• Subscribe
    1. Discussions
    2. Anyone using fine tuned openai models?

    Anyone using fine tuned openai models?

    Dhananjoy
    1 reply
    Curious if anyone is using fine-tuned models in production and what does the performance and cost look like.

    Replies

    Richard Gao
    The main issue with fine tuning is that if you're fine tuning the larger models (which are the most useful ones), it takes an immense amount of text for your fine tuned model to actually be useful. Not only is selecting what info to feed the AI quite prohibitive for most, but the cost rises linearly with the amount of data being fed in for training. From what I've seen about your product, it'll make fine tuning itself much easier, though I haven't done it myself, and don't know about the cost, but it could be an issue if you want a larger, useful model, especially if you're adding more cost on top of that with your product. I've heard a lot of people say embeddings seem to be more useful than fine tuning, but only testing will tell.
    Share