OmegaPRM & Consensus-Filtered PRM (Qwen2.5-Math-PRM)

Scales automated process supervision efficiently using a Divide-and-Conquer Monte Carlo Tree Search (OmegaPRM) that performs binary search over a failing trajectory to locate the exact first error step in $\mathcal{O}(\log T)$ rollouts instead of $\mathcal{O}(T)$ linear rollouts. Qwen2.5-Math-PRM further發現 that pure MC estimation has high false-positive rates (a wrong intermediate step can still reach the right answer by luck) and intersects MC labels with an LLM-as-a-Judge critique filter (Consensus Filtering).