Add paddlelite backend support (#260)

* Add paddlelite backend support * Update CMakeLists.txt * Update __init__.py
2025-10-05 00:33:03 +08:00 · 2022-09-21 13:22:34 +08:00
parent 32596b3b89
commit addce837bc
22 changed files with 370 additions and 56 deletions
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -42,7 +42,8 @@ option(WITH_GPU "Whether WITH_GPU=ON, will enable onnxruntime-gpu/paddle-infernc
 option(ENABLE_ORT_BACKEND "Whether to enable onnxruntime backend." OFF)
 option(ENABLE_TRT_BACKEND "Whether to enable tensorrt backend." OFF)
 option(ENABLE_PADDLE_BACKEND "Whether to enable paddle backend." OFF)
-option(ENABLE_OPENVINO_BACKEND "Whether to enable paddle backend." OFF)
+option(ENABLE_OPENVINO_BACKEND "Whether to enable openvino backend." OFF)
+option(ENABLE_LITE_BACKEND "Whether to enable paddle lite backend." OFF)
 option(CUDA_DIRECTORY "If build tensorrt backend, need to define path of cuda library.")
 option(TRT_DIRECTORY "If build tensorrt backend, need to define path of tensorrt library.")
 option(ENABLE_VISION "Whether to enable vision models usage." OFF)
@@ -120,10 +121,11 @@ file(GLOB_RECURSE DEPLOY_ORT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastde
 file(GLOB_RECURSE DEPLOY_PADDLE_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/paddle/*.cc)
 file(GLOB_RECURSE DEPLOY_TRT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/tensorrt/*.cc ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/tensorrt/*.cpp)
 file(GLOB_RECURSE DEPLOY_OPENVINO_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/openvino/*.cc)
+file(GLOB_RECURSE DEPLOY_LITE_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/lite/*.cc)
 file(GLOB_RECURSE DEPLOY_VISION_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/vision/*.cc)
 file(GLOB_RECURSE DEPLOY_TEXT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/text/*.cc)
 file(GLOB_RECURSE DEPLOY_PYBIND_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/pybind/*.cc ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/*_pybind.cc)
-list(REMOVE_ITEM ALL_DEPLOY_SRCS ${DEPLOY_ORT_SRCS} ${DEPLOY_PADDLE_SRCS} ${DEPLOY_TRT_SRCS} ${DEPLOY_OPENVINO_SRCS} ${DEPLOY_VISION_SRCS} ${DEPLOY_TEXT_SRCS})
+list(REMOVE_ITEM ALL_DEPLOY_SRCS ${DEPLOY_ORT_SRCS} ${DEPLOY_PADDLE_SRCS} ${DEPLOY_TRT_SRCS} ${DEPLOY_OPENVINO_SRCS} ${DEPLOY_LITE_SRCS} ${DEPLOY_VISION_SRCS} ${DEPLOY_TEXT_SRCS})

 set(DEPEND_LIBS "")

@@ -147,6 +149,12 @@ if(ENABLE_ORT_BACKEND)
  list(APPEND DEPEND_LIBS external_onnxruntime)
 endif()

