pine grail
Seems like the past hours the API is producing a generic rate limit error. I cannot find another reason but to assume the service is down or intermittent:
openai.error.RateLimitError: The server had an error while processing your request. Sorry about that!
Using python 3.10
python openapi client 0.26.4
I'm located in EEST (+2 GMT) Anyone else seeing this? Could it be a local issue?
hazy drift
I just started getting it
grave zephyrBOT
@hazy drift
camil0's message blocked by AutoMod
pine grail
This is happening all day for me. The model is text-davinci-003.
i'm on the paid plan and most def not hitting any limits. Also, the error is that the server had an error...
sour folio
Are you still getting this
strong sail
@pine grail ^
Also, are you on a pay-as-you-go account? And can we see the request that you're sending?
pine grail
Hi @strong sail ,
Thanks for reaching out. Yes I'm a pay as you go account. How would you like me to share the request? Like the python code with the prompt?
strong sail
Hi <@213045272048041984> ,
Thanks for reaching out. Yes I'm a pay as you go acc...
Yeah! The python code and the prompt 🙂
pine grail
def call_the_model(prompt, max_tokens=550):
result = openai.Completion.create(
engine="text-davinci-003",
prompt=prompt,
temperature=0.9,
max_tokens=max_tokens,
top_p=1,
frequency_penalty=0,
presence_penalty=0.6,
)
return result['choices'][0]['text']
company = "OpenAI"
call_the_model(
f"Generate a short company description for {company}:"
)
This is pretty much a snippet out of the app. Having said that, the error comes and goes, sometimes the code works and sometimes I get the aforementioned rate limit error.
By the way I have also opened this topic on the forums, with several other people responding to the thread on this:
https://community.openai.com/t/ratelimiterror-the-server-had-an-error-with-no-reason-given/50146/15
grave zephyrBOT
sour folio
This error means the server is having issues
grave zephyrBOT
warm blaze
This error means the server is having issues
This error has been happening consistently for almost 72 hours now… it’s a universal error and not prompt-based. ~30-40% of outputs fail and result in “server_error.” It seems to only happen with davinci-003.
“The server had an error while processing your request. Sorry about that!”
It has completely destroyed my product and many others, and we’d really appreciate if you guys could try and fix it asap :/
I have tried just about everything including adding up to 20-second delays on each request, it does nothing. 30-40% of the requests still fail.
P.s I’ve been developing products/working with the API for well over a year, so I can guarantee with absolute certainty that this is not user error.
There are also other threads with the same error:
Lower in the thread:
devout snow
This error has been happening consistently for almost 72 hours now… it’s a unive...
Out of curiosity, are you from USA?
warm blaze
Out of curiosity, are you from USA?
Yes why?
tulip pivot
I used the following libraries to work around this issue:
ratelimiter https://github.com/RazerM/ratelimiter
retrying https://github.com/rholder/retrying
from ratelimiter import RateLimiter
from retrying import retry
from secret import OPEN_AI_API_KEY
openai.api_key = OPEN_AI_API_KEY
@retry(stop_max_attempt_number=10)
@RateLimiter(max_calls=20, period=60)
def generate_openai_completion(
prompt,
model="text-davinci-003",
max_tokens=100,
temperature=0.68,
number=3,
top_p=1,
best_of=10,
frequency_penalty=0.8,
presence_penalty=0,
):
response = openai.Completion.create(
model=model,
prompt=prompt,
temperature=temperature,
max_tokens=max_tokens,
n=number,
top_p=top_p,
best_of=best_of,
frequency_penalty=frequency_penalty,
presence_penalty=presence_penalty,
)
return response```This doesn’t make sense for paid accounts as the limit is 3000/minute, not 20/minute. This library does nothing but slow down the calls which is unnecessary if the server was working as expected.
warm blaze
The error rate seems to be getting worse, we’re up to ~40-50% of requests now. The only way to counteract is setting a cURL retry limit of ~5+. Often, 3/5 fail, but eventually there is one that works. This is undoubtedly a severe problem with OpenAI severs.
warm blaze
Example of failures... Hopefully I'm not being charged for the prompt tokens on the failures, because every other request requires 2-3+ retries before I get a successful one.
pulsar jackal
What's the purpose of pay as you go when service is most of the time down
Monthly payment will always be almost nill
Each and every request is getting Error 429
pine grail
I can double everything @warm blaze says.
