diff --git a/cmake/inference_lib.cmake b/cmake/inference_lib.cmake
index 2a5595307ca27a8d3a93cfa74458980c3a5d8af4..059c3a04487cce103bf55de965cea2400997a7e8 100644
--- a/cmake/inference_lib.cmake
+++ b/cmake/inference_lib.cmake
@@ -137,7 +137,7 @@ function(copy_part_of_thrid_party TARGET DST)
 endfunction()
 
 # inference library for only inference
-set(inference_lib_deps third_party paddle_fluid paddle_fluid_c paddle_fluid_shared paddle_fluid_c_shared)
+set(inference_lib_deps third_party paddle_inference paddle_inference_c paddle_inference_shared paddle_inference_c_shared)
 add_custom_target(inference_lib_dist DEPENDS ${inference_lib_deps})
 
 
@@ -164,20 +164,20 @@ copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_INSTALL_DIR})
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 if(WIN32)
     if(WITH_STATIC_LIB)
-        set(paddle_fluid_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/libpaddle_fluid.lib
-                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_fluid.*)
+        set(paddle_inference_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/libpaddle_inference.lib
+                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.*)
     else()
-        set(paddle_fluid_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_fluid.dll
-                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_fluid.lib)
+        set(paddle_inference_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.dll
+                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.lib)
     endif()
     copy(inference_lib_dist
-            SRCS  ${src_dir}/inference/api/paddle_*.h ${paddle_fluid_lib}
+            SRCS  ${src_dir}/inference/api/paddle_*.h ${paddle_inference_lib}
             DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib
             ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib)
 else(WIN32)
-    set(paddle_fluid_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*)
+    set(paddle_inference_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_inference.*)
     copy(inference_lib_dist
-                SRCS  ${src_dir}/inference/api/paddle_*.h ${paddle_fluid_lib}
+                SRCS  ${src_dir}/inference/api/paddle_*.h ${paddle_inference_lib}
                 DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib)
 endif(WIN32)
 
@@ -196,13 +196,13 @@ copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_C_INSTALL_DIR})
 
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 if(WIN32)
-  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/${CMAKE_BUILD_TYPE}/paddle_fluid_c.*)
+  set(paddle_inference_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/${CMAKE_BUILD_TYPE}/paddle_inference_c.*)
 else(WIN32)
-  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_fluid_c.*)
+  set(paddle_inference_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_inference_c.*)
 endif(WIN32)
 
 copy(inference_lib_dist
-      SRCS  ${src_dir}/inference/capi/paddle_c_api.h  ${paddle_fluid_c_lib}
+      SRCS  ${src_dir}/inference/capi/paddle_c_api.h  ${paddle_inference_c_lib}
       DSTS  ${PADDLE_INFERENCE_C_INSTALL_DIR}/paddle/include ${PADDLE_INFERENCE_C_INSTALL_DIR}/paddle/lib)
 
 # fluid library for both train and inference
@@ -213,12 +213,12 @@ set(dst_dir "${PADDLE_INSTALL_DIR}/paddle/fluid")
 set(module "inference")
 if(WIN32)
         copy(fluid_lib_dist
-                SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/api/paddle_*.h ${paddle_fluid_lib}
+                SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/api/paddle_*.h ${paddle_inference_lib}
                 DSTS ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module}
                 )
         else()
         copy(fluid_lib_dist
-                SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/api/paddle_*.h ${paddle_fluid_lib}
+                SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/api/paddle_*.h ${paddle_inference_lib}
                 DSTS ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module} 
                 )
 endif()
diff --git a/paddle/fluid/inference/CMakeLists.txt b/paddle/fluid/inference/CMakeLists.txt
index fb55d5463621ec1fa4e6fb8dcbfa03a85fad51ee..8ef6bcd8600c8521afd4eafe2d0943eca77b877a 100644
--- a/paddle/fluid/inference/CMakeLists.txt
+++ b/paddle/fluid/inference/CMakeLists.txt
@@ -17,8 +17,7 @@ if(WITH_TESTING)
   include(tests/test.cmake) # some generic cmake function for inference
 endif()
 
