Adaptive Layerwise Perturbation: Unifying Off-Policy Corrections for LLM RL

TL;DR

This paper introduces Adaptive Layerwise Perturbation (ALP), a method that injects controlled noise into layer inputs during training to stabilize off-policy learning in large language models, reducing importance ratio tails and improving performance.

Contribution

ALP is a novel approach that applies learnable perturbations at each layer to unify off-policy corrections, enhancing training stability and exploration in LLM RL.

Findings

ALP reduces importance ratio tail heaviness and KL spikes during training.

ALP improves final performance on math and reasoning tasks.

Representation-level perturbations across all layers are most effective.

Abstract

Off-policy problems such as policy staleness and training--inference mismatch have become a major bottleneck for training stability and further exploration in LLM RL. The distribution gap between the inference and updated policies grows because of the techniques to enhance inference efficiency, leading to heavy-tailed importance ratios. Heavy-tailed ratios arise when the policy is locally sharp, which further inflates gradients and can push updates outside the trust region. To address this, we propose Adaptive Layerwise Perturbation (ALP), which injects small learnable perturbations into the input hidden states of each layer during updates and uses the resulting perturbed policy as the numerator of the importance ratio against the unchanged inference policy in the objective. Intuitively, by adding controlled noise to intermediate representations, ALP prevents the updated policy from deviating too sharply from the inference policy and enlarges the policy family to cover inference-time mismatch noise. Hence, the flattened distribution can naturally tighten the gap between the updated and inference policies and reduce the tail of importance ratios, thus maintaining training stability. This is further validated empirically. Experiments on single-turn math and multi-turn tool-integrated reasoning tasks show that ALP not only improves final performance, but also avoids blow-up in the importance-ratio tail and KL spikes during iterative training, along with boosted exploration. Ablations show that representation-level perturbations across all layers are most effective, substantially outperforming partial-layer and logits-only variants.