Amazon Titan Text Embeddings V2 Now Available in Amazon Bedrock

Amazon Titan Text Embeddings V2 is the second-generation text embeddings model from Amazon, designed for Retrieval-Augmented Generation (RAG). It is pre-trained on over 100 languages and on code, and allows users to choose the size of the output vector to balance detail against computational time and storage costs. It also features an improved unit vector normalization to enhance accuracy when measuring vector similarity.

In Amazon Titan Text Embeddings V2, you can choose the size of the output vector from three options: 256, 512, or 1024 dimensions. The implications of this choice are:

• Larger vector sizes (1024) create more detailed responses but increase computational time.
• Shorter vector lengths (256) are less detailed but improve response time and reduce storage costs.
• Vectors with 512 dimensions retain approximately 99% of the accuracy of the 1024 dimension vectors, while 256 dimensions keep about 97% of the accuracy.

Amazon Titan Text Embeddings V2 improves the accuracy of Retrieval-Augmented Generation (RAG) by providing high-quality embeddings or vectors that act as condensed summaries of text. These vectors help RAG to fetch the most relevant and up-to-date information from a custom source, which in turn allows a large language model (LLM) to generate more accurate and relevant answers.

To use Amazon Titan Text Embeddings V2 in your code, you can invoke the model through the Amazon Bedrock Runtime API. Here's a simple example using the Swift programming language:

import Foundation
import AWSBedrockRuntime 

let text = "This is the text to transform in a vector"

// create an API client
let client = try BedrockRuntimeClient(region: "us-east-1")

// create the request 
let request = InvokeModelInput(
   accept: "application/json",
   body: """
   {
      "inputText": "\(text)",
      "dimensions": 256,
      "normalize": true
   }
   """.data(using: .utf8), 
   contentType: "application/json",
   modelId: "amazon.titan-embed-text-v2:0")

// send the request 
let response = try await client.invokeModel(input: request)

// decode the response
let response = String(data: (response.body!), encoding: .utf8)

print(response ?? "")

This code snippet shows how to set up the client, create a request with the desired text, vector size, and normalization option, and then send the request to get the embeddings.

Amazon Titan Text Embeddings V2 will soon become the default large language model (LLM) proposed by Knowledge Bases for Amazon Bedrock. Existing knowledge bases created with the original Amazon Titan Text Embeddings model will continue to operate without any required changes.