From Decoding to Meta Generation: Inference Time Algorithms for Language Models - Lecture 22

Learn about inference-time algorithms for Language Models through a comprehensive guest lecture delivered by Sean Welleck as part of Carnegie Mellon University's Advanced Natural Language Processing course. Explore various decoding methods and meta-generation techniques that are crucial for optimizing language model performance during inference. Delve into cutting-edge approaches that enhance the capabilities of language models, making them more efficient and effective for real-world applications. Gain valuable insights into the theoretical foundations and practical implementations of these algorithms, essential knowledge for anyone working with modern NLP systems and large language models.