Advanced RAG: Enhancing Model Efficiency and Accuracy

While naive RAG provides a foundation for using LLMs with document retrieval, it has limitations. For instance, the retrieval accuracy might not be perfect, potentially leading to irrelevant information being used as context. Additionally, the LLM might struggle to integrate retrieved information seamlessly, resulting in disjointed responses. In this lesson, we'll look at advanced RAG and explore its workflow.

Improving retrieval process

Advanced RAG builds on naive RAG by introducing improvements in retrieval quality. It adds pre-retrieval and post-retrieval strategies before and after the retrieval process. Let’s look at the figure below:

Pre-retrieval process

Advanced RAG leverages a two-pronged approach during the pre-retrieval process to enhance retrieval accuracy before the actual retrieval happens at inference time. Here’s a breakdown of the key techniques employed:

• Optimizing indexing

• Optimizing query

Optimizing indexing

The following are some important optimizing indexing techniques:

• Enhancing data granularity (chunking strategies): Before delving into retrieval intricacies, grasp the significance of text processing strategies like chunking, which play a pivotal role in breaking down large ...