Vision requests are counting double?

When making a vision completion request call, my x-ratelimit-remaining-requests value is going down by 2 every time. Does one completion call count as 2 requests? I’m trying to manage the meager 100/day we get, and noticing that it’s actually only 50 is making me sad. :slight_smile:

Interesting: I suffered one failure, and then two successes. I’m at 95 remaining.

‘x-ratelimit-limit-requests’: ‘100’, ‘x-ratelimit-limit-tokens’: ‘40000’, ‘x-ratelimit-remaining-requests’: ‘95’, ‘x-ratelimit-remaining-tokens’: ‘39968’, ‘x-ratelimit-reset-requests’: ‘58m43.712s’

another identical call? No change reported. Maybe they cache? So I resize. Another error:

openai.InternalServerError: Error code: 500 - {‘error’: {‘message’: ‘Something went wrong processing one of your images.’, ‘type’: ‘server_error’, ‘param’: None, ‘code’: None}}

‘x-ratelimit-limit-requests’: ‘100’, ‘x-ratelimit-limit-tokens’: ‘40000’, ‘x-ratelimit-remaining-requests’: ‘90’, ‘x-ratelimit-reset-requests’: ‘2h19m49.595s’

Dropped from 95 to 90 with me getting the one with the same limit shown, another failure, then an image description. And as others reported, bizarre reset times you cannot possibly calculate against.

So your failure or any server error costs you limits too - and python can retry.

input = 1
output = 1

Confirmed.

cheapest connection test:
completion_tokens=10, prompt_tokens=286, total_tokens=296

2 Likes

Yeah, I turned off retries but have to wait to do more requests to try again… I was seeing the same drop by 5 on a 500 error. It’s the drop by 2 on a success that has me confused.

After turning off retries, my counts now decrease by one for every API call, as expected. That explains why a failure would result in 5 calls being charged, but I still don’t understand why every successful call was charging 2. What are the odds that the first one was failing and second succeeding every time?