RayOrch v0.0.1 Release Note

RayOrch v0.0.1 首次发布

RayOrch 是一个基于 Ray 的轻量级编排框架，用于构建高吞吐、多阶段推理流水线。
这个版本的目标是把「阶段化推理」从“能跑”推进到“可编排、可并行、可优化”。

在 v0.0.1 中，我们重点解决了以下问题：

• 用统一抽象组织多阶段模块（RayModule）
• 支持异步重叠执行（Overlapped Pipeline）以提升吞吐
• 支持 DAG 依赖调度以表达非线性流水线
• 提供更清晰的 dispatch/collect 语义与环境管理能力
• 增加示例与测试，便于复现和扩展

---

RayOrch v0.0.1 — First Public Release

RayOrch is a lightweight orchestration layer built on Ray for high-throughput, stage-based inference pipelines.

This release establishes the core foundations for:

• reusable stage abstraction with RayModule
• async microbatch overlap execution
• DAG-style dependency scheduling
• explicit dispatch/collect semantics for replica-aware workloads
• practical examples and tests for multi-engine serving workflows

---

What's Changed

• feat(example): add SGLang ↔ vLLM ping-pong demo by @SunnyHaze in #1
• refactor(example): make ping-pong flow explicit and move prompts to separate module by @SunnyHaze in #2
• [test] add sample of dual vllm by @SunnyHaze in #3
• add nvtx profile for dev mode by @MOLYHECI in #4
• update env_registry and test it by @Lavender1 in #5
• Refactor RayModule and improve DAG pipeline clarity by @SunnyHaze in #6

New Contributors

• @SunnyHaze made their first contribution in #1
• @MOLYHECI made their first contribution in #4
• @Lavender1 made their first contribution in #5

Full Changelog: https://github.com/OpenDCAI/RayOrch/commits/v0.0.1