Add inference model (#327)

* add-inference-model

Author: Steffy-zxf

paddlehub/finetune/checkpoint.py

@@ -74,10 +74,6 @@ def save_checkpoint(checkpoint_dir,
     ckpt = checkpoint_pb2.CheckPoint()
 
     model_saved_dir = os.path.join(checkpoint_dir, "step_%d" % global_step)
-    logger.info("Saving model checkpoint to {}".format(model_saved_dir))
-    fluid.io.save_persistables(
-        exe, dirname=model_saved_dir, main_program=main_program)
-
     ckpt.current_epoch = current_epoch
     ckpt.global_step = global_step
     ckpt.latest_model_dir = model_saved_dir

paddlehub/finetune/task/base_task.py

@@ -662,11 +662,7 @@ def _default_eval_end_event(self, run_states):
                                            "best_model")
             logger.eval("best model saved to %s [best %s=%.5f]" %
                         (model_saved_dir, main_metric, main_value))
-
-            save_result = fluid.io.save_persistables(
-                executor=self.exe,
-                dirname=model_saved_dir,
-                main_program=self.main_program)
+            self.save_inference_model(dirname=model_saved_dir)
 
     def _default_log_interval_event(self, run_states):
         scores, avg_loss, run_speed = self._calculate_metrics(run_states)
@@ -717,6 +713,10 @@ def _calculate_metrics(self, run_states):
     # NOTE: current saved checkpoint machanism is not completed,
     # it can't restore dataset training status
     def save_checkpoint(self):
+        model_saved_dir = os.path.join(self.config.checkpoint_dir,
+                                       "step_%d" % self.current_step)
+        logger.info("Saving model checkpoint to {}".format(model_saved_dir))
+        self.save_inference_model(dirname=model_saved_dir)
         save_checkpoint(
             checkpoint_dir=self.config.checkpoint_dir,
             current_epoch=self.current_epoch,

paddlehub/finetune/task/classifier_task.py

@@ -317,7 +317,7 @@ def _calculate_metrics(self, run_states):
     def fetch_list(self):
         if self.is_train_phase or self.is_test_phase:
             return [metric.name for metric in self.metrics] + [self.loss.name]
-        return self.outputs
+        return [output.name for output in self.outputs]
 
     def _postprocessing(self, run_states):
         results = []