diff --git a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
index ec8471ef960a2fc44af23c52be09cd678fab3f70..03f0f726eb61c2619c7719a865383090f86b5b7f 100644
--- a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
+++ b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
@@ -77,7 +77,7 @@ endif(NOT WIN32)
 link_directories("${PADDLE_LIB}/third_party/install/protobuf/lib")
 link_directories("${PADDLE_LIB}/third_party/install/glog/lib")
 link_directories("${PADDLE_LIB}/third_party/install/gflags/lib")
-link_directories("${PADDLE_LIB}/paddle/fluid/inference")
+link_directories("${PADDLE_LIB}/paddle/lib")
 
 add_executable(${DEMO_NAME} ${DEMO_NAME}.cc)
 
@@ -97,10 +97,10 @@ endif()
 # Note: libpaddle_inference_api.so/a must put before libpaddle_fluid.so/a
 if(WITH_STATIC_LIB)
   set(DEPS
-      ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
+      ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
 else()
   set(DEPS
-      ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
+      ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
 endif()
 
 if (NOT WIN32)
diff --git a/paddle/fluid/inference/api/demo_ci/run.sh b/paddle/fluid/inference/api/demo_ci/run.sh
index 65c95f0834a9356fc14faed8342f5d1e474edf8f..67994aad70a40c0e0c8a311914d4ea40b96eaf1e 100755
--- a/paddle/fluid/inference/api/demo_ci/run.sh
+++ b/paddle/fluid/inference/api/demo_ci/run.sh
@@ -5,12 +5,13 @@ TEST_GPU_CPU=$3 # test both GPU/CPU mode or only CPU mode
 DATA_DIR=$4 # dataset
 TENSORRT_INCLUDE_DIR=$5 # TensorRT header file dir, defalut to /usr/local/TensorRT/include
 TENSORRT_LIB_DIR=$6 # TensorRT lib file dir, default to /usr/local/TensorRT/lib
+inference_install_dir=${PADDLE_ROOT}/build/fluid_inference_install_dir
 
 cd `dirname $0`
 current_dir=`pwd`
 if [ $2 == ON ]; then
   # You can export yourself if move the install path
-  MKL_LIB=${PADDLE_ROOT}/build/fluid_install_dir/third_party/install/mklml/lib
+  MKL_LIB=${inference_install_dir}/third_party/install/mklml/lib
   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${MKL_LIB}
 fi
 if [ $3 == ON ]; then
@@ -55,7 +56,7 @@ cd build
 for WITH_STATIC_LIB in ON OFF; do
   # -----simple_on_word2vec-----
   rm -rf *
-  cmake .. -DPADDLE_LIB=${PADDLE_ROOT}/build/fluid_install_dir/ \
+  cmake .. -DPADDLE_LIB=${inference_install_dir} \
     -DWITH_MKL=$TURN_ON_MKL \
     -DDEMO_NAME=simple_on_word2vec \
     -DWITH_GPU=$TEST_GPU_CPU \
@@ -75,7 +76,7 @@ for WITH_STATIC_LIB in ON OFF; do
   fi
   # ---------vis_demo---------
   rm -rf *
-  cmake .. -DPADDLE_LIB=${PADDLE_ROOT}/build/fluid_install_dir/ \
+  cmake .. -DPADDLE_LIB=${inference_install_dir} \
     -DWITH_MKL=$TURN_ON_MKL \
     -DDEMO_NAME=vis_demo \
     -DWITH_GPU=$TEST_GPU_CPU \
@@ -98,7 +99,7 @@ for WITH_STATIC_LIB in ON OFF; do
   # --------tensorrt mobilenet------
   if [ $USE_TENSORRT == ON -a $TEST_GPU_CPU == ON ]; then
     rm -rf *
-    cmake .. -DPADDLE_LIB=${PADDLE_ROOT}/build/fluid_install_dir/ \
+    cmake .. -DPADDLE_LIB=${inference_install_dir} \
       -DWITH_MKL=$TURN_ON_MKL \
       -DDEMO_NAME=trt_mobilenet_demo \
       -DWITH_GPU=$TEST_GPU_CPU \
diff --git a/paddle/fluid/inference/api/demo_ci/simple_on_word2vec.cc b/paddle/fluid/inference/api/demo_ci/simple_on_word2vec.cc
index 8058d7e881025b1d806efe187d4523adadff367d..5ab45360e7ea85a26d5f9c0057aebee6b5788d7d 100644
--- a/paddle/fluid/inference/api/demo_ci/simple_on_word2vec.cc
+++ b/paddle/fluid/inference/api/demo_ci/simple_on_word2vec.cc
@@ -23,7 +23,7 @@ limitations under the License. */
 #include <memory>
 #include <thread>  //NOLINT
 
