diff --git a/paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc b/paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc
index defa0a525f6885a6206f54e6a27190ba181b74a2..5132b3b5e72ca24c04e53d0157d33676d45b2a2a 100644
--- a/paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc
+++ b/paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc
@@ -96,6 +96,7 @@ void MemoryOptimizePass::CollectVarMemorySize(
   const int fake_batch_size = 1;
 
   auto valid_var = [&](framework::ir::Node* node) -> bool {
+    // lod operator reuse may cause unknown errors.
     std::set<std::string> invalid_op = {"while",
                                         "conditional_block",
                                         "tensorrt_engine",
@@ -103,6 +104,7 @@ void MemoryOptimizePass::CollectVarMemorySize(
                                         "merge_lod_tensor_infer",
                                         "merge_lod_tensor",
                                         "equal",
+                                        "sequence_pool",
                                         "lod_reset"};
     for (auto* tmp : node->inputs) {
       CHECK(tmp->IsOp());
diff --git a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
index b7e8f40e40859fb54d502d612cdb3ce2c4aff614..a09f5776c71f570a77a5162d5070c2f65ff24596 100644
--- a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
+++ b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
@@ -83,14 +83,24 @@ if (USE_TENSORRT AND WITH_GPU)
   endif()
   set(TENSORRT_INCLUDE_DIR ${TENSORRT_ROOT}/include)
   set(TENSORRT_LIB_DIR ${TENSORRT_ROOT}/lib)
-endif()
-
-if (NOT WIN32)
-  if (USE_TENSORRT AND WITH_GPU)
-      include_directories("${TENSORRT_INCLUDE_DIR}")
-      link_directories("${TENSORRT_LIB_DIR}")
+  file(READ ${TENSORRT_INCLUDE_DIR}/NvInfer.h TENSORRT_VERSION_FILE_CONTENTS)
+  string(REGEX MATCH "define NV_TENSORRT_MAJOR +([0-9]+)" TENSORRT_MAJOR_VERSION
+    "${TENSORRT_VERSION_FILE_CONTENTS}")
+  if("${TENSORRT_MAJOR_VERSION}" STREQUAL "")
+    file(READ ${TENSORRT_INCLUDE_DIR}/NvInferVersion.h TENSORRT_VERSION_FILE_CONTENTS)
+    string(REGEX MATCH "define NV_TENSORRT_MAJOR +([0-9]+)" TENSORRT_MAJOR_VERSION
+      "${TENSORRT_VERSION_FILE_CONTENTS}")
   endif()
-endif(NOT WIN32)
+  if("${TENSORRT_MAJOR_VERSION}" STREQUAL "")
+    message(SEND_ERROR "Failed to detect TensorRT version.")
+  endif()
+  string(REGEX REPLACE "define NV_TENSORRT_MAJOR +([0-9]+)" "\\1"
+    TENSORRT_MAJOR_VERSION "${TENSORRT_MAJOR_VERSION}")
+  message(STATUS "Current TensorRT header is ${TENSORRT_INCLUDE_DIR}/NvInfer.h. "
+    "Current TensorRT version is v${TENSORRT_MAJOR_VERSION}. ")
+  include_directories("${TENSORRT_INCLUDE_DIR}")
+  link_directories("${TENSORRT_LIB_DIR}")
+endif()
 
 if(WITH_MKL)
   set(MATH_LIB_PATH "${PADDLE_LIB_THIRD_PARTY_PATH}mklml")
@@ -147,14 +157,17 @@ endif(NOT WIN32)
 if(WITH_GPU)
   if(NOT WIN32)
     if (USE_TENSORRT)
-      set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer${CMAKE_STATIC_LIBRARY_SUFFIX})
-      set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer_plugin${CMAKE_STATIC_LIBRARY_SUFFIX})
+      set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer${CMAKE_SHARED_LIBRARY_SUFFIX})
+      set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer_plugin${CMAKE_SHARED_LIBRARY_SUFFIX})
     endif()
     set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX})
   else()
     if(USE_TENSORRT)
       set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/nvinfer${CMAKE_STATIC_LIBRARY_SUFFIX})
       set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/nvinfer_plugin${CMAKE_STATIC_LIBRARY_SUFFIX})
+      if(${TENSORRT_MAJOR_VERSION} GREATER_EQUAL 7)
+        set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/myelin64_1${CMAKE_STATIC_LIBRARY_SUFFIX})
+      endif()
     endif()
     set(DEPS ${DEPS} ${CUDA_LIB}/cudart${CMAKE_STATIC_LIBRARY_SUFFIX} )
     set(DEPS ${DEPS} ${CUDA_LIB}/cublas${CMAKE_STATIC_LIBRARY_SUFFIX} )
@@ -172,6 +185,11 @@ if(WIN32)
             COMMAND ${CMAKE_COMMAND} -E copy ${TENSORRT_LIB_DIR}/nvinfer_plugin${CMAKE_SHARED_LIBRARY_SUFFIX}
               ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
     )
+    if(${TENSORRT_MAJOR_VERSION} GREATER_EQUAL 7)
+      add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
+              COMMAND ${CMAKE_COMMAND} -E copy ${TENSORRT_LIB_DIR}/myelin64_1${CMAKE_SHARED_LIBRARY_SUFFIX}
+                ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE})
+    endif()
   endif()
   if(WITH_MKL)
     add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
diff --git a/paddle/fluid/platform/flags.cc b/paddle/fluid/platform/flags.cc
index 378071964fc6bfc86617b72cfb861e1d384f9f75..20be80b17617432e81bd70cebb6eeeae4626e5ef 100644
--- a/paddle/fluid/platform/flags.cc
+++ b/paddle/fluid/platform/flags.cc
@@ -498,8 +498,14 @@ DEFINE_bool(use_mkldnn, false, "Use MKLDNN to run");
  * If FLAGS_call_stack_level == 2, the python stack, c++ stack, and error
  * message summary will be shown.
  */
+#ifdef PADDLE_ON_INFERENCE
+static const int32_t kDefaultCallStackLevel = 2;
+#else
+static const int32_t kDefaultCallStackLevel = 1;
+#endif
+
 DEFINE_int32(
-    call_stack_level, 1,
+    call_stack_level, kDefaultCallStackLevel,
     "Determine the call stack to print when error or exeception happens."
     // TODO(zhiqiu): implement logic of FLAGS_call_stack_level==0
     // "If FLAGS_call_stack_level == 0, only the error message summary will be "
diff --git a/paddle/fluid/platform/port.h b/paddle/fluid/platform/port.h
index b2f26ba9581e08a6f0c1f16756d3708673805d6d..453bea625b0ab22bf1f5bea0c6530f40f4993752 100644
--- a/paddle/fluid/platform/port.h
+++ b/paddle/fluid/platform/port.h
@@ -47,6 +47,7 @@ static void *dlsym(void *handle, const char *symbol_name) {
   found_symbol = GetProcAddress((HMODULE)handle, symbol_name);
 
   if (found_symbol == NULL) {
+    LOG(ERROR) << "Load symbol " << symbol_name << " failed.";
     throw std::runtime_error(std::string(symbol_name) + " not found.");
   }
   return reinterpret_cast<void *>(found_symbol);