No matter what I do, I can never select an AI Foundry embedding model in the Import and vectorize data in Azure AI Search. I wanted to try the SByte vector encoding via Cohere. I use East US for: resource group foundry project/hub ai search and all its related resources The documentation says I should deploy Cohere to a "serverless endpoint" option which isn't there at all: So I'm at a loss as to how to make this work.

Hello kzu , Greetings and Welcome to Microsoft Q&A! Thanks for posting the question. As I understand that you're having trouble with using Cohere embeddings in the Azure AI Search wizard, There might be a restriction on the types of embedding models supported in the vectorization process within the Azure AI Search wizard. Currently, the wizard supports only text-embedding-ada-002, text-embedding-3-large, and text-embedding-3-small models. Internally, the wizard utilizes the AzureOpenAIEmbedding skill to connect to Azure OpenAI, ensuring seamless integration with these specific embedding models. If you are attempting to use a different embedding model, it may not be supported within the wizard and may require a manual setup or alternative approach, the wizard calls the AML skill to connect to the catalog. Please refer this Set up embedding models , Supported embedding models . I hope you understand. And, if you have any further query do let us know. If this answers your query, do click Accept Answer and Yes for was this answer helpful. Thank you!

AI Foundry (Cohere) embeddings in AI search wizard not working

kzu 0

No matter what I do, I can never select an AI Foundry embedding model in the Import and vectorize data in Azure AI Search. I wanted to try the SByte vector encoding via Cohere.

I use East US for:

resource group
foundry project/hub
ai search and all its related resources

The documentation says I should deploy Cohere to a "serverless endpoint" option which isn't there at all:

So I'm at a loss as to how to make this work.

1 answer

SriLakshmi C 2,640 Reputation points Microsoft Vendor

2025-02-04T09:34:57.67+00:00

Hello kzu,

Greetings and Welcome to Microsoft Q&A! Thanks for posting the question.

As I understand that you're having trouble with using Cohere embeddings in the Azure AI Search wizard,

There might be a restriction on the types of embedding models supported in the vectorization process within the Azure AI Search wizard. Currently, the wizard supports only text-embedding-ada-002, text-embedding-3-large, and text-embedding-3-small models. Internally, the wizard utilizes the AzureOpenAIEmbedding skill to connect to Azure OpenAI, ensuring seamless integration with these specific embedding models. If you are attempting to use a different embedding model, it may not be supported within the wizard and may require a manual setup or alternative approach, the wizard calls the AML skill to connect to the catalog.

Please refer this Set up embedding models, Supported embedding models.

I hope you understand. And, if you have any further query do let us know.

If this answers your query, do click Accept Answer and Yes for was this answer helpful.

Thank you!
Please sign in to rate this answer.
SriLakshmi C 2,640 Reputation points Microsoft Vendor

2025-02-05T08:45:34.31+00:00

Hi kzu,

Just checking to see if you had a chance to review the below response.

Do let me know if that helps or have any other queries.

If the response helped, please do click Accept Answer and Yes for was this answer helpful. Thank you!

kzu 0 Reputation points

2025-02-05T22:03:36.17+00:00

I'm looking to reproduce what was shown by Pablo Castro back in MSBuild 2024 quite a few months ago. Relevant video (at precise time): https://youtu.be/T8pB_3vUkuE?t=1413

Saideep Anchuri 2,030 Reputation points Microsoft Vendor

2025-02-06T06:07:37.6266667+00:00

Hi kzu,

To implement vector search for text using Cohere's text embedding models, you have two main options:

Cohere-embed-v3-english: This model generates high-quality embeddings specifically for English text. It's highly effective for various Natural Language Processing (NLP) tasks, including vector search.

Cohere-embed-v3-multilingual: This model supports multiple languages, making it ideal for cross-lingual tasks. It creates embeddings that maintain consistency across different languages, which can be particularly useful in a multilingual environment.

For a solution utilizing these models, follow these steps:

Text Embedding:

Use Cohere-embed-v3-english for English texts.

Use Cohere-embed-v3-multilingual for texts in various languages.

Vector Search:

Generate embeddings for your text data using the appropriate Cohere model.

Store these embeddings in a vector database.

Perform vector searches by querying the vector database with new text embeddings to find similar texts.To implement vector search for text using Cohere's text embedding models, you have two main options:

Cohere-embed-v3-english: This model generates high-quality embeddings specifically for English text. It's highly effective for various Natural Language Processing (NLP) tasks, including vector search.

Cohere-embed-v3-multilingual: This model supports multiple languages, making it ideal for cross-lingual tasks. It creates embeddings that maintain consistency across different languages, which can be particularly useful in a multilingual environment.

For a solution utilizing these models, follow these steps:

Text Embedding:

Use Cohere-embed-v3-english for English texts.

Use Cohere-embed-v3-multilingual for texts in various languages.

Vector Search:

Generate embeddings for your text data using the appropriate Cohere model.

Store these embeddings in a vector database.

Perform vector searches by querying the vector database with new text embeddings to find similar texts.

Thank You.

Javier Díez 0 Reputation points

2025-02-07T09:23:35.9866667+00:00

Hi,

I am facing the same problem. When I am this point an with an embedding model created, in this case: text-embedding-3-small:

And I have the model created in muy Azure Foundry Project:

Thanks!

kzu 0 Reputation points

2025-02-08T03:56:19.33+00:00

@saideep anchuri you seem to have completely lost the point of the question. I'm not asking what model and how to use it and why one or the other. If you re-read the question and my screenshots, you can clearly see that the problem is that EITHER MODEL NEVER SHOWS UP in the wizard. I tried with all four (two text-embeddings-3 from OAI and the two Cohere), and the only ones that ever show up are the OAI ones.

Manas Mohanty 295 Reputation points Microsoft Vendor

2025-02-10T16:22:14.71+00:00

Hi kzu,

Sorry for the inconvenience.

Pre-requisite for models to show in drop down for public resource is

Regions for AI search and OpenAI need to be same region.

AI Search should have relevant permission (Azure AI Developer) for Azure AI foundry to read models

If it is a private resource.

You need to have OpenAI, AI search and Storage in same regions and connected through endpoint. DNS should resolve to private IPS through nslookup command.

OpenAI allowed as trusted resource from AI search firewall

Should be configured through managed virtual network

Permission of Azure AI Developer for AI search over Azure AI foundry.

I hope you understand.

Could you please take a moment to retake the survey on the above response? Your feedback is greatly appreciated.

Thank you.

Manas Mohanty 295 Reputation points Microsoft Vendor

2025-02-11T16:46:11.2+00:00

Hi kzu,

We are still looking for your inputs on any additional queries. Hope the pointer shared were useful.

Thank you.
Sign in to comment

Use comments to ask for clarification, additional information, or improvements to the question.

Share via

AI Foundry (Cohere) embeddings in AI search wizard not working

1 answer

Your answer