3 ms·
I've seen many YouTube videos claiming that fine tuning can significantly reduce costs or make a smaller model perform like a larger one. Most of them were not
by dist-epoch 1y ago
I've seen many YouTube videos claiming that fine tuning can significantly reduce costs or make a smaller model perform like a larger one.
Most of them were not from fine-tuning tools or model sellers.
> how it was fine-tuned - all of the training data that was used
It's not that sophisticated. You just need a dataset of prompts and the expected answer. And obviously a way to score the results, so you can guide the fine tuning.
- simonw 1y agoI've seen those same claims, in videos and articles all over the place. Which is why it's so weird that I can't find a convincing live demo to see the results for myself!
- dist-epoch 1y agoMaybe just give it a go on OpenAI? An example on how to train (a presumably small) model to call a get_current_weather function: https://platform.openai.com/docs/guides/supervised-fine-tuning#formatting-your-data https://platform.openai.com/docs/guides/supervised-fine-tuni... It's not such a sexy subject, it's mostly done by companies to reduce costs, which is maybe why there is not much written about it.
- simonw 1y agoThat is exactly the problem: I do not need to save money on my LLM calls, so any experiment I do along those lines won't really benefit me very much. I'm deeply curious, but not quite enough to put the work in if I don't have a practical need for it. I'm constantly surprised at how hard it is to find somebody who can show me a demo! That's why I keep on hassling any company that appears to be selling fine-tuning tooling: if you want people to buy your product, giving them a convincing demo feels like it should be table stakes.