Mv inference ut (#52987)

* mv inference/api infer_ut * mv test * merge develop fix error * fix * fix build error * fix build error * fix bug * fix tester_helper.h * fix analyzer_transformer_profile_tester.cc * fix * fix mac * fix mac * fix error * fix * fix

Mv inference ut (#52987)
* mv inference/api infer_ut * mv test * merge develop fix error * fix * fix build error * fix build error * fix bug * fix tester_helper.h * fix analyzer_transformer_profile_tester.cc * fix * fix mac * fix mac * fix error * fix * fix
008debe7 · tianshuo78520a · GitHub · 4efca9fb · 008debe7 · 008debe7
121 changed file
--- a/paddle/fluid/inference/CMakeLists.txt
+++ b/paddle/fluid/inference/CMakeLists.txt
@@ -13,10 +13,6 @@
 # limitations under the License.
 #
-if(WITH_TESTING)
-  include(tests/test.cmake) # some generic cmake function for inference
-endif()
 cc_library(
  paddle_inference_io
  SRCS io.cc
@@ -91,10 +87,6 @@ endif()
 # C inference API
 add_subdirectory(capi_exp)
-if(WITH_TESTING AND WITH_INFERENCE_API_TEST)
-  add_subdirectory(tests/api)
-endif()
 set(SHARED_INFERENCE_SRCS
    io.cc
    ${CMAKE_CURRENT_SOURCE_DIR}/../framework/data_feed.cc

--- a/paddle/fluid/inference/analysis/CMakeLists.txt
+++ b/paddle/fluid/inference/analysis/CMakeLists.txt
@@ -36,84 +36,3 @@ cc_library(
  analysis
  SRCS analyzer.cc
  DEPS ${analysis_deps} analysis_helper analysis_pass ${INFER_IR_PASSES})
-function(inference_analysis_test_build TARGET)
-  if(WITH_TESTING)
-    set(options "")
-    set(oneValueArgs "")
-    set(multiValueArgs SRCS EXTRA_DEPS)
-    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
-                          "${multiValueArgs}" ${ARGN})
-    inference_base_test_build(
-      ${TARGET}
-      SRCS
-      ${analysis_test_SRCS}
-      DEPS
-      ${analysis_test_EXTRA_DEPS}
-      analysis
-      pass
-      ${GLOB_PASS_LIB})
-  endif()
-endfunction()
-function(inference_analysis_test_run TARGET)
-  if(WITH_TESTING)
-    set(options "")
-    set(oneValueArgs "")
-    set(multiValueArgs COMMAND ARGS)
-    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
-                          "${multiValueArgs}" ${ARGN})
-    inference_base_test_run(${TARGET} COMMAND ${analysis_test_COMMAND} ARGS
-                            ${analysis_test_ARGS})
-    set_tests_properties(${TARGET} PROPERTIES LABELS "RUN_TYPE=INFER")
-  endif()
-endfunction()
-function(inference_analysis_test TARGET)
-  if(WITH_TESTING)
-    set(options "")
-    set(oneValueArgs "")
-    set(multiValueArgs SRCS ARGS EXTRA_DEPS)
-    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
-                          "${multiValueArgs}" ${ARGN})
-    inference_base_test_build(
-      ${TARGET}
-      SRCS
-      ${analysis_test_SRCS}
-      DEPS
-      ${analysis_test_EXTRA_DEPS}
-      analysis
-      pass
-      ${GLOB_PASS_LIB})
-    inference_base_test_run(${TARGET} COMMAND ${TARGET} ARGS
-                            ${analysis_test_ARGS})
-    set_tests_properties(${TARGET} PROPERTIES LABELS "RUN_TYPE=INFER")
-  endif()
-endfunction()
-if(NOT APPLE AND NOT WIN32)
-  inference_analysis_test(
-    test_analyzer
-    SRCS
-    analyzer_tester.cc
-    EXTRA_DEPS
-    reset_tensor_array
-    paddle_inference_shared
-    ARGS
-    --inference_model_dir=${WORD2VEC_MODEL_DIR})
-elseif(WIN32)
-  inference_analysis_test(
-    test_analyzer
-    SRCS
-    analyzer_tester.cc
-    EXTRA_DEPS
-    reset_tensor_array
-    paddle_inference_api
-    ARGS
-    --inference_model_dir=${WORD2VEC_MODEL_DIR})
-  if(WITH_ONNXRUNTIME AND WIN32)
-    # Copy onnxruntime for some c++ test in Windows, since the test will
-    # be build only in CI, so suppose the generator in Windows is Ninja.
-    copy_onnx(test_analyzer)
-  endif()
-endif()
--- a/paddle/fluid/inference/api/CMakeLists.txt
+++ b/paddle/fluid/inference/api/CMakeLists.txt
@@ -97,85 +97,8 @@ else()
         infer_io_utils model_utils)
 endif()
-cc_test(
-  test_paddle_inference_api
-  SRCS api_tester.cc
-  DEPS paddle_inference_api)
-cc_test(
-  inference_api_helper_test
-  SRCS helper_test.cc
-  DEPS paddle_inference_api)
 if(WITH_ONNXRUNTIME AND WIN32)
  # Copy onnxruntime for some c++ test in Windows, since the test will
  # be build only in CI, so suppose the generator in Windows is Ninja.
  copy_onnx(test_paddle_inference_api)
 endif()
-if(WITH_TESTING)
-  if(NOT APPLE AND NOT WIN32)
-    inference_base_test(
-      test_api_impl
-      SRCS
-      api_impl_tester.cc
-      DEPS
-      paddle_inference_shared
-      ARGS
-      --word2vec_dirname=${WORD2VEC_MODEL_DIR}
-      --book_dirname=${IMG_CLS_RESNET_INSTALL_DIR})
-  elseif(WIN32)
-    inference_base_test(
-      test_api_impl
-      SRCS
-      api_impl_tester.cc
-      DEPS
-      ${inference_deps}
-      ARGS
-      --word2vec_dirname=${WORD2VEC_MODEL_DIR}
-      --book_dirname=${IMG_CLS_RESNET_INSTALL_DIR})
-  endif()
-endif()
-if(NOT APPLE AND NOT WIN32)
-  cc_test_old(
-    test_analysis_predictor
-    SRCS
-    analysis_predictor_tester.cc
-    DEPS
-    paddle_inference_shared
-    ARGS
-    --dirname=${WORD2VEC_MODEL_DIR})
-elseif(WIN32)
-  cc_test_old(
-    test_analysis_predictor
-    SRCS
-    analysis_predictor_tester.cc
-    DEPS
-    analysis_predictor
-    benchmark
-    ${inference_deps}
-    ARGS
-    --dirname=${WORD2VEC_MODEL_DIR})
-endif()
-if(WITH_TESTING AND WITH_MKLDNN)
-  if(NOT APPLE AND NOT WIN32)
-    cc_test(
-      test_mkldnn_quantizer
-      SRCS mkldnn_quantizer_tester.cc
-      DEPS paddle_inference_shared ARGS --dirname=${WORD2VEC_MODEL_DIR})
-  elseif(WIN32)
-    cc_test(
-      test_mkldnn_quantizer
-      SRCS mkldnn_quantizer_tester.cc
-      DEPS analysis_predictor benchmark ${inference_deps} ARGS
-           --dirname=${WORD2VEC_MODEL_DIR})
-  endif()
-endif()
-if(WITH_TESTING AND TEST test_api_impl)
-  if(NOT APPLE)
-    set_tests_properties(test_api_impl PROPERTIES TIMEOUT 120)
-  endif()
-endif()
--- a/paddle/fluid/inference/api/onnxruntime_predictor_tester.cc
+++ b/paddle/fluid/inference/api/onnxruntime_predictor_tester.cc
@@ -25,9 +25,9 @@
 #include "paddle/fluid/inference/api/onnxruntime_predictor.h"
 #include "paddle/fluid/inference/api/paddle_api.h"
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/io_utils.h"
 #include "paddle/phi/backends/cpu/cpu_info.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_string(dirname, "", "dirname to tests.");

