SkyRL

Detailed Introduction

SkyRL is a modular full-stack reinforcement learning (RL) library maintained by NovaSky-AI, focused on building scalable training and evaluation pipelines for large language models (LLMs). The project includes subpackages such as skyrl-agent, skyrl-train, and skyrl-gym, covering environment construction, training stack, agent layers, and deployment tooling to support reproducible research and engineering for long-horizon, real-world tasks.

Main Features

• Modular components: split into training, agent, and environment libraries for easy composition and extension.
• Reproducible training pipelines: high-performance training stack and configurable experiment management for large-scale training.
• Rich environment suite: skyrl-gym provides tool-use environments implemented with the Gymnasium API.
• Open collaboration: Apache-2.0 license with comprehensive docs and examples for community contributions.

Use Cases

• Training long-horizon agents for multi-turn tool-use and dialog tasks.
• Benchmarking and evaluating training algorithms and model performance in realistic environments.
• Teaching and research: reproducing experiments, building baselines, and tuning performance.

Technical Features

• Implemented in Python and compatible with common deep-learning and distributed training toolchains, with a focus on performance and scalability.
• Command-line and configuration-driven interfaces enable running large-scale training on cloud or local clusters.
• Integrated monitoring and evaluation modules export experiment metrics to support reproducibility.