TGRL: Temperature-Grouped Reinforcement Learning for Efficient Exploration in LLMs

2026-09-27

AI Summary

This study addresses the challenges of inefficient exploration and the unquantified benefits of temperature parameters in reinforcement learning for large language models. To this end, we propose Temperature-Grouped Reinforcement Learning, a method that transforms temperature diversity into an explicit training signal. By contrasting rewards across groups to estimate exploration gains, it introduces the first use of Jensen-Shannon divergence to map these gains into token-level credit assignment, thereby enhancing training efficiency without increasing the rollout budget. Experimental results demonstrate that our approach accelerates training by 36% over baselines, yields an average improvement of 1.6% on mathematical tasks, and increases CodeForces ratings by 196.7 points, significantly outperforming existing methods across multiple benchmarks.

Abstract

Efficient exploration often remains a central bottleneck in reinforcement learning with verifiable rewards (RLVR). Although temperature control and test-time scaling strategies can increase rollout diversity of large language models (LLMs), they either expand the sample budget at rollout time or leave the benefit of exploration unquantified. To this end, we propose Temperature-Grouped Reinforcement Learning (TGRL), which turns temperature-induced diversity into an explicit training signal. For each prompt, TGRL partitions its rollout group into low- and high-temperature subsets, estimates exploration gain through their reward contrast, and allocates this group-level signal as token-level credit using Jensen--Shannon (JS) divergence between the corresponding temperature-scaled next-token distributions induced by the same logits. Notably, TGRL reaches equivalent accuracy up to 36% faster than strong RLVR baselines without expanding the rollout budget. Across 11 benchmarks from diverse domains, TGRL broadly improves over strong RLVR baselines: it improves the six-benchmark math average by 1.6% at 32B, raises CodeForces rating by 196.7 points and LiveCodeBench Pass@16 by 4.4%, and improves ALFWorld/WebShop success rates by 6.3%/4.9%. Comprehensive ablations and wall-clock analysis confirm the efficacy of all proposed components. Code is available at https://github.com/1229095296/TGRL/tree/main.

Problem

• Reinforcement Learning with Verifiable Rewards
• Efficient Exploration
• Large Language Models
• Temperature Control
• Rollout Diversity

Innovation

• Temperature-Grouped Reinforcement Learning
• Exploration Gain
• Jensen-Shannon Divergence
• Token-level Credit
• Reinforcement Learning with Verifiable Rewards