Ant Group open-sources AState: a high-performance state management system for reinforcement learning
Original title:inclusionAI/asystem-astate
AI summary
Ant Group has open-sourced AState, a high-performance general-purpose state data management system designed for reinforcement learning. It aims to address core challenges in RL training and inference, including low I/O efficiency, insufficient weight synchronization performance, and non-robust state recovery. The system uses a three-layer architecture: an API layer with tensor-native interfaces, a service layer supporting multiple deployment modes, and a foundation layer with efficient and scalable transmission capabilities. Key features include a unified tensor-level API, high-performance weight synchronization, and topology-aware design. At trillion-parameter scale, AState can complete weight synchronization in about 6 seconds, far below the industry's common minute-level latency. It is currently deployed in Ant's internal production environment as a key component of ASystem.
Key Takeaways
- Ant Group has open-sourced AState,a high-performance general-purpose state data management system…
- It aims to address core challenges in RL training and inference,including low I/O…
- The system uses a three-layer architecture: an API layer with tensor-native interfaces,a service…
- Key features include a unified tensor-level API,high-performance weight synchronization,and…
- At trillion-parameter scale,AState can complete weight synchronization in about 6 seconds,far…
Quick Facts
- Published
- Updated
- Source
- 蚂蚁 inclusionAI:GitHub 新仓库
- Category
- Products & Apps
- AI score
- 63
蚂蚁把万亿参数 RL 训练的权重同步从分钟级压到 6 秒,这套 AState 系统是真刀真枪的工程解法,做大规模 RL infra 的团队值得拆一拆它的 RDMA P2P 架构。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Ant Group open-sources AState: a high-performance state management system for reinforcement learning. December 3, 2025. https://new.nat6.net/en/news/cmorb7ik3006wslhfydetkk7m/