I noticed that I can easily create chatbox where connects the backend to gpt 4. However, I have a finetuned model based on pre-trained gpt model, is there a way to use my own finetuned gpt model for my chatbox? Thanks!
Would love an answer here as well.
@huytool157 do you happen to know how the Retool chat component can be used in combination with one's own fine-tuned model?
Huy will know for sure, but I don't think this is possible with AI actions right now.
IIRC fined-tuned models need to have the exact name of the model input into the API call. And right now model names are a preconfigured list by Retool.
You might be able to use a REST api query and build out a chat-like component yourself, but IMO using built in vectors with gpt-4 yields much better results than fine-tuning gpt-3.5 (seems gpt-4 is now finally fine-tuneable on an experimental build but I haven't tested this). Vectors is also much easier to configure and setup / automate the addition of new context( vector documents) with workflows as needed.
Adding a +1 here that we will need the ability to access fine tuned models (even if we need to copy and paste the exact name in ourselves). They are critical for aligning models in high accuracy, customer facing workflows beyond vectors.
Thanks all for the feedback! (And thanks, Isaac, for the context! )
We'll post back here if this request gets picked up
In the meantime, please consider sharing additional AI feature requests (or Retool AI solutions, if you have any) with the ai tag as we are currently running a monthly challenge that is focused on Retool AI features, like AI Actions & Vectors.