How Large Language Models Work: Inside an LLM (Field Guide)
A plain-English tour of how a large language model is built, trained, shrunk, and run.
Inside a Large Language Model
From CR Labs (crlab.ca) — by Zubair Ashraf. A two-host field guide to the full lifecycle of a large language model: what parameters mean, how pre-training builds the brain, how post-training and efficient methods like LoRA make it helpful, how quantization shrinks it, how retrieval keeps it current, and how inference serves it — with the real hardware and cost behind every stage.
Chapters:
00:00 Inside a Large Language Model
00:14 A model's life in five acts
00:27 Build once, deliver forever
00:46 What parameters really are
00:56 Parameters are knobs
01:36 Pre-training: building the brain
01:41 The whole trick: guess the next token
02:19 Why not at home?
02:39 Post-training: from autocomplete to assistant
02:44 Step one: supervised fine-tuning
02:59 Step two: learning from preferences
03:32 Efficient fine-tuning
04:05 Who uses what
04:09 Quantization: making it fit
04:53 Retrieval: knowledge without retraining
05:17 Inference: the recurring bill
05:22 Training is once. Serving is forever.
05:45 A simple decision guide
05:49 Thanks for watching
Watch the full series:
Inside a Large Language Model on crlab.ca: https://crlab.ca/videos/inside-a-larg...
YouTube playlist: • Inside a Large Language Model — A Field Guide
More AI explainers from CR Labs: https://crlab.ca/videos/