【fluid api clear】Remove reduce sum (#48330)

* remove fluid.reduce_sum

* remove fluid.reduce_sum

* modify axis and import paddle

* modify keepdim and out_name

* modift unittest

* modift unittest

* modify CI_static and loss.py

* modify test_mse_loss

* modify static ci

* modify static ci datatype

* add import paddle in test

* fix conflict

* fix conflict

* modify ci

* modify ci

* fix_conflict

* fix bug

* code_style

paddle/fluid/imperative/README.md

@@ -136,6 +136,7 @@ TODO
 # Examples
 
 ```python
+import paddle
 class MyLayer(fluid.imperative.Layer):
     def __init__(self):
         super(MyLayer, self).__init__()
@@ -143,7 +144,7 @@ class MyLayer(fluid.imperative.Layer):
     def forward(self, inputs):
         x = fluid.layers.relu(inputs)
         x = fluid.layers.elementwise_mul(x, x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return [x]
 
 
@@ -184,7 +185,7 @@ class MLP(fluid.Layer):
     def forward(self, inputs):
         x = self._linear1(inputs)
         x = self._linear2(x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return x
 
 

python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_utils.py

@@ -69,7 +69,7 @@ class GroupShardedClipGrad:
                     layers.merge_selected_rows(g)
                 )
             square = paddle.square(merge_grad)
-            sum_square = layers.reduce_sum(square)
+            sum_square = paddle.sum(square)
 
             if p.dtype == paddle.float16:
                 if p_slice:
@@ -87,7 +87,7 @@ class GroupShardedClipGrad:
             global_norm_fp16 = paddle.to_tensor([0.0], dtype=paddle.float32)
         else:
             global_norm_fp16 = layers.concat(sum_square_fp16)
-            global_norm_fp16 = layers.reduce_sum(global_norm_fp16)
+            global_norm_fp16 = paddle.sum(global_norm_fp16)
             global_norm_fp16 = paddle.cast(
                 global_norm_fp16, dtype=paddle.float32
             )
@@ -97,7 +97,7 @@ class GroupShardedClipGrad:
             global_unslice_fp16 = paddle.to_tensor([0.0], dtype=paddle.float32)
         else:
             global_unslice_fp16 = layers.concat(unslice_params_fp16)
-            global_unslice_fp16 = layers.reduce_sum(global_unslice_fp16)
+            global_unslice_fp16 = paddle.sum(global_unslice_fp16)
             global_unslice_fp16 = paddle.cast(
                 global_unslice_fp16, dtype=paddle.float32
             )
@@ -108,7 +108,7 @@ class GroupShardedClipGrad:
             if len(sum_square_fp32) != 0
             else paddle.to_tensor([0.0], dtype=paddle.float32)
         )
-        global_norm_fp32 = layers.reduce_sum(global_norm_fp32)
+        global_norm_fp32 = paddle.sum(global_norm_fp32)
 
         # global norm of non-distributed FP32 params_and_grads for unslice parameters
         global_unslice_fp32 = (
@@ -116,7 +116,7 @@ class GroupShardedClipGrad:
             if len(unslice_params_fp32) != 0
             else paddle.to_tensor([0.0], dtype=paddle.float32)
         )
-        global_unslice_fp32 = layers.reduce_sum(global_unslice_fp32)
+        global_unslice_fp32 = paddle.sum(global_unslice_fp32)
         global_unslice_var = global_unslice_fp16 + global_unslice_fp32
 
         global_norm_var = (

python/paddle/distributed/fleet/meta_parallel/sharding/sharding_utils.py

@@ -70,8 +70,7 @@ class ShardingClipGrad:
                     layers.merge_selected_rows(g)
                 )
             square = paddle.square(merge_grad)
-            sum_square = layers.reduce_sum(square)
-
+            sum_square = paddle.sum(square)
             if p.dtype == paddle.float16:
                 if p_slice:
                     sum_square_fp16.append(sum_square)
@@ -88,7 +87,7 @@ class ShardingClipGrad:
             global_norm_fp16 = paddle.to_tensor([0.0], dtype=paddle.float32)
         else:
             global_norm_fp16 = layers.concat(sum_square_fp16)
-            global_norm_fp16 = layers.reduce_sum(global_norm_fp16)
+            global_norm_fp16 = paddle.sum(global_norm_fp16)
             global_norm_fp16 = paddle.cast(
                 global_norm_fp16, dtype=paddle.float32
             )
@@ -98,7 +97,7 @@ class ShardingClipGrad:
             global_unslice_fp16 = paddle.to_tensor([0.0], dtype=paddle.float32)
         else:
             global_unslice_fp16 = layers.concat(unslice_params_fp16)
-            global_unslice_fp16 = layers.reduce_sum(global_unslice_fp16)
+            global_unslice_fp16 = paddle.sum(global_unslice_fp16)
             global_unslice_fp16 = paddle.cast(
                 global_unslice_fp16, dtype=paddle.float32
             )
@@ -109,7 +108,7 @@ class ShardingClipGrad:
             if len(sum_square_fp32) != 0
             else paddle.to_tensor([0.0], dtype=paddle.float32)
         )
-        global_norm_fp32 = layers.reduce_sum(global_norm_fp32)
+        global_norm_fp32 = paddle.sum(global_norm_fp32)
 
         # global norm of non-distributed FP32 params_and_grads for unslice parameter
         global_unslice_fp32 = (
@@ -117,7 +116,7 @@ class ShardingClipGrad:
             if len(unslice_params_fp32) != 0
             else paddle.to_tensor([0.0], dtype=paddle.float32)
         )
-        global_unslice_fp32 = layers.reduce_sum(global_unslice_fp32)
+        global_unslice_fp32 = paddle.sum(global_unslice_fp32)
         global_unslice_var = global_unslice_fp16 + global_unslice_fp32
 
         global_norm_var = (

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

@@ -37,7 +37,7 @@ def sum(input, scope=None, util=None):
 
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
-          cnt = fluid.layers.reduce_sum(input)
+          cnt = paddle.sum(input)
           global_cnt = fluid.layers.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)
@@ -77,7 +77,7 @@ def max(input, scope=None, util=None):
 
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
-          cnt = fluid.layers.reduce_sum(input)
+          cnt = paddle.sum(input)
           global_cnt = fluid.layers.create_global_var(persistable=True, dtype='float32', shape=[1], value=0)
           tmp = paddle.maximum(cnt, global_cnt)
           fluid.layers.assign(tmp, global_cnt)
@@ -117,7 +117,7 @@ def min(input, scope=None, util=None):
 
           # in model.py
           input = fluid.layers.cast(some_input, dtype='float32')
-          cnt = fluid.layers.reduce_sum(input)
+          cnt = paddle.sum(input)
           global_cnt = fluid.layers.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)

python/paddle/fluid/clip.py

@@ -73,7 +73,7 @@ def _squared_l2_norm(x):
         or x.dtype == core.VarDesc.VarType.BF16
     ):
         square = paddle.square(x)
-        sum_square = layers.reduce_sum(square)
+        sum_square = paddle.sum(square)
         return sum_square
 
     if in_dygraph_mode():

python/paddle/fluid/contrib/slim/quantization/adaround.py

@@ -64,9 +64,7 @@ class AdaRoundLoss:
         square_cost = fluid.layers.square_error_cost(
             ada_quantized_output, orig_output
         )
-        recon_loss = fluid.layers.reduce_mean(
-            fluid.layers.reduce_sum(square_cost, dim=-1)
-        )
+        recon_loss = fluid.layers.reduce_mean(paddle.sum(square_cost, axis=-1))
         return recon_loss
 
     def compute_round_loss(self, alpha_v, warm_start, beta):
@@ -76,7 +74,7 @@ class AdaRoundLoss:
 
             # calculate regularization term - which ensures parameter to converge to exactly zeros and ones
