I am using Keras in R with "Keras" package and having a lot of fun.
I heard that in python, using float16 instead of default float32 could
save precious GPU memory and use has the effect as if doubling the vram (for supported
RTX cards, at least).
However, I could not find anyway to use float16 in Keras for R.
Is this even possible? Or, is python my option for the time being.