ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agents

arXiv cs.AI / 3/20/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical Usage

共有:

Key Points

ProRL Agent proposes a rollout-as-a-service infrastructure that serves the full agentic rollout lifecycle via an API, enabling scalable RL training for multi-turn LLM agents.
It provides standardized and extensible sandbox environments for diverse agentic tasks in rootless HPC settings, easing deployment and maintenance.
The approach decouples rollout orchestration from the training loop, addressing integration, migration, and maintenance challenges in existing RL pipelines.
The solution is open-sourced and integrated with NVIDIA NeMo Gym, with validation through RL training on software engineering, math, STEM, and coding tasks.

Abstract

Multi-turn LLM agents are increasingly important for solving complex, interactive tasks, and reinforcement learning (RL) is a key ingredient for improving their long-horizon behavior. However, RL training requires generating large numbers of sandboxed rollout trajectories, and existing infrastructures often couple rollout orchestration with the training loop, making systems hard to migrate and maintain. Under the rollout-as-a-service philosophy, we present ProRL Agent , a scalable infrastructure that serves the full agentic rollout lifecycle through an API service. ProRL Agent also provides standardized and extensible sandbox environments that support diverse agentic tasks in rootless HPC settings. We validate ProRL Agent through RL training on software engineering, math, STEM, and coding tasks. ProRL Agent is open-sourced and integrated as part of NVIDIA NeMo Gym.