             # at the end of optimization
-            reg_term = fluid.layers.reduce_sum(
+            reg_term = paddle.sum(
                 -paddle.pow(paddle.abs(2 * h_v - 1), beta) + 1
             )
 

python/paddle/fluid/contrib/tests/test_multi_precision_fp16_train.py

@@ -113,7 +113,7 @@ def train(use_pure_fp16=True, use_nesterov=False, optimizer=""):
         cost = fluid.layers.softmax_with_cross_entropy(
             logits, label, return_softmax=False
         )
-        sum_cost = fluid.layers.reduce_sum(cost)
+        sum_cost = paddle.sum(cost)
 
         # Test program
         test_program = train_program.clone(for_test=True)

python/paddle/fluid/evaluator.py

@@ -15,6 +15,7 @@
 import warnings
 import numpy as np
 
+import paddle
 from . import layers
 from .framework import Program, Variable, program_guard
 from . import unique_name

python/paddle/fluid/framework.py

@@ -1640,6 +1640,7 @@ class Variable(metaclass=VariableMetaClass):
         Examples:
             .. code-block:: python
 
+                import paddle
                 import paddle.fluid as fluid
                 import numpy as np
 
@@ -1652,7 +1653,7 @@ class Variable(metaclass=VariableMetaClass):
                         tmp.stop_gradient=False
                         inputs2.append(tmp)
                     ret2 = fluid.layers.sums(inputs2)
-                    loss2 = fluid.layers.reduce_sum(ret2)
+                    loss2 = paddle.sum(ret2)
                     loss2.backward()
                     print(loss2.gradient())
 
@@ -1687,6 +1688,7 @@ class Variable(metaclass=VariableMetaClass):
         Examples:
             .. code-block:: python
 
+                import paddle
                 import paddle.fluid as fluid
                 import numpy as np
 
@@ -1698,7 +1700,7 @@ class Variable(metaclass=VariableMetaClass):
                         tmp.stop_gradient=False
                         inputs2.append(tmp)
                     ret2 = fluid.layers.sums(inputs2)
-                    loss2 = fluid.layers.reduce_sum(ret2)
+                    loss2 = paddle.sum(ret2)
                     loss2.backward()
                     print(loss2.gradient())
                     loss2.clear_gradient()

python/paddle/fluid/install_check.py

@@ -45,7 +45,7 @@ class SimpleLayer(Layer):
 
     def forward(self, inputs):
         x = self._linear1(inputs)
-        x = layers.reduce_sum(x)
+        x = paddle.sum(x)
         return x
 
 

python/paddle/fluid/layers/control_flow.py

@@ -3399,7 +3399,7 @@ class IfElse:
         output = ie() #  [array([[-7.], [-9.], [ 8.], [ 7.]], dtype=float32)]
 
         # Get the first Variable in the output List and add all elements.
-        out = fluid.layers.reduce_sum(output[0])
+        out = paddle.sum(output[0])
 
         exe = fluid.Executor(fluid.CPUPlace())
         exe.run(fluid.default_startup_program())

python/paddle/fluid/layers/detection.py

@@ -582,6 +582,7 @@ def sigmoid_focal_loss(x, label, fg_num, gamma=2.0, alpha=0.25):
     Examples:
         .. code-block:: python
 
+            import paddle
             import numpy as np
             import paddle.fluid as fluid
 
@@ -591,7 +592,7 @@ def sigmoid_focal_loss(x, label, fg_num, gamma=2.0, alpha=0.25):
             batch_size = 32
             max_iter = 20
 
-
+            paddle.enable_static()
             def gen_train_data():
                 x_data = np.random.uniform(0, 255, (batch_size, 3, image_height,
                                                     image_width)).astype('float64')
@@ -601,12 +602,12 @@ def sigmoid_focal_loss(x, label, fg_num, gamma=2.0, alpha=0.25):
 
 
             def get_focal_loss(pred, label, fg_num, num_classes):
-                pred = fluid.layers.reshape(pred, [-1, num_classes])
-                label = fluid.layers.reshape(label, [-1, 1])
+                pred = paddle.reshape(pred, [-1, num_classes])
+                label = paddle.reshape(label, [-1, 1])
                 label.stop_gradient = True
                 loss = fluid.layers.sigmoid_focal_loss(
                     pred, label, fg_num, gamma=2.0, alpha=0.25)
-                loss = fluid.layers.reduce_sum(loss)
+                loss = paddle.sum(loss)
                 return loss
 
 
@@ -628,7 +629,7 @@ def sigmoid_focal_loss(x, label, fg_num, gamma=2.0, alpha=0.25):
                     data = fluid.layers.fill_constant(shape=[1], value=1, dtype='int32')
                     fg_label = fluid.layers.greater_equal(label, data)
                     fg_label = fluid.layers.cast(fg_label, dtype='int32')
-                    fg_num = fluid.layers.reduce_sum(fg_label)
+                    fg_num = paddle.sum(fg_label, dtype='int32')
                     fg_num.stop_gradient = True
                     avg_loss = get_focal_loss(output, label, fg_num, num_classes)
                     return avg_loss
@@ -1847,9 +1848,9 @@ def ssd_loss(
     # shape=(-1, 0) is set for compile-time, the correct shape is set by
     # actual_shape in runtime.
     loss = paddle.reshape(x=loss, shape=actual_shape)
-    loss = nn.reduce_sum(loss, dim=1, keep_dim=True)
+    loss = paddle.sum(loss, axis=1, keepdim=True)
     if normalize:
-        normalizer = nn.reduce_sum(target_loc_weight)
+        normalizer = paddle.sum(target_loc_weight)
         loss = loss / normalizer
 
     return loss

python/paddle/fluid/layers/distributions.py

@@ -538,13 +538,13 @@ class Categorical(Distribution):
         )
         e_logits = paddle.exp(logits)
         other_e_logits = paddle.exp(other_logits)
-        z = nn.reduce_sum(e_logits, dim=-1, keep_dim=True)
-        other_z = nn.reduce_sum(other_e_logits, dim=-1, keep_dim=True)
+        z = paddle.sum(e_logits, axis=-1, keepdim=True)
+        other_z = paddle.sum(other_e_logits, axis=-1, keepdim=True)
         prob = e_logits / z
-        kl = nn.reduce_sum(
+        kl = paddle.sum(
             prob * (logits - nn.log(z) - other_logits + nn.log(other_z)),
-            dim=-1,
-            keep_dim=True,
+            axis=-1,
+            keepdim=True,
         )
 
         return kl
@@ -558,10 +558,11 @@ class Categorical(Distribution):
         """
         logits = self.logits - paddle.max(self.logits, axis=-1, keepdim=True)
         e_logits = paddle.exp(logits)
-        z = nn.reduce_sum(e_logits, dim=-1, keep_dim=True)
+        z = paddle.sum(e_logits, axis=-1, keepdim=True)
+
         prob = e_logits / z
-        entropy = -1.0 * nn.reduce_sum(
-            prob * (logits - nn.log(z)), dim=-1, keep_dim=True
+        entropy = -1.0 * paddle.sum(
+            prob * (logits - nn.log(z)), axis=-1, keepdim=True
         )
 
         return entropy
@@ -703,7 +704,7 @@ class MultivariateNormalDiag(Distribution):
         """
         check_type(other, 'other', MultivariateNormalDiag, 'kl_divergence')
 
-        tr_cov_matmul = nn.reduce_sum(self._inv(other.scale) * self.scale)
+        tr_cov_matmul = paddle.sum(self._inv(other.scale) * self.scale)
         loc_matmul_cov = nn.matmul(
             (other.loc - self.loc), self._inv(other.scale)
         )

python/paddle/fluid/layers/nn.py

@@ -75,7 +75,6 @@ __all__ = [
     'batch_norm',
     'instance_norm',
     'data_norm',
-    'reduce_sum',
     'reduce_mean',
     'reduce_all',
     'reduce_any',

python/paddle/fluid/optimizer.py

@@ -1788,14 +1788,16 @@ class LarsMomentumOptimizer(Optimizer):
     Examples:
