English
Share
Sign In
Combination of RAG, Embedding Model, and VectorDB
Naver Webtoon, Yang Young-soon <Denma>
This is a combination that makes me want to say that it is a combination of three technologies. If you have read this guide without missing it, you will remember about Retrieval -Augmented Generation (RAG). Then, you may be wondering naturally. Well, RAG also refers to existing information to provide richer answers? Then, what are the roles of Embedding Model and VectorDB? I am sorry if you are not curious. But I was curious.
Embedding models are used to convert words or sentences into vectors, which are arrays of numbers that computers can understand. This is similar to teaching a computer a language, and the computer can understand the meaning of sentences or words through these vectors.
VectorDB is a database that stores and manages these converted vectors. In other words, it can be said to be like a 'vocabulary' of the language learned by the computer. Many words and sentences are organized in the form of vectors, so they can be quickly looked up when needed.
RAG) is an artificial intelligence model that looks up previously stored information (such as information stored in VectorDB) and generates an answer based on this when answering a question. It is similar to a student recalling what he studied to solve a test question. RAG references existing information to generate more accurate and useful answers.
In other words, VectorDB can be a repository of information that the RAG model can reference. Of course, other databases are the same. RAG searches for vector information stored in VectorDB to generate answers to specific questions, and uses that information to create new answers. In this process, RAG 'understands' the information in VectorDB and constructs its own answers based on it.
In this way, RAG uses existing knowledge to provide richer and more accurate information. However, the reason RAG is attracting attention is that it is innovative in that artificial intelligence does not simply provide predetermined answers, but creates new answers based on stored knowledge.
ⓒ 2023. Haebom, all rights reserved.
It may be used for commercial purposes with permission from the copyright holder, provided the source is cited.
The images used are for the purpose of helping understanding, and all rights to the images belong to the original authors indicated in the source.