Created by: gromzhu

将训练好的transformer模型save_inference_model之后，加载保存的inference_model预测，报错shape不同，我用paddle保存模型的时候tensor shape设置的是(none,1)，然后预测的时候我传shape（-1，1）的变量进去报错shape不相等。

报错信息 File ".jumbo/lib/python2.7/site-packages/paddle/fluid/executor.py", line 225, in check_feed_shape_type (var.name, len(var.shape), var.shape, diff_shape)) ValueError: The fed Variable u'init_score' should have dimensions = 1, shape = (0L,), but received fed shape [-1L, 1L] on each device