llama.cpp

server : avoid breaking KV cache when prompt >= n_ctx (#6855)

#8359

Closed