DARTS: Distribution-Aware Active Rollout Trajectory Shaping for Accelerating LLM Reinforcement Learning

2026-05-29

AI Summary

This work addresses the inefficiency in large language model reinforcement learning caused by the long-tailed distribution of response lengths, which leads to suboptimal rollout efficiency. The authors propose an active distribution shaping paradigm that explicitly models the fine-grained long-tailed characteristics inherent in prompts. By integrating distribution-aware trajectory sampling with an adaptive redundancy allocation mechanism, the method actively steers rollouts toward more concise and deterministic responses. Without compromising model performance, this approach achieves up to a 1.77× speedup in training compared to state-of-the-art systems, substantially mitigating the computational overhead induced by response-length long tails.

Abstract

Reinforcement Learning (RL) has become pivotal for improving model capabilities yet suffers from rollout efficiency bottlenecks due to the long-tail response length distribution. While existing works mitigate the impact of long tails via prompt-level tail scheduling, we focus on the root source of inefficiency: the distribution itself. Specifically, we characterize the long-tail distribution at a finer granularity, identifying intra-prompt long tails, and revealing that they frequently consist of ineffective verbosity. To address this, we propose a novel paradigm of active distribution shaping to shape the rollout distribution towards conciseness and certainty, thereby fundamentally resolving tail-induced overheads. We achieve this through a distribution-aware trajectory sampling mechanism, which selects trajectories from a redundant exploration space for each prompt, and an adaptive redundancy allocation scheme to maximize both shaping effectiveness and system efficiency. Experiments demonstrate significant acceleration over state-of-the-art systems by up to 1.77x without compromising model performance.

Problem

• reinforcement learning
• long-tail distribution
• rollout efficiency
• response length
• trajectory sampling

Innovation

• Distribution-Aware
• Active Rollout Shaping
• Intra-Prompt Long Tail
• Trajectory Sampling
• Redundancy Allocation