Losses end up becoming NAN during training. how to debug and fix them?

Besides that, you could also try the already suggested approach to normalize the target for training and “unnormalize” it for predictions from this thread.

2 Likes