+if(ENABLE_LITE_BACKEND)
+  add_definitions(-DENABLE_LITE_BACKEND)
+  include(${PROJECT_SOURCE_DIR}/cmake/paddlelite.cmake)
+  list(APPEND ALL_DEPLOY_SRCS ${DEPLOY_LITE_SRCS})
+endif()
+
 if(ENABLE_PADDLE_BACKEND)
  add_definitions(-DENABLE_PADDLE_BACKEND)
  list(APPEND ALL_DEPLOY_SRCS ${DEPLOY_PADDLE_SRCS})
--- a/cmake/paddlelite.cmake
+++ b/cmake/paddlelite.cmake
@@ -0,0 +1,72 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+include(ExternalProject)
+
+set(PADDLELITE_PROJECT "extern_paddlelite")
+set(PADDLELITE_PREFIX_DIR ${THIRD_PARTY_PATH}/paddlelite)
+set(PADDLELITE_SOURCE_DIR
+    ${THIRD_PARTY_PATH}/paddlelite/src/${PADDLELITE_PROJECT})
+set(PADDLELITE_INSTALL_DIR ${THIRD_PARTY_PATH}/install/paddlelite)
+set(PADDLELITE_INC_DIR
+    "${PADDLELITE_INSTALL_DIR}/include"
+    CACHE PATH "paddlelite include directory." FORCE)
+set(PADDLELITE_LIB_DIR
+    "${PADDLELITE_INSTALL_DIR}/lib"
+    CACHE PATH "paddlelite lib directory." FORCE)
+set(CMAKE_BUILD_RPATH "${CMAKE_BUILD_RPATH}" "${PADDLELITE_LIB_DIR}")
+
+#set(PADDLELITE_URL "https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.11/inference_lite_lib.armlinux.armv8.gcc.with_extra.with_cv.tar.gz")
+set(PADDLELITE_URL "https://bj.bcebos.com/fastdeploy/third_libs/lite-linux-arm64-20220920.tgz")
+
+if(WIN32)
+  message(FATAL_ERROR "Doesn't support windows platform with backend Paddle-Lite.")
+elseif(APPLE)
+  message(FATAL_ERROR "Doesn't support Mac OSX platform with backend Paddle-Lite.")
+else()
+  if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
+    set(PADDLELITE_URL "https://bj.bcebos.com/fastdeploy/third_libs/lite-linux-arm64-20220920.tgz")
+  else()
+    message(FATAL_ERROR "Only support Linux aarch64 now, x64 is not supported with backend Paddle-Lite.")
+  endif()
+endif()
+
+include_directories(${PADDLELITE_INC_DIR})
+
+if(WIN32)
+elseif(APPLE)
+else()
+  set(PADDLELITE_LIB "${PADDLELITE_INSTALL_DIR}/lib/libpaddle_full_api_shared.so")
+endif()
+
+ExternalProject_Add(
+  ${PADDLELITE_PROJECT}
+  ${EXTERNAL_PROJECT_LOG_ARGS}
+  URL ${PADDLELITE_URL}
+  PREFIX ${PADDLELITE_PREFIX_DIR}
+  DOWNLOAD_NO_PROGRESS 1
+  CONFIGURE_COMMAND ""
+  BUILD_COMMAND ""
+  UPDATE_COMMAND ""
+  INSTALL_COMMAND
+    ${CMAKE_COMMAND} -E remove_directory ${PADDLELITE_INSTALL_DIR} &&
+    ${CMAKE_COMMAND} -E make_directory ${PADDLELITE_INSTALL_DIR} &&
+    ${CMAKE_COMMAND} -E rename ${PADDLELITE_SOURCE_DIR}/lib/ ${PADDLELITE_INSTALL_DIR}/lib &&
+    ${CMAKE_COMMAND} -E copy_directory ${PADDLELITE_SOURCE_DIR}/include
+    ${PADDLELITE_INC_DIR}
+  BUILD_BYPRODUCTS ${PADDLELITE_LIB})
+
+add_library(external_paddle_lite STATIC IMPORTED GLOBAL)
+set_property(TARGET external_paddle_lite PROPERTY IMPORTED_LOCATION ${PADDLELITE_LIB})
+add_dependencies(external_paddle_lite ${PADDLELITE_PROJECT})
--- a/fastdeploy/backends/lite/lite_backend.cc
+++ b/fastdeploy/backends/lite/lite_backend.cc
@@ -0,0 +1,141 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/backends/lite/lite_backend.h"
+#include <cstring>
+
+namespace fastdeploy {
+
+// Convert data type from paddle lite to fastdeploy
+FDDataType LiteDataTypeToFD(const paddle::lite_api::PrecisionType& dtype) {
+  if (dtype == paddle::lite_api::PrecisionType::kFloat) {
+    return FDDataType::FP32;
+  } else if (dtype == paddle::lite_api::PrecisionType::kInt8) {
+    return FDDataType::INT8;
+  } else if (dtype == paddle::lite_api::PrecisionType::kInt32) {
+    return FDDataType::INT32;
+  } else if (dtype == paddle::lite_api::PrecisionType::kInt64) {
+    return FDDataType::INT64;
+  } else if (dtype == paddle::lite_api::PrecisionType::kInt16) {
+    return FDDataType::INT16;
+  } else if (dtype == paddle::lite_api::PrecisionType::kUInt8) {
+    return FDDataType::UINT8;
+  } else if (dtype == paddle::lite_api::PrecisionType::kFP64) {
+    return FDDataType::FP64;
+  }
+  FDASSERT(false, "Unexpected data type of %d.", dtype);
+  return FDDataType::FP32;
+}
+
+void LiteBackend::BuildOption(const LiteBackendOption& option) {
+  std::vector<paddle::lite_api::Place> valid_places;
+  valid_places.push_back(paddle::lite_api::Place{TARGET(kARM), PRECISION(kFloat)});
+  config_.set_valid_places(valid_places);
+}
+
+bool LiteBackend::InitFromPaddle(const std::string& model_file,
+                               const std::string& params_file,
+                               const LiteBackendOption& option) {
+  if (initialized_) {
+    FDERROR << "LiteBackend is already initialized, cannot initialize again."
+            << std::endl;
+    return false;
+  }
+
+  config_.set_model_file(model_file);
+  config_.set_param_file(params_file);
+  BuildOption(option);
