GPT 3.5i Turbo & Davinci models very slow and requests timeout | OpenAI | Page 1

chrome crystal Mar 19, 2023, 4:21 AM

#

I have been having issues with davinci and turbo models because they have been slow and I get timeout errors. We have users on our platform and would like them to have a reliable service to use. Is there just a constant influx of requests happening that is causing these issues to the openai servers. Also I saw there was an outage less than an hour ago is that still impacting the requests?

charred jackal Mar 19, 2023, 9:07 AM

#

Same here - the API is currently incredibly slow and often times out. Sadly this tends to happen every time OpenAI launches a new product. It's so inconsistent that it's almost impossible to use right now.

chrome crystal Mar 19, 2023, 3:13 PM

#

This is not fair at all. We’re paying for this service so that we can offer something reliable to a user base and new things keep releasing and screwing up the apis.

shut forum Mar 19, 2023, 3:56 PM

#

It's very slow right now

tropic solstice Mar 19, 2023, 4:21 PM

#

same here. very delayed response if any at all

stuck snow Mar 19, 2023, 5:31 PM

#

Also seeing this. The specific error I'm getting is: APIConnectionError: Error communicating with OpenAI: ('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer'))

#

In this specific case I had a 5 second buffer between requests

wraith walrus Mar 19, 2023, 6:03 PM

#

Also seeing this since last night - responses on Davinci completion are taking >20s and timing out

#

I was giving a live demo of my app this morning and had it fail due to exceeding the timeout. It was so embarrassing dalle_looking

echo cradle Mar 19, 2023, 6:12 PM

#

Same here. Terribly inconsistent connection since last night.(gptturbo3.5) But my servers located in different locations request at varying speeds and some being unaffected at all. I guess Openai has cut down the servers on certain parts of the API since yesterday,

rich sand Mar 19, 2023, 7:34 PM

#

Same here! I hope that in the future the API will be stable on response times. For production projects that rely heavily on this API, you'll have to trust their server

stable flume Mar 19, 2023, 8:57 PM

#

same here, sometimes it is fast, and sometimes it takes like 3 minutes for a simple prompt

heavy plover Mar 19, 2023, 9:36 PM

#

Same here

#

Extremely slow

#

Since GPT-4 and even a day before the announcement, everything has been slow for me on the API end with the exception of older models, e.g. Davinci 003

#

Can't be a coincidence anymore

shut forum Mar 19, 2023, 10:25 PM

#

feels bad spending money on ads to have user potentially get no response

chrome crystal Mar 19, 2023, 11:27 PM

#

This is actually ridiculous why do the early adopters have to get the short end of the stick

rapid otter Mar 20, 2023, 3:29 AM

#

The API is very slow

hybrid mulch Mar 20, 2023, 3:43 AM

#

Sometimes fast sometimes slow doesn't seem to matter what I send it

const response = await axios.post(
"https://api.openai.com/v1/chat/completions",
{
model: "gpt-3.5-turbo",
messages: [
...updatedMessages.map((message) => ({
role: message.sender,
content: message.text,
})),
],
},
{
headers: {
"Content-Type": "application/json",
Authorization: Bearer ${apikey},
},
}
);

charred jackal Mar 20, 2023, 5:51 AM

#

I can't even get an output it's so slow. I thought for sure it would be fixed by today... but nothing.

#

It just took 3 minutes to get a 1 paragraph response from davinci-003, and the other paragraph timed out. Setting cURL to 10 retries works, but I hope I'm not being charged for these retry attempts.

rough sapphire Mar 20, 2023, 6:33 AM

#

5 min response time on testing simple prompts in the playground

mellow copper Mar 20, 2023, 11:18 AM

#

Hey, I'm getting a lot of timeout requests today. This has been happening a lot the past few days where the request just hangs. Anyone else having this issue?

hearty pilot Mar 20, 2023, 11:33 AM

#

Yep, getting timeouts all the time. Started this morning for me...

low eagle Mar 20, 2023, 11:43 AM

#

is this issue getting solved soon ? my request via python and chat complete api using turbo freeze all the time here in germany today and yesterday... ?? can we get a stable working api back please ?

