Stanford CS336 Spring 2026: Learn to Build an LLM from Scratch with Public Assignments
stanfordnlp · x · 2026-10-03
Stanford's CS336 (Language Modeling from Scratch), taught by Percy Liang and Tatsunori Hashimoto, has published its Spring 2026 syllabus with publicly available materials for self-study.
- Coverage: the full pipeline from tokenizer to training and evaluation, including transformer basics, GPU optimization, scaling laws, data cleaning, and reasoning RL.
- Approach: OS-course-style, students build their own LM end to end — data collection/cleaning, model construction, training, evaluation — with implementation-heavy assignments.
- Prerequisites: strong Python and PyTorch plus ML/math foundations; GPU compute is separate, and the course recommends debugging correctness on CPU before paid training runs.
- Lecture recordings are on a YouTube playlist; assignment links are public.
More from Companies & People
- OpenRouter co-founder on Stripe deal: payments and inference will blend together — a16z · 2026-10-03
- Agentic AI Course on LinkedIn Learning Surpasses 300,000 Learners — DavidLinthicum · 2026-10-03
- Stanford's free 2-hour course explains how modern LLMs are built from scratch — CurieuxExplorer · 2026-10-03
- OpenAI safety researcher David Robinson quits, says 'nuclear level' safeguards are needed — Always_Curious_One2 · 2026-10-03
- NYC Agentic AI meetup to recap September AI moves with live PowerShell decision demo — dfinke · 2026-10-03
- Splice CEO Kakul Srivastava thinks AI emails are killing conversations — The Verge AI · 2026-10-03