--- a/paddle/fluid/inference/tests/api/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/api/CMakeLists.txt
-# In Windows, c_api test link must link both 2 shared to avoid symbols redefinition,
-# in Linux, c_api test cant do like this or graph_to_program register more than once.
-# Both Windows and Linux can only use paddle_inference_c, but this will increase size
-# of build folder by 30G.
-if(WIN32)
-  set(INFERENCE_C_EXTRA_DEPS paddle_inference_shared paddle_inference_c_shared)
-else()
-  set(INFERENCE_C_EXTRA_DEPS paddle_inference_shared paddle_inference_c)
-endif()
-function(download_data install_dir data_file check_sum)
-  string(REGEX MATCH "[^/\\]+$" file_name ${data_file})
-  if(NOT EXISTS ${install_dir}/${file_name})
-    inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}
-                                      ${data_file} ${check_sum})
-  endif()
-endfunction()
-function(download_data_without_verify install_dir data_file)
-  string(REGEX MATCH "[^/\\]+$" file_name ${data_file})
-  if(NOT EXISTS ${install_dir}/${file_name})
-    inference_download_and_uncompress_without_verify(
-      ${install_dir} ${INFERENCE_URL} ${data_file})
-  endif()
-endfunction()
-function(download_int8_data install_dir data_file check_sum)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}/int8
-                                      ${data_file} ${check_sum})
-  endif()
-endfunction()
-function(download_int8_data_without_verify install_dir data_file)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress_without_verify(
-      ${install_dir} ${INFERENCE_URL}/int8 ${data_file})
-  endif()
-endfunction()
-function(download_bfloat16_data install_dir data_file check_sum)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}/bfloat16
-                                      ${data_file} ${check_sum})
-  endif()
-endfunction()
-function(download_bfloat16_data_without_verify install_dir data_file)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress_without_verify(
-      ${install_dir} ${INFERENCE_URL}/bfloat16 ${data_file})
-  endif()
-endfunction()
-function(download_GRU_data install_dir data_file check_sum)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}/gru
-                                      ${data_file} ${check_sum})
-  endif()
-endfunction()
-function(download_GRU_data_without_verify install_dir data_file)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress_without_verify(
-      ${install_dir} ${INFERENCE_URL}/gru ${data_file})
-  endif()
-endfunction()
-function(download_quant_data install_dir data_file check_sum)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress(
-      ${install_dir} ${INFERENCE_URL}/int8/QAT_models ${data_file} ${check_sum})
-  endif()
-endfunction()
-function(download_quant_data_without_verify install_dir data_file)
-  if(NOT EXISTS ${install_dir}/${data_file})
-    inference_download_and_uncompress_without_verify(
-      ${install_dir} ${INFERENCE_URL}/int8/QAT_models ${data_file})
-  endif()
-endfunction()
-function(download_model_and_data install_dir model_name model_check_sum
-         data_name data_check_sum)
-  download_data(${install_dir} ${model_name} ${model_check_sum})
-  download_data(${install_dir} ${data_name} ${data_check_sum})
-endfunction()
-function(download_model_and_data_without_verify install_dir model_name
-         data_name)
-  download_data_without_verify(${install_dir} ${model_name})
-  download_data_without_verify(${install_dir} ${data_name})
-endfunction()
-function(download_result install_dir result_name check_sum)
-  download_data(${install_dir} ${result_name} ${check_sum})
-endfunction()
-function(download_result_without_verify install_dir result_name)
-  download_data_without_verify(${install_dir} ${result_name})
-endfunction()
-function(inference_analysis_api_test target install_dir filename)
-  inference_analysis_test(
-    ${target}
-    SRCS
-    ${filename}
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${install_dir}/model
-    --infer_data=${install_dir}/data.txt
-    --refer_result=${install_dir}/result.txt)
-endfunction()
-function(inference_analysis_api_int8_test target install_dir filename)
-  inference_analysis_test(
-    ${target}
-    SRCS
-    ${filename}
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${install_dir}/model
-    --infer_data=${install_dir}/data.txt
-    --refer_result=${install_dir}/result.txt
-    --accuracy=0.8
-    --batch_size=5
-    --enable_int8_ptq=true)
-endfunction()
-function(inference_multiple_models_analysis_api_test target install_dir
-         filename)
-  inference_analysis_test(
-    ${target}
-    SRCS
-    ${filename}
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${install_dir}/mobilenet_v2_models/1
-    --infer_model2=${install_dir}/mobilenet_v2_models/xx
-    --infer_model3=${install_dir}/mobilenet_v2_models/3)
-endfunction()
-function(inference_analysis_api_test_build TARGET_NAME filename)
-  inference_analysis_test_build(${TARGET_NAME} SRCS ${filename} EXTRA_DEPS
-                                paddle_inference_shared)
-endfunction()
-function(inference_analysis_api_int8_test_run TARGET_NAME test_binary model_dir
-         data_path)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${model_dir}/model
-    --infer_data=${data_path}
-    --warmup_batch_size=${WARMUP_BATCH_SIZE}
-    --batch_size=50
-    --enable_int8_ptq=true
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --iterations=2)
-endfunction()
-function(inference_analysis_api_int8_test_run_custom_warmup_batch_size
-         TARGET_NAME test_binary model_dir data_path warmup_batch_size)
-  set(WARMUP_BATCH_SIZE ${warmup_batch_size})
-  inference_analysis_api_int8_test_run(${TARGET_NAME} ${test_binary}
-                                       ${model_dir} ${data_path})
-endfunction()
-function(inference_analysis_api_bfloat16_test_run TARGET_NAME test_binary
-         model_dir data_path)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${model_dir}/model
-    --infer_data=${data_path}
-    --batch_size=50
-    --enable_bf16=true
-    --paddle_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --iterations=2)
-endfunction()
-function(inference_analysis_api_object_dection_int8_test_run TARGET_NAME
-         test_binary model_dir data_path)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${model_dir}/model
-    --infer_data=${data_path}
-    --warmup_batch_size=10
-    --batch_size=300
-    --enable_int8_ptq=true
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --iterations=1)
-endfunction()
-function(inference_analysis_api_test_with_fake_data_build TARGET_NAME filename)
-  inference_analysis_test_build(${TARGET_NAME} SRCS ${filename} EXTRA_DEPS
-                                paddle_inference_shared)
-endfunction()
-function(inference_analysis_api_test_with_fake_data_run TARGET_NAME test_binary
-         model_dir disable_fc)
-  inference_analysis_test_run(
-    ${TARGET_NAME} COMMAND ${test_binary} ARGS --infer_model=${model_dir}/model
-    --disable_mkldnn_fc=${disable_fc})
-endfunction()
-function(
-  inference_analysis_api_quant_test_run
-  TARGET_NAME
-  test_binary
-  fp32_model_dir
-  int8_model_dir
-  data_path
-  enable_int8_qat)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --fp32_model=${fp32_model_dir}
-    --int8_model=${int8_model_dir}
-    --infer_data=${data_path}
-    --batch_size=50
-    --enable_int8_qat=${enable_int8_qat}
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --with_accuracy_layer=false
-    --iterations=2)
-endfunction()
-function(inference_analysis_api_lexical_test_run TARGET_NAME test_binary
-         infer_model data_path)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${infer_model}
-    --infer_data=${data_path}
-    --batch_size=50
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --with_accuracy_layer=true
-    --use_analysis=true
-    --iterations=2)
-endfunction()
-function(inference_analysis_api_lexical_bfloat16_test_run TARGET_NAME
-         test_binary infer_model data_path)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${infer_model}
-    --infer_data=${data_path}
-    --batch_size=50
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --with_accuracy_layer=true
-    --use_analysis=true
-    --enable_bf16=true
-    --iterations=2)
-endfunction()
-function(
-  inference_analysis_api_lexical_int8_test_run
-  TARGET_NAME
-  test_binary
-  infer_model
-  data_path
-  enable_int8_ptq
-  enable_int8_qat
-  fuse_multi_gru)
-  inference_analysis_test_run(
-    ${TARGET_NAME}
-    COMMAND
-    ${test_binary}
-    ARGS
-    --infer_model=${infer_model}
-    --infer_data=${data_path}
-    --batch_size=100
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --with_accuracy_layer=true
-    --use_analysis=true
-    --enable_int8_ptq=${enable_int8_ptq}
-    --enable_int8_qat=${enable_int8_qat}
-    --quantized_accuracy=0.015
-    --fuse_multi_gru=${fuse_multi_gru}
-    --iterations=4)
-endfunction()
-function(preprocess_data2bin_test_run target py_script_source data_dir
-         output_file)
-  py_test(${target}
-          SRCS ${CMAKE_CURRENT_SOURCE_DIR}/${py_script_source} ARGS
-               --data_dir=${data_dir} --output_file=${output_file} --local)
-endfunction()
-if(NOT APPLE AND WITH_MKLML)
-  # RNN1
-  set(RNN1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/rnn1")
-  download_model_and_data_without_verify(
-    ${RNN1_INSTALL_DIR} "rnn1/model.tar.gz" "rnn1/data.txt.tar.gz")
-  inference_analysis_api_test(test_analyzer_rnn1 ${RNN1_INSTALL_DIR}
-                              analyzer_rnn1_tester.cc)
-  # seq_pool1
-  set(SEQ_POOL1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/seq_pool")
-  download_model_and_data_without_verify(
-    ${SEQ_POOL1_INSTALL_DIR} "seq_pool1_model_.tar.gz"
-    "seq_pool1_data.txt.tar.gz")
-  inference_analysis_api_test(
-    test_analyzer_seq_pool1_compare_determine ${SEQ_POOL1_INSTALL_DIR}
-    analyzer_seq_pool1_compare_determine_tester.cc)
-  inference_analysis_api_test(test_analyzer_seq_pool1 ${SEQ_POOL1_INSTALL_DIR}
-                              analyzer_seq_pool1_compare_tester.cc)
-  inference_analysis_api_test(
-    test_analyzer_seq_pool1_fuse_compare_zero_copy ${SEQ_POOL1_INSTALL_DIR}
-    analyzer_seq_pool1_fuse_compare_zero_copy_tester.cc)
-  inference_analysis_api_test(
-    test_analyzer_seq_pool1_fuse_statis ${SEQ_POOL1_INSTALL_DIR}
-    analyzer_seq_pool1_fuse_statis_tester.cc)
-  inference_analysis_api_test(
-    test_analyzer_seq_pool1_profile ${SEQ_POOL1_INSTALL_DIR}
-    analyzer_seq_pool1_profile_tester.cc)
-  if(NOT WIN32)
-    set_tests_properties(test_analyzer_seq_pool1_compare_determine
-                         PROPERTIES TIMEOUT 120)
-    set_tests_properties(test_analyzer_seq_pool1 PROPERTIES TIMEOUT 120)
-    set_tests_properties(test_analyzer_seq_pool1_fuse_compare_zero_copy
-                         PROPERTIES TIMEOUT 120)
-    set_tests_properties(test_analyzer_seq_pool1_fuse_statis PROPERTIES TIMEOUT
-                                                                        120)
-    set_tests_properties(test_analyzer_seq_pool1_profile PROPERTIES TIMEOUT 120)
-  endif()
-else()
-  # TODO: fix this test on MACOS and OPENBLAS, the reason is that
-  # fusion_seqexpand_concat_fc_op is not supported on MACOS and OPENBLAS
-  message(
-    WARNING
-      "These tests has been disabled in OSX or WITH_MKL=OFF before being fixed: \n test_analyzer_rnn1"
-  )
-  message(
-    WARNING
-      "These tests has been disabled in OSX or WITH_MKL=OFF before being fixed: \n test_analyzer_seq_pool1"
-  )
-endif()
-# RNN2
-set(RNN2_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/rnn2")
-download_model_and_data_without_verify(${RNN2_INSTALL_DIR} "rnn2_model.tar.gz"
-                                       "rnn2_data.txt.tar.gz")
-inference_analysis_api_test(test_analyzer_rnn2 ${RNN2_INSTALL_DIR}
-                            analyzer_rnn2_tester.cc)
-# TODO(luotao, Superjom) Disable DAM test, temporarily fix
-# https://github.com/PaddlePaddle/Paddle/issues/15032#issuecomment-455990914.
-# After inference framework refactor, will reopen it.
-# normal DAM
-set(DAM_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/dam")
-download_model_and_data_without_verify(${DAM_INSTALL_DIR} "DAM_model.tar.gz"
-                                       "DAM_data.txt.tar.gz")
-#inference_analysis_api_test(test_analyzer_dam ${DAM_INSTALL_DIR} analyzer_dam_tester.cc EXTRA_DEPS legacy_allocator)
-# small DAM
-set(DAM_SMALL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/small_dam")
-download_model_and_data_without_verify(
-  ${DAM_SMALL_INSTALL_DIR} "dam_small_model.tar.gz" "dam_small_data.txt.tar.gz")
-inference_analysis_test(
-  test_analyzer_small_dam
-  SRCS
-  analyzer_dam_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${DAM_SMALL_INSTALL_DIR}/model
-  --infer_data=${DAM_SMALL_INSTALL_DIR}/data.txt)
-#save model
-inference_analysis_api_test(test_analyzer_save_model ${DAM_SMALL_INSTALL_DIR}
-                            analyzer_save_model_tester.cc)
-# chinese_ner
-set(CHINESE_NER_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/chinese_ner")
-download_model_and_data_without_verify(
-  ${CHINESE_NER_INSTALL_DIR} "chinese_ner_model.tar.gz"
-  "chinese_ner-data.txt.tar.gz")
-inference_analysis_api_test(test_analyzer_ner ${CHINESE_NER_INSTALL_DIR}
-                            analyzer_ner_tester.cc)
-# lac
-set(LAC_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/lac")
-download_model_and_data(
-  ${LAC_INSTALL_DIR} "lac_model.tar.gz" 419ca6eb85f57a01bfe173591910aec5
-  "lac_data.txt.tar.gz" 9983539cd6b34fbdc411e43422776bfd)
-inference_analysis_api_test(test_analyzer_lac ${LAC_INSTALL_DIR}
-                            analyzer_lac_tester.cc)
-# Pyramid DNN
-set(PYRAMID_DNN_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/pyramid_dnn")
-download_model_and_data_without_verify(
-  ${PYRAMID_DNN_INSTALL_DIR} "PyramidDNN_model.tar.gz"
-  "PyramidDNN_data.txt.tar.gz")
-inference_analysis_api_test(
-  test_analyzer_pyramid_dnn ${PYRAMID_DNN_INSTALL_DIR}
-  analyzer_pyramid_dnn_tester.cc)
-# Ernie
-set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie")
-download_model_and_data(
-  ${ERNIE_INSTALL_DIR} "Ernie_model.tar.gz" aa59192dd41ed377f9f168e3a1309fa6
-  "Ernie_data.txt.tar.gz" 5396e63548edad7ca561e7e26a9476d1)
-download_result(${ERNIE_INSTALL_DIR} "Ernie_result.txt.tar.gz"
-                73beea65abda2edb61c1662cd3180c62)
-if(WITH_GPU)
-  inference_analysis_api_test(test_analyzer_ernie ${ERNIE_INSTALL_DIR}
-                              analyzer_ernie_tester.cc)
-  inference_analysis_api_test(gpu_ernie_half_test ${ERNIE_INSTALL_DIR}
-                              gpu_ernie_half_test.cc)
-  set_tests_properties(gpu_ernie_half_test PROPERTIES TIMEOUT 60)
-endif()
-inference_analysis_api_int8_test(test_analyzer_ernie_int8 ${ERNIE_INSTALL_DIR}
-                                 analyzer_ernie_int8_tester.cc)
-# Ernie large
-set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie_Large")
-download_model_and_data(
-  ${ERNIE_INSTALL_DIR} "Ernie_large_model.tar.gz"
-  af7715245ed32cc77374625d4c80f7ef "Ernie_large_data.txt.tar.gz"
-  edb2113eec93783cad56ed76d47ba57f)
-download_result(${ERNIE_INSTALL_DIR} "Ernie_large_result.txt.tar.gz"
-                1facda98eef1085dc9d435ebf3f23a73)
-inference_analysis_test(
-  test_analyzer_ernie_large
-  SRCS
-  analyzer_ernie_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${ERNIE_INSTALL_DIR}/model
-  --infer_data=${ERNIE_INSTALL_DIR}/data.txt
-  --refer_result=${ERNIE_INSTALL_DIR}/result.txt
-  --ernie_large=true)
-if(NOT WIN32
-   AND NOT APPLE
-   AND TEST test_analyzer_ernie_large)
-  set_tests_properties(test_analyzer_ernie_large PROPERTIES TIMEOUT 150 LABELS
-                                                            "RUN_TYPE=NIGHTLY")
-endif()
-if(WIN32 AND TEST test_analyzer_ernie_large)
-  set_tests_properties(test_analyzer_ernie_large PROPERTIES TIMEOUT 200)
-endif()
-# text_classification
-set(TEXT_CLASSIFICATION_INSTALL_DIR
-    "${INFERENCE_DEMO_INSTALL_DIR}/text_classification")
-download_model_and_data(
-  ${TEXT_CLASSIFICATION_INSTALL_DIR} "text-classification-Senta.tar.gz"
-  3f0f440313ca50e26184e65ffd5809ab "text_classification_data.txt.tar.gz"
-  36ae620020cc3377f45ed330dd36238f)
-inference_analysis_api_test(
-  test_analyzer_text_classification ${TEXT_CLASSIFICATION_INSTALL_DIR}
-  analyzer_text_classification_tester.cc)
-# seq_conv1
-set(SEQ_CONV1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/seq_conv1")
-download_model_and_data_without_verify(
-  ${SEQ_CONV1_INSTALL_DIR} "seq_conv1_model.tar.gz" "seq_conv1_data.txt.tar.gz")
-inference_analysis_api_test(test_analyzer_seq_conv1 ${SEQ_CONV1_INSTALL_DIR}
-                            analyzer_seq_conv1_tester.cc)
-# transformer, the dataset only works on batch_size=8 now
-set(TRANSFORMER_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/transformer")
-download_model_and_data_without_verify(
-  ${TRANSFORMER_INSTALL_DIR} "temp/transformer_model.tar.gz"
-  "temp/transformer_data.txt.tar.gz")
-inference_analysis_test(
-  test_analyzer_transformer
-  SRCS
-  analyzer_transformer_compare_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${TRANSFORMER_INSTALL_DIR}/model
-  --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
-  --batch_size=8
-  --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
-inference_analysis_test(
-  test_analyzer_transformer_fuse
-  SRCS
-  analyzer_transformer_fuse_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${TRANSFORMER_INSTALL_DIR}/model
-  --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
-  --batch_size=8
-  --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
-inference_analysis_test(
-  test_analyzer_transformer_profile
-  SRCS
-  analyzer_transformer_profile_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${TRANSFORMER_INSTALL_DIR}/model
-  --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
-  --batch_size=8
-  --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
-# VIT-OCR
-set(VIT_OCR_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/vit")
-if(NOT EXISTS ${VIT_OCR_INSTALL_DIR}/vit_ocr.tgz)
-  inference_download_and_uncompress_without_verify(
-    ${VIT_OCR_INSTALL_DIR} ${INFERENCE_URL} "ocr/vit_ocr.tgz")
-endif()
-inference_analysis_test(
-  test_analyzer_vit_ocr
-  SRCS
-  analyzer_vit_ocr_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${VIT_OCR_INSTALL_DIR}/vit_ocr/model
-  --infer_data=${VIT_OCR_INSTALL_DIR}/vit_ocr/datavit.txt)
-# ocr
-set(OCR_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/ocr")
-if(NOT EXISTS ${OCR_INSTALL_DIR}/ocr.tar.gz)
-  inference_download_and_uncompress_without_verify(
-    ${OCR_INSTALL_DIR} "http://paddlemodels.bj.bcebos.com/"
-    "inference-vis-demos/ocr.tar.gz")
-endif()
-inference_analysis_api_test(test_analyzer_ocr ${OCR_INSTALL_DIR}
-                            analyzer_vis_tester.cc)
-# densebox
-set(DENSEBOX_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/densebox")
-download_data_without_verify(${DENSEBOX_INSTALL_DIR} "densebox.tar.gz")
-inference_analysis_test(
-  test_analyzer_detect_functional_mkldnn
-  SRCS
-  analyzer_detect_functional_mkldnn_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${DENSEBOX_INSTALL_DIR}/model
-  --infer_data=${DENSEBOX_INSTALL_DIR}/detect_input_50.txt
-  --infer_shape=${DENSEBOX_INSTALL_DIR}/shape_50.txt)
-# mobilenet with transpose op
-set(MOBILENET_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/mobilenet")
-if(NOT EXISTS ${MOBILENET_INSTALL_DIR}/mobilenet.tar.gz)
-  inference_download_and_uncompress_without_verify(
-    ${MOBILENET_INSTALL_DIR} "http://paddlemodels.bj.bcebos.com/"
-    "inference-vis-demos/mobilenet.tar.gz")
-endif()
-inference_analysis_api_test(test_analyzer_mobilenet_transpose
-                            ${MOBILENET_INSTALL_DIR} analyzer_vis_tester.cc)
-### Image classification tests with fake data
-set(IMG_CLASS_TEST_APP "test_analyzer_image_classification")
-set(IMG_CLASS_TEST_APP_SRC "analyzer_image_classification_tester.cc")
-# build test binary to be used in subsequent tests
-inference_analysis_api_test_with_fake_data_build(${IMG_CLASS_TEST_APP}
-                                                 ${IMG_CLASS_TEST_APP_SRC})
-# googlenet
-set(GOOGLENET_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/googlenet")
-download_data_without_verify(${GOOGLENET_MODEL_DIR} "googlenet.tar.gz")
-inference_analysis_api_test_with_fake_data_run(
-  test_analyzer_googlenet ${IMG_CLASS_TEST_APP} ${GOOGLENET_MODEL_DIR} false)
-# resnet50
-set(RESNET50_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/resnet50")
-download_data_without_verify(${RESNET50_MODEL_DIR} "resnet50_model.tar.gz")
-inference_analysis_api_test_with_fake_data_run(
-  test_analyzer_resnet50 ${IMG_CLASS_TEST_APP} ${RESNET50_MODEL_DIR} true)
-if(WIN32)
-  set_tests_properties(test_analyzer_resnet50 PROPERTIES TIMEOUT 200)
-endif()
-# mobilenet with depthwise_conv op
-set(MOBILENET_MODEL_DIR
-    "${INFERENCE_DEMO_INSTALL_DIR}/mobilenet_depthwise_conv")
-download_data_without_verify(${MOBILENET_MODEL_DIR} "mobilenet_model.tar.gz")
-inference_analysis_api_test_with_fake_data_run(
-  test_analyzer_mobilenet_depthwise_conv ${IMG_CLASS_TEST_APP}
-  ${MOBILENET_MODEL_DIR} false)
-if(WITH_MKLDNN)
-  ### INT8 tests
-  set(INT8_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/int8v2")
-  ## Image classification models
-  # ImageNet small dataset
-  # It may be already downloaded for Quant & INT8 unit tests
-  set(IMAGENET_DATA_ARCHIVE "imagenet_val_100_tail.tar.gz")
-  set(IMAGENET_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/imagenet")
-  set(IMAGENET_DATA_PATH "${IMAGENET_DATA_DIR}/data.bin")
-  download_int8_data_without_verify(${IMAGENET_DATA_DIR}
-                                    ${IMAGENET_DATA_ARCHIVE})
-  # build test binary to be used in subsequent tests
-  set(INT8_IMG_CLASS_TEST_APP "test_analyzer_int8_image_classification")
-  set(INT8_IMG_CLASS_TEST_APP_SRC
-      "analyzer_int8_image_classification_tester.cc")
-  inference_analysis_api_test_build(${INT8_IMG_CLASS_TEST_APP}
-                                    ${INT8_IMG_CLASS_TEST_APP_SRC})
-  # resnet50 int8
-  set(INT8_RESNET50_MODEL_DIR "${INT8_DATA_DIR}/resnet50")
-  download_int8_data_without_verify(${INT8_RESNET50_MODEL_DIR}
-                                    "resnet50_int8_model.tar.gz")
-  inference_analysis_api_int8_test_run(
-    test_analyzer_int8_resnet50 ${INT8_IMG_CLASS_TEST_APP}
-    ${INT8_RESNET50_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # mobilenetv1 int8
-  set(INT8_MOBILENETV1_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv1")
-  download_int8_data_without_verify(${INT8_MOBILENETV1_MODEL_DIR}
-                                    "mobilenetv1_int8_model.tar.gz")
-  inference_analysis_api_int8_test_run(
-    test_analyzer_int8_mobilenetv1 ${INT8_IMG_CLASS_TEST_APP}
-    ${INT8_MOBILENETV1_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # mobilenetv2 int8
-  set(INT8_MOBILENETV2_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv2")
-  download_int8_data_without_verify(${INT8_MOBILENETV2_MODEL_DIR}
-                                    "mobilenet_v2_int8_model.tar.gz")
-  inference_analysis_api_int8_test_run(
-    test_analyzer_int8_mobilenetv2 ${INT8_IMG_CLASS_TEST_APP}
-    ${INT8_MOBILENETV2_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # resnet101 int8
-  set(INT8_RESNET101_MODEL_DIR "${INT8_DATA_DIR}/resnet101")
-  download_int8_data_without_verify(${INT8_RESNET101_MODEL_DIR}
-                                    "Res101_int8_model.tar.gz")
-  #   inference_analysis_api_int8_test_run(test_analyzer_int8_resnet101 ${INT8_IMG_CLASS_TEST_APP} ${INT8_RESNET101_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # vgg16 int8
-  set(INT8_VGG16_MODEL_DIR "${INT8_DATA_DIR}/vgg16")
-  download_int8_data_without_verify(${INT8_VGG16_MODEL_DIR}
-                                    "VGG16_int8_model.tar.gz")
-  #  inference_analysis_api_int8_test_run(test_analyzer_int8_vgg16 ${INT8_IMG_CLASS_TEST_APP} ${INT8_VGG16_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # vgg19 int8
-  set(INT8_VGG19_MODEL_DIR "${INT8_DATA_DIR}/vgg19")
-  download_int8_data_without_verify(${INT8_VGG19_MODEL_DIR}
-                                    "VGG19_int8_model.tar.gz")
-  #   inference_analysis_api_int8_test_run(test_analyzer_int8_vgg19 ${INT8_IMG_CLASS_TEST_APP} ${INT8_VGG19_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # googlenet int8
-  set(INT8_GOOGLENET_MODEL_DIR "${INT8_DATA_DIR}/googlenet")
-  download_int8_data_without_verify(${INT8_GOOGLENET_MODEL_DIR}
-                                    "GoogleNet_int8_model.tar.gz")
-  inference_analysis_api_int8_test_run_custom_warmup_batch_size(
-    test_analyzer_int8_googlenet ${INT8_IMG_CLASS_TEST_APP}
-    ${INT8_GOOGLENET_MODEL_DIR} ${IMAGENET_DATA_PATH} 10)
-  # mobilenetv3_large_x1_0 int8
-  set(INT8_MOBILENETV3_LARGE_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv3_large")
-  set(INT8_MOBILENETV3_FILE_NAME "MobileNetV3_large_x1_0_infer.tar")
-  if(NOT EXISTS
-     ${INT8_MOBILENETV3_LARGE_MODEL_DIR}/${INT8_MOBILENETV3_FILE_NAME})
-    inference_download_and_uncompress_without_verify(
-      ${INT8_MOBILENETV3_LARGE_MODEL_DIR}
-      "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/"
-      ${INT8_MOBILENETV3_FILE_NAME})
