Parameter Exploration for RLVR via Variational Learning
2026-08-10 • Machine Learning
Machine LearningArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors studied how large language models (LLMs) learn better by exploring different options during training. Instead of just changing how the model picks words (action-space exploration), they changed the model's parameters themselves (parameter-space exploration) to explore more diverse behaviors. They introduced a method called 3PO, which samples different policies to improve learning. Their tests on math and code tasks showed this approach helps the models perform better and makes training more stable compared to older methods. Overall, the authors showed that exploring through changing parameters can be a useful way to improve LLM reinforcement learning.
Reinforcement LearningLarge Language Models (LLMs)ExplorationAction-space ExplorationParameter-space ExplorationPolicy OptimizationPerturbed Parameter Policy Optimization (3PO)RolloutsReward EstimationTraining Stability
Authors
Vatsal Venkatkrishna, Nico Daheim, Iryna Gurevych
Abstract
Exploration has been a focus of reinforcement learning research for a long time. Recently, there has been growing evidence that it is also an important ingredient in LLM reinforcement learning recipes that can significantly impact downstream performance. Many existing methods control exploration in the action-space, for example, using temperature scaling. However, these methods cannot reorder tokens but only influence the variance in the output distribution. This limits exploration and can lead to divergence or stalled training. Here, we investigate parameter-space exploration, where rollouts are generated by sampling different policies from a posterior that may each explore different rollouts. Sampling less or more diverse policies is then a complementary control lever over exploration. We introduce a family of methods called Perturbed Parameter Policy Optimization (3PO) which use different sampling strategies and different rollout grouping for reward estimation. Experiments on OLMo-3-1025-7B and Qwen2.5-Math-7B across mathematical reasoning and code generation tasks show that these approaches consistently improve average downstream performance over standard GRPO at a near-identical FLOPs cost. Moreover, using multiple parameter samples consistently produces fewer zero-advantage groups and malformed or incorrect rollouts during training than GRPO and action-space baselines. Overall, our work presents evidence that parameter-space exploration can improve reinforcement learning for LLMs.