         .. code-block:: python
 
+            import paddle
             import paddle.fluid as fluid
             import numpy as np
 
+            paddle.enable_static()
             np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
             inp = fluid.layers.data(
                 name="inp", shape=[2, 2], append_batch_size=False)
             out = fluid.layers.fc(inp, size=3)
-            out = fluid.layers.reduce_sum(out)
+            out = paddle.sum(out)
             optimizer = fluid.optimizer.LarsMomentumOptimizer(learning_rate=0.001, momentum=0.9)
             optimizer.minimize(out)
 
@@ -2046,13 +2048,15 @@ class AdagradOptimizer(Optimizer):
     Examples:
         .. code-block:: python
 
+            import paddle
             import numpy as np
             import paddle.fluid as fluid
 
+            paddle.enable_static()
             np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
             inp = fluid.data(name="inp", shape=[2, 2])
             out = fluid.layers.fc(inp, size=3)
-            out = fluid.layers.reduce_sum(out)
+            out = paddle.sum(out)
             optimizer = fluid.optimizer.AdagradOptimizer(learning_rate=0.2)
             optimizer.minimize(out)
 

python/paddle/fluid/tests/test_if_else_op.py

@@ -186,7 +186,7 @@ class TestIfElse(unittest.TestCase):
                 false_target = paddle.tanh(false_target)
                 ie.output(false_target)
             if_out = ie()
-            out = layers.reduce_sum(if_out[0])
+            out = paddle.sum(if_out[0])
 
             exe = fluid.Executor(place)
             exe.run(fluid.default_startup_program())

python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_transformer.py

@@ -947,8 +947,8 @@ class TransFormer(Layer):
             soft_label=True if self._label_smooth_eps else False,
         )
         weighted_cost = cost * weights
-        sum_cost = fluid.layers.reduce_sum(weighted_cost)
-        token_num = fluid.layers.reduce_sum(weights)
+        sum_cost = paddle.sum(weighted_cost)
+        token_num = paddle.sum(weights)
         token_num.stop_gradient = True
         avg_cost = sum_cost / token_num
         return sum_cost, avg_cost, predict, token_num

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

@@ -56,7 +56,7 @@ def fake_simnet_reader():
 def get_acc(cos_q_nt, cos_q_pt, batch_size):
     cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
     cond = fluid.layers.cast(cond, dtype='float64')
-    cond_3 = fluid.layers.reduce_sum(cond)
+    cond_3 = paddle.sum(cond)
     acc = fluid.layers.elementwise_div(
         cond_3,
         fluid.layers.fill_constant(

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

@@ -1591,8 +1591,8 @@ def transformer(
         soft_label=True if label_smooth_eps else False,
     )
     weighted_cost = cost * weights
-    sum_cost = layers.reduce_sum(weighted_cost)
-    token_num = layers.reduce_sum(weights)
+    sum_cost = paddle.sum(weighted_cost)
+    token_num = paddle.sum(weights)
     avg_cost = sum_cost / token_num
     avg_cost.stop_gradient = True
     return sum_cost, avg_cost, predict, token_num

python/paddle/fluid/tests/unittests/dygraph_to_static/seq2seq_dygraph_model.py

@@ -302,7 +302,7 @@ class BaseModel(fluid.dygraph.Layer):
         )
         loss = loss * tar_mask
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss
 
@@ -405,7 +405,7 @@ class BaseModel(fluid.dygraph.Layer):
         parent_ids = []
 
         for step_idx in range(paddle.to_tensor(self.beam_max_step_num)):
-            if fluid.layers.reduce_sum(1 - beam_finished).numpy()[0] == 0:
+            if paddle.sum(1 - beam_finished).numpy()[0] == 0:
                 break
             step_input = self._merge_batch_beams(step_input)
             new_dec_hidden, new_dec_cell = [], []
@@ -830,6 +830,6 @@ class AttentionModel(fluid.dygraph.Layer):
         )
         loss = loss * tar_mask
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss

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

@@ -510,8 +510,8 @@ class BOW(Layer):
             right_emb, shape=[-1, self.seq_len, self.bow_dim]
         )
 
-        bow_left = fluid.layers.reduce_sum(left_emb, dim=1)
-        bow_right = fluid.layers.reduce_sum(right_emb, dim=1)
+        bow_left = paddle.sum(left_emb, axis=1)
+        bow_right = paddle.sum(right_emb, axis=1)
         softsign_layer = SoftsignLayer()
         left_soft = softsign_layer.ops(bow_left)
         right_soft = softsign_layer.ops(bow_right)

python/paddle/fluid/tests/unittests/dygraph_to_static/simnet_dygraph_model_v2.py

@@ -497,8 +497,8 @@ class BOW(paddle.nn.Layer):
             right_emb, shape=[-1, self.seq_len, self.bow_dim]
         )
 
-        bow_left = paddle.fluid.layers.reduce_sum(left_emb, dim=1)
-        bow_right = paddle.fluid.layers.reduce_sum(right_emb, dim=1)
+        bow_left = paddle.sum(left_emb, axis=1)
+        bow_right = paddle.sum(right_emb, axis=1)
         softsign_layer = SoftsignLayer()
         left_soft = softsign_layer.ops(bow_left)
         right_soft = softsign_layer.ops(bow_right)

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

@@ -324,9 +324,7 @@ def bmn_loss_func(
             num_entries = fluid.layers.cast(
                 fluid.layers.shape(pmask), dtype=DATATYPE
             )
-            num_positive = fluid.layers.cast(
-                fluid.layers.reduce_sum(pmask), dtype=DATATYPE
-            )
+            num_positive = fluid.layers.cast(paddle.sum(pmask), dtype=DATATYPE)
             ratio = num_entries / num_positive
             coef_0 = 0.5 * ratio / (ratio - 1)
             coef_1 = 0.5 * ratio
@@ -359,15 +357,9 @@ def bmn_loss_func(
         u_lmask = fluid.layers.cast(x=u_lmask, dtype=DATATYPE)
         u_lmask = fluid.layers.elementwise_mul(u_lmask, mask)
 
-        num_h = fluid.layers.cast(
-            fluid.layers.reduce_sum(u_hmask), dtype=DATATYPE
-        )
-        num_m = fluid.layers.cast(
-            fluid.layers.reduce_sum(u_mmask), dtype=DATATYPE
-        )
-        num_l = fluid.layers.cast(
-            fluid.layers.reduce_sum(u_lmask), dtype=DATATYPE
-        )
+        num_h = fluid.layers.cast(paddle.sum(u_hmask), dtype=DATATYPE)
+        num_m = fluid.layers.cast(paddle.sum(u_mmask), dtype=DATATYPE)
+        num_l = fluid.layers.cast(paddle.sum(u_lmask), dtype=DATATYPE)
 
         r_m = num_h / num_m
         u_smmask = fluid.layers.assign(
@@ -391,11 +383,7 @@ def bmn_loss_func(
         weights.stop_gradient = True
         loss = fluid.layers.square_error_cost(pred_score, gt_iou_map)
         loss = fluid.layers.elementwise_mul(loss, weights)
-        loss = (
-            0.5
-            * fluid.layers.reduce_sum(loss)
-            / fluid.layers.reduce_sum(weights)
-        )
+        loss = 0.5 * paddle.sum(loss) / paddle.sum(weights)
 
         return loss
 
@@ -406,8 +394,8 @@ def bmn_loss_func(
         nmask = fluid.layers.cast(x=(gt_iou_map <= 0.9), dtype=DATATYPE)
         nmask = fluid.layers.elementwise_mul(nmask, mask)
 
-        num_positive = fluid.layers.reduce_sum(pmask)
-        num_entries = num_positive + fluid.layers.reduce_sum(nmask)
+        num_positive = paddle.sum(pmask)
+        num_entries = num_positive + paddle.sum(nmask)
         ratio = num_entries / num_positive
         coef_0 = 0.5 * ratio / (ratio - 1)
         coef_1 = 0.5 * ratio
@@ -415,11 +403,11 @@ def bmn_loss_func(
