Sebastian Raschka has released an 11-hour video course titled 'Build a Reasoning Model (From Scratch)', teaching how to implement reasoning models using PyTorch. The course covers inference scaling, reinforcement learning, and distillation, starting from a pre-trained Qwen3 model. It is designed as a sequel to his 'Coding LLMs from the Ground Up' series but can be taken standalone.
The videos focus on evaluation, inference-time scaling, RL, and distillation, with early sections runnable on CPU and later sections requiring GPU. Raschka also explains the relationship between reasoning models and AI agents, describing reasoning models as the 'engine' inside an agent harness—a software framework that provides tool access and context management. While reasoning models are typically developed for math and code, their capabilities generalize to other domains.
Most modern LLMs now function as reasoning models with adjustable reasoning effort settings. The course complements Raschka's book 'Build a Large Language Model (From Scratch)' and his Substack article 'The Components of a Coding Agent'.
Source: Hacker News · Summarized by HeadlinesBriefing