Llama.cpp running qwen 3.8 on a mini PC with 32GB ram and 8GB vram. Not the must speedy but it works. I only access it within my network and I’m also set up with tailscale.