Created by: yc556600

如果您没有查询到相似问题，为快速解决您的提问，建立issue时请提供如下细节信息：

• 版本、环境信息：    1）PaddlePaddle版本：1.6.2 同样一个文本，预测出的概率以及对应标签答案不一致 预测代码如下： place = fluid.CUDAPlace(0) if args.use_cuda == True else fluid.CPUPlace() exe = fluid.Executor(place) exe.run(predict_startup)

  if args.init_checkpoint: #init_pretraining_params(exe, args.init_checkpoint, predict_prog, args.use_fp16) init_checkpoint(exe, args.init_checkpoint, predict_prog, args.use_fp16) else: raise ValueError("args 'init_checkpoint' should be set for prediction!")

  Due to the design that ParallelExecutor would drop small batches (mostly the last batch)

  So using ParallelExecutor may left some data unpredicted

  if prediction of each and every example is needed, please use Executor instead

  predict_exe = fluid.ParallelExecutor( use_cuda=args.use_cuda, main_program=predict_prog)

  predict_data_loader.set_batch_generator( processor.data_generator( batch_size=args.batch_size, phase='test', epoch=1, shuffle=False))

  predict_data_loader.start() all_results = [] xxx_results = [] time_begin = time.time() while True: try: results = predict_exe.run(fetch_list=[probs.name]) all_results.extend(results[0]) xxx_results.extend(results) except fluid.core.EOFException: predict_data_loader.reset() break  

保存模型代码如下： if args.save_inference_model_path: _, ckpt_dir = os.path.split(args.init_checkpoint.rstrip('/')) dir_name = ckpt_dir + '_inference_model' model_path = os.path.join(args.save_inference_model_path, dir_name) print("save inference model to %s" % model_path) fluid.io.save_inference_model( model_path, feed_target_names, [probs], exe, params_filename="params", model_filename="model", main_program=predict_prog)

加载模型代码： def load_model(self): print (os.path.join(self.init_checkpoint, "params")) print (os.path.join(self.init_checkpoint, "model")) self.config = AnalysisConfig(os.path.join(self.init_checkpoint, "model"), os.path.join(self.init_checkpoint, "params"))

    self.config.enable_use_gpu(20000, 0)
    print (self.config.gpu_device_id())
    # self.config.disable_gpu()

    # 创建PaddlePredictor
    self.predictor = create_paddle_predictor(self.config)