Chollet Recommends His Book's Chapters 15-16 for Building LLMs from Scratch
On August 25, Keras creator François Chollet publicly recommended chapters 15-16 of his own book Deep Learning with Python for anyone wanting to build a large language model from scratch, a recommendation that spread further through retweets, including reader endorsements and Chollet's own promotion of the chapter 15 attention-mechanism explanation.
Confirmed
- Chollet's recommended path is reading chapters 15-16 of Deep Learning with Python (third edition), aimed at readers building an LLM from scratch, and he noted it suits 17-year-olds as well as learners of any age.
- He emphasized that chapter 15 offers one of the most intuitive and clearest explanations of why dot-product attention works.
- The book's code examples are written in Keras and can run under frameworks such as JAX and PyTorch.
- The book is freely readable online, keeping the entry barrier low.
Why it matters
- Dot-product attention is the core mechanism behind Transformers and modern LLMs, yet intuitive explanations are rare; a chapter personally endorsed by Chollet is a reliable, low-cost starting point for beginners.
- The combination of Chollet's recommendation and his reader community's echo spread this learning path widely within a single day, offering strong reference value for self-learners choosing materials.
2026-08-25 ~ 2026-08-25 · 7 related posts
Primary sources
- [source] François Chollet Recommends Chapters for Building LLMs from Scratch — fchollet · 2026-08-25
- Chollet recommends Deep Learning with Python ch. 15-16 for learning LLMs from scratch — fchollet · 2026-08-25
5 near-duplicate retellings: fchollet · fchollet · fchollet · fchollet · fchollet