diff --git a/paddle/fluid/inference/tensorrt/convert/ut_helper.h b/paddle/fluid/inference/tensorrt/convert/ut_helper.h
index 833f35167b91cecb2ab7e06c52b920d79d1cec0b..0a6f171fc40a838fd81d6a51aca0430d5526f188 100644
--- a/paddle/fluid/inference/tensorrt/convert/ut_helper.h
+++ b/paddle/fluid/inference/tensorrt/convert/ut_helper.h
@@ -163,6 +163,9 @@ class TRTConvertValidation {
     }
   }
 
+  // We use the set 'neglected_output' here, because some Ops like batch norm,
+  // the outputs specified in the op des are only used during training,
+  // so we should neglect those output during inference.
   void Execute(int batch_size,
                std::unordered_set<std::string> neglected_output = {}) {
     // Execute Fluid Op
@@ -177,9 +180,6 @@ class TRTConvertValidation {
     ASSERT_FALSE(op_desc_->OutputArgumentNames().empty());
     const size_t output_space_size = 3000;
     for (const auto& output : op_desc_->OutputArgumentNames()) {
-      // Some Ops like batch norm, the output specified
-      // in the op des is only used during training, so we should
-      // neglect those ouput during inference.
       if (neglected_output.count(output)) continue;
       std::vector<float> fluid_out;
       std::vector<float> trt_out(output_space_size);