ARL-Tangram: Unleash the Resource Efficiency in Agentic Reinforcement Learning
This work addresses resource management inefficiencies for cloud-based agentic RL systems, offering significant performance gains and cost savings, though it is incremental as it builds on existing frameworks with a novel orchestration approach.
The paper tackles the problem of resource inefficiency in agentic reinforcement learning (RL) due to static over-provisioning of external cloud resources, and proposes ARL-Tangram, a unified resource management system that improves average action completion time by up to 4.3×, speeds up RL training step duration by up to 1.5×, and saves external resources by up to 71.2%.
Agentic reinforcement learning (RL) has emerged as a transformative workload in cloud clusters, enabling large language models (LLMs) to solve complex problems through interactions with real world. However, unlike traditional RL, agentic RL demands substantial external cloud resources, e.g., CPUs for code execution and GPUs for reward models, that exist outside the primary training cluster. Existing agentic RL framework typically rely on static over-provisioning, i.e., resources are often tied to long-lived trajectories or isolated by tasks, which leads to severe resource inefficiency. We propose the action-level orchestration, and incorporate it into ARL-Tangram, a unified resource management system that enables fine-grained external resource sharing and elasticity. ARL-Tangram utilizes a unified action-level formulation and an elastic scheduling algorithm to minimize action completion time (ACT) while satisfying heterogeneous resource constraints. Further, heterogeneous resource managers are tailored to efficiently support the action-level execution on resources with heterogeneous characteristics and topologies. Evaluation on real-world agentic RL tasks demonstrates that ARL-Tangram improves average ACT by up to 4.3$\times$, speeds up the step duration of RL training by up to 1.5$\times$, and saves the external resources by up to 71.2$\%$. This system has been deployed to support the training of the MiMo series models.