#

is there any rate limit for the request ? i have put a 3 second delay for every request now

#

i got again the error "Traceback (most recent call last):
File "C:\dev\content\head5.py", line 55, in <module>
response = openai.ChatCompletion.create(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\admin\AppData\Local\Programs\Python\Python311\Lib\site-packages\openai\api_resources\chat_completion.py", line 25, in create
return super().create(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\admin\AppData\Local\Programs\Python\Python311\Lib\site-packages\openai\api_resources\abstract\engine_api_resource.py", line 153, in create
response, _, api_key = requestor.request(
^^^^^^^^^^^^^^^^^^
File "C:\Users\admin\AppData\Local\Programs\Python\Python311\Lib\site-packages\openai\api_requestor.py", line 226, in request
resp, got_stream = self._interpret_response(result, stream)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\admin\AppData\Local\Programs\Python\Python311\Lib\site-packages\openai\api_requestor.py", line 619, in _interpret_response
self._interpret_response_line(
File "C:\Users\admin\AppData\Local\Programs\Python\Python311\Lib\site-packages\openai\api_requestor.py", line 679, in _interpret_response_line
raise self.handle_error_response(
openai.error.RateLimitError: The server had an error while processing your request. Sorry about that!
"

#

this time it seems to be really a ratelimit error

charred jackal Mar 20, 2023, 12:01 PM

#

Sadly these errors seem to happen every time OpenAI launches a new product… the last major one lasted a week without the API status saying there was a problem at all. The only hope we have is an OpenAI team member seeing this post.

#

If you use PHP, you can make curl retry until it works. Sometimes it takes 5-10 retries but it does eventually work.

dapper stratus Mar 20, 2023, 12:20 PM

#

Hi, question regarding long form text generation using the chatgpt3.5-turbo Api.
I've created a dynamic prompt that gets variables from a DB. The avg length of prompt is around ~430 tokens and the expected output should be around ~ 2000 tokens. I get a much smaller output than expected, I'm wondering if keep runing with a single long prompt to provide the output (better in terms of keeping the context) or either design the prompt to "remember" the previous m/ and run multiple prompts for each output. Any toughts regarding this? Tnx

low eagle Mar 20, 2023, 12:23 PM

#

dapper stratus Hi, question regarding long form text generation using the chatgpt3.5-turbo Api....

do an own thread for this ..would be the better idea i think...

tardy spear Mar 20, 2023, 12:34 PM

#

Been extremely slow for the past 2 days, half of the requests from my app are failing…

mellow copper Mar 20, 2023, 12:41 PM

#

tardy spear Been extremely slow for the past 2 days, half of the requests from my app are fa...

Same here. So annoying.

fast agate Mar 20, 2023, 1:06 PM

#

mellow copper Same here. So annoying.

I have been having the same issue, including internal 502 errors and general timeouts

finite marten Mar 20, 2023, 1:19 PM

#

If OpenAI wants developers to build their businesses on top of this API, it's going to have to get a whole lot more reliable than this.

normal cradle Mar 20, 2023, 1:59 PM

#

Yes, this issue started yesterday and we're still facing slow responses. Is there anyone here from OpenAI support?

tardy spear Mar 20, 2023, 2:08 PM

#

3 days of silence

finite marten Mar 20, 2023, 2:20 PM

#

In the meantime, I suggest making it apparent to your end-users which party is at fault for the errors and slow responses :P

heavy plover Mar 20, 2023, 2:22 PM

#

finite marten In the meantime, I suggest making it apparent to your end-users which party is a...

facts

normal cradle Mar 20, 2023, 2:52 PM

#

I tried asking ChatGPT to formulate an error message that blames OpenAI for the recent degradation in service but it wouldn't let me 😅
https://twitter.com/ofirski_/status/1637829046611460096?s=20

ofirski (@ofirski_)

@sama will be proud 😍

Trying to blame #ChatGPT's API which is extremely slow since yesterday.

shut forum Mar 20, 2023, 3:41 PM

#

finite marten In the meantime, I suggest making it apparent to your end-users which party is a...

ngl was thinking about this

floral cairn Mar 20, 2023, 4:11 PM

#

https://tenor.com/view/do-something-comeon-stickman-bored-gif-22722853

Tenor

#

me waiting for api to work

slow spoke Mar 20, 2023, 4:32 PM

#

User: How long does it take until our Sun starts expanding and we all burn do death?

