The current crop of services and pdf chatbots that you see going around using "train" on your own data aren't doing anything of the sort. They are essentially f

The current crop of services and pdf chatbots that you see going around using "train" on your own data aren't doing anything of the sort. They are essentially finding relevant sections of your docs and stuffing that into the prompt context window. Think of llama as the trained base model (not particularly useful for any one thing by itself) and alpaca is a fine tuned variant of that to give it chatgpt like "powers".