Dynamic memory allocation. Drop Baichuan/InternLM support in favor of llama.cpp.

Dynamic memory allocation. Drop Baichuan/InternLM support in favor of llama.cpp. #248