You can rent out your GPU to vast.ai

But the issue with large language models is that you need ms-latency access to TBs of data, or else you won't be able to saturate your GPUs with useful work.