Sebastian Raschka's from-scratch video course on building reasoning models
Web page · https://magazine.sebastianraschka.com/p/building-reasoning-models
First seen by Oriel on .
AI summary
This page presents an eight-video, 11-hour PyTorch course that builds and improves a reasoning model from a pre-trained Qwen3 0.6B model.
- The standalone course covers evaluation, inference-time scaling, reinforcement learning with verifiable rewards, and distillation, with Python and PyTorch implementations.
- The series implements streaming generation, KV caching, torch.compile, sampling, self-consistency, self-refinement, and mathematical verification on MATH-500.
- A simplified GRPO training loop uses verifier-derived rewards, relative advantages, clipping, KL terms, format rewards, checkpoints, and training metrics.
- The final lecture distills reasoning capabilities from teacher-generated DeepSeek R1 answers using supervised training and compares results on MATH-500.
- Early exercises can run on a CPU, while later training experiments are more practical with a GPU; Python and PyTorch familiarity is recommended.
#reasoning models #pytorch #reinforcement learning #llm distillation
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 849, latest observation ; +838 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 116, latest observation ; +115 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50245138 | https://magazine.sebastianraschka.com/p/building-reasoning-models Read discussion |