Finetune/train on csvs with personal data as personal assistant


I’d like to make one or multiple presets within open ai gpt-3 davinci-003 based upon some csv files.

Note that I never used fine tuning before, but played around with the playground quite a bit.

The csv files would be something like:

Name, First Name, Address, City, Birthday, Hobbies BucketList, Likes, Dislikes, Birthday, LinkedIn Page, Education High School, father of, daughter, of, food likes, food dislikes etc…

I d then use the data all my friends, neighboors, family, colleagues as input for this csv and train gpt-3 with it.

That way I could ask the trained set all kinds of insights like “how many uncles x has or whose Birthdays this month and what should I buy, y comes to dinner what should I cook”

Obviously I’d update the csv data regularely.

Not sure howto address that. Please advise.

1 Like

I’m not sure it would work as the model “hallucinates” (makes up) data at times - even with a good fine-tune.

Have you read the docs on fine-tuning? What’s your experience level with programming?

PS - welcome to the forums!

Try using a JSON format…