         loss_pos = fluid.layers.elementwise_mul(
             fluid.layers.log(pred_score + epsilon), pmask
         )
-        loss_pos = coef_1 * fluid.layers.reduce_sum(loss_pos)
+        loss_pos = coef_1 * paddle.sum(loss_pos)
         loss_neg = fluid.layers.elementwise_mul(
             fluid.layers.log(1.0 - pred_score + epsilon), nmask
         )
-        loss_neg = coef_0 * fluid.layers.reduce_sum(loss_neg)
+        loss_neg = coef_0 * paddle.sum(loss_neg)
         loss = -1 * (loss_pos + loss_neg) / num_entries
         return loss
 

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

@@ -220,7 +220,7 @@ class PtbModel(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_ptb_lm_v2.py

@@ -214,7 +214,7 @@ class PtbModel(paddle.nn.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = paddle.mean(loss, axis=[0])
-        loss = paddle.fluid.layers.reduce_sum(loss)
+        loss = paddle.paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

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

@@ -123,7 +123,7 @@ def train(args, place, to_static):
 
             loss_probs = fluid.layers.log(loss_probs)
             loss_probs = fluid.layers.elementwise_mul(loss_probs, mask)
-            loss_probs = fluid.layers.reduce_sum(loss_probs, dim=-1)
+            loss_probs = paddle.sum(loss_probs, axis=-1)
 
             policy.saved_log_probs.append(loss_probs)
             return action, loss_probs
@@ -153,7 +153,7 @@ def train(args, place, to_static):
                 policy_loss.append(cur_loss)
 
             policy_loss = fluid.layers.concat(policy_loss)
-            policy_loss = fluid.layers.reduce_sum(policy_loss)
+            policy_loss = paddle.sum(policy_loss)
 
             policy_loss.backward()
             optimizer.minimize(policy_loss)

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

@@ -144,7 +144,7 @@ class BOW(fluid.dygraph.Layer):
         mask_emb = paddle.expand(o_np_mask, [-1, self.hid_dim])
         emb = emb * mask_emb
         emb = paddle.reshape(emb, shape=[-1, self.seq_len, self.hid_dim])
-        bow_1 = fluid.layers.reduce_sum(emb, dim=1)
+        bow_1 = paddle.sum(emb, axis=1)
         bow_1 = paddle.tanh(bow_1)
         fc_1 = self._fc1(bow_1)
         fc_2 = self._fc2(fc_1)

python/paddle/fluid/tests/unittests/dygraph_to_static/test_word2vec.py

@@ -15,6 +15,7 @@
 import math
 import random
 import numpy as np
+import paddle
 import paddle.fluid as fluid
 import unittest
 
@@ -259,7 +260,7 @@ class SkipGram(fluid.dygraph.Layer):
         word_sim = fluid.layers.elementwise_mul(
             center_words_emb, target_words_emb
         )
-        word_sim = fluid.layers.reduce_sum(word_sim, dim=-1)
+        word_sim = paddle.sum(word_sim, axis=-1)
 
         pred = paddle.nn.functional.sigmoid(word_sim)
 

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

@@ -586,8 +586,8 @@ class CrossEntropyCriterion:
             soft_label=True if self.label_smooth_eps else False,
         )
         weighted_cost = cost * weights
-        sum_cost = layers.reduce_sum(weighted_cost)
-        token_num = layers.reduce_sum(weights)
+        sum_cost = paddle.sum(weighted_cost)
+        token_num = paddle.sum(weights)
         token_num.stop_gradient = True
         avg_cost = sum_cost / token_num
         return sum_cost, avg_cost, token_num

python/paddle/fluid/tests/unittests/ipu/test_reduce_x_op_ipu.py

@@ -134,7 +134,7 @@ class TestMin(TestMean):
 
 class TestSum(TestMean):
     def set_test_op(self):
-        self.op = paddle.fluid.layers.reduce_sum
+        self.op = paddle.paddle.sum
 
 
 class TestLogsumexp(TestMean):

python/paddle/fluid/tests/unittests/ir/inference/test_trt_reduce_sum_op.py

@@ -17,6 +17,7 @@ import unittest
 import numpy as np
 from inference_pass_test import InferencePassTest
 
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.core import AnalysisConfig, PassVersionChecker
@@ -28,9 +29,7 @@ class TRTReduceSumTest(InferencePassTest):
             data = fluid.data(
                 name="data", shape=[-1, 3, 10, 192], dtype="float32"
             )
-            reduce_sum = fluid.layers.reduce_sum(
-                data, dim=[2, -1], keep_dim=True
-            )
+            reduce_sum = paddle.sum(data, axis=[2, -1], keepdim=True)
             out = fluid.layers.batch_norm(reduce_sum, is_test=True)
 
         self.feeds = {
@@ -63,7 +62,7 @@ class TRTReduceSumAllTest(InferencePassTest):
             data = fluid.data(
                 name="data", shape=[-1, 3, 10, 192], dtype="float32"
             )
-            reduce_sum = fluid.layers.reduce_sum(data, keep_dim=True)
+            reduce_sum = paddle.sum(data, keepdim=True)
             out = fluid.layers.batch_norm(reduce_sum, is_test=True)
 
         self.feeds = {

python/paddle/fluid/tests/unittests/mlu/sync_batch_norm_op_mlu.py

@@ -98,7 +98,7 @@ class TestSyncBatchNormOpTraining(TestSyncBatchNormRunnerBase):
                 if self.bn_dtype == np.float16:
                     bn = fluid.layers.cast(bn, 'float32')
                 sigmoid = paddle.nn.functional.sigmoid(bn)
-                out = fluid.layers.reduce_sum(sigmoid)
+                out = paddle.sum(sigmoid)
                 # if not sync_bn:
                 #     out = out / core.get_mlu_device_count()
                 if not only_forward:

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

@@ -100,7 +100,7 @@ class TestSyncBatchNormOpTraining(TestSyncBatchNormRunnerBase):
                 # if self.dtype == np.float16:
                 #     bn = fluid.layers.cast(bn, 'float32')
                 sigmoid = paddle.nn.functional.sigmoid(bn)
-                out = fluid.layers.reduce_sum(sigmoid)
+                out = paddle.sum(sigmoid)
                 # if not sync_bn:
                 #     out = out / core.get_npu_device_count()
                 if not only_forward:

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

@@ -85,7 +85,7 @@ class TestReduceSum2(OpTest):
 class TestReduceSumNet(unittest.TestCase):
     def set_reduce_sum_function(self, x):
         # keep_dim = False
-        return paddle.fluid.layers.reduce_sum(x, dim=-1)
+        return paddle.sum(x, axis=-1)
 
     def _test(self, run_npu=True):
         main_prog = paddle.static.Program()
@@ -153,7 +153,7 @@ class TestReduceSumNet(unittest.TestCase):
 class TestReduceSumNet2(TestReduceSumNet):
     def set_reduce_sum_function(self, x):
         # keep_dim = True
-        return paddle.fluid.layers.reduce_sum(x, dim=-1, keep_dim=True)
+        return paddle.sum(x, axis=-1, keepdim=True)
 
 
 class TestReduceSumNet3(TestReduceSumNet):
@@ -172,7 +172,7 @@ class TestReduceSumNet3(TestReduceSumNet):
             b = paddle.static.data(name="b", shape=[2, 3, 4], dtype='float32')
 
             z = paddle.add(a, b)
-            loss = fluid.layers.reduce_sum(z)
+            loss = paddle.sum(z)
             sgd = fluid.optimizer.SGD(learning_rate=0.01)
             sgd.minimize(loss)
 

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

@@ -73,7 +73,7 @@ class SimpleNet(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss
 

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

@@ -95,7 +95,7 @@ class TestArgsortOpCPU(unittest.TestCase):
             )
             self.sorted_x.stop_gradient = False
             loss = fluid.layers.elementwise_mul(self.sorted_x, label)
-            self.loss = fluid.layers.reduce_sum(loss)
