llama.cpp

9dcac6cf - llama-fit-params: lower ctx size for multi GPU (#18101)