CODEWITHSUNDEEP
azureopenai-apiazure-openai
user19771002
user19771002

Reputation: 31

How can I perform an Azure OpenAI Chat Completion on Private Data with Vector Search

I am trying to build a chat experience on private data in Azure OpenAI. Currently in OpenAI Studio I can deploy a web app against private data housed in Azure Cognitive Search. I'd like that Search to be a vector embedding search, but cannot see any reasonable way to achieve that out of the box. The DataSources property on the api request doesn't seem to accept anything that would perform a vector search.

I can build this manually by doing the embeddings query manually, then doing the search query manually, and finally prompt engineering the Chat request. However, this seems like it should be something out of the box. This is all pretty new, and documentation is sparse on the Azure OpenAI stuff so im trying to decide if this is a feature gap or a documentation gap.

Upvotes: 0

Views: 1785

Answers (2)

matt1616
matt1616

Reputation: 502

steps to enable the vector/semantic search:

  • turn on the semantic ranker in the cognitive search component (left menu)
  • in the openai component, when adding your data, use the text-embedding-ada-002 model, then add vector search. select the embedding model (ada), and then the search type (should have a semantic option)

model info:

Upvotes: 0

user19771002
user19771002

Reputation: 31

For those curious about this, just yesterday (the day before the question was posted) a PR was accepted into the sample app repository which demonstrates the Vector Search capability.

Upvotes: 0

Related Questions