Essential models and datasets used to build the IPDA debate canonical model. Includes ORPO, GRPO iterations, SFT distillation, and golden samples.