在 Amazon EKS 上用 EFA 与 DeepEP 扩展 MoE 强化学习，吞吐量提升 40%
Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput
AWS 官方给出了在 EKS 上跑 MoE 强化学习的架构，用 EFA 加 DeepEP 把 RLHF/GRPO 训练吞吐量提高了 40%，自建训练集群的可以抄作业。
AWS 发布了一套在 Amazon EKS 上训练 MoE 强化学习的架构方案，组合了 Elastic Fabric Adapter（EFA）与 DeepEP 通信库，并搭配 Amazon S3 存储检查点。该方案面向大规模 RLHF 与 GRPO 训练场景，将聚合 rollout 吞吐量提升了 40%。文章给出了具体的架构设计与部署方式，适合自行搭建强化学习训练集群的团队参考。
Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput
Learn how to scale Mixture-of-Experts (MoE) reinforcement learning on Amazon EKS using Elastic Fabric Adapter (EFA) and DeepEP. This post presents an architecture that combines Amazon EKS, EFA, and Amazon S3 and increased aggregate reinforcement learning rollout throughput by 40% for large-scale RLHF and GRPO training.
