diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/auto_parallel_parallelizer.py b/python/paddle/fluid/tests/unittests/collective/fleet/auto_parallel_parallelizer.py
index ffb413fca2c87..b1c00c61057ca 100755
--- a/python/paddle/fluid/tests/unittests/collective/fleet/auto_parallel_parallelizer.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/auto_parallel_parallelizer.py
@@ -22,7 +22,6 @@
 import paddle.utils as utils
 from paddle.distributed import fleet
 from paddle.distributed.fleet import auto
-from paddle.fluid import layers
 
 paddle.enable_static()
 _global_parallel_strategy = None
@@ -93,7 +92,9 @@ def mlp_pretrain_forward(train_program, start_program):
 
         predict = mlp(input)
 
-        cost = layers.cross_entropy(input=predict, label=label)
+        cost = paddle.nn.functional.cross_entropy(
+            input=predict, label=label, reduction='none', use_softmax=False
+        )
         avg_cost = paddle.mean(x=cost)
 
     return avg_cost, train_program, start_program
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/dist_mnist_gradient_merge.py b/python/paddle/fluid/tests/unittests/collective/fleet/dist_mnist_gradient_merge.py
index 85394ea89da4e..01d6d970c6cd2 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/dist_mnist_gradient_merge.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/dist_mnist_gradient_merge.py
@@ -34,7 +34,9 @@ def get_model(self, batch_size=2):
 
         # Train program
         predict = cnn_model(images)
-        cost = fluid.layers.cross_entropy(input=predict, label=label)
+        cost = paddle.nn.functional.cross_entropy(
+            input=predict, label=label, reduction='none', use_softmax=False
+        )
         avg_cost = paddle.mean(x=cost)
 
         # Evaluator
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py b/python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py
index f484a7930059b..9515c987a50f8 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py
@@ -350,7 +350,9 @@ def run_one_loop(self, model, opt, data):
 
         out = model(img)
         softmax_out = paddle.nn.functional.softmax(out, use_cudnn=False)
-        loss = fluid.layers.cross_entropy(input=softmax_out, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=softmax_out, label=label, reduction='none', use_softmax=False
+        )
         avg_loss = paddle.mean(x=loss)
         return avg_loss
 
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist.py b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist.py
index 4530d8e24216a..f974709ce87ab 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist.py
@@ -100,7 +100,9 @@ def get_model(self, batch_size=2, use_dgc=False, dist_strategy=None):
             # Train program
             predict = cnn_model(images)
         with fluid.device_guard("gpu:1"):
-            cost = fluid.layers.cross_entropy(input=predict, label=label)
+            cost = paddle.nn.functional.cross_entropy(
+                input=predict, label=label, reduction='none', use_softmax=False
+            )
             avg_cost = paddle.mean(x=cost)
 
         # Evaluator
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_multi_device.py b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_multi_device.py
index 0b75b034ce46e..ecc71abe6252c 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_multi_device.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_multi_device.py
@@ -100,7 +100,9 @@ def get_model(self, batch_size=2, use_dgc=False, dist_strategy=None):
             # Train program
             predict = cnn_model(images)
         with fluid.device_guard("gpu:1"):
-            cost = fluid.layers.cross_entropy(input=predict, label=label)
+            cost = paddle.nn.functional.cross_entropy(
+                input=predict, label=label, reduction='none', use_softmax=False
+            )
             avg_cost = paddle.mean(x=cost)
 
         # Evaluator
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_one_device.py b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_one_device.py
index 5b1e590fc0058..a6971e210af0c 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_one_device.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_one_device.py
@@ -92,7 +92,9 @@ def get_model(self, batch_size=2, use_dgc=False, dist_strategy=None):
             # Train program
             predict = cnn_model(images)
         with fluid.device_guard("gpu:0"):
-            cost = fluid.layers.cross_entropy(input=predict, label=label)
+            cost = paddle.nn.functional.cross_entropy(
+                input=predict, label=label, reduction='none', use_softmax=False
+            )
             avg_cost = paddle.mean(x=cost)
 
         # Evaluator
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/test_fleet_checkpoint.py b/python/paddle/fluid/tests/unittests/collective/fleet/test_fleet_checkpoint.py
index d1e24149c2df2..b030b6121464d 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/test_fleet_checkpoint.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/test_fleet_checkpoint.py
@@ -41,7 +41,9 @@ def _test_checkpoint(self, fs, dir_path):
             feed_list=[image, label], place=fluid.CPUPlace()
         )
         predict = fluid.layers.fc(input=image, size=10, act='softmax')
-        loss = fluid.layers.cross_entropy(input=predict, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=predict, label=label, reduction='none', use_softmax=False
+        )
         avg_loss = paddle.mean(loss)
         optimizer = fluid.optimizer.AdamOptimizer(learning_rate=0.001)
 
