Setting up the data needed to train and evaluate simple self-distillation for competitive math.

Improving LLM performance on competitive math through unverified self-distillation.