diff --git a/mindspore/lite/examples/train_lenet/model/train_utils.py b/mindspore/lite/examples/train_lenet/model/train_utils.py
index a9a0b175b7..9e8e3fa315 100644
--- a/mindspore/lite/examples/train_lenet/model/train_utils.py
+++ b/mindspore/lite/examples/train_lenet/model/train_utils.py
@@ -28,7 +28,7 @@ def TrainWrap(net, loss_fn=None, optimizer=None, weights=None):
     if weights is None:
         weights = ParameterTuple(net.trainable_params())
     if optimizer is None:
-        optimizer = nn.Adam(weights, learning_rate=1e-2, beta1=0.9, beta2=0.999, eps=1e-8, use_locking=False,
-                            use_nesterov=False, weight_decay=0.0, loss_scale=1.0)
+        optimizer = nn.Adam(weights, learning_rate=0.003, beta1=0.9, beta2=0.999, eps=1e-5, use_locking=False,
+                            use_nesterov=False, weight_decay=4e-5, loss_scale=1.0)
     train_net = nn.TrainOneStepCell(loss_net, optimizer)
     return train_net
diff --git a/mindspore/lite/examples/train_lenet/src/net_runner.cc b/mindspore/lite/examples/train_lenet/src/net_runner.cc
index 13c12baefd..a44551aeaf 100644
--- a/mindspore/lite/examples/train_lenet/src/net_runner.cc
+++ b/mindspore/lite/examples/train_lenet/src/net_runner.cc
@@ -159,7 +159,7 @@ int NetRunner::InitDB() {
 }
 
 int NetRunner::TrainLoop() {
-  struct mindspore::lite::StepLRLambda step_lr_lambda(1, 0.8);
+  struct mindspore::lite::StepLRLambda step_lr_lambda(1, 0.7);
   mindspore::lite::LRScheduler step_lr_sched(mindspore::lite::StepLRLambda, static_cast<void *>(&step_lr_lambda), 1);
 
   mindspore::lite::LossMonitor lm(100);
diff --git a/mindspore/lite/examples/transfer_learning/model/prepare_model.sh b/mindspore/lite/examples/transfer_learning/model/prepare_model.sh
index 21f9dcb1ac..cc049cf48a 100755
--- a/mindspore/lite/examples/transfer_learning/model/prepare_model.sh
+++ b/mindspore/lite/examples/transfer_learning/model/prepare_model.sh
@@ -8,7 +8,7 @@ fi
 echo "============Exporting=========="
 if [ -n "$1" ]; then
   DOCKER_IMG=$1
-  rm *.so*
+  rm -f *.so*
   docker run -w $PWD --runtime=nvidia -v /home/$USER:/home/$USER --privileged=true ${DOCKER_IMG} /bin/bash -c "python transfer_learning_export.py; chmod 444 transfer_learning_tod*.mindir; rm -rf __pycache__"
 else
   echo "MindSpore docker was not provided, attempting to run locally"
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp32_grad/bn_grad.cc b/mindspore/lite/src/runtime/kernel/arm/fp32_grad/bn_grad.cc
index 67459b0e6f..b8b0428d84 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp32_grad/bn_grad.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp32_grad/bn_grad.cc
@@ -18,7 +18,7 @@
 #include <math.h>
 #include <algorithm>
 #include <vector>
-
+#include <string>
 #include <thread>
 #include <fstream>
 
@@ -51,6 +51,12 @@ int BNGradCPUKernel::Execute(int task_id) {
   auto *input_scale = in_tensors_.at(2);
   auto *input_mean = in_tensors_.at(3);
   auto *input_var = in_tensors_.at(4);
+
+  auto kernel_name = this->name();
+  if (kernel_name.find("FusedBatchNormGradCPU") != std::string::npos) {
+    input_mean = in_tensors_.at(4);
+    input_var = in_tensors_.at(5);
+  }
   auto bn_param = reinterpret_cast<BNGradParameter *>(op_parameter_);
   int stage = stage_;
   int thread_num = thread_num_;
diff --git a/mindspore/lite/src/train/transfer_session.h b/mindspore/lite/src/train/transfer_session.h
index ee6ea56a1a..69a08c8c54 100644
--- a/mindspore/lite/src/train/transfer_session.h
+++ b/mindspore/lite/src/train/transfer_session.h
@@ -64,11 +64,11 @@ class TransferSession : public lite::TrainSession {
   int CompileTransferGraph();
 
  protected:
-  lite::LiteSession *backbone_session_;
-  char *lite_model_;
+  lite::LiteSession *backbone_session_ = nullptr;
+  char *lite_model_ = nullptr;
   std::vector<mindspore::tensor::MSTensor *> combined_inputs_;
   std::vector<std::pair<mindspore::tensor::MSTensor *, mindspore::tensor::MSTensor *>> backbone_head_map_;
-  bool is_valid_;
+  bool is_valid_ = false;
 
  private:
   bool CompileFormatTransform(tensor::MSTensor *out, tensor::MSTensor *in, int *mask);