Hi I'm starting to use Azure OpenAI embeddings text-embedding-ada-002 model, but it seems to have a limit of ***2048 ***tokens, while OpenAI says it should be 8192 . Thanks! Simon

Thanks for your additional information, I got you now @Simon Breslav So you are mentioning the limit for Azure open AI and Open AI are not the same. For Azure Open AI, the limit of text-embedding-ada-002 is 2048 but for Open AI is 8191. I think this is the current product limitation and I am sorry for any inconvenience at this moment. Azure Open AI Service is releasing step by step and it is not fully releasing as for now. I find some supportive information shows currently in Azure documentation as below - The Azure Open AI is not releasing the same tokens limit as Open AI at this moment. I have forwarded this feedback to product manager to confirm any plan to release the same limit as Open AI, and I will let you know once I have any response. I hope this helps. Regards, Yutong -Please kindly accept the answer if you feel helpful to support the community, thanks a lot.

text-embedding-ada-002 token context length

Accepted answer

YutongTie-MSFT 46,986 Reputation points

2023-03-09T22:11:30.1833333+00:00

Thanks for your additional information, I got you now @Simon Breslav

So you are mentioning the limit for Azure open AI and Open AI are not the same. For Azure Open AI, the limit of text-embedding-ada-002 is 2048 but for Open AI is 8191.

I think this is the current product limitation and I am sorry for any inconvenience at this moment. Azure Open AI Service is releasing step by step and it is not fully releasing as for now.

I find some supportive information shows currently in Azure documentation as below -

The Azure Open AI is not releasing the same tokens limit as Open AI at this moment. I have forwarded this feedback to product manager to confirm any plan to release the same limit as Open AI, and I will let you know once I have any response.

I hope this helps.

Regards,

Yutong

-Please kindly accept the answer if you feel helpful to support the community, thanks a lot.
Please sign in to rate this answer.

3 people found this answer helpful.
Simon Breslav 50 Reputation points

2023-03-09T22:14:52.5733333+00:00

@YutongTie-MSFT Got it, thank you for the information.

Simon Breslav 50 Reputation points

2023-03-11T17:22:57.2233333+00:00

@YutongTie-MSFT just to clarify, just want to point out that Azure docs do say it should be 8k... so I assume that's something product teams are working on.... although I'm not sure why the model id says text-ada-embeddings-002 instead of text-embeddings-ada-002

https://learn.microsoft.com/en-ca/azure/cognitive-services/openai/concepts/models#embeddings-models-1

YutongTie-MSFT 46,986 Reputation points

2023-03-11T19:54:12.7466667+00:00

Thanks @Simon Breslav I got this information from another post. I have seen some inconsistency from different docs. Yesterday, I already sent an email to the product manager and I am waiting for the double confirmation. I will let you know the plan and the official setting when I get the information. Thanks for your understanding again.

Simon Breslav 50 Reputation points

2023-03-16T13:02:24.2433333+00:00

Hi Yutong @YutongTie-MSFT I saw that the docs have been updated (https://learn.microsoft.com/en-ca/azure/cognitive-services/openai/concepts/models#embeddings-models-1) to reflect 2,046 token limit for text-embeddings-ada-002 model, do you know if there are any plans to increase that to 8k limit that OpenAI version has? Or Azure will keep 2K limit for foreseeable future?

YutongTie-MSFT 46,986 Reputation points

2023-03-16T15:13:33.41+00:00

I have check with product team and will let you know soon

pj 0 Reputation points

2023-04-08T18:41:48.6166667+00:00

Hi @YutongTie-MSFT any update on this model? It looks like the documentation has been updated again:

Simon Breslav 50 Reputation points

2023-04-08T20:41:19.6466667+00:00

Yep, when you create a new text-embedding-ada-002 model deployment there is now an additional version dropdown....

YutongTie-MSFT 46,986 Reputation points

2023-04-08T21:48:28.6966667+00:00

Hello @Simon Breslav Sorry I missed your message, thanks for your feedback for the limit, product team has released a version 2 model which provide latest weights/updated token limit @pj As Simon said, there is a new version we highly recommend you to try, you will see the dropdown of version as below screenshot -

Please let me know if you have any questions, feel free to open a new thread. Thanks for your understanding.

pj 0 Reputation points

2023-04-12T17:35:29.4966667+00:00

Thanks @YutongTie-MSFT , one of the software engineers on our team applied to gain access to the APIs about a week ago, but he says there is a hold-up with address verification taking a long time. Any idea how we could speed that up?

YutongTie-MSFT 46,986 Reputation points

2023-04-12T17:43:54.8933333+00:00

Hello @pj The normal waiting time for a response is around two weeks from the access team due to the capacity limit. Currently we have no way to speed it up if no special situation happens, sorry about that.
Sign in to comment

text-embedding-ada-002 token context length

0 additional answers