Assistant is repeating itself in a single run

Hi there,

My custom assistant seems to have changed behaviour in the past hours.
Note: I did not change anything in its configuration.

Here is my use case:

  • I create a new thread for a specific assistant.
  • I add a User message.
  • I start a run with special instructions
  • Assistant is adding a new message in thread. It’s a correct/expected answer.
  • I add a new User message.
  • I start a new run with special instructions (not the same as previously)
  • The assistant is adding multiple messages in the thread. All the same, within 2 minutes. It’s always the same message. It’s always adding exactly 20 of them. The message is correct and expected though.

That has the critical side effect that my run API call times out.

Note:

  • Of course, my custom instructions are not asking my assistant to repeat itself.
1 Like

I’m having the same problem, with the same model.
I start a thread, add a “user” message. Start a run on the thread, but the run keeps being “in_progress”. When looking at the thread, the assistant answers the “users” message 20 times, and then completes. Why? Can I stop this somehow? It eats a lot of tokens. I only need one response! :smiley:

1 Like

Fascinating. Did you try creating a separate assistant that is completely the same except that it has the custom instruction as the Assistant instructions and no custom instructions in the run?

I finally fixed this.
Changed the model from “gpt-3.5-turbo” to “gpt-3.5-turbo-1106” and the problem went away.

1 Like

Same here, switching model fixes the 20 messages issue.

1 Like