Using Evaluations to Optimize a RAG Pipeline: from Chunkings and Embeddings to LLMs | by Christy Bergman | Jul, 2024 – Niraranra

Retrieval Augmented Know-how (RAG) is a useful strategy for using your private data in an AI-powered Chatbot. On this weblog submit, I’ll stroll by way of three key strategies to get basically probably the most out of RAG and think about each method to go looking out top-of-the-line combos.

For readers who merely have to know the TL;DR conclusion: basically probably the most RAG accuracy enchancment acquired right here from exploring completely completely different chunking strategies.

• 89% Enchancment by altering the Chunking Approach 📦
• 20% Enchancment by altering the Embedding Model 🤖
• 6% Enchancment by altering the LLM Model 🧪

Let’s dive into each method and uncover the best-performers for a real-world RAG utility using RAG ingredient evaluations! 🚀📚

I’ll use Milvus documentation public web pages as a result of the docs data and Ragas as a result of the evaluation methodology. See my earlier blog about how to use RAGAS. The rest of this weblog is organized as follows:

1. Textual content material Chunking Strategies
2. Embedding Fashions
3. LLM (Generative) Fashions