How aburkov/thelmbook works
Project type and purpose
• This repo is not a traditional application or service—it’s an educational/book project (“The Hundred-Page Language Models Book”) made up of example code, Jupyter notebooks, and Markdown documentation. Its purpose is to teach language-model concepts through runnable experiments and narrative wiki pages.
File structure
• Top level:
• A set of .py scripts (e.g. GRPO.py, quadratic_loss.py, embedding_vs_linear.py) that implement core experiments or utilities.
• A set of Jupyter notebooks (.ipynb) demonstrating specific language-model workflows (e.g. sampling_method.ipynb, emotion_GPT2_as_classifier.ipynb).
• README.md with book title and cover image.
• “wiki/”: directory of Markdown pages forming the narrative text of the book (each topic from tokenization to distributed training).
• No dedicated build scripts or static-site generator config—this repo is structured as source artifacts (notebooks + docs) served directly by GitHub.
README
• Confirms this is the “Hundred-Page Language Models Book” and shows the cover. No further architectural details.