move fluid.layers.create_global_var to static.create_global_var (#48777)

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

@@ -23,9 +23,9 @@ from paddle import framework
 from paddle.common_ops_import import LayerHelper
 from paddle.fluid.clip import GradientClipByNorm, append_gradient_clip_ops
 from paddle.fluid.dygraph import base as imperative_base
-from paddle.fluid.layers import tensor
 from paddle.fluid.optimizer import Momentum, Optimizer
 from paddle.framework import core
+from paddle.static import create_global_var
 
 
 class DGCMomentumOptimizer(Optimizer):
@@ -217,7 +217,7 @@ class DGCMomentumOptimizer(Optimizer):
         )
 
         # rampup begin step var for all_reduce_op_handle
-        self._rampup_begin_step_var = tensor.create_global_var(
+        self._rampup_begin_step_var = create_global_var(
             shape=[1],
             dtype=core.VarDesc.VarType.FP32,
             persistable=True,
@@ -237,7 +237,7 @@ class DGCMomentumOptimizer(Optimizer):
 
             v_var = self._add_accumulator(self._v_velocity_acc_str, param_var)
 
-            k_var = tensor.create_global_var(
+            k_var = create_global_var(
                 shape=[1],
                 dtype=param_var.dtype,
                 persistable=True,
@@ -246,7 +246,7 @@ class DGCMomentumOptimizer(Optimizer):
                 force_cpu=True,
             )
 
-            encoded_var = tensor.create_global_var(
+            encoded_var = create_global_var(
                 shape=[1],
                 dtype=param_var.dtype,
                 persistable=True,
@@ -255,7 +255,7 @@ class DGCMomentumOptimizer(Optimizer):
                 force_cpu=False,
             )
 
-            gather_var = tensor.create_global_var(
+            gather_var = create_global_var(
                 shape=[1],
                 dtype=param_var.dtype,
                 persistable=True,

python/paddle/distributed/fleet/metrics/metric.py

@@ -40,7 +40,7 @@ def sum(input, scope=None, util=None):
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
           cnt = paddle.sum(input)
-          global_cnt = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
+          global_cnt = paddle.static.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp = fluid.layers.elementwise_add(cnt, global_cnt)
           fluid.layers.assign(tmp, global_cnt)
 
@@ -80,7 +80,7 @@ def max(input, scope=None, util=None):
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
           cnt = paddle.sum(input)
-          global_cnt = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
+          global_cnt = paddle.static.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp = paddle.maximum(cnt, global_cnt)
           fluid.layers.assign(tmp, global_cnt)
 
@@ -120,7 +120,7 @@ def min(input, scope=None, util=None):
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
           cnt = paddle.sum(input)
-          global_cnt = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
+          global_cnt = paddle.static.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp = fluid.layers.elementwise_min(cnt, global_cnt)
           fluid.layers.assign(tmp, global_cnt)
 
@@ -391,15 +391,15 @@ def acc(correct, total, scope=None, util=None):
         .. code-block:: python
 
           # in model.py
-          correct = fluid.layers.create_global_var(dtype='float32', shape=[1], value=0)
-          total = fluid.layers.create_global_var(dtype='float32', shape=[1], value=0)
+          correct = paddle.static.create_global_var(dtype='float32', shape=[1], value=0)
+          total = paddle.static.create_global_var(dtype='float32', shape=[1], value=0)
           acc = fluid.layers.acc(predict, label, k=1, correct=correct, total=total)
 
-          global_correct = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
+          global_correct = paddle.static.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp1 = fluid.layers.elementwise_min(correct, global_correct)
           fluid.layers.assign(tmp1, global_correct)
 
-          global_total = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
+          global_total = paddle.static.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp2 = fluid.layers.elementwise_min(total, global_total)
           fluid.layers.assign(tmp2, global_total)
 

python/paddle/distributed/passes/auto_parallel_gradient_merge.py

@@ -64,7 +64,7 @@ def _remove_and_get_optimizer_op(main_program, dist_context):
 def _get_gm_cond_var(main_program, k_steps, dist_context):
     main_block = main_program.global_block()
     # Add const var
-    k_step_var = layers.create_global_var(
+    k_step_var = paddle.static.create_global_var(
         name="gradient_merge_k",
         shape=[1],
         value=int(k_steps),
@@ -74,7 +74,7 @@ def _get_gm_cond_var(main_program, k_steps, dist_context):
     )
     set_var_dist_attr(dist_context, k_step_var, [-1], world_process_group.ranks)
 
-    zero_var = layers.create_global_var(
+    zero_var = paddle.static.create_global_var(
         name="gradient_merge_zero",
         shape=[1],
         value=int(0),
@@ -85,7 +85,7 @@ def _get_gm_cond_var(main_program, k_steps, dist_context):
     set_var_dist_attr(dist_context, zero_var, [-1], world_process_group.ranks)
 
