Alot of API credits expiring this month - any ideas on finetuning datasets I should create?

Per the title, I have quite a number of openai credits expiring this month.

Hope to make create some finetuning datasets to upload at Huggingface.

Any ideas would be appreciated!

That’s an interesting offer.
Do you have the option to fine-tune GPT-4 or 3.5 only?

Only GPT 3.5 :frowning:

Any ideas on Q&A datasets?

No need to worry. I updated the tags to properly match your request.
Hopefully this will help you find some nice ideas!

1 Like

Edit- I thought it was disallowed to use outputs from OpenAI as training data for other models, but I guess thats only if you’re going to “compete with OpenAI’s models”

1 Like

How about:

Run embeddings on a chunked database of everything you have that could inform an AI application better.

1 Like

Do one on this Discourse forum.

There are many questions that are being asked here and while there is no AI that can currently answer them all, there are a few like the ask-ai channel on the OpenAI Discord forum that can answer about one every now and then. The bot is by and uses RAG.

What would be interesting to see is if a fine tuning can do a better job.

A better way to access the post on this forum for use with RAG or fine tuning would be to pull them as JSON instead of HTML. Just add .json to the URL.

The first post of this topic is

to get the JSON reply use

I will warn you that this forum is an echo chamber.


1 Like

Really great idea! Thanks for sharing.