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision.py b/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision.py
index 3141ed81a458b..d30466d9fc957 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision.py
@@ -1308,7 +1308,9 @@ def train_resnet(self, enable_amp=True, level='O1'):
                 ):
                     out = resnet(img)
 
-                loss = fluid.layers.cross_entropy(input=out, label=label)
+                loss = paddle.nn.functional.cross_entropy(
+                    input=out, label=label, reduction='none', use_softmax=False
+                )
                 avg_loss = paddle.mean(x=loss)
 
                 dy_out = avg_loss.numpy()
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py b/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
index dbcdf3ac46c9e..f688d28b85603 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/test_imperative_auto_mixed_precision_for_eager.py
@@ -1298,7 +1298,9 @@ def train_resnet(self, enable_amp=True, level='O1'):
                 ):
                     out = resnet(img)
 
-                loss = fluid.layers.cross_entropy(input=out, label=label)
+                loss = paddle.nn.functional.cross_entropy(
+                    input=out, label=label, reduction='none', use_softmax=False
+                )
                 avg_loss = paddle.mean(x=loss)
 
                 dy_out = avg_loss.numpy()
diff --git a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_quant_dequant_pass.py b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_quant_dequant_pass.py
index 05244e3627b69..9451a214aadf3 100644
--- a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_quant_dequant_pass.py
+++ b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_quant_dequant_pass.py
@@ -53,7 +53,12 @@ def network():
             elif self.conv_groups == 4:
                 cout = paddle.reshape(conv_out, shape=[1, 1, 10816])
             result = fluid.layers.relu(cout)
-            loss = fluid.layers.cross_entropy(input=result, label=label_shape)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=label_shape,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -156,7 +161,12 @@ def network():
             )
             cout = paddle.reshape(conv_out, shape=[1, 1, 10816])
             result = fluid.layers.relu(cout)
-            loss = fluid.layers.cross_entropy(input=result, label=label_shape)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=label_shape,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -257,7 +267,12 @@ def network():
             elif self.conv_groups == 4:
                 cout = paddle.reshape(conv_out, shape=[1, 1, 10816])
             result = fluid.layers.relu(cout)
-            loss = fluid.layers.cross_entropy(input=result, label=label_shape)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=label_shape,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
diff --git a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_fc_fuse_quant_dequant_pass.py b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_fc_fuse_quant_dequant_pass.py
index ca491a06c6860..372a4ea23ba35 100644
--- a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_fc_fuse_quant_dequant_pass.py
+++ b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_fc_fuse_quant_dequant_pass.py
@@ -38,7 +38,12 @@ def network():
                 act="relu",
             )
             result = fluid.layers.relu(fc_out)
-            loss = fluid.layers.cross_entropy(input=result, label=self.label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=self.label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -105,7 +110,12 @@ def network():
             )
             c_out = paddle.reshape(fc_out, shape=[0, 784])
             result = fluid.layers.relu(c_out)
-            loss = fluid.layers.cross_entropy(input=result, label=self.label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=self.label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -174,7 +184,12 @@ def network():
             )
             c_out = paddle.reshape(fc_out, shape=[1, 1, 2744])
             result = fluid.layers.relu(c_out)
-            loss = fluid.layers.cross_entropy(input=result, label=label_shape)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=label_shape,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
diff --git a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_matmul_quant_dequant.py b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_matmul_quant_dequant.py
index 4e2b3e0ae2420..ca6a0d98d89f9 100644
--- a/python/paddle/fluid/tests/unittests/ir/inference/test_trt_matmul_quant_dequant.py
+++ b/python/paddle/fluid/tests/unittests/ir/inference/test_trt_matmul_quant_dequant.py
@@ -47,7 +47,12 @@ def network():
                 act=None,
             )
             result = fluid.layers.relu(fc_out)
-            loss = fluid.layers.cross_entropy(input=result, label=self.label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=self.label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -144,7 +149,12 @@ def network():
                 act=None,
             )
             result = fluid.layers.relu(fc_out)
-            loss = fluid.layers.cross_entropy(input=result, label=self.label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=self.label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result
 
@@ -240,7 +250,12 @@ def network():
                 act=None,
             )
             result = fluid.layers.relu(fc_out)
-            loss = fluid.layers.cross_entropy(input=result, label=self.label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=result,
+                label=self.label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss = paddle.mean(loss)
             return avg_loss, result