Vector Search external embedding model (OpenAI) example

This notebook shows how to use the Vector Search Python SDK, which provides a VectorSearchClient as a primary API for working with Vector Search.

This notebook uses Databricks support of external models (AWS|Azure) to access an OpenAI embeddings model to generate embeddings.

Load toy dataset into source Delta table

The following creates the source Delta table.

Chunk sample dataset

Chunking the sample dataset helps you avoid exceeding the context limit of the embedding model. The OpenAI model supports up to 8192 tokens. However, Databricks recommends that you split the data into smaller context chunks so that you can feed a wider variety of examples into the reasoning model for your RAG application.

Create vector search endpoint

Register OpenAI embedding model endpoint

For detailed usage information, see the external model documentation for configuring an OpenAI endpoint (AWS|Azure).

To provide credentials, use the Databricks secret manager (AWS|Azure).

Create vector index

Similarity search

The following cells show how to query the Vector Index to find similar documents.

Delete vector index