llama.cpp

dfc29b64 - context : autoscale n_ctx_train when yarn scaling specified (#28030)