Fine tuning text-davinci-3


Is there any info available about the possibility to fine tune other models than the base ones (davinci in my case) ?
The API calls are great with text-davinci-3 but the token limitation makes it very difficult to do a correct few shot examples method and the fine tuning is not available…

In my case, and I believe it is the same for many people, I am stuck because fine tuned text-davinci-3 is very likely to work perfectly my case but “prompt designed text-davinci-3” and “fine tuned base davinci” are both unable to perform correctly.

Thanks all

Hey! Which model would you want to fine tune?

1 Like

This model: text-davinci-003

I just pushed an update to the docs that should be live soon, the TLDR is that you can only fine tune the base models like davinci not text-davinci-003. The main difference is that text-davinci-003 has been tuned to work well based on instructions so it’s less useful after being fine-tuned which is why we only expose the base models for fine-tuning.


And to go back to your original question, there is no plan to allow other models to be fine-tuned beyond the base models at this time.