{
  "id": 16318093,
  "title": "Building Reasoning Models from the Ground Up: A Video Course",
  "url": "https://urgent.news/2026/10/24/building-reasoning-models-from-the-ground-up-a-video-course",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-24T11:14:32.000Z",
  "source": {
    "name": "Sebastian Raschka",
    "slug": "sebastian-raschka",
    "url": "https://magazine.sebastianraschka.com/p/building-reasoning-models"
  },
  "original_language": "en",
  "account": "In the past two years, I have written extensively about reasoning models on Ahead of AI, including a recent article on controlling reasoning effort levels in contemporary LLMs. Reasoning models remain highly relevant, as almost all current-generation LLMs offer some form of reasoning capability. To help those interested in the inner workings of reasoning models, including inference scaling, reinforcement learning, and distillation, I recently completed a comprehensive video course titled \"Build Reasoning Model (From Scratch).\" This course implements these concepts from the ground up in PyTorch, offering an in-depth understanding of how they function. Additionally, accompanying lectures in the form of eight coding videos (totaling 11 hours of content) are available for viewers to follow along. The course begins with a pre-trained LLM and gradually implements methods to evaluate and enhance it using Python and PyTorch. While familiarity with the \"Coding LLMs from the Ground Up\" course from last year is beneficial, it is not mandatory to watch this series as a prerequisite. This sequel course builds upon those foundations and expands the scope to cover reasoning models in greater depth. The accompanying resources include the \"Build a Large Language Model (From Scratch)\" book and videos, as well as the \"Build A Reasoning Model (From Scratch)\" book and videos. The \"Components of a Coding Agent\" article on Substack further explores how reasoning models fit into the broader landscape of AI agents. The videos themselves focus on developing the reasoning model using a small pre-trained Qwen3 model. By starting with a pre-trained model, the course emphasizes evaluation, inference-time scaling, reinforcement learning, and distillation, which are the core aspects of improving a reasoning model. The course assumes a solid understanding of Python and PyTorch, although prior knowledge of \"Building a Large Language Model (From Scratch)\" is recommended but not strictly necessary. For those working through the early text-generation and evaluation examples, a CPU is sufficient. However, for more advanced experiments, particularly training the model, a GPU is highly recommended. The first video in the series covers the setup process and provides additional recommendations, while the repository also contains hardware and environment recommendations to ensure a smooth learning experience.",
  "summary": "Over 11 hours of coding videos on evaluation, inference-time scaling, reinforcement learning, and distillation, with Python and PyTorch.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}