Building Reasoning Models from the Ground Up: A Video Course ↗

Summary

Sebastian Raschka releases an 11-hour video course implementing reasoning models from scratch using PyTorch. The series covers evaluation on MATH-500, inference-time scaling (self-consistency, self-refinement), Reinforcement Learning with Verifiable Rewards (simplified GRPO), and distillation from a teacher model. It uses a pre-trained Qwen3 0.6B base model. Raschka provides code for each step, including environment setup, answer verification, and training loops. The content is designed as a sequel to his previous LLM course but stands alone. He distinguishes reasoning models (the engine) from agent harnesses (the machinery). The post links to eight specific videos covering these topics in detail.