HLSL has no concept of 16-bit floats anymore (except for an extension): docs.microsoft.com/en-us/windows/desktop...graphics-hlsl-scalar ("This data type is provided only for language compatibility. Direct3D 10 shader targets map all half data types to float data types.").
GLSL has, but those are mapped to 32-bit on desktop as well, like Marty said.
But most hardware still map halfs to float. It is going to take a while before using anything other than single precision in Reshade will make any sense.
On the latest hardware you could technically use min16float and min16int (and variants), but Reshade would have to support them first and check to see if the user was on Win7 or earlier as min16float support was first introduced with Win8.
I just doubt you are going to see any real improvement in performance ATM.