Blog
Release notes and notes from the build.
Porting part of llama.cpp's ggml core from C to C+ proved something narrower and more useful than a rewrite: C+ can enter a live C/C++ build one object file at a time, with the surrounding code none the wiser.