Sebastian Raschka's from-scratch video course on building reasoning models

	Sebastian Raschka's from-scratch video course on building reasoning models(magazine.sebastianraschka.com)
	722 points by sgiannopoulos 20 hours ago \| 100 comments
	tl;dr: Sebastian Raschka released a free 11-hour video course (8 lectures) accompanying his "Build a Reasoning Model (From Scratch)" book, walking through implementation in Python and PyTorch starting from a pre-trained Qwen3 0.6B model. The course covers evaluation with verifiers, inference-time scaling (temperature sampling, self-consistency, self-refinement), reinforcement learning with verifiable rewards (a simplified GRPO), and distillation from a stronger teacher model (DeepSeek R1). Prerequisites are Python/PyTorch familiarity; early examples run on CPU, but training sections benefit from a GPU.
	HN Discussion: ↑Praise for Raschka's clear pedagogy and recommendation of his books/courses ↑From-scratch tutorials are valuable for understanding fundamentals, not for building production models ~Learning ML/LLMs feels frustrating due to math-heavy theory or compute-heavy practice ↓Questioning the article's definition of reasoning models as just RL post-training ↓Build-your-own-model guides are impractical and of limited real-world use