dawuchen

Results 2 issues of dawuchen

有2个地方想请问下: 1. 这里[capsule_weight](https://github.com/THUDM/ComiRec/blob/master/src/model.py#L147)是用的普通的tensor而不是variable,请问是对于每个batch的每个样本,一次前向计算时都会做随机初始化然后做3次迭代吗?那inference的时候也是需要做迭代过程? 2. 我这边的几个训练任务,得到的capsule_weight以及之后的dnn的各层输出的norm值都非常大,请问这是正常的吗? ![image](https://user-images.githubusercontent.com/13624034/115815109-f0c4ab80-a428-11eb-874e-26f9967d203d.png)

I know --executor-cores is the param of spark-submit to specify the cpu cores per executor .But in [CLI tutorial](https://github.com/yahoo/CaffeOnSpark/wiki/CLI) also says -devices is the number of devices (CPU or GPU)...