tender goblet
I built a small PHP/curl feature for my own use which takes large chunks of text, breaks them up into pieces with the same instructions on the front of the text, and submits to the GPT API as prompts in pieces that will not go over token limits. The results are collated back into a big file. It was all working GREAT, then pieces started failing here and there, returning only HTTP Error 400. Now, all pieces return that. I have plenty in my token budget, the failing prompts work fine in playground, I've got it sleeping 2 seconds between requests so I'm not hitting it hard at all... any ideas? What can I share? Verbose curl just shows it stopping after MAX_CONCURRENT_STREAMS:
* We are completely uploaded and fine
* old SSL session ID is stale, removing
* Connection state changed (MAX_CONCURRENT_STREAMS == 256)!
< HTTP/2 400
< date: Wed, 05 Apr 2023 21:30:57 GMT
< content-type: application/json
< content-length: 294
< access-control-allow-origin: *
< openai-model: text-davinci-003
< openai-organization: user-xxxxx
< openai-processing-ms: 4
< openai-version: 2020-10-01
< strict-transport-security: max-age=15724800; includeSubDomains
< x-ratelimit-limit-requests: 3000
< x-ratelimit-limit-tokens: 250000
< x-ratelimit-remaining-requests: 2999
< x-ratelimit-remaining-tokens: 246170
< x-ratelimit-reset-requests: 20ms
< x-ratelimit-reset-tokens: 919ms```
I can't find much info on that MAX_CONCURRENT_STREAMS line, but I really don't think I'm hitting it hard with one request every 2 seconds at most. Ideas of what to look for next? Is the API just overloaded? It all worked at first!