diff --git a/paddle/fluid/train/custom_trainer/feed/accessor/dense_input_accessor.cc b/paddle/fluid/train/custom_trainer/feed/accessor/dense_input_accessor.cc
index b9a40b0a032377b660a859398e51a12a7c808749..1957b695c17a73c876fc36c0586a8b7559b75e56 100644
--- a/paddle/fluid/train/custom_trainer/feed/accessor/dense_input_accessor.cc
+++ b/paddle/fluid/train/custom_trainer/feed/accessor/dense_input_accessor.cc
@@ -70,6 +70,52 @@ int32_t DenseInputAccessor::pull_dense(size_t table_id) {
 
 int32_t DenseInputAccessor::forward(SampleInstance* samples, size_t num,
     paddle::framework::Scope* scope) {
+    collect_persistables(scope);
+
+    if (_need_async_pull) {
+        ++_pull_request_num;
+    }
+    return 0;
+}
+
+int32_t DenseInputAccessor::backward(SampleInstance* samples, size_t num,
+        paddle::framework::Scope* scope) {
+    if (!_need_gradient) {
+        return 0;
+    }
+    size_t data_buffer_idx = 0;
+    std::vector<paddle::ps::Region> regions;
+    for (auto& variable : _x_variables) {
+        auto* tensor = scope->Var(variable.gradient_name)->
+            GetMutable<paddle::framework::LoDTensor>();
+        auto* grad_data = tensor->mutable_data<float>(_trainer_context->cpu_place);
+        regions.emplace_back(grad_data, variable.dim);
+    }
+    auto* ps_client = _trainer_context->pslib->ps_client();
+    auto push_status = ps_client->push_dense(regions.data(), regions.size(), _table_id);
+    //push_status.get();
+    if (!FLAGS_feed_trainer_debug_dense_name.empty()) {
+        std::stringstream ssm;
+        for (auto& variable : _x_variables) {
+            ssm.str("");
+            if (variable.name != FLAGS_feed_trainer_debug_dense_name) {
+                continue;
+            }
+            auto& tensor = scope->Var(variable.gradient_name)->
+                Get<paddle::framework::LoDTensor>();
+            const auto* var_data = tensor.data<float>();
+            for (size_t data_idx = 0; data_idx < variable.dim; ++data_idx) {
+                if (data_idx > 0)
+                    ssm << ",";
+                ssm << var_data[data_idx];
+            }
+            VLOG(2) << "[DEBUG]push_dense: " << ssm.str();
+        }
+    }
+    return 0;
+}
+
+int32_t DenseInputAccessor::collect_persistables(paddle::framework::Scope* scope) {
     // 首次同步pull，之后异步pull
     if (_data_buffer == nullptr) {
         _pull_mutex.lock();
@@ -95,7 +141,9 @@ int32_t DenseInputAccessor::forward(SampleInstance* samples, size_t num,
         paddle::framework::DDim ddim(shape_ptr, variable.shape.size());
         auto* tensor = ScopeHelper::resize_lod_tensor(scope, variable.name, ddim);  
         auto* grad_tensor = ScopeHelper::resize_lod_tensor(scope, variable.gradient_name, ddim);
-        VLOG(5) << "fill scope variable:" << variable.name << ", " << variable.gradient_name;
+        VLOG(5) << "fill scope variable:" << variable.name << ", " << variable.gradient_name
+            << ", data_buffer: " << _data_buffer + data_buffer_idx
+            << ", dim: " << variable.dim * sizeof(float);
         auto* var_data = tensor->mutable_data<float>(_trainer_context->cpu_place);
         memcpy(var_data, _data_buffer + data_buffer_idx, variable.dim * sizeof(float));
         data_buffer_idx += variable.dim;
@@ -120,45 +168,12 @@ int32_t DenseInputAccessor::forward(SampleInstance* samples, size_t num,
             VLOG(2) << "[DEBUG]pull_dense: " << ssm.str();
         }
     }
-    if (_need_async_pull) {
-        ++_pull_request_num;
-    }
     return 0;
 }
 
