I got to the point of figuring out that none of the precompiled binaries will work on my Xeon (it's a Haswell-era processor with no AVX-512) and I'd have to rebuild everything from scratch. But the repo doesn't include clear instructions or scripts to rebuild its special llama.cpp variant, only a few patch files (which don't seem to be enough on their own?). So I'm not going to pursue it further.