A free textbook that actually covers LLMs end-to-end

What it does

This repository hosts a full textbook—ten chapters plus preliminaries—tracing NLP from classic neural foundations through modern large language models. Each chapter is available as a standalone PDF, with a complete bound version also provided. The authors, Tong Xiao and Jingbo Zhu, expanded earlier published survey papers (on transformers and LLM foundations) into a broader curriculum.

The interesting bit

The book is deliberately structured as a single narrative arc: Part I covers machine learning and neural net basics, Part II walks through RNNs, CNNs, seq2seq, and transformers, then Part III dives straight into the contemporary stack—pre-training, prompting, alignment, and inference. That progression is rarer than it sounds; most free resources either stop at 2017 or assume you already speak transformer.

Key highlights

• Covers the full pipeline from word embeddings to RLHF-style alignment and efficient inference
• Derived from peer-reviewed survey papers, not blog posts
• Complete PDF and per-chapter PDFs available directly in the repo
• Machine-translated versions exist in six languages (Chinese, Japanese, French, German, Italian, Portuguese) via a separate repo
• Published 2025, so the LLM material is genuinely recent

Caveats

• The translations are LLM-generated; quality control is unclear
• No exercises, code, or interactive elements—this is strictly a reading text
• The repo itself is a file host; issues go to the authors’ personal email

Verdict

Worth bookmarking if you’re a student or practitioner who wants one coherent reference instead of fifty scattered papers. Skip it if you need hands-on notebooks or are already deep enough in LLM research that survey material feels redundant.