-# TODO(panyx0718): Should this be called paddle_fluid_inference_api_internal?
-cc_library(paddle_fluid_api
+cc_library(paddle_inference_io
     SRCS io.cc
     DEPS paddle_framework ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
 
@@ -46,15 +45,15 @@ set(STATIC_INFERENCE_API paddle_inference_api analysis_predictor
         analysis_config paddle_pass_builder activation_functions ${mkldnn_quantizer_cfg})
 #TODO(wilber, T8T9): Do we still need to support windows gpu static library?
 if(WIN32 AND WITH_GPU)
-  cc_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_API})
+  cc_library(paddle_inference DEPS ${fluid_modules} ${STATIC_INFERENCE_API})
 else()
-  create_static_lib(paddle_fluid ${fluid_modules} ${STATIC_INFERENCE_API})
+  create_static_lib(paddle_inference ${fluid_modules} ${STATIC_INFERENCE_API})
 endif()
 
 if(NOT APPLE)
   # TODO(liuyiqu: Temporarily disable the link flag because it is not support on Mac.
-  set(LINK_FLAGS "-Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/paddle_fluid.sym")
-  set_target_properties(paddle_fluid PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
+  set(LINK_FLAGS "-Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/paddle_inference.sym")
+  set_target_properties(paddle_inference PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
 endif()
 
 # C inference API
@@ -88,30 +87,30 @@ if (WITH_PSCORE)
 endif ()
 
 # Create shared inference library
-cc_library(paddle_fluid_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
+cc_library(paddle_inference_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
     DEPS ${SHARED_INFERENCE_DEPS})
 
 get_property(os_dependency_modules GLOBAL PROPERTY OS_DEPENDENCY_MODULES)
-target_link_libraries(paddle_fluid_shared ${os_dependency_modules})
+target_link_libraries(paddle_inference_shared ${os_dependency_modules})
 if(WIN32)
-    target_link_libraries(paddle_fluid_shared gflags)
+    target_link_libraries(paddle_inference_shared gflags)
 endif()
 
-set_target_properties(paddle_fluid_shared PROPERTIES OUTPUT_NAME paddle_fluid)
+set_target_properties(paddle_inference_shared PROPERTIES OUTPUT_NAME paddle_inference)
 if(NOT APPLE AND NOT WIN32)
   # TODO(liuyiqun): Temporarily disable the link flag because it is not support on Mac.
-  set(LINK_FLAGS "-Wl,--version-script ${CMAKE_CURRENT_SOURCE_DIR}/paddle_fluid.map")
-  set_target_properties(paddle_fluid_shared PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
+  set(LINK_FLAGS "-Wl,--version-script ${CMAKE_CURRENT_SOURCE_DIR}/paddle_inference.map")
+  set_target_properties(paddle_inference_shared PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
   # check symbol hidden
   FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/check_symbol.cmake
     "execute_process(COMMAND sh -c \"${CMAKE_CURRENT_SOURCE_DIR}/check_symbol.sh"
-    " ${CMAKE_CURRENT_BINARY_DIR}/libpaddle_fluid.so\" RESULT_VARIABLE symbol_res)\n"
+    " ${CMAKE_CURRENT_BINARY_DIR}/libpaddle_inference.so\" RESULT_VARIABLE symbol_res)\n"
     "if(NOT \"\${symbol_res}\" STREQUAL \"0\")\n"
     "  message(FATAL_ERROR \"Check symbol failed.\")\n"
     "endif()\n")
   add_custom_command(
     OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/.check_symbol"
     COMMAND ${CMAKE_COMMAND} -P "${CMAKE_CURRENT_BINARY_DIR}/check_symbol.cmake"
-    DEPENDS paddle_fluid_shared)
+    DEPENDS paddle_inference_shared)
   add_custom_target(check_symbol ALL DEPENDS "${CMAKE_CURRENT_BINARY_DIR}/.check_symbol")
 endif()
diff --git a/paddle/fluid/inference/analysis/CMakeLists.txt b/paddle/fluid/inference/analysis/CMakeLists.txt
index 98554ed04976670c1a846cbeab69815417c0a998..dab1b9f7b113523b12f468886b56d5038fe3f165 100644
--- a/paddle/fluid/inference/analysis/CMakeLists.txt
+++ b/paddle/fluid/inference/analysis/CMakeLists.txt
@@ -1,13 +1,13 @@
 unset(analysis_deps CACHE)
 set(analysis_deps # analysis_deps can be extended accross the project
-        framework_proto proto_desc graph pass paddle_fluid_api executor pretty_log
+        framework_proto proto_desc graph pass paddle_inference_io executor pretty_log
         ir_pass_manager
         CACHE INTERNAL "")
 
 add_subdirectory(ir_passes)
 add_subdirectory(passes)
 