+  predictor_ = paddle::lite_api::CreatePaddlePredictor<paddle::lite_api::CxxConfig>(config_);
+  
+  inputs_desc_.clear();
+  outputs_desc_.clear();
+  inputs_order_.clear();
+  std::vector<std::string> input_names = predictor_->GetInputNames();
+  std::vector<std::string> output_names = predictor_->GetOutputNames();
+  for (size_t i = 0; i < input_names.size(); ++i) {
+    inputs_order_[input_names[i]] = i;
+    TensorInfo info;
+    auto tensor = predictor_->GetInput(i);
+    auto shape = tensor->shape();
+    info.shape.assign(shape.begin(), shape.end());
+    info.name = input_names[i];
+    info.dtype = LiteDataTypeToFD(tensor->precision());
+    inputs_desc_.emplace_back(info);
+  }
+  for (size_t i = 0; i < output_names.size(); ++i) {
+    TensorInfo info;
+    auto tensor = predictor_->GetOutput(i);
+    auto shape = tensor->shape();
+    info.shape.assign(shape.begin(), shape.end());
+    info.name = output_names[i];
+    info.dtype = LiteDataTypeToFD(tensor->precision());
+    outputs_desc_.emplace_back(info);
+  }
+  
+  initialized_ = true;
+  return true;
+}
+
+TensorInfo LiteBackend::GetInputInfo(int index) {
+  FDASSERT(index < NumInputs(),
+           "The index: %d should less than the number of inputs: %d.", index,
+           NumInputs());
+  return inputs_desc_[index];
+}
+
+std::vector<TensorInfo> LiteBackend::GetInputInfos() { return inputs_desc_; }
+
+TensorInfo LiteBackend::GetOutputInfo(int index) {
+  FDASSERT(index < NumOutputs(),
+           "The index: %d should less than the number of outputs %d.", index,
+           NumOutputs());
+  return outputs_desc_[index];
+}
+
+std::vector<TensorInfo> LiteBackend::GetOutputInfos() {
+  return outputs_desc_;
+}
+
+bool LiteBackend::Infer(std::vector<FDTensor>& inputs,
+                          std::vector<FDTensor>* outputs) {
+  if (inputs.size() != inputs_desc_.size()) {
+    FDERROR << "[LiteBackend] Size of inputs(" << inputs.size()
+            << ") should keep same with the inputs of this model("
+            << inputs_desc_.size() << ")." << std::endl;
+    return false;
+  }
+
+  for (size_t i = 0; i < inputs.size(); ++i) {
+    auto iter = inputs_order_.find(inputs[i].name);
+    if (iter == inputs_order_.end()) {
+      FDERROR << "Cannot find input with name:" << inputs[i].name << " in loaded model." << std::endl;
+      return false;
+    }
+    auto tensor = predictor_->GetInput(iter->second);
+    tensor->Resize(inputs[i].shape);
+    tensor->ShareExternalMemory(const_cast<void*>(inputs[i].CpuData()), inputs[i].Nbytes(), paddle::lite_api::TargetType::kARM);
+  }
+
+  predictor_->Run();
+
+  outputs->resize(outputs_desc_.size());
+  for (size_t i = 0; i < outputs_desc_.size(); ++i) {
+    auto tensor = predictor_->GetOutput(i);
+    (*outputs)[i].Resize(tensor->shape(), outputs_desc_[i].dtype, outputs_desc_[i].name);
+    memcpy((*outputs)[i].MutableData(), tensor->data<void>(), (*outputs)[i].Nbytes());
+  }
+  return true;
+}
+
+}  // namespace fastdeploy
--- a/fastdeploy/backends/lite/lite_backend.h
+++ b/fastdeploy/backends/lite/lite_backend.h
@@ -0,0 +1,61 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <iostream>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "fastdeploy/backends/backend.h"
+#include "paddle_api.h"  // NOLINT
+
+namespace fastdeploy {
+
+struct LiteBackendOption {
+};
+
+// Convert data type from paddle lite to fastdeploy
+FDDataType LiteDataTypeToFD(const paddle::lite_api::PrecisionType& dtype);
+
+class LiteBackend : public BaseBackend {
+ public:
+  LiteBackend() {}
+  virtual ~LiteBackend() = default;
+  void BuildOption(const LiteBackendOption& option);
+
+  bool InitFromPaddle(const std::string& model_file,
+                      const std::string& params_file,
+                      const LiteBackendOption& option = LiteBackendOption());
+
+  bool Infer(std::vector<FDTensor>& inputs, std::vector<FDTensor>* outputs);
+
+  int NumInputs() const { return inputs_desc_.size(); }
+
+  int NumOutputs() const { return outputs_desc_.size(); }
+
+  TensorInfo GetInputInfo(int index);
+  TensorInfo GetOutputInfo(int index);
+  std::vector<TensorInfo> GetInputInfos() override;
+  std::vector<TensorInfo> GetOutputInfos() override;
+
+ private:
+  paddle::lite_api::CxxConfig config_;
+  std::shared_ptr<paddle::lite_api::PaddlePredictor> predictor_;
+  std::vector<TensorInfo> inputs_desc_;
+  std::vector<TensorInfo> outputs_desc_;
+  std::map<std::string, int> inputs_order_;
+};
+}  // namespace fastdeploy
--- a/fastdeploy/core/allocate.cc
+++ b/fastdeploy/core/allocate.cc
@@ -11,7 +11,6 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
-#pragma once

