ljjb

> However, the loss didn't seems to be convergent. It descends rapidly at first, but with the increase of the epoch, it looks like a cosine function and the amplitude...