Snowflake BUILD | Retrieval Augmented Generation In The Real World скачать в хорошем качестве

Snowflake BUILD | Retrieval Augmented Generation In The Real World

Retrieval augmented generation (RAG) enhances an LLM with external data that is private, recent, or otherwise excluded from training. While prototyping RAG is simple, especially with the LlamaIndex framework, creating robust pipelines is challenging. RAG performance tends to worsen when the data source is large or contains disparate information, like a FAQ with questions and answers scattered throughout. In this session, we'll explore two methods to boost RAG performance: 1. Use more sophisticated retrieval techniques such as “big to small”, sub-question generation, re-ranking, and hybrid search 2. Finetune the embedding model with relevant data We'll showcase these enhancements using a Streamlit Chatbot, allowing users to compare RAG-enhanced responses with the standard model’s responses. By the end of this session, you'll have the skills to develop production-ready RAG applications with LlamaIndex. Code for demo app: https://github.com/carolinedlu/llamai... Subscribe for more! http://www.snowflake.com/YTsubscribe/ Explore sample code, download tools, and connect with peers: https://developers.snowflake.com/