llama.cpp

f4626d08 - openai: test features of templates at runtime, to make sure no bits of intel are lost