-int32_t DenseInputAccessor::backward(SampleInstance* samples, size_t num,
-        paddle::framework::Scope* scope) {
-    if (!_need_gradient) {
-        return 0;
-    } 
-    size_t data_buffer_idx = 0;
-    std::vector<paddle::ps::Region> regions;
+int32_t DenseInputAccessor::collect_persistables_name(std::vector<std::string>& persistables) {
     for (auto& variable : _x_variables) {
-        auto* tensor = scope->Var(variable.gradient_name)->
-            GetMutable<paddle::framework::LoDTensor>(); 
-        auto* grad_data = tensor->mutable_data<float>(_trainer_context->cpu_place);
-        regions.emplace_back(grad_data, variable.dim);
-    }
-    auto* ps_client = _trainer_context->pslib->ps_client();
-    auto push_status = ps_client->push_dense(regions.data(), regions.size(), _table_id);
-    //push_status.get();
-    if (!FLAGS_feed_trainer_debug_dense_name.empty()) {
-        std::stringstream ssm;
-        for (auto& variable : _x_variables) {
-            ssm.str("");
-            if (variable.name != FLAGS_feed_trainer_debug_dense_name) {
-                continue;
-            }
-            auto& tensor = scope->Var(variable.gradient_name)->
-                Get<paddle::framework::LoDTensor>(); 
-            const auto* var_data = tensor.data<float>();
-            for (size_t data_idx = 0; data_idx < variable.dim; ++data_idx) {
-                if (data_idx > 0)
-                    ssm << ",";
-                ssm << var_data[data_idx];
-            }
-            VLOG(2) << "[DEBUG]push_dense: " << ssm.str();
-        }
+        persistables.push_back(variable.name);
     }
     return 0;
 }
diff --git a/paddle/fluid/train/custom_trainer/feed/accessor/input_data_accessor.h b/paddle/fluid/train/custom_trainer/feed/accessor/input_data_accessor.h
index 45c3d051a1d0d0584f193aad35f722dc96cf9371..52d07b6b0598f2a1f4f87f44bba2d88b874b7582 100644
--- a/paddle/fluid/train/custom_trainer/feed/accessor/input_data_accessor.h
+++ b/paddle/fluid/train/custom_trainer/feed/accessor/input_data_accessor.h
@@ -38,6 +38,12 @@ public:
     // 后向，一般用于更新梯度，在训练网络执行后调用
     virtual int32_t backward(SampleInstance* samples, size_t num,
         ::paddle::framework::Scope* scope) = 0;
+
+    // 收集持久化变量的名称, 并将值拷贝到Scope
+    virtual int32_t collect_persistables_name(std::vector<std::string>& persistables) {return 0;}
+
+    // 填充持久化变量的值，用于保存
+    virtual int32_t collect_persistables(paddle::framework::Scope* scope) {return 0;}
 protected:
     size_t _table_id = 0;
     bool _need_gradient = false;
@@ -144,6 +150,11 @@ public:
 
     virtual int32_t backward(SampleInstance* samples, size_t num,
         paddle::framework::Scope* scope);
+
+
+    virtual int32_t collect_persistables_name(std::vector<std::string>& persistables);
+
+    virtual int32_t collect_persistables(paddle::framework::Scope* scope);
 protected:
     virtual int32_t pull_dense(size_t table_id);
 
diff --git a/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.cc b/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.cc
index af3d4f7cbf18bde80dadf10fcd11ba5135110b97..e7ef07b0fd4ad5acc5f7990a49dc4826a797d6f8 100644
--- a/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.cc
+++ b/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.cc
@@ -52,6 +52,7 @@ int MultiThreadExecutor::initialize(YAML::Node exe_config,
     CHECK(_trainer_context->file_system->exists(model_config_path)) 
         << "miss model config file:" << model_config_path;
     _model_config = YAML::LoadFile(model_config_path);
+    _persistables.clear();
     for (const auto& accessor_config : _model_config["input_accessor"]) {
         auto accessor_class = accessor_config["class"].as<std::string>();
         auto* accessor_ptr = CREATE_INSTANCE(DataInputAccessor, accessor_class);
@@ -66,7 +67,10 @@ int MultiThreadExecutor::initialize(YAML::Node exe_config,
                 _table_to_accessors[table_id] = {accessor_ptr};
             }
         }
-    } 
+        CHECK(accessor_ptr->collect_persistables_name(_persistables) == 0)
+            << "collect_persistables Failed, class:" << accessor_class;
+    }
+    std::sort(_persistables.begin(), _persistables.end()); // 持久化变量名一定要排序
 
     // Monitor组件
     for (const auto& monitor_config : _model_config["monitor"]) {
@@ -79,6 +83,27 @@ int MultiThreadExecutor::initialize(YAML::Node exe_config,
     return ret;
 }
 
