master
libs : rename libcommon -> libllama-common (#21936)
llama: automatically set parameters not set by the user in such a way that maximizes GPU utilization (#16653)
fit-params : refactor + add option to output estimated memory per device (#22171)