Stanford CME295 Transformers & LLMs | Autumn 2025 | Lecture 3 - Tranformers & Large Language Models

For more information about Stanford’s graduate programs, visit: https://online.stanford.edu/graduate-education October 10, 2025 This lecture covers: • Definition and architecture • Mixture of experts • Context length, temperature • Sampling strategies • Prompting, in-context learning • Chain of thought • Self-consistency To follow along with the course schedule and syllabus, visit: https://cme295.stanford.edu/syllabus/ Chapters: 00:00:00 Introduction 00:01:02 Recap of Transformers-based models 00:03:43 LLM definition 00:07:37 Mixture of Experts 00:12:43 Dense & Sparse MoE 00:15:47 MoE in LLMs 00:36:35 Response generation 00:38:34 Greedy decoding & beam search 00:46:36 Sampling-based methods 00:52:24 Impact of temperature on predictions 01:04:53 Guided decoding 01:07:07 Prompting strategies 01:14:09 In-context learning 01:18:35 Chain-of-thought, self-consistency 01:25:00 Inference optimizations with KV cache 01:33:09 PagedAttention, MLA Afshine Amidi is an Adjunct Lecturer at Stanford University. Shervine Amidi is an Adjunct Lecturer at Stanford University. View the course playlist: https://www.youtube.com/playlist?list=PLoROMvodv4rOCXd21gf0CF4xr35yINeOy