AIS: Adaptive Importance Sampling for Quantized RL

TL;DR

This paper introduces Adaptive Importance Sampling (AIS), a correction framework for quantized reinforcement learning that balances exploration and stability, improving training efficiency for large language models.

Contribution

AIS dynamically adjusts gradient correction during training, mitigating bias from low-precision rollouts while maintaining exploration benefits.

Findings

AIS matches BF16 baseline performance on most tasks.

AIS retains 1.5 to 2.76x rollout speedup of FP8.

AIS improves training stability on reasoning benchmarks.

Abstract

Reinforcement learning (RL) for large language models (LLMs) is dominated by the cost of rollout generation, which has motivated the use of low-precision rollouts (e.g., FP8) paired with a BF16 trainer to improve throughput and reduce memory pressure. This introduces a rollout-training mismatch that biases the policy gradient and can cause training to collapse outright on reasoning benchmarks. We show that the mismatch is non-stationary and acts as a double-edged sword: early in training it provides a stochastic exploration bonus, exposing the gradient to trajectories the trainer would otherwise under-sample, but the same perturbation transitions into a destabilizing source of bias as the policy concentrates. To solve this, we propose Adaptive Importance Sampling (AIS), a correction framework that adjusts the strength of its intervention on a per-batch basis. AIS combines three real-time diagnostics, namely weight reliability, divergence severity, and variance amplification, into a single mixing coefficient that interpolates between the uncorrected and fully importance-weighted gradients, suppressing the destabilizing component of the mismatch while preserving its exploratory benefit. We integrate AIS into GRPO and evaluate it on the diffusion-based LLaDA-8B-Instruct and the autoregressive Qwen3-8B and Qwen3.5-9B across mathematical reasoning and planning benchmarks. AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8.