tensorflow regression: sample_weight option not working, loss is coming out to be nan

Question

I\'m trying to build an autoencoder with weighted MSE as loss function. But the loss is coming out to be nan. I tried removing the sample_weight op