-cc_library(analysis_helper SRCS helper.cc DEPS framework_proto proto_desc graph paddle_fluid_api)
+cc_library(analysis_helper SRCS helper.cc DEPS framework_proto proto_desc graph paddle_inference_io)
 
 cc_library(ir_pass_manager SRCS ir_pass_manager.cc DEPS graph pass ${INFER_IR_PASSES} analysis_helper)
 
@@ -62,7 +62,7 @@ endfunction(inference_analysis_test)
 if (NOT APPLE AND NOT WIN32)
   inference_analysis_test(test_analyzer
     SRCS analyzer_tester.cc
-    EXTRA_DEPS reset_tensor_array paddle_fluid_shared
+    EXTRA_DEPS reset_tensor_array paddle_inference_shared
     ARGS --inference_model_dir=${WORD2VEC_MODEL_DIR})
 elseif(WIN32)
     inference_analysis_test(test_analyzer
diff --git a/paddle/fluid/inference/analysis/ir_passes/lite_subgraph_pass.cc b/paddle/fluid/inference/analysis/ir_passes/lite_subgraph_pass.cc
index 4402d5c595a2370d7d6c451c8f051d18d239d207..c697914904b3e949050d6e61a7edb521ad6dd0e5 100644
--- a/paddle/fluid/inference/analysis/ir_passes/lite_subgraph_pass.cc
+++ b/paddle/fluid/inference/analysis/ir_passes/lite_subgraph_pass.cc
@@ -271,9 +271,14 @@ void LiteSubgraphPass::SetUpEngine(
       paddle::lite_api::Place({target_type, precision_type}),
       paddle::lite_api::Place({target_type, PRECISION(kInt64)}),
       paddle::lite_api::Place({target_type, PRECISION(kFloat)}),
-      paddle::lite_api::Place({TARGET(kHost), PRECISION(kFloat)}),
+#ifdef PADDLE_WITH_ARM
+      paddle::lite_api::Place({TARGET(kARM), precision_type}),
+      paddle::lite_api::Place({TARGET(kARM), PRECISION(kFloat)}),
+#else
       paddle::lite_api::Place({TARGET(kX86), precision_type}),
       paddle::lite_api::Place({TARGET(kX86), PRECISION(kFloat)}),
+#endif
+      paddle::lite_api::Place({TARGET(kHost), PRECISION(kFloat)}),
   };
   config.cpu_math_library_num_threads = cpu_math_library_num_threads;
   config.xpu_l3_workspace_size = xpu_l3_workspace_size;
diff --git a/paddle/fluid/inference/api/CMakeLists.txt b/paddle/fluid/inference/api/CMakeLists.txt
index 8bf4c5499db85fcfd64bd5350667c0e71fb6d220..22aa210c97ef821f6d18b704c71d8b23eeb74944 100755
--- a/paddle/fluid/inference/api/CMakeLists.txt
+++ b/paddle/fluid/inference/api/CMakeLists.txt
@@ -50,7 +50,7 @@ cc_test(test_paddle_inference_api SRCS api_tester.cc DEPS paddle_inference_api)
 
 if(WITH_TESTING)
   if (NOT APPLE AND NOT WIN32)
-    inference_base_test(test_api_impl SRCS api_impl_tester.cc DEPS paddle_fluid_shared
+    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})
     set_tests_properties(test_api_impl PROPERTIES DEPENDS test_image_classification)
   elseif(WIN32)
@@ -62,7 +62,7 @@ if(WITH_TESTING)
 endif()
 