 #ifdef WITH_GPU
 #include <cuda_runtime_api.h>
--- a/fastdeploy/fastdeploy_model.cc
+++ b/fastdeploy/fastdeploy_model.cc
@@ -166,7 +166,7 @@ std::map<std::string, float> FastDeployModel::PrintStatisInfoOfRuntime() {
  std::map<std::string, float> statis_info_of_runtime_dict;

  if (time_of_runtime_.size() < 10) {
-    std::cout << "[FastDeploy] [WARNING] PrintStatisInfoOfRuntime require the runtime ran 10 times at "
+    FDWARNING << "PrintStatisInfoOfRuntime require the runtime ran 10 times at "
                 "least, but now you only ran "
              << time_of_runtime_.size() << " times." << std::endl;
  }
--- a/fastdeploy/fastdeploy_model.h
+++ b/fastdeploy/fastdeploy_model.h
@@ -75,7 +75,7 @@ class FASTDEPLOY_DECL FastDeployModel {
 #define TIMERECORD_END(id, prefix)                                           \
  if (DebugEnabled()) {                                                      \
    tc_##id.End();                                                           \
-    FDINFO << __FILE__ << "(" << __LINE__ << "):" << __FUNCTION__ << " " \
+    FDLogger() << __FILE__ << "(" << __LINE__ << "):" << __FUNCTION__ << " " \
               << prefix << " duration = " << tc_##id.Duration() << "s."     \
               << std::endl;                                                 \
  }
--- a/fastdeploy/fastdeploy_runtime.cc
+++ b/fastdeploy/fastdeploy_runtime.cc
@@ -33,6 +33,10 @@
 #include "fastdeploy/backends/openvino/ov_backend.h"
 #endif

