Provide feedback when selected model is no longer available

Thibaultmol

For ex, GPT-4 was removed recently. So i'm wondering: what's the plan for people who had a url with gpt-4 in it via a url parameter.

Currently it just falls back to whatever you had as your previous model you used, which is... ok I guess.

But I'm wondering if there should be a fallback for each model. for ex: if someone had llama 3 in their url paramter.... when llama 3 got removed, it should fallback to llama 3.1

And then give the user a toast/fading-notification that says "The model specified in your URL is no longer available. We have automatically updated your request to use the closest available model." or something like that

Luis

I agree with giving users feedback about such change, we'll add that. Regarding the fallback model, I believe it's more intuitive to select the default, currently it's the last used model, instead of creating a mapping table for this edge case. Does that make sense to you?

I will rename the thread to reflect the change we will implement.

Thibaultmol

Luis I mean that depends. Some people might put in certain data with certain custom assistants that they don't want going to Google (gemini). they might use gemini for regular Assistant use, but for certain custom assistants they input data that they only want to go to openai for example or something like that.

And some people might have found that one llm company works better over an other when it comes to certain custom assitant tasks

So I don't think switching to default assistant is the best option here. I understand that's the easiest option. But I think you'll get a bunch "HUH? why did my custom assistant change from gpt4 to gemini pro" style messages.