     # Add step var & cond var
-    step_var = layers.create_global_var(
+    step_var = paddle.static.create_global_var(
         name="gradient_merge_step",
         shape=[1],
         value=int(0),

python/paddle/fluid/contrib/mixed_precision/bf16/decorator.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import paddle
 from paddle.fluid import (
     core,
     default_main_program,
@@ -68,7 +69,7 @@ class OptimizerWithMixedPrecision:
         if isinstance(self._optimizer._learning_rate, float):
             self._optimizer._learning_rate_map[
                 default_main_program()
-            ] = layers.create_global_var(
+            ] = paddle.static.create_global_var(
                 name=unique_name.generate("learning_rate"),
                 shape=[1],
                 value=float(self._optimizer._learning_rate),

python/paddle/fluid/contrib/mixed_precision/decorator.py

@@ -122,7 +122,7 @@ class OptimizerWithMixedPrecision:
         return getattr(self._optimizer, "_supports_check_nan_inf", False)
 
     def _init_amp_var(self):
-        self._loss_scaling = layers.create_global_var(
+        self._loss_scaling = paddle.static.create_global_var(
             name=unique_name.generate("loss_scaling"),
             shape=[1],
             value=self._init_loss_scaling,
@@ -131,14 +131,14 @@ class OptimizerWithMixedPrecision:
         )
 
         if self._use_dynamic_loss_scaling:
-            self._num_good_steps = layers.create_global_var(
+            self._num_good_steps = paddle.static.create_global_var(
                 name=unique_name.generate("num_good_steps"),
                 shape=[1],
                 value=0,
                 dtype='int32',
                 persistable=True,
             )
-            self._num_bad_steps = layers.create_global_var(
+            self._num_bad_steps = paddle.static.create_global_var(
                 name=unique_name.generate("num_bad_steps"),
                 shape=[1],
                 value=0,
@@ -151,7 +151,7 @@ class OptimizerWithMixedPrecision:
         if isinstance(self._optimizer._learning_rate, float):
             self._optimizer._learning_rate_map[
                 default_main_program()
-            ] = layers.create_global_var(
+            ] = paddle.static.create_global_var(
                 name=unique_name.generate("learning_rate"),
                 shape=[1],
                 value=float(self._optimizer._learning_rate),

python/paddle/fluid/contrib/optimizer.py

@@ -143,7 +143,7 @@ class Momentum(Optimizer):
 
         var_name = param.name + "_fp32_master"
         var_name = unique_name.generate(var_name)
-        var = layers.create_global_var(
+        var = paddle.static.create_global_var(
             name=var_name,
             shape=param.shape,
             value=0,

python/paddle/fluid/dygraph/learning_rate_scheduler.py

@@ -68,7 +68,7 @@ class LearningRateDecay:
         """
         from .. import layers
 
-        lr = layers.create_global_var(
+        lr = paddle.static.create_global_var(
             name=unique_name.generate("learning_rate"),
             shape=[1],
             value=float(lr),

python/paddle/fluid/layers/control_flow.py

@@ -2368,9 +2368,10 @@ class Switch:
     Examples:
         .. code-block:: python
 
+            import paddle
             import paddle.fluid as fluid
 
-            lr = fluid.layers.create_global_var(
+            lr = paddle.static.create_global_var(
                 shape=[1],
                 value=0.0,
                 dtype='float32',

python/paddle/fluid/layers/learning_rate_scheduler.py

@@ -420,7 +420,7 @@ def piecewise_decay(boundaries, values):
         else:
             global_step = _decay_step_counter()
 
-            lr = tensor.create_global_var(
+            lr = paddle.static.create_global_var(
                 shape=[1],
                 value=0.0,
                 dtype='float32',
@@ -575,7 +575,7 @@ def linear_lr_warmup(learning_rate, warmup_steps, start_lr, end_lr):
             )
             return lr
         else:
-            lr = tensor.create_global_var(
+            lr = paddle.static.create_global_var(
                 shape=[1],
                 value=0.0,
                 dtype=dtype,

python/paddle/fluid/layers/tensor.py

@@ -12,25 +12,19 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import math
 import numpy
 import warnings
 
 from ..layer_helper import LayerHelper
-from ..param_attr import ParamAttr
-from ..initializer import Initializer
 from ..framework import (
     _current_expected_place,
     convert_np_dtype_to_dtype_,
     _non_static_mode,
     _varbase_creator,
-    device_guard,
     _in_legacy_dygraph,
     in_dygraph_mode,
-    _get_paddle_place,
 )
 from ..framework import Variable
-from ..initializer import Constant
 from ..core import VarDesc
 from .. import core
 from .layer_function_generator import templatedoc
@@ -47,7 +41,6 @@ from .utils import check_shape
 from paddle import _C_ops, _legacy_C_ops
 