How about a response from OpenAI, so that we can plan our next steps? I cannot see how silence is helping anyone.
cc @strong sail
grave zephyrBOT
I can double everything <@389317598577164291> says.
How about a response from ...
@pine grail
Reason: Reached maximum mentions (3 per message)
warm blaze
I can double everything <@389317598577164291> says.
How about a response from ...
The current solution is to implement retries and looking for the $result['error'] as a sign to retry. It's a really frustrating way to do it (and I seriously hope I'm not being charged for the prompt tokens), but it does work.
It is quite sad that this has been going on for 72 hours, while OpenAI's "status monitoring" page has shown "fully operational" the entire time. I can't even imagine how many products have been broken because of this error...
inland scroll
i reinstall python. then i get right result only first
warm blaze
i reinstall python. then i get right result only first
It's random, but 40-50% of all requests to davinci-003 are failing at the moment, and so far, still no word from the OpenAI team... which is almost hard to believe at this point.
nimble cape
same here. i use Python 3.11.1
cursive hemlock
The current solution is to implement retries and looking for the $result['error'...
My prompt is long, when the request doesn't fail, connection lasts more than 1 minute.. catch and try wouldn't be a solution in my use-case.. too much time. I was ready to base new implementations on these APIs.. really bad news today for my company
cursive hemlock
Out of curiosity, are you from USA?
I'm from Europe and still have these errors, so I suppose this is error is not happening in some countries only.
devout snow
I'm from Europe and still have these errors, so I suppose this is error is not h...
No, it doesn't seem to be. Hopefully it settles soon
cursive hemlock
Did someone see if the request fails immediatly when calling or after some time?
I can temporary use a retry system in the try-catch block but it makes sense only if the requests fails immediatly.. my prompts are too long
coarse fable
Is there anyone from @grave zephyr that can respond to this?
I’m building an app that uses embeddings and forward facing davinci, and it’s causing problems. The support has not been great so far.
It’s at the point where I’d base my app on other tooling as soon as it’s available
rustic drum
Seems like the past hours the API is producing a generic rate limit error. I can...
Same here. Change all the API_KEYS. No impact!
open lintel
Same here. Change all the API_KEYS. No impact!
does changing the api keys fix the problem?
somber tartan
Same problem for me for the past 4 / 5 days. Very annoying. Sometimes a time.sleep of 1' between calls mitigates the error rate but even with 1' between calls I end up getting the error.
open lintel
i'm getting the error but it isn't because i'm hammering the api...i'm going way beyond 30 seconds a call...just trying to work with it
somber tartan
Is there anyone from OpenAI on this Discord server? 🙂
coarse fable
I truly don’t even know. It took them a week to respond to my last support request on their site
It’s like I’m literally trying to give you money
somber tartan
I can't imagine how painful this error is for the startups which have built services on top of GPT-3...
At the moment you can't even make 1 API call per minute (MINUTE) without an error. Inacceptable.
rustic drum
I changed all my functions to implement a while attempts try / except + sleep 15 seconds calls... it seems to work.
rustic drum
Is there anyone from OpenAI on this Discord server? 🙂
not sure
somber tartan
I changed all my functions to implement a while attempts try / except + sleep 15...
So basically you retry the errors until it works? Hopefully you don't pay for missed calls...
rustic drum
hahaha... hope so too
rustic drum
So basically you retry the errors until it works? Hopefully you don't pay for mi...
It's seems not
peak musk
ok but how many seconds between two calls with response?
rancid olive
I changed all my functions to implement a while attempts try / except + sleep 15...
can you explain how you do this i have the same problem
lament moon
too many requests error is annoying me for 2 straight days now lol. even if I create a new key, this error still comes out.
rustic drum
can you explain how you do this i have the same problem
attempts = 0
while attempts < 5:
try:
completions = openai.Completion.create(
engine=model_engine,
prompt=prompt,
max_tokens=size,
n=1,
stop=None,
temperature=temp,
api_key=api_key
)
message = completions.choices[0].text
return message.strip()
except:
attempts += 1
time.sleep(5)
print("Error")
sys.exit()
coarse fable
I changed all my functions to implement a while attempts try / except + sleep 15...
