Llamafile did great work on improving cpu inferencing and generosly upstreamed it into llama.cpp

Wonderful project, please check it out.