Training for Reasoning
Policy gradients and GRPO-style training, distillation, tool use and a cost-versus-accuracy lab.
0 of 3 complete
- ▶ Up nextPro24 min+50 XPPolicy gradients and GRPO-style training, on a problem you can runReinforcement learning with verifiable rewards from first principles: the policy-gradient identity, baselines and…
- Pro18 min+50 XPDistillation, curricula and tool-augmented reasoningTransfer reasoning from a strong model to a small one by supervised distillation of verified traces, rejection…
- LabPro90 min+100 XPLab: cost versus accuracy for reasoning strategiesRun a simulator that compares direct answers, chain of thought, voting, verifier best-of-n and adaptive sampling on…