+#ifdef ENABLE_LITE_BACKEND
+#include "fastdeploy/backends/lite/lite_backend.h"
+#endif
+
 namespace fastdeploy {

 std::vector<Backend> GetAvailableBackends() {
@@ -48,6 +52,9 @@ std::vector<Backend> GetAvailableBackends() {
 #endif
 #ifdef ENABLE_OPENVINO_BACKEND
  backends.push_back(Backend::OPENVINO);
+#endif
+#ifdef ENABLE_LITE_BACKEND
+  backends.push_back(Backend::LITE);
 #endif
  return backends;
 }
@@ -71,6 +78,8 @@ std::string Str(const Backend& b) {
    return "Backend::PDINFER";
  } else if (b == Backend::OPENVINO) {
    return "Backend::OPENVINO";
+  } else if (b == Backend::LITE) {
+    return "Backend::LITE";
  }
  return "UNKNOWN-Backend";
 }
@@ -194,6 +203,15 @@ void RuntimeOption::UseOpenVINOBackend() {
  FDASSERT(false, "The FastDeploy didn't compile with OpenVINO.");
 #endif
 }
+
+void RuntimeOption::UseLiteBackend() {
+#ifdef ENABLE_LITE_BACKEND
+  backend = Backend::LITE;
+#else
+  FDASSERT(false, "The FastDeploy didn't compile with Paddle Lite.");
+#endif
+}
+
 void RuntimeOption::EnablePaddleMKLDNN() { pd_enable_mkldnn = true; }

 void RuntimeOption::DisablePaddleMKLDNN() { pd_enable_mkldnn = false; }
@@ -262,12 +280,12 @@ bool Runtime::Init(const RuntimeOption& _option) {
    FDASSERT(option.device == Device::CPU || option.device == Device::GPU,
             "Backend::ORT only supports Device::CPU/Device::GPU.");
    CreateOrtBackend();
-    FDINFO << "Runtime initialized with Backend::ORT in device " << Str(option.device) << "." << std::endl;
+    FDINFO << "Runtime initialized with Backend::ORT in " << Str(option.device) << "." << std::endl;
  } else if (option.backend == Backend::TRT) {
    FDASSERT(option.device == Device::GPU,
             "Backend::TRT only supports Device::GPU.");
    CreateTrtBackend();
-    FDINFO << "Runtime initialized with Backend::TRT in device " << Str(option.device) << "." << std::endl;
+    FDINFO << "Runtime initialized with Backend::TRT in " << Str(option.device) << "." << std::endl;
  } else if (option.backend == Backend::PDINFER) {
    FDASSERT(option.device == Device::CPU || option.device == Device::GPU,
             "Backend::TRT only supports Device::CPU/Device::GPU.");
@@ -275,12 +293,16 @@ bool Runtime::Init(const RuntimeOption& _option) {
        option.model_format == Frontend::PADDLE,
        "Backend::PDINFER only supports model format of Frontend::PADDLE.");
    CreatePaddleBackend();
-    FDINFO << "Runtime initialized with Backend::PDINFER in device " << Str(option.device) << "." << std::endl;
+    FDINFO << "Runtime initialized with Backend::PDINFER in " << Str(option.device) << "." << std::endl;
  } else if (option.backend == Backend::OPENVINO) {
    FDASSERT(option.device == Device::CPU,
             "Backend::OPENVINO only supports Device::CPU");
    CreateOpenVINOBackend();
-    FDINFO << "Runtime initialized with Backend::OPENVINO in device " << Str(option.device) << "." << std::endl;
+    FDINFO << "Runtime initialized with Backend::OPENVINO in " << Str(option.device) << "." << std::endl;
+  } else if (option.backend == Backend::LITE) {
+    FDASSERT(option.device == Device::CPU, "Backend::LITE only supports Device::CPU");
+    CreateLiteBackend();
+    FDINFO << "Runtime initialized with Backend::LITE in " << Str(option.device) << "." << std::endl;
  } else {
    FDERROR << "Runtime only support "
               "Backend::ORT/Backend::TRT/Backend::PDINFER as backend now."
@@ -433,4 +455,21 @@ void Runtime::CreateTrtBackend() {
           "ENABLE_TRT_BACKEND=ON.");
 #endif
 }
+
+void Runtime::CreateLiteBackend() {
+#ifdef ENABLE_LITE_BACKEND
+  auto lite_option = LiteBackendOption();
+  FDASSERT(option.model_format == Frontend::PADDLE,
+           "LiteBackend only support model format of Frontend::PADDLE");
+  backend_ = utils::make_unique<LiteBackend>();
+  auto casted_backend = dynamic_cast<LiteBackend*>(backend_.get());
+  FDASSERT(casted_backend->InitFromPaddle(option.model_file, option.params_file, lite_option),
+           "Load model from nb file failed while initializing LiteBackend.");
+#else
+  FDASSERT(false,
+           "LiteBackend is not available, please compiled with "
+           "ENABLE_LITE_BACKEND=ON.");
+#endif
+}
+
 }  // namespace fastdeploy
--- a/fastdeploy/fastdeploy_runtime.h
+++ b/fastdeploy/fastdeploy_runtime.h
@@ -21,7 +21,7 @@

