DARTS: Distribution-Aware Active Rollout Trajectory Shaping for Accelerating LLM Reinforcement Learning
This work tackles the rollout efficiency bottleneck in LLM reinforcement learning, a critical issue for researchers and practitioners training large language models.
This paper addresses the inefficiency in LLM reinforcement learning caused by long-tail response length distributions, specifically identifying intra-prompt long tails of ineffective verbosity. The authors propose DARTS, a method that actively shapes the rollout distribution towards conciseness and certainty, achieving up to 1.77x acceleration over state-of-the-art systems without performance compromise.
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.