+            self.loss = paddle.sum(loss)
 
     def forward(self):
         self.feed_map = {

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

@@ -151,7 +151,7 @@ class TestDistFleetHeterProgram(unittest.TestCase):
         with fluid.device_guard("gpu"):
             labels = fluid.layers.cast(inputs[-1], dtype="int64")
             cost = fluid.layers.cross_entropy(input=predict, label=labels)
-            avg_cost = fluid.layers.reduce_sum(cost)
+            avg_cost = paddle.sum(cost)
 
         return avg_cost
 

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

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps11.py

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps12.py

@@ -41,7 +41,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps13.py

@@ -42,7 +42,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps2.py

@@ -41,7 +41,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps3.py

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps4.py

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps5.py

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

python/paddle/fluid/tests/unittests/test_dist_fleet_ps6.py

@@ -38,7 +38,7 @@ class TestPSPassWithBow(unittest.TestCase):
         def get_acc(cos_q_nt, cos_q_pt, batch_size):
             cond = fluid.layers.less_than(cos_q_nt, cos_q_pt)
             cond = fluid.layers.cast(cond, dtype='float64')
-            cond_3 = fluid.layers.reduce_sum(cond)
+            cond_3 = paddle.sum(cond)
             acc = fluid.layers.elementwise_div(
                 cond_3,
                 fluid.layers.fill_constant(

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

@@ -405,7 +405,7 @@ class TestFakeInit(TranspilerTest):
         neg_emb_b_vec = paddle.reshape(neg_emb_b, shape=[-1, neg_num])
 
         true_logits = fluid.layers.elementwise_add(
-            fluid.layers.reduce_sum(
+            paddle.sum(
                 fluid.layers.elementwise_mul(input_emb, true_emb_w),
                 dim=1,
                 keep_dim=True,
@@ -435,8 +435,8 @@ class TestFakeInit(TranspilerTest):
             neg_logits, label_zeros
         )
         cost = fluid.layers.elementwise_add(
-            fluid.layers.reduce_sum(true_xent, dim=1),
-            fluid.layers.reduce_sum(neg_xent, dim=1),
+            paddle.sum(true_xent, axis=1),
+            paddle.sum(neg_xent, axis=1),
         )
         avg_cost = fluid.layers.reduce_mean(cost)
 

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

@@ -468,7 +468,7 @@ def lm_model(
 
     loss = paddle.reshape(loss, shape=[-1, num_steps])
     loss = layers.reduce_mean(loss, dim=[0])
-    loss = layers.reduce_sum(loss)
+    loss = paddle.sum(loss)
 
     loss.persistable = True
     last_cell.persistable = True

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

@@ -592,7 +592,7 @@ class EagerDeletionTwoRecurrentOpsTest(EagerDeletionRecurrentOpTest1):
             mem_pre = rnn_1.memory(shape=[-1, self.input_dim], batch_ref=x)
             x_t = rnn_1.step_input(x)
             last_rnn_output = rnn_0()
-            last_rnn_sum = fluid.layers.reduce_sum(last_rnn_output)
+            last_rnn_sum = paddle.sum(last_rnn_output)
             mem = layers.elementwise_add(x=x_t, y=last_rnn_sum)
             y = layers.elementwise_add(x=mem_pre, y=mem)
             rnn_1.update_memory(mem_pre, mem)

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

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import unittest
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.incubate.fleet.base.role_maker as role_maker
 from paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler import (
@@ -55,7 +56,7 @@ class TestPyramidHashOpApi(unittest.TestCase):
             name=None,
         )
 
-        cost = fluid.layers.reduce_sum(hash_embd)
+        cost = paddle.sum(hash_embd)
 
         role = role_maker.UserDefinedRoleMaker(
             current_id=0,

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

@@ -34,7 +34,7 @@ class MyLayer(fluid.Layer):
         x = fluid.layers.relu(inputs)
         self._x_for_debug = x
         x = fluid.layers.elementwise_mul(x, x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return [x]
 
 
@@ -65,7 +65,7 @@ class MLP(fluid.Layer):
     def forward(self, inputs):
         x = self._linear1(inputs)
         x = self._linear2(x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return x
 
 
@@ -108,7 +108,7 @@ class SimpleRNNCell(fluid.Layer):
         hidden = self._helper.append_activation(hidden, act='tanh')
         out = paddle.fluid.layers.nn.mul(hidden, self._h2o_w)
         softmax_out = paddle.nn.functional.softmax(out)
-        reduce_out = paddle.fluid.layers.nn.reduce_sum(softmax_out)
+        reduce_out = paddle.sum(softmax_out)
         return reduce_out, hidden
 
 
@@ -342,7 +342,7 @@ class TestImperative(unittest.TestCase):
                 tmp.stop_gradient = False
                 inputs.append(tmp)
             ret = paddle.add_n(inputs)
-            loss = fluid.layers.reduce_sum(ret)
+            loss = paddle.sum(ret)
             loss.backward()
         with fluid.dygraph.guard():
             inputs2 = []
@@ -351,7 +351,7 @@ class TestImperative(unittest.TestCase):
                 tmp.stop_gradient = False
                 inputs2.append(tmp)
             ret2 = paddle.add_n(inputs2)
-            loss2 = fluid.layers.reduce_sum(ret2)
+            loss2 = paddle.sum(ret2)
             fluid.set_flags({'FLAGS_sort_sum_gradient': True})
             loss2.backward()
 
@@ -739,11 +739,11 @@ class TestImperative(unittest.TestCase):
             )
 
             a = paddle.expand(
-                paddle.reshape(fluid.layers.reduce_sum(inp_data1), [1, 1]),
+                paddle.reshape(paddle.sum(inp_data1), [1, 1]),
                 [4, -1],
             )
             b = paddle.expand(
-                paddle.reshape(fluid.layers.reduce_sum(inp_data2), [1, 1]),
+                paddle.reshape(paddle.sum(inp_data2), [1, 1]),
                 [4, -1],
             )
             cond = fluid.layers.less_than(x=a, y=b)

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

@@ -254,9 +254,7 @@ class TestDygraphDeepCF(unittest.TestCase):
 
             deepcf = DeepCF(num_users, num_items, matrix)
             prediction = deepcf(users, items)
-            loss = fluid.layers.reduce_sum(
-                fluid.layers.log_loss(prediction, labels)
-            )
+            loss = paddle.sum(fluid.layers.log_loss(prediction, labels))
             adam = fluid.optimizer.AdamOptimizer(0.01)
             adam.minimize(loss)
 
@@ -309,7 +307,7 @@ class TestDygraphDeepCF(unittest.TestCase):
                         to_variable(users_np[slice : slice + self.batch_size]),
                         to_variable(items_np[slice : slice + self.batch_size]),
                     )
-                    loss = fluid.layers.reduce_sum(
+                    loss = paddle.sum(
                         fluid.layers.log_loss(
                             prediction,
                             to_variable(
@@ -343,7 +341,7 @@ class TestDygraphDeepCF(unittest.TestCase):
                         to_variable(users_np[slice : slice + self.batch_size]),
                         to_variable(items_np[slice : slice + self.batch_size]),
                     )
-                    loss2 = fluid.layers.reduce_sum(