 if (NOT APPLE AND NOT WIN32)
-  cc_test(test_analysis_predictor SRCS analysis_predictor_tester.cc DEPS paddle_fluid_shared
+  cc_test(test_analysis_predictor SRCS analysis_predictor_tester.cc DEPS paddle_inference_shared
           ARGS --dirname=${WORD2VEC_MODEL_DIR})
 elseif (WIN32)
   cc_test(test_analysis_predictor SRCS analysis_predictor_tester.cc DEPS analysis_predictor benchmark ${inference_deps}
diff --git a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
index a09f5776c71f570a77a5162d5070c2f65ff24596..e24d83af2f3685f94c79ecf7b6be0dcb11b72f7d 100644
--- a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
+++ b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
@@ -132,12 +132,12 @@ else()
 endif()
 
 if(WITH_STATIC_LIB)
-  set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
+  set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_inference${CMAKE_STATIC_LIBRARY_SUFFIX})
 else()
   if(WIN32)
-    set(DEPS ${PADDLE_LIB}/paddle/lib/paddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
+    set(DEPS ${PADDLE_LIB}/paddle/lib/paddle_inference${CMAKE_STATIC_LIBRARY_SUFFIX})
   else()
-    set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
+    set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_inference${CMAKE_SHARED_LIBRARY_SUFFIX})
   endif()
 endif()
 
@@ -204,7 +204,7 @@ if(WIN32)
   endif()
   if(NOT WITH_STATIC_LIB)
       add_custom_command(TARGET ${DEMO_NAME} POST_BUILD 
-        COMMAND ${CMAKE_COMMAND} -E copy "${PADDLE_LIB}/paddle/lib/paddle_fluid.dll" ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
+        COMMAND ${CMAKE_COMMAND} -E copy "${PADDLE_LIB}/paddle/lib/paddle_inference.dll" ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
       )
   endif()
 endif()
diff --git a/paddle/fluid/inference/api/demo_ci/windows_inference.md b/paddle/fluid/inference/api/demo_ci/windows_inference.md
index 44b2586ad6d33ce7cbd2bb3080acc96b5e27f660..73938cb995f17a0ccd8df7effce18c4f81c03916 100644
--- a/paddle/fluid/inference/api/demo_ci/windows_inference.md
+++ b/paddle/fluid/inference/api/demo_ci/windows_inference.md
@@ -1,14 +1,14 @@
 # windows inference
-本文介绍windows inference，目前只提供了静态编译，编译出paddle_fluid.lib，包含了除openblas.dll之外的所有第三方依赖库。
+本文介绍windows inference，目前只提供了静态编译，编译出paddle_inference.lib，包含了除openblas.dll之外的所有第三方依赖库。
 
-1. 下载最新的paddle_fluid.lib和openblas.dll，并把它们放在同一个目录下。
+1. 下载最新的paddle_inference.lib和openblas.dll，并把它们放在同一个目录下。
 
 2. 准备预训练好的模型文件，例如models中的模型，可以将模型用safe_inference_model接口保存下来。将模型文件放到该目录下
 
 3. 进入Paddle/paddle/fluid/inference/api/demo_ci目录，新建build目录，然后使用cmake生成vs2015的solution文件。
-其中PADDLE_LIB是前面的paddle_fluid.lib对应文件夹, CUDA_LIB指定为x64格式下的cuda系统库目录文件夹。
+其中PADDLE_LIB是前面的paddle_inference.lib对应文件夹, CUDA_LIB指定为x64格式下的cuda系统库目录文件夹。
 ```shell
- cmake .. -G "Visual Studio 14 2015 Win64" -DWITH_GPU=ON -DWITH_MKL=OFF -DWITH_STATIC_LIB=ON -DCMAKE_BUILD_TYPE=Release -DDEMO_NAME=inference_icnet -DPADDLE_LIB=D:\to_the_paddle_fluid.lib -DCUDA_LIB=D:\tools\v8.0\lib\x64
+ cmake .. -G "Visual Studio 14 2015 Win64" -DWITH_GPU=ON -DWITH_MKL=OFF -DWITH_STATIC_LIB=ON -DCMAKE_BUILD_TYPE=Release -DDEMO_NAME=inference_icnet -DPADDLE_LIB=D:\to_the_paddle_inference.lib -DCUDA_LIB=D:\tools\v8.0\lib\x64
 ```
 然后用vs2015打开对应的项目文件，注意使用静态链接 "/MT"，生成对应的exe。将openblas.dll放到exe所在目录。
 
