Fine-tune mode

[nilsherzig]

Accepts a list of LLMs / settings and basically does table testing with them. Uses user input to vote on the results. Used to find good configurations by brute force.

Maybe crowd source this, since it will take hours / days to run.