 namespace fastdeploy {

-enum FASTDEPLOY_DECL Backend { UNKNOWN, ORT, TRT, PDINFER, OPENVINO };
+enum FASTDEPLOY_DECL Backend { UNKNOWN, ORT, TRT, PDINFER, OPENVINO, LITE };
 // AUTOREC will according to the name of model file
 // to decide which Frontend is
 enum FASTDEPLOY_DECL Frontend { AUTOREC, PADDLE, ONNX };
@@ -66,6 +66,9 @@ struct FASTDEPLOY_DECL RuntimeOption {
  // use openvino backend
  void UseOpenVINOBackend();

+  // use paddle lite backend
+  void UseLiteBackend();
+
  // enable mkldnn while use paddle inference in CPU
  void EnablePaddleMKLDNN();
  // disable mkldnn while use paddle inference in CPU
@@ -161,6 +164,8 @@ struct FASTDEPLOY_DECL Runtime {

  void CreateOpenVINOBackend();

+  void CreateLiteBackend();
+
  int NumInputs() { return backend_->NumInputs(); }
  int NumOutputs() { return backend_->NumOutputs(); }
  TensorInfo GetInputInfo(int index);
--- a/fastdeploy/pybind/fastdeploy_runtime.cc
+++ b/fastdeploy/pybind/fastdeploy_runtime.cc
@@ -27,6 +27,7 @@ void BindRuntime(pybind11::module& m) {
      .def("use_ort_backend", &RuntimeOption::UseOrtBackend)
      .def("use_trt_backend", &RuntimeOption::UseTrtBackend)
      .def("use_openvino_backend", &RuntimeOption::UseOpenVINOBackend)
+      .def("use_lite_backend", &RuntimeOption::UseLiteBackend)
      .def("enable_paddle_mkldnn", &RuntimeOption::EnablePaddleMKLDNN)
      .def("disable_paddle_mkldnn", &RuntimeOption::DisablePaddleMKLDNN)
      .def("enable_paddle_log_info", &RuntimeOption::EnablePaddleLogInfo)
@@ -109,7 +110,8 @@ void BindRuntime(pybind11::module& m) {
      .value("UNKOWN", Backend::UNKNOWN)
      .value("ORT", Backend::ORT)
      .value("TRT", Backend::TRT)
-      .value("PDINFER", Backend::PDINFER);
+      .value("PDINFER", Backend::PDINFER)
+      .value("LITE", Backend::LITE);
  pybind11::enum_<Frontend>(m, "Frontend", pybind11::arithmetic(),
                            "Frontend for inference.")
      .value("PADDLE", Frontend::PADDLE)
--- a/fastdeploy/pybind/main.cc.in
+++ b/fastdeploy/pybind/main.cc.in
@@ -137,10 +137,6 @@ PYBIND11_MODULE(@PY_LIBRARY_NAME@, m) {
      "Make programer easier to deploy deeplearning model, save time to save "
      "the world!";

-  pybind11::class_<FDLogger>(m, "FDLogger")
-      .def_readwrite_static("disable_info", &FDLogger::disable_info)
-      .def_readwrite_static("disable_warning", &FDLogger::disable_warning);
-
  BindRuntime(m);
  BindFDModel(m);
 #ifdef ENABLE_VISION
--- a/fastdeploy/utils/perf.h
+++ b/fastdeploy/utils/perf.h
@@ -38,7 +38,7 @@ class FASTDEPLOY_DECL TimeCounter {
    if (!print_out) {
      return;
    }
-    std::cout << prefix << " duration = " << Duration() << "s." << std::endl;
+    FDLogger() << prefix << " duration = " << Duration() << "s." << std::endl;
  }

 private:
--- a/fastdeploy/utils/utils.cc
+++ b/fastdeploy/utils/utils.cc
@@ -16,17 +16,14 @@

 namespace fastdeploy {

-bool FDLogger::disable_info = false;
-bool FDLogger::disable_warning = false;
-
-FDLogger::FDLogger(int level, const std::string& prefix) {
+FDLogger::FDLogger(bool verbose, const std::string& prefix) {
+  verbose_ = verbose;
  line_ = "";
-  level_ = level;
  prefix_ = prefix;
 }

