llama.cpp

3d9a5518 - Fixed mmap prefetch for GPU offloading (#2529)