GPT:
skeletonconfused

fast agate Mar 20, 2023, 4:40 PM

#

Still waiting...

brazen ore Mar 20, 2023, 5:08 PM

#

finite marten In the meantime, I suggest making it apparent to your end-users which party is a...

I’m literally going to implement this thank you 😂

shut forum Mar 20, 2023, 5:51 PM

#

I had luck with the API locally but it seems my prod site is totally dead

long marten Mar 20, 2023, 5:53 PM

#

shut forum I had luck with the API locally but it seems my prod site is totally dead

mine too.. been 12+ hours and counting.. not even 1 response from it

floral cairn Mar 20, 2023, 5:56 PM

#

havent heard anything from the openai team about anything either...

fast agate Mar 20, 2023, 6:04 PM

#

https://tenor.com/view/song-thesongthatneverends-lambchop-forever-gif-19132536

Tenor

fast agate Mar 20, 2023, 7:08 PM

#

Just to document, still, same issues and errors

#

Confirm if you can

teal quartz Mar 20, 2023, 7:53 PM

#

It seems largely resolved for at least text-davinci-003. I was seeing up to 2m timeouts on roughly half or a third of trivial requests, and now I can run 40 requests in a row all completing in under two seconds.

tardy spear Mar 20, 2023, 8:09 PM

#

Seems to be getting slightly better on my end too

floral cairn Mar 20, 2023, 8:12 PM

#

yep good here as well now.

#

ive also added backoff

#

https://platform.openai.com/docs/guides/rate-limits/error-mitigation

OpenAI API

An API for accessing new AI models developed by OpenAI

#

is def helping

long marten Mar 20, 2023, 9:27 PM

#

my api still not getting a response

#

tried 2 accounts, different keys

#

3.5 turbo or 4

rich sand Mar 21, 2023, 6:44 AM

#

Erros again and again last night but with this code this time:

data: {
      error: {
        message: 'The server had an error while processing your request. Sorry about that!',
        type: 'server_error',
        param: null,
        code: null
      }
    }

4th day of bug, it becomes difficult to trust

low eagle Mar 21, 2023, 8:44 AM

#

im using some delay between api requests.

dapper stratus Mar 21, 2023, 9:19 AM

#

low eagle im using some delay between api requests.

of how long?

low eagle Mar 21, 2023, 9:20 AM

#

not sure some seconds

#

5 - 10

loud rampart Mar 21, 2023, 10:01 AM

#

same with gpt4, almost not working, waiting for respon 160+ seconds

rich sand Mar 21, 2023, 11:41 AM

#

For me it seems to be fixed, no error since this morning

#

For gpt-3.5

#

And fast responses

red mica Mar 21, 2023, 2:36 PM

#

Rediculous trying to get anything done over the API during working day, they need to just end Free ChatGPT period. So its not impacting other services

floral cairn Mar 21, 2023, 5:00 PM

#

low eagle im using some delay between api requests.

look into backoff lib:

https://platform.openai.com/docs/guides/rate-limits/request-increase

OpenAI API

An API for accessing new AI models developed by OpenAI

#

might help

finite marten Mar 22, 2023, 12:58 AM

#

red mica Rediculous trying to get anything done over the API during working day, they nee...

I would be fine with a small business tier if it meant reliable service on separate infrastructure

prime temple Mar 22, 2023, 2:40 AM

#

curious - if locally it works fine, but in production seems to time out - any suggestions on what could be causing it? (streaming, PHP)

#GPT 3.5i Turbo & Davinci models very slow and requests timeout