 FDLogger& FDLogger::operator<<(std::ostream& (*os)(std::ostream&)) {
-  if (!verbose()) {
+  if (!verbose_) {
    return *this;
  }
  std::cout << prefix_ << " " << line_ << std::endl;
--- a/fastdeploy/utils/utils.h
+++ b/fastdeploy/utils/utils.h
@@ -37,26 +37,16 @@ namespace fastdeploy {

 class FASTDEPLOY_DECL FDLogger {
 public:
-  static bool disable_info;
-  static bool disable_warning;
-  // 0: INFO
-  // 1: WARNING
-  // 2: ERROR
-  explicit FDLogger(int level = 0, const std::string& prefix = "[FastDeploy]");
-
-  bool verbose() {
-    if (disable_info && level_ == 0) {
-      return false;
-    }
-    if (disable_warning && level_ == 1) {
-      return false;
-    }
-    return true;
+  FDLogger() {
+    line_ = "";
+    prefix_ = "[FastDeploy]";
+    verbose_ = true;
  }
+  explicit FDLogger(bool verbose, const std::string& prefix = "[FastDeploy]");

  template <typename T>
  FDLogger& operator<<(const T& val) {
-    if (!verbose()) {
+    if (!verbose_) {
      return *this;
    }
    std::stringstream ss;
@@ -66,7 +56,7 @@ class FASTDEPLOY_DECL FDLogger {
  }
  FDLogger& operator<<(std::ostream& (*os)(std::ostream&));
  ~FDLogger() {
-    if (!verbose() && line_ != "") {
+    if (!verbose_ && line_ != "") {
      std::cout << line_ << std::endl;
    }
  }
@@ -74,7 +64,7 @@ class FASTDEPLOY_DECL FDLogger {
 private:
  std::string line_;
  std::string prefix_;
-  int level_ = 0;
+  bool verbose_ = true;
 };

 FASTDEPLOY_DECL bool ReadBinaryFromFile(const std::string& file,
@@ -85,15 +75,15 @@ FASTDEPLOY_DECL bool ReadBinaryFromFile(const std::string& file,
 #endif

 #define FDERROR                                                \
-  FDLogger(2, "[ERROR]") << __REL_FILE__ << "(" << __LINE__ \
+  FDLogger(true, "[ERROR]") << __REL_FILE__ << "(" << __LINE__ \
                            << ")::" << __FUNCTION__ << "\t"

 #define FDWARNING                                                \
-  FDLogger(1, "[WARNING]") << __REL_FILE__ << "(" << __LINE__ \
+  FDLogger(true, "[WARNING]") << __REL_FILE__ << "(" << __LINE__ \
                              << ")::" << __FUNCTION__ << "\t"

 #define FDINFO                                                \
-  FDLogger(0, "[INFO]") << __REL_FILE__ << "(" << __LINE__ \
+  FDLogger(true, "[INFO]") << __REL_FILE__ << "(" << __LINE__ \
                           << ")::" << __FUNCTION__ << "\t"

 #define FDASSERT(condition, format, ...)                        \
--- a/fastdeploy/vision/classification/ppcls/model.cc
+++ b/fastdeploy/vision/classification/ppcls/model.cc
@@ -26,7 +26,7 @@ PaddleClasModel::PaddleClasModel(const std::string& model_file,
                                 const RuntimeOption& custom_option,
                                 const Frontend& model_format) {
  config_file_ = config_file;
-  valid_cpu_backends = {Backend::ORT, Backend::OPENVINO, Backend::PDINFER};
+  valid_cpu_backends = {Backend::ORT, Backend::OPENVINO, Backend::PDINFER, Backend::LITE};
  valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
  runtime_option = custom_option;
  runtime_option.model_format = model_format;
--- a/fastdeploy/vision/detection/ppdet/picodet.cc
+++ b/fastdeploy/vision/detection/ppdet/picodet.cc
@@ -24,7 +24,7 @@ PicoDet::PicoDet(const std::string& model_file, const std::string& params_file,
                 const RuntimeOption& custom_option,
                 const Frontend& model_format) {
  config_file_ = config_file;
-  valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
+  valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::LITE};
  valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
  runtime_option = custom_option;
  runtime_option.model_format = model_format;
--- a/fastdeploy/vision/detection/ppdet/ppyoloe.cc
+++ b/fastdeploy/vision/detection/ppdet/ppyoloe.cc
@@ -24,6 +24,7 @@ PPYOLOE::PPYOLOE(const std::string& model_file, const std::string& params_file,
 }

