llama.cpp

server : make n_cache_reuse configurable per request

#17858

Merged