Running local LLMs on a laptop for research — tips?

Got access to a lab that wants me to experiment with open-weight models on consumer hardware.

What I'm working with:

• RTX 4060 8GB (my personal machine)
• Lab server: 2x A5000 (48GB each)
• Budget for API calls is tight

Looking for:

1. Best quantization for Qwen/LLaMA 3 on 8GB
2. vLLM vs llama.cpp for throughput
3. Any good RAG frameworks that don't need a PhD to set up

Appreciate any pointers!