fix load_checkpoint bug (#600)

paddlehub/finetune/task/base_task.py

@@ -833,8 +833,9 @@ class BaseTask(object):
             self.exe,
             main_program=self.main_program)
         # Revise max_train_steps when incremental training
-        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)
+        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
 
     def load_parameters(self, dirname):