 __all__ = [
-    'create_global_var',
     'cast',
     'tensor_array_to_tensor',
     'concat',
@@ -61,86 +54,6 @@ __all__ = [
 ]
 
 
-def create_global_var(
-    shape, value, dtype, persistable=False, force_cpu=False, name=None
-):
-    """
-    This function creates a new tensor variable with value in the global block(block 0).
-
-    Parameters:
-        shape (list[int]|tuple[int]): Shape of the variable
-        value (float): The value of the variable. The new created
-                      variable will be filled with it.
-        dtype (str): Data type of the variable
-        persistable (bool, optional): If this variable is persistable.
-                           Default: False
-        force_cpu (bool, optional): Force this variable to be on CPU.
-                         Default: False
-        name (str, optional): For detailed information, please refer to
-           :ref:`api_guide_Name` . Usually name is no need to set and None by default.
-
-    Returns:
-        Variable: The created Variable
-
-    Examples:
-        .. code-block:: python
-
-            import paddle
-            paddle.enable_static()
-            var = paddle.static.create_global_var(shape=[2,3], value=1.0, dtype='float32',
-                                           persistable=True, force_cpu=True, name='new_var')
-    """
-    check_type(
-        shape, 'shape', (list, tuple, numpy.ndarray), 'create_global_var'
-    )
-    for item in shape:
-        check_type(
-            item,
-            'item of shape',
-            (
-                int,
-                numpy.uint8,
-                numpy.int8,
-                numpy.int16,
-                numpy.int32,
-                numpy.int64,
-            ),
-            'create_global_var',
-        )
-
-    check_dtype(
-        dtype,
-        'dtype',
-        [
-            'bool',
-            'float16',
-            'float32',
-            'float64',
-            'int8',
-            'int16',
-            'int32',
-            'int64',
-            'uint8',
-            'uint16',
-        ],
-        'create_global_var',
-    )
-
-    helper = LayerHelper("global_var", **locals())
-    var = helper.create_global_variable(
-        dtype=dtype,
-        shape=shape,
-        persistable=persistable,
-        name=name,
-        stop_gradient=True,
-    )
-    helper.set_variable_initializer(
-        var, initializer=Constant(value=float(value), force_cpu=force_cpu)
-    )
-
-    return var
-
-
 def cast(x, dtype):
     """
 

python/paddle/fluid/optimizer.py

@@ -418,7 +418,7 @@ class Optimizer:
                 else:
                     self._learning_rate_map[
                         framework.default_main_program()
-                    ] = layers.create_global_var(
+                    ] = paddle.static.create_global_var(
                         name=unique_name.generate("learning_rate"),
                         shape=[1],
                         value=float(self._learning_rate),
@@ -449,7 +449,7 @@ class Optimizer:
             # create learning rate in the current main program
             self._learning_rate_map[
                 framework.default_main_program()
-            ] = layers.create_global_var(
+            ] = paddle.static.create_global_var(
                 name=unique_name.generate("learning_rate"),
                 shape=[1],
                 value=float(self._learning_rate),
@@ -474,6 +474,7 @@ class Optimizer:
         Examples:
             .. code-block:: python
 
+                import paddle
                 import paddle.fluid as fluid
                 import paddle
 
@@ -497,7 +498,7 @@ class Optimizer:
 
 
                     # set learning rate manually by framework Variable
-                    lr_var = fluid.layers.create_global_var(
+                    lr_var = paddle.static.create_global_var(
                         shape=[1], value=0.7, dtype='float32')
                     adam.set_lr(lr_var)
                     lr = adam.current_step_lr()
@@ -1498,7 +1499,7 @@ class SGDOptimizer(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,
@@ -1859,7 +1860,7 @@ class LarsMomentumOptimizer(Optimizer):
 
             var_name = param.name + '_fp32_master'
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,
@@ -2267,21 +2268,21 @@ class AdamOptimizer(Optimizer):
                 def get_decayed_betas(beta1_init, beta2_init, decay_steps, decay_rate, epsilon_init):
                     global_step = lr_scheduler._decay_step_counter()
 
