Sebastian Raschka's free 11-hour course trains a reasoning model from Qwen3 0.6B
Eight YouTube videos build reasoning-model techniques in PyTorch: evaluation, inference-time scaling, GRPO reinforcement learning and distillation.
The course improves a pre-trained Qwen3 0.6B model rather than training from scratch, runs early sections on a CPU and ships code in a GitHub repository.
Commenters split between readers of his LLM book who loved it and skeptics who asked who the course is for.
What the eight videos cover
Raschka's announcement: about 11 hours of coding across eight videos, published October 24, using Python and PyTorch. Topics are evaluation, inference-time scaling, reinforcement learning with GRPO and distillation. A GPU is recommended for the later training parts. A companion book and a shorter 90-minute LinkedIn Learning course exist.
Who it's for
"I got his book Build Your Own LLM From Scratch and it's super readable and does a great job at holding your hand through the entire process" — a reader of Raschka's earlier LLM book
"All of these build your own model guides are like someone putting out a build your own CPU guide 10 years ago. Interesting, but not of any practical use." — a commenter
The deciding detail is the 0.6B model: small enough that you can reproduce the techniques on a laptop and see what each does, even if the result will not compete with anything you would deploy.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Someone built all of Switzerland in Minecraft at one block per meter, for CHF 15
Blocks generates 41,285 square kilometers of Swiss terrain from open geodata, with the Dufour peak at 4,634 blocks tall.