-  endif()
-  inference_analysis_test_run(
-    test_analyzer_int8_mobilenetv3_large
-    COMMAND
-    ${INT8_IMG_CLASS_TEST_APP}
-    ARGS
-    --infer_model=${INT8_MOBILENETV3_LARGE_MODEL_DIR}/MobileNetV3_large_x1_0_infer
-    --infer_data=${IMAGENET_DATA_PATH}
-    --warmup_batch_size=50
-    --batch_size=1
-    --enable_int8_ptq=true
-    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --iterations=100
-    --with_accuracy_layer=false)
-  ### BFLOAT16 tests
-  # build test binary to be used in subsequent tests
-  set(BF16_IMG_CLASS_TEST_APP "test_analyzer_bfloat16_image_classification")
-  set(BF16_IMG_CLASS_TEST_APP_SRC
-      "analyzer_bfloat16_image_classification_tester.cc")
-  inference_analysis_api_test_build(${BF16_IMG_CLASS_TEST_APP}
-                                    ${BF16_IMG_CLASS_TEST_APP_SRC})
-  # resnet50 bfloat16
-  inference_analysis_api_bfloat16_test_run(
-    test_analyzer_bfloat16_resnet50 ${BF16_IMG_CLASS_TEST_APP}
-    ${INT8_RESNET50_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # googlenet bfloat16
-  inference_analysis_api_bfloat16_test_run(
-    test_analyzer_bfloat16_googlenet ${BF16_IMG_CLASS_TEST_APP}
-    ${INT8_GOOGLENET_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # mobilenetv1 bfloat16
-  inference_analysis_api_bfloat16_test_run(
-    test_analyzer_bfloat16_mobilenetv1 ${BF16_IMG_CLASS_TEST_APP}
-    ${INT8_MOBILENETV1_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # mobilenetv2 bfloat16
-  inference_analysis_api_bfloat16_test_run(
-    test_analyzer_bfloat16_mobilenetv2 ${BF16_IMG_CLASS_TEST_APP}
-    ${INT8_MOBILENETV2_MODEL_DIR} ${IMAGENET_DATA_PATH})
-  # mobilenetv3_large
-  inference_analysis_test_run(
-    test_analyzer_bfloat16_mobilenetv3_large
-    COMMAND
-    ${BF16_IMG_CLASS_TEST_APP}
-    ARGS
-    --infer_model=${INT8_MOBILENETV3_LARGE_MODEL_DIR}/MobileNetV3_large_x1_0_infer
-    --infer_data=${IMAGENET_DATA_PATH}
-    --batch_size=1
-    --enable_bf16=true
-    --paddle_num_threads=${CPU_NUM_THREADS_ON_CI}
-    --iterations=100
-    --with_accuracy_layer=false)
-  ### Object detection models
-  set(PASCALVOC_DATA_PATH "${INT8_DATA_DIR}/pascalvoc_val_head_300.bin")
-  set(INT8_OBJ_DETECT_TEST_APP "test_analyzer_int8_object_detection")
-  set(INT8_OBJ_DETECT_TEST_APP_SRC "analyzer_int8_object_detection_tester.cc")
-  # download dataset if necessary
-  download_int8_data_without_verify(${INT8_DATA_DIR}
-                                    "pascalvoc_val_head_300.tar.gz")
-  # build test binary to be used in subsequent tests
-  inference_analysis_api_test_build(${INT8_OBJ_DETECT_TEST_APP}
-                                    ${INT8_OBJ_DETECT_TEST_APP_SRC})
-  # mobilenet-ssd int8
-  set(INT8_MOBILENET_SSD_MODEL_DIR "${INT8_DATA_DIR}/mobilenet-ssd")
-  download_int8_data_without_verify(${INT8_MOBILENET_SSD_MODEL_DIR}
-                                    "mobilenet_ssd_int8_model.tar.gz")
-  inference_analysis_api_object_dection_int8_test_run(
-    test_analyzer_int8_mobilenet_ssd ${INT8_OBJ_DETECT_TEST_APP}
-    ${INT8_MOBILENET_SSD_MODEL_DIR} ${PASCALVOC_DATA_PATH})
-  ### Lexcial analysis GRU model
-  set(GRU_PATH "${INFERENCE_DEMO_INSTALL_DIR}/gru")
-  download_gru_data_without_verify("${GRU_PATH}" "GRU_eval_data.tar.gz")
-  download_gru_data_without_verify("${GRU_PATH}" "GRU_eval_model_v2.tar.gz")
-  set(GRU_DATA_PATH "${GRU_PATH}/GRU_eval_data.bin")
-  set(GRU_MODEL_PATH "${GRU_PATH}/GRU_eval_model_v2")
-  set(LEXICAL_TEST_APP "test_analyzer_lexical_analysis")
-  set(LEXICAL_TEST_APP_SRC "analyzer_lexical_analysis_gru_tester.cc")
-  # build test binary to be used in subsequent tests
-  inference_analysis_api_test_build(${LEXICAL_TEST_APP} ${LEXICAL_TEST_APP_SRC})
-  # run lexcial analysis test
-  inference_analysis_api_lexical_test_run(
-    test_analyzer_lexical_gru ${LEXICAL_TEST_APP} ${GRU_MODEL_PATH}
-    ${GRU_DATA_PATH})
-  # run bfloat16 lexical analysis test
-  inference_analysis_api_lexical_bfloat16_test_run(
-    test_analyzer_lexical_gru_bfloat16 ${LEXICAL_TEST_APP} ${GRU_MODEL_PATH}
-    ${GRU_DATA_PATH})
-  # run post-training quantization lexical analysis test
-  inference_analysis_api_lexical_int8_test_run(
-    test_analyzer_lexical_gru_int8
-    ${LEXICAL_TEST_APP}
-    ${GRU_MODEL_PATH}
-    ${GRU_DATA_PATH}
-    true # enable_int8_ptq
-    false # enable_int8_qat
-    false) # fuse_multi_gru
-  # run post-training quantization lexical analysis test with multi_gru fuse
-  inference_analysis_api_lexical_int8_test_run(
-    test_analyzer_lexical_gru_int8_multi_gru
-    ${LEXICAL_TEST_APP}
-    ${GRU_MODEL_PATH}
-    ${GRU_DATA_PATH}
-    true # enable_int8_ptq
-    false # enable_int8_qat
-    true) # fuse_multi_gru
-  # run qat gru test
-  set(QAT_GRU_MODEL_ARCHIVE "GRU_quant_acc.tar.gz")
-  set(QAT_GRU_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/quant/GRU_quant2")
-  download_quant_data(${QAT_GRU_MODEL_DIR} ${QAT_GRU_MODEL_ARCHIVE}
-                      cf207f8076dcfb8b74d8b6bdddf9090c)
-  inference_analysis_api_lexical_int8_test_run(
-    test_analyzer_lexical_gru_qat_int8
-    ${LEXICAL_TEST_APP}
-    "${QAT_GRU_MODEL_DIR}/GRU_quant_acc"
-    ${GRU_DATA_PATH}
-    false # enable_int8_ptq
-    true # enable_int8_qat
-    false) # fuse_multi_gru
-  ### optimized FP32 vs. Quant INT8 tests
-  set(QUANT_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/quant")
-  set(QUANT_IMG_CLASS_TEST_APP "test_analyzer_quant_image_classification")
-  set(QUANT_IMG_CLASS_TEST_APP_SRC
-      "analyzer_quant_image_classification_tester.cc")
-  # build test binary to be used in subsequent tests
-  inference_analysis_api_test_build(${QUANT_IMG_CLASS_TEST_APP}
-                                    ${QUANT_IMG_CLASS_TEST_APP_SRC})
-  # MobileNetV1 FP32 vs. Quant INT8
-  # The FP32 model should already be downloaded for slim Quant unit tests on Linux
-  set(QUANT2_MobileNetV1_MODEL_DIR "${QUANT_DATA_DIR}/MobileNetV1_quant2")
-  set(QUANT2_INT8_MobileNetV1_MODEL_DIR
-      "${QUANT_DATA_DIR}/MobileNetV1_quant2_int8")
-  if(NOT LINUX)
-    download_quant_data_without_verify(${QUANT2_MobileNetV1_MODEL_DIR}
-                                       "MobileNet_qat_perf.tar.gz")
-  endif()
-  download_quant_data_without_verify(${QUANT2_INT8_MobileNetV1_MODEL_DIR}
-                                     "MobileNet_qat_perf_int8.tar.gz")
-  inference_analysis_api_quant_test_run(
-    test_analyzer_quant_performance_benchmark
-    ${QUANT_IMG_CLASS_TEST_APP}
-    ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
-    ${QUANT2_INT8_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf_int8
-    ${IMAGENET_DATA_PATH}
-    false)
-  # Quant2 MobileNetV1
-  inference_analysis_api_quant_test_run(
-    test_analyzer_quant2_mobilenetv1_mkldnn
-    ${QUANT_IMG_CLASS_TEST_APP}
-    ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
-    ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
-    ${IMAGENET_DATA_PATH}
-    true)
-  # Quant2 ResNet50 with input/output scales in `fake_quantize_range_abs_max` operators and the `out_threshold` attributes,
-  # with weight scales in `fake_channel_wise_dequantize_max_abs` operators
-  set(QUANT2_RESNET50_CHANNELWISE_MODEL_DIR
-      "${QUANT_DATA_DIR}/ResNet50_quant2_channelwise")
-  set(QUANT2_RESNET50_CHANNELWISE_MODEL_ARCHIVE
-      "ResNet50_qat_channelwise.tar.gz")
-  if(NOT LINUX)
-    download_quant_data_without_verify(
-      ${QUANT2_RESNET50_CHANNELWISE_MODEL_DIR}
-      ${QUANT2_RESNET50_CHANNELWISE_MODEL_ARCHIVE})
-  endif()
-  set(QUANT2_RESNET50_MODEL
-      ${QUANT2_RESNET50_CHANNELWISE_MODEL_DIR}/ResNet50_qat_channelwise)
-  inference_analysis_api_quant_test_run(
-    test_analyzer_quant2_resnet50_channelwise_mkldnn
-    ${QUANT_IMG_CLASS_TEST_APP} ${QUANT2_RESNET50_MODEL}
-    ${QUANT2_RESNET50_MODEL} ${IMAGENET_DATA_PATH} true)
-  ### Other tests
-  # MKLDNN quantizer config
-  set(MKLDNN_QUANTIZER_CONFIG_TEST_APP "test_mkldnn_quantizer_config")
-  set(MKLDNN_QUANTIZER_CONFIG_TEST_APP_SRC "mkldnn_quantizer_config_tester.cc")
-  inference_analysis_api_test_build(${MKLDNN_QUANTIZER_CONFIG_TEST_APP}
-                                    ${MKLDNN_QUANTIZER_CONFIG_TEST_APP_SRC})
-  inference_analysis_test_run(test_mkldnn_quantizer_config COMMAND
-                              ${MKLDNN_QUANTIZER_CONFIG_TEST_APP})
-  # preprocess data2bin imagenet
-  download_int8_data_without_verify(${INT8_DATA_DIR} "imagenet_small.tar.gz")
-  set(IMAGENET_SMALL_DATA_DIR "${INT8_DATA_DIR}/imagenet_small")
-  set(IMAGENET_SMALL_OUTPUT_FILE "imagenet_small.bin")
-  preprocess_data2bin_test_run(
-    preprocess_local_imagenet "full_ILSVRC2012_val_preprocess.py"
-    ${IMAGENET_SMALL_DATA_DIR} ${IMAGENET_SMALL_OUTPUT_FILE})
-  # preprocess data2bin pascalvoc
-  download_int8_data_without_verify(${INT8_DATA_DIR} "pascalvoc_small.tar.gz")
-  set(PASCALVOC_SMALL_DATA_DIR "${INT8_DATA_DIR}/pascalvoc_small")
-  set(PASCALVOC_SMALL_OUTPUT_FILE "pascalvoc_small.bin")
-  preprocess_data2bin_test_run(
-    preprocess_local_pascalvoc "full_pascalvoc_test_preprocess.py"
-    ${PASCALVOC_SMALL_DATA_DIR} ${PASCALVOC_SMALL_OUTPUT_FILE})
-endif()
-# bert, max_len=20, embedding_dim=128
-set(BERT_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/bert_emb128")
-download_model_and_data_without_verify(
-  ${BERT_INSTALL_DIR} "bert_emb128_model.tar.gz" "bert_data_len20.txt.tar.gz")
-inference_analysis_api_test(test_analyzer_bert ${BERT_INSTALL_DIR}
-                            analyzer_bert_tester.cc)
-# multiple models prediction
-set(MMP_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/multi_model_prediction")
-download_data_without_verify(${MMP_INSTALL_DIR}
-                             PaddleInference/mobilenet_v2_models.tar.gz)
-inference_multiple_models_analysis_api_test(
-  test_analyzer_multi_model_prediction ${MMP_INSTALL_DIR}
-  analyzer_mmp_tester.cc)
-if(WITH_GPU AND TENSORRT_FOUND)
-  set(TRT_MODEL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/trt_models")
-  if(NOT EXISTS ${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models.tar.gz)
-    inference_download_and_uncompress(
-      ${TRT_MODEL_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
-      "trt_inference_test_models.tar.gz" 3dcccdc38b549b6b1b4089723757bd98)
-  endif()
-  set(TEST_SPLIT_CONVERTER_MODEL
-      "${TRT_MODEL_INSTALL_DIR}/trt_split_op_converter_test")
-  if(NOT EXISTS ${TEST_SPLIT_CONVERTER_MODEL}/split_converter.tgz)
-    inference_download_and_uncompress_without_verify(
-      ${TEST_SPLIT_CONVERTER_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "split_converter.tgz")
-  endif()
-  inference_analysis_test(
-    trt_mobilenet_test
-    SRCS
-    trt_mobilenet_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  inference_analysis_test(
-    trt_resnet50_test
-    SRCS
-    trt_resnet50_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  inference_analysis_test(
-    trt_resnext_test
-    SRCS
-    trt_resnext_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  inference_analysis_test(
-    trt_fc_prelu_test
-    SRCS
-    trt_fc_prelu_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  inference_analysis_test(
-    trt_cascade_rcnn_test
-    SRCS
-    trt_cascade_rcnn_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  inference_analysis_test(
-    trt_split_converter_test
-    SRCS
-    trt_split_converter_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TEST_SPLIT_CONVERTER_MODEL}/)
-  inference_analysis_test(
-    test_analyzer_capi_exp_gpu
-    SRCS
-    analyzer_capi_exp_gpu_tester.cc
-    EXTRA_DEPS
-    ${INFERENCE_C_EXTRA_DEPS}
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  if(WIN32)
-    target_link_libraries(test_analyzer_capi_exp_gpu paddle_inference_c_shared)
-  else()
-    target_link_libraries(test_analyzer_capi_exp_gpu paddle_inference_c)
-  endif()
-  inference_analysis_test(
-    test_analyzer_capi_exp_xpu
-    SRCS
-    analyzer_capi_exp_xpu_tester.cc
-    EXTRA_DEPS
-    ${INFERENCE_C_EXTRA_DEPS}
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
-  if(WIN32)
-    target_link_libraries(test_analyzer_capi_exp_xpu paddle_inference_c_shared)
-  else()
-    target_link_libraries(test_analyzer_capi_exp_xpu paddle_inference_c)
-  endif()
-  set(TRT_MODEL_QUANT_RESNET_DIR
-      "${INFERENCE_DEMO_INSTALL_DIR}/small_quant_model")
-  if(NOT EXISTS ${INFERENCE_DEMO_INSTALL_DIR}/small_quant_model.tgz)
-    inference_download_and_uncompress_without_verify(
-      ${INFERENCE_DEMO_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
-      "small_quant_model.tgz")
-  endif()
-  inference_analysis_test(
-    trt_quant_int8_test
-    SRCS
-    trt_quant_int8_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_QUANT_RESNET_DIR})
-  set(TRT_MODEL_QUANT_YOLOV3_DIR
-      "${INFERENCE_DEMO_INSTALL_DIR}/yolov3_r50_quant_aware")
-  if(NOT EXISTS ${INFERENCE_DEMO_INSTALL_DIR}/yolov3_r50_quant_aware.tgz)
-    inference_download_and_uncompress_without_verify(
-      ${INFERENCE_DEMO_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
-      "yolov3_r50_quant_aware.tgz")
-  endif()
-  inference_analysis_test(
-    trt_quant_int8_yolov3_r50_test
-    SRCS
-    trt_quant_int8_yolov3_r50_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_QUANT_YOLOV3_DIR})
-  set(TEST_TRT_DYNAMIC_MODEL2 "${TRT_MODEL_INSTALL_DIR}/complex_model_dynamic")
-  if(NOT EXISTS ${TEST_TRT_DYNAMIC_MODEL2}/complex_model_dynamic2.tar.gz)
-    inference_download_and_uncompress_without_verify(
-      ${TEST_TRT_DYNAMIC_MODEL2} ${INFERENCE_URL}/tensorrt_test
-      "complex_model_dynamic2.tar.gz")
-  endif()
-  set(TEST_TRT_DYNAMIC_MODEL
-      "${TRT_MODEL_INSTALL_DIR}/conv_bn_swish_split_gelu")
-  if(NOT EXISTS ${TEST_TRT_DYNAMIC_MODEL}/conv_bn_swish_split_gelu.tar.gz)
-    inference_download_and_uncompress(
-      ${TEST_TRT_DYNAMIC_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "conv_bn_swish_split_gelu.tar.gz" 2a5e8791e47b221b4f782151d76da9c6)
-  endif()
-  inference_analysis_test(
-    trt_dynamic_shape_test
-    SRCS
-    trt_dynamic_shape_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TRT_MODEL_INSTALL_DIR})
-  set(TEST_TRT_ERNIE_MODEL "${TRT_MODEL_INSTALL_DIR}/ernie_test")
-  if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4.tar.gz)
-    inference_download_and_uncompress(
-      ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "ernie_model_4.tar.gz" 5fa371efa75706becbaad79195d2ca68)
-  endif()
-  inference_analysis_test(
-    test_trt_dynamic_shape_ernie
-    SRCS
-    trt_dynamic_shape_ernie_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4)
-  set(TEST_TRT_TRANSFORMER_PRUNE_MODEL
-      "${TRT_MODEL_INSTALL_DIR}/transformer_prune")
-  if(NOT EXISTS ${TEST_TRT_TRANSFORMER_PRUNE_MODEL}/transformer_prune.tar.gz)
-    inference_download_and_uncompress(
-      ${TEST_TRT_TRANSFORMER_PRUNE_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "transformer_prune.tar.gz" 77b56dc73ff0cf44ddb1ce9ca0b0f471)
-  endif()
-  inference_analysis_test(
-    test_trt_dynamic_shape_transformer_prune
-    SRCS
-    trt_dynamic_shape_transformer_prune_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TEST_TRT_TRANSFORMER_PRUNE_MODEL}/transformer_prune)
-  if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4_unserialized.tgz)
-    inference_download_and_uncompress(
-      ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "ernie_model_4_unserialized.tgz" 833d73fc6a7f7e1ee4a1fd6419209e55)
-  endif()
-  inference_analysis_test(
-    test_trt_dynamic_shape_ernie_ser_deser
-    SRCS
-    trt_dynamic_shape_ernie_serialize_deserialize_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4_unserialized)
-  if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4_fp16_unserialized.tgz)
-    inference_download_and_uncompress(
-      ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
-      "ernie_model_4_fp16_unserialized.tgz" c5ff2d0cad79953ffbf2b8b9e2fae6e4)
-  endif()
-  inference_analysis_test(
-    test_trt_dynamic_shape_ernie_fp16_ser_deser
-    SRCS
-    trt_dynamic_shape_ernie_fp16_serialize_deserialize_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4_fp16_unserialized)
-endif()
-set(LITE_MODEL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/lite")
-download_data_without_verify(${LITE_MODEL_INSTALL_DIR} "mul_model_fp32.tgz")
-inference_analysis_test(
-  lite_mul_model_test
-  SRCS
-  lite_mul_model_test.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${LITE_MODEL_INSTALL_DIR})
-inference_analysis_test(
-  lite_resnet50_test
-  SRCS
-  lite_resnet50_test.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${RESNET50_MODEL_DIR})
-inference_analysis_test(
-  test_analyzer_capi_exp
-  SRCS
-  analyzer_capi_exp_tester.cc
-  EXTRA_DEPS
-  ${INFERENCE_C_EXTRA_DEPS}
-  ARGS
-  --infer_model=${RESNET50_MODEL_DIR}/model)
-if(WIN32)
-  target_link_libraries(test_analyzer_capi_exp paddle_inference_c_shared)
-else()
-  target_link_libraries(test_analyzer_capi_exp paddle_inference_c)
-endif()
-inference_analysis_test(
-  test_analyzer_capi_exp_pd_config
-  SRCS
-  analyzer_capi_exp_pd_config_tester.cc
-  EXTRA_DEPS
-  ${INFERENCE_C_EXTRA_DEPS}
-  ARGS
-  --infer_model=${MOBILENET_INSTALL_DIR}/model)
-if(WIN32)
-  target_link_libraries(test_analyzer_capi_exp_pd_config
-                        paddle_inference_c_shared)
-else()
-  target_link_libraries(test_analyzer_capi_exp_pd_config paddle_inference_c)
-endif()
-inference_analysis_test(
-  test_analyzer_capi_exp_pd_tensor
-  SRCS
-  analyzer_capi_exp_pd_tensor_tester.cc
-  EXTRA_DEPS
-  ${INFERENCE_C_EXTRA_DEPS}
-  ARGS
-  --infer_model=${MOBILENET_INSTALL_DIR}/model)
-if(WIN32)
-  target_link_libraries(test_analyzer_capi_exp_pd_tensor
-                        paddle_inference_c_shared)
-else()
-  target_link_libraries(test_analyzer_capi_exp_pd_tensor paddle_inference_c)
-endif()
-if(NOT APPLE AND NOT WIN32)
-  inference_analysis_test(
-    test_analyzer_capi_exp_pd_threads
-    SRCS
-    analyzer_capi_exp_pd_threads_tester.cc
-    EXTRA_DEPS
-    ${INFERENCE_C_EXTRA_DEPS}
-    ARGS
-    --infer_model=${MOBILENET_INSTALL_DIR}/model)
-  if(WIN32)
-    target_link_libraries(test_analyzer_capi_exp_pd_threads
-                          paddle_inference_c_shared)
-  else()
-    target_link_libraries(test_analyzer_capi_exp_pd_threads paddle_inference_c)
-  endif()
-endif()
-inference_analysis_test(
-  test_analyzer_zerocopytensor_tensor
-  SRCS
-  analyzer_zerocopy_tensor_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${OCR_INSTALL_DIR}/model)
-if(WITH_DISTRIBUTE AND WITH_PSCORE)
-  inference_analysis_test(
-    test_analyzer_dist_model
-    SRCS
-    analyzer_dist_model_tester.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${OCR_INSTALL_DIR}/model)
-endif()
-if(WITH_DISTRIBUTE
-   AND WITH_PSCORE
-   AND WITH_XPU
-   AND WITH_XPU_BKCL)
-  inference_analysis_test(
-    test_analyzer_dist_model_xpu
-    SRCS
-    analyzer_dist_model_xpu_tester.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${OCR_INSTALL_DIR}/model)
-endif()
-inference_analysis_test(
-  test_analyzer_paddletensor_tensor
-  SRCS
-  analyzer_paddle_tensor_tester.cc
-  EXTRA_DEPS
-  paddle_inference_shared
-  ARGS
-  --infer_model=${OCR_INSTALL_DIR}/model
-  --infer_data=${OCR_INSTALL_DIR}/data.txt
-  --refer_result=${OCR_INSTALL_DIR}/result.txt)
-if(WITH_MKLDNN)
-  inference_analysis_test(
-    test_analyzer_capi_exp_int
-    SRCS
-    analyzer_capi_exp_int_tester.cc
-    EXTRA_DEPS
-    ${INFERENCE_C_EXTRA_DEPS}
-    ARGS
-    --infer_model=${INT8_DATA_DIR}/resnet50/model)
-  if(WIN32)
-    target_link_libraries(test_analyzer_capi_exp_int paddle_inference_c_shared)
-  else()
-    target_link_libraries(test_analyzer_capi_exp_int paddle_inference_c)
-  endif()
-endif()
-inference_analysis_test(
-  test_analyzer_capi_exp_ner
-  SRCS
-  analyzer_capi_exp_ner_tester.cc
-  EXTRA_DEPS
-  ${INFERENCE_C_EXTRA_DEPS}
-  ARGS
-  --infer_model=${CHINESE_NER_INSTALL_DIR}/model)
-if(WIN32)
-  target_link_libraries(test_analyzer_capi_exp_ner paddle_inference_c_shared)
-else()
-  target_link_libraries(test_analyzer_capi_exp_ner paddle_inference_c)
-endif()
-if(WITH_GPU)
-  inference_analysis_test(
-    paddle_infer_api_test
-    SRCS
-    paddle_infer_api_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${RESNET50_MODEL_DIR})
-  inference_analysis_test(
-    paddle_infer_api_copy_tensor_tester
-    SRCS
-    paddle_infer_api_copy_tensor_tester.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${RESNET50_MODEL_DIR})
-  set_tests_properties(paddle_infer_api_copy_tensor_tester PROPERTIES TIMEOUT
-                                                                      30)
-endif()
-cc_test(
-  paddle_infer_api_errors_test
-  SRCS paddle_infer_api_errors_tester.cc
-  DEPS paddle_inference_api)
-if(WITH_GPU AND TENSORRT_FOUND)
-  set_tests_properties(trt_resnext_test PROPERTIES TIMEOUT 300)
-  set_tests_properties(trt_quant_int8_yolov3_r50_test PROPERTIES TIMEOUT 400)
-  set_tests_properties(trt_resnet50_test PROPERTIES TIMEOUT 300)
-  set_tests_properties(trt_cascade_rcnn_test PROPERTIES TIMEOUT 300)
-  set_tests_properties(test_trt_dynamic_shape_ernie_ser_deser PROPERTIES TIMEOUT
-                                                                         300)
-  set_tests_properties(test_trt_dynamic_shape_ernie_fp16_ser_deser
-                       PROPERTIES TIMEOUT 300)
-  set_tests_properties(test_trt_dynamic_shape_ernie PROPERTIES TIMEOUT 300)
-endif()
-if(WITH_MKLDNN)
-  set_tests_properties(test_analyzer_int8_resnet50 PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_int8_mobilenet_ssd PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_quant_performance_benchmark
-                       PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_int8_mobilenetv2 PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_int8_mobilenetv1 PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_int8_mobilenetv3_large PROPERTIES TIMEOUT
-                                                                       120)
-  set_tests_properties(test_analyzer_quant2_mobilenetv1_mkldnn
-                       PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_quant2_resnet50_channelwise_mkldnn
-                       PROPERTIES TIMEOUT 120)
-endif()
-set_tests_properties(lite_resnet50_test PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_mobilenet_transpose PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_resnet50 PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_ner PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_ernie_int8 PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_googlenet PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_small_dam PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_transformer PROPERTIES TIMEOUT 120)
-set_tests_properties(test_analyzer_mobilenet_depthwise_conv PROPERTIES TIMEOUT
-                                                                       120)
-if(WITH_GPU)
-  set_tests_properties(test_analyzer_bert PROPERTIES TIMEOUT 120)
-  set_tests_properties(test_analyzer_ernie PROPERTIES TIMEOUT 120)
-endif()
-if(WITH_GPU AND TENSORRT_FOUND)
-  set_tests_properties(trt_mobilenet_test PROPERTIES TIMEOUT 120)
-  if(WITH_MKLDNN)
-    set_tests_properties(test_analyzer_bfloat16_resnet50 PROPERTIES TIMEOUT 120)
-  endif()
-endif()
-if(ON_INFER OR WITH_GPU)
-  set_tests_properties(test_analyzer_transformer_profile PROPERTIES TIMEOUT 120)
-endif()
-if(WITH_IPU)
-  #word2vec sample
-  set(WORD2VEC_INSTALL_DIR
-      "${INFERENCE_DEMO_INSTALL_DIR}/word2vec/word2vec.inference.model")
-  inference_analysis_test(
-    ipu_word2vec_sample
-    SRCS
-    ipu_word2vec_sample.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${WORD2VEC_INSTALL_DIR})
-  # ERNIE
-  set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie")
-  inference_analysis_api_test(ipu_ernie_test ${ERNIE_INSTALL_DIR}
-                              ipu_ernie_test.cc ARGS --warmup=true --repeat=10)
-  inference_analysis_api_test(
-    ipu_ernie_fp16_test ${ERNIE_INSTALL_DIR} ipu_ernie_fp16_test.cc ARGS
-    --warmup=true --repeat=10)
-  # Resnet50
-  set(RESNET50_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/resnet50")
-  inference_analysis_test(
-    ipu_resnet50_test
-    SRCS
-    ipu_resnet50_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${RESNET50_MODEL_DIR}
-    --warmup=true
-    --repeat=10)
-  inference_analysis_test(
-    ipu_resnet50_fp16_test
-    SRCS
-    ipu_resnet50_fp16_test.cc
-    EXTRA_DEPS
-    paddle_inference_shared
-    ARGS
-    --infer_model=${RESNET50_MODEL_DIR}
-    --warmup=true
-    --repeat=10)
-  # Only support Resnet50 and Ernie currently
-  inference_analysis_api_test(
-    ipu_multi_model_profile
-    SRCS
-    ipu_multi_model_profile.cc
-    ARGS
-    --model_name="Resnet50"
-    --infer_model=${RESNET50_MODEL_DIR}
-    --warmup=true
-    --repeat=10)
-endif()
--- a/paddle/scripts/paddle_build.sh
+++ b/paddle/scripts/paddle_build.sh
@@ -3008,7 +3008,7 @@ EOF
    echo "ipipe_log_param_Demo_Ci_Tests_Total_Time: $[ $demo_ci_endTime_s - $demo_ci_startTime_s ]s" >> ${PADDLE_ROOT}/build/build_summary.txt
    infer_ut_startTime_s=`date +%s`
-    cd ${PADDLE_ROOT}/paddle/fluid/inference/tests/infer_ut
+    cd ${PADDLE_ROOT}/test/cpp/inference/infer_ut
    ./run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF} ${INFERENCE_DEMO_INSTALL_DIR} \
             ${TENSORRT_ROOT_DIR:-/usr} ${WITH_ONNXRUNTIME:-ON}
    TEST_EXIT_CODE=$?