diff --git a/paddle/fluid/inference/api/high_level_api.md b/paddle/fluid/inference/api/high_level_api.md
index 6c4471d868acde0b897db31a1aaa1a01f592aa8e..5b90c7d369c57a9f5ebf76e201dbfcd7b9fc790e 100644
--- a/paddle/fluid/inference/api/high_level_api.md
+++ b/paddle/fluid/inference/api/high_level_api.md
@@ -1,7 +1,7 @@
 # Inference High-level APIs
 This document describes the high-level inference APIs, one can use them to deploy a Paddle model for an application quickly.
 
-The APIs are described in `paddle_inference_api.h`, just one header file, and two libaries `libpaddle_fluid.so` and `libpaddle_fluid_api.so` are needed for a deployment.
+The APIs are described in `paddle_inference_api.h`, just one header file, and two libaries `libpaddle_inference.so` and `libpaddle_inference_io.so` are needed for a deployment.
 
 ## PaddleTensor
 We provide the `PaddleTensor` data structure to give a general tensor interface.
diff --git a/paddle/fluid/inference/api/high_level_api_cn.md b/paddle/fluid/inference/api/high_level_api_cn.md
index 5a15bf7893f1106222ffa5d7eb0cafaf994322f8..0d420f3369742a73f2592cd780ed4bc3bebe9439 100644
--- a/paddle/fluid/inference/api/high_level_api_cn.md
+++ b/paddle/fluid/inference/api/high_level_api_cn.md
@@ -5,7 +5,7 @@
 预测库包含:
 
 - 头文件 `paddle_inference_api.h` 定义了所有的接口
-- 库文件 `libpaddle_fluid.so/.a(Linux/Mac)` `libpaddle_fluid.lib/paddle_fluid.dll(Windows)` 
+- 库文件 `libpaddle_inference.so/.a(Linux/Mac)` `libpaddle_inference.lib/paddle_inference.dll(Windows)` 
 
