API to gpt-3.5-turbo-16k getting stuck in a loop until it reaches max tokens

API
1

This only started today: part of a recipe was in the json file then got stuck in what looks like a loop until it reached my max tokens of 8000:

  • 1/4 teaspoon black pepper
    • \n
  • 1/4 teaspoon dried dill
    • \n
  • 1/4 teaspoon dried parsley
    • \n
  • 1/4 teaspoon dried thyme
    • \n
  • 1/4 teaspoon dried oregano
    • \n
  • 1/4 teaspoon dried basil
    • \n
  • 1/4 teaspoon dried rosemary
    • \n
  • 1/4 teaspoon dried sage
    • \n
  • 1/4 teaspoon dried marjoram
    • \n
  • 1/4 teaspoon dried tarragon
    • \n
  • 1/4 teaspoon dried cilantro
    • \n
  • 1/4 teaspoon dried mint
    • \n
  • 1/4 teaspoon dried chives
    • \n
  • 1/4 teaspoon dried thyme
    • \n
  • 1/4 teaspoon dried parsley
    • \n
  • 1/4 teaspoon dried dill
    • \n
  • 1/4 teaspoon dried basil
    • \n
  • 1/4 teaspoon dried rosemary
    • \n
  • 1/4 teaspoon dried sage
    • \n
  • 1/4 teaspoon dried marjoram
    • \n
  • 1/4 teaspoon dried tarragon
    • \n
  • 1/4 teaspoon dried cilantro
    • \n
  • 1/4 teaspoon dried mint
    • \n
  • 1/4 teaspoon dried chives
    • \n
  • 1/4 teaspoon dried thyme
    • \n
  • 1/4 teaspoon dried parsley
    • \n
  • 1/4 teaspoon dried dill
    • \n
  • 1/4 teaspoon dried basil
    • \n
  • 1/4 teaspoon dried rosemary
    • \n
  • 1/4 teaspoon dried sage
    • \n
  • 1/4 teaspoon dried marjoram
    • \n
  • 1/4 teaspoon dried tarragon
    • \n
  • 1/4 teaspoon dried cilantro
    • \n
    2

    Welcome to the forum.

    We need more info. What settings (temperature, etc.) are you using? What’s your prompt look like?

    2 Likes
    3

    This happened to me also yesterday with the gpt-3.5-turbo-16k model. I presume it’s some sort of bug that they will fix very soon.

    1 Like
    4

    Happens in code too sometimes. Nothing new really. Happened months ago as well.

    5

    Try decreasing temperature a little.
    Some more text here

    1 Like
    6

    temperature is at 0 so I can’t go lower. I’m changing my prompt because I was giving it very specific and long instructions to have it format responses in html, but now I’ll just get the json and do it myself.

    1 Like
    7

    Just yesterday, i was trying to transcribe some audio with Whisper. My temp was 0.9 or something like that, and it did return a lot of repeated lines from some point in the audio onwards.

    Then, I changed to temp = 0. While it did improve the transcription a little (not everything from a certain point was repeating), what really made the transcription quite good was setting temp = 0.2 or so.

    Cheers!

    1 Like
    8

    I think I have a very similar issue and wrote a detailed report.
    On the GPT-3.5-turbo model messages were re-generated in a loop several times. Here is my issue / bug report. Maybe you want to check your usage and see if you have the same issue: