[sometimelurker]

llama.cpp + pi.dev for me, using small Gemma and qwen llms