Tensorflow sampling from float32 and float64

Question

I have a memory issue in running the float64 in the cluster so want to try out float32. However with float32 I get nan af