Reward Models and Best-of-N - CMU LLM Inference 12

Learn about reward models and best-of-n sampling techniques in this 53-minute lecture from CMU's Advanced NLP course, delivered by Amanda Bertsch as part of the LLM Inference series. Explore the theoretical foundations and practical applications of reward models, understand how best-of-n sampling works in practice, and discover how Monte Carlo Tree Search can be applied to language model inference. Gain insights into advanced techniques for improving language model outputs through sophisticated sampling and search strategies used in modern NLP systems.