 下面是详细的一些 API 概念介绍
 
@@ -76,7 +76,7 @@ CHECK(predictor->Run(slots, &outputs));
 // 获取 outputs ...
 ```
 
-编译时，联编 `libpaddle_fluid.a/.so(Linux/Mac)` 或 `libpaddle_fluid.lib/paddle_fluid.dll(Windows)` 便可。
+编译时，联编 `libpaddle_inference.a/.so(Linux/Mac)` 或 `libpaddle_inference.lib/paddle_inference.dll(Windows)` 便可。
 
 ## 详细代码参考
 
diff --git a/paddle/fluid/inference/api/paddle_pass_builder.cc b/paddle/fluid/inference/api/paddle_pass_builder.cc
index 4d40334cbc0b1f9f7cd1339fc87fdd368700c7f0..b7291ef3077df9120730676d80cdfacd768a699a 100644
--- a/paddle/fluid/inference/api/paddle_pass_builder.cc
+++ b/paddle/fluid/inference/api/paddle_pass_builder.cc
@@ -134,11 +134,15 @@ GpuPassStrategy::GpuPassStrategy() : PassStrategy({}) {
         "fc_elementwise_layernorm_fuse_pass",        //
 #if CUDNN_VERSION >= 7100  // To run conv_fusion, the version of cudnn must be
                            // guaranteed at least v7
+// cudnn8.0 has memory leak problem in conv + eltwise + act, so we
+// disable the pass.
+#if !(CUDNN_VERSION >= 8000 && CUDNN_VERSION < 8100)
         "conv_elementwise_add_act_fuse_pass",   //
         "conv_elementwise_add2_act_fuse_pass",  //
-        "conv_elementwise_add_fuse_pass",       //
-#endif                                          //
-        "transpose_flatten_concat_fuse_pass",   //
+#endif
+        "conv_elementwise_add_fuse_pass",      //
+#endif                                         //
+        "transpose_flatten_concat_fuse_pass",  //
         // following pass should be located in the last, since it will
         // work on all fused ops.
         "runtime_context_cache_pass"
diff --git a/paddle/fluid/inference/capi/CMakeLists.txt b/paddle/fluid/inference/capi/CMakeLists.txt
index 7a555279f85089331d3e66f11f62574c4ae8b697..32f780122bcd61e0ef9ec53ad3380d11f54345aa 100644
--- a/paddle/fluid/inference/capi/CMakeLists.txt
+++ b/paddle/fluid/inference/capi/CMakeLists.txt
@@ -15,15 +15,15 @@
 
 set(C_API_SRCS pd_config.cc pd_predictor.cc pd_tensor.cc c_api.cc)
 
-cc_library(paddle_fluid_c SRCS ${C_API_SRCS} DEPS paddle_fluid)
+cc_library(paddle_inference_c SRCS ${C_API_SRCS} DEPS paddle_inference)
 
 if(NOT ON_INFER)
     return()
 endif()
 
 # Create inference capi shared library
-cc_library(paddle_fluid_c_shared SHARED SRCS ${C_API_SRCS} DEPS paddle_fluid)
-set_target_properties(paddle_fluid_c_shared PROPERTIES OUTPUT_NAME paddle_fluid_c)
+cc_library(paddle_inference_c_shared SHARED SRCS ${C_API_SRCS} DEPS paddle_inference)
+set_target_properties(paddle_inference_c_shared PROPERTIES OUTPUT_NAME paddle_inference_c)
 if(WIN32)
-    target_link_libraries(paddle_fluid_c_shared shlwapi.lib)
+    target_link_libraries(paddle_inference_c_shared shlwapi.lib)
 endif()
diff --git a/paddle/fluid/inference/paddle_fluid.map b/paddle/fluid/inference/paddle_inference.map
similarity index 100%
rename from paddle/fluid/inference/paddle_fluid.map
rename to paddle/fluid/inference/paddle_inference.map
diff --git a/paddle/fluid/inference/paddle_fluid.sym b/paddle/fluid/inference/paddle_inference.sym
similarity index 100%
rename from paddle/fluid/inference/paddle_fluid.sym
rename to paddle/fluid/inference/paddle_inference.sym
diff --git a/paddle/fluid/inference/tests/api/CMakeLists.txt b/paddle/fluid/inference/tests/api/CMakeLists.txt
index 2fa076b002715e79f50f95ca25ffad23e9cb2cf5..a173328e64ae5ec682fe6b729b8f1ee4f86fb867 100644
--- a/paddle/fluid/inference/tests/api/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/api/CMakeLists.txt
@@ -1,7 +1,7 @@
 if (NOT APPLE AND NOT WIN32)
-    set(INFERENCE_EXTRA_DEPS paddle_fluid_shared)
+    set(INFERENCE_EXTRA_DEPS paddle_inference_shared)
 else()
-    set(INFERENCE_EXTRA_DEPS paddle_inference_api paddle_fluid_api ir_pass_manager analysis_predictor benchmark)
+    set(INFERENCE_EXTRA_DEPS paddle_inference_api paddle_inference_io ir_pass_manager analysis_predictor benchmark)
 endif()
 
 if(WITH_GPU AND TENSORRT_FOUND)
@@ -508,10 +508,10 @@ if(WITH_GPU AND TENSORRT_FOUND)
             EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} 
