AWS 发布 Ray Train DLC：在 EKS 上用 EFA 做分布式训练
原标题：Distributed training with Ray Train DLCs and EFA on EKS
AI 摘要
AWS 发布 Ray Train 深度学习容器（DLC），将 Ray、PyTorch、CUDA、NCCL、EFA 用户态库及 AWS OFI NCCL 插件等组件预先构建并统一测试，发布在 Amazon ECR Public 并持续打补丁。该镜像在 EKS 上通过 KubeRay 运行，同一标签可同时充当 Ray head 和 worker，用户只需配置实例、GPU、网络传输和额外 Python 包，无需自行构建镜像。示例在 2 台 g6.12xlarge 的 8 张 L4 GPU 上使用 PyTorch FSDP 微调 Qwen2.5-1.5B，跨实例通信直接走 EFA。
正文节选
Distributed training with Ray Train DLCs and EFA on EKS What usually stops PyTorch training from scaling out to a cluster is the container, not the model code. Every container needs the same CUDA runtime, a PyTorch build compiled against it, NCCL (NVIDIA Collective Communications Library), the Elastic Fabric Adapter (EFA) userspace libraries, the AWS OFI NCCL plugin that connects NCCL to EFA, and a Ray version the cluster operator agrees with. Building that image and rebuilding it each time one