Fetching the paper…

Smaller, Weaker, Yet Better: Training LLM Reasoners via Compute-Optimal Sampling · Around