diff --git a/paddle/fluid/pybind/imperative.cc b/paddle/fluid/pybind/imperative.cc
index b5dad4034a5f34a9626886c5e2c0830e3c74b3ad..4c46af3199e29e50e661c660e81665a404beaa67 100644
--- a/paddle/fluid/pybind/imperative.cc
+++ b/paddle/fluid/pybind/imperative.cc
@@ -712,7 +712,7 @@ void BindImperative(py::module *m_ptr) {
                     tmp.stop_gradient=False
                     inputs.append(tmp)
                 ret = paddle.sums(inputs2)
-                loss = paddle.fluid.layers.reduce_sum(ret)
+                loss = paddle.sum(ret)
                 loss.backward()
                 print("Before clear_gradient {}".format(loss.grad))
                 loss.clear_gradient()
diff --git a/python/paddle/nn/functional/loss.py b/python/paddle/nn/functional/loss.py
index 2ca2015b729a8ed43708e91ee5d0fc3aedb034b7..b056029fb5aa1b1d039ba4cc6d9d747d96c9fc14 100644
--- a/python/paddle/nn/functional/loss.py
+++ b/python/paddle/nn/functional/loss.py
@@ -305,13 +305,15 @@ def binary_cross_entropy_with_logits(logit,
     out = paddle.fluid.layers.sigmoid_cross_entropy_with_logits(
         logit, label, name=sigmoid_name)
 
-    one = paddle.fluid.layers.fill_constant(shape=[1], value=1.0, dtype=logit.dtype)
+    one = paddle.fluid.layers.fill_constant(
+        shape=[1], value=1.0, dtype=logit.dtype)
     if pos_weight is not None:
         fluid.data_feeder.check_variable_and_dtype(
             pos_weight, 'pos_weight', ['float32', 'float64'],
             'binary_cross_entropy_with_logits')
         log_weight = paddle.add(
-            paddle.multiply(label, paddle.fluid.layers.elementwise_sub(pos_weight, one)),
+            paddle.multiply(
+                label, paddle.fluid.layers.elementwise_sub(pos_weight, one)),
             one)
         pos_weight_name = name if reduction == 'none' and weight is None else None
         out = paddle.multiply(out, log_weight, name=pos_weight_name)
@@ -618,7 +620,8 @@ def margin_ranking_loss(input,
 
     if margin != 0.0:
         margin_var = out.block.create_var(dtype=out.dtype)
-        paddle.fluid.layers.fill_constant([1], out.dtype, margin, out=margin_var)
+        paddle.fluid.layers.fill_constant(
+            [1], out.dtype, margin, out=margin_var)
         out = paddle.add(out, margin_var)
 
     result_out = helper.create_variable_for_type_inference(input.dtype)
@@ -729,7 +732,8 @@ def l1_loss(input, label, reduction='mean', name=None):
         unreduced = paddle.fluid.layers.elementwise_sub(input, label, act='abs')
         return paddle.mean(unreduced, name=name)
     else:
-        return paddle.fluid.layers.elementwise_sub(input, label, act='abs', name=name)
+        return paddle.fluid.layers.elementwise_sub(
+            input, label, act='abs', name=name)
 
 
 def nll_loss(input,
@@ -1342,7 +1346,7 @@ def sigmoid_focal_loss(logit,
             label = paddle.to_tensor([[1.0, 0.0, 0.0], [0.0, 1.0, 0.0]], dtype='float32')
             one = paddle.to_tensor([1.], dtype='float32')
             fg_label = paddle.greater_equal(label, one)
-            fg_num = paddle.fluid.layers.reduce_sum(paddle.cast(fg_label, dtype='float32'))
+            fg_num = paddle.sum(paddle.cast(fg_label, dtype='float32'))
             output = paddle.nn.functional.sigmoid_focal_loss(logit, label, normalizer=fg_num)
             print(output.numpy())  # [0.65782464]