Fine-Tune Multiple Paragraphs Of Text - Best Practice?

API
1

Hey there,

I did a search but couldn’t find any similar questions…

I’ve been successful in fine-tuning via the API using the following format suggested by the docs (OpenAI API):

{"prompt": "<prompt text>", "completion": "<ideal generated text>"}

Now, if I want to upload multiple paragraphs of text per prompt, do I follow the same guidelines?

For example:

{"prompt": "How do I do XZY proprietary process?", "completion": "Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. Proprietary process text. "}

That completion could be several pages long.

Is that correct?

I’d assume, that each time a variation of the prompt wording of “How do I do XZY proprietary process” is entered my fine-tuned model will give a remixed variation of the completion wording?

Thanks in advance to people who have done this and know the answer :slight_smile: