Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models
This work addresses efficiency issues in reinforcement learning for large reasoning models, offering a domain-specific solution that is incremental but impactful for computational resource allocation.
The paper tackles the high computational cost of reinforcement learning for large language models by introducing Generalizable Predictive Prompt Selection (GPS), which uses a lightweight generative model to prioritize informative prompts, resulting in substantial improvements in training efficiency, final performance, and test-time efficiency across varied reasoning benchmarks.
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.