 void PPYOLOE::GetNmsInfo() {
+#ifdef ENABLE_PADDLE_FRONTEND
  if (runtime_option.model_format == Frontend::PADDLE) {
    std::string contents;
    if (!ReadBinaryFromFile(runtime_option.model_file, &contents)) {
@@ -41,6 +42,7 @@ void PPYOLOE::GetNmsInfo() {
      normalized = reader.nms_params.normalized;
    }
  }
+#endif
 }

 bool PPYOLOE::Initialize() {
--- a/fastdeploy/vision/detection/ppdet/ppyoloe.h
+++ b/fastdeploy/vision/detection/ppdet/ppyoloe.h
@@ -56,7 +56,7 @@ class FASTDEPLOY_DECL PPYOLOE : public FastDeployModel {
  float score_threshold = 0.01;
  int64_t nms_top_k = 10000;
  bool normalized = true;
-  bool has_nms_ = false;
+  bool has_nms_ = true;

  // This function will used to check if this model contains multiclass_nms
  // and get parameters from the operator
--- a/fastdeploy/vision/ocr/ppocr/classifier.cc
+++ b/fastdeploy/vision/ocr/ppocr/classifier.cc
@@ -26,12 +26,11 @@ Classifier::Classifier(const std::string& model_file,
                       const RuntimeOption& custom_option,
                       const Frontend& model_format) {
  if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT,
-                          Backend::OPENVINO};  // 指定可用的CPU后端
+    valid_cpu_backends = {Backend::ORT, Backend::OPENVINO};  // 指定可用的CPU后端
    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
  } else {
    valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO};
-    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_gpu_backends = {Backend::PDINFER, Backend::TRT, Backend::ORT};
  }
  runtime_option = custom_option;
  runtime_option.model_format = model_format;
--- a/fastdeploy/vision/ocr/ppocr/recognizer.cc
+++ b/fastdeploy/vision/ocr/ppocr/recognizer.cc
@@ -44,12 +44,12 @@ Recognizer::Recognizer(const std::string& model_file,
                       const RuntimeOption& custom_option,
                       const Frontend& model_format) {
  if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT,
-                          Backend::OPENVINO};  // 指定可用的CPU后端
+    valid_cpu_backends = {Backend::ORT, Backend::OPENVINO};  // 指定可用的CPU后端
    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
  } else {
-    valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO};
-    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    // NOTE:此模型暂不支持paddle-inference-Gpu推理
+    valid_cpu_backends = {Backend::ORT, Backend::PDINFER, Backend::OPENVINO};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
  }

  runtime_option = custom_option;
--- a/python/fastdeploy/init.py
+++ b/python/fastdeploy/init.py
@@ -17,7 +17,7 @@ import os
 import sys

 from .c_lib_wrap import (Frontend, Backend, FDDataType, TensorInfo, Device,
-                         FDLogger, is_built_with_gpu, is_built_with_ort,
+                         is_built_with_gpu, is_built_with_ort,
                         is_built_with_paddle, is_built_with_trt,
                         get_default_cuda_directory)
 from .runtime import Runtime, RuntimeOption
--- a/python/fastdeploy/runtime.py
+++ b/python/fastdeploy/runtime.py
@@ -79,6 +79,9 @@ class RuntimeOption:
    def use_openvino_backend(self):
        return self._option.use_openvino_backend()

+    def use_lite_backend(self):
+        return self._option.use_lite_backend()
+
    def enable_paddle_mkldnn(self):
        return self._option.enable_paddle_mkldnn()