diff --git a/cmake/external/lite.cmake b/cmake/external/lite.cmake
index 6c837c97089b7f3997bc1f669cd9e71976cb6980..2617bdd834d1867eb6c2c40dac944aefe0f2a783 100644
--- a/cmake/external/lite.cmake
+++ b/cmake/external/lite.cmake
@@ -44,7 +44,7 @@ if (NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
       ${LITE_PROJECT}
       ${EXTERNAL_PROJECT_LOG_ARGS}
       GIT_REPOSITORY      "https://github.com/PaddlePaddle/Paddle-Lite.git"
-      GIT_TAG             922ace19a45f30075618f71428523e7a2d5898d6
+      GIT_TAG             0f875ef367bd2dbfa2e557eb2a2fc841bacdf6cf
       PREFIX              ${LITE_SOURCES_DIR}
       UPDATE_COMMAND      ""
       BUILD_COMMAND       ${LITE_BUILD_COMMAND}
diff --git a/paddle/fluid/inference/lite/op_teller.cc b/paddle/fluid/inference/lite/op_teller.cc
index fd7a5da7cec6f12d39bd06e70deed5989ef5dc43..c5f1eccc3334a109aa0a6a21ae8c189d42e18447 100644
--- a/paddle/fluid/inference/lite/op_teller.cc
+++ b/paddle/fluid/inference/lite/op_teller.cc
@@ -28,7 +28,7 @@ namespace lite {
 struct SimpleOpTeller : public Teller {
   SimpleOpTeller() {
     const std::map<std::string, std::string>& op2path =
-        OpKernelInfoCollector::Global().GetOp2PathDict();
+        paddle::lite::GetOp2PathDict();
     auto is_non_inst = [](const std::string& op) -> bool {
       const std::vector<std::string> ops = {"feed", "fetch", "while"};
       return std::find(ops.begin(), ops.end(), op) != ops.end();
diff --git a/paddle/fluid/inference/lite/tensor_utils.cc b/paddle/fluid/inference/lite/tensor_utils.cc
index 9f361d563b054652f44ded5ab015d3d9cf683ee0..8e7ebd3119e976555b6281a73601723f48dcf12a 100644
--- a/paddle/fluid/inference/lite/tensor_utils.cc
+++ b/paddle/fluid/inference/lite/tensor_utils.cc
@@ -157,7 +157,10 @@ void TensorCopyAsync(paddle::lite::Tensor* dst, const framework::LoDTensor& src,
   dst->Resize(framework::vectorize(src.dims()));
   const void* src_data = src.data<void>();
   void* dst_data = dst->mutable_data(bytes);
+  VLOG(3) << "[CopyAsync fluid -> lite] Bytes = " << bytes << ", src = " << &src
+          << ", dst = " << dst << ", src_type = " << src.type();
   MemoryCopyAsync(dst_place, dst_data, src_place, src_data, bytes, ctx);
+  VLOG(3) << "[Lite memory size] Bytes = " << dst->memory_size();
 }
 
 template <>
@@ -172,7 +175,10 @@ void TensorCopyAsync(framework::LoDTensor* dst, const paddle::lite::Tensor& src,
   const void* src_data = src.raw_data();
   // When Lite is ready, the source type needs to be modified here.
   void* dst_data = dst->mutable_data(dst_place, dst->type());
+  VLOG(3) << "[CopyAsync lite -> fluid] Bytes = " << bytes << ", src = " << &src
+          << ", dst = " << dst << ", src_type = " << dst->type();
   MemoryCopyAsync(dst_place, dst_data, src_place, src_data, bytes, ctx);
+  VLOG(3) << "[Lite memory size] Bytes = " << src.memory_size();
 }
 
 }  // namespace utils
diff --git a/paddle/fluid/operators/lite/lite_engine_op.h b/paddle/fluid/operators/lite/lite_engine_op.h
index 62bbef663235a7a930ede47cc3f8a31921276796..3b48615338f729a56db133a2072ceea5e8e94b22 100644
--- a/paddle/fluid/operators/lite/lite_engine_op.h
+++ b/paddle/fluid/operators/lite/lite_engine_op.h
@@ -77,7 +77,8 @@ class LiteEngineOp : public framework::OperatorBase {
           inference::analysis::GetFromScope<framework::LoDTensor>(scope,
                                                                   in_names_[i]);
       paddle::lite::Tensor *dst_t = engine_->GetInput(i);
-      VLOG(3) << "fluid -> lite: " << in_names_[i];
+      VLOG(3) << "[Copy] fluid -> lite (" << in_names_[i] << " -> "
+              << engine_->GetInputNames()[i] << ")";
       inference::lite::utils::TensorCopyAsync(dst_t, src_t, *ctx);
     }
 #ifdef PADDLE_WITH_CUDA
@@ -94,7 +95,8 @@ class LiteEngineOp : public framework::OperatorBase {
       framework::LoDTensor *dst_t =
           &inference::analysis::GetFromScope<framework::LoDTensor>(
               scope, out_names_[i]);
-      VLOG(3) << "lite -> fluid: " << out_names_[i];
+      VLOG(3) << "[Copy] lite -> fluid (" << out_names_[i] << " -> "
+              << engine_->GetOutputNames()[i] << ")";
       inference::lite::utils::TensorCopyAsync(dst_t, src_t, *ctx);
     }
 #ifdef PADDLE_WITH_CUDA