A Visual Guide to Reasoning LLMs (Extends Chapters 6 and 12)
In the book, we cover the concept of chain-of-thoughs to improve the quality of outputs through extended reasoning.
A Visual Guide to Reasoning LLMs dives deeper into making LLMs demonstrate extensive reasoning process. It describes the paradigm shift from train-time compute (more data, longer training, larger models) which is a focus of this book to the test-time compute (extend inference through "reasoning").
This guide explores various technique, both during training and during inference, to distill reasoning into LLMs. There is also a section describing an extremely impactful model in 2024, namely DeepSeek-R1.
