Large Language Models (LLMs) - Stanford CME295 Transformers & LLMs | Autumn 2025 | Lecture 2 - Transformer-Based Models & Tricks

For more information about Stanford’s graduate programs, visit: https://online.stanford.edu/graduate-education October 3, 2025 This lecture covers: • Attention approximation • MHA, MQA, GQA • Position embeddings (regular, learned) • RoPE and applications • Transformer-based architectures • BERT and its derivatives To follow along with the course schedule and syllabus, visit: https://cme295.stanford.edu/syllabus/ Chapters: 00:00:00 Introduction 00:01:30 Recap of Transformers 00:10:37 Overview of position embeddings 00:15:36 Sinusoidal embeddings 00:25:56 T5 bias, ALiBi 00:31:02 RoPE 00:43:42 Layer normalization 00:50:39 Sparse attention 00:55:38 Sharing attention heads 01:02:42 Transformer-based models 01:11:38 BERT deep dive 01:33:24 BERT finetuning 01:43:30 Extensions of BERT 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