The distributed AI laboratory Gradient today released the Echo-2 distributed reinforcement learning framework (arxiv.org/pdf/2602.02192), aimed at breaking the training efficiency barrier in AI research. By achieving a complete decoupling of Learner and Actor at the architectural level, Echo-2 reduced the post-training cost of a 30B model from $4,500 to $425. It delivers over 10 times the scientific throughput under the same budget.

The framework utilizes computing and storage separation technology for asynchronous training (Async RL), offloading massive sampling computing power to unstable GPU instances and Parallax-based heterogeneous GPUs. Coupled with breakthroughs in bounded staleness, instance fault-tolerant scheduling, and self-developed Lattica communication protocol, it significantly improves training efficiency while ensuring model accuracy. Alongside the framework release, Gradient is also set to launch the RLaaS platform Logits, promoting a shift in AI research from "capital accumulation" to "efficiency iteration." Logits is now open for reservation to students and researchers worldwide (logits.dev).

About Gradient

Gradient is an AI laboratory dedicated to building distributed infrastructure, focusing on distributed training, servicing, and deployment of cutting-edge large models. Gradient has received support from top investment institutions and is building an open and efficient future intelligent era.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。