Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents
The article discusses a new approach called Hera for coordinating device-cloud collaborative large language model (LLM) agents. Hera aims to optimize the performance and cost of LLMs by addressing the device-cloud dilemma through a two-stage training process. The method has shown promising results in various evaluations, outperforming existing techniques in efficiency and success rates.
- ▪Hera is designed to improve long-horizon task performance for LLM agents by optimizing device-cloud coordination.
- ▪The approach utilizes a two-stage training paradigm involving imitation learning and cost-aware reinforcement learning.
- ▪Hera consistently outperforms prior methods, achieving a 92.5% success rate with only 46.3% cloud usage in evaluations.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24598 (cs) [Submitted on 23 May 2026] Title:Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents Authors:Yuxin Zhang, Mengxue Hu, Zheng Lin, Xiaoyi Fan, Fan Xie, Zihan Fang, Jing Yang, Wenjun Zhu, Zhiwen Chen, Chengfei Lv, Zhe Chen View a PDF of the paper titled Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents, by Yuxin Zhang and 10 other authors View PDF HTML (experimental) Abstract:Large language model (LLM) agents excel at solving complex long-horizon tasks through autonomous interaction with environments.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
