Max Length for Fine-tuning Completions?

Hi,

I’m trying to finetune with some of my own fiction writing. I’ve been using full stories (some ~6k-10k words), and I’m wondering if those completions are too long for the finetune process? How long can a completion be? Is there a tool available to break up long text into max length more easily?

Sorry if this is super basic! Thanks for your help!

1 Like

You should those and split them. Completions typically have a max length of around 2048 tokens, and you will not be able to fit those full stories into that length. I suggest splitting up the stories and using part of the story as a prompt, and the next section as a completion. This will help keep the stories within the boundary of the 2048 token limit.
For the prompt to generate the story, add the user prompt and beginning paragraph or so of the story, so the model can build upon itself by appending the generated beginning into the same API call.