llama.cpp

Support requantizing kvcache while model is loaded

#24367

Open