Build A Reasoning Model From Scratch 8: Distillation

In this final video, we look into model distillation as a more efficient alternative to reinforcement learning for developing small reasoning models. - Reasoning playlist: https://www.youtube.com/watch?v=Kh9mqTzjuEQ&list=PLfrHAQCRxtUs - Reasoning Book: https://amzn.to/4aAKiFY - Reasoning GitHub repo: https://github.com/rasbt/reasoning-from-scratch - LLMs from Scratch book: https://amzn.to/4fqvn0D - LLMs from Scratch repo: https://github.com/rasbt/LLMs-from-scratch - LLMs from Scratch playlist: https://www.youtube.com/watch?v=yAcWnfsZhzo&list=PLTKMiZHVd_2IIEsoJrWACkIxLRdfMlw11&index=1 00:00 Introduction and series recap 01:07 Why distill a reasoning model? 02:51 Soft vs. hard distillation 11:06 On-policy vs. off-policy distillation 18:09 Preparing data for reasoning distillation 22:35 Loading the DeepSeek R1 MATH dataset 28:04 Inspecting reasoning traces and final answers 31:27 Tokenization and chat templates 35:58 Pre-tokenizing the training dataset 39:01 Sequence lengths and the 2,048-token limit 42:38 Creating training and validation splits 45:32 Loading the Qwen3-0.6B base model 46:24 Computing answer-token log probabilities 54:10 Cross-entropy loss for distillation 57:25 Training and validation loss helpers 1:02:03 Implementing the distillation training loop 1:07:26 Running the standalone training script 1:09:41 Plotting training and validation losses 1:11:31 Evaluating checkpoints on MATH-500 1:14:09 Comparing teachers and distillation results 1:16:34 Future directions for reasoning models 1:19:41 Bonus exercises and next steps