypeError(
                     "in fleet.save_inference_model() function, main_program must be as Program type, CompiledProgram is not allowed"
                 )
-            fluid.io.save_inference_model(
+            paddle.static.save_inference_model(
                 dirname,
                 feeded_var_names,
                 target_vars,
@@ -451,7 +453,7 @@ class FleetTranspiler(Fleet):
                 export_for_deployment,
             )
         else:
-            fluid.io.save_inference_model(
+            paddle.static.save_inference_model(
                 dirname,
                 feeded_var_names,
                 target_vars,
@@ -733,7 +735,7 @@ class FleetTranspiler(Fleet):
             )
         )
 
-        fluid.io.save_vars(
+        paddle.static.save_vars(
             executor,
             main_program=main_program,
             dirname=dirname,
@@ -766,8 +768,8 @@ class FleetTranspiler(Fleet):
                 "in fleet.save_persistables() function, executor must be as Executor type"
             )
         # Todo(MrChengmo): support recv&save GPU-Kernel for ps-gpu model save
-        if not isinstance(executor.place, fluid.CPUPlace):
-            save_executor = Executor(fluid.CPUPlace())
+        if not isinstance(executor.place, paddle.CPUPlace):
+            save_executor = Executor(paddle.CPUPlace())
         else:
             save_executor = executor
 
@@ -894,8 +896,8 @@ class ParameterServerOptimizer(DistributedOptimizer):
         return _main, _startup
 
     def _build_pserver_programs(self, compiled_config):
-        _main = fluid.Program()
-        _startup = fluid.Program()
+        _main = paddle.static.Program()
+        _startup = paddle.static.Program()
 
         if not compiled_config.is_geo_mode():
             _main = server.add_listen_and_serv_pass(_main, compiled_config)

python/paddle/fluid/incubate/fleet/parameter_server/ir/heter_trainer_pass.py

@@ -14,10 +14,9 @@
 
 import warnings
 
-import paddle.fluid.core as core
-import paddle.fluid.framework as framework
+import paddle.framework.core as core
+import paddle
 
-from paddle.fluid.transpiler.details.program_utils import delete_ops
 from paddle.fluid.incubate.fleet.parameter_server.ir.trainer_pass import (
     find_heter_ops,
 )
@@ -60,7 +59,7 @@ def split_heter_worker_ops_pass(program, config, stage_id, device):
 
     program_block_ops = union_forward_gradient_op(program_block_ops)
     block_vars_detail = find_block_joints(program, program_block_ops, heter_ops)
-    heter_program = framework.Program()
+    heter_program = paddle.static.Program()
     create_heter_program(
         program,
         config,

python/paddle/fluid/incubate/fleet/parameter_server/ir/pserver_pass.py

@@ -14,8 +14,7 @@
 
 import collections
 
-from paddle.fluid import core
-from paddle.fluid.framework import Block
+from paddle.framework import core, Block
 
 from paddle.fluid.incubate.fleet.parameter_server.ir.public import (
     _get_optimize_ops,

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

@@ -14,15 +14,13 @@
 
 from functools import reduce
 
+import paddle
 import collections
 import math
 import os
 import warnings
 import logging
-import paddle.fluid as fluid
-from paddle.fluid import core
-from paddle.fluid.core import CommContext
-import paddle.fluid.framework as framework
+from paddle.framework import core
 from paddle.fluid.incubate.fleet.parameter_server.mode import DistributedMode
 from paddle.fluid.incubate.fleet.parameter_server.ir import vars_metatools
 from paddle.fluid.incubate.fleet.parameter_server.ir.ps_dispatcher import (
@@ -415,7 +413,7 @@ class CompileTimeStrategy:
 
         trainer_id = self.get_role_id()
         aggregate = True
-        ctx = CommContext(
+        ctx = core.CommContext(
             name,
             names,
             eps,
@@ -493,7 +491,7 @@ class CompileTimeStrategy:
                     is_distributed,
                 )
 
