What happen when all losses turn to nan????

[chen-erqi]

I use the latest pykan. Multikan is amazing to fit more complicted physical problems. But when I use the multiKan, I find a weired problem. The multiKan seems to turn to nan easily. Given a dataset and a multiKan, it works well. But when I reduce the sample number of this dataset to feed into a same multiKan, all losses turn to nan. Could anyone please to figure out this problem? thanks a lot!!!