Tried a tiny AI model for recipe scaling, got better results than the big one
I run a small kitchen in Portland and kept messing up batch conversions for our specials. Last week I fed the same problem to a 7B parameter local model and the big cloud one I pay $20 a month for. The little guy caught a decimal error the big one missed, and it did it in 3 seconds offline. Has anyone else found smaller models winning on narrow tasks like this, or was it just luck?