-                    beta1 = fluid.layers.create_global_var(
+                    beta1 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta1_init),
                         dtype='float32',
                         # set persistable for save checkpoints and resume
                         persistable=True,
                         name="beta1")
-                    beta2 = fluid.layers.create_global_var(
+                    beta2 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta2_init),
                         dtype='float32',
                         # set persistable for save checkpoints and resume
                         persistable=True,
                         name="beta2")
-                    epsilon = fluid.layers.create_global_var(
+                    epsilon = paddle.static.create_global_var(
                         shape=[1],
                         value=float(epsilon_init),
                         dtype='float32',
@@ -4326,7 +4327,7 @@ class ExponentialMovingAverage:
 
     def _get_ema_decay(self):
         with default_main_program()._lr_schedule_guard():
-            decay_var = layers.tensor.create_global_var(
+            decay_var = paddle.static.create_global_var(
                 shape=[1],
                 value=self._decay,
                 dtype='float32',
@@ -4346,7 +4347,7 @@ class ExponentialMovingAverage:
         return decay_var
 
     def _get_decay_pow(self, block):
-        global_step = layers.create_global_var(
+        global_step = paddle.static.create_global_var(
             name=self._step_counter_name,
             shape=[1],
             value=0,
@@ -4359,7 +4360,7 @@ class ExponentialMovingAverage:
         return decay_pow_acc, global_step
 
     def _create_ema_vars(self, param):
-        param_ema = layers.create_global_var(
+        param_ema = paddle.static.create_global_var(
             name=unique_name.generate(self._name + param.name + '_ema'),
             shape=param.shape,
             value=0.0,
@@ -7273,7 +7274,7 @@ class LookaheadOptimizer:
 
         with framework.program_guard(main_block.program, startup_program):
             # Add Var k to main prog and startup prog
-            k = layers.create_global_var(
+            k = paddle.static.create_global_var(
                 name="lookahead_k",
                 shape=[1],
                 value=int(self.k),
@@ -7282,7 +7283,7 @@ class LookaheadOptimizer:
             )
 
             # Add Var alpha to main prog and startup prog
-            alpha = layers.create_global_var(
+            alpha = paddle.static.create_global_var(
                 name="lookahead_alpha",
                 shape=[1],
                 value=float(self.alpha),
@@ -7291,7 +7292,7 @@ class LookaheadOptimizer:
             )
 
             # Add Var step
-            step = layers.create_global_var(
+            step = paddle.static.create_global_var(
                 name="lookahead_step",
                 shape=[1],
                 value=int(0),
@@ -7498,7 +7499,7 @@ class GradientMergeOptimizer:
 
     def _get_gm_cond_var(self, main_block):
         # Add const var
-        k_step_var = layers.create_global_var(
+        k_step_var = paddle.static.create_global_var(
             name="gradient_merge_k",
             shape=[1],
             value=int(self.k_steps),
@@ -7507,7 +7508,7 @@ class GradientMergeOptimizer:
             force_cpu=True,
         )
 
-        zero_var = layers.create_global_var(
+        zero_var = paddle.static.create_global_var(
             name="gradient_merge_zero",
             shape=[1],
             value=int(0),
@@ -7517,7 +7518,7 @@ class GradientMergeOptimizer:
         )
 
         # Add step var & cond var
-        step_var = layers.create_global_var(
+        step_var = paddle.static.create_global_var(
             name="gradient_merge_step",
             shape=[1],
             value=int(0),

python/paddle/fluid/tests/test_python_operator_overriding.py

@@ -19,7 +19,6 @@ import numpy as np
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.framework as framework
-import paddle.fluid.layers as layers
 
 paddle.enable_static()
 
@@ -32,10 +31,10 @@ class TestPythonOperatorOverride(unittest.TestCase):
         y_data = np.random.random(size=shape).astype(dtype)
         python_out = fn(x_data, y_data)
 
-        x_var = layers.create_global_var(
+        x_var = paddle.static.create_global_var(
             name='x', shape=shape, value=0.0, dtype=dtype, persistable=True
         )
-        y_var = layers.create_global_var(
+        y_var = paddle.static.create_global_var(
             name='y', shape=shape, value=0.0, dtype=dtype, persistable=True
         )
         out = fn(x_var, y_var)

python/paddle/fluid/tests/unittests/dist_transformer.py

@@ -289,7 +289,7 @@ class LearningRateScheduler:
         self.warmup_steps = warmup_steps
         self.d_model = d_model
         self.static_lr = learning_rate
-        self.learning_rate = layers.create_global_var(
+        self.learning_rate = paddle.static.create_global_var(
             name=name,
             shape=[1],
             value=float(learning_rate),

python/paddle/fluid/tests/unittests/dygraph_to_static/test_bmn.py

@@ -308,7 +308,7 @@ def bmn_loss_func(
             ]
             bm_mask.append(mask_vector)
         bm_mask = np.array(bm_mask, dtype=np.float32)
-        self_bm_mask = fluid.layers.create_global_var(
+        self_bm_mask = paddle.static.create_global_var(
             shape=[dscale, tscale], value=0, dtype=DATATYPE, persistable=True
         )
         fluid.layers.assign(bm_mask, self_bm_mask)

