[CodeStyle] Remove some code with paddle.fluid (#2633)

* remove fluid

Author: ZHUI

examples/language_model/gpt-3/dygraph/modeling.py

@@ -22,7 +22,6 @@
 import paddle.tensor as tensor
 from paddle.fluid import layers
 from paddle.nn.layer.transformer import _convert_param_attr_to_list
-from paddle.fluid.initializer import Normal, Constant, NumpyArrayInitializer
 
 from paddlenlp.transformers import PretrainedModel, register_base_model
 

examples/language_model/gpt-3/static/dataset.py

@@ -363,7 +363,7 @@ def data_gen():
                     yield tuple(
                         [np.expand_dims(np.array(x), axis=0) for x in data])
 
-            data_loader = paddle.fluid.io.DataLoader.from_generator(
+            data_loader = paddle.io.DataLoader.from_generator(
                 feed_list=data_holders, capacity=70, iterable=False)
             data_loader.set_sample_generator(data_gen,
                                              batch_size=args.micro_batch_size,

examples/language_model/gpt-3/static/run_pretrain_static.py

@@ -77,7 +77,7 @@ def dist_optimizer(args, topo):
         args.global_batch_size, micro_batch_size)
     acc_steps = bsz_per_dp // micro_batch_size
 
-    exec_strategy = paddle.fluid.ExecutionStrategy()
+    exec_strategy = paddle.static.ExecutionStrategy()
     exec_strategy.num_threads = 2
     exec_strategy.num_iteration_per_drop_scope = 1
 
@@ -320,8 +320,7 @@ def do_train(args):
 
             clip = None
             if args.grad_clip > 0:
-                clip = paddle.fluid.clip.GradientClipByGlobalNorm(
-                    clip_norm=args.grad_clip)
+                clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=args.grad_clip)
 
             decay_param = [
                 p.name for n, p in model.named_parameters()

examples/language_model/moe/dygraph/dataset.py

@@ -106,7 +106,7 @@ def construct_samples_and_shuffle_data(name, data_prefix, documents, sizes,
     np_rng.set_state(savedState)
 
     if paddle.distributed.get_world_size() > 1:
-        if paddle.fluid.framework.in_dygraph_mode():
+        if paddle.in_dynamic_mode():
             paddle.distributed.barrier()
 
     # Load mappings.

examples/language_model/moe/dygraph/framework/adamw.py

@@ -14,9 +14,9 @@
 
 from paddle.optimizer import Optimizer
 from paddle.optimizer.adam import Adam
-from paddle.fluid import core
+from paddle.framework import core
 from paddle.fluid import framework
-from paddle.fluid.framework import Variable
+from paddle.static import Variable
 from paddle.fluid.dygraph import base as imperative_base
 from collections.abc import Callable
 import paddle

examples/language_model/moe/dygraph/framework/storage_process.py

@@ -12,11 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle.fluid import core
+from paddle.framework import core
 import numpy as np
 from collections import OrderedDict
 
-from paddle.fluid.framework import in_dygraph_mode, _in_legacy_dygraph
+from paddle.framework import in_dygraph_mode, _in_legacy_dygraph
 
 if in_dygraph_mode():
     from paddle.distributed.fleet.meta_parallel.sharding.group_sharded_storage import ParamStorage, GradStorage

examples/language_model/moe/dygraph/modeling.py

@@ -21,9 +21,7 @@
 import paddle.nn.functional as F
 import paddle.tensor as tensor
 from paddle.fluid import layers
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.nn.layer.transformer import _convert_param_attr_to_list
-from paddle.fluid.initializer import Normal, Constant, NumpyArrayInitializer
 
 from paddlenlp.transformers import PretrainedModel, register_base_model
 

examples/language_model/moe/dygraph/run_moe_pretrain.py

@@ -35,12 +35,10 @@
 from utils import get_timers, set_timers
 from types import MethodType
 from paddle import _C_ops
-from paddle.fluid import core
-from paddle.fluid.dygraph import to_variable
+from paddle.framework import core
 import paddle.distributed as dist
 from framework import assign_group_by_size, flatten_dense_tensors, obtain_storage, AdamW, group_sharded_parallel
 from paddle.incubate.distributed.models import moe
-from paddle.fluid.framework import in_dygraph_mode
 from paddle.distributed.fleet.meta_parallel.sharding.sharding_utils import ShardingScaler
 from paddle.distributed.fleet.meta_parallel.sharding.group_sharded_utils import GroupShardedScaler
 
