Unified User Interface Usage

This chapter introduces the fundamental static components of RLinf from a user’s perspective. These components lay the groundwork for configuring and launching scalable RL workloads.

• YAML Configuration

  A comprehensive guide to all YAML configuration parameters used throughout RLinf scripts. Learn how to structure your configuration files for clarity, flexibility, and reproducibility.

• Worker-Based Programming Interface

  Introduces the concept of a Worker, the modular execution unit in RLinf, each handling a specific task in the RL pipeline. Multiple similar Workers form a WorkerGroup, simplifying distributed execution and promoting scalability.

• Worker Placement Strategy

  Explains how RLinf strategically assigns hardware resources across different tasks and workers to ensure efficient hardware utilization and balanced execution, including not only accelerator hardware (e.g., GPUs, NPUs) but also robotic hardware and CPU-only nodes.

• Ray-Based Cluster Launching

  Describes the globally unique Cluster object, responsible for coordinating all roles, processes, and communication across distributed nodes in a training job.

• High-Level Programming Flow

  Integrates the concepts of WorkerGroup, Placement, and Cluster, and presents the complete programming flow of RLinf to illustrate its high-level workflow.