-                ctx = CommContext(
+                ctx = core.CommContext(
                     param_ctx.var_name(),
                     param_ctx.split_varnames(),
                     param_ctx.split_endpoints(),
@@ -659,7 +657,7 @@ class CompileTimeStrategy:
                 ]
                 var_numel = reduce(lambda x, y: x * y, var.shape[1:])
 
-                sparse_ctx = CommContext(
+                sparse_ctx = core.CommContext(
                     grad_name,
                     [grad_name],
                     ["127.0.0.1:6071"],
@@ -714,7 +712,7 @@ class CompileTimeStrategy:
             grad_name = "Dense@Grad"
             trainer_id = self.get_role_id()
             aggregate = True
-            dense_ctx = CommContext(
+            dense_ctx = core.CommContext(
                 grad_name,
                 [grad_name],
                 ["127.0.0.1:6071"],
@@ -742,7 +740,7 @@ class CompileTimeStrategy:
                 var_numel = reduce(lambda x, y: x * y, var.shape)
                 grad_name = origin_varname
                 aggregate = True
-                dense_ctx = CommContext(
+                dense_ctx = core.CommContext(
                     grad_name,
                     [grad_name],
                     ["127.0.0.1:6071"],
@@ -809,7 +807,7 @@ class CompileTimeStrategy:
             shape = list(var.shape)
             shape[0] = 0 if is_distributed else shape[0]
 
-            sparse_ctx = CommContext(
+            sparse_ctx = core.CommContext(
                 grad_name,
                 splited_varname,
                 ep_list,
@@ -901,7 +899,7 @@ class CompileTimeStrategy:
         endpoints = self.get_ps_endpoints()
         sections = [1] * len(endpoints)
         names = [name] * len(endpoints)
-        ctx = CommContext(
+        ctx = core.CommContext(
             name,
             names,
             endpoints,
@@ -1417,7 +1415,7 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
         NaturalExpDecay,
         InverseTimeDecay,
     )
-    from paddle.fluid.layers.learning_rate_scheduler import (
+    from paddle.static.learning_rate_scheduler import (
         exponential_decay,
         noam_decay,
         piecewise_decay,
@@ -1425,12 +1423,14 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
         inverse_time_decay,
     )
 
-    decay_main_program = fluid.framework.Program()
-    decay_startup_program = fluid.framework.Program()
+    decay_main_program = paddle.static.Program()
+    decay_startup_program = paddle.static.Program()
     lr_name = ""
 
     if isinstance(lr_sheduler, ExponentialDecay):
-        with fluid.program_guard(decay_main_program, decay_startup_program):
+        with paddle.static.program_guard(
+            decay_main_program, decay_startup_program
+        ):
             lr = exponential_decay(1.0, lr_decay_steps, lr_sheduler.gamma, True)
             lr_name = lr.name
             logging.warn(
@@ -1441,7 +1441,9 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
                 % lr_decay_steps
             )
     elif isinstance(lr_sheduler, NoamDecay):
-        with fluid.program_guard(decay_main_program, decay_startup_program):
+        with paddle.static.program_guard(
+            decay_main_program, decay_startup_program
+        ):
             lr = noam_decay(lr_sheduler.d_model, lr_sheduler.warmup_steps, 1.0)
             lr_name = lr.name
             logging.warn(
@@ -1449,7 +1451,9 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
                 % lr_sheduler.warmup_steps
             )
     elif isinstance(lr_sheduler, NaturalExpDecay):
-        with fluid.program_guard(decay_main_program, decay_startup_program):
+        with paddle.static.program_guard(
+            decay_main_program, decay_startup_program
+        ):
             lr = natural_exp_decay(1.0, lr_decay_steps, lr_sheduler.gamma, True)
             lr_name = lr.name
             logging.warn(
@@ -1460,7 +1464,9 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
                 % lr_decay_steps
             )
     elif isinstance(lr_sheduler, InverseTimeDecay):
-        with fluid.program_guard(decay_main_program, decay_startup_program):
+        with paddle.static.program_guard(
+            decay_main_program, decay_startup_program
+        ):
             lr = inverse_time_decay(
                 1.0, lr_decay_steps, lr_sheduler.gamma, True
             )

