llama.cpp

e0e806f5 - kv-cache : fix unified::seq_rm to work with seq_id < 0 (#13985)