Is My Project Worth Trying to Fine-Tune?

API
1

I was thinking about running an experimental fine-tuning on ChatGPT 3.5.

I would feed it ~100,000 song lyrics (tagged by genre and year) in hopes of having the model write “high-quality” lyrics.

But after reading what I could find on fine-tuning and chatting with ChatGPT (4) itself, I’m doubting that a fine-tuned 3.5 can do any better than ChatGPT 4. Can anyone tell me if that’s the case?

2

Welcome to the Forum!

Interesting question and project.

Have you run any tests so far with GPT-4 and if so, what if any remaining concerns, are you looking to address through fine-tuning?