--- a/test/cpp/CMakeLists.txt
+++ b/test/cpp/CMakeLists.txt
@@ -4,3 +4,4 @@ add_subdirectory(new_executor)
 add_subdirectory(prim)
 add_subdirectory(imperative)
 add_subdirectory(ir)
+add_subdirectory(inference)
--- a/test/cpp/inference/CMakeLists.txt
+++ b/test/cpp/inference/CMakeLists.txt
+if(WITH_TESTING)
+  include(test.cmake) # some generic cmake function for inference
+endif()
+add_subdirectory(analysis)
+add_subdirectory(api)
--- a/test/cpp/inference/analysis/CMakeLists.txt
+++ b/test/cpp/inference/analysis/CMakeLists.txt
+function(inference_analysis_test_build TARGET)
+  if(WITH_TESTING)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS EXTRA_DEPS)
+    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
+                          "${multiValueArgs}" ${ARGN})
+    inference_base_test_build(
+      ${TARGET}
+      SRCS
+      ${analysis_test_SRCS}
+      DEPS
+      ${analysis_test_EXTRA_DEPS}
+      analysis
+      pass
+      ${GLOB_PASS_LIB})
+  endif()
+endfunction()
+function(inference_analysis_test_run TARGET)
+  if(WITH_TESTING)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs COMMAND ARGS)
+    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
+                          "${multiValueArgs}" ${ARGN})
+    inference_base_test_run(${TARGET} COMMAND ${analysis_test_COMMAND} ARGS
+                            ${analysis_test_ARGS})
+    set_tests_properties(${TARGET} PROPERTIES LABELS "RUN_TYPE=INFER")
+  endif()
+endfunction()
+function(inference_analysis_test TARGET)
+  if(WITH_TESTING)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS ARGS EXTRA_DEPS)
+    cmake_parse_arguments(analysis_test "${options}" "${oneValueArgs}"
+                          "${multiValueArgs}" ${ARGN})
+    inference_base_test_build(
+      ${TARGET}
+      SRCS
+      ${analysis_test_SRCS}
+      DEPS
+      ${analysis_test_EXTRA_DEPS}
+      analysis
+      pass
+      ${GLOB_PASS_LIB})
+    inference_base_test_run(${TARGET} COMMAND ${TARGET} ARGS
+                            ${analysis_test_ARGS})
+    set_tests_properties(${TARGET} PROPERTIES LABELS "RUN_TYPE=INFER")
+  endif()
+endfunction()
+if(NOT APPLE AND NOT WIN32)
+  inference_analysis_test(
+    test_analyzer
+    SRCS
+    analyzer_tester.cc
+    EXTRA_DEPS
+    reset_tensor_array
+    paddle_inference_shared
+    ARGS
+    --inference_model_dir=${WORD2VEC_MODEL_DIR})
+elseif(WIN32)
+  inference_analysis_test(
+    test_analyzer
+    SRCS
+    analyzer_tester.cc
+    EXTRA_DEPS
+    reset_tensor_array
+    paddle_inference_api
+    ARGS
+    --inference_model_dir=${WORD2VEC_MODEL_DIR})
+  if(WITH_ONNXRUNTIME AND WIN32)
+    # Copy onnxruntime for some c++ test in Windows, since the test will
+    # be build only in CI, so suppose the generator in Windows is Ninja.
+    copy_onnx(test_analyzer)
+  endif()
+endif()
--- a/paddle/fluid/inference/analysis/analyzer_tester.cc
+++ b/paddle/fluid/inference/analysis/analyzer_tester.cc
--- a/test/cpp/inference/api/CMakeLists.txt
+++ b/test/cpp/inference/api/CMakeLists.txt
+# In Windows, c_api test link must link both 2 shared to avoid symbols redefinition,
+# in Linux, c_api test cant do like this or graph_to_program register more than once.
+# Both Windows and Linux can only use paddle_inference_c, but this will increase size
+# of build folder by 30G.
+cc_test(
+  test_paddle_inference_api
+  SRCS api_tester.cc
+  DEPS paddle_inference_api)
+cc_test(
+  inference_api_helper_test
+  SRCS helper_test.cc
+  DEPS paddle_inference_api)
+if(WITH_ONNXRUNTIME AND WIN32)
+  # Copy onnxruntime for some c++ test in Windows, since the test will
+  # be build only in CI, so suppose the generator in Windows is Ninja.
+  copy_onnx(test_paddle_inference_api)
+endif()
+if(WITH_TESTING AND WITH_INFERENCE_API_TEST)
+  if(WIN32)
+    set(INFERENCE_C_EXTRA_DEPS paddle_inference_shared
+                               paddle_inference_c_shared)
+  else()
+    set(INFERENCE_C_EXTRA_DEPS paddle_inference_shared paddle_inference_c)
+  endif()
+  function(download_data install_dir data_file check_sum)
+    string(REGEX MATCH "[^/\\]+$" file_name ${data_file})
+    if(NOT EXISTS ${install_dir}/${file_name})
+      inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}
+                                        ${data_file} ${check_sum})
+    endif()
+  endfunction()
+  function(download_data_without_verify install_dir data_file)
+    string(REGEX MATCH "[^/\\]+$" file_name ${data_file})
+    if(NOT EXISTS ${install_dir}/${file_name})
+      inference_download_and_uncompress_without_verify(
+        ${install_dir} ${INFERENCE_URL} ${data_file})
+    endif()
+  endfunction()
+  function(download_int8_data install_dir data_file check_sum)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}/int8
+                                        ${data_file} ${check_sum})
+    endif()
+  endfunction()
+  function(download_int8_data_without_verify install_dir data_file)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress_without_verify(
+        ${install_dir} ${INFERENCE_URL}/int8 ${data_file})
+    endif()
+  endfunction()
+  function(download_bfloat16_data install_dir data_file check_sum)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress(
+        ${install_dir} ${INFERENCE_URL}/bfloat16 ${data_file} ${check_sum})
+    endif()
+  endfunction()
+  function(download_bfloat16_data_without_verify install_dir data_file)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress_without_verify(
+        ${install_dir} ${INFERENCE_URL}/bfloat16 ${data_file})
+    endif()
+  endfunction()
+  function(download_GRU_data install_dir data_file check_sum)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress(${install_dir} ${INFERENCE_URL}/gru
+                                        ${data_file} ${check_sum})
+    endif()
+  endfunction()
+  function(download_GRU_data_without_verify install_dir data_file)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress_without_verify(
+        ${install_dir} ${INFERENCE_URL}/gru ${data_file})
+    endif()
+  endfunction()
+  function(download_quant_data install_dir data_file check_sum)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress(
+        ${install_dir} ${INFERENCE_URL}/int8/QAT_models ${data_file}
+        ${check_sum})
+    endif()
+  endfunction()
+  function(download_quant_data_without_verify install_dir data_file)
+    if(NOT EXISTS ${install_dir}/${data_file})
+      inference_download_and_uncompress_without_verify(
+        ${install_dir} ${INFERENCE_URL}/int8/QAT_models ${data_file})
+    endif()
+  endfunction()
+  function(download_model_and_data install_dir model_name model_check_sum
+           data_name data_check_sum)
+    download_data(${install_dir} ${model_name} ${model_check_sum})
+    download_data(${install_dir} ${data_name} ${data_check_sum})
+  endfunction()
+  function(download_model_and_data_without_verify install_dir model_name
+           data_name)
+    download_data_without_verify(${install_dir} ${model_name})
+    download_data_without_verify(${install_dir} ${data_name})
+  endfunction()
+  function(download_result install_dir result_name check_sum)
+    download_data(${install_dir} ${result_name} ${check_sum})
+  endfunction()
+  function(download_result_without_verify install_dir result_name)
+    download_data_without_verify(${install_dir} ${result_name})
+  endfunction()
+  function(inference_analysis_api_test target install_dir filename)
+    inference_analysis_test(
+      ${target}
+      SRCS
+      ${filename}
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${install_dir}/model
+      --infer_data=${install_dir}/data.txt
+      --refer_result=${install_dir}/result.txt)
+  endfunction()
+  function(inference_analysis_api_int8_test target install_dir filename)
+    inference_analysis_test(
+      ${target}
+      SRCS
+      ${filename}
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${install_dir}/model
+      --infer_data=${install_dir}/data.txt
+      --refer_result=${install_dir}/result.txt
+      --accuracy=0.8
+      --batch_size=5
+      --enable_int8_ptq=true)
+  endfunction()
+  function(inference_multiple_models_analysis_api_test target install_dir
+           filename)
+    inference_analysis_test(
+      ${target}
+      SRCS
+      ${filename}
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${install_dir}/mobilenet_v2_models/1
+      --infer_model2=${install_dir}/mobilenet_v2_models/xx
+      --infer_model3=${install_dir}/mobilenet_v2_models/3)
+  endfunction()
+  function(inference_analysis_api_test_build TARGET_NAME filename)
+    inference_analysis_test_build(${TARGET_NAME} SRCS ${filename} EXTRA_DEPS
+                                  paddle_inference_shared)
+  endfunction()
+  function(inference_analysis_api_int8_test_run TARGET_NAME test_binary
+           model_dir data_path)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${model_dir}/model
+      --infer_data=${data_path}
+      --warmup_batch_size=${WARMUP_BATCH_SIZE}
+      --batch_size=50
+      --enable_int8_ptq=true
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --iterations=2)
+  endfunction()
+  function(inference_analysis_api_int8_test_run_custom_warmup_batch_size
+           TARGET_NAME test_binary model_dir data_path warmup_batch_size)
+    set(WARMUP_BATCH_SIZE ${warmup_batch_size})
+    inference_analysis_api_int8_test_run(${TARGET_NAME} ${test_binary}
+                                         ${model_dir} ${data_path})
+  endfunction()
+  function(inference_analysis_api_bfloat16_test_run TARGET_NAME test_binary
+           model_dir data_path)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${model_dir}/model
+      --infer_data=${data_path}
+      --batch_size=50
+      --enable_bf16=true
+      --paddle_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --iterations=2)
+  endfunction()
+  function(inference_analysis_api_object_dection_int8_test_run TARGET_NAME
+           test_binary model_dir data_path)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${model_dir}/model
+      --infer_data=${data_path}
+      --warmup_batch_size=10
+      --batch_size=300
+      --enable_int8_ptq=true
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --iterations=1)
+  endfunction()
+  function(inference_analysis_api_test_with_fake_data_build TARGET_NAME
+           filename)
+    inference_analysis_test_build(${TARGET_NAME} SRCS ${filename} EXTRA_DEPS
+                                  paddle_inference_shared)
+  endfunction()
+  function(inference_analysis_api_test_with_fake_data_run TARGET_NAME
+           test_binary model_dir disable_fc)
+    inference_analysis_test_run(
+      ${TARGET_NAME} COMMAND ${test_binary} ARGS
+      --infer_model=${model_dir}/model --disable_mkldnn_fc=${disable_fc})
+  endfunction()
+  function(
+    inference_analysis_api_quant_test_run
+    TARGET_NAME
+    test_binary
+    fp32_model_dir
+    int8_model_dir
+    data_path
+    enable_int8_qat)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --fp32_model=${fp32_model_dir}
+      --int8_model=${int8_model_dir}
+      --infer_data=${data_path}
+      --batch_size=50
+      --enable_int8_qat=${enable_int8_qat}
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --with_accuracy_layer=false
+      --iterations=2)
+  endfunction()
+  function(inference_analysis_api_lexical_test_run TARGET_NAME test_binary
+           infer_model data_path)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${infer_model}
+      --infer_data=${data_path}
+      --batch_size=50
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --with_accuracy_layer=true
+      --use_analysis=true
+      --iterations=2)
+  endfunction()
+  function(inference_analysis_api_lexical_bfloat16_test_run TARGET_NAME
+           test_binary infer_model data_path)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${infer_model}
+      --infer_data=${data_path}
+      --batch_size=50
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --with_accuracy_layer=true
+      --use_analysis=true
+      --enable_bf16=true
+      --iterations=2)
+  endfunction()
+  function(
+    inference_analysis_api_lexical_int8_test_run
+    TARGET_NAME
+    test_binary
+    infer_model
+    data_path
+    enable_int8_ptq
+    enable_int8_qat
+    fuse_multi_gru)
+    inference_analysis_test_run(
+      ${TARGET_NAME}
+      COMMAND
+      ${test_binary}
+      ARGS
+      --infer_model=${infer_model}
+      --infer_data=${data_path}
+      --batch_size=100
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --with_accuracy_layer=true
+      --use_analysis=true
+      --enable_int8_ptq=${enable_int8_ptq}
+      --enable_int8_qat=${enable_int8_qat}
+      --quantized_accuracy=0.015
+      --fuse_multi_gru=${fuse_multi_gru}
+      --iterations=4)
+  endfunction()
+  function(preprocess_data2bin_test_run target py_script_source data_dir
+           output_file)
+    py_test(${target}
+            SRCS ${CMAKE_CURRENT_SOURCE_DIR}/${py_script_source} ARGS
+                 --data_dir=${data_dir} --output_file=${output_file} --local)
+  endfunction()
+  if(NOT APPLE AND WITH_MKLML)
+    # RNN1
+    set(RNN1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/rnn1")
+    download_model_and_data_without_verify(
+      ${RNN1_INSTALL_DIR} "rnn1/model.tar.gz" "rnn1/data.txt.tar.gz")
+    inference_analysis_api_test(test_analyzer_rnn1 ${RNN1_INSTALL_DIR}
+                                analyzer_rnn1_tester.cc)
+    # seq_pool1
+    set(SEQ_POOL1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/seq_pool")
+    download_model_and_data_without_verify(
+      ${SEQ_POOL1_INSTALL_DIR} "seq_pool1_model_.tar.gz"
+      "seq_pool1_data.txt.tar.gz")
+    inference_analysis_api_test(
+      test_analyzer_seq_pool1_compare_determine ${SEQ_POOL1_INSTALL_DIR}
+      analyzer_seq_pool1_compare_determine_tester.cc)
+    inference_analysis_api_test(
+      test_analyzer_seq_pool1 ${SEQ_POOL1_INSTALL_DIR}
+      analyzer_seq_pool1_compare_tester.cc)
+    inference_analysis_api_test(
+      test_analyzer_seq_pool1_fuse_compare_zero_copy ${SEQ_POOL1_INSTALL_DIR}
+      analyzer_seq_pool1_fuse_compare_zero_copy_tester.cc)
+    inference_analysis_api_test(
+      test_analyzer_seq_pool1_fuse_statis ${SEQ_POOL1_INSTALL_DIR}
+      analyzer_seq_pool1_fuse_statis_tester.cc)
+    inference_analysis_api_test(
+      test_analyzer_seq_pool1_profile ${SEQ_POOL1_INSTALL_DIR}
+      analyzer_seq_pool1_profile_tester.cc)
+    if(NOT WIN32)
+      set_tests_properties(test_analyzer_seq_pool1_compare_determine
+                           PROPERTIES TIMEOUT 120)
+      set_tests_properties(test_analyzer_seq_pool1 PROPERTIES TIMEOUT 120)
+      set_tests_properties(test_analyzer_seq_pool1_fuse_compare_zero_copy
+                           PROPERTIES TIMEOUT 120)
+      set_tests_properties(test_analyzer_seq_pool1_fuse_statis
+                           PROPERTIES TIMEOUT 120)
+      set_tests_properties(test_analyzer_seq_pool1_profile PROPERTIES TIMEOUT
+                                                                      120)
+    endif()
+  else()
+    # TODO: fix this test on MACOS and OPENBLAS, the reason is that
+    # fusion_seqexpand_concat_fc_op is not supported on MACOS and OPENBLAS
+    message(
+      WARNING
+        "These tests has been disabled in OSX or WITH_MKL=OFF before being fixed: \n test_analyzer_rnn1"
+    )
+    message(
+      WARNING
+        "These tests has been disabled in OSX or WITH_MKL=OFF before being fixed: \n test_analyzer_seq_pool1"
+    )
+  endif()
+  # RNN2
+  set(RNN2_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/rnn2")
+  download_model_and_data_without_verify(
+    ${RNN2_INSTALL_DIR} "rnn2_model.tar.gz" "rnn2_data.txt.tar.gz")
+  inference_analysis_api_test(test_analyzer_rnn2 ${RNN2_INSTALL_DIR}
+                              analyzer_rnn2_tester.cc)
+  # TODO(luotao, Superjom) Disable DAM test, temporarily fix
+  # https://github.com/PaddlePaddle/Paddle/issues/15032#issuecomment-455990914.
+  # After inference framework refactor, will reopen it.
+  # normal DAM
+  set(DAM_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/dam")
+  download_model_and_data_without_verify(${DAM_INSTALL_DIR} "DAM_model.tar.gz"
+                                         "DAM_data.txt.tar.gz")
+  #inference_analysis_api_test(test_analyzer_dam ${DAM_INSTALL_DIR} analyzer_dam_tester.cc EXTRA_DEPS legacy_allocator)
+  # small DAM
+  set(DAM_SMALL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/small_dam")
+  download_model_and_data_without_verify(
+    ${DAM_SMALL_INSTALL_DIR} "dam_small_model.tar.gz"
+    "dam_small_data.txt.tar.gz")
+  inference_analysis_test(
+    test_analyzer_small_dam
+    SRCS
+    analyzer_dam_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${DAM_SMALL_INSTALL_DIR}/model
+    --infer_data=${DAM_SMALL_INSTALL_DIR}/data.txt)
+  #save model
+  inference_analysis_api_test(test_analyzer_save_model ${DAM_SMALL_INSTALL_DIR}
+                              analyzer_save_model_tester.cc)
+  # chinese_ner
+  set(CHINESE_NER_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/chinese_ner")
+  download_model_and_data_without_verify(
+    ${CHINESE_NER_INSTALL_DIR} "chinese_ner_model.tar.gz"
+    "chinese_ner-data.txt.tar.gz")
+  inference_analysis_api_test(test_analyzer_ner ${CHINESE_NER_INSTALL_DIR}
+                              analyzer_ner_tester.cc)
+  # lac
+  set(LAC_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/lac")
+  download_model_and_data(
+    ${LAC_INSTALL_DIR} "lac_model.tar.gz" 419ca6eb85f57a01bfe173591910aec5
+    "lac_data.txt.tar.gz" 9983539cd6b34fbdc411e43422776bfd)
+  inference_analysis_api_test(test_analyzer_lac ${LAC_INSTALL_DIR}
+                              analyzer_lac_tester.cc)
+  # Pyramid DNN
+  set(PYRAMID_DNN_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/pyramid_dnn")
+  download_model_and_data_without_verify(
+    ${PYRAMID_DNN_INSTALL_DIR} "PyramidDNN_model.tar.gz"
+    "PyramidDNN_data.txt.tar.gz")
+  inference_analysis_api_test(
+    test_analyzer_pyramid_dnn ${PYRAMID_DNN_INSTALL_DIR}
+    analyzer_pyramid_dnn_tester.cc)
+  # Ernie
+  set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie")
+  download_model_and_data(
+    ${ERNIE_INSTALL_DIR} "Ernie_model.tar.gz" aa59192dd41ed377f9f168e3a1309fa6
+    "Ernie_data.txt.tar.gz" 5396e63548edad7ca561e7e26a9476d1)
+  download_result(${ERNIE_INSTALL_DIR} "Ernie_result.txt.tar.gz"
+                  73beea65abda2edb61c1662cd3180c62)
+  if(WITH_GPU)
+    inference_analysis_api_test(test_analyzer_ernie ${ERNIE_INSTALL_DIR}
+                                analyzer_ernie_tester.cc)
+    inference_analysis_api_test(gpu_ernie_half_test ${ERNIE_INSTALL_DIR}
+                                gpu_ernie_half_test.cc)
+    set_tests_properties(gpu_ernie_half_test PROPERTIES TIMEOUT 60)
+  endif()
+  inference_analysis_api_int8_test(
+    test_analyzer_ernie_int8 ${ERNIE_INSTALL_DIR} analyzer_ernie_int8_tester.cc)
+  # Ernie large
+  set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie_Large")
+  download_model_and_data(
+    ${ERNIE_INSTALL_DIR} "Ernie_large_model.tar.gz"
+    af7715245ed32cc77374625d4c80f7ef "Ernie_large_data.txt.tar.gz"
+    edb2113eec93783cad56ed76d47ba57f)
+  download_result(${ERNIE_INSTALL_DIR} "Ernie_large_result.txt.tar.gz"
+                  1facda98eef1085dc9d435ebf3f23a73)
+  inference_analysis_test(
+    test_analyzer_ernie_large
+    SRCS
