Fine Tuning the latest 4o model

GPT builders
1

We are extremely excited with the new capabilities gpt-4o has on image recognition and generation.

We have been working on a image + text → text fine tuned model and have been experimenting with fine-tuning 4o-mini and have observed mixed results.

The latest updates on 4o have been explosive and greatly improved the performance of our task! In fact it is a significant improvement of even our fine-tuned approach. We would very much like to fine-tune this iteration of the model.

The only selection for fine tuning 4o is currently 2024-08-06. Is there a prediction (even a loose one) on when fine tuning will become available with the latest 4o update? I assume 2024-08-06 will not have the same capabilities as the newest update, and so fine tuning will not be worth the cost currently.

2

  • Image generation API is being worked on, but no specific timeline to share just yet.

  • The 4o you’re using in ChatGPT is chatgpt-4o-latest, which does not support fine-tuning. https://platform.openai.com/docs/models/chatgpt-4o-latest

  • The team’s working on a dated 4o model (i.e. will likely support fine-tuning) that will include some of the the improvements you see in ChatGPT, which we hope to release in the coming weeks.

2 Likes
3

Thank you so much Edwin! Very exciting stuff!

2 Likes