             ARGS --infer_model=${TEST_INSTANCE_NORM_MODEL}/)
     inference_analysis_test(test_analyzer_capi_gpu SRCS analyzer_capi_gpu_tester.cc
-            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
             ARGS --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
     inference_analysis_test(test_analyzer_capi_xpu SRCS analyzer_capi_xpu_tester.cc
-            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
             ARGS --infer_model=${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models)
             
     set(TRT_MODEL_QUANT_RESNET_DIR "${INFERENCE_DEMO_INSTALL_DIR}/small_quant_model")
@@ -593,11 +593,11 @@ download_data(${LITE_MODEL_INSTALL_DIR} "mul_model_fp32.tgz")
 #        ARGS --infer_model=${RESNET50_MODEL_DIR})
 
 inference_analysis_test(test_analyzer_capi SRCS analyzer_capi_tester.cc
-            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
             ARGS --infer_model=${RESNET50_MODEL_DIR}/model)
 
 inference_analysis_test(test_analyzer_capi_pd_tensor SRCS analyzer_capi_pd_tensor_tester.cc
-            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
             ARGS --infer_model=${MOBILENET_INSTALL_DIR}/model)
 
 inference_analysis_test(test_analyzer_zerocopytensor_tensor SRCS analyzer_zerocopy_tensor_tester.cc
@@ -610,12 +610,12 @@ inference_analysis_test(test_analyzer_paddletensor_tensor SRCS analyzer_paddle_t
             
 if(WITH_MKLDNN)
   inference_analysis_test(test_analyzer_capi_int SRCS analyzer_capi_int_tester.cc
-            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+            EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
             ARGS --infer_model=${INT8_DATA_DIR}/resnet50/model)
  endif()
 
 inference_analysis_test(test_analyzer_capi_ner SRCS analyzer_capi_ner_tester.cc 
-        EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_fluid_c
+        EXTRA_DEPS ${INFERENCE_EXTRA_DEPS} paddle_inference_c
         ARGS --infer_model=${CHINESE_NER_INSTALL_DIR}/model)
 
 if(WITH_GPU)
diff --git a/paddle/fluid/train/CMakeLists.txt b/paddle/fluid/train/CMakeLists.txt
index 8f360d77967050e926a71b8b3cd2ddc1bbd25c8a..0688c63cac3f3fdd9e8b2a215c5384d8e175bd3c 100644
--- a/paddle/fluid/train/CMakeLists.txt
+++ b/paddle/fluid/train/CMakeLists.txt
@@ -7,12 +7,12 @@ function(train_test TARGET_NAME)
     if (NOT APPLE AND NOT WIN32)
         cc_test(test_train_${TARGET_NAME}
                 SRCS test_train_${TARGET_NAME}.cc
-                DEPS paddle_fluid_shared
+                DEPS paddle_inference_shared
                 ARGS --dirname=${PYTHON_TESTS_DIR}/book/)
     else()
         cc_test(test_train_${TARGET_NAME}
                 SRCS test_train_${TARGET_NAME}.cc
-                DEPS paddle_fluid_api
+                DEPS paddle_inference_io
                 ARGS --dirname=${PYTHON_TESTS_DIR}/book/)
     endif()
     if(TEST test_train_${TARGET_NAME})
diff --git a/paddle/fluid/train/demo/CMakeLists.txt b/paddle/fluid/train/demo/CMakeLists.txt
index 57fda493a81102096ab6c152379c9557e50d6828..95da77d68d482afd8e25c79d8d1ec5e045bbe738 100644
--- a/paddle/fluid/train/demo/CMakeLists.txt
+++ b/paddle/fluid/train/demo/CMakeLists.txt
@@ -69,7 +69,7 @@ endif(APPLE)
 target_link_libraries(demo_trainer
         ${MACOS_LD_FLAGS}
         ${ARCHIVE_START}
-        ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid.so
+        ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_inference.so
         ${ARCHIVE_END}
         ${MATH_LIB}
         ${MKLDNN_LIB}
diff --git a/paddle/fluid/train/imdb_demo/CMakeLists.txt b/paddle/fluid/train/imdb_demo/CMakeLists.txt
index 29d54d0d2fbf61cae9f1cf8505124a969c029005..e943d6bc78eab08caa1d513973639ea9dad13b2c 100644
--- a/paddle/fluid/train/imdb_demo/CMakeLists.txt
+++ b/paddle/fluid/train/imdb_demo/CMakeLists.txt
@@ -68,7 +68,7 @@ endif(APPLE)
 target_link_libraries(demo_trainer
 	${MACOS_LD_FLAGS}
 	${ARCHIVE_START}
-	${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid.so
+	${PADDLE_LIB}/paddle/fluid/inference/libpaddle_inference.so
 	${ARCHIVE_END}
 	${MATH_LIB}
 	${MKLDNN_LIB}