python/paddle/fluid/tests/unittests/npu/test_adam_op_npu.py

@@ -354,21 +354,21 @@ class TestNetWithEpsilonTensor(unittest.TestCase):
                 beta2_init = 0.999
                 epsilon_init = 1e-8
                 if use_tensor:
-                    beta1 = fluid.layers.create_global_var(
+                    beta1 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta1_init),
                         dtype='float32',
                         persistable=True,
                         name="beta1",
                     )
-                    beta2 = fluid.layers.create_global_var(
+                    beta2 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta2_init),
                         dtype='float32',
                         persistable=True,
                         name="beta2",
                     )
-                    epsilon = fluid.layers.create_global_var(
+                    epsilon = paddle.static.create_global_var(
                         shape=[1],
                         value=float(epsilon_init),
                         dtype='float32',

python/paddle/fluid/tests/unittests/test_adam_op.py

@@ -616,10 +616,10 @@ class TestAdamOpV2(unittest.TestCase):
                 conv = fluid.layers.conv2d(data, 8, 3)
                 loss = paddle.mean(conv)
 
-                beta1 = fluid.layers.create_global_var(
+                beta1 = paddle.static.create_global_var(
                     shape=[1], value=0.85, dtype='float32', persistable=True
                 )
-                beta2 = fluid.layers.create_global_var(
+                beta2 = paddle.static.create_global_var(
                     shape=[1], value=0.95, dtype='float32', persistable=True
                 )
                 betas = [beta1, beta2]
@@ -711,7 +711,7 @@ class TestAdamOpV2(unittest.TestCase):
         cur_lr = adam.get_lr()
         assert lr == cur_lr
         with self.assertRaises(TypeError):
-            lr_var = paddle.fluid.layers.create_global_var(
+            lr_var = paddle.static.create_global_var(
                 shape=[1], value=lr, dtype='float32'
             )
             adam.set_lr(lr_var)
@@ -817,21 +817,21 @@ class TestAdamOptimizer(unittest.TestCase):
                 beta2_init = 0.999
                 epsilon_init = 1e-8
                 if use_tensor:
-                    beta1 = fluid.layers.create_global_var(
+                    beta1 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta1_init),
                         dtype='float32',
                         persistable=True,
                         name="beta1",
                     )
-                    beta2 = fluid.layers.create_global_var(
+                    beta2 = paddle.static.create_global_var(
                         shape=[1],
                         value=float(beta2_init),
                         dtype='float32',
                         persistable=True,
                         name="beta2",
                     )
-                    epsilon = fluid.layers.create_global_var(
+                    epsilon = paddle.static.create_global_var(
                         shape=[1],
                         value=float(epsilon_init),
                         dtype='float32',

python/paddle/fluid/tests/unittests/test_adamw_op.py

@@ -212,10 +212,10 @@ class TestAdamWOp(unittest.TestCase):
                 conv = fluid.layers.conv2d(data, 8, 3)
                 loss = paddle.mean(conv)
 
-                beta1 = fluid.layers.create_global_var(
+                beta1 = paddle.static.create_global_var(
                     shape=[1], value=0.85, dtype='float32', persistable=True
                 )
-                beta2 = fluid.layers.create_global_var(
+                beta2 = paddle.static.create_global_var(
                     shape=[1], value=0.95, dtype='float32', persistable=True
                 )
                 betas = [beta1, beta2]

python/paddle/fluid/tests/unittests/test_create_global_var.py

@@ -16,7 +16,7 @@ import unittest
 
 import numpy as np
 
-import paddle.fluid as fluid
+import paddle
 from paddle.fluid import Program, program_guard
 
 
@@ -25,19 +25,19 @@ class TestCreateGlobalVarError(unittest.TestCase):
         with program_guard(Program(), Program()):
 
             def test_shape():
-                fluid.layers.create_global_var(1, 2.0, np.float32)
+                paddle.static.create_global_var(1, 2.0, np.float32)
 
             self.assertRaises(TypeError, test_shape)
 
             def test_shape_item():
-                fluid.layers.create_global_var([1.0, 2.0, 3.0], 2.0, 'float32')
+                paddle.static.create_global_var([1.0, 2.0, 3.0], 2.0, 'float32')
 
             self.assertRaises(TypeError, test_shape_item)
 
             # Since create_global_var support all dtype in convert_dtype().
             # Hence, assertRaises ValueError not TypeError.
             def test_dtype():
-                fluid.layers.create_global_var([1, 2, 3], 2.0, np.complex128)
+                paddle.static.create_global_var([1, 2, 3], 2.0, np.complex128)
 
             self.assertRaises(TypeError, test_dtype)
 

python/paddle/fluid/tests/unittests/test_eager_deletion_padding_rnn.py

@@ -542,7 +542,7 @@ class PaddingRNNTestBase(unittest.TestCase):
                     )
                 )
 
