Fine Tune GPT Model - Any Sucess?

Alexander301

Alexander301

Junior Member
Joined
Dec 25, 2024
Messages
130
Reaction score
58
Hello folks,
Has anyone tried to use finetuned AI/GPT models to have a more unique author tone? If so, what kind of training data did you use? I tried myself a few times based on training data from real authors but the models seems to become very overfitted and its response do not stay on track.

What I am hoping to get is same level of prompt adherence as the 4o/mini models but with custom author tone built into the model.
 
Advertise on BHW
Alexander301 said:
Has anyone tried to use finetuned AI/GPT models to have a more unique author tone? If so, what kind of training data did you use? I tried myself a few times based on training data from real authors but the models seems to become very overfitted and its response do not stay on track.

What I am hoping to get is same level of prompt adherence as the 4o/mini models but with custom author tone built into the model.
Click to expand...
Instead of full model fine-tuning, try careful prompt crafting. You can steer the model toward the tone you want by providing detailed instructions like:
  • “Write in the style of [Author] while discussing [Topic].”
  • "Use humor and vivid metaphors in [Author]'s tone."
 
Alexander301 said:
Hello folks,
Has anyone tried to use finetuned AI/GPT models to have a more unique author tone? If so, what kind of training data did you use? I tried myself a few times based on training data from real authors but the models seems to become very overfitted and its response do not stay on track.

What I am hoping to get is same level of prompt adherence as the 4o/mini models but with custom author tone built into the model.
Click to expand...
Hi
I parsed auto subtitles of youtube bloggers from Nigeria to get their slang then restored to normal text via mini from GPT.
 
David Boost said:
  • “Write in the style of [Author] while discussing [Topic].”
  • "Use humor and vivid metaphors in [Author]'s tone."
Click to expand...
Has anyone had any success with fine-tuned models? I am looking for something beyond simple "Write in [authors] tone". I tried some other suggestions which mentioned using literature and empty user responses but the model seemed to be overfitted and too constrained to the trained text.
 
Advertise on BHW
You must log in or register to reply here.
Sign up now!

Main Menu

Home Main Forum List Black Hat SEO White Hat SEO BHW Newbie Guide Blogging Black Hat Tools Social Media Downloads

Marketplace

Account Selling Content / Copywriting EBooks / Methods Hosting Misc Proxies For Sale SEO - Link building SEO - Packages Social Media Social Media - Panels Web Design More

Making Money

Affiliate Programs Hire a Freelancer IM Journeys Making Money Pay Per Click Site Flipping Want to Buy

BlackHat World

Dispute Resolution Forum Suggestions & Feedback Introductions Lounge
Back
Top