This is a waste of time and money, and isn’t the SLA for the api
Let’s not push sub optimal solutions that don’t solve the problem for a lot of people so Openai actually responds
lament moon
attempts = 0
while attempts < 5:
try:
completions = openai.Completio...
Thanks. I did something else for my bot. But still, it's really bad when I'm paying for the API 🥹
north flame
😵 every employee is on it now right ? using heroku and their 30s limit just delete my AI feature
sour folio
I’m going to look at this error today. For anyone getting it and wants me to look at their account specifically please dm me your account email
sour folio
Thanks everyone
sour folio
this should be resolving, please let me know if its still happening for you
grave zephyrBOT
coarse fable
Moxi any idea if there’s a request per second cap for the API too? Everything is listed as request per minute, but I’m assuming that gets rolled out to requests per second at some stage?
For example, 3000 rpm translates to 50 requests per second — if I go above 50 will I be limited?
lyric forge
I have posted an update on the community pages for people that are not on Discord
sour folio
Moxi any idea if there’s a request per second cap for the API too? Everything is...
I don’t think there’s a per second
Though the limit may be a slighting window
(Haven’t checked)
Sliding***
FYI you can request a higher limit
coarse fable
yeah, I sent one in…sadly seems to take a bit of time so I’m getting a little antsy
but I think I might be ok for the time being if there isn’t an RPS cap
rustic dust
The current solution is to implement retries and looking for the $result['error'...
My WhatsApp bot with 8000 users
sour folio
if you see this again after now, please post here. im getting dm's but unsure if it's stil happening. everything looks fine on my end
leaden flower
Is there a way to increase the codex-davinci-002 rate limit even with paid subscription? Currently davinci-002 is in beta.
warm blaze
if you see this again after now, please post here. im getting dm's but unsure if...
I will be testing for the next hours and report back with what I find.
coarse fable
I've been good, but haven't had as much burst again today
cursive hemlock
if you see this again after now, please post here. im getting dm's but unsure if...
Working fine right now, seems to be fixed, today I'll work full time on gpt-3 implementation in our software so I'll tell you
vast charm
Is there anyone from <@1046280307462123561> that can respond to this?
I’m build...
curious what kind of issues you're running into? have been experimenting on a similar project using embeddings with some friends
warm blaze
It seems the error has finally been resolved...
north flame
I still have hight latency, and sometime some 503
is it because we got a big prompt ?
warm blaze
is it because we got a big prompt ?
Sadly these type of errors are fairly common with davinci-003... but the error being discussed in this thread was not.
north flame
Sadly these type of errors are fairly common with davinci-003... but the error b...
dang, and did you know if an issue I could follow up on is opened about this ?
lunar ledge
This is unrelated to the API Ratelimit Error, look up the new Version of the OpenAI Library to get async Completion.
sour folio
dang, and did you know if an issue I could follow up on is opened about this ?
when server is overwhelmed you'll see 5xx
weirdly i dont see it at all in my bot, but i use max tokens 200, though even at 500 it was fine
sour folio
dang, and did you know if an issue I could follow up on is opened about this ?
I’ll check your account today
there is elevated 5xxs, ill fix today
sour folio
The errors seem to resolve but let me know if you’re still seeing issues
north flame
unfortunately I still experience issues on my side, I see that other thread has been opened with the same concern today
in a general basis do you recommend to lighter the prompt to increase perf ?
lament moon
I did try that already but still getting errors. I even did text embedding.
serene ocean
Im definitely still getting these errors.
cursive hemlock
I still have hight latency, and sometime some 503
same here. Sometimes random 503 responses
sour folio
Should have resolved
lyric forge
Thanks
warm blaze
Should have resolved
The error is back and worse than ever...
warm blaze
Should have resolved
The error is back and worse than ever. 15 in a row! I have 20 retries and it almost hit the limit.
