Can someone help me how to resolve this issue. I created a custom assistant and it works fine in playground. But it fails when I am trying to call it from API. The error message is below:
“The model ‘asst_****’ does not exist or you do not have access to it.”
I checked all prelim check list to confirm I am calling the right one. And all are good. Except I am failing to call it from API.
Try to reach the support, still waiting on their reply. ( I felt it is live support. But unfortunately it is not and slow in responding;()
They need to be used through a different API endpoint, in multiple calls to construct a message thread, run it, poll and wait for the response, retrieve it.
Here’s where you can get an idea of how they work, and the coding it would take to interact with them.
Thanks for your response.
I thought it is a straight call like models. But it raises new concerns for me that would the performance be an issue when we have to deal with multiple API calls? Would it be better to have fine-tuned models as an alternative?
An assistant in Assistants isn’t making a new model or a fine-tuning.
It is just a way of having the “instruction” placed into a “developer” role message for you, and having a server-side conversation history maintained instead of using your own database to record and re-send past chats along with the latest message.
Take the text from your instruction box, and then use it as the first developer message in a list of messages that you send to chat completions, using the same model name, and your latest user question last. You’ll get the same kind of response.
Assistants adds value in a built-in file search function for retrieving knowledge, and a Python code interpreter than can run code the AI writes, also things you can write for yourself as Chat Completions functions.
So while there are utilities and conveniences in Assistants, it also comes with a higher coding requirement, and also the multiple API calls when you could simply make a single Chat Completions call and immediately start streaming the AI response.
Now I see the flow. And looks like custom assistants may become expensive than models, why because the assistant I have created has a very long instruction set and every call I make is having good token count.
Also one I thing I noticed is after reading the messages, the output is not having the actual expected result (in my case I was requesting a list in a json).