GenSelect: A Generative Approach to Best-of-N
This addresses a bottleneck in test-time scaling for reasoning tasks, offering a more efficient method for selecting optimal solutions from large candidate sets.
The paper tackles the problem of inefficient scaling in generative reward models for reasoning tasks by introducing GenSelect, where LLMs use long reasoning to select the best solution among N candidates. This approach outperforms existing scoring methods on math reasoning tasks, leveraging LLMs' comparative strengths while scaling efficiently across parallel sampling budgets.
Generative reward models with parallel sampling have enabled effective test-time scaling for reasoning tasks. Current approaches employ pointwise scoring of individual solutions or pairwise comparisons. However, pointwise methods underutilize LLMs' comparative abilities, while pairwise methods scale inefficiently with larger sampling budgets. We introduce GenSelect, where the LLM uses long reasoning to select the best solution among N candidates. This leverages LLMs' comparative strengths while scaling efficiently across parallel sampling budgets. For math reasoning, we demonstrate that reasoning models, such as QwQ and DeepSeek-R1-0528, excel at GenSelect, outperforming existing scoring approaches with simple prompting.