This is the exact same error as last time, surely there must be some sort of permanent fix...
cursive hemlock
Should have resolved
Not for me. 429 came back. One request after not sending requests for different minutes. I'm going to write you my account email
warm blaze
Not for me. 429 came back. One request after not sending requests for different ...
It's back for everyone, not just you.
cursive hemlock
thought she fixed it this night.
warm blaze
Sadly it seems it was just a temporary fix... because it's than ever for me right now.
fringe dome
openai.error.RateLimitError: The server had an error while processing your request. Sorry about that!
Any soultion for this? @grave zephyr
cursive hemlock
openai.error.RateLimitError: The server had an error while processing your reque...
We are already talking about this. Read the chat history
fringe dome
And I found no soultion from the chat.
That's why I asked for the latest update from officials.
cursive hemlock
Moxi is from OpenAI and 6 hours ago said it was fixed (again), we're telling her this is not fixed, so tagging them again is not needed
This is right now the number of attempts for each completion (try-catch under while block).
As you can see is insane
fringe dome
Hmmm Okay ✅
cinder vessel
same
gritty birch
I have the same error. Moreover, sometimes it gives me answers and prompts which are unrelated to what I send. Did any of you had this issue?
cursive hemlock
I have the same error. Moreover, sometimes it gives me answers and prompts which...
I had a strange response. I asked openAI to translate a text to spanish and answered with a part of text in english without spaces but that was not "unrelated" that was like it went crazy for a moment.
"Divanes y Camas a BargainPrice: Modelos y Precios"
Makes no sense.. why completion contains 2 english words without Space, in the middle of a spanish translation? The from language was italian so makes no sense
vivid steeple
I have the same error.
{"message":"The server had an error while processing your request. Sorry about that!","type":"server_error"}
restive lily
Mine are either returning 429, or timing out even for very simple requests (and even with a 90 second timeout).
fringe dome
And the strange thing is, in playground it keeps giving network error and stops completing text. Although there's no connectivity issue. API error and playground error is different but the behaviour is same for text-davinci-003. Sometimes work and sometimes don't.
sage pewter
Each 2 request with 429 error, (paid account with extended limits)
plucky quarry
There is a easy way to solve this issure temporarily that create a new SECRET KEY. I think the reason is the prompt is too long. I read above the information just now, and there are several like me.
warm blaze
We are all still waiting on a reply from the team... I just got 18 retries before a successful submission lol. OpenAI is really dropping the ball on davinci-003. Every other day this error is happening.
warm blaze
There is a easy way to solve this issure temporarily that create a new SECRET KE...
Why would creating a new secret key do anything? They've fixed this error multiple times in the past few days... and it keeps reoccurring. It's OpenAI servers.
sage pewter
There is a easy way to solve this issure temporarily that create a new SECRET KE...
We tried to change it, doesn't help
plucky quarry
Why would creating a new secret key do anything? They've fixed this error multip...
😩
warm blaze
We tried to change it, doesn't help
They must be doing some sort of temporary fix for this issue... but it really needs a permanent one. I don't see how people can realistically create products around davinci-003 when the OpenAI servers have errors every other day.
plucky quarry
So how do we get their attention to solve this problem once and for all.
sage pewter
We temporarily switch to davinci-002, it works much faster and more stable, but worse in quality
warm blaze
So how do we get their attention to solve this problem once and for all.
Moxi will see these replies in the morning I hope, and maybe she/the team will realize that it needs more than a temporary fix.
warm blaze
We temporarily switch to davinci-002, it works much faster and more stable, but ...
Yes it seems to be only davinci-003 that has the issue, you're right, 002 works but is significantly worse quality, and much shorter replies. The prompts have to be crafted in an entirely different way, and I've already converted my prompts to be for davinci-003.
On a positive note, one request took 18 retries, and I'm 99% sure I wasn't charged for the errors. So it seems the prompts are not costing tokens when the error occurs.
