Stanford CS336 posts 18 free lectures building LLMs from scratch, Triton to RLHF
techNmak · x · 2026-10-09
Stanford's CS336: Language Modeling from Scratch (Spring 2026, 18 lectures) is free on YouTube, covering the full LLM stack: tokenization, Transformers, MoE, FLOPs/GPUs, Triton kernels and distributed training, scaling laws and data curation, inference/evals, and SFT/RLHF/RLVR plus multimodality. Students implement tokenizers, optimizers, FlashAttention2 in Triton, build memory-efficient distributed training, process raw Common Crawl into pretraining data, fit scaling laws, and train a math-reasoning model. Stanford says students write an order of magnitude more code than in most AI classes.
More from Companies & People
- University of Bristol Recruits AI Faculty, Touting UK's Largest Supercomputer Isambard AI — dimadamen · 2026-10-09
- Weekly AI roundup: ChatGPT generates its own interfaces, Claude Haiku 5.5 reclaims the cheap-model frontier — The AI Daily Brief · 2026-10-09
- PartyKit founder shuts down free hosted platform 2.5 years after Cloudflare acquisition — threepointone · 2026-10-09
- AI industry's obsession with ARR: an easily manipulated metric with no standard definition — zijing_wu · 2026-10-09
- ChatGPT hit 1M users in 5 days; rivals took 5 weeks just to respond — letandrewcook · 2026-10-09
- Four community demos locked for Open Together in SF on Oct 16, applications close Oct 12 — erhannah · 2026-10-09