+                    loss2 = paddle.sum(
                         fluid.layers.log_loss(
                             prediction2,
                             to_variable(
@@ -386,7 +384,7 @@ class TestDygraphDeepCF(unittest.TestCase):
                                 items_np[slice : slice + self.batch_size]
                             ),
                         )
-                        loss = fluid.layers.reduce_sum(
+                        loss = paddle.sum(
                             fluid.layers.log_loss(
                                 prediction,
                                 to_variable(

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

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import unittest
+import paddle
 import paddle.fluid as fluid
 import numpy as np
 from test_imperative_base import new_program_scope
@@ -46,7 +47,7 @@ class MLP(fluid.Layer):
     def forward(self, inputs):
         x = self._linear1(inputs)
         x = self._linear2(x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return x
 
 

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

@@ -96,7 +96,7 @@ class TestDygraphGNN(unittest.TestCase):
             # In other example, it's nll with log_softmax. However, paddle's
             # log_loss only supports binary classification now.
             loss = fluid.layers.softmax_with_cross_entropy(logits, labels)
-            loss = fluid.layers.reduce_sum(loss)
+            loss = paddle.sum(loss)
 
             adam = AdamOptimizer(learning_rate=1e-3)
             adam.minimize(loss)
@@ -136,7 +136,7 @@ class TestDygraphGNN(unittest.TestCase):
             loss = fluid.layers.softmax_with_cross_entropy(
                 logits, to_variable(labels)
             )
-            loss = fluid.layers.reduce_sum(loss)
+            loss = paddle.sum(loss)
             loss.backward()
             adam = AdamOptimizer(
                 learning_rate=1e-3, parameter_list=model.parameters()
@@ -164,7 +164,7 @@ class TestDygraphGNN(unittest.TestCase):
             loss2 = fluid.layers.softmax_with_cross_entropy(
                 logits2, to_variable(labels2)
             )
-            loss2 = fluid.layers.reduce_sum(loss2)
+            loss2 = paddle.sum(loss2)
             loss2.backward()
             adam2 = AdamOptimizer(
                 learning_rate=1e-3, parameter_list=model2.parameters()

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

@@ -72,7 +72,7 @@ class SimpleNet(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss
 

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

@@ -325,7 +325,7 @@ class SimpleAttention(fluid.dygraph.Layer):
         scaled = fluid.layers.elementwise_mul(
             x=encoder_vec, y=weights_reshape, axis=0
         )
-        context = fluid.layers.reduce_sum(scaled, dim=1)
+        context = paddle.sum(scaled, axis=1)
 
         return context
 
@@ -498,7 +498,7 @@ class TestDygraphOCRAttention(unittest.TestCase):
                     loss = fluid.layers.cross_entropy(
                         input=dy_prediction, label=label_out
                     )
-                    avg_loss = fluid.layers.reduce_sum(loss)
+                    avg_loss = paddle.sum(loss)
 
                     dy_out = avg_loss.numpy()
 
@@ -576,7 +576,7 @@ class TestDygraphOCRAttention(unittest.TestCase):
             cost = fluid.layers.cross_entropy(
                 input=static_prediction, label=static_label_out
             )
-            static_avg_loss = fluid.layers.reduce_sum(cost)
+            static_avg_loss = paddle.sum(cost)
             # param_grad_list = fluid.backward.append_backward(static_avg_loss)
             optimizer.minimize(static_avg_loss)
 

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

@@ -16,6 +16,7 @@ import unittest
 import numpy as np
 from collections import OrderedDict
 
+import paddle
 import paddle.fluid as fluid
 from paddle.fluid import core
 from paddle.fluid.dygraph.parallel import DataParallel
@@ -34,7 +35,7 @@ class MyLayer(fluid.Layer):
     def forward(self, inputs):
         x = fluid.layers.relu(inputs)
         x = fluid.layers.elementwise_mul(x, x)
-        x = fluid.layers.reduce_sum(x)
+        x = paddle.sum(x)
         return [x]
 
 

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

@@ -231,7 +231,7 @@ class PtbModel(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

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

@@ -28,7 +28,7 @@ class RecurrentTest(fluid.Layer):
 
     def forward(self, in1, in2):
         out = fluid.layers.mul(in1, in2)
-        sum_out = fluid.layers.reduce_sum(out)
+        sum_out = paddle.sum(out)
         return sum_out, out
 
 

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

@@ -73,13 +73,13 @@ class TestImperativeMnist(unittest.TestCase):
 
             loss_probs = fluid.layers.log(loss_probs)
             loss_probs = fluid.layers.elementwise_mul(loss_probs, dy_mask)
-            loss_probs = fluid.layers.reduce_sum(loss_probs, dim=-1)
+            loss_probs = paddle.sum(loss_probs, axis=-1)
 
             dy_reward = fluid.dygraph.base.to_variable(reward)
             dy_reward.stop_gradient = True
 
             loss_probs = fluid.layers.elementwise_mul(dy_reward, loss_probs)
-            loss = fluid.layers.reduce_sum(loss_probs)
+            loss = paddle.sum(loss_probs)
 
             sgd = SGDOptimizer(
                 learning_rate=1e-3, parameter_list=policy.parameters()
@@ -141,12 +141,12 @@ class TestImperativeMnist(unittest.TestCase):
 
             st_loss_probs = fluid.layers.log(st_loss_probs)
             st_loss_probs = fluid.layers.elementwise_mul(st_loss_probs, st_mask)
-            st_loss_probs = fluid.layers.reduce_sum(st_loss_probs, dim=-1)
+            st_loss_probs = paddle.sum(st_loss_probs, axis=-1)
 
             st_loss_probs = fluid.layers.elementwise_mul(
                 st_reward, st_loss_probs
             )
-            st_loss = fluid.layers.reduce_sum(st_loss_probs)
+            st_loss = paddle.sum(st_loss_probs)
 
             st_sgd.minimize(st_loss)
 

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

@@ -227,7 +227,7 @@ class PtbModel(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

python/paddle/fluid/tests/unittests/test_imperative_save_load_v2.py

@@ -230,7 +230,7 @@ class PtbModel(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

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

@@ -81,7 +81,7 @@ class SimpleNet(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss
 

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

@@ -379,9 +379,7 @@ def loss_cls(cls, label, cfg):
     cls_shape = cls.shape
     cls = paddle.reshape(cls, [-1, cls_shape[1] * cls_shape[2] * cls_shape[3]])
     return (
-        fluid.layers.reduce_sum(
-            fluid.layers.sigmoid_cross_entropy_with_logits(cls, label)
-        )
+        paddle.sum(fluid.layers.sigmoid_cross_entropy_with_logits(cls, label))
         / cfg.batch_size
     )
 

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

@@ -1102,8 +1102,8 @@ class TransFormer(Layer):
             soft_label=True if self._label_smooth_eps else False,
         )
         weighted_cost = cost * weights
-        sum_cost = fluid.layers.reduce_sum(weighted_cost)
-        token_num = fluid.layers.reduce_sum(weights)
+        sum_cost = paddle.sum(weighted_cost)
+        token_num = paddle.sum(weights)