But it took... 12 minutes to complete one article because of so many retries lol.
sage pewter
Yes, but it is a better to have a working production today. I'm also thinking about finetuning previous version to get more stable and cheap solution.
warm blaze
Should have resolved
I'm going to sleep... but I'm going to post this one more time so maybe Moxi sees it.
The error is back, and worse than ever. Last time the most I saw was 3 retries... now I've seen 15-18 in a row, which is insane.
wild flicker
Same here ! A lot of "The server had an error while processing your request. Sorry about that!"
We have a paid account with extended limits also
restive lily
We are all still waiting on a reply from the team... I just got 18 retries befor...
Very well put. It's very, very tough to monetise any product linked to d-003 at the moment.
tranquil pagoda
It's disaster, and its always broken just exactly when it hurt the most
waxen oasis
plucky scroll
same here: all day long
openai.error.RateLimitError: The server had an error while processing your request. Sorry about that!
north flame
I don't know what to do...I can't launch my product
I've tried to switch API key, the prompt in a Lite version isn't reliable
fierce peak
More Fun, No stress. Yeah, sure!
oblique owl
i am a plus account but i get <Response [429]> {
"error": {
"message": "The server had an error while processing your request. Sorry about that!",
"type": "server_error",
"param": null,
"code": null
}
}
rigid sedge
This issue kept coming up today but never happened to me before. I doubt if it's because I'm calling API too often within a short amount of time? (though it's still < 50)
cursive hemlock
Doesn't appear to be a real rate limit, there is something wrong in their servers. We all get this error even if sending one completion request/minute
fierce peak
Doesn't appear to be a real rate limit, there is something wrong in their server...
but apparently https://status.openai.com/ doesn't report any issue
cursive hemlock
Yes we know. This problem started some days ago and some OpenAI staff member already replied this discord thread
They fixed it about 3 days ago but the problem came back yesterday
So this is not "officially" recognized as openAI problem but the staff confirmed they have internal issues
restive lily
Yep, and is considerably worse this time.
cursive hemlock
Yep, and is considerably worse this time.
Yes.. way worse
restive lily
I have paying customers sending me the link to OpenAI's status page 😓
cursive hemlock
Today i've moved to another poject because I was not able to test completions without getting 3/4 errors before
cursive hemlock
I have paying customers sending me the link to OpenAI's status page 😓
I was ready to change the entire content creation and translations departments of the company thanks to openAI API implementation in our internal softwares. I'm glad I said to my boss I wanted to test some days before doing that..
restive lily
Oooof, good dodge! It's brutal for things like marketing efforts, too - paying $10 to bring someone in the door, only for the product to not work for them...
sour folio
Ugh ok looking
This is text davinci 003??
I’ll look into why this isn’t paging today
cursive hemlock
This is text davinci 003??
Yes, we talk about davinci 003.
Thanks. I sent to you in PVT my account email so you can check logs if it can help
sour folio
should be better now? anyone still seeing issues
grave zephyrBOT
RayD#9979 said:
lyric forge
I am sure everyone here is asking the same question but doesn't want to ask. Moxi are you the ONLY tech person working on Davinci. Are you also the ONLY tech person monitoring for downtime. How do we get things like this fixed when you are sleeping. WHO do we contact and HOW. Rant over (for now) .....
sour folio
we have oncall rotation but there is another fire theyre fighting
oncalls do get paged through the night and they were woken up for other issues there's just a lot of fires
we will be more stable soon
lyric forge
It is really tough for us all out here. We are trying to use your tech and presenting to investors that take a long time to get on side and then the tech blows upmonnour face with no response for 12 hours+
sour folio
we understand, we're making this better
lyric forge
I'm not the only one with this problem . I am sure other would chip in but I would prefer the forum to be a productive place. So I will end my rant here and hope you take this on board as how everyone is feeling. Logan is not replying either and he is our contact on the community forum pages. Thanks for fixing the issue again Moxi. Appreciated
sour folio
is it resolved for you now btw
lyric forge
Suggestion: If it is such an easy fix can we have a page somewhere that we can get someone's attention to flip the same switch you did. It seems it was fixed in a minute or two and we just needed to get the right person (your) attention..thanks again
sour folio
its not an easy fix
its a different problem every time
(i didnt fix it the oncall was already on it)
fierce peak
its a different problem every time
i can imagine the complexity of managing a LLM as service, with several critical points of failure.
