llama.cpp

fbe7dfa5 - ggml : add max buffer sizes to opencl and metal backends (#5181)