Rate limit reached token count wrong?

[marble bridge]

I'm getting the following error message from GPT somwhat consistently from my application:

 "Rate limit reached for default-text-davinci-003 in organization org-NQ6krFT3xTPVZKaqFuTvGx5N on tokens per min. Limit: 250000.000000 / min. Current: 260000.000000 / min. Contact [e-mail] if you continue to have issues."

When I look at usage on the https://platform.openai.com/account/usage page I dont see any 5 minute periods with over 100,000 tokens used let alone 250,000 in a single minute. Wondering whats the cause of this discrepancy?

[misty needle]

I get the same thing. Sending no more than 10K token per 5 minutes. The only workaround is setting a more conservative retry policy.

[acoustic cloak]

First, please check whether your api key has been leaked, just delete the original one and try again. But getting rate limit error without actually reaching the limit is indeed a problem met by many people

I think you can temporarily solve this by using try{} in your code to generate the response several times before getting a response

You can also contact the support in the main website

[marble bridge]

Thanks, we already have backoffs built in, it’s not clear to me it’s really helping us though as it seems to often continue failing and just backs things up further

I don’t think there’s a leak, the usage page all matches what we see on our end and is below the backoff limits and spending is as expected