TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning

Lizhou Cai·Yuhang Jiang·Weijie Liu·Ru Peng·Heming Zou·Qi Wang·Kai Yang·Xin Xu+4 more

Jun 9, 2026submitted

180citations

View PDF arXiv page Code4.1k

Preview of TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning

TL;DR

AI-GENERATED

State-of-the-art research on TRACE: A Unified Rollout Budget Allocati achieving benchmark breakthroughs.

TASKS

Foundation Models

METHODS

Quantum Neural Network

Related Papers

Back to papers

Repository

lucidrains/magvit2-pytorch

Official

4.1kstars71forks

Official implementation from the authors

Open Repository

Citation

@article{cai2026,
  author = {Lizhou Cai, Yuhang Jiang, Weijie Liu, Ru Peng, Heming Zou, Qi Wang, Kai Yang, Xin Xu, Yixiu Mao, Saiyong Yang, Xiangyang Ji, Yun Qu},
  title = {TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning},
  year = {2026},
  eprint = {2606.11119},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2606.11119},
}

Paper Details

Published

Jun 9, 2026

Updated

Jun 26, 2026, 01:06 PM

DOI

10.48550/arXiv.2606.11119

arXiv

2606.11119

License

CC-BY-4.0

Source

Verified Archive

References

Citations

180