Fine tuning Davinci01 or prompting Davinci03

API
1

Hi all

I am working on a simple copy writing app and I am wondering what’s the most efficient option:

Prompting davinci03
Or fine tuning davinci01

I have a dataset of 20-25 entries (prompt-completion)

Thanks
:pray:

1 Like
2

You might not have enough samples in your dataset for fine-tuning, so I would try a two-shot (give two examples in your prompt)…

Hope that helps!

Welcome to the community…

2 Likes
3

@PaulBellow is correct.

Prompting is the way, unless you have a large dataset and request volume, where the economics would support fine-tuning

2 Likes
4

Thank you. I should have a dataset of 200 - 300 entries. It’s a series of Q&A about a specific topic that GPT3.5 doesn’t “know.”

What parameters: Temperature, Top P, Frequency, and Presence, should I use to force the system to be more deterministic and compliant with the fined-tuned model?