Hello, does anyone know if .mar files are architecture specific?
E.g. you train a model and create a .mar file on an x86 CPU and then you want to serve/deploy it on another architecture, e.g. ppc64le or ARM. Is that possible or can the .mar file only be served on x86 CPUs then?
To make it even more complex:
Assuming that the previous would work (so no binaries are compiled) what about the endianess? Trained on x86 little endian and then serve/deploy on big endian systems - would that work?
Thanks!