Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models

TL;DR

This paper proposes GPS, a lightweight, generalizable predictive model for efficient prompt selection in reinforcement learning of large language models, improving training and inference efficiency.

Contribution

Introduction of GPS, a Bayesian inference-based prompt selection method that generalizes across prompts and incorporates diversity, reducing computational costs in RL training of language models.

Findings

GPS significantly improves training efficiency over baselines.

GPS enhances final reasoning performance.

GPS reduces test-time computational costs.

Abstract

Reinforcement learning enhances the reasoning capabilities of large language models but often involves high computational costs due to rollout-intensive optimization. Online prompt selection presents a plausible solution by prioritizing informative prompts to improve training efficiency. However, current methods either depend on costly, exact evaluations or construct prompt-specific predictive models lacking generalization across prompts. This study introduces Generalizable Predictive Prompt Selection (GPS), which performs Bayesian inference towards prompt difficulty using a lightweight generative model trained on the shared optimization history. Intermediate-difficulty prioritization and history-anchored diversity are incorporated into the batch acquisition principle to select informative prompt batches. The small predictive model also generalizes at test-time for efficient computational allocation. Experiments across varied reasoning benchmarks indicate GPS's substantial improvements in training efficiency, final performance, and test-time efficiency over superior baseline methods.