crossmedia loss只下降了一次.

crossmedia这个例子数据格式有很多错误, fix了以后, data.txt中只留一条数据, 发现loss只下降了一次, 然后就无法下降了, 打印如下:
loss: [array(0.6933602, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]
loss: [array(0.6931472, dtype=float32)]

1. 能不能提供一个目前在官方镜像可以跑的crossmedia

Start with the crossmedia example and the data.txt input described in the issue, then reproduce the reported loss output in the official image. Compare behavior after correcting the data format and determine whether a runnable example can be provided with loss continuing to decrease.

Written by the indexing model from the issue text.