Before you start using Vector Search, you need to choose an embeddingmodel, prepare your data, and decide what type of endpoint you'll use. This pageprovides some information about doing those things.

Prepare your embeddings

To use Vector Search, you need to have your embeddings ready.If you already have your embeddings, skip toChoose anendpoint.

To create your embeddings, do the following:

1. Choose an embedding model: There are many external embeddingmodels available, which offer different features.

   Vector Search supports dense embeddings, sparse embeddings,and hybrid search. Hybrid search uses dense andsparse embeddings according to the weight that you specify for thoseembedding types.

   Depending on your use case, choose one of the following type of model:

   • Ready-to-use:If you want to semantically match text to text or text to images by therelevance of the text or image alone. This is a standard use case, so youdon't need to train or tune the model.Generative AI on Vertex AIis a recommended option for this use case. Generative AI on Vertex AI usesdense embedding models.

   • Custom model for embeddings: If you want to match based onyour own data or specific use case.

2. Prepare your data: Clean and preprocess your data toensure that it's in a format that can be used by the embedding model.

3. Train the embedding model if you use a custom model: If you choose touse a custom embeddings model (tuning), you need to train it on your data.This can be a time-consuming process that depends on the size and complexityof your data. If you use a pretrained model from theModel Garden,then you can skip this step.

4. Generate embeddings: After the model is trained, use it to generateembeddings for your data.

Choose an endpoint

After you have created your index, you'll deploy it to an endpoint. Formore information, seeDeploy and manage public indexendpoints andDeploy and manageindex endpoints in a VPC network. It'shelpful to decide what kind of endpoint you'll need before youcreate your index.

You can deploy your query index to one of the following:

• Public endpoint: If you deploy to a public endpoint, you don't need toset up your network. Public networks have slightly higher latency, but arefaster to set up and easier to maintain.

• Private Endpoint: If you want to use a VPC, you must firstset up networking. Vector Search supports two types of privatenetwork.

  • VPC Network Peering connectionfor reduced network latency.

  • Private serviceconnectfor private consumption of services across VPCnetworks that belong to different groups, teams, projects, ororganizations.

What's next

After you've generated your embeddings and decided where to deploy yourindex, the next step is to configure your index.

• Learn how to configureinput data format and structure
• Learn how to create a Vector Search index usingnotebook tutorials
• Learn how tomanage indexes