diff --git a/lite/kernels/mlu/bridges/act_op_test.cc b/lite/kernels/mlu/bridges/act_op_test.cc
index 7cec0529e49e694c362b3e0a550948f7855c85a2..038e0a92075b5544ab6c5d63fb4a1d12085514cf 100644
--- a/lite/kernels/mlu/bridges/act_op_test.cc
+++ b/lite/kernels/mlu/bridges/act_op_test.cc
@@ -116,7 +116,7 @@ void test_act(std::vector<int64_t> x_shape, std::string op_type) {
     opdesc.SetAttr("offset", 0.5f);
   }
 
-  // create and convert op to NPU model, then run it on NPU
+  // create and convert op to MLU model, then run it on MLU
   auto op = CreateOp<operators::ActivationOp>(opdesc, &scope);
   // execute reference implementation and save to output tensor
   act_ref(op);
diff --git a/lite/kernels/mlu/bridges/elementwise_ops_test.cc b/lite/kernels/mlu/bridges/elementwise_ops_test.cc
index e5087dd708eee3ba255fbfa0383d31b12a6b6870..7844e5b1b57567f72750b21ba288547cb165eb54 100644
--- a/lite/kernels/mlu/bridges/elementwise_ops_test.cc
+++ b/lite/kernels/mlu/bridges/elementwise_ops_test.cc
@@ -153,7 +153,7 @@ void test_elementwise_add(const std::vector<int64_t>& x_shape,
   opdesc.SetOutput("Out", {out_var_name});
   opdesc.SetAttr("axis", axis);
 
-  // create and convert op to NPU model, then run it on NPU
+  // create and convert op to MLU model, then run it on MLU
   auto op = CreateOp<operators::ElementwiseOp>(opdesc, &scope);
 
   // execute reference implementation and save to output tensor
diff --git a/lite/kernels/mlu/bridges/softmax_op_test.cc b/lite/kernels/mlu/bridges/softmax_op_test.cc
index a5251ed43c9187fc2874f9b01853b45b8abf7f1c..d5d7251205a0f60b9e5c8568a58ba48661c9df3e 100644
--- a/lite/kernels/mlu/bridges/softmax_op_test.cc
+++ b/lite/kernels/mlu/bridges/softmax_op_test.cc
@@ -93,7 +93,7 @@ void test_softmax(const std::vector<int64_t>& input_shape, int axis) {
   opdesc.SetOutput("Out", {out_var_name});
   opdesc.SetAttr("axis", axis);
 
-  // create and convert op to NPU model, then run it on NPU
+  // create and convert op to MLU model, then run it on MLU
   auto op = CreateOp<operators::SoftmaxOp>(opdesc, &scope);
   // execute reference implementation and save to output tensor
   softmax_ref<float>(op);