It is January 6, 2024.
I got this kind of server error 7 out of 10 times when calling function with Assistant API. I read some threads on this forum and people said it was about the quality as well as the ability of the “instruction” and “user message”. I know it is still in beta and we don’t have any guidelines on how to debug effectively.
Does anyone experience this error consistently?
- It’s about using the right model, with 1106 version models being the only ones to properly support the specification of retrieval with parallel tool call.
- It’s about using the right model, with 1106 version models have significant problems calling functions and tools with accented or Unicode characters.
If those cases overlap, and unless you have new information about a previous application failing today, the solution may be to build a chat-completion based information retrieval system to enhance AI knowledge instead.
Or wait for new models to replace those with issues.
Maybe you can share more details about the function(s), the function repsonse , the model you are using …
Hey @_j , thanks a lot for sharing your experience and your time. A few hours after this post, I tweaked my “instruction” and “description” to clear out any possible confusion and keep it as simple and as short as possible. I stopped getting server error.
@jlvanhulst For the functions, they were simply 2 external API calls to get some info. As I said above, something things just started working again after some tweaks lol.
The model I am using is GPT-4-Preview-1106 or GPT-4.5-Turbo as they called it. I build my own RAG so I don’t use the built-in retrieval tool by OpenAI.
I’ll try out parallel functionc calls tommorow.
Thanks again guys, really appreciate