Author Releases 11-Hour Course Teaching 8 Reasoning-Model Builds From Scratch
Updated
Updated · magazine.sebastianraschka.com · Oct 24
Author Releases 11-Hour Course Teaching 8 Reasoning-Model Builds From Scratch
1 articles · Updated · magazine.sebastianraschka.com · Oct 24
Summary
Eight lecture videos totaling 11 hours were published as a new coding course on building reasoning models from scratch in Python and PyTorch.
The series starts from a pre-trained Qwen3 0.6B model and walks through evaluation, inference-time scaling, reinforcement learning with verifiable rewards, and distillation rather than full pretraining.
MATH-500 serves as the main benchmark, with the course showing how to build a verifier, test self-consistency and self-refinement, implement GRPO-style RLVR, and compare distilled checkpoints.
The videos accompany the author's book of the same theme, are designed as standalone material for learners comfortable with Python and PyTorch, and suggest GPU use for later training-heavy sections.
The release reflects the author's view that reasoning models remain central to modern AI agents, functioning as the core engine inside broader agent software harnesses.