+    analyzer_ernie_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${ERNIE_INSTALL_DIR}/model
+    --infer_data=${ERNIE_INSTALL_DIR}/data.txt
+    --refer_result=${ERNIE_INSTALL_DIR}/result.txt
+    --ernie_large=true)
+  if(NOT WIN32
+     AND NOT APPLE
+     AND TEST test_analyzer_ernie_large)
+    set_tests_properties(test_analyzer_ernie_large
+                         PROPERTIES TIMEOUT 150 LABELS "RUN_TYPE=NIGHTLY")
+  endif()
+  if(WIN32 AND TEST test_analyzer_ernie_large)
+    set_tests_properties(test_analyzer_ernie_large PROPERTIES TIMEOUT 200)
+  endif()
+  # text_classification
+  set(TEXT_CLASSIFICATION_INSTALL_DIR
+      "${INFERENCE_DEMO_INSTALL_DIR}/text_classification")
+  download_model_and_data(
+    ${TEXT_CLASSIFICATION_INSTALL_DIR} "text-classification-Senta.tar.gz"
+    3f0f440313ca50e26184e65ffd5809ab "text_classification_data.txt.tar.gz"
+    36ae620020cc3377f45ed330dd36238f)
+  inference_analysis_api_test(
+    test_analyzer_text_classification ${TEXT_CLASSIFICATION_INSTALL_DIR}
+    analyzer_text_classification_tester.cc)
+  # seq_conv1
+  set(SEQ_CONV1_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/seq_conv1")
+  download_model_and_data_without_verify(
+    ${SEQ_CONV1_INSTALL_DIR} "seq_conv1_model.tar.gz"
+    "seq_conv1_data.txt.tar.gz")
+  inference_analysis_api_test(test_analyzer_seq_conv1 ${SEQ_CONV1_INSTALL_DIR}
+                              analyzer_seq_conv1_tester.cc)
+  # transformer, the dataset only works on batch_size=8 now
+  set(TRANSFORMER_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/transformer")
+  download_model_and_data_without_verify(
+    ${TRANSFORMER_INSTALL_DIR} "temp/transformer_model.tar.gz"
+    "temp/transformer_data.txt.tar.gz")
+  inference_analysis_test(
+    test_analyzer_transformer
+    SRCS
+    analyzer_transformer_compare_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${TRANSFORMER_INSTALL_DIR}/model
+    --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
+    --batch_size=8
+    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
+  inference_analysis_test(
+    test_analyzer_transformer_fuse
+    SRCS
+    analyzer_transformer_fuse_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${TRANSFORMER_INSTALL_DIR}/model
+    --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
+    --batch_size=8
+    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
+  inference_analysis_test(
+    test_analyzer_transformer_profile
+    SRCS
+    analyzer_transformer_profile_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${TRANSFORMER_INSTALL_DIR}/model
+    --infer_data=${TRANSFORMER_INSTALL_DIR}/data.txt
+    --batch_size=8
+    --cpu_num_threads=${CPU_NUM_THREADS_ON_CI})
+  # VIT-OCR
+  set(VIT_OCR_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/vit")
+  if(NOT EXISTS ${VIT_OCR_INSTALL_DIR}/vit_ocr.tgz)
+    inference_download_and_uncompress_without_verify(
+      ${VIT_OCR_INSTALL_DIR} ${INFERENCE_URL} "ocr/vit_ocr.tgz")
+  endif()
+  inference_analysis_test(
+    test_analyzer_vit_ocr
+    SRCS
+    analyzer_vit_ocr_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${VIT_OCR_INSTALL_DIR}/vit_ocr/model
+    --infer_data=${VIT_OCR_INSTALL_DIR}/vit_ocr/datavit.txt)
+  # ocr
+  set(OCR_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/ocr")
+  if(NOT EXISTS ${OCR_INSTALL_DIR}/ocr.tar.gz)
+    inference_download_and_uncompress_without_verify(
+      ${OCR_INSTALL_DIR} "http://paddlemodels.bj.bcebos.com/"
+      "inference-vis-demos/ocr.tar.gz")
+  endif()
+  inference_analysis_api_test(test_analyzer_ocr ${OCR_INSTALL_DIR}
+                              analyzer_vis_tester.cc)
+  # densebox
+  set(DENSEBOX_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/densebox")
+  download_data_without_verify(${DENSEBOX_INSTALL_DIR} "densebox.tar.gz")
+  inference_analysis_test(
+    test_analyzer_detect_functional_mkldnn
+    SRCS
+    analyzer_detect_functional_mkldnn_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${DENSEBOX_INSTALL_DIR}/model
+    --infer_data=${DENSEBOX_INSTALL_DIR}/detect_input_50.txt
+    --infer_shape=${DENSEBOX_INSTALL_DIR}/shape_50.txt)
+  # mobilenet with transpose op
+  set(MOBILENET_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/mobilenet")
+  if(NOT EXISTS ${MOBILENET_INSTALL_DIR}/mobilenet.tar.gz)
+    inference_download_and_uncompress_without_verify(
+      ${MOBILENET_INSTALL_DIR} "http://paddlemodels.bj.bcebos.com/"
+      "inference-vis-demos/mobilenet.tar.gz")
+  endif()
+  inference_analysis_api_test(test_analyzer_mobilenet_transpose
+                              ${MOBILENET_INSTALL_DIR} analyzer_vis_tester.cc)
+  ### Image classification tests with fake data
+  set(IMG_CLASS_TEST_APP "test_analyzer_image_classification")
+  set(IMG_CLASS_TEST_APP_SRC "analyzer_image_classification_tester.cc")
+  # build test binary to be used in subsequent tests
+  inference_analysis_api_test_with_fake_data_build(${IMG_CLASS_TEST_APP}
+                                                   ${IMG_CLASS_TEST_APP_SRC})
+  # googlenet
+  set(GOOGLENET_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/googlenet")
+  download_data_without_verify(${GOOGLENET_MODEL_DIR} "googlenet.tar.gz")
+  inference_analysis_api_test_with_fake_data_run(
+    test_analyzer_googlenet ${IMG_CLASS_TEST_APP} ${GOOGLENET_MODEL_DIR} false)
+  # resnet50
+  set(RESNET50_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/resnet50")
+  download_data_without_verify(${RESNET50_MODEL_DIR} "resnet50_model.tar.gz")
+  inference_analysis_api_test_with_fake_data_run(
+    test_analyzer_resnet50 ${IMG_CLASS_TEST_APP} ${RESNET50_MODEL_DIR} true)
+  if(WIN32)
+    set_tests_properties(test_analyzer_resnet50 PROPERTIES TIMEOUT 200)
+  endif()
+  # mobilenet with depthwise_conv op
+  set(MOBILENET_MODEL_DIR
+      "${INFERENCE_DEMO_INSTALL_DIR}/mobilenet_depthwise_conv")
+  download_data_without_verify(${MOBILENET_MODEL_DIR} "mobilenet_model.tar.gz")
+  inference_analysis_api_test_with_fake_data_run(
+    test_analyzer_mobilenet_depthwise_conv ${IMG_CLASS_TEST_APP}
+    ${MOBILENET_MODEL_DIR} false)
+  if(WITH_MKLDNN)
+    ### INT8 tests
+    set(INT8_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/int8v2")
+    ## Image classification models
+    # ImageNet small dataset
+    # It may be already downloaded for Quant & INT8 unit tests
+    set(IMAGENET_DATA_ARCHIVE "imagenet_val_100_tail.tar.gz")
+    set(IMAGENET_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/imagenet")
+    set(IMAGENET_DATA_PATH "${IMAGENET_DATA_DIR}/data.bin")
+    download_int8_data_without_verify(${IMAGENET_DATA_DIR}
+                                      ${IMAGENET_DATA_ARCHIVE})
+    # build test binary to be used in subsequent tests
+    set(INT8_IMG_CLASS_TEST_APP "test_analyzer_int8_image_classification")
+    set(INT8_IMG_CLASS_TEST_APP_SRC
+        "analyzer_int8_image_classification_tester.cc")
+    inference_analysis_api_test_build(${INT8_IMG_CLASS_TEST_APP}
+                                      ${INT8_IMG_CLASS_TEST_APP_SRC})
+    # resnet50 int8
+    set(INT8_RESNET50_MODEL_DIR "${INT8_DATA_DIR}/resnet50")
+    download_int8_data_without_verify(${INT8_RESNET50_MODEL_DIR}
+                                      "resnet50_int8_model.tar.gz")
+    inference_analysis_api_int8_test_run(
+      test_analyzer_int8_resnet50 ${INT8_IMG_CLASS_TEST_APP}
+      ${INT8_RESNET50_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # mobilenetv1 int8
+    set(INT8_MOBILENETV1_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv1")
+    download_int8_data_without_verify(${INT8_MOBILENETV1_MODEL_DIR}
+                                      "mobilenetv1_int8_model.tar.gz")
+    inference_analysis_api_int8_test_run(
+      test_analyzer_int8_mobilenetv1 ${INT8_IMG_CLASS_TEST_APP}
+      ${INT8_MOBILENETV1_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # mobilenetv2 int8
+    set(INT8_MOBILENETV2_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv2")
+    download_int8_data_without_verify(${INT8_MOBILENETV2_MODEL_DIR}
+                                      "mobilenet_v2_int8_model.tar.gz")
+    inference_analysis_api_int8_test_run(
+      test_analyzer_int8_mobilenetv2 ${INT8_IMG_CLASS_TEST_APP}
+      ${INT8_MOBILENETV2_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # resnet101 int8
+    set(INT8_RESNET101_MODEL_DIR "${INT8_DATA_DIR}/resnet101")
+    download_int8_data_without_verify(${INT8_RESNET101_MODEL_DIR}
+                                      "Res101_int8_model.tar.gz")
+    #   inference_analysis_api_int8_test_run(test_analyzer_int8_resnet101 ${INT8_IMG_CLASS_TEST_APP} ${INT8_RESNET101_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # vgg16 int8
+    set(INT8_VGG16_MODEL_DIR "${INT8_DATA_DIR}/vgg16")
+    download_int8_data_without_verify(${INT8_VGG16_MODEL_DIR}
+                                      "VGG16_int8_model.tar.gz")
+    #  inference_analysis_api_int8_test_run(test_analyzer_int8_vgg16 ${INT8_IMG_CLASS_TEST_APP} ${INT8_VGG16_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # vgg19 int8
+    set(INT8_VGG19_MODEL_DIR "${INT8_DATA_DIR}/vgg19")
+    download_int8_data_without_verify(${INT8_VGG19_MODEL_DIR}
+                                      "VGG19_int8_model.tar.gz")
+    #   inference_analysis_api_int8_test_run(test_analyzer_int8_vgg19 ${INT8_IMG_CLASS_TEST_APP} ${INT8_VGG19_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # googlenet int8
+    set(INT8_GOOGLENET_MODEL_DIR "${INT8_DATA_DIR}/googlenet")
+    download_int8_data_without_verify(${INT8_GOOGLENET_MODEL_DIR}
+                                      "GoogleNet_int8_model.tar.gz")
+    inference_analysis_api_int8_test_run_custom_warmup_batch_size(
+      test_analyzer_int8_googlenet ${INT8_IMG_CLASS_TEST_APP}
+      ${INT8_GOOGLENET_MODEL_DIR} ${IMAGENET_DATA_PATH} 10)
+    # mobilenetv3_large_x1_0 int8
+    set(INT8_MOBILENETV3_LARGE_MODEL_DIR "${INT8_DATA_DIR}/mobilenetv3_large")
+    set(INT8_MOBILENETV3_FILE_NAME "MobileNetV3_large_x1_0_infer.tar")
+    if(NOT EXISTS
+       ${INT8_MOBILENETV3_LARGE_MODEL_DIR}/${INT8_MOBILENETV3_FILE_NAME})
+      inference_download_and_uncompress_without_verify(
+        ${INT8_MOBILENETV3_LARGE_MODEL_DIR}
+        "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/"
+        ${INT8_MOBILENETV3_FILE_NAME})
+    endif()
+    inference_analysis_test_run(
+      test_analyzer_int8_mobilenetv3_large
+      COMMAND
+      ${INT8_IMG_CLASS_TEST_APP}
+      ARGS
+      --infer_model=${INT8_MOBILENETV3_LARGE_MODEL_DIR}/MobileNetV3_large_x1_0_infer
+      --infer_data=${IMAGENET_DATA_PATH}
+      --warmup_batch_size=50
+      --batch_size=1
+      --enable_int8_ptq=true
+      --cpu_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --iterations=100
+      --with_accuracy_layer=false)
+    ### BFLOAT16 tests
+    # build test binary to be used in subsequent tests
+    set(BF16_IMG_CLASS_TEST_APP "test_analyzer_bfloat16_image_classification")
+    set(BF16_IMG_CLASS_TEST_APP_SRC
+        "analyzer_bfloat16_image_classification_tester.cc")
+    inference_analysis_api_test_build(${BF16_IMG_CLASS_TEST_APP}
+                                      ${BF16_IMG_CLASS_TEST_APP_SRC})
+    # resnet50 bfloat16
+    inference_analysis_api_bfloat16_test_run(
+      test_analyzer_bfloat16_resnet50 ${BF16_IMG_CLASS_TEST_APP}
+      ${INT8_RESNET50_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # googlenet bfloat16
+    inference_analysis_api_bfloat16_test_run(
+      test_analyzer_bfloat16_googlenet ${BF16_IMG_CLASS_TEST_APP}
+      ${INT8_GOOGLENET_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # mobilenetv1 bfloat16
+    inference_analysis_api_bfloat16_test_run(
+      test_analyzer_bfloat16_mobilenetv1 ${BF16_IMG_CLASS_TEST_APP}
+      ${INT8_MOBILENETV1_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # mobilenetv2 bfloat16
+    inference_analysis_api_bfloat16_test_run(
+      test_analyzer_bfloat16_mobilenetv2 ${BF16_IMG_CLASS_TEST_APP}
+      ${INT8_MOBILENETV2_MODEL_DIR} ${IMAGENET_DATA_PATH})
+    # mobilenetv3_large
+    inference_analysis_test_run(
+      test_analyzer_bfloat16_mobilenetv3_large
+      COMMAND
+      ${BF16_IMG_CLASS_TEST_APP}
+      ARGS
+      --infer_model=${INT8_MOBILENETV3_LARGE_MODEL_DIR}/MobileNetV3_large_x1_0_infer
+      --infer_data=${IMAGENET_DATA_PATH}
+      --batch_size=1
+      --enable_bf16=true
+      --paddle_num_threads=${CPU_NUM_THREADS_ON_CI}
+      --iterations=100
+      --with_accuracy_layer=false)
+    ### Object detection models
+    set(PASCALVOC_DATA_PATH "${INT8_DATA_DIR}/pascalvoc_val_head_300.bin")
+    set(INT8_OBJ_DETECT_TEST_APP "test_analyzer_int8_object_detection")
+    set(INT8_OBJ_DETECT_TEST_APP_SRC "analyzer_int8_object_detection_tester.cc")
+    # download dataset if necessary
+    download_int8_data_without_verify(${INT8_DATA_DIR}
+                                      "pascalvoc_val_head_300.tar.gz")
+    # build test binary to be used in subsequent tests
+    inference_analysis_api_test_build(${INT8_OBJ_DETECT_TEST_APP}
+                                      ${INT8_OBJ_DETECT_TEST_APP_SRC})
+    # mobilenet-ssd int8
+    set(INT8_MOBILENET_SSD_MODEL_DIR "${INT8_DATA_DIR}/mobilenet-ssd")
+    download_int8_data_without_verify(${INT8_MOBILENET_SSD_MODEL_DIR}
+                                      "mobilenet_ssd_int8_model.tar.gz")
+    inference_analysis_api_object_dection_int8_test_run(
+      test_analyzer_int8_mobilenet_ssd ${INT8_OBJ_DETECT_TEST_APP}
+      ${INT8_MOBILENET_SSD_MODEL_DIR} ${PASCALVOC_DATA_PATH})
+    ### Lexcial analysis GRU model
+    set(GRU_PATH "${INFERENCE_DEMO_INSTALL_DIR}/gru")
+    download_gru_data_without_verify("${GRU_PATH}" "GRU_eval_data.tar.gz")
+    download_gru_data_without_verify("${GRU_PATH}" "GRU_eval_model_v2.tar.gz")
+    set(GRU_DATA_PATH "${GRU_PATH}/GRU_eval_data.bin")
+    set(GRU_MODEL_PATH "${GRU_PATH}/GRU_eval_model_v2")
+    set(LEXICAL_TEST_APP "test_analyzer_lexical_analysis")
+    set(LEXICAL_TEST_APP_SRC "analyzer_lexical_analysis_gru_tester.cc")
+    # build test binary to be used in subsequent tests
+    inference_analysis_api_test_build(${LEXICAL_TEST_APP}
+                                      ${LEXICAL_TEST_APP_SRC})
+    # run lexcial analysis test
+    inference_analysis_api_lexical_test_run(
+      test_analyzer_lexical_gru ${LEXICAL_TEST_APP} ${GRU_MODEL_PATH}
+      ${GRU_DATA_PATH})
+    # run bfloat16 lexical analysis test
+    inference_analysis_api_lexical_bfloat16_test_run(
+      test_analyzer_lexical_gru_bfloat16 ${LEXICAL_TEST_APP} ${GRU_MODEL_PATH}
+      ${GRU_DATA_PATH})
+    # run post-training quantization lexical analysis test
+    inference_analysis_api_lexical_int8_test_run(
+      test_analyzer_lexical_gru_int8
+      ${LEXICAL_TEST_APP}
+      ${GRU_MODEL_PATH}
+      ${GRU_DATA_PATH}
+      true # enable_int8_ptq
+      false # enable_int8_qat
+      false) # fuse_multi_gru
+    # run post-training quantization lexical analysis test with multi_gru fuse
+    inference_analysis_api_lexical_int8_test_run(
+      test_analyzer_lexical_gru_int8_multi_gru
+      ${LEXICAL_TEST_APP}
+      ${GRU_MODEL_PATH}
+      ${GRU_DATA_PATH}
+      true # enable_int8_ptq
+      false # enable_int8_qat
+      true) # fuse_multi_gru
+    # run qat gru test
+    set(QAT_GRU_MODEL_ARCHIVE "GRU_quant_acc.tar.gz")
+    set(QAT_GRU_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/quant/GRU_quant2")
+    download_quant_data(${QAT_GRU_MODEL_DIR} ${QAT_GRU_MODEL_ARCHIVE}
+                        cf207f8076dcfb8b74d8b6bdddf9090c)
+    inference_analysis_api_lexical_int8_test_run(
+      test_analyzer_lexical_gru_qat_int8
+      ${LEXICAL_TEST_APP}
+      "${QAT_GRU_MODEL_DIR}/GRU_quant_acc"
+      ${GRU_DATA_PATH}
+      false # enable_int8_ptq
+      true # enable_int8_qat
+      false) # fuse_multi_gru
+    ### optimized FP32 vs. Quant INT8 tests
+    set(QUANT_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/quant")
+    set(QUANT_IMG_CLASS_TEST_APP "test_analyzer_quant_image_classification")
+    set(QUANT_IMG_CLASS_TEST_APP_SRC
+        "analyzer_quant_image_classification_tester.cc")
+    # build test binary to be used in subsequent tests
+    inference_analysis_api_test_build(${QUANT_IMG_CLASS_TEST_APP}
+                                      ${QUANT_IMG_CLASS_TEST_APP_SRC})
+    # MobileNetV1 FP32 vs. Quant INT8
+    # The FP32 model should already be downloaded for slim Quant unit tests on Linux
+    set(QUANT2_MobileNetV1_MODEL_DIR "${QUANT_DATA_DIR}/MobileNetV1_quant2")
+    set(QUANT2_INT8_MobileNetV1_MODEL_DIR
+        "${QUANT_DATA_DIR}/MobileNetV1_quant2_int8")
+    if(NOT LINUX)
+      download_quant_data_without_verify(${QUANT2_MobileNetV1_MODEL_DIR}
+                                         "MobileNet_qat_perf.tar.gz")
+    endif()
+    download_quant_data_without_verify(${QUANT2_INT8_MobileNetV1_MODEL_DIR}
+                                       "MobileNet_qat_perf_int8.tar.gz")
+    inference_analysis_api_quant_test_run(
+      test_analyzer_quant_performance_benchmark
+      ${QUANT_IMG_CLASS_TEST_APP}
+      ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
+      ${QUANT2_INT8_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf_int8
+      ${IMAGENET_DATA_PATH}
+      false)
+    # Quant2 MobileNetV1
+    inference_analysis_api_quant_test_run(
+      test_analyzer_quant2_mobilenetv1_mkldnn
+      ${QUANT_IMG_CLASS_TEST_APP}
+      ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
+      ${QUANT2_MobileNetV1_MODEL_DIR}/MobileNet_qat_perf/float
+      ${IMAGENET_DATA_PATH}
+      true)
+    # Quant2 ResNet50 with input/output scales in `fake_quantize_range_abs_max` operators and the `out_threshold` attributes,
+    # with weight scales in `fake_channel_wise_dequantize_max_abs` operators
+    set(QUANT2_RESNET50_CHANNELWISE_MODEL_DIR
+        "${QUANT_DATA_DIR}/ResNet50_quant2_channelwise")
+    set(QUANT2_RESNET50_CHANNELWISE_MODEL_ARCHIVE
+        "ResNet50_qat_channelwise.tar.gz")
+    if(NOT LINUX)
+      download_quant_data_without_verify(
+        ${QUANT2_RESNET50_CHANNELWISE_MODEL_DIR}
+        ${QUANT2_RESNET50_CHANNELWISE_MODEL_ARCHIVE})
+    endif()
+    set(QUANT2_RESNET50_MODEL
+        ${QUANT2_RESNET50_CHANNELWISE_MODEL_DIR}/ResNet50_qat_channelwise)
+    inference_analysis_api_quant_test_run(
+      test_analyzer_quant2_resnet50_channelwise_mkldnn
+      ${QUANT_IMG_CLASS_TEST_APP} ${QUANT2_RESNET50_MODEL}
+      ${QUANT2_RESNET50_MODEL} ${IMAGENET_DATA_PATH} true)
+    ### Other tests
+    # MKLDNN quantizer config
+    set(MKLDNN_QUANTIZER_CONFIG_TEST_APP "test_mkldnn_quantizer_config")
+    set(MKLDNN_QUANTIZER_CONFIG_TEST_APP_SRC
+        "mkldnn_quantizer_config_tester.cc")
+    inference_analysis_api_test_build(${MKLDNN_QUANTIZER_CONFIG_TEST_APP}
+                                      ${MKLDNN_QUANTIZER_CONFIG_TEST_APP_SRC})
+    inference_analysis_test_run(test_mkldnn_quantizer_config COMMAND
+                                ${MKLDNN_QUANTIZER_CONFIG_TEST_APP})
+    # preprocess data2bin imagenet
+    download_int8_data_without_verify(${INT8_DATA_DIR} "imagenet_small.tar.gz")
+    set(IMAGENET_SMALL_DATA_DIR "${INT8_DATA_DIR}/imagenet_small")
+    set(IMAGENET_SMALL_OUTPUT_FILE "imagenet_small.bin")
+    preprocess_data2bin_test_run(
+      preprocess_local_imagenet "full_ILSVRC2012_val_preprocess.py"
+      ${IMAGENET_SMALL_DATA_DIR} ${IMAGENET_SMALL_OUTPUT_FILE})
+    # preprocess data2bin pascalvoc