+int32_t MultiThreadExecutor::save_persistables(const std::string& filename) {
+    // auto fs = _trainer_context->file_system;
+    // fs->mkdir(fs->path_split(filename).first);
+    auto scope_obj = _scope_obj_pool->get();
+    for (size_t i = 0; i < _input_accessors.size(); ++i) {
+        _input_accessors[i]->collect_persistables(scope_obj.get());
+    }
+    framework::ProgramDesc prog;
+    auto* block = prog.MutableBlock(0);
+    auto* op = block->AppendOp();
+    op->SetType("save_combine");
+    op->SetInput("X", _persistables);
+    op->SetAttr("file_path", filename);
+    op->CheckAttrs();
+
+    platform::CPUPlace place;
+    framework::Executor exe(place);
+    exe.Run(prog, scope_obj.get(), 0, true, true);
+    return 0;
+}
+
 paddle::framework::Channel<DataItem> MultiThreadExecutor::run(
     paddle::framework::Channel<DataItem> input, const DataParser* parser) {
 
diff --git a/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.h b/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.h
index e1fbc42d699ed852d7bee9f9c6d64221e55b824b..becfabea5c32db1ea766fbb09ad9b83f34f0d7cd 100644
--- a/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.h
+++ b/paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.h
@@ -46,6 +46,8 @@ public:
     //执行训练
     virtual paddle::framework::Channel<DataItem> run(
         paddle::framework::Channel<DataItem> input, const DataParser* parser);
+
+    virtual int32_t save_persistables(const std::string& filename);
     
     virtual bool is_dump_all_model() {
         return _need_dump_all_model;
@@ -79,6 +81,7 @@ protected:
     std::vector<std::shared_ptr<DataInputAccessor>> _input_accessors;
     std::map<uint32_t, std::vector<DataInputAccessor*>> _table_to_accessors;
     std::shared_ptr<paddle::ps::ObjectPool<::paddle::framework::Scope>> _scope_obj_pool;
+    std::vector<std::string> _persistables;
 };
 
 }  // namespace feed
diff --git a/paddle/fluid/train/custom_trainer/feed/io/file_system.h b/paddle/fluid/train/custom_trainer/feed/io/file_system.h
index 0ef5a37b0c0a3e04d2f20d2b036ff2541b4f0f48..4531cbf1125d0a3af54a503fab82e1b9db970f2d 100644
--- a/paddle/fluid/train/custom_trainer/feed/io/file_system.h
+++ b/paddle/fluid/train/custom_trainer/feed/io/file_system.h
@@ -25,6 +25,10 @@ public:
     virtual bool exists(const std::string& path) = 0;
     virtual void mkdir(const std::string& path) = 0;
     virtual std::string path_join(const std::string& dir, const std::string& path);
+    template<class... STRS>
+    std::string path_join(const std::string& dir, const std::string& path, const STRS&... paths) {
+        return path_join(path_join(dir, path), paths...);
+    }
     virtual std::pair<std::string, std::string> path_split(const std::string& path);
 protected:
 };
diff --git a/paddle/fluid/train/custom_trainer/feed/process/learner_process.cc b/paddle/fluid/train/custom_trainer/feed/process/learner_process.cc
index 367f340b20c434f5f103292b12f90f161e3cb555..40cd0651512ba0a007416176ef336474e6feb129 100755
--- a/paddle/fluid/train/custom_trainer/feed/process/learner_process.cc
+++ b/paddle/fluid/train/custom_trainer/feed/process/learner_process.cc
@@ -27,6 +27,7 @@ int LearnerProcess::initialize(std::shared_ptr<TrainerContext> context_ptr) {
 }
 
 int LearnerProcess::wait_save_model(uint64_t epoch_id, ModelSaveWay way) {
+    auto fs = _context_ptr->file_system;
     auto* ps_client = _context_ptr->pslib->ps_client();
     auto* environment = _context_ptr->environment.get();
     auto* epoch_accessor = _context_ptr->epoch_accessor.get();
@@ -39,18 +40,21 @@ int LearnerProcess::wait_save_model(uint64_t epoch_id, ModelSaveWay way) {
     paddle::platform::Timer timer;
     timer.Start();
     std::set<uint32_t> table_set;
+    auto model_dir = epoch_accessor->model_save_path(epoch_id, way);
     for (auto& executor : _executors) {
         const auto& table_accessors = executor->table_accessors();
         for (auto& itr : table_accessors) {
             table_set.insert(itr.first);
         }
+        auto save_path = fs->path_join(model_dir, executor->train_exe_name() + "_param");
+        VLOG(2) << "Start save model, save_path:" << save_path;
+        executor->save_persistables(save_path);
     }
     int ret_size = 0;
     auto table_num = table_set.size();
     std::future<int> rets[table_num];
     for (auto table_id : table_set) {
         VLOG(2) << "Start save model, table_id:" << table_id;
-        auto model_dir = epoch_accessor->model_save_path(epoch_id, way);
         rets[ret_size++] = ps_client->save(table_id, model_dir, std::to_string((int)way));
     }
     int all_ret = 0;