Statically linking against musl

Is it possible to statically link MXNet training (not just amalgamated inference!) using a more portable stdlib like musl.

I tried compiling using -nostdlib -static to get libmxnet.a, but then when I try compiling+linking a test program, I get infinite errors which suggest that linking has failed.
Am I trying to do something particularly unreasonable? In any case, it’d be useful, I think, to have a ML library that can actually do some basic training on smart/mobile devices!