-                self.learning_rate = fluid.layers.create_global_var(
+                self.learning_rate = paddle.static.create_global_var(
                     name="learning_rate",
                     shape=[1],
                     value=1.0,

python/paddle/fluid/tests/unittests/test_fleet_metric.py

@@ -79,14 +79,14 @@ class TestFleetMetric(unittest.TestCase):
         train = fluid.Program()
         startup = fluid.Program()
         with fluid.program_guard(train, startup):
-            t = fluid.layers.create_global_var(
+            t = paddle.static.create_global_var(
                 shape=[1, 1],
                 value=1,
                 dtype='int64',
                 persistable=True,
                 force_cpu=True,
             )
-            t1 = fluid.layers.create_global_var(
+            t1 = paddle.static.create_global_var(
                 shape=[1, 1],
                 value=1,
                 dtype='int64',

python/paddle/fluid/tests/unittests/test_imperative_optimizer.py

@@ -595,7 +595,7 @@ class TestOptimizerLearningRate(unittest.TestCase):
                 lr = adam.current_step_lr()
                 np.testing.assert_allclose(lr, lr_list[i], rtol=1e-06, atol=0.0)
 
-            lr_var = fluid.layers.create_global_var(
+            lr_var = paddle.static.create_global_var(
                 shape=[1], value=0.7, dtype='float32'
             )
             adam.set_lr(lr_var)

python/paddle/fluid/tests/unittests/test_imperative_optimizer_v2.py

@@ -721,7 +721,7 @@ class TestOptimizerLearningRate(unittest.TestCase):
                 np.testing.assert_allclose(lr, lr_list[i], rtol=1e-06, atol=0.0)
 
             with self.assertRaises(TypeError):
-                lr_var = fluid.layers.create_global_var(
+                lr_var = paddle.static.create_global_var(
                     shape=[1], value=0.7, dtype='float32'
                 )
                 adam.set_lr(lr_var)

python/paddle/fluid/tests/unittests/test_lr_scheduler.py

@@ -109,7 +109,7 @@ class TestReduceOnPlateauDecay:
         main_prog = paddle.static.Program()
         start_prog = paddle.static.Program()
         with paddle.static.program_guard(main_prog, start_prog):
-            x = fluid.layers.create_global_var(
+            x = paddle.static.create_global_var(
                 [1], 1, 'float32', persistable=True
             )
             paddle.increment(x)

python/paddle/fluid/tests/unittests/test_parallel_executor_feed_persistable_var.py

@@ -39,7 +39,7 @@ class TestFeedPersistableVar(unittest.TestCase):
         }
 
     def optimizer(self):
-        learning_rate = fluid.layers.create_global_var(
+        learning_rate = paddle.static.create_global_var(
             name="learning_rate",
             shape=[1],
             value=1.0,

python/paddle/fluid/tests/unittests/test_queue.py

@@ -16,9 +16,9 @@ import unittest
 
 import numpy as np
 
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
-import paddle.fluid.layers as layers
 
 
 class TestQueue(unittest.TestCase):
@@ -31,14 +31,14 @@ class TestQueue(unittest.TestCase):
         startup_program = fluid.Program()
         value = np.random.rand(1)
         with fluid.program_guard(main_program, startup_program):
-            data_in = layers.create_global_var(
+            data_in = paddle.static.create_global_var(
                 shape=[2, 3],
                 value=value,
                 dtype="float32",
                 persistable=True,
                 name='var_in',
             )
-            data_out = layers.create_global_var(
+            data_out = paddle.static.create_global_var(
                 shape=[2, 3],
                 value=value - 1.0,
                 dtype="float32",

python/paddle/fluid/tests/unittests/test_switch.py

@@ -30,7 +30,7 @@ class TestSwitch(unittest.TestCase):
         two_var = layers.fill_constant(shape=[1], dtype='float32', value=2.0)
         three_var = layers.fill_constant(shape=[1], dtype='float32', value=3.0)
 
-        result = layers.create_global_var(
+        result = paddle.static.create_global_var(
             shape=[1], value=-1.0, dtype='float32', persistable=True
         )
 
@@ -71,7 +71,7 @@ class TestSwitchCaseError(unittest.TestCase):
                 shape=[1], dtype='float32', value=0.0
             )
 
-            result = layers.create_global_var(
+            result = paddle.static.create_global_var(
                 shape=[1], value=-1.0, dtype='float32', persistable=True
             )
 

python/paddle/fluid/tests/unittests/test_var_info.py

@@ -20,7 +20,7 @@ import unittest
 
 import numpy as np
 
-import paddle.fluid as fluid
+import paddle
 
 
 class TestVarInfo(unittest.TestCase):
@@ -29,7 +29,7 @@ class TestVarInfo(unittest.TestCase):
     def test_var_info(self):
         """Testcase for get and set info for variable."""
         value = np.random.randn(1)
-        var = fluid.layers.create_global_var([1], value, "float32")
+        var = paddle.static.create_global_var([1], value, "float32")
         var._set_info("name", "test")
         ret = var._get_info("name")
         assert ret == "test"

python/paddle/fluid/tests/unittests/xpu/test_adamw_op_xpu.py

@@ -199,13 +199,13 @@ class XPUTestAdamwOp2(XPUOpTestWrapper):
                     conv = fluid.layers.conv2d(data, 8, 3)
                     loss = paddle.mean(conv)
 
