llama.cpp

a885dcff - batched-bench : fix llama_synchronize usage during prompt processing (#15835)