diff --git a/paddle/fluid/inference/tests/api/CMakeLists.txt b/paddle/fluid/inference/tests/api/CMakeLists.txt
index c8074b7665f6e5a4dfae525d1494e9b0c2aa7b51..45ba4fbe5a369327292e04acb09fc1d80e6a8adf 100644
--- a/paddle/fluid/inference/tests/api/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/api/CMakeLists.txt
@@ -298,12 +298,12 @@ if(WITH_MKLDNN)
   # build test binary to be used in subsequent tests
   inference_analysis_api_test_build(${QAT_IMG_CLASS_TEST_APP} ${QAT_IMG_CLASS_TEST_APP_SRC})
 
-  # ResNet50 FP32 vs. QAT INT8
-  set(QAT2_RESNET50_MODEL_DIR "${QAT_DATA_DIR}/ResNet50_qat_perf")
-  download_qat_data(${QAT2_RESNET50_MODEL_DIR} "ResNet50_qat_perf.tar.gz")
-  set(QAT2_INT8_RESNET50_MODEL_DIR "${QAT_DATA_DIR}/ResNet50_qat_perf_int8")
-  download_qat_data(${QAT2_INT8_RESNET50_MODEL_DIR} "ResNet50_qat_perf_int8.tar.gz")
-  inference_analysis_api_qat_test_run(test_analyzer_qat_performance_benchmark ${QAT_IMG_CLASS_TEST_APP} ${QAT2_RESNET50_MODEL_DIR}/ResNet50_qat_perf/float ${QAT2_INT8_RESNET50_MODEL_DIR}/ResNet50_qat_perf_int8 ${IMAGENET_DATA_PATH})
+  # MobileNet FP32 vs. QAT INT8
+  set(QAT2_MobileNet_MODEL_DIR "${QAT_DATA_DIR}/MobileNet_qat_perf")
+  download_qat_data(${QAT2_MobileNet_MODEL_DIR} "MobileNet_qat_perf.tar.gz")
+  set(QAT2_INT8_MobileNet_MODEL_DIR "${QAT_DATA_DIR}/MobileNet_qat_perf_int8")
+  download_qat_data(${QAT2_INT8_MobileNet_MODEL_DIR} "MobileNet_qat_perf_int8.tar.gz")
+  inference_analysis_api_qat_test_run(test_analyzer_qat_performance_benchmark ${QAT_IMG_CLASS_TEST_APP} ${QAT2_MobileNet_MODEL_DIR}/MobileNet_qat_perf/float ${QAT2_INT8_MobileNet_MODEL_DIR}/MobileNet_qat_perf_int8 ${IMAGENET_DATA_PATH})
 
 endif()
 
diff --git a/python/paddle/fluid/contrib/slim/tests/CMakeLists.txt b/python/paddle/fluid/contrib/slim/tests/CMakeLists.txt
index 9d43ed4dbe3db14c1fc19df754a25bbbb20ae0f7..24839170922ec99ade60e9226eee9a99a6520b34 100644
--- a/python/paddle/fluid/contrib/slim/tests/CMakeLists.txt
+++ b/python/paddle/fluid/contrib/slim/tests/CMakeLists.txt
@@ -32,6 +32,8 @@ function(inference_qat_int8_test target model_dir data_dir test_script use_mkldn
                  --acc_diff_threshold 0.1)
 endfunction()
 
+
+# set batch_size 10 for UT only (avoid OOM). For whole dataset, use batch_size 25 
 function(inference_qat2_int8_test target model_dir data_dir test_script use_mkldnn)
     py_test(${target} SRCS ${test_script}
             ENVS FLAGS_OMP_NUM_THREADS=${CPU_NUM_THREADS_ON_CI}
@@ -39,7 +41,7 @@ function(inference_qat2_int8_test target model_dir data_dir test_script use_mkld
                  FLAGS_use_mkldnn=${use_mkldnn}
             ARGS --qat_model ${model_dir}/float
                  --infer_data ${data_dir}/data.bin
-                 --batch_size 25
+                 --batch_size 10
                  --batch_num 2
                  --acc_diff_threshold 0.1
                  --qat2)