AsyncFlow: An Asynchronous Streaming RL Framework for Efficient LLM Post-Training

TL;DR

AsyncFlow is a novel asynchronous streaming reinforcement learning framework designed for efficient post-training of large language models, overcoming scalability bottlenecks and resource imbalances in existing RL systems.

Contribution

It introduces a distributed data management module and a producer-consumer asynchronous workflow, enabling scalable, flexible, and engine-agnostic RL post-training.

Findings

Achieves 1.59x throughput improvement over baselines

Supports automated pipeline overlapping and dynamic load balancing

Decouples RL framework from underlying training and inference engines

Abstract

Reinforcement learning (RL) has become a pivotal technology in the post-training phase of large language models (LLMs). Traditional task-colocated RL frameworks suffer from significant scalability bottlenecks, while task-separated RL frameworks face challenges in complex dataflows and the corresponding resource idling and workload imbalance. Moreover, most existing frameworks are tightly coupled with LLM training or inference engines, making it difficult to support custom-designed engines. To address these challenges, we propose AsyncFlow, an asynchronous streaming RL framework for efficient post-training. Specifically, we introduce a distributed data storage and transfer module that provides a unified data management and fine-grained scheduling capability in a fully streamed manner. This architecture inherently facilitates automated pipeline overlapping among RL tasks and dynamic load balancing. Moreover, we propose a producer-consumer-based asynchronous workflow engineered to minimize computational idleness by strategically deferring parameter update process within staleness thresholds. Finally, the core capability of AsynFlow is architecturally decoupled from underlying training and inference engines and encapsulated by service-oriented user interfaces, offering a modular and customizable user experience. Extensive experiments demonstrate an average of 1.59 throughput improvement compared with state-of-the-art baseline. The presented architecture in this work provides actionable insights for next-generation RL training system designs.