diff --git a/lite/kernels/arm/norm_compute.cc b/lite/kernels/arm/norm_compute.cc
index 3cc1645fc6823c4c3276cd1f22f4be8a584d2073..fb8b4bbe0773b808a0f6942d1120ebc7d4e844d2 100644
--- a/lite/kernels/arm/norm_compute.cc
+++ b/lite/kernels/arm/norm_compute.cc
@@ -47,4 +47,5 @@ REGISTER_LITE_KERNEL(
     norm, kARM, kFloat, kNCHW, paddle::lite::kernels::arm::NormCompute, def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindOutput("Norm", {LiteType::GetTensorTy(TARGET(kARM))})
     .Finalize();
diff --git a/lite/operators/op_params.h b/lite/operators/op_params.h
index f7e0bcd4e498741ab911c5daf4358a72b646f009..394a83f4f4d60a39b3be0b6bfe3c13135ec3e22c 100644
--- a/lite/operators/op_params.h
+++ b/lite/operators/op_params.h
@@ -612,6 +612,7 @@ struct SequenceSoftmaxParam {
 struct NormParam {
   const lite::Tensor* X{};
   lite::Tensor* Out{};
+  lite::Tensor* Norm{};
   int axis{1};
   float epsilon{1e-10};
 };
diff --git a/lite/tests/kernels/norm_compute_test.cc b/lite/tests/kernels/norm_compute_test.cc
index 830bac062784a8c16752f4e43a23ed8157cc6c0f..6aee1758c19cd793de709921c4733b8892e5f3d9 100644
--- a/lite/tests/kernels/norm_compute_test.cc
+++ b/lite/tests/kernels/norm_compute_test.cc
@@ -46,7 +46,7 @@ class NormComputeTester : public arena::TestCase {
 
     auto* x = scope->FindTensor(input_);
     const auto* x_data = x->data<float>();
-    int axis = axis_ < 0 ? axis + dims_.size() : axis_;
+    int axis = axis_ < 0 ? axis_ + dims_.size() : axis_;
     int pre_n = dims_.count(0, axis);
     int n = dims_[axis];
     int post_n = dims_.count(axis + 1, dims_.size());