I tried self hosting LLM for commandline instant completion and guidance utility: https://github.com/zerocorebeta/Option-K

But problem is even 7b models are too slow on my pc.

Hosted models are lightening fast. I considered possibility of buying hardware but decided against it.