/* *Author: xiexionghang *Train样本 */ #pragma once #include "paddle/fluid/train/custom_trainer/feed/process/process.h" #include "paddle/fluid/train/custom_trainer/feed/executor/multi_thread_executor.h" namespace paddle { namespace custom_trainer { namespace feed { class LearnerProcess : public Process { public: LearnerProcess() {} virtual ~LearnerProcess() {} virtual int run(); virtual int initialize(std::shared_ptr context_ptr); protected: // 加载所有模型 virtual int load_model(uint64_t epoch_id); // 同步保存所有模型, is_force_dump:不判断dump条件,强制dump出模型 virtual int wait_save_model(uint64_t epoch_id, ModelSaveWay way, bool is_force_dump = false); private: bool _startup_dump_inference_base; //启动立即dump base std::vector> _executors; }; } // namespace feed } // namespace custom_trainer } // namespace paddle