         token_num.stop_gradient = True
         avg_cost = sum_cost / token_num
         return sum_cost, avg_cost, predict, token_num

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

@@ -73,9 +73,8 @@ class TestInplaceANBOpTraining(unittest.TestCase):
                 # may have same name, multiply 1. to generate
                 # a new Variable for fetch
                 bn = bn * 1.0
-
                 sigmoid = paddle.nn.functional.sigmoid(bn)
-                out = fluid.layers.reduce_sum(sigmoid)
+                out = paddle.sum(sigmoid)
                 if not only_forward:
                     sgd_opt = fluid.optimizer.SGD(learning_rate=0.0)
                     sgd_opt.backward(out)

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

@@ -69,7 +69,7 @@ def create_program(data_format="NCHW"):
         )
         y = conv(x) + x
 
-        loss = fluid.layers.reduce_sum(y)
+        loss = paddle.sum(y)
 
         sgd = fluid.optimizer.SGD(learning_rate=0.01)
         sgd.minimize(loss)

python/paddle/fluid/tests/unittests/test_lookup_table_v2_op.py

@@ -213,7 +213,7 @@ class TestLookupTableIsSparse(unittest.TestCase):
                 ),
                 is_sparse=is_sparse,
             )
-            y = fluid.layers.reduce_sum(emb, dim=-1)
+            y = paddle.sum(emb, axis=-1)
 
             loss = fluid.layers.square_error_cost(input=y, label=y_)
             loss = paddle.mean(loss)

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

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import unittest
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import numpy as np
@@ -50,7 +51,7 @@ class TestMultiheadAttention(unittest.TestCase):
             num_heads=8,
             dropout_rate=0.0,
         )
-        out = fluid.layers.reduce_sum(contexts, dim=None)
+        out = paddle.sum(contexts, axis=None)
         fluid.backward.append_backward(loss=out)
 
         self.fetch_list = [contexts]

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

@@ -98,7 +98,7 @@ class TestReduceSumWithDimDoubleGradCheck(unittest.TestCase):
 
         x = layers.data('x', shape, False, dtype)
         x.persistable = True
-        y = layers.reduce_sum(x, dim=0)
+        y = paddle.sum(x, axis=0)
         x_arr = np.random.uniform(-1, 1, shape).astype(dtype)
 
         gradient_checker.double_grad_check(

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

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import unittest
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import numpy as np
@@ -37,7 +38,7 @@ class TestNormalization(unittest.TestCase):
         )
         data.stop_gradient = False
         l2_norm = fluid.layers.l2_normalize(x=data, axis=axis, epsilon=epsilon)
-        out = fluid.layers.reduce_sum(l2_norm, dim=None)
+        out = paddle.sum(l2_norm, axis=None)
 
         fluid.backward.append_backward(loss=out)
         self.fetch_list = [l2_norm]

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

@@ -914,10 +914,10 @@ class TestReduceSumOpError(unittest.TestCase):
             x1 = fluid.create_lod_tensor(
                 np.array([[-1]]), [[1]], fluid.CPUPlace()
             )
-            self.assertRaises(TypeError, fluid.layers.reduce_sum, x1)
+            self.assertRaises(TypeError, paddle.sum, x1)
             # The input dtype of reduce_sum_op  must be float32 or float64 or int32 or int64.
             x2 = fluid.layers.data(name='x2', shape=[4], dtype="uint8")
-            self.assertRaises(TypeError, fluid.layers.reduce_sum, x2)
+            self.assertRaises(TypeError, paddle.sum, x2)
 
 
 class API_TestSumOp(unittest.TestCase):

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

@@ -226,7 +226,7 @@ class TestRegularizer(unittest.TestCase):
             para_sum = []
             for para in param_list:
                 para_mul = paddle.square(x=para)
-                para_sum.append(fluid.layers.reduce_sum(input=para_mul))
+                para_sum.append(paddle.sum(para_mul))
             avg_cost_l2 += fluid.layers.sums(para_sum) * 0.5
 
             optimizer = fluid.optimizer.Adagrad(learning_rate=0.1)
@@ -261,7 +261,7 @@ class TestRegularizer(unittest.TestCase):
         with fluid.program_guard(fluid.Program(), fluid.Program()):
             x = fluid.layers.uniform_random([2, 2, 3])
             out = fluid.layers.fc(x, 5, param_attr=fc_param_attr)
-            loss = fluid.layers.reduce_sum(out)
+            loss = paddle.sum(out)
             sgd = fluid.optimizer.SGD(learning_rate=0.1, regularization=l2)
             sgd.minimize(loss)
         with fluid.dygraph.guard():

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

@@ -134,7 +134,7 @@ class TestRegularizer(unittest.TestCase):
             para_sum = []
             for para in param_list:
                 para_mul = paddle.square(x=para)
-                para_sum.append(fluid.layers.reduce_sum(input=para_mul))
+                para_sum.append(paddle.sum(para_mul))
             avg_cost_l2 += fluid.layers.sums(para_sum) * 0.5
 
             optimizer = fluid.optimizer.Adagrad(learning_rate=0.1)
@@ -171,7 +171,7 @@ class TestRegularizer(unittest.TestCase):
         with fluid.program_guard(fluid.Program(), fluid.Program()):
             x = fluid.layers.uniform_random([2, 2, 3])
             out = fluid.layers.fc(x, 5, param_attr=fc_param_attr)
-            loss = fluid.layers.reduce_sum(out)
+            loss = paddle.sum(out)
             sgd = fluid.optimizer.SGD(learning_rate=0.1, regularization=l2)
             sgd.minimize(loss)
         with fluid.dygraph.guard():

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

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import unittest
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.layers.control_flow import lod_rank_table
@@ -51,7 +52,7 @@ class TestReorderLoDTensor(unittest.TestCase):
         new_dat = fluid.layers.reorder_lod_tensor_by_rank(
             x=dat, rank_table=table
         )
-        loss = fluid.layers.reduce_sum(new_dat)
+        loss = paddle.sum(new_dat)
         fluid.backward.append_backward(loss=loss)
         cls.fetch_list = [new_dat, cls.data_desc[0][0] + '@GRAD']
 

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

@@ -148,7 +148,7 @@ class TestReverseLoDTensorArray(unittest.TestCase):
 
             reverse_array = fluid.layers.reverse(tensor_array, axis=axis)
             output, _ = fluid.layers.tensor_array_to_tensor(reverse_array)
-            loss = fluid.layers.reduce_sum(output)
+            loss = paddle.sum(output)
             fluid.backward.append_backward(loss)
             input_grads = list(
                 map(

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

@@ -644,7 +644,7 @@ def def_seq2seq_model(
     )
     loss = loss * tar_mask
     loss = layers.reduce_mean(loss, dim=[0])
-    loss = layers.reduce_sum(loss)
+    loss = paddle.sum(loss)
 
     # optimizer
     optimizer = fluid.optimizer.Adam(0.001)

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

@@ -318,7 +318,7 @@ class PolicyGradient:
         neg_log_prob = layers.cross_entropy(act_prob, action)
         cost = neg_log_prob * reward
         cost = (
-            (layers.reduce_sum(cost) / layers.reduce_sum(length))
+            (paddle.sum(cost) / paddle.sum(length))
             if length is not None
             else layers.reduce_mean(cost)
