Claude Haiku API: 429, 'message': 'Quota exceeded for aiplatform.googleapis.com/

[mimiwu]

I want to get access with Claude Haiku's API and got this error:

C:\Users\Administrator\tool\02_bard\vertex ai>python haiku02.py
无法生成文本: Error code: 429 - {'error': {'code': 429, 'message': 'Quota exceeded for aiplatform.googleapis.com/online_prediction_output_tokens_per_minute_per_base_model with base model: anthropic-claude-3-haiku. Please submit a quota increase request. https://cloud.google.com/vertex-ai/docs/generative-ai/quotas-genai.', 'status': 'RESOURCE_EXHAUSTED'}}
生成文本失败

C:\Users\Administrator\tool\02_bard\vertex ai>
It is just my first try and impossible to exceed any limit. Why?

[spaceship_and_d]

Same problem: 

"error": {
"code": 429,
"message": "Quota exceeded for aiplatform.googleapis.com/online_prediction_output_tokens_per_minute_per_base_model with base model: anthropic-claude-3-sonnet. Please submit a quota increase request. https://cloud.google.com/vertex-ai/docs/generative-ai/quotas-genai.",
"status": "RESOURCE_EXHAUSTED"
}

[Mikkel]

Same problem here

[JadielMiles]

Quotas are set to ensure fair use of resources and prevent misuse. Your API request has exceeded your allocated quota, resulting in a 429 "Quota Exceeded" error.
You need to request a quota increase to use Claude Haiku's API. Google usually gives you the option to increase your quota according to your needs.

[mimiwu]

I think this is impossible since this is the first time I get access to it and made only one API call following the official instructions.

[mary99]

I have the same problem too - but for CPUs. It's the first pipeline i'm trying to kick off having just started a google cloud account (free $300 in credits). I use all the default values for a regression analysis in AutoML. Get exceeded limits on the CPUs. When I drop everything down to using just 1 cpu (which appears on my quotas page) - same error. 

[mathai70]

anthropic.RateLimitError: Error code: 429 - {'error': {'code': 429, 'message': 'Quota exceeded for aiplatform.googleapis.com/online_prediction_output_tokens_per_minute_per_base_model with base model: anthropic-claude-3-haiku. Please submit a quota increase request. https://cloud.google.com/vertex-ai/docs/generative-ai/quotas-genai.', 'status': 'RESOURCE_EXHAUSTED'}}
Im getting the same error. I've yet to complete a single request.

[mary99]

Hi @mathai70 I am in exactly the same situation. I spent many hours trying to figure out what I am doing wrong. I suspect it may be because I am on a free trial (albeit with "$300" credit). Even so, I am unsure why Google has so heavily marketed the free trial. One would assume they would want to ensure the best experience possible. As much as I would like to continue with Vertex AI, am moving over to AWS. 

[mathai70]

I have a paid GCP account and it still does not work. I moved from the claude api on their website because it was needed to buy credits beforehand which is terrible when trying to scale an application and none of these large companies ever response to requests.

I thought this would be a great work around but i guess not.

[spaceship_and_d]

I know. I've been trying to get this to work for the past one week. Might move to oai on azure. 

[Poala_Tenorio]

The error message you received indicates that the quota for generating text using Claude Haiku's API has been exceeded. This quota limit is associated with the specific base model "anthropic-claude-3-haiku" and the limit for online prediction output tokens per minute.

Even though it's your first try, it's possible that the quota for this API has already been reached by other users or processes. Quota limits are set by the API provider (in this case, Google Cloud's Vertex AI) to manage resource usage and ensure fair access to services.

Rate limits of Claude API are currently measured in requests per minute, tokens per minute, and tokens per day. For a free tier for comparison, requests per minute (RPM) is limited to 5, tokens per minute (TPM) is to 25,000, and tokens per day (TPD) is to 300,000. You can browse for more information through this documentation.

If you believe that you haven't actually exceeded the quota or if you think there might be an error, you can also reach out to the API provider's support for assistance. They can help investigate the issue further and provide guidance on how to proceed.

[mimiwu]

I can't even find the haiku model you mentioned here to set quota: This quota limit is associated with the specific base model "anthropic-claude-3-haiku" and the limit for online prediction output tokens per minute.

Could you pls show how I can set the quota for Haiku? Thanks!

[mathai70]

I've used the Haiku API on Anthropic and it works fine, so not sure why you would suggest that we reach out to them. You should raise this issue with the Vertex AI Team and give us back a response on how to navigate through this issue. Why would my quota limit be impacted by others? That means requesting a quota increase would not actually solve the problem. None of us have been able to complete a single request, I would say this needs to be escalated. 

[Mikkel]

So I share limits with other GCP users? If I then request a limit increase even though I havent had a single request made yet I risk other GCP users have used up my limit before I manage to make my own requests?

[Mikkel]

Hi Poala I can see some people succeeded in requesting a qouta increase, where is do you do that? it seems like I am not able to click anything, you offered us a bunch of credits to build out part of our genai here, am I missing something, should I write our a account manager for a increase?

[spaceship_and_d]

I think they are saying we gotta reach out to Anthropic to solve it ourselves. Writing the email now...

[spaceship_and_d]

 In the image below you can see how all claude models are set to '0' for quotas. 

I am now trying to change it, but they don't allow values above zero: 

And here's GCP's link i followed: https://cloud.google.com/docs/quotas/view-manage

 

[mimiwu]

Indeed. We are asked to send request to sales team to increase quota, but can't get access to sales. Can't believe this if Google is a startup.

[mathai70]

Its a problem with the Vertex AI team, a quota increase would not make a difference if you can't even send one simple short request. I've never used this forum, does anyone useful from Google ever reply?

[mathai70]

i was able to make a increase in quota request. I am on a paid business account though. 

[Zorik16]

Exceeding quota is a common problem when using cloud APIs, as in your case with Google Cloud. This means that you limit the use of withdrawal tokens for online forecasting. The solution includes contacting the provider or checking documentation to increase the quota.

[spaceship_and_d]

Anyone solved this issue yet?

[MichaelQuinlan]

Also getting the same error, in the api/quota explorer 'online_prediction_output_tokens_per_minute_per_base_model' isn't an option so you can't find the current quota nor ask for an increase.

On Anthropic directly they have 4 tiers I believe GCP/Vertex default is near Tier 1 which is not sufficient for us.