diff --git a/paddle/fluid/inference/api/paddle_pass_builder.cc b/paddle/fluid/inference/api/paddle_pass_builder.cc
index 87e02a02caebd93d701dfd9e51c35fb974c770ed..3d72295be4b779693a56d1ddb6bc4aad7c2c82c9 100644
--- a/paddle/fluid/inference/api/paddle_pass_builder.cc
+++ b/paddle/fluid/inference/api/paddle_pass_builder.cc
@@ -86,7 +86,8 @@ const std::vector<std::string> kAnakinSubgraphPasses({
 
 GpuPassStrategy::GpuPassStrategy() : PassStrategy({}) {
   passes_.assign({
-    "infer_clean_graph_pass",  //
+    "infer_clean_graph_pass",          //
+        "runtime_context_cache_pass",  //
         //   "identity_scale_op_clean_pass",              //
         "conv_affine_channel_fuse_pass",             //
         "conv_eltwiseadd_affine_channel_fuse_pass",  //
@@ -96,7 +97,6 @@ GpuPassStrategy::GpuPassStrategy() : PassStrategy({}) {
         "conv_elementwise_add_act_fuse_pass",   //
         "conv_elementwise_add2_act_fuse_pass",  //
         "conv_elementwise_add_fuse_pass",       //
-        "runtime_context_cache_pass",           //
 #endif                                          //
         "transpose_flatten_concat_fuse_pass",
   });
@@ -116,7 +116,11 @@ CpuPassStrategy::CpuPassStrategy() : PassStrategy({}) {
   // NOTE the large fusions should be located in the front, so that they will
   // not be damaged by smaller ones.
   passes_.assign({
-      "infer_clean_graph_pass",         //
+      "infer_clean_graph_pass",  //
+      // TODO(luotao): runtime_context_cache_pass should be located in the
+      // front, see https://github.com/PaddlePaddle/Paddle/issues/16609,
+      // will enhance this pass later.
+      "runtime_context_cache_pass",     //
       "attention_lstm_fuse_pass",       //
       "seqpool_concat_fuse_pass",       //
       "seqconv_eltadd_relu_fuse_pass",  //
@@ -132,8 +136,6 @@ CpuPassStrategy::CpuPassStrategy() : PassStrategy({}) {
       "conv_bn_fuse_pass",             //
       "conv_eltwiseadd_bn_fuse_pass",  //
       "is_test_pass",                  //
-      "identity_scale_op_clean_pass",  //
-      "runtime_context_cache_pass",    //
   });
   use_gpu_ = false;
 }
diff --git a/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc b/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc
index ece094717b8076321c68d7fdd29f07c4da6b0ed4..fbf67d933786e3ee2baab7a20911da2837cdce4d 100644
--- a/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc
@@ -23,18 +23,11 @@ namespace analysis {
 
 void SetConfig(AnalysisConfig *cfg) {
   cfg->SetModel(FLAGS_infer_model);
-  cfg->SetProgFile("__model__");
   cfg->DisableGpu();
   cfg->SwitchIrOptim();
-  cfg->SwitchSpecifyInputNames(false);
+  cfg->SwitchSpecifyInputNames();
   cfg->SetCpuMathLibraryNumThreads(FLAGS_paddle_num_threads);
   cfg->EnableMKLDNN();
-  cfg->pass_builder()->SetPasses(
-      {"infer_clean_graph_pass", "mkldnn_placement_pass",
-       "depthwise_conv_mkldnn_pass", "conv_bn_fuse_pass",
-       "conv_eltwiseadd_bn_fuse_pass", "conv_bias_mkldnn_fuse_pass",
-       "conv_elementwise_add_mkldnn_fuse_pass", "conv_relu_mkldnn_fuse_pass",
-       "fc_fuse_pass", "is_test_pass"});
 }
 
 template <typename T>
@@ -84,13 +77,13 @@ std::shared_ptr<std::vector<PaddleTensor>> GetWarmupData(
           std::to_string(num_images) + " is bigger than all test data size.");
 
   PaddleTensor images;
-  images.name = "input";
+  images.name = "image";
   images.shape = {num_images, 3, 224, 224};
   images.dtype = PaddleDType::FLOAT32;
   images.data.Resize(sizeof(float) * num_images * 3 * 224 * 224);
 
   PaddleTensor labels;
-  labels.name = "labels";
+  labels.name = "label";
   labels.shape = {num_images, 1};
   labels.dtype = PaddleDType::INT64;
   labels.data.Resize(sizeof(int64_t) * num_images);
@@ -132,7 +125,7 @@ void SetInput(std::vector<std::vector<PaddleTensor>> *inputs,
       images_offset_in_file + sizeof(float) * total_images * 3 * 224 * 224;
 
   TensorReader<float> image_reader(file, images_offset_in_file,
-                                   image_batch_shape, "input");
+                                   image_batch_shape, "image");
   TensorReader<int64_t> label_reader(file, labels_offset_in_file,
                                      label_batch_shape, "label");
 
diff --git a/paddle/fluid/inference/tests/api/tester_helper.h b/paddle/fluid/inference/tests/api/tester_helper.h
index 9a0dcc722cf00984b8c0e3ac20f13849e2904102..5cc54ed299c50b48c83de2742b715b16cf1f8cd0 100644
--- a/paddle/fluid/inference/tests/api/tester_helper.h
+++ b/paddle/fluid/inference/tests/api/tester_helper.h
@@ -316,7 +316,8 @@ void PredictionRun(PaddlePredictor *predictor,
                    int num_threads, int tid) {
   int num_times = FLAGS_repeat;
   int iterations = inputs.size();  // process the whole dataset ...
-  if (FLAGS_iterations > 0 && FLAGS_iterations < inputs.size())
+  if (FLAGS_iterations > 0 &&
+      FLAGS_iterations < static_cast<int64_t>(inputs.size()))
     iterations =
         FLAGS_iterations;  // ... unless the number of iterations is set
   outputs->resize(iterations);
@@ -329,14 +330,14 @@ void PredictionRun(PaddlePredictor *predictor,
 #endif
   if (!FLAGS_zero_copy) {
     run_timer.tic();
-    for (size_t i = 0; i < iterations; i++) {
+    for (int i = 0; i < iterations; i++) {
       for (int j = 0; j < num_times; j++) {
         predictor->Run(inputs[i], &(*outputs)[i], FLAGS_batch_size);
       }
     }
     elapsed_time = run_timer.toc();
   } else {
-    for (size_t i = 0; i < iterations; i++) {
+    for (int i = 0; i < iterations; i++) {
       ConvertPaddleTensorToZeroCopyTensor(predictor, inputs[i]);
       run_timer.tic();
       for (int j = 0; j < num_times; j++) {
@@ -366,9 +367,8 @@ void TestOneThreadPrediction(
     const std::vector<std::vector<PaddleTensor>> &inputs,
     std::vector<std::vector<PaddleTensor>> *outputs, bool use_analysis = true) {
   auto predictor = CreateTestPredictor(config, use_analysis);
-  PredictionWarmUp(predictor.get(), inputs, outputs, FLAGS_paddle_num_threads,
-                   0);
-  PredictionRun(predictor.get(), inputs, outputs, FLAGS_paddle_num_threads, 0);
+  PredictionWarmUp(predictor.get(), inputs, outputs, 1, 0);
+  PredictionRun(predictor.get(), inputs, outputs, 1, 0);
 }
 
 void TestMultiThreadPrediction(