Is there a way to pass NaNs as label ot keras.Sequential's fit()?

Question

I have a sequential dataset. But some labels (at the end of my label-set), are too far in the future and therefore cant be calculated.

Since I dont want to cut the en