Ray 2.56.0 发布：增强数据稳定性与 LLM 服务性能

AI 摘要

Ray 2.56.0 发布，重点提升 Ray Data 稳定性、Ray Serve LLM 性能和 Ray Core 的 GPU 域感知调度。新版本引入多数据集支持、自动批大小选择、逻辑内存配置，并重构了 LLM 服务路径，新增一致性哈希路由和容量队列路由。此外，支持 Kubernetes 原地 Pod 调整，增强集群弹性。

正文节选

# Highlights * **Ray Data Stability:** In this Ray release, we've added a variety of stability improvements, including running multiple datasets in a cluster, adding automatic batch size selection to CPU-based map-batches, and default logical memory configuration to prevent OOMs. We've also tightened `iter_batches` stability by reducing hidden buffering and shutting down the executor when consumers exit early (#63660, #63682, #62949). This reduces object-store spilling for common training wor

发布时间：2026-06-30 04:32

抓取时间：2026-08-02 00:31

来源机构：Anyscale