+    download_int8_data_without_verify(${INT8_DATA_DIR} "pascalvoc_small.tar.gz")
+    set(PASCALVOC_SMALL_DATA_DIR "${INT8_DATA_DIR}/pascalvoc_small")
+    set(PASCALVOC_SMALL_OUTPUT_FILE "pascalvoc_small.bin")
+    preprocess_data2bin_test_run(
+      preprocess_local_pascalvoc "full_pascalvoc_test_preprocess.py"
+      ${PASCALVOC_SMALL_DATA_DIR} ${PASCALVOC_SMALL_OUTPUT_FILE})
+  endif()
+  # bert, max_len=20, embedding_dim=128
+  set(BERT_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/bert_emb128")
+  download_model_and_data_without_verify(
+    ${BERT_INSTALL_DIR} "bert_emb128_model.tar.gz" "bert_data_len20.txt.tar.gz")
+  inference_analysis_api_test(test_analyzer_bert ${BERT_INSTALL_DIR}
+                              analyzer_bert_tester.cc)
+  # multiple models prediction
+  set(MMP_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/multi_model_prediction")
+  download_data_without_verify(${MMP_INSTALL_DIR}
+                               PaddleInference/mobilenet_v2_models.tar.gz)
+  inference_multiple_models_analysis_api_test(
+    test_analyzer_multi_model_prediction ${MMP_INSTALL_DIR}
+    analyzer_mmp_tester.cc)
+  if(WITH_GPU AND TENSORRT_FOUND)
+    set(TRT_MODEL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/trt_models")
+    if(NOT EXISTS ${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models.tar.gz)
+      inference_download_and_uncompress(
+        ${TRT_MODEL_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
+        "trt_inference_test_models.tar.gz" 3dcccdc38b549b6b1b4089723757bd98)
+    endif()
+    set(TEST_SPLIT_CONVERTER_MODEL
+        "${TRT_MODEL_INSTALL_DIR}/trt_split_op_converter_test")
+    if(NOT EXISTS ${TEST_SPLIT_CONVERTER_MODEL}/split_converter.tgz)
+      inference_download_and_uncompress_without_verify(
+        ${TEST_SPLIT_CONVERTER_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "split_converter.tgz")
+    endif()
+    inference_analysis_test(
+      trt_mobilenet_test
+      SRCS
+      trt_mobilenet_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    inference_analysis_test(
+      trt_resnet50_test
+      SRCS
+      trt_resnet50_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    inference_analysis_test(
+      trt_resnext_test
+      SRCS
+      trt_resnext_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    inference_analysis_test(
+      trt_fc_prelu_test
+      SRCS
+      trt_fc_prelu_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    inference_analysis_test(
+      trt_cascade_rcnn_test
+      SRCS
+      trt_cascade_rcnn_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    inference_analysis_test(
+      trt_split_converter_test
+      SRCS
+      trt_split_converter_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TEST_SPLIT_CONVERTER_MODEL}/)
+    inference_analysis_test(
+      test_analyzer_capi_exp_gpu
+      SRCS
+      analyzer_capi_exp_gpu_tester.cc
+      EXTRA_DEPS
+      ${INFERENCE_C_EXTRA_DEPS}
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    if(WIN32)
+      target_link_libraries(test_analyzer_capi_exp_gpu
+                            paddle_inference_c_shared)
+    else()
+      target_link_libraries(test_analyzer_capi_exp_gpu paddle_inference_c)
+    endif()
+    inference_analysis_test(
+      test_analyzer_capi_exp_xpu
+      SRCS
+      analyzer_capi_exp_xpu_tester.cc
+      EXTRA_DEPS
+      ${INFERENCE_C_EXTRA_DEPS}
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
+    if(WIN32)
+      target_link_libraries(test_analyzer_capi_exp_xpu
+                            paddle_inference_c_shared)
+    else()
+      target_link_libraries(test_analyzer_capi_exp_xpu paddle_inference_c)
+    endif()
+    set(TRT_MODEL_QUANT_RESNET_DIR
+        "${INFERENCE_DEMO_INSTALL_DIR}/small_quant_model")
+    if(NOT EXISTS ${INFERENCE_DEMO_INSTALL_DIR}/small_quant_model.tgz)
+      inference_download_and_uncompress_without_verify(
+        ${INFERENCE_DEMO_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
+        "small_quant_model.tgz")
+    endif()
+    inference_analysis_test(
+      trt_quant_int8_test
+      SRCS
+      trt_quant_int8_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_QUANT_RESNET_DIR})
+    set(TRT_MODEL_QUANT_YOLOV3_DIR
+        "${INFERENCE_DEMO_INSTALL_DIR}/yolov3_r50_quant_aware")
+    if(NOT EXISTS ${INFERENCE_DEMO_INSTALL_DIR}/yolov3_r50_quant_aware.tgz)
+      inference_download_and_uncompress_without_verify(
+        ${INFERENCE_DEMO_INSTALL_DIR} ${INFERENCE_URL}/tensorrt_test
+        "yolov3_r50_quant_aware.tgz")
+    endif()
+    inference_analysis_test(
+      trt_quant_int8_yolov3_r50_test
+      SRCS
+      trt_quant_int8_yolov3_r50_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_QUANT_YOLOV3_DIR})
+    set(TEST_TRT_DYNAMIC_MODEL2
+        "${TRT_MODEL_INSTALL_DIR}/complex_model_dynamic")
+    if(NOT EXISTS ${TEST_TRT_DYNAMIC_MODEL2}/complex_model_dynamic2.tar.gz)
+      inference_download_and_uncompress_without_verify(
+        ${TEST_TRT_DYNAMIC_MODEL2} ${INFERENCE_URL}/tensorrt_test
+        "complex_model_dynamic2.tar.gz")
+    endif()
+    set(TEST_TRT_DYNAMIC_MODEL
+        "${TRT_MODEL_INSTALL_DIR}/conv_bn_swish_split_gelu")
+    if(NOT EXISTS ${TEST_TRT_DYNAMIC_MODEL}/conv_bn_swish_split_gelu.tar.gz)
+      inference_download_and_uncompress(
+        ${TEST_TRT_DYNAMIC_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "conv_bn_swish_split_gelu.tar.gz" 2a5e8791e47b221b4f782151d76da9c6)
+    endif()
+    inference_analysis_test(
+      trt_dynamic_shape_test
+      SRCS
+      trt_dynamic_shape_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TRT_MODEL_INSTALL_DIR})
+    set(TEST_TRT_ERNIE_MODEL "${TRT_MODEL_INSTALL_DIR}/ernie_test")
+    if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4.tar.gz)
+      inference_download_and_uncompress(
+        ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "ernie_model_4.tar.gz" 5fa371efa75706becbaad79195d2ca68)
+    endif()
+    inference_analysis_test(
+      test_trt_dynamic_shape_ernie
+      SRCS
+      trt_dynamic_shape_ernie_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4)
+    set(TEST_TRT_TRANSFORMER_PRUNE_MODEL
+        "${TRT_MODEL_INSTALL_DIR}/transformer_prune")
+    if(NOT EXISTS ${TEST_TRT_TRANSFORMER_PRUNE_MODEL}/transformer_prune.tar.gz)
+      inference_download_and_uncompress(
+        ${TEST_TRT_TRANSFORMER_PRUNE_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "transformer_prune.tar.gz" 77b56dc73ff0cf44ddb1ce9ca0b0f471)
+    endif()
+    inference_analysis_test(
+      test_trt_dynamic_shape_transformer_prune
+      SRCS
+      trt_dynamic_shape_transformer_prune_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TEST_TRT_TRANSFORMER_PRUNE_MODEL}/transformer_prune)
+    if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4_unserialized.tgz)
+      inference_download_and_uncompress(
+        ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "ernie_model_4_unserialized.tgz" 833d73fc6a7f7e1ee4a1fd6419209e55)
+    endif()
+    inference_analysis_test(
+      test_trt_dynamic_shape_ernie_ser_deser
+      SRCS
+      trt_dynamic_shape_ernie_serialize_deserialize_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4_unserialized)
+    if(NOT EXISTS ${TEST_TRT_ERNIE_MODEL}/ernie_model_4_fp16_unserialized.tgz)
+      inference_download_and_uncompress(
+        ${TEST_TRT_ERNIE_MODEL} ${INFERENCE_URL}/tensorrt_test
+        "ernie_model_4_fp16_unserialized.tgz" c5ff2d0cad79953ffbf2b8b9e2fae6e4)
+    endif()
+    inference_analysis_test(
+      test_trt_dynamic_shape_ernie_fp16_ser_deser
+      SRCS
+      trt_dynamic_shape_ernie_fp16_serialize_deserialize_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${TEST_TRT_ERNIE_MODEL}/ernie_model_4_fp16_unserialized)
+  endif()
+  set(LITE_MODEL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/lite")
+  download_data_without_verify(${LITE_MODEL_INSTALL_DIR} "mul_model_fp32.tgz")
+  inference_analysis_test(
+    lite_mul_model_test
+    SRCS
+    lite_mul_model_test.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${LITE_MODEL_INSTALL_DIR})
+  inference_analysis_test(
+    lite_resnet50_test
+    SRCS
+    lite_resnet50_test.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${RESNET50_MODEL_DIR})
+  inference_analysis_test(
+    test_analyzer_capi_exp
+    SRCS
+    analyzer_capi_exp_tester.cc
+    EXTRA_DEPS
+    ${INFERENCE_C_EXTRA_DEPS}
+    ARGS
+    --infer_model=${RESNET50_MODEL_DIR}/model)
+  if(WIN32)
+    target_link_libraries(test_analyzer_capi_exp paddle_inference_c_shared)
+  else()
+    target_link_libraries(test_analyzer_capi_exp paddle_inference_c)
+  endif()
+  inference_analysis_test(
+    test_analyzer_capi_exp_pd_config
+    SRCS
+    analyzer_capi_exp_pd_config_tester.cc
+    EXTRA_DEPS
+    ${INFERENCE_C_EXTRA_DEPS}
+    ARGS
+    --infer_model=${MOBILENET_INSTALL_DIR}/model)
+  if(WIN32)
+    target_link_libraries(test_analyzer_capi_exp_pd_config
+                          paddle_inference_c_shared)
+  else()
+    target_link_libraries(test_analyzer_capi_exp_pd_config paddle_inference_c)
+  endif()
+  inference_analysis_test(
+    test_analyzer_capi_exp_pd_tensor
+    SRCS
+    analyzer_capi_exp_pd_tensor_tester.cc
+    EXTRA_DEPS
+    ${INFERENCE_C_EXTRA_DEPS}
+    ARGS
+    --infer_model=${MOBILENET_INSTALL_DIR}/model)
+  if(WIN32)
+    target_link_libraries(test_analyzer_capi_exp_pd_tensor
+                          paddle_inference_c_shared)
+  else()
+    target_link_libraries(test_analyzer_capi_exp_pd_tensor paddle_inference_c)
+  endif()
+  if(NOT APPLE AND NOT WIN32)
+    inference_analysis_test(
+      test_analyzer_capi_exp_pd_threads
+      SRCS
+      analyzer_capi_exp_pd_threads_tester.cc
+      EXTRA_DEPS
+      ${INFERENCE_C_EXTRA_DEPS}
+      ARGS
+      --infer_model=${MOBILENET_INSTALL_DIR}/model)
+    if(WIN32)
+      target_link_libraries(test_analyzer_capi_exp_pd_threads
+                            paddle_inference_c_shared)
+    else()
+      target_link_libraries(test_analyzer_capi_exp_pd_threads
+                            paddle_inference_c)
+    endif()
+  endif()
+  inference_analysis_test(
+    test_analyzer_zerocopytensor_tensor
+    SRCS
+    analyzer_zerocopy_tensor_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${OCR_INSTALL_DIR}/model)
+  if(WITH_DISTRIBUTE AND WITH_PSCORE)
+    inference_analysis_test(
+      test_analyzer_dist_model
+      SRCS
+      analyzer_dist_model_tester.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${OCR_INSTALL_DIR}/model)
+  endif()
+  if(WITH_DISTRIBUTE
+     AND WITH_PSCORE
+     AND WITH_XPU
+     AND WITH_XPU_BKCL)
+    inference_analysis_test(
+      test_analyzer_dist_model_xpu
+      SRCS
+      analyzer_dist_model_xpu_tester.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${OCR_INSTALL_DIR}/model)
+  endif()
+  inference_analysis_test(
+    test_analyzer_paddletensor_tensor
+    SRCS
+    analyzer_paddle_tensor_tester.cc
+    EXTRA_DEPS
+    paddle_inference_shared
+    ARGS
+    --infer_model=${OCR_INSTALL_DIR}/model
+    --infer_data=${OCR_INSTALL_DIR}/data.txt
+    --refer_result=${OCR_INSTALL_DIR}/result.txt)
+  if(WITH_MKLDNN)
+    inference_analysis_test(
+      test_analyzer_capi_exp_int
+      SRCS
+      analyzer_capi_exp_int_tester.cc
+      EXTRA_DEPS
+      ${INFERENCE_C_EXTRA_DEPS}
+      ARGS
+      --infer_model=${INT8_DATA_DIR}/resnet50/model)
+    if(WIN32)
+      target_link_libraries(test_analyzer_capi_exp_int
+                            paddle_inference_c_shared)
+    else()
+      target_link_libraries(test_analyzer_capi_exp_int paddle_inference_c)
+    endif()
+  endif()
+  inference_analysis_test(
+    test_analyzer_capi_exp_ner
+    SRCS
+    analyzer_capi_exp_ner_tester.cc
+    EXTRA_DEPS
+    ${INFERENCE_C_EXTRA_DEPS}
+    ARGS
+    --infer_model=${CHINESE_NER_INSTALL_DIR}/model)
+  if(WIN32)
+    target_link_libraries(test_analyzer_capi_exp_ner paddle_inference_c_shared)
+  else()
+    target_link_libraries(test_analyzer_capi_exp_ner paddle_inference_c)
+  endif()
+  if(WITH_GPU)
+    inference_analysis_test(
+      paddle_infer_api_test
+      SRCS
+      paddle_infer_api_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${RESNET50_MODEL_DIR})
+    inference_analysis_test(
+      paddle_infer_api_copy_tensor_tester
+      SRCS
+      paddle_infer_api_copy_tensor_tester.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${RESNET50_MODEL_DIR})
+    set_tests_properties(paddle_infer_api_copy_tensor_tester PROPERTIES TIMEOUT
+                                                                        30)
+  endif()
+  cc_test(
+    paddle_infer_api_errors_test
+    SRCS paddle_infer_api_errors_tester.cc
+    DEPS paddle_inference_api)
+  if(WITH_GPU AND TENSORRT_FOUND)
+    set_tests_properties(trt_resnext_test PROPERTIES TIMEOUT 300)
+    set_tests_properties(trt_quant_int8_yolov3_r50_test PROPERTIES TIMEOUT 400)
+    set_tests_properties(trt_resnet50_test PROPERTIES TIMEOUT 300)
+    set_tests_properties(trt_cascade_rcnn_test PROPERTIES TIMEOUT 300)
+    set_tests_properties(test_trt_dynamic_shape_ernie_ser_deser
+                         PROPERTIES TIMEOUT 300)
+    set_tests_properties(test_trt_dynamic_shape_ernie_fp16_ser_deser
+                         PROPERTIES TIMEOUT 300)
+    set_tests_properties(test_trt_dynamic_shape_ernie PROPERTIES TIMEOUT 300)
+  endif()
+  if(WITH_MKLDNN)
+    set_tests_properties(test_analyzer_int8_resnet50 PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_int8_mobilenet_ssd PROPERTIES TIMEOUT
+                                                                     120)
+    set_tests_properties(test_analyzer_quant_performance_benchmark
+                         PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_int8_mobilenetv2 PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_int8_mobilenetv1 PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_int8_mobilenetv3_large PROPERTIES TIMEOUT
+                                                                         120)
+    set_tests_properties(test_analyzer_quant2_mobilenetv1_mkldnn
+                         PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_quant2_resnet50_channelwise_mkldnn
+                         PROPERTIES TIMEOUT 120)
+  endif()
+  set_tests_properties(lite_resnet50_test PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_mobilenet_transpose PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_resnet50 PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_ner PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_ernie_int8 PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_googlenet PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_small_dam PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_transformer PROPERTIES TIMEOUT 120)
+  set_tests_properties(test_analyzer_mobilenet_depthwise_conv PROPERTIES TIMEOUT
+                                                                         120)
+  if(WITH_GPU)
+    set_tests_properties(test_analyzer_bert PROPERTIES TIMEOUT 120)
+    set_tests_properties(test_analyzer_ernie PROPERTIES TIMEOUT 120)
+  endif()
+  if(WITH_GPU AND TENSORRT_FOUND)
+    set_tests_properties(trt_mobilenet_test PROPERTIES TIMEOUT 120)
+    if(WITH_MKLDNN)
+      set_tests_properties(test_analyzer_bfloat16_resnet50 PROPERTIES TIMEOUT
+                                                                      120)
+    endif()
+  endif()
+  if(ON_INFER OR WITH_GPU)
+    set_tests_properties(test_analyzer_transformer_profile PROPERTIES TIMEOUT
+                                                                      120)
+  endif()
+  if(WITH_IPU)
+    #word2vec sample
+    set(WORD2VEC_INSTALL_DIR
+        "${INFERENCE_DEMO_INSTALL_DIR}/word2vec/word2vec.inference.model")
+    inference_analysis_test(
+      ipu_word2vec_sample
+      SRCS
+      ipu_word2vec_sample.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${WORD2VEC_INSTALL_DIR})
+    # ERNIE
+    set(ERNIE_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/Ernie")
+    inference_analysis_api_test(
+      ipu_ernie_test ${ERNIE_INSTALL_DIR} ipu_ernie_test.cc ARGS --warmup=true
+      --repeat=10)
+    inference_analysis_api_test(
+      ipu_ernie_fp16_test ${ERNIE_INSTALL_DIR} ipu_ernie_fp16_test.cc ARGS
+      --warmup=true --repeat=10)
+    # Resnet50
+    set(RESNET50_MODEL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/resnet50")
+    inference_analysis_test(
+      ipu_resnet50_test
+      SRCS
+      ipu_resnet50_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${RESNET50_MODEL_DIR}
+      --warmup=true
+      --repeat=10)
+    inference_analysis_test(
+      ipu_resnet50_fp16_test
+      SRCS
+      ipu_resnet50_fp16_test.cc
+      EXTRA_DEPS
+      paddle_inference_shared
+      ARGS
+      --infer_model=${RESNET50_MODEL_DIR}
+      --warmup=true
+      --repeat=10)
+    # Only support Resnet50 and Ernie currently
+    inference_analysis_api_test(
+      ipu_multi_model_profile
+      SRCS
+      ipu_multi_model_profile.cc
+      ARGS
+      --model_name="Resnet50"
+      --infer_model=${RESNET50_MODEL_DIR}
+      --warmup=true
+      --repeat=10)
+  endif()
+  set(inference_deps ${analysis_deps} paddle_inference_api analysis
+                     naive_executor ${GLOB_PASS_LIB})
+  if(WITH_TESTING)
+    if(NOT APPLE AND NOT WIN32)
+      inference_base_test(
+        test_api_impl
+        SRCS
+        api_impl_tester.cc
+        DEPS
+        paddle_inference_shared
+        ARGS
+        --word2vec_dirname=${WORD2VEC_MODEL_DIR}
+        --book_dirname=${IMG_CLS_RESNET_INSTALL_DIR})
+    elseif(WIN32)
+      inference_base_test(
+        test_api_impl
+        SRCS
+        api_impl_tester.cc
+        DEPS
+        ${inference_deps}
+        ARGS
+        --word2vec_dirname=${WORD2VEC_MODEL_DIR}
+        --book_dirname=${IMG_CLS_RESNET_INSTALL_DIR})
+    endif()
+  endif()
+  if(NOT APPLE AND NOT WIN32)
+    cc_test_old(
+      test_analysis_predictor
+      SRCS
+      analysis_predictor_tester.cc
+      DEPS
+      paddle_inference_shared
+      ARGS
+      --dirname=${WORD2VEC_MODEL_DIR})
+  elseif(WIN32)
+    cc_test_old(
+      test_analysis_predictor
+      SRCS
+      analysis_predictor_tester.cc
+      DEPS
+      analysis_predictor
+      benchmark
+      ${inference_deps}
+      ARGS
+      --dirname=${WORD2VEC_MODEL_DIR})
+  endif()
+  if(WITH_TESTING AND WITH_MKLDNN)
+    if(NOT APPLE AND NOT WIN32)
+      cc_test(
+        test_mkldnn_quantizer
+        SRCS mkldnn_quantizer_tester.cc
+        DEPS paddle_inference_shared ARGS --dirname=${WORD2VEC_MODEL_DIR})
+    elseif(WIN32)
+      cc_test(
+        test_mkldnn_quantizer
+        SRCS mkldnn_quantizer_tester.cc
+        DEPS analysis_predictor benchmark ${inference_deps} ARGS
+             --dirname=${WORD2VEC_MODEL_DIR})
+    endif()
+  endif()
+  if(WITH_TESTING AND TEST test_api_impl)
+    if(NOT APPLE)
+      set_tests_properties(test_api_impl PROPERTIES TIMEOUT 120)
+    endif()
+  endif()
+endif()
--- a/paddle/fluid/inference/api/analysis_predictor_tester.cc
+++ b/paddle/fluid/inference/api/analysis_predictor_tester.cc
@@ -27,9 +27,9 @@
 #include "paddle/fluid/inference/api/helper.h"
 #include "paddle/fluid/inference/api/paddle_api.h"
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/io_utils.h"
 #include "paddle/phi/backends/cpu/cpu_info.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_string(dirname, "", "dirname to tests.");

