作者位次 #7/16OSDI 2026Oral· Operating Systems & Distributed Systems
DynaRL: Flexible and Dynamic Scheduling of Large-Scale Reinforcement Learning Training
Yuanqing Wang · Hao Lin · Junhao Hu · Chunyang Zhu · Quanlu Zhang · Zhen Guo · Yuchen Zhang · Xu Fu · Si Xu · Bo Dai · Zixiao Huang · Chao Yu · Boxun Li · Guohao Dai · Tmall Group of Alibaba) · Yu Wang
摘要
Modern reinforcement learning (RL) workloads, powering large language models, long‑horizon reasoning, and agentic systems, exhibit extreme dynamicity due to heavy‑tailed rollouts, irregular multi‑turn tool interactions, and time‑varying bottlenecks. Static resource allocations in today’s distribu...