Search papers, labs, and topics across Lattice.
The paper introduces ECHO-2, a distributed reinforcement learning framework designed to improve the cost-efficiency of post-training large language models by distributing rollout execution across remote inference workers. ECHO-2 overlaps rollout generation, dissemination, and training by treating policy staleness as a controllable parameter and uses a capacity model to optimize learner utilization given dissemination latency. Experiments on GRPO post-training of 4B and 8B models demonstrate that ECHO-2 achieves significant cost efficiency gains while maintaining comparable RL reward performance.
Achieve significant cost savings in LLM reinforcement learning by overlapping rollout generation, dissemination, and training with a framework that tolerates bounded policy staleness.
Reinforcement learning (RL) is a critical stage in post-training large language models (LLMs), involving repeated interaction between rollout generation, reward evaluation, and centralized learning. Distributing rollout execution offers opportunities to leverage more cost-efficient inference resources, but introduces challenges in wide-area coordination and policy dissemination. We present ECHO-2, a distributed RL framework for post-training with remote inference workers and non-negligible dissemination latency. ECHO-2 combines centralized learning with distributed rollouts and treats bounded policy staleness as a user-controlled parameter, enabling rollout generation, dissemination, and training to overlap. We introduce an overlap-based capacity model that relates training time, dissemination latency, and rollout throughput, yielding a practical provisioning rule for sustaining learner utilization. To mitigate dissemination bottlenecks and lower cost, ECHO-2 employs peer-assisted pipelined broadcast and cost-aware activation of heterogeneous workers. Experiments on GRPO post-training of 4B and 8B models under real wide-area bandwidth regimes show that ECHO-2 significantly improves cost efficiency while preserving RL reward comparable to strong baselines.
