HPC-AI Framework for RL and Agentic Workflows

As scientific research grows more complex, a wide range of HPC and AI agentic workflows have been proposed in materials design, drug screening, and molecular design. However, existing distributed frameworks and solutions fall short in facilitating intensive interactions between HPC and AI tasks, struggling with challenges like non-trivial diverse task coupling, heterogeneous task management, and inefficient I/O operations. To address these challenges, we present RTAI, a distributed runtime system aiming at promoting the performance of emerging HPC-AI workflows on HPC clusters. Our experiments demonstrate that RTAI significantly improves HPC and AI workflows, reducing workflow makespan by up to 1.8× and achieving 5× higher I/O throughput than standard baselines. Based on RTAI, an AI-driven drug screening workflow can scale up to 12.45 million processor cores and efficiently screen of up to 10¹³ synthesizable compounds. RTAI can reduce the runtime of drug screening from several months to just 22.6 hours, showing strong scalability and capability on large-scale scientific workflows.