AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

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

Yujie Wang, Siwei Chen, Longzan Luo, Xinyi Liu, Xupeng Miao, Fangcheng Fu, Bin CUI

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add token-level streaming (ablation)VeRL · Ablation, Qwen2.5-14Bspeedup: 1.00x -> 1.09xImproved
add distribution-aware (dual-end) sampling with uniform redundancy (ablation)VeRL + token streaming · Ablation, Qwen2.5-14Bspeedup: 1.09x -> 1.40xImproved
add variance-based adaptive redundancy allocation (ablation)DARTS sampling · Ablation, Qwen2.5-14Bspeedup: 1.40x -> 1.63xImproved
shift dual-end ratio L:K from 1:7 to 4:4 (ablation)DARTS · Qwen2.5-Math-7Bspeedup: 1.45x -> 1.25xHurt
shift dual-end ratio L:K from 1:7 to 4:4 (ablation)DARTS · Qwen2.5-3Bspeedup: 1.29x -> 1.15xHurt
shift dual-end ratio L:K from 1:7 to 2:6 (ablation)DARTS · Qwen2.5-3Bspeedup: 1.29x -> 1.35xNo clear effect
scale cluster from 16 to 64 GPUs (ablation)DARTS · Qwen2.5-Math-7Bspeedup: 1.21x -> 1.55xImproved
replace VeRL with DARTS on BBHQwen2.5-32B · BBH zero-shotaccuracy: 78.1 -> 84.7Improved
replace VeRL with DARTS on BBHQwen3-30B-A3B · BBH zero-shotaccuracy: 85.8 -> 84.4Hurt
replace VeRL with DARTS on BBHQwen2.5-3B · BBH zero-shotaccuracy: 34.7 -> 38.7Improved
apply a larger explicit length penalty instead of distribution shapingGRPO · downstream math tasksperformance drop: 2%-7% performance dropHurt
use DARTS instead of Tail BatchingVeRL-based RL · end-to-end throughputspeedup over VeRL: 1.07x-1.62x -> 1.29x-1.77xImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy