VLLM or llama.cpp: Choosing the right LLM inference engine for your use case