ROLL: sample-level rollout orchestration and resource mapping

What changed

Integrated parallel workers, sample-level rollout scheduling, environment/reward workers and automatic device mapping in a configurable RL training library for multi-domain and agentic workloads.

Scientific scope

Integrated parallel workers, sample-level rollout scheduling, environment/reward workers and automatic device mapping in a configurable RL training library for multi-domain and agentic workloads. Inherits HybridFlow's single-controller design, Ray, Megatron/DeepSpeed, vLLM/SGLang and RL algorithms. Two-week200B-scale run is a reported deployment, not a fault-free theorem or independent audit.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits HybridFlow's single-controller design, Ray, Megatron/DeepSpeed, vLLM/SGLang and RL algorithms. Two-week200B-scale run is a reported deployment, not a fault-free theorem or independent audit.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Reinforcement Learning Optimization for Large-Scale Learning: An Efficient and User-Friendly Scaling Library

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction