Announcing Amazon Titan Text V2 now available in Amazon Bedrock, optimized for improving RAG

The Amazon Titan family, available in Amazon Bedrock, offers powerful pre-trained models for AI and ML tasks. The latest addition, Amazon Titan Text Embeddings V2, is optimized for Retrieval-Augmented Generation (RAG) and supports vector sizes of 256, 512, or 1024 dimensions. By choosing smaller vectors, users can reduce storage costs and latency while maintaining accuracy.

These embeddings improve RAG accuracy by summarizing text effectively. For instance, they help a large language model (LLM) fetch up-to-date information from custom sources like knowledge bases. Amazon Titan Text Embeddings V2 maintains high accuracy even at smaller dimensions, with 512-dimensional vectors retaining 99% of the accuracy of 1024-dimensional ones.

Users can access Amazon Titan Text Embeddings V2 through Knowledge Bases for Amazon Bedrock or directly via the Bedrock Runtime API. The API request includes parameters like input text, normalization flag, and vector dimensions. Users can invoke the model from their code, as demonstrated in Swift. Existing knowledge bases created with the original Amazon Titan Text Embeddings model will continue to function seamlessly.

Source :- https://aws.amazon.com/it/blogs/aws/amazon-titan-text-v2-now-available-in-amazon-bedrock-optimized-for-improving-rag/