--- a/paddle/fluid/inference/tests/api/analyzer_bert_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_bert_tester.cc
@@ -13,7 +13,7 @@
 // limitations under the License.
 #include "paddle/fluid/framework/transfer_scope_cache.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_bfloat16_image_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_bfloat16_image_classification_tester.cc
@@ -13,8 +13,8 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_analysis_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/phi/backends/cpu/cpu_info.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(enable_mkldnn, true, "Enable MKLDNN");

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_gpu_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_gpu_tester.cc
@@ -24,7 +24,7 @@ limitations under the License. */
 #endif
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_int_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_int_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_ner_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_ner_tester.cc
@@ -20,7 +20,7 @@
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_config_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_config_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_tensor_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_tensor_tester.cc
@@ -23,7 +23,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_threads_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_pd_threads_tester.cc
@@ -23,7 +23,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_tester.cc
@@ -21,7 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/inference/capi_exp/pd_config.h"
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
 #include "paddle/fluid/inference/capi_exp/pd_utils.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_exp_xpu_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_exp_xpu_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi_exp/pd_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_gpu_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_gpu_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_int_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_int_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_ner_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_ner_tester.cc
@@ -20,7 +20,7 @@
 #include <vector>
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_pd_tensor_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_pd_tensor_tester.cc
@@ -24,7 +24,7 @@ limitations under the License. */
 #include "paddle/fluid/inference/capi/c_api_internal.h"
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_capi_xpu_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_capi_xpu_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/inference/capi/paddle_c_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_dam_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_dam_tester.cc
@@ -15,7 +15,7 @@
 #include <vector>
 #include "paddle/fluid/inference/analysis/helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 const int FLAGS_max_turn_num = 1;

