I was about to make a topic asking about this, but I figured I’d search around first and found this post from @Logan_Kilpatrick today,
Buried in there is this awesome tidbit,
To support that, we will also be rolling out tuning support for Gemini 1.5 Flash on June 17th.
Which is amazing and exactly what I’ve been waiting for!
@Logan_Kilpatrick a couple of questions,
- Can you confirm fine-tuning for Gemini Flash will be multi-modal?
- If the fine-tuning is multi-modal, will there be a cookbook or best-practices guide for fine-tuning with mixed inputs?
- Are there plans to allow using previously fine-tuned models as base models for further fine-tuning?
- We talked about this briefly in the other place, but have there been any discussions about the possibility of publishing or otherwise sharing fine-tuned models?