-                    beta1 = fluid.layers.create_global_var(
+                    beta1 = paddle.static.create_global_var(
                         shape=[1],
                         value=0.85,
                         dtype=self.in_type_str,
                         persistable=True,
                     )
-                    beta2 = fluid.layers.create_global_var(
+                    beta2 = paddle.static.create_global_var(
                         shape=[1],
                         value=0.95,
                         dtype=self.in_type_str,

python/paddle/incubate/optimizer/distributed_fused_lamb.py

@@ -15,7 +15,7 @@
 import os
 
 import paddle
-from paddle.fluid import core, framework, layers, unique_name
+from paddle.fluid import core, framework, unique_name
 from paddle.fluid.clip import ClipGradByGlobalNorm
 from paddle.fluid.executor import global_scope
 from paddle.fluid.framework import Variable, name_scope
@@ -172,7 +172,7 @@ class DistributedFusedLamb(Optimizer):
 
     def _create_scale_from_constant(self, value):
         name = unique_name.generate('global_scale')
-        return layers.create_global_var(
+        return paddle.static.create_global_var(
             name=name,
             shape=[1],
             dtype='float32',

python/paddle/incubate/optimizer/lookahead.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import paddle
-from paddle.fluid import framework, layers, unique_name
+from paddle.fluid import framework, unique_name
 from paddle.fluid.dygraph import base as imperative_base
 from paddle.fluid.framework import Variable
 from paddle.fluid.layer_helper import LayerHelper
@@ -192,7 +192,7 @@ class LookAhead(Optimizer):
 
     def _increment_global_var(self):
         if self._global_step_var is None:
-            self._global_step_var = layers.create_global_var(
+            self._global_step_var = paddle.static.create_global_var(
                 name=unique_name.generate("lookahead_step"),
                 shape=[1],
                 value=0,
@@ -212,7 +212,7 @@ class LookAhead(Optimizer):
         zero_var = paddle.zeros(
             shape=[1], dtype='int32', name='lookahead_zeros'
         )
-        k_var = layers.create_global_var(
+        k_var = paddle.static.create_global_var(
             name=unique_name.generate("lookahead_k"),
             shape=[1],
             value=self.k,

python/paddle/optimizer/adam.py

@@ -18,7 +18,7 @@ from collections import defaultdict
 import paddle
 from paddle import _C_ops, _legacy_C_ops
 
-from ..fluid import core, framework, layers, unique_name
+from ..fluid import core, framework, unique_name
 from ..fluid.dygraph import base as imperative_base
 from ..fluid.framework import Variable, in_dygraph_mode
 from ..fluid.layer_helper import LayerHelper
@@ -233,7 +233,7 @@ class Adam(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,

python/paddle/optimizer/adamw.py

@@ -19,7 +19,7 @@ from collections.abc import Callable
 import paddle
 
 from .. import _C_ops, _legacy_C_ops
-from ..fluid import core, framework, layers, unique_name
+from ..fluid import core, framework, unique_name
 from ..fluid.clip import GradientClipBase
 from ..fluid.dygraph import base as imperative_base
 from ..fluid.framework import Parameter, Variable
@@ -338,7 +338,7 @@ class AdamW(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,

python/paddle/optimizer/lamb.py

@@ -12,10 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import paddle
 from paddle import _C_ops, _legacy_C_ops
 from paddle.fluid.executor import global_scope
 
-from ..fluid import core, framework, layers, unique_name
+from ..fluid import core, framework, unique_name
 from ..fluid.framework import Variable
 from ..fluid.layer_helper import LayerHelper
 from .optimizer import Optimizer
@@ -162,7 +163,7 @@ class Lamb(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,

python/paddle/optimizer/momentum.py

@@ -19,7 +19,7 @@ from paddle import _C_ops, _legacy_C_ops
 from paddle.fluid.framework import _in_legacy_dygraph, in_dygraph_mode
 from paddle.fluid.regularizer import L2DecayRegularizer
 
-from ..fluid import core, framework, layers, unique_name
+from ..fluid import core, framework, unique_name
 from ..fluid.layer_helper import LayerHelper
 from .optimizer import Optimizer
 
@@ -209,7 +209,7 @@ class Momentum(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,

python/paddle/optimizer/optimizer.py

@@ -31,7 +31,7 @@ from paddle.fluid.framework import (
     name_scope,
 )
 
