RAG WARS - Advancing AI: Enhancing LLMs and RAG for Improved Performance & Reliability
🚀 Join Us for a Dynamic Event (Food will be served 6-7 pm)!
🔍 Topic: "Advancing AI: Enhancing LLMs and RAG for Improved Performance & Reliability"
📅 June 19th, Time: 6:00 PM PST/ 9:00 PM CET
This meetup explores advanced techniques to enhance the utility and reliability of Large Language Models (LLMs) across diverse applications. From structured outputs and external function integration to robust enterprise data architecture and strategies for reducing hallucinations, the talks cover a spectrum of methods to optimize both the performance and accuracy of LLM and RAG-based systems in real-world settings.
Talks are listed in order of presentation:
[1] Structured Output and Function Calling for Large Language Models
Suleman Kazi, ML @ Vectara
Ever wanted your LLM to produce output in a particular format (JSON, CSV, XML….) so you can easily parse it out or use it in a downstream task? How about giving it access to external functions that perform a task or return information that the LLM does not have access to? In this talk, you’ll learn about doing both of these tasks, known as structured output and function calling, respectively. We’ll talk about how they are useful and how you can enable their use with open-source LLMs on HuggingFace.
[2] Enterprise data architecture in machine learning and RAG systems
Nikhil Bysani, Engineering @ Vectara
Best practices in storing and consuming data to be used in ML systems, like a data lake/warehouse, s3, event driven systems
Talk about data lifecycle and best ingestion practices with vectara
Talk about managing state such that and synchronization of data between Vectara and other data systems
[3] Strategies for Mitigating Hallucination in Large Language Models
Rogger Luo, ML @ Vectara
Hallucination poses a significant challenge to the usability and reliability of LLM applications. In this presentation, we offer an insightful overview of contemporary methods aimed at mitigating hallucination in summarization, drawing from our own practical experiences with these techniques. Our examination reveals that these methods can be broadly categorized into three main approaches: Alignment with Fine-tuning (DPO), Control at Inference (DoLA), and Post-Editing(FAVA).
The whole conversation will be moderated by:
- Ofer Mendelevitch, Head of Developer Relations at Vectara
This event is open for everyone to join so save the date and meet us 6 pm PST on June 19th. Let's explore the cutting-edge of RAG together while networking and enjoying food and drinks! 🚀