llama.cpp

5a4ff43e - llama : disable pipeline parallelism if compute buffer allocation fails (#16748)