Using AI Search vector index in chat playground fails

Michael Johansen 0

I have deployed a GPT4-32k model in ai.azure.com.
As the AI Search Service does not yet have a web site import function I have made an Azure Table with id, title, content and url and pasted static web site data from our company website into data rows.
I then used Import and vectorize to create a vector index.

When I try to use the index in Chat Playground I get this error:

"An error occurred when calling Azure Cognitive Search: Azure Search Error: 400, message='Server responded with status 400. Error message: {"error":{"code":"InvalidRequestParameter","message":"There's a mismatch in vector dimensions. The vector field 'text_vector', with dimension of '1024', expects a length of '1024'. However, the provided vector has a length of '1536'. Please ensure that the vector length matches the expected length of the vector field."

I have read the the documentation and tried to re-create the index but I cannot see how to change the dimensions so it fits.

Laxman Reddy Revuri 1,755 Reputation points Microsoft Vendor

2025-01-09T10:51:07.67+00:00

Hi @Michael Johansen
Thanks for the question and using MS Q&A platform.
The issue seems to arise due to a mismatch in the vector dimensions between the text_vector field in your index and the vector being provided. The text_vector field is currently defined with a dimension of 1024, whereas the vector you are using has a dimension of 1536.
1.Check the Vector Field Configuration: During the creation or the updating of the vector index, it is critical that the vector field ‘text_vector’ is updated to a vector field dimension of ‘1536’, so that it is consistent with the vectors that are requested.

2.Create the index again: To change the dimension of a pre-existing vector field, the index needs to be created again with the updated dimensions, this requires the addition of a schema to the index which provides the new dimension of the vector.

3.Use the correct model: It is imperative that the model that is employed for vector generation is in line with the dimensions that have been included in the index schema.

References:
Create a vector index
Configure a vectorizer in a search index
Vector index size and staying under limits
Michael Johansen 0 Reputation points

2025-01-09T16:05:38.74+00:00
Thanks for your reply.

I have re-rerun the indexer and when I look at the index json it says:

{ "name": "text_vector", "type": "Collection(Edm.Single)", "key": false, "retrievable": true, "stored": true, "searchable": true, "filterable": false, "sortable": false, "facetable": false, "synonymMaps": [], "dimensions": 1024,

But when I go to the Chat Playground and re-create the data connection and start over it still ends up with the same error:
There's a mismatch in vector dimensions. The vector field 'text_vector', with dimension of '1024', expects a length of '1024'. However, the provided vector has a length of '1536'.

This doesn't really make sense...

"with a dimension of 1024 expects 1024".

I cannot find out where these 1536 is configured or gets into the system.

Best regards,

Michael Johansen
Michael Johansen 0 Reputation points

2025-01-13T20:11:02.6933333+00:00

Thanks for your reply. I have now created a new project and deployed a gpt-4 (not 32k) and created a new vector index. But I get the same error. I will contact support.

1 answer

Laxman Reddy Revuri 1,755 Reputation points Microsoft Vendor

2025-01-13T11:19:21.14+00:00

Hi @Michael Johansen
I Apologize for the continuous issue that you have regarding the vector dimensions of your index. ensure that the Chat Playground is using the latest version of the index schema. It is possible that the issue is with the GPT4-32k model not generating vectors of the correct length. I would recommend reaching out to Azure support
Please sign in to rate this answer.

1 person found this answer helpful.

0 comments No comments
Sign in to comment

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

Share via

Using AI Search vector index in chat playground fails

1 answer

Your answer