How to build custom reasoning agents with a fraction of the compute

The technique, called Reinforcement Learning with Verifiable Rewards with Self-Distillation (RLSD), combines the reliable ...