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle/include/paddle_inference_api.h"
 
 DEFINE_string(dirname, "", "Directory of the inference model.");
 DEFINE_bool(use_gpu, false, "Whether use gpu.");
diff --git a/paddle/fluid/inference/api/demo_ci/trt_mobilenet_demo.cc b/paddle/fluid/inference/api/demo_ci/trt_mobilenet_demo.cc
index ffb12b5871f088f15e43a1b0ff7e2a8b2f5fd079..4a8404f21c6ec6a1647e964ac3538b4b49151009 100644
--- a/paddle/fluid/inference/api/demo_ci/trt_mobilenet_demo.cc
+++ b/paddle/fluid/inference/api/demo_ci/trt_mobilenet_demo.cc
@@ -18,7 +18,7 @@ limitations under the License. */
 
 #include <gflags/gflags.h>
 #include <glog/logging.h>  // use glog instead of CHECK to avoid importing other paddle header files.
-#include "paddle/fluid/inference/demo_ci/utils.h"
+#include "utils.h"  // NOLINT
 
 DECLARE_double(fraction_of_gpu_memory_to_use);
 DEFINE_string(modeldir, "", "Directory of the inference model.");
diff --git a/paddle/fluid/inference/api/demo_ci/utils.h b/paddle/fluid/inference/api/demo_ci/utils.h
index 4792c97fe7d0a3f9c904774ad4a8e580cefcf237..d70c6aea791219a40c3164b51499f9d5e562be71 100644
--- a/paddle/fluid/inference/api/demo_ci/utils.h
+++ b/paddle/fluid/inference/api/demo_ci/utils.h
@@ -18,7 +18,7 @@
 #include <iostream>
 #include <string>
 #include <vector>
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle/include/paddle_inference_api.h"
 
 namespace paddle {
 namespace demo {
diff --git a/paddle/fluid/inference/api/demo_ci/vis_demo.cc b/paddle/fluid/inference/api/demo_ci/vis_demo.cc
index db61786e2fefda29256d84b5357028ec9c39b014..a694a4e0feae5cc9ec7f216b1fc711555c44605e 100644
--- a/paddle/fluid/inference/api/demo_ci/vis_demo.cc
+++ b/paddle/fluid/inference/api/demo_ci/vis_demo.cc
@@ -18,7 +18,7 @@ limitations under the License. */
 
 #include <gflags/gflags.h>
 #include <glog/logging.h>  // use glog instead of CHECK to avoid importing other paddle header files.
-#include "paddle/fluid/inference/demo_ci/utils.h"
+#include "utils.h"  // NOLINT
 
 #ifdef PADDLE_WITH_CUDA
 DECLARE_double(fraction_of_gpu_memory_to_use);
diff --git a/paddle/scripts/paddle_build.sh b/paddle/scripts/paddle_build.sh
index da6f5ca1586a570fb548d7b987330a8d58156e24..6f127611577e65c19227ca73a79474874cb75add 100755
--- a/paddle/scripts/paddle_build.sh
+++ b/paddle/scripts/paddle_build.sh
@@ -659,6 +659,7 @@ function gen_fluid_lib() {
 EOF
         cmake .. -DWITH_DISTRIBUTE=OFF
         make -j `nproc` fluid_lib_dist
+        make -j `nproc` inference_lib_dist
       fi
 }
 
@@ -672,6 +673,8 @@ EOF
         cd ${PADDLE_ROOT}/build
         cp -r fluid_install_dir fluid
         tar -czf fluid.tgz fluid
+        cp -r fluid_inference_install_dir fluid_inference
+        tar -czf fluid_inference.tgz fluid_inference
       fi
 }
 
@@ -683,7 +686,9 @@ function test_fluid_lib() {
     ========================================
 EOF
         cd ${PADDLE_ROOT}/paddle/fluid/inference/api/demo_ci
-        ./run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF} ${INFERENCE_DEMO_INSTALL_DIR} ${TENSORRT_INCLUDE_DIR:-/usr/local/TensorRT/include} ${TENSORRT_LIB_DIR:-/usr/local/TensorRT/lib}
+        ./run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF} ${INFERENCE_DEMO_INSTALL_DIR} \
+                 ${TENSORRT_INCLUDE_DIR:-/usr/local/TensorRT/include} \
+                 ${TENSORRT_LIB_DIR:-/usr/local/TensorRT/lib}
         ./clean.sh
       fi
 }