north flame
I'm still being kicked out by 503 sometimes (most of). I'm located in France if it could help.
sour folio
I'm still being kicked out by 503 sometimes (most of). I'm located in France if ...
i see it, looking
between you and me if everyone uses our api appropriately (follow limits on rate and billing) we wouldnt have nearly as much issues like this
north flame
I didn't do the math yet but if you see my account, would you say we don't use your api well ?
grave zephyrBOT
Paul Bellow#0488 said:
sour folio
no, you guys are all good
foggy charm
Moxi is it a matter of hitting the API in loops? On the retro forum, we have a lot of questions about people wanting to hit the API 50+ times in a row rather than do long paragraphs... because they said they got "better results"
sour folio
if you dont know if you're doing it wrong, you're doing it right
people are doing it on purpose
as long as you stay under your rate limits (what the numbers in real 429s say) you're good
foggy charm
I'm not looping myself, but I see a lot of devs talking about it...
sour folio
i think detaul is 60 per minute so in a row is ok
glacial flume
Hey Moxi, still getting the error when using davinci-002 it's ok, but use davinci-003 it's basically impossible.
sour folio
Hey Moxi, still getting the error when using davinci-002 it's ok, but use davinc...
500 errors?
glacial flume
500 errors?
"Could not connect to remote server"
sour folio
what is the error code
glacial flume
what is the error code
500
fringe dome
We didn’t get 429 error but “server had an error while processing your request”. Has it been fixed for all text-davinci-003 api users?
wild flicker
Thank you for your response Moxi. The teams need to update your status page, because when we explain to our customers that the problem is with OpenAI, they send us your status page in response, which says that everything went well today...
This is a real problem for customer relations
sour folio
should be going up
restive lily
should be better now? anyone still seeing issues
Thanks for the comms - you're the best! I can't even imagine the volume of fires with this sort of scaling 🔥
I imagine you're also trying to hire ### engineers, which itself takes ages
and then they have to onboard/upskill
I haven't seen a 429 for a while...
The request times on davinci-003 seem real slow though
sour folio
yeah we are hiring a lot and onboarding takes time, we are looking at 003
restive lily
Awesome to hear. 2 min response for a 1k max tokens seems a bit excessive. I'm new-ish to using these APIs, but some of the feedback on the forums in early Jan was that responses were much faster.
I did wonder if prepping to publish the ChatGPT API was pulling focus off maintenance.
upbeat patio
There was an error generating a response
foggy charm
I'm getting 504 gateway timeout on TD3
seems to be back... thanks for all the hard work! it's appreciated!
cursive hemlock
i'll check it tomorrow in the morning (CET)
grave zephyrBOT
Pandal#3955 said:
warm blaze
I'm getting 504 gateway timeout on TD3
The particular error being discussed here seems to be resolved, but I am getting timeout errors now off and on as well. I waited 15 minutes for one response, and it ended up timing out. I'd almost rather have the "Sorry!" error than this... at least with that one, after 18 retries it would end up working.
fringe dome
Hey Moxi same error occurring again.
tranquil pagoda
From yesterday we have even 35 timeouts / 10 minutes. Completions now took more then 2x of the time compering to the moment i build the app and design prompts month ago. Api on heroku limit time to 30 sec :/ Someone have advice haw to make it longer? Where you host your openai apps?
warm blaze
From yesterday we have even 35 timeouts / 10 minutes. Completions now took more ...
I'm experiencing the same thing, ridiculously long response times and frequent timeouts. I truly hope that one day we will be able to use davinci-003 without near-constant errors and timeouts.
fringe dome
I'm experiencing the same thing, ridiculously long response times and frequent t...
Doesn't seem like near future plans 🙂
warm blaze
yeah we are hiring a lot and onboarding takes time, we are looking at 003
These timeouts are 100x worse than the other error. At least with that error I could keep retrying and would eventually get a response, even if it took 18 retries.
