Hi,
Run status is hung in “in_progress” state for many threads (in playground and direct API testing).
New threads on the playground are also hung. I have been trying for many hours on this.
Thanks.
Tool calling has been damaged and degraded in several models, with the AI emitting tools that were never needed and can’t work, and I have yet to get official feedback to my bug report and other thread.
It may be that the AI has emitted functions to nonsensical recipients and gotten stuck in an error state or is burning through retries on your dime.
You can retrieve the run steps and see how many iterations have been run on a thread.
I would ensure that you have a watchdog that can terminate and cancel any runs that would be left going crazy.
gpt-4-turbo-preview may be the best and only choice to attempt using assistants.
For whatever its worth I have been doing quite a few runs today with extensive function calling on gpt-4-0125-preview without any problems
I have exact same problem right now. It was working well until 6 days ago and now suddenly it is broken.
Seems like a universal issue with the assistant API at the moment
In my case, it suddenly getting stuck on in_progress when polling. It is pretty much consistent and the API is practically not usable anymore.
I’m facing the exact same issue. It suddenly stopped working, and it’s stuck in progress forever. The API status tho on openAI says it’s operational?
I have the same problem. It is also happening when I test it on the playground.
I dont know about that maybe status.openai have some problem too or maybe they didnt updated it for today. Look at the comment above, that dude has the same problem too
Ya, definitely something on openAI’s end. Will give it a try again later.
I implemented proper timeout handling with retries for querying the run status, and it resolved the issue. Now, if the API endpoint hangs and doesn’t return the next state in the current async call, I initiate a retry, to get actual status and everything continues smoothly.
How do you initiate a retry can you help me about that?
Getting the same issue on my end, my run can return one message but subsequent messages never get out of in_progress state.
Your answer is very good, it will help here
I’m having the same issue, and its costing me alot of money… Can we get a refund for threads that get hungup or stuck in a run?
As has been mentioned by many before, this is beta. It…is…NOT…production quality. Yes, you’re paying OpenAI to be a beta user.
Until it is stable, expect these kinds of issues. Do not run your prod on OpenAI beta.
That said, the AssistantApi will be a future winner; because the semantics are pretty awesome. Most of the issues are in the execution plane(getting llms to respond). The data plane (like creating an assistant, thread etc) just works.
Started facing this issue all of sudden today. After one or two back to back response, the run hangs.
The work around I am doing is to select an older gpt-4 model. Luckily, not only does it perform better, but it isn’t hanging.
This is a temporary solution. I think a good fallback behavior in general (automated by the assistants api if an agent doesn’t respond in time).
False alarm. I added a tool to my assistant and forgot about it and upon retesting the workflow the assistant started calling the tool. The way I had designed my logic, my program would hang. I added handling for this case (which I should have done right off the bat).
My runs are moving to expired status and i am not able to find out why. CAn someone please provide if there is a documentation on when run moves to expired status.