         )
@@ -407,7 +407,7 @@ class MLE:
         mask = layers.sequence_mask(length, maxlen=max_seq_len, dtype="float32")
         loss = loss * mask
         loss = layers.reduce_mean(loss, dim=[0])
-        loss = layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
         optimizer = fluid.optimizer.Adam(self.lr)
         optimizer.minimize(loss)
         return loss

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

@@ -417,7 +417,7 @@ class TestRunProgramOpWithEmbedding(RunProgramOpTest):
         self.op_type = "run_program"
         self.dtype = np.float32
         self.input_names = {'X': ['x'], 'Params': ['emb_weight']}
-        self.output_names = {'Out': ['reduce_sum_0.tmp_0']}
+        self.output_names = {'Out': ['sum_0.tmp_0']}
 
         self.inputs = {
             'X': {'x': np.array([[1, 3, 0, 4, 7]]).astype("int64")},
@@ -456,7 +456,7 @@ class TestRunProgramOpWithEmbedding(RunProgramOpTest):
             ),
             is_sparse=True,
         )
-        y = fluid.layers.reduce_sum(emb, dim=-1)
+        y = paddle.sum(emb, axis=-1)
         # 2. get forward op num
         fwd_op_num = fluid.default_main_program().global_block().desc.op_size()
         # 3. append backward

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

@@ -1436,7 +1436,7 @@ class TestGradientTruncated(unittest.TestCase):
             # set_value_grad_op will not be run during backward.
             y, value = op(x)
             y2 = y + 1
-            loss = paddle.fluid.layers.reduce_sum(y2)
+            loss = paddle.paddle.sum(y2)
             sgd = paddle.optimizer.Adam()
             sgd.minimize(loss)
             place = (

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

@@ -716,7 +716,7 @@ class TestSliceApiWithLoDTensorArray(unittest.TestCase):
                     slice_arr, axis=self.axis, use_stack=True
                 )
 
-            loss = fluid.layers.reduce_sum(output)
+            loss = paddle.sum(output)
             fluid.backward.append_backward(loss)
             g_vars = list(
                 map(

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

@@ -241,7 +241,7 @@ class PtbModel(fluid.Layer):
         )
         loss = paddle.reshape(loss, shape=[-1, self.num_steps])
         loss = fluid.layers.reduce_mean(loss, dim=[0])
-        loss = fluid.layers.reduce_sum(loss)
+        loss = paddle.sum(loss)
 
         return loss, last_hidden, last_cell
 

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

@@ -95,7 +95,7 @@ class TestSyncBatchNormOpTraining(unittest.TestCase):
                 else:
                     bn = fluid.layers.cast(bn, 'float64')
                 sigmoid = paddle.nn.functional.sigmoid(bn)
-                out = fluid.layers.reduce_sum(sigmoid)
+                out = paddle.sum(sigmoid)
                 if not sync_bn:
                     out = out / core.get_cuda_device_count()
                 if not only_forward:

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

@@ -14,6 +14,7 @@
 
 import unittest
 import numpy as np
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid import Program, program_guard
@@ -196,7 +197,7 @@ class TestLoDTensorArrayStack(unittest.TestCase):
             output, output_index = fluid.layers.tensor_array_to_tensor(
                 input=array, **self.attrs
             )
-            loss = fluid.layers.reduce_sum(output)
+            loss = paddle.sum(output)
             fluid.backward.append_backward(loss)
         self.output_vars = [output, output_index]
 

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

@@ -15,6 +15,7 @@
 import unittest
 import numpy as np
 import collections
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.initializer import ConstantInitializer
@@ -46,7 +47,7 @@ class TestWeightNormalization(unittest.TestCase):
             bias_attr=False,
             act=None,
         )
-        loss = fluid.layers.reduce_sum(out)
+        loss = paddle.sum(out)
         fluid.backward.append_backward(loss=loss)
         cls.fetch_list = [
             'weight_norm_param_g',

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

@@ -158,7 +158,7 @@ def multi_head_attention(
 
         def __softmax(x, eps=1e-9):
             exp_out = paddle.exp(x=x)
-            sum_out = layers.reduce_sum(exp_out, dim=-1, keep_dim=False)
+            sum_out = paddle.sum(exp_out, axis=-1, keepdim=False)
             return layers.elementwise_div(x=exp_out, y=sum_out, axis=0)
 
         scaled_q = paddle.scale(x=q, scale=d_model**-0.5)
@@ -595,4 +595,4 @@ def transformer(
 
     cost = layers.cross_entropy(input=predict, label=gold)
     weighted_cost = cost * weights
-    return layers.reduce_sum(weighted_cost)
+    return paddle.sum(weighted_cost)

python/paddle/incubate/distributed/models/moe/grad_clip.py

@@ -143,21 +143,21 @@ class ClipGradForMOEByGlobalNorm(ClipGradBase):
         global_norm_var = []
         if len(sum_square_list_fp16) > 0:
             global_norm_var_fp16 = layers.concat(sum_square_list_fp16)
-            global_norm_var_fp16 = layers.reduce_sum(global_norm_var_fp16)
+            global_norm_var_fp16 = paddle.sum(global_norm_var_fp16)
             global_norm_var.append(global_norm_var_fp16.astype(sum_dtype))
         if len(sum_square_list_fp32) > 0:
             global_norm_var_fp32 = layers.concat(sum_square_list_fp32)
-            global_norm_var_fp32 = layers.reduce_sum(global_norm_var_fp32)
+            global_norm_var_fp32 = paddle.sum(global_norm_var_fp32)
             if sum_dtype == 'float32':
                 global_norm_var.append(global_norm_var_fp32)
             else:
                 global_norm_var.append(global_norm_var_fp32.astype(sum_dtype))
         if len(sum_square_list) > 0:
             global_norm_var_fp64 = layers.concat(sum_square_list)
-            global_norm_var_fp64 = layers.reduce_sum(global_norm_var_fp64)
+            global_norm_var_fp64 = paddle.sum(global_norm_var_fp64)
             global_norm_var.append(global_norm_var_fp64)
         global_norm_var = layers.concat(global_norm_var)
-        global_norm_var = layers.reduce_sum(global_norm_var)
+        global_norm_var = paddle.sum(global_norm_var)
         return global_norm_var, sum_dtype
 
     @imperative_base.no_grad

python/paddle/nn/layer/loss.py

@@ -593,7 +593,8 @@ class MSELoss(Layer):
 
         reduce_op = 'reduce_mean'
         if self.reduction == 'sum':
-            reduce_op = 'reduce_sum'
+            square_out = paddle.sum(square_out)
+            return square_out
 
         return getattr(fluid.layers, reduce_op)(square_out)
 

python/paddle/tests/test_model.py

@@ -163,7 +163,7 @@ def dynamic_train(model, dataloader):
     for inputs, labels in dataloader:
         outputs = model(inputs)
         loss = CrossEntropyLoss(reduction="sum")(outputs, labels)
-        avg_loss = fluid.layers.reduce_sum(loss)
+        avg_loss = paddle.sum(loss)
         avg_loss.backward()
         optim.minimize(avg_loss)
         model.clear_gradients()
@@ -510,7 +510,7 @@ class TestModelFunction(unittest.TestCase):
             m.train()
             output = m(to_tensor(data))
             loss = CrossEntropyLoss(reduction='sum')(output, to_tensor(label))
-            avg_loss = fluid.layers.reduce_sum(loss)
+            avg_loss = paddle.sum(loss)
             avg_loss.backward()
             optim.minimize(avg_loss)
             m.clear_gradients()