Photo by writer | Kanele
If you work in a data related field, you must always renew yourself. Data scientists use different tasks such as viewing data, data modulation, and goods storage programs.
Like this, AI changed data science from A to Him. Once you are on the way to search for data science, you may have heard the name Rag.
In this article, we will break down the rag. Starting with a presentation article and how it is now used to cut the cost when working with large languages of languages (LLMS). But first, let's cover the foundations.
What is the generation of retrievysval-ficted (RAG)?
Patrick Lewis began to be introduced for the RAG for the education text first by 2020. It includes two important things: the retriover and generator.
Concept after this is easy. Instead of extracting answers from parameters, the RAG can collect relevant information from the document.
What is the retrieved?
Retriver is used to collect relevant information from the document. But how?
Let us consider this. Has a large Excel sheet. Suppose it was 20 MB, and thousands of lines. You want to search for Call_Date for user_id = 10234.
As a result of this retriev, instead of monitoring the entire document, the RAG will only search the proper part.
But how does this help? If you search through the whole volume, you will use multiple tokens. As you may know, the use of APM's API is calculated using tokens.
Let's visit and see how this is calculated. For example, when you attach the presentation of this article. Cost 123 tokens.
You should look at this calculation cost using API Kallm. For example, if you think using a word document, name 10 MB, can be thousands of tokens. Every time you load this text using the LLM API, Features of Cost.
Using the Rag, you can choose the correct part of the document, reduces the number of tokens to pay less. It is straight.
How does this return this?
Before you start retrieval, the documents are separated by small chunks, paragraphs. Each chunk is converted into a vector that is running using the embedded model (to move ACNANDERIns, beta-bear, etc.).
So when the user wants to go about to ask what the phone day is, reinstatement compared to the Vector for question in all chunk's spears and select the same. Light, of course?
What is the generator?
As we define above, after regaining the most relevant documents, the generator takes over. Produces a response using a user's question and the text received.
By using this method, you also reduce the risk of understanding. Because instead of releasing the answer freely from the information AI was trained in, the model sets its answer in the original document you provided.
The appearing of Windows context
The first models, such as GPT-2 with smaller windows content, about 2048 tokens. That is why these models do not have features uploading the file. If you remember, after a few models, ChatGPT provides data loading feature because the context window appeared in that.
Advanced Models like GPT-4O have 128K token limit, supporting data loading and can show Rag Retand, in case the context window. But that is where applications reducing the costs enter.
So now, one of the reasons users use the rag to reduce costs, but not that. Because the cost of the use of the LLM is decreasing, GPT 4.1 present the content window up to 1 million tokens, delicious increase. Now, the Rag also appears.
A practice related to industry
Now, the llms appear to be agents. They have to use your jobs instead of just letting the answers. Some companies improve the control models and even your keywords and mouse.
In these cases, you should not take the opportunity to understand. So here the rag enters the scene. At this stage, we will deeply analyze one example in the real world.
Companies are looking for a talent to improve agents. Are not only large companies; Maximum dimensions or small companies and implementation requires their options. You can find these functions on Freelancer websites such as PWork and FIVER.
Invited commercial
Suppose a middle company from Europe wants you to create an agent, the agent issuing the proposals of their customers by using company documents.
In addition, this Angel should use the content by involving the relevant hotel details in this business event proposals or campaigns.
But there is a problem: The agent you usually tease. Why does this happen? Because instead of depending solely on the company document, the model is dragging information from its original training information. That training data can expire, because as you know, these llm are updated regularly.
Therefore, as a result, AI keeps adding the names of wrong hotels or just wrong details. You now point to the cause of the problem: lack of reliable information.
This is where the rag let us look at this immediately.
“Produce a suggestion, according to voice and company data, and use Web Search for hotels.
This web search feature turns the rag route.
The last thoughts
In this article, we received the evolution of AI models and why rag has been using it. As you can see, the reason has changed over time, but the problem remains: efficiency.
No matter the reason for cost or speed, this approach will continue to be used in AI related activities. And by “Ai-related,” I don't include data science, because, as you know it, in the coming Summer of Ai, the data science is already affected by AI.
If you want to follow the same articles, solve 700+ dialects related to data science, and 50 data projects, visit my platform.
Nate Rosid He is a data scientist and product plan. He is a person who is an educated educator, and the Founder of Stratascratch, a stage that helps data scientists prepare their conversations with the highest discussion of the chat. Nate writes the latest stylies in the work market, offers chat advice, sharing data science projects, and covered everything SQL.