--- a/paddle/fluid/inference/tests/api/analyzer_detect_functional_mkldnn_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_detect_functional_mkldnn_tester.cc
@@ -17,8 +17,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/phi/common/place.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_string(infer_shape, "", "data shape file");
 DEFINE_int32(sample, 20, "number of sample");

--- a/paddle/fluid/inference/tests/api/analyzer_detect_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_detect_tester.cc
@@ -17,7 +17,7 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_string(infer_shape, "", "data shape file");
 DEFINE_int32(sample, 20, "number of sample");

--- a/paddle/fluid/inference/tests/api/analyzer_dist_model_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_dist_model_tester.cc
@@ -17,8 +17,8 @@
 #include "paddle/fluid/framework/op_desc.h"
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/singleton.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_dist_model_xpu_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_dist_model_xpu_tester.cc
@@ -17,8 +17,8 @@
 #include "paddle/fluid/framework/op_desc.h"
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/singleton.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_ernie_int8_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_ernie_int8_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/analyzer_ernie_tester.h"
+#include "test/cpp/inference/api/analyzer_ernie_tester.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_ernie_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_ernie_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/analyzer_ernie_tester.h"
+#include "test/cpp/inference/api/analyzer_ernie_tester.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_ernie_tester.h
+++ b/paddle/fluid/inference/tests/api/analyzer_ernie_tester.h
@@ -14,7 +14,7 @@
 #pragma once
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_image_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_image_classification_tester.cc
@@ -15,7 +15,7 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(disable_mkldnn_fc, false, "Disable usage of MKL-DNN's FC op");

--- a/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_int8_image_classification_tester.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_analysis_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(enable_mkldnn, true, "Enable MKLDNN");

--- a/paddle/fluid/inference/tests/api/analyzer_int8_object_detection_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_int8_object_detection_tester.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_analysis_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(enable_mkldnn, true, "Enable MKLDNN");

--- a/paddle/fluid/inference/tests/api/analyzer_lac_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_lac_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_lexical_analysis_gru_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_lexical_analysis_gru_tester.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_analysis_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 // setting iterations to 0 means processing the whole dataset
 namespace paddle {

--- a/paddle/fluid/inference/tests/api/analyzer_mmp_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_mmp_tester.cc
@@ -15,7 +15,7 @@
 #include <random>
 #include "paddle/fluid/framework/transfer_scope_cache.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 // Here add missing commands
 DEFINE_string(infer_model2, "", "model path");
@@ -35,7 +35,9 @@ void SetConfig(AnalysisConfig* config, const std::string& infer_model) {
 }
 std::unique_ptr<PaddlePredictor> InitializePredictor(
-    const std::string& infer_model, std::vector<float>& data, bool use_mkldnn) {
+    const std::string& infer_model,
+    const std::vector<float>& data,
+    bool use_mkldnn) {
  AnalysisConfig cfg;
  SetConfig(&cfg, infer_model);
  if (use_mkldnn) {

--- a/paddle/fluid/inference/tests/api/analyzer_ner_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_ner_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_paddle_tensor_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_paddle_tensor_tester.cc
@@ -16,8 +16,8 @@
 #include "paddle/fluid/framework/op_desc.h"
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/singleton.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_quant_image_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_quant_image_classification_tester.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_analysis_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(enable_mkldnn, true, "Enable MKLDNN");

--- a/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 DEFINE_bool(with_precision_check, true, "turn on test");

--- a/paddle/fluid/inference/tests/api/analyzer_rnn2_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_rnn2_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_save_model_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_save_model_tester.cc
@@ -13,7 +13,7 @@
 // limitations under the License.
 #include "paddle/fluid/inference/analysis/helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_conv1_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_conv1_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_compare_determine_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_compare_determine_tester.cc
@@ -16,8 +16,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_seq_pool1_tester_helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_compare_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_compare_tester.cc
@@ -16,8 +16,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_seq_pool1_tester_helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_fuse_compare_zero_copy_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_fuse_compare_zero_copy_tester.cc
@@ -16,8 +16,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_seq_pool1_tester_helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_fuse_statis_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_fuse_statis_tester.cc
@@ -16,8 +16,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_seq_pool1_tester_helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_profile_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_profile_tester.cc
@@ -16,8 +16,8 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_seq_pool1_tester_helper.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester_helper.h
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <utility>
 #include <vector>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_text_classification_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_text_classification_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_transformer_compare_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_transformer_compare_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/analyzer_transformer_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_transformer_tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_transformer_fuse_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_transformer_fuse_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/analyzer_transformer_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_transformer_tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_transformer_profile_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_transformer_profile_tester.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/analyzer_transformer_tester_helper.h"
+#include "test/cpp/inference/api/analyzer_transformer_tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_transformer_tester_helper.h
+++ b/paddle/fluid/inference/tests/api/analyzer_transformer_tester_helper.h
@@ -16,7 +16,7 @@
 #include <utility>
 #include <vector>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_vis_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_vis_tester.cc
@@ -17,7 +17,7 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_vit_ocr_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_vit_ocr_tester.cc
@@ -15,7 +15,7 @@ limitations under the License. */
 #include <fstream>
 #include <iostream>
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/analyzer_zerocopy_tensor_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_zerocopy_tensor_tester.cc
@@ -16,8 +16,8 @@
 #include "paddle/fluid/framework/op_desc.h"
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
 #include "paddle/fluid/inference/utils/singleton.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/api/api_impl_tester.cc
+++ b/paddle/fluid/inference/api/api_impl_tester.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include "gflags/gflags.h"
 #include "paddle/fluid/framework/convert_utils.h"
 #include "paddle/fluid/inference/api/api_impl.h"
-#include "paddle/fluid/inference/tests/test_helper.h"
+#include "test/cpp/inference/test_helper.h"
 #ifdef __clang__
 #define ACC_DIFF 4e-3

--- a/paddle/fluid/inference/api/api_tester.cc
+++ b/paddle/fluid/inference/api/api_tester.cc
--- a/paddle/fluid/inference/tests/api/config_printer.h
+++ b/paddle/fluid/inference/tests/api/config_printer.h
--- a/paddle/fluid/inference/tests/api/full_ILSVRC2012_val_preprocess.py
+++ b/paddle/fluid/inference/tests/api/full_ILSVRC2012_val_preprocess.py
--- a/paddle/fluid/inference/tests/api/full_pascalvoc_test_preprocess.py
+++ b/paddle/fluid/inference/tests/api/full_pascalvoc_test_preprocess.py
--- a/paddle/fluid/inference/tests/api/gpu_ernie_half_test.cc
+++ b/paddle/fluid/inference/tests/api/gpu_ernie_half_test.cc
@@ -13,7 +13,7 @@
 // limitations under the License.
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/api/helper_test.cc
+++ b/paddle/fluid/inference/api/helper_test.cc
--- a/paddle/fluid/inference/tests/api/int8_mkldnn_quantization.md
+++ b/paddle/fluid/inference/tests/api/int8_mkldnn_quantization.md
--- a/paddle/fluid/inference/tests/api/ipu_ernie_fp16_test.cc
+++ b/paddle/fluid/inference/tests/api/ipu_ernie_fp16_test.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/ipu_ernie_test.cc
+++ b/paddle/fluid/inference/tests/api/ipu_ernie_test.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/ipu_multi_model_profile.cc
+++ b/paddle/fluid/inference/tests/api/ipu_multi_model_profile.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/ipu_resnet50_fp16_test.cc
+++ b/paddle/fluid/inference/tests/api/ipu_resnet50_fp16_test.cc
@@ -18,7 +18,7 @@ limitations under the License. */
 #include <cmath>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/ipu_resnet50_test.cc
+++ b/paddle/fluid/inference/tests/api/ipu_resnet50_test.cc
@@ -15,7 +15,7 @@ limitations under the License. */
 #include <cmath>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/ipu_word2vec_sample.cc
+++ b/paddle/fluid/inference/tests/api/ipu_word2vec_sample.cc
--- a/paddle/fluid/inference/tests/api/lite_mul_model_test.cc
+++ b/paddle/fluid/inference/tests/api/lite_mul_model_test.cc
@@ -20,7 +20,7 @@ limitations under the License. */
 #include <thread>  // NOLINT
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/lite_resnet50_test.cc
+++ b/paddle/fluid/inference/tests/api/lite_resnet50_test.cc
@@ -18,7 +18,7 @@ limitations under the License. */
 #include <cmath>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/mkldnn_quantizer_config_tester.cc
+++ b/paddle/fluid/inference/tests/api/mkldnn_quantizer_config_tester.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <iostream>
 #include "paddle/fluid/inference/api/paddle_mkldnn_quantizer_config.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/api/mkldnn_quantizer_tester.cc
+++ b/paddle/fluid/inference/api/mkldnn_quantizer_tester.cc
--- a/paddle/fluid/inference/tests/api/paddle_infer_api_copy_tensor_tester.cc
+++ b/paddle/fluid/inference/tests/api/paddle_infer_api_copy_tensor_tester.cc
@@ -22,8 +22,8 @@ limitations under the License. */
 #include "gflags/gflags.h"
 #include "glog/logging.h"
 #include "paddle/fluid/inference/api/paddle_infer_contrib.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
 #include "paddle/fluid/platform/float16.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle_infer {

--- a/paddle/fluid/inference/tests/api/paddle_infer_api_errors_tester.cc
+++ b/paddle/fluid/inference/tests/api/paddle_infer_api_errors_tester.cc
--- a/paddle/fluid/inference/tests/api/paddle_infer_api_test.cc
+++ b/paddle/fluid/inference/tests/api/paddle_infer_api_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle_infer {

--- a/paddle/fluid/inference/tests/api/test_detection_dataset_preprocess.py
+++ b/paddle/fluid/inference/tests/api/test_detection_dataset_preprocess.py
--- a/paddle/fluid/inference/tests/api/tester_helper.h
+++ b/paddle/fluid/inference/tests/api/tester_helper.h
@@ -35,10 +35,10 @@
 #include "paddle/fluid/inference/api/helper.h"
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #include "paddle/fluid/inference/api/paddle_inference_pass.h"
-#include "paddle/fluid/inference/tests/api/config_printer.h"
-#include "paddle/fluid/inference/tests/test_helper.h"
 #include "paddle/fluid/inference/utils/benchmark.h"
 #include "paddle/fluid/platform/profiler/event_tracing.h"
+#include "test/cpp/inference/api/config_printer.h"
+#include "test/cpp/inference/test_helper.h"
 DEFINE_string(model_name, "", "model name");
 DEFINE_string(infer_model, "", "model path");

--- a/paddle/fluid/inference/tests/api/trt_cascade_rcnn_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_cascade_rcnn_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_fp16_serialize_deserialize_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_fp16_serialize_deserialize_test.cc
@@ -24,7 +24,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h"
+#include "test/cpp/inference/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.cc
@@ -24,7 +24,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h"
+#include "test/cpp/inference/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h
@@ -26,7 +26,7 @@ limitations under the License. */
 #include <vector>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_test.cc
@@ -17,7 +17,7 @@ limitations under the License. */
 #include "gflags/gflags.h"
 #include "paddle/fluid/inference/tensorrt/helper.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_dynamic_shape_transformer_prune_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_dynamic_shape_transformer_prune_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_fc_prelu_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_fc_prelu_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_instance_norm_converter_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_instance_norm_converter_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_mobilenet_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_mobilenet_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_quant_int8_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_quant_int8_test.cc
@@ -18,7 +18,7 @@ limitations under the License. */
 #include <numeric>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_quant_int8_yolov3_r50_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_quant_int8_yolov3_r50_test.cc
@@ -15,7 +15,7 @@ limitations under the License. */
 #include <numeric>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_resnet50_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_resnet50_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_resnext_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_resnext_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_split_converter_test.cc
+++ b/paddle/fluid/inference/tests/api/trt_split_converter_test.cc
@@ -16,7 +16,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include "gflags/gflags.h"
-#include "paddle/fluid/inference/tests/api/trt_test_helper.h"
+#include "test/cpp/inference/api/trt_test_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/api/trt_test_helper.h
+++ b/paddle/fluid/inference/tests/api/trt_test_helper.h
@@ -20,7 +20,7 @@ limitations under the License. */
 #include "gflags/gflags.h"
 #include "glog/logging.h"
 #include "gtest/gtest.h"
-#include "paddle/fluid/inference/tests/api/tester_helper.h"
+#include "test/cpp/inference/api/tester_helper.h"
 namespace paddle {
 namespace inference {

--- a/paddle/fluid/inference/tests/infer_ut/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/infer_ut/CMakeLists.txt
--- a/paddle/fluid/inference/tests/infer_ut/README.md
+++ b/paddle/fluid/inference/tests/infer_ut/README.md
--- a/paddle/fluid/inference/tests/infer_ut/external-cmake/gtest-cpp.cmake
+++ b/paddle/fluid/inference/tests/infer_ut/external-cmake/gtest-cpp.cmake
--- a/paddle/fluid/inference/tests/infer_ut/run.sh
+++ b/paddle/fluid/inference/tests/infer_ut/run.sh
--- a/paddle/fluid/inference/tests/infer_ut/test_LeViT.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_LeViT.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_det_mv3_db.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_det_mv3_db.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_ernie_text_cls.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_ernie_text_cls.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_ernie_xnli_int8.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_ernie_xnli_int8.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_helper.h
+++ b/paddle/fluid/inference/tests/infer_ut/test_helper.h
--- a/paddle/fluid/inference/tests/infer_ut/test_mobilnetv1.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_mobilnetv1.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_ppyolo_mbv3.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_ppyolo_mbv3.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_ppyolov2_r50vd.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_ppyolov2_r50vd.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_resnet50.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_resnet50.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_resnet50_quant.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_resnet50_quant.cc
--- a/paddle/fluid/inference/tests/infer_ut/test_suite.h
+++ b/paddle/fluid/inference/tests/infer_ut/test_suite.h
--- a/paddle/fluid/inference/tests/infer_ut/test_yolov3.cc
+++ b/paddle/fluid/inference/tests/infer_ut/test_yolov3.cc
--- a/paddle/fluid/inference/tests/test.cmake
+++ b/paddle/fluid/inference/tests/test.cmake
--- a/paddle/fluid/inference/tests/test_helper.h
+++ b/paddle/fluid/inference/tests/test_helper.h