Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms

TL;DR

Echo introduces a novel RL system that decouples inference and training phases across heterogeneous hardware, maintaining efficiency and enabling large-scale RL for LLMs on distributed resources.

Contribution

The paper presents Echo, a system that separates inference and training in large-scale RL, using lightweight synchronization protocols to improve hardware utilization and scalability.

Findings

Echo matches baseline convergence speed and reward.

Offloads trajectory generation to edge hardware.

Enables large-scale RL with heterogeneous resources.

Abstract

Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This serial context switching violates the single-program-multiple-data (SPMD) assumption underlying today's distributed training systems. We present Echo, the RL system that cleanly decouples these two phases across heterogeneous "inference" and "training" swarms while preserving statistical efficiency. Echo introduces two lightweight synchronization protocols: a sequential pull mode that refreshes policy weights according to API call for minimal bias, and an asynchronous push-pull mode that streams version-tagged rollouts through a replay buffer to maximise hardware utilisation. Training four representative RL workloads with Qwen3-4B, Qwen2.5-7B, Qwen3-30B-A3B-Thinking-2507 and Qwen3-32B on a geographically distributed cluster, Echo matches a fully co-located Verl baseline in convergence speed and final reward while off-loading trajectory generation to commodity edge hardware. These promising results demonstrate that large-scale RL for LLMs could achieve datacentre-grade performance using decentralised, heterogeneous resources.