text-davinci-002 creates a completion but the content filter (
content-filter-alpha) returns a 503.
What should I do in this scenario? The completion will count against my quota but I can’t use it in my app because it can’t go through the content filter.
That’s a good question. 503 means the server is not ready to serve the request - not necessarily that it is down. I would say you should implement a loop in your app that will check repeatedly with a delay between each check. I’ve implemented this in all my GPT-3 completions because sometimes the API just isn’t available. I have a 1 second delay between checks, and right now, my testing apps will make the attempt 5 times before giving up. It almost always goes through on the second try in the rare cases where it does fail on the first attempt.
Thank you for your response! The OpenAI requests happen on my Node server and I count them against my user’s quota. The problem is, I don’t want to increase the user’s quota usage when they don’t get an actual result. But the call to
text-davinci-002 still costs me money. So I guess what I need is a way to stop the API requests completely until the content filter is back up.
So maybe I should implement some sort of timer that suspends the API calls for a while whenever the content filter returns an error.
What do you think, does this make sense? Or is there a better way?
Hi @florianwalther, apologies for any issues this might have caused you. Implementing retries with exponential backoff is often a good idea when working with APIs like ours. We’re evaluating ways to make the experience smoother here. I’ll reach out directly to you to help sort out any quota issues stemming from this.
Thank you for getting back to me!
Just to make this clear, this didn’t actually cause me any damage. I don’t have many active users yet. I just wanted to prepare for similar situations in the future.