fix load_checkpoint bug (#600)

paddlehub/finetune/task/base_task.py

@@ -833,6 +833,7 @@ class BaseTask(object):
             self.exe,
             main_program=self.main_program)
         # Revise max_train_steps when incremental training
+        if is_load_successful:
             self.max_train_steps = self.env.current_step + self.max_train_steps / self.config.num_epoch * (
                 self.config.num_epoch - self.env.current_epoch + 1)
         return is_load_successful