python/paddle/fluid/incubate/fleet/parameter_server/ir/trainer_pass.py

@@ -19,9 +19,9 @@ import warnings
 import math
 
 from functools import reduce
-import paddle.fluid as fluid
-import paddle.fluid.core as core
-import paddle.fluid.framework as framework
+import paddle
+from paddle.framework import core
+import paddle.framework as framework
 
 from paddle.fluid.transpiler.details.program_utils import delete_ops
 from paddle.fluid.incubate.fleet.parameter_server.ir.public import (
@@ -962,7 +962,7 @@ def find_heter_ops(program, default_device="cpu"):
     if len(heter_ops) == 0:
         warnings.warn(
             "No heterogeneous OP was found in your program , "
-            " please using fluid.device_guard() to run OPs on different device."
+            " please using paddle.static.device_guard() to run OPs on different device."
         )
 
     total_heter_ops = 0
@@ -1824,7 +1824,7 @@ def screen_persistables(program, var_list):
         else:
             var = program.global_block().vars[var_name]
 
-        if fluid.io.is_persistable(var):
+        if paddle.static.is_persistable(var):
             need_remove.append(var_name)
 
     for var_name in need_remove:

python/paddle/fluid/incubate/fleet/parameter_server/ir/vars_metatools.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 from functools import reduce
 
-from paddle.fluid.framework import Variable
-from paddle.fluid import core
+from paddle.framework.io import Variable
+from paddle.framework import core
 
 dtype_to_size = {
     core.VarDesc.VarType.FP16: 2,

python/paddle/fluid/incubate/fleet/parameter_server/pslib/__init__.py

@@ -16,8 +16,7 @@ import os
 import sys
 from .optimizer_factory import *
 from google.protobuf import text_format
-import paddle.fluid as fluid
-from paddle.fluid.framework import Program
+from paddle.framework import core
 
 from paddle.fluid.incubate.fleet.base.fleet_base import Fleet
 from paddle.fluid.incubate.fleet.base.mode import Mode
@@ -44,10 +43,10 @@ class PSLib(Fleet):
         if role_maker is None:
             role_maker = MPISymetricRoleMaker()
         super().init(role_maker)
-        self._fleet_ptr = fluid.core.Fleet()
+        self._fleet_ptr = core.Fleet()
         self._heter_ptr = None
         if isinstance(role_maker, HeterRoleMaker):
-            self._heter_ptr = fluid.core.Heter()
+            self._heter_ptr = core.Heter()
 
     def _set_client_communication_config(
         self, request_timeout_ms, connect_timeout_ms, max_retry

python/paddle/fluid/incubate/fleet/parameter_server/pslib/optimizer_factory.py

@@ -14,8 +14,8 @@
 """Optimizer Factory."""
 
 __all__ = ["DistributedAdam", "FLEET_GLOBAL_DICT"]
-import paddle.fluid as fluid
-from paddle.fluid import core
+import paddle
+from paddle.framework import core
 from paddle.fluid.distribute_lookup_table import find_distributed_lookup_table
 from paddle.fluid.distribute_lookup_table import (
     find_distributed_lookup_table_inputs,
@@ -504,7 +504,7 @@ class DistributedAdam(DistributedOptimizerImplBase):
             prog_id = str(id(loss.block.program))
             # param_grads of program
             params_grads = sorted(
-                fluid.backward.append_backward(loss, parameters, no_grad_set),
+                paddle.static.append_backward(loss, parameters, no_grad_set),
                 key=lambda x: x[0].name,
             )
 

python/paddle/framework/__init__.py

@@ -61,6 +61,7 @@ from ..fluid.framework import (
     OpProtoHolder,
 )  # noqa: F401
 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

python/paddle/static/__init__.py

@@ -76,6 +76,7 @@ from ..fluid.io import batch  # noqa: F401
 
 from ..fluid.contrib.layers import ctr_metric_bundle  # 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