-from ..fluid import framework, layers, unique_name
+from ..fluid import framework, unique_name
 from ..fluid.backward import _get_no_grad_set_name, append_backward
 from ..fluid.clip import (
     GradientClipBase,
@@ -469,7 +469,7 @@ class Optimizer:
             else:
                 self._learning_rate_map[
                     framework.default_main_program()
-                ] = layers.create_global_var(
+                ] = paddle.static.create_global_var(
                     name=unique_name.generate("learning_rate"),
                     shape=[1],
                     value=float(self._learning_rate),

python/paddle/optimizer/sgd.py

@@ -14,9 +14,10 @@
 
 import warnings
 
+import paddle
 from paddle import _C_ops, _legacy_C_ops
 
-from ..fluid import core, framework, layers, unique_name
+from ..fluid import core, framework, unique_name
 from ..fluid.dygraph import no_grad
 from ..fluid.framework import _in_legacy_dygraph, in_dygraph_mode
 from ..fluid.layer_helper import LayerHelper
@@ -101,7 +102,7 @@ class SGD(Optimizer):
 
             var_name = param.name + "_fp32_master"
             var_name = unique_name.generate(var_name)
-            var = layers.create_global_var(
+            var = paddle.static.create_global_var(
                 name=var_name,
                 shape=param.shape,
                 value=0,

python/paddle/static/__init__.py

@@ -33,6 +33,7 @@ from .input import data  # noqa: F401
 from .input import InputSpec  # noqa: F401
 
 from ..tensor.creation import create_parameter  # noqa: F401
+from ..tensor.creation import create_global_var  # noqa: F401
 
 from ..fluid.executor import Executor  # noqa: F401
 from ..fluid.executor import global_scope  # noqa: F401
@@ -70,7 +71,6 @@ from ..fluid.io import load_vars  # noqa: F401
 from ..fluid.io import save_vars  # noqa: F401
 from ..fluid.io import batch  # noqa: F401
 
-from ..fluid.layers import create_global_var  # noqa: F401
 from ..fluid.contrib.layers import ctr_metric_bundle  # noqa: F401
 from ..fluid.layers import exponential_decay  # noqa: F401
 

python/paddle/tensor/creation.py

@@ -36,7 +36,7 @@ from ..fluid.framework import (
     _in_legacy_dygraph,
     device_guard,
 )
-from ..fluid.initializer import Initializer
+from ..fluid.initializer import Constant, Initializer
 from ..fluid.layers import utils
 from ..fluid.param_attr import ParamAttr
 from ..framework import (
@@ -70,6 +70,84 @@ def _real_to_complex_dtype(dtype):
         return dtype
 
 
+def create_global_var(
+    shape, value, dtype, persistable=False, force_cpu=False, name=None
+):
+    """
+    This function creates a new tensor variable with value in the global block(block 0).
+
+    Args:
+        shape (list[int]|tuple[int]): Shape of the variable
+        value (float): The value of the variable. The new created
+                      variable will be filled with it.
+        dtype (str): Data type of the variable
+        persistable (bool, optional): If this variable is persistable.
+                           Default: False
+        force_cpu (bool, optional): Force this variable to be on CPU.
+                         Default: False
+        name (str, optional): For detailed information, please refer to
+           :ref:`api_guide_Name` . Usually name is no need to set and None by default.
+
+    Returns:
+        Variable: The created Variable
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+            paddle.enable_static()
+            var = paddle.static.create_global_var(shape=[2,3], value=1.0, dtype='float32',
+                                           persistable=True, force_cpu=True, name='new_var')
+    """
+    check_type(shape, 'shape', (list, tuple, np.ndarray), 'create_global_var')
+    for item in shape:
+        check_type(
+            item,
+            'item of shape',
+            (
+                int,
+                np.uint8,
+                np.int8,
+                np.int16,
+                np.int32,
+                np.int64,
+            ),
+            'create_global_var',
+        )
+
+    check_dtype(
+        dtype,
+        'dtype',
+        [
+            'bool',
+            'float16',
+            'float32',
+            'float64',
+            'int8',
+            'int16',
+            'int32',
+            'int64',
+            'uint8',
+            'uint16',
+        ],
+        'create_global_var',
+    )
+
+    helper = LayerHelper("global_var", **locals())
+    var = helper.create_global_variable(
+        dtype=dtype,
+        shape=shape,
+        persistable=persistable,
+        name=name,
+        stop_gradient=True,
+    )
+    helper.set_variable_initializer(
+        var, initializer=Constant(value=float(value), force_cpu=force_cpu)
+    )
+
+    return var
+
+
 def create_parameter(
     shape, dtype, name=None, attr=None, is_bias=False, default_initializer=None
 ):