This is my current attempt to generate an article with 5 sections, 200 tokens per section in a loop with a 2 second sleep in between requests... That's 17 minutes... and nothing. I had to restart apache and kill the process.
It seems to be happening frequently as well.
tranquil pagoda
@warm blaze I think that manually setting the timeout to 30-60 sec paradoxically may be better in this case
warm blaze
<@389317598577164291> I think that manually setting the timeout to 30-60 sec par...
You're probably right, though, it's a tough call as the davinci-003 response time can be quite slow even without the timeout bug that's happening now. At some point I increased the curl timeout to 1000 seconds because the responses were so slow (but would eventually go through).
cursive hemlock
I still have 429 and 502 random even if I call the api 1/minute
And really high response time too
sour folio
i am looking
and i swear im gonna fix alerting for this today, i think its not being alerted on properly
sour folio
Doesn't seem like near future plans 🙂
we are working on it 😭 ive been so swamped with dalle
Thank you for your patience and for helping make this better!!!
rustic drum
Hi, i have a huge latency right now on davinci03.
more than 1 min / request. Can you confirm ?
sour folio
yeah looking
fringe dome
If our output is supposed to be 100 tokens but we keep it 800 tokens in API call, will it result in delaying or loading the server Moxi?
rustic drum
yeah looking
For your information my ID is mlevypro@gmail.com I have a batch working right now. Around 90 requests. at least 5min/req since 15 min
sour folio
For your information my ID is mlevypro@gmail.com I have a batch working right no...
i cannot find your requests. can you share a request id (should be in response header)
grave zephyrBOT
AthenaAI#6202 said:
restive lily
Hey again Moxi, I definitely think the latency has improved in the last 24-48 hours, c011105580a7c95c5b9a02bebe8d3521 (just now) still took 40 seconds though.
Could this be the complexity of the request? The same request on ChatGPT takes a fraction of the time.
restive lily
Same request just took 22 seconds, and then 28. I'll take a ~35% improvement at a time when you must be totally swamped.
Not sure if you can answer this, but presumably the latency on the ChatGPT API will be similar...?
sour folio
Really depends on how much resources we can give it at any time
And the general load
It shouldn’t take that long unless you are requesting from it 4000 token requests
Share a request id?
warm blaze
Everything seems to be working fairly well at the moment thankfully... not getting any timeouts.
restive lily
Agree - great work Moxi and team 🤩 🙏
rustic drum
Latency I mean
Sorry, I am back. This morning (based in France), it seems a lot better 🙏🏻
mystic badger
I am free trial user but I am not able to make requests through API
it says that you exceeded your current quota, please check your plan and billing details
Weirdly, I haven’t been allocated those free 18 dollars
Can anyone help?
restive lily
@mystic badger, this thread is specific to 429 Rate Limit Errors. This also sounds like a general billing customer support query.
mystic badger
<@885529740667134002>, this thread is specific to 429 Rate Limit Errors. This al...
Is there a separate channel for that? I have asked in another channel too BTW
strong sail
I am free trial user but I am not able to make requests through API
Try upgrading to a pay-as-you-go account, free trial accounts have low limits
mystic badger
Try upgrading to a pay-as-you-go account, free trial accounts have low limits
What is the limit exactly?
sour folio
What is the limit exactly?
However much free credit you have left. You have to attach a credit card to keep using
mystic badger
However much free credit you have left. You have to attach a credit card to keep...
I want to ask one more thing. I have account attached with organisation. Is that the reason for no free credits?
sour folio
I want to ask one more thing. I have account attached with organisation. Is that...
You either used it or it expired or you already made accounts with the same phone number
Free credit is associated with the org not the user
Please contact support
restive lily
This issue (429s) is unfortunately back this morning.
Would be great to get some eyes on it as/when you get a moment.
Latency seems unusually high, too.
night pine
I try to apply on their website but always error
Me too for one of my account. Maybe it's because of the region of your account.
soft mesa
Yes. I also use VPN but still not work. I can’t even open the official website of new bing now.
soft mesa
Me too for one of my account. Maybe it's because of the region of your account.
