API fine-tune model help (I pay for help)

[wraith galleon]

For the last 2 weeks I've been trying to make a fine tune model on gpt's api that can match my writing style based on examples I give it, I've gotten close to the results I'm looking for, but it's too inconsistent. I'm an absolute noob when it comes to fine tuning and prompting and honestly have no clue what I'm doing, I have been watching some tutorials and even reaching out on freelance platforms without success since there is so less information online and not a lot of people offer their expertise, coming here is honestly my last hope and would pay a nice amount for someone's help in creating this fine tune model, if interested, please send me a dm or react to this message and I can elaborate further, thank you! 🙏

[rigid hill]

Is there a reason you want fine tune as opposed to in-context learning (prompting as you call it) ? What does "mostly working but inconsistent" mean? How many example are you feeding for fine tuning? How long do you fineutne? What model are you finetuning?

[wraith galleon]

without fine tuning it doesn't take my writing style in consideration and gives different results than my examples, with mostly working I mean that it sometimes gives a result that I can use, or can be happy with, but that success rate is around 20% and less of the generations, for fine tuning I tried, gpt4 mini and 4o, I have given it 10, 50 and up to 160 inputs, different types of system and user prompts

[rigid hill]

Writing style is an odd thing. Would another human be able to mimic your "style" based on those example? Like a "Pirate style" is very distict, right? Is your style distinct enough from general writing that those number of example would make you expect different "style" ?

[wraith galleon]

currently I just want generations that make sense, even if it isn't my writing style, but it's not even capable of doing that atm

[rigid hill]

If you fine tune something and it's performance is worse, then the fine tuning went wrong. Overfitting, regression, catestrophic forgetting are some of the terms i'd look up. Finetuning is hard and not a lot of people know how to do it properly, including me 🙂

[wraith galleon]

that's exactly why I'm reaching out to professionals who want to help me out and get payed for it since I don't have the expertise to do this

[hidden bluff]

Fine tuning would a lot of data, how do you think to have document based chatbot that users would upload documents and query based on that?

[slate socket]

In all honesty, I've been trying to do this exact thing for weeks as well.

oh wait this is a 6 month old thread