"Build A Reasoning Model (From Scratch)" is a technical textbook by Sebastian Raschka, published by Manning Publications, that teaches step-by-step how to build Reasoning capabilities in LLMs using Python/PyTorch. Starting from a pretrained Open Source base model (Qwen3), the book implements methods such as Chain-of-Thought, Self-Consistency, Reinforcement Learning with verifiable rewards (GRPO), and Distillation completely from scratch, without third-party LLM libraries. The accompanying code is available as an Open Source repository on GitHub and runs primarily on consumer hardware. This is not a standalone commercial AI model with API pricing, but rather educational supporting material and code for building a small Reasoning model.

Zdroje (2)