llama.cpp

15267192 - llama : refactor tensor offloading as callback