I just want to try to see how the bot having the emotion things. Recently, there’re many wired news about that
night pine
Yes. I also use VPN but still not work. I can’t even open the official website o...
VPN might be the major reason because Bing will think you might try to hide your location
I’ve tried to apply without VPN and succeed
But indeed many factors lead to this
soft mesa
I will try it later without VPN👀
north flame
I use this thread to follow-up on rate limit errors on davinci api, are you talking about the same issue ? 🕵️
restive lily
Likewise...
nimble cape
Just try resending the request to the api. I haven't encountered any rate limits since except for some errors. i think moxy fixed the issue already 🫡
restive lily
Yeah, it's been much more robust for a few days now. Kinda clinging on to this tenuous direct link to our best pal Moxi in case it returns 😅
rigid sedge
Guys, I'm using text-davinci-edit-001 API and I thought it was free to use as it's a beta version. But I'm continuously getting the RateLimitError. Any suggestions?
plucky fossil
Guys, I'm using text-davinci-edit-001 API and I thought it was free to use as it...
Whats that endpoint never heard of that model name
glacial arrow
Can someone confirm that fine tuning is working fine today? I have been trying since last 3 hours the fine tune status is pending
Thanks
rigid sedge
Whats that endpoint never heard of that model name
It’s a beta model specializing in editing tasks while davinvi-text-003 is more of a text generating model.
plucky fossil
It’s a beta model specializing in editing tasks while davinvi-text-003 is more o...
oh you trained ur own?
or you got into a beta program
restive lily
There's definitely some sort of incident ongoing
Constant 429s and 500s last two hours
fair verge
getting 429 and 500s for text-davinci-003
restive lily
Yep
dusky quail
@grave zephyr this might be happening again, I got a 429 (might've been a few, not sure) on gpt-3.5-turbo when not even close to reaching the actual API rate limit.
vast goblet
Happening for me too
night pine
<@1046280307462123561> this might be happening again, I got a 429 (might've been...
It seems that it’s a common issue now, for more help you can contact the support on the website. There isn’t no point @grave zephyr, it’s just a bot🤣
warm blaze
We're back again, except this time davinci-003 straight up times out like 50% of the time. It's unusable again. I pray Moxi sees this and they can fix it...
rustic dust
We're back again, except this time davinci-003 straight up times out like 50% of...
why are you still using davinci 3 bro
warm blaze
why are you still using davinci 3 bro
Because it's much more consistent than gpt3.5, which was designed to be a chatbot, not an instruct engine. I'm writing articles, and gpt3.5 is far more repetitive and doesn't work as well.
warm blaze
checking
Moxi please notice this and tell the team to fix the API timeout/server_error issues that are happening very frequently right now (I posted this here because I believe it's related to the problem from February).
I wouldn't be posting this unless is was a major issue. This isn't a timeout here or there. This is 5+ timeouts/errors in a row, even with 5-10 second sleep between them.
My customers are complaining because the OpenAI status page says there's no issue, when there clearly is.
I can assure you with 100% certainty this is not user-error or my code, because everyone in my circle of business acquaintances is experiencing the same thing.
I'm using gpt-3.5-trubo, but it's also happening on davinci-003.
When there is an error (sometimes it just flat times out with no error returned) this is it:
{
"error": {
"message": "The server had an error while processing your request. Sorry about that!",
"type": "server_error",
"param": null,
"code": null
}
}
Array
(
[error] => Array
(
[message] => The server had an error while processing your request. Sorry about that!
[type] => server_error
[param] =>
[code] =>
)
warm blaze
New error as well on gpt-3.5-turbo:
"That model is currently overloaded with other requests. You can retry your request, or contact us through our help center at help.openai.com if the error persists."
wispy pendant
We're getting a lot of these as well, definitely within Organisation limits.
RateLimitError: The server had an error while processing your request. Sorry about that! on text-davinci-003
dawn dock
I'm also experiencing this error
waxen oasis
Me too! Today in particular it is happening for us!
sweet steeple
Same here, i am the only person testing my app and I have regenerated my apiKey did all that I could but still the server is responding with error