Using Playground Chats as Training Data for Modeling

From when can chats from the Playground be used as training data for a model?
Or are there already possibilities?

Welcome to the community!

What do you mean with training? Fine-tunes? The playground chats aren’t really there for anything really, other than testing the API.

1 Like

API Platform (after March 1, 2023) isn’t used for training our models.

1 Like

+1 to dignity_for_all if you are asking about OpenAI training their models with your prompts through the Playground.

Otherwise if you are asking if you can take the information you gain from the Playground and train it further. You could use the Playground to test a prompt and then pick what responses you like the best. If there are specific responses you like better than others you could take those and apply it to a fine tuned model.

As a reminder you can only fine tune up to GPT 3.5 as of now.

1 Like

Some users seem to be able to fine-tune GPT-4o as well :slightly_smiling_face:

1 Like

Thank you. That means I would have to copy a complete good chat history from the Playground and insert it into my Fine Tuning model. Can you tell me how to do that?

You would need to prepare a jsonl with your prompts and the expected responses. Once this is complete you can then run this input against a newly created fine tuned model from one of OpenAI’s existing models.

Please read the docs below on how to prep your dataset to feed into a fine tuned model

Thank you very much! That works perfectly!

1 Like