diff --git a/RELEASE.md b/RELEASE.md
index 5490fa143a1d9104ca8f072832b683c227d83147..35f1710b127ea1013f04df72e85c10aad4c205ac 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -1,3 +1,15 @@
+# `v1.5.4`
+
+* 修复Fine-tune中断，checkpoint文件恢复训练失败的问题
+
+# `v1.5.3`
+
+* 优化口罩模型输出结果，提供更加灵活的部署及调用方式
+
+# `v1.5.2`
+
+* 优化pyramidbox_lite_server_mask、pyramidbox_lite_mobile_mask模型的服务化部署性能
+
 # `v1.5.1`
 
 * 修复加载module缺少cache目录的问题
diff --git a/paddlehub/finetune/task/base_task.py b/paddlehub/finetune/task/base_task.py
index 9ce64d6d23e3db5b2c5508f0b253dced741e5da6..f24f20e9afbdb017e2ca2aa88b0724eb7e5cbfde 100644
--- a/paddlehub/finetune/task/base_task.py
+++ b/paddlehub/finetune/task/base_task.py
@@ -807,10 +807,16 @@ class BaseTask(object):
     # NOTE: current saved checkpoint machanism is not completed,
     # it can't restore dataset training status
     def save_checkpoint(self):
+        """
+        save the program of the last step in training
+        """
         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)
+        # to resume traning by loading ckpt, it must be save program (save_persistables)
+        fluid.io.save_persistables(
+            self.exe, dirname=model_saved_dir, main_program=self.main_program)
         save_checkpoint(
             checkpoint_dir=self.config.checkpoint_dir,
             current_epoch=self.current_epoch,