@@ -179,8 +177,8 @@ def unscale_method(self, optimizer):
             if (param._grad_ivar() is not None) and (
                 param._grad_ivar().dtype == core.VarDesc.VarType.FP32)
         ]
-    temp_found_inf_fp16 = to_variable(np.array([0]).astype(np.bool))
-    temp_found_inf_fp32 = to_variable(np.array([0]).astype(np.bool))
+    temp_found_inf_fp16 = paddle.to_tensor(np.array([0]).astype(np.bool))
+    temp_found_inf_fp32 = paddle.to_tensor(np.array([0]).astype(np.bool))
 
     if len(param_grads_fp16):
         _C_ops.check_finite_and_unscale(param_grads_fp16, self._scale,
@@ -443,7 +441,7 @@ def do_train(args):
             scaler = fleet.distributed_scaler(scaler)
             scaler._unscale = MethodType(unscale_method, scaler)
         else:
-            wrap_scale_func = GroupShardedScaler if in_dygraph_mode(
+            wrap_scale_func = GroupShardedScaler if paddle.in_dynamic_mode(
             ) else ShardingScaler
             scaler = wrap_scale_func(scaler)
 

examples/language_model/moe/dygraph/userdefine_optimizer.py

@@ -16,11 +16,10 @@
 import sys
 import paddle
 from paddle.optimizer import Optimizer
-from paddle.fluid.clip import ClipGradByGlobalNorm
-from paddle.fluid.dygraph import base as imperative_base
-from paddle.fluid import framework
-from paddle.fluid.framework import Variable
-from paddle.fluid import core
+from paddle.nn import ClipGradByGlobalNorm
+from paddle import framework
+from paddle.static import Variable
+from paddle.framework import core
 from paddle.fluid import layers
 from paddle.distributed.fleet.utils.hybrid_parallel_util import fused_allreduce_gradients
 
@@ -46,7 +45,7 @@ def __init__(self, clip, hcg):
         self._clip = clip
         self._hcg = hcg
 
-    @imperative_base.no_grad
+    @paddle.no_grad
     def _dygraph_clip(self, params_grads):
         params_and_grads = []
         sum_square_list_dist = []
@@ -141,7 +140,7 @@ def __init__(self, optimizer, hcg):
             self._inner_opt._grad_clip = HybridParallelClipGrad(
                 self._inner_opt._grad_clip, hcg)
 
-    @imperative_base.no_grad
+    @paddle.no_grad
     @framework.dygraph_only
     def step(self):
         parameters_list = _obtain_optimizer_parameters_list(self._inner_opt)

model_zoo/ernie-1.0/run_pretrain_static.py

@@ -182,7 +182,7 @@ def dist_optimizer(args, topo):
         args.global_batch_size, micro_batch_size)
     accumulate_steps = bsz_per_dp // micro_batch_size
 
-    exec_strategy = paddle.fluid.ExecutionStrategy()
+    exec_strategy = paddle.static.ExecutionStrategy()
     exec_strategy.num_threads = 1
     exec_strategy.num_iteration_per_drop_scope = 10000
 
@@ -466,8 +466,7 @@ def do_train(args):
 
         clip = None
         if args.grad_clip > 0:
-            clip = paddle.fluid.clip.GradientClipByGlobalNorm(
-                clip_norm=args.grad_clip)